diff --git a/api/except.txt b/api/except.txt index 662fa16f813..798b2c2f413 100644 --- a/api/except.txt +++ b/api/except.txt @@ -2,12 +2,56 @@ pkg encoding/json, method (*RawMessage) MarshalJSON() ([]uint8, error) pkg math/big, const MaxBase = 36 pkg math/big, type Word uintptr pkg net, func ListenUnixgram(string, *UnixAddr) (*UDPConn, error) +pkg os (linux-arm), const O_SYNC = 1052672 +pkg os (linux-arm), const O_SYNC = 4096 +pkg os (linux-arm-cgo), const O_SYNC = 1052672 +pkg os (linux-arm-cgo), const O_SYNC = 4096 +pkg os, const ModeAppend FileMode +pkg os, const ModeCharDevice FileMode +pkg os, const ModeDevice FileMode +pkg os, const ModeDir FileMode +pkg os, const ModeExclusive FileMode +pkg os, const ModeIrregular FileMode +pkg os, const ModeNamedPipe FileMode +pkg os, const ModePerm FileMode +pkg os, const ModeSetgid FileMode +pkg os, const ModeSetuid FileMode +pkg os, const ModeSocket FileMode +pkg os, const ModeSticky FileMode +pkg os, const ModeSymlink FileMode +pkg os, const ModeTemporary FileMode pkg os, const ModeType = 2399141888 pkg os, const ModeType = 2399666176 -pkg os (linux-arm), const O_SYNC = 4096 -pkg os (linux-arm-cgo), const O_SYNC = 4096 -pkg os (linux-arm), const O_SYNC = 1052672 -pkg os (linux-arm-cgo), const O_SYNC = 1052672 +pkg os, const ModeType FileMode +pkg os, func Chmod(string, FileMode) error +pkg os, func Lstat(string) (FileInfo, error) +pkg os, func Mkdir(string, FileMode) error +pkg os, func MkdirAll(string, FileMode) error +pkg os, func OpenFile(string, int, FileMode) (*File, error) +pkg os, func SameFile(FileInfo, FileInfo) bool +pkg os, func Stat(string) (FileInfo, error) +pkg os, method (*File) Chmod(FileMode) error +pkg os, method (*File) Readdir(int) ([]FileInfo, error) +pkg os, method (*File) Stat() (FileInfo, error) +pkg os, method (*PathError) Error() string +pkg os, method (*PathError) Timeout() bool +pkg os, method (*PathError) Unwrap() error +pkg os, method (FileMode) IsDir() bool +pkg os, method (FileMode) IsRegular() bool +pkg os, method (FileMode) Perm() FileMode +pkg os, method (FileMode) String() string +pkg os, type FileInfo interface { IsDir, ModTime, Mode, Name, Size, Sys } +pkg os, type FileInfo interface, IsDir() bool +pkg os, type FileInfo interface, ModTime() time.Time +pkg os, type FileInfo interface, Mode() FileMode +pkg os, type FileInfo interface, Name() string +pkg os, type FileInfo interface, Size() int64 +pkg os, type FileInfo interface, Sys() interface{} +pkg os, type FileMode uint32 +pkg os, type PathError struct +pkg os, type PathError struct, Err error +pkg os, type PathError struct, Op string +pkg os, type PathError struct, Path string pkg syscall (darwin-amd64), const ImplementsGetwd = false pkg syscall (darwin-amd64), func Fchflags(string, int) error pkg syscall (darwin-amd64-cgo), const ImplementsGetwd = false @@ -18,22 +62,72 @@ pkg syscall (freebsd-386), const ELAST = 94 pkg syscall (freebsd-386), const ImplementsGetwd = false pkg syscall (freebsd-386), const O_CLOEXEC = 0 pkg syscall (freebsd-386), func Fchflags(string, int) error +pkg syscall (freebsd-386), func Mknod(string, uint32, int) error +pkg syscall (freebsd-386), type Dirent struct, Fileno uint32 +pkg syscall (freebsd-386), type Dirent struct, Namlen uint8 +pkg syscall (freebsd-386), type Stat_t struct, Blksize uint32 +pkg syscall (freebsd-386), type Stat_t struct, Dev uint32 +pkg syscall (freebsd-386), type Stat_t struct, Gen uint32 +pkg syscall (freebsd-386), type Stat_t struct, Ino uint32 +pkg syscall (freebsd-386), type Stat_t struct, Lspare int32 +pkg syscall (freebsd-386), type Stat_t struct, Nlink uint16 +pkg syscall (freebsd-386), type Stat_t struct, Pad_cgo_0 [8]uint8 +pkg syscall (freebsd-386), type Stat_t struct, Rdev uint32 +pkg syscall (freebsd-386), type Statfs_t struct, Mntfromname [88]int8 +pkg syscall (freebsd-386), type Statfs_t struct, Mntonname [88]int8 pkg syscall (freebsd-386-cgo), const AF_MAX = 38 pkg syscall (freebsd-386-cgo), const DLT_MATCHING_MAX = 242 pkg syscall (freebsd-386-cgo), const ELAST = 94 pkg syscall (freebsd-386-cgo), const ImplementsGetwd = false pkg syscall (freebsd-386-cgo), const O_CLOEXEC = 0 +pkg syscall (freebsd-386-cgo), func Mknod(string, uint32, int) error +pkg syscall (freebsd-386-cgo), type Dirent struct, Fileno uint32 +pkg syscall (freebsd-386-cgo), type Dirent struct, Namlen uint8 +pkg syscall (freebsd-386-cgo), type Stat_t struct, Blksize uint32 +pkg syscall (freebsd-386-cgo), type Stat_t struct, Dev uint32 +pkg syscall (freebsd-386-cgo), type Stat_t struct, Gen uint32 +pkg syscall (freebsd-386-cgo), type Stat_t struct, Ino uint32 +pkg syscall (freebsd-386-cgo), type Stat_t struct, Lspare int32 +pkg syscall (freebsd-386-cgo), type Stat_t struct, Nlink uint16 +pkg syscall (freebsd-386-cgo), type Stat_t struct, Pad_cgo_0 [8]uint8 +pkg syscall (freebsd-386-cgo), type Stat_t struct, Rdev uint32 +pkg syscall (freebsd-386-cgo), type Statfs_t struct, Mntfromname [88]int8 +pkg syscall (freebsd-386-cgo), type Statfs_t struct, Mntonname [88]int8 pkg syscall (freebsd-amd64), const AF_MAX = 38 pkg syscall (freebsd-amd64), const DLT_MATCHING_MAX = 242 pkg syscall (freebsd-amd64), const ELAST = 94 pkg syscall (freebsd-amd64), const ImplementsGetwd = false pkg syscall (freebsd-amd64), const O_CLOEXEC = 0 pkg syscall (freebsd-amd64), func Fchflags(string, int) error +pkg syscall (freebsd-amd64), func Mknod(string, uint32, int) error +pkg syscall (freebsd-amd64), type Dirent struct, Fileno uint32 +pkg syscall (freebsd-amd64), type Dirent struct, Namlen uint8 +pkg syscall (freebsd-amd64), type Stat_t struct, Blksize uint32 +pkg syscall (freebsd-amd64), type Stat_t struct, Dev uint32 +pkg syscall (freebsd-amd64), type Stat_t struct, Gen uint32 +pkg syscall (freebsd-amd64), type Stat_t struct, Ino uint32 +pkg syscall (freebsd-amd64), type Stat_t struct, Lspare int32 +pkg syscall (freebsd-amd64), type Stat_t struct, Nlink uint16 +pkg syscall (freebsd-amd64), type Stat_t struct, Rdev uint32 +pkg syscall (freebsd-amd64), type Statfs_t struct, Mntfromname [88]int8 +pkg syscall (freebsd-amd64), type Statfs_t struct, Mntonname [88]int8 pkg syscall (freebsd-amd64-cgo), const AF_MAX = 38 pkg syscall (freebsd-amd64-cgo), const DLT_MATCHING_MAX = 242 pkg syscall (freebsd-amd64-cgo), const ELAST = 94 pkg syscall (freebsd-amd64-cgo), const ImplementsGetwd = false pkg syscall (freebsd-amd64-cgo), const O_CLOEXEC = 0 +pkg syscall (freebsd-amd64-cgo), func Mknod(string, uint32, int) error +pkg syscall (freebsd-amd64-cgo), type Dirent struct, Fileno uint32 +pkg syscall (freebsd-amd64-cgo), type Dirent struct, Namlen uint8 +pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Blksize uint32 +pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Dev uint32 +pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Gen uint32 +pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Ino uint32 +pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Lspare int32 +pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Nlink uint16 +pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Rdev uint32 +pkg syscall (freebsd-amd64-cgo), type Statfs_t struct, Mntfromname [88]int8 +pkg syscall (freebsd-amd64-cgo), type Statfs_t struct, Mntonname [88]int8 pkg syscall (freebsd-arm), const AF_MAX = 38 pkg syscall (freebsd-arm), const BIOCGRTIMEOUT = 1074545262 pkg syscall (freebsd-arm), const BIOCSRTIMEOUT = 2148287085 @@ -62,10 +156,22 @@ pkg syscall (freebsd-arm), const SizeofSockaddrDatalink = 56 pkg syscall (freebsd-arm), const SizeofSockaddrUnix = 108 pkg syscall (freebsd-arm), const TIOCTIMESTAMP = 1074558041 pkg syscall (freebsd-arm), func Fchflags(string, int) error +pkg syscall (freebsd-arm), func Mknod(string, uint32, int) error pkg syscall (freebsd-arm), type BpfHdr struct, Pad_cgo_0 [2]uint8 +pkg syscall (freebsd-arm), type Dirent struct, Fileno uint32 +pkg syscall (freebsd-arm), type Dirent struct, Namlen uint8 pkg syscall (freebsd-arm), type RawSockaddrDatalink struct, Pad_cgo_0 [2]uint8 pkg syscall (freebsd-arm), type RawSockaddrUnix struct, Pad_cgo_0 [2]uint8 +pkg syscall (freebsd-arm), type Stat_t struct, Blksize uint32 +pkg syscall (freebsd-arm), type Stat_t struct, Dev uint32 +pkg syscall (freebsd-arm), type Stat_t struct, Gen uint32 +pkg syscall (freebsd-arm), type Stat_t struct, Ino uint32 +pkg syscall (freebsd-arm), type Stat_t struct, Lspare int32 +pkg syscall (freebsd-arm), type Stat_t struct, Nlink uint16 pkg syscall (freebsd-arm), type Stat_t struct, Pad_cgo_0 [4]uint8 +pkg syscall (freebsd-arm), type Stat_t struct, Rdev uint32 +pkg syscall (freebsd-arm), type Statfs_t struct, Mntfromname [88]int8 +pkg syscall (freebsd-arm), type Statfs_t struct, Mntonname [88]int8 pkg syscall (freebsd-arm-cgo), const AF_MAX = 38 pkg syscall (freebsd-arm-cgo), const BIOCGRTIMEOUT = 1074545262 pkg syscall (freebsd-arm-cgo), const BIOCSRTIMEOUT = 2148287085 @@ -94,10 +200,22 @@ pkg syscall (freebsd-arm-cgo), const SizeofSockaddrDatalink = 56 pkg syscall (freebsd-arm-cgo), const SizeofSockaddrUnix = 108 pkg syscall (freebsd-arm-cgo), const TIOCTIMESTAMP = 1074558041 pkg syscall (freebsd-arm-cgo), func Fchflags(string, int) error +pkg syscall (freebsd-arm-cgo), func Mknod(string, uint32, int) error pkg syscall (freebsd-arm-cgo), type BpfHdr struct, Pad_cgo_0 [2]uint8 +pkg syscall (freebsd-arm-cgo), type Dirent struct, Fileno uint32 +pkg syscall (freebsd-arm-cgo), type Dirent struct, Namlen uint8 pkg syscall (freebsd-arm-cgo), type RawSockaddrDatalink struct, Pad_cgo_0 [2]uint8 pkg syscall (freebsd-arm-cgo), type RawSockaddrUnix struct, Pad_cgo_0 [2]uint8 +pkg syscall (freebsd-arm-cgo), type Stat_t struct, Blksize uint32 +pkg syscall (freebsd-arm-cgo), type Stat_t struct, Dev uint32 +pkg syscall (freebsd-arm-cgo), type Stat_t struct, Gen uint32 +pkg syscall (freebsd-arm-cgo), type Stat_t struct, Ino uint32 +pkg syscall (freebsd-arm-cgo), type Stat_t struct, Lspare int32 +pkg syscall (freebsd-arm-cgo), type Stat_t struct, Nlink uint16 pkg syscall (freebsd-arm-cgo), type Stat_t struct, Pad_cgo_0 [4]uint8 +pkg syscall (freebsd-arm-cgo), type Stat_t struct, Rdev uint32 +pkg syscall (freebsd-arm-cgo), type Statfs_t struct, Mntfromname [88]int8 +pkg syscall (freebsd-arm-cgo), type Statfs_t struct, Mntonname [88]int8 pkg syscall (linux-386), type Cmsghdr struct, X__cmsg_data [0]uint8 pkg syscall (linux-386-cgo), type Cmsghdr struct, X__cmsg_data [0]uint8 pkg syscall (linux-amd64), type Cmsghdr struct, X__cmsg_data [0]uint8 @@ -109,10 +227,10 @@ pkg syscall (netbsd-386-cgo), const ImplementsGetwd = false pkg syscall (netbsd-amd64), const ImplementsGetwd = false pkg syscall (netbsd-amd64-cgo), const ImplementsGetwd = false pkg syscall (netbsd-arm), const ImplementsGetwd = false -pkg syscall (netbsd-arm-cgo), const ImplementsGetwd = false pkg syscall (netbsd-arm), const SizeofIfData = 132 pkg syscall (netbsd-arm), func Fchflags(string, int) error pkg syscall (netbsd-arm), type IfMsghdr struct, Pad_cgo_1 [4]uint8 +pkg syscall (netbsd-arm-cgo), const ImplementsGetwd = false pkg syscall (netbsd-arm-cgo), const SizeofIfData = 132 pkg syscall (netbsd-arm-cgo), func Fchflags(string, int) error pkg syscall (netbsd-arm-cgo), type IfMsghdr struct, Pad_cgo_1 [4]uint8 @@ -140,6 +258,7 @@ pkg syscall (openbsd-386), const SYS_GETITIMER = 86 pkg syscall (openbsd-386), const SYS_GETRUSAGE = 117 pkg syscall (openbsd-386), const SYS_GETTIMEOFDAY = 116 pkg syscall (openbsd-386), const SYS_KEVENT = 270 +pkg syscall (openbsd-386), const SYS_KILL = 37 pkg syscall (openbsd-386), const SYS_LSTAT = 293 pkg syscall (openbsd-386), const SYS_NANOSLEEP = 240 pkg syscall (openbsd-386), const SYS_SELECT = 93 @@ -193,6 +312,7 @@ pkg syscall (openbsd-386-cgo), const SYS_GETITIMER = 86 pkg syscall (openbsd-386-cgo), const SYS_GETRUSAGE = 117 pkg syscall (openbsd-386-cgo), const SYS_GETTIMEOFDAY = 116 pkg syscall (openbsd-386-cgo), const SYS_KEVENT = 270 +pkg syscall (openbsd-386-cgo), const SYS_KILL = 37 pkg syscall (openbsd-386-cgo), const SYS_LSTAT = 293 pkg syscall (openbsd-386-cgo), const SYS_NANOSLEEP = 240 pkg syscall (openbsd-386-cgo), const SYS_SELECT = 93 @@ -257,6 +377,7 @@ pkg syscall (openbsd-amd64), const SYS_GETITIMER = 86 pkg syscall (openbsd-amd64), const SYS_GETRUSAGE = 117 pkg syscall (openbsd-amd64), const SYS_GETTIMEOFDAY = 116 pkg syscall (openbsd-amd64), const SYS_KEVENT = 270 +pkg syscall (openbsd-amd64), const SYS_KILL = 37 pkg syscall (openbsd-amd64), const SYS_LSTAT = 293 pkg syscall (openbsd-amd64), const SYS_NANOSLEEP = 240 pkg syscall (openbsd-amd64), const SYS_SELECT = 93 @@ -320,6 +441,7 @@ pkg syscall (openbsd-amd64-cgo), const SYS_GETITIMER = 86 pkg syscall (openbsd-amd64-cgo), const SYS_GETRUSAGE = 117 pkg syscall (openbsd-amd64-cgo), const SYS_GETTIMEOFDAY = 116 pkg syscall (openbsd-amd64-cgo), const SYS_KEVENT = 270 +pkg syscall (openbsd-amd64-cgo), const SYS_KILL = 37 pkg syscall (openbsd-amd64-cgo), const SYS_LSTAT = 293 pkg syscall (openbsd-amd64-cgo), const SYS_NANOSLEEP = 240 pkg syscall (openbsd-amd64-cgo), const SYS_SELECT = 93 @@ -348,20 +470,6 @@ pkg syscall (openbsd-amd64-cgo), type Statfs_t struct, F_spare [3]uint32 pkg syscall (openbsd-amd64-cgo), type Statfs_t struct, Pad_cgo_1 [4]uint8 pkg syscall (openbsd-amd64-cgo), type Timespec struct, Pad_cgo_0 [4]uint8 pkg syscall (openbsd-amd64-cgo), type Timespec struct, Sec int32 -pkg testing, func RegisterCover(Cover) -pkg testing, func MainStart(func(string, string) (bool, error), []InternalTest, []InternalBenchmark, []InternalExample) *M -pkg testing, func MainStart(testDeps, []InternalTest, []InternalBenchmark, []InternalExample) *M -pkg text/template/parse, type DotNode bool -pkg text/template/parse, type Node interface { Copy, String, Type } -pkg unicode, const Version = "6.2.0" -pkg unicode, const Version = "6.3.0" -pkg unicode, const Version = "7.0.0" -pkg unicode, const Version = "8.0.0" -pkg syscall (openbsd-386), const SYS_KILL = 37 -pkg syscall (openbsd-386-cgo), const SYS_KILL = 37 -pkg syscall (openbsd-amd64), const SYS_KILL = 37 -pkg syscall (openbsd-amd64-cgo), const SYS_KILL = 37 -pkg unicode, const Version = "9.0.0" pkg syscall (windows-386), const TOKEN_ALL_ACCESS = 983295 pkg syscall (windows-386), type AddrinfoW struct, Addr uintptr pkg syscall (windows-386), type CertChainPolicyPara struct, ExtraPolicyPara uintptr @@ -380,81 +488,17 @@ pkg syscall (windows-amd64), type CertRevocationInfo struct, CrlInfo uintptr pkg syscall (windows-amd64), type CertRevocationInfo struct, OidSpecificInfo uintptr pkg syscall (windows-amd64), type CertSimpleChain struct, TrustListInfo uintptr pkg syscall (windows-amd64), type RawSockaddrAny struct, Pad [96]int8 -pkg syscall (freebsd-386), func Mknod(string, uint32, int) error -pkg syscall (freebsd-386), type Dirent struct, Fileno uint32 -pkg syscall (freebsd-386), type Dirent struct, Namlen uint8 -pkg syscall (freebsd-386), type Stat_t struct, Blksize uint32 -pkg syscall (freebsd-386), type Stat_t struct, Dev uint32 -pkg syscall (freebsd-386), type Stat_t struct, Gen uint32 -pkg syscall (freebsd-386), type Stat_t struct, Ino uint32 -pkg syscall (freebsd-386), type Stat_t struct, Lspare int32 -pkg syscall (freebsd-386), type Stat_t struct, Nlink uint16 -pkg syscall (freebsd-386), type Stat_t struct, Pad_cgo_0 [8]uint8 -pkg syscall (freebsd-386), type Stat_t struct, Rdev uint32 -pkg syscall (freebsd-386), type Statfs_t struct, Mntfromname [88]int8 -pkg syscall (freebsd-386), type Statfs_t struct, Mntonname [88]int8 -pkg syscall (freebsd-386-cgo), func Mknod(string, uint32, int) error -pkg syscall (freebsd-386-cgo), type Dirent struct, Fileno uint32 -pkg syscall (freebsd-386-cgo), type Dirent struct, Namlen uint8 -pkg syscall (freebsd-386-cgo), type Stat_t struct, Blksize uint32 -pkg syscall (freebsd-386-cgo), type Stat_t struct, Dev uint32 -pkg syscall (freebsd-386-cgo), type Stat_t struct, Gen uint32 -pkg syscall (freebsd-386-cgo), type Stat_t struct, Ino uint32 -pkg syscall (freebsd-386-cgo), type Stat_t struct, Lspare int32 -pkg syscall (freebsd-386-cgo), type Stat_t struct, Nlink uint16 -pkg syscall (freebsd-386-cgo), type Stat_t struct, Pad_cgo_0 [8]uint8 -pkg syscall (freebsd-386-cgo), type Stat_t struct, Rdev uint32 -pkg syscall (freebsd-386-cgo), type Statfs_t struct, Mntfromname [88]int8 -pkg syscall (freebsd-386-cgo), type Statfs_t struct, Mntonname [88]int8 -pkg syscall (freebsd-amd64), func Mknod(string, uint32, int) error -pkg syscall (freebsd-amd64), type Dirent struct, Fileno uint32 -pkg syscall (freebsd-amd64), type Dirent struct, Namlen uint8 -pkg syscall (freebsd-amd64), type Stat_t struct, Blksize uint32 -pkg syscall (freebsd-amd64), type Stat_t struct, Dev uint32 -pkg syscall (freebsd-amd64), type Stat_t struct, Gen uint32 -pkg syscall (freebsd-amd64), type Stat_t struct, Ino uint32 -pkg syscall (freebsd-amd64), type Stat_t struct, Lspare int32 -pkg syscall (freebsd-amd64), type Stat_t struct, Nlink uint16 -pkg syscall (freebsd-amd64), type Stat_t struct, Rdev uint32 -pkg syscall (freebsd-amd64), type Statfs_t struct, Mntfromname [88]int8 -pkg syscall (freebsd-amd64), type Statfs_t struct, Mntonname [88]int8 -pkg syscall (freebsd-amd64-cgo), func Mknod(string, uint32, int) error -pkg syscall (freebsd-amd64-cgo), type Dirent struct, Fileno uint32 -pkg syscall (freebsd-amd64-cgo), type Dirent struct, Namlen uint8 -pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Blksize uint32 -pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Dev uint32 -pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Gen uint32 -pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Ino uint32 -pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Lspare int32 -pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Nlink uint16 -pkg syscall (freebsd-amd64-cgo), type Stat_t struct, Rdev uint32 -pkg syscall (freebsd-amd64-cgo), type Statfs_t struct, Mntfromname [88]int8 -pkg syscall (freebsd-amd64-cgo), type Statfs_t struct, Mntonname [88]int8 -pkg syscall (freebsd-arm), func Mknod(string, uint32, int) error -pkg syscall (freebsd-arm), type Dirent struct, Fileno uint32 -pkg syscall (freebsd-arm), type Dirent struct, Namlen uint8 -pkg syscall (freebsd-arm), type Stat_t struct, Blksize uint32 -pkg syscall (freebsd-arm), type Stat_t struct, Dev uint32 -pkg syscall (freebsd-arm), type Stat_t struct, Gen uint32 -pkg syscall (freebsd-arm), type Stat_t struct, Ino uint32 -pkg syscall (freebsd-arm), type Stat_t struct, Lspare int32 -pkg syscall (freebsd-arm), type Stat_t struct, Nlink uint16 -pkg syscall (freebsd-arm), type Stat_t struct, Rdev uint32 -pkg syscall (freebsd-arm), type Statfs_t struct, Mntfromname [88]int8 -pkg syscall (freebsd-arm), type Statfs_t struct, Mntonname [88]int8 -pkg syscall (freebsd-arm-cgo), func Mknod(string, uint32, int) error -pkg syscall (freebsd-arm-cgo), type Dirent struct, Fileno uint32 -pkg syscall (freebsd-arm-cgo), type Dirent struct, Namlen uint8 -pkg syscall (freebsd-arm-cgo), type Stat_t struct, Blksize uint32 -pkg syscall (freebsd-arm-cgo), type Stat_t struct, Dev uint32 -pkg syscall (freebsd-arm-cgo), type Stat_t struct, Gen uint32 -pkg syscall (freebsd-arm-cgo), type Stat_t struct, Ino uint32 -pkg syscall (freebsd-arm-cgo), type Stat_t struct, Lspare int32 -pkg syscall (freebsd-arm-cgo), type Stat_t struct, Nlink uint16 -pkg syscall (freebsd-arm-cgo), type Stat_t struct, Rdev uint32 -pkg syscall (freebsd-arm-cgo), type Statfs_t struct, Mntfromname [88]int8 -pkg syscall (freebsd-arm-cgo), type Statfs_t struct, Mntonname [88]int8 +pkg testing, func MainStart(func(string, string) (bool, error), []InternalTest, []InternalBenchmark, []InternalExample) *M +pkg testing, func MainStart(testDeps, []InternalTest, []InternalBenchmark, []InternalExample) *M +pkg testing, func RegisterCover(Cover) pkg text/scanner, const GoTokens = 1012 +pkg text/template/parse, type DotNode bool +pkg text/template/parse, type Node interface { Copy, String, Type } pkg unicode, const Version = "10.0.0" pkg unicode, const Version = "11.0.0" pkg unicode, const Version = "12.0.0" +pkg unicode, const Version = "6.2.0" +pkg unicode, const Version = "6.3.0" +pkg unicode, const Version = "7.0.0" +pkg unicode, const Version = "8.0.0" +pkg unicode, const Version = "9.0.0" diff --git a/api/next.txt b/api/next.txt index 71fdaddc08c..7e5378da5ee 100644 --- a/api/next.txt +++ b/api/next.txt @@ -1,22 +1,420 @@ -pkg unicode, const Version = "13.0.0" -pkg unicode, var Chorasmian *RangeTable -pkg unicode, var Dives_Akuru *RangeTable -pkg unicode, var Khitan_Small_Script *RangeTable -pkg unicode, var Yezidi *RangeTable -pkg text/template/parse, const NodeComment = 20 -pkg text/template/parse, const NodeComment NodeType -pkg text/template/parse, const ParseComments = 1 -pkg text/template/parse, const ParseComments Mode -pkg text/template/parse, method (*CommentNode) Copy() Node -pkg text/template/parse, method (*CommentNode) String() string -pkg text/template/parse, method (CommentNode) Position() Pos -pkg text/template/parse, method (CommentNode) Type() NodeType -pkg text/template/parse, type CommentNode struct -pkg text/template/parse, type CommentNode struct, Text string -pkg text/template/parse, type CommentNode struct, embedded NodeType -pkg text/template/parse, type CommentNode struct, embedded Pos -pkg text/template/parse, type Mode uint -pkg text/template/parse, type Tree struct, Mode Mode +pkg archive/zip, method (*ReadCloser) Open(string) (fs.File, error) +pkg archive/zip, method (*Reader) Open(string) (fs.File, error) +pkg debug/elf, const DT_ADDRRNGHI = 1879047935 +pkg debug/elf, const DT_ADDRRNGHI DynTag +pkg debug/elf, const DT_ADDRRNGLO = 1879047680 +pkg debug/elf, const DT_ADDRRNGLO DynTag +pkg debug/elf, const DT_AUDIT = 1879047932 +pkg debug/elf, const DT_AUDIT DynTag +pkg debug/elf, const DT_AUXILIARY = 2147483645 +pkg debug/elf, const DT_AUXILIARY DynTag +pkg debug/elf, const DT_CHECKSUM = 1879047672 +pkg debug/elf, const DT_CHECKSUM DynTag +pkg debug/elf, const DT_CONFIG = 1879047930 +pkg debug/elf, const DT_CONFIG DynTag +pkg debug/elf, const DT_DEPAUDIT = 1879047931 +pkg debug/elf, const DT_DEPAUDIT DynTag +pkg debug/elf, const DT_FEATURE = 1879047676 +pkg debug/elf, const DT_FEATURE DynTag +pkg debug/elf, const DT_FILTER = 2147483647 +pkg debug/elf, const DT_FILTER DynTag +pkg debug/elf, const DT_FLAGS_1 = 1879048187 +pkg debug/elf, const DT_FLAGS_1 DynTag +pkg debug/elf, const DT_GNU_CONFLICT = 1879047928 +pkg debug/elf, const DT_GNU_CONFLICT DynTag +pkg debug/elf, const DT_GNU_CONFLICTSZ = 1879047670 +pkg debug/elf, const DT_GNU_CONFLICTSZ DynTag +pkg debug/elf, const DT_GNU_HASH = 1879047925 +pkg debug/elf, const DT_GNU_HASH DynTag +pkg debug/elf, const DT_GNU_LIBLIST = 1879047929 +pkg debug/elf, const DT_GNU_LIBLIST DynTag +pkg debug/elf, const DT_GNU_LIBLISTSZ = 1879047671 +pkg debug/elf, const DT_GNU_LIBLISTSZ DynTag +pkg debug/elf, const DT_GNU_PRELINKED = 1879047669 +pkg debug/elf, const DT_GNU_PRELINKED DynTag +pkg debug/elf, const DT_MIPS_AUX_DYNAMIC = 1879048241 +pkg debug/elf, const DT_MIPS_AUX_DYNAMIC DynTag +pkg debug/elf, const DT_MIPS_BASE_ADDRESS = 1879048198 +pkg debug/elf, const DT_MIPS_BASE_ADDRESS DynTag +pkg debug/elf, const DT_MIPS_COMPACT_SIZE = 1879048239 +pkg debug/elf, const DT_MIPS_COMPACT_SIZE DynTag +pkg debug/elf, const DT_MIPS_CONFLICT = 1879048200 +pkg debug/elf, const DT_MIPS_CONFLICT DynTag +pkg debug/elf, const DT_MIPS_CONFLICTNO = 1879048203 +pkg debug/elf, const DT_MIPS_CONFLICTNO DynTag +pkg debug/elf, const DT_MIPS_CXX_FLAGS = 1879048226 +pkg debug/elf, const DT_MIPS_CXX_FLAGS DynTag +pkg debug/elf, const DT_MIPS_DELTA_CLASS = 1879048215 +pkg debug/elf, const DT_MIPS_DELTA_CLASS DynTag +pkg debug/elf, const DT_MIPS_DELTA_CLASSSYM = 1879048224 +pkg debug/elf, const DT_MIPS_DELTA_CLASSSYM DynTag +pkg debug/elf, const DT_MIPS_DELTA_CLASSSYM_NO = 1879048225 +pkg debug/elf, const DT_MIPS_DELTA_CLASSSYM_NO DynTag +pkg debug/elf, const DT_MIPS_DELTA_CLASS_NO = 1879048216 +pkg debug/elf, const DT_MIPS_DELTA_CLASS_NO DynTag +pkg debug/elf, const DT_MIPS_DELTA_INSTANCE = 1879048217 +pkg debug/elf, const DT_MIPS_DELTA_INSTANCE DynTag +pkg debug/elf, const DT_MIPS_DELTA_INSTANCE_NO = 1879048218 +pkg debug/elf, const DT_MIPS_DELTA_INSTANCE_NO DynTag +pkg debug/elf, const DT_MIPS_DELTA_RELOC = 1879048219 +pkg debug/elf, const DT_MIPS_DELTA_RELOC DynTag +pkg debug/elf, const DT_MIPS_DELTA_RELOC_NO = 1879048220 +pkg debug/elf, const DT_MIPS_DELTA_RELOC_NO DynTag +pkg debug/elf, const DT_MIPS_DELTA_SYM = 1879048221 +pkg debug/elf, const DT_MIPS_DELTA_SYM DynTag +pkg debug/elf, const DT_MIPS_DELTA_SYM_NO = 1879048222 +pkg debug/elf, const DT_MIPS_DELTA_SYM_NO DynTag +pkg debug/elf, const DT_MIPS_DYNSTR_ALIGN = 1879048235 +pkg debug/elf, const DT_MIPS_DYNSTR_ALIGN DynTag +pkg debug/elf, const DT_MIPS_FLAGS = 1879048197 +pkg debug/elf, const DT_MIPS_FLAGS DynTag +pkg debug/elf, const DT_MIPS_GOTSYM = 1879048211 +pkg debug/elf, const DT_MIPS_GOTSYM DynTag +pkg debug/elf, const DT_MIPS_GP_VALUE = 1879048240 +pkg debug/elf, const DT_MIPS_GP_VALUE DynTag +pkg debug/elf, const DT_MIPS_HIDDEN_GOTIDX = 1879048231 +pkg debug/elf, const DT_MIPS_HIDDEN_GOTIDX DynTag +pkg debug/elf, const DT_MIPS_HIPAGENO = 1879048212 +pkg debug/elf, const DT_MIPS_HIPAGENO DynTag +pkg debug/elf, const DT_MIPS_ICHECKSUM = 1879048195 +pkg debug/elf, const DT_MIPS_ICHECKSUM DynTag +pkg debug/elf, const DT_MIPS_INTERFACE = 1879048234 +pkg debug/elf, const DT_MIPS_INTERFACE DynTag +pkg debug/elf, const DT_MIPS_INTERFACE_SIZE = 1879048236 +pkg debug/elf, const DT_MIPS_INTERFACE_SIZE DynTag +pkg debug/elf, const DT_MIPS_IVERSION = 1879048196 +pkg debug/elf, const DT_MIPS_IVERSION DynTag +pkg debug/elf, const DT_MIPS_LIBLIST = 1879048201 +pkg debug/elf, const DT_MIPS_LIBLIST DynTag +pkg debug/elf, const DT_MIPS_LIBLISTNO = 1879048208 +pkg debug/elf, const DT_MIPS_LIBLISTNO DynTag +pkg debug/elf, const DT_MIPS_LOCALPAGE_GOTIDX = 1879048229 +pkg debug/elf, const DT_MIPS_LOCALPAGE_GOTIDX DynTag +pkg debug/elf, const DT_MIPS_LOCAL_GOTIDX = 1879048230 +pkg debug/elf, const DT_MIPS_LOCAL_GOTIDX DynTag +pkg debug/elf, const DT_MIPS_LOCAL_GOTNO = 1879048202 +pkg debug/elf, const DT_MIPS_LOCAL_GOTNO DynTag +pkg debug/elf, const DT_MIPS_MSYM = 1879048199 +pkg debug/elf, const DT_MIPS_MSYM DynTag +pkg debug/elf, const DT_MIPS_OPTIONS = 1879048233 +pkg debug/elf, const DT_MIPS_OPTIONS DynTag +pkg debug/elf, const DT_MIPS_PERF_SUFFIX = 1879048238 +pkg debug/elf, const DT_MIPS_PERF_SUFFIX DynTag +pkg debug/elf, const DT_MIPS_PIXIE_INIT = 1879048227 +pkg debug/elf, const DT_MIPS_PIXIE_INIT DynTag +pkg debug/elf, const DT_MIPS_PLTGOT = 1879048242 +pkg debug/elf, const DT_MIPS_PLTGOT DynTag +pkg debug/elf, const DT_MIPS_PROTECTED_GOTIDX = 1879048232 +pkg debug/elf, const DT_MIPS_PROTECTED_GOTIDX DynTag +pkg debug/elf, const DT_MIPS_RLD_MAP = 1879048214 +pkg debug/elf, const DT_MIPS_RLD_MAP DynTag +pkg debug/elf, const DT_MIPS_RLD_MAP_REL = 1879048245 +pkg debug/elf, const DT_MIPS_RLD_MAP_REL DynTag +pkg debug/elf, const DT_MIPS_RLD_TEXT_RESOLVE_ADDR = 1879048237 +pkg debug/elf, const DT_MIPS_RLD_TEXT_RESOLVE_ADDR DynTag +pkg debug/elf, const DT_MIPS_RLD_VERSION = 1879048193 +pkg debug/elf, const DT_MIPS_RLD_VERSION DynTag +pkg debug/elf, const DT_MIPS_RWPLT = 1879048244 +pkg debug/elf, const DT_MIPS_RWPLT DynTag +pkg debug/elf, const DT_MIPS_SYMBOL_LIB = 1879048228 +pkg debug/elf, const DT_MIPS_SYMBOL_LIB DynTag +pkg debug/elf, const DT_MIPS_SYMTABNO = 1879048209 +pkg debug/elf, const DT_MIPS_SYMTABNO DynTag +pkg debug/elf, const DT_MIPS_TIME_STAMP = 1879048194 +pkg debug/elf, const DT_MIPS_TIME_STAMP DynTag +pkg debug/elf, const DT_MIPS_UNREFEXTNO = 1879048210 +pkg debug/elf, const DT_MIPS_UNREFEXTNO DynTag +pkg debug/elf, const DT_MOVEENT = 1879047674 +pkg debug/elf, const DT_MOVEENT DynTag +pkg debug/elf, const DT_MOVESZ = 1879047675 +pkg debug/elf, const DT_MOVESZ DynTag +pkg debug/elf, const DT_MOVETAB = 1879047934 +pkg debug/elf, const DT_MOVETAB DynTag +pkg debug/elf, const DT_PLTPAD = 1879047933 +pkg debug/elf, const DT_PLTPAD DynTag +pkg debug/elf, const DT_PLTPADSZ = 1879047673 +pkg debug/elf, const DT_PLTPADSZ DynTag +pkg debug/elf, const DT_POSFLAG_1 = 1879047677 +pkg debug/elf, const DT_POSFLAG_1 DynTag +pkg debug/elf, const DT_PPC64_GLINK = 1879048192 +pkg debug/elf, const DT_PPC64_GLINK DynTag +pkg debug/elf, const DT_PPC64_OPD = 1879048193 +pkg debug/elf, const DT_PPC64_OPD DynTag +pkg debug/elf, const DT_PPC64_OPDSZ = 1879048194 +pkg debug/elf, const DT_PPC64_OPDSZ DynTag +pkg debug/elf, const DT_PPC64_OPT = 1879048195 +pkg debug/elf, const DT_PPC64_OPT DynTag +pkg debug/elf, const DT_PPC_GOT = 1879048192 +pkg debug/elf, const DT_PPC_GOT DynTag +pkg debug/elf, const DT_PPC_OPT = 1879048193 +pkg debug/elf, const DT_PPC_OPT DynTag +pkg debug/elf, const DT_RELACOUNT = 1879048185 +pkg debug/elf, const DT_RELACOUNT DynTag +pkg debug/elf, const DT_RELCOUNT = 1879048186 +pkg debug/elf, const DT_RELCOUNT DynTag +pkg debug/elf, const DT_SPARC_REGISTER = 1879048193 +pkg debug/elf, const DT_SPARC_REGISTER DynTag +pkg debug/elf, const DT_SYMINENT = 1879047679 +pkg debug/elf, const DT_SYMINENT DynTag +pkg debug/elf, const DT_SYMINFO = 1879047935 +pkg debug/elf, const DT_SYMINFO DynTag +pkg debug/elf, const DT_SYMINSZ = 1879047678 +pkg debug/elf, const DT_SYMINSZ DynTag +pkg debug/elf, const DT_SYMTAB_SHNDX = 34 +pkg debug/elf, const DT_SYMTAB_SHNDX DynTag +pkg debug/elf, const DT_TLSDESC_GOT = 1879047927 +pkg debug/elf, const DT_TLSDESC_GOT DynTag +pkg debug/elf, const DT_TLSDESC_PLT = 1879047926 +pkg debug/elf, const DT_TLSDESC_PLT DynTag +pkg debug/elf, const DT_USED = 2147483646 +pkg debug/elf, const DT_USED DynTag +pkg debug/elf, const DT_VALRNGHI = 1879047679 +pkg debug/elf, const DT_VALRNGHI DynTag +pkg debug/elf, const DT_VALRNGLO = 1879047424 +pkg debug/elf, const DT_VALRNGLO DynTag +pkg debug/elf, const DT_VERDEF = 1879048188 +pkg debug/elf, const DT_VERDEF DynTag +pkg debug/elf, const DT_VERDEFNUM = 1879048189 +pkg debug/elf, const DT_VERDEFNUM DynTag +pkg debug/elf, const PT_AARCH64_ARCHEXT = 1879048192 +pkg debug/elf, const PT_AARCH64_ARCHEXT ProgType +pkg debug/elf, const PT_AARCH64_UNWIND = 1879048193 +pkg debug/elf, const PT_AARCH64_UNWIND ProgType +pkg debug/elf, const PT_ARM_ARCHEXT = 1879048192 +pkg debug/elf, const PT_ARM_ARCHEXT ProgType +pkg debug/elf, const PT_ARM_EXIDX = 1879048193 +pkg debug/elf, const PT_ARM_EXIDX ProgType +pkg debug/elf, const PT_GNU_EH_FRAME = 1685382480 +pkg debug/elf, const PT_GNU_EH_FRAME ProgType +pkg debug/elf, const PT_GNU_MBIND_HI = 1685386580 +pkg debug/elf, const PT_GNU_MBIND_HI ProgType +pkg debug/elf, const PT_GNU_MBIND_LO = 1685382485 +pkg debug/elf, const PT_GNU_MBIND_LO ProgType +pkg debug/elf, const PT_GNU_PROPERTY = 1685382483 +pkg debug/elf, const PT_GNU_PROPERTY ProgType +pkg debug/elf, const PT_GNU_RELRO = 1685382482 +pkg debug/elf, const PT_GNU_RELRO ProgType +pkg debug/elf, const PT_GNU_STACK = 1685382481 +pkg debug/elf, const PT_GNU_STACK ProgType +pkg debug/elf, const PT_MIPS_ABIFLAGS = 1879048195 +pkg debug/elf, const PT_MIPS_ABIFLAGS ProgType +pkg debug/elf, const PT_MIPS_OPTIONS = 1879048194 +pkg debug/elf, const PT_MIPS_OPTIONS ProgType +pkg debug/elf, const PT_MIPS_REGINFO = 1879048192 +pkg debug/elf, const PT_MIPS_REGINFO ProgType +pkg debug/elf, const PT_MIPS_RTPROC = 1879048193 +pkg debug/elf, const PT_MIPS_RTPROC ProgType +pkg debug/elf, const PT_OPENBSD_BOOTDATA = 1705253862 +pkg debug/elf, const PT_OPENBSD_BOOTDATA ProgType +pkg debug/elf, const PT_OPENBSD_RANDOMIZE = 1705237478 +pkg debug/elf, const PT_OPENBSD_RANDOMIZE ProgType +pkg debug/elf, const PT_OPENBSD_WXNEEDED = 1705237479 +pkg debug/elf, const PT_OPENBSD_WXNEEDED ProgType +pkg debug/elf, const PT_PAX_FLAGS = 1694766464 +pkg debug/elf, const PT_PAX_FLAGS ProgType +pkg debug/elf, const PT_S390_PGSTE = 1879048192 +pkg debug/elf, const PT_S390_PGSTE ProgType +pkg debug/elf, const PT_SUNWSTACK = 1879048187 +pkg debug/elf, const PT_SUNWSTACK ProgType +pkg debug/elf, const PT_SUNW_EH_FRAME = 1685382480 +pkg debug/elf, const PT_SUNW_EH_FRAME ProgType +pkg embed, method (FS) Open(string) (fs.File, error) +pkg embed, method (FS) ReadDir(string) ([]fs.DirEntry, error) +pkg embed, method (FS) ReadFile(string) ([]uint8, error) +pkg embed, type FS struct +pkg flag, func Func(string, string, func(string) error) +pkg flag, method (*FlagSet) Func(string, string, func(string) error) +pkg go/build, type Package struct, EmbedPatterns []string +pkg go/build, type Package struct, IgnoredOtherFiles []string +pkg go/build, type Package struct, TestEmbedPatterns []string +pkg go/build, type Package struct, XTestEmbedPatterns []string +pkg html/template, func ParseFS(fs.FS, ...string) (*Template, error) +pkg html/template, method (*Template) ParseFS(fs.FS, ...string) (*Template, error) +pkg io, func NopCloser(Reader) ReadCloser +pkg io, func ReadAll(Reader) ([]uint8, error) +pkg io, type ReadSeekCloser interface { Close, Read, Seek } +pkg io, type ReadSeekCloser interface, Close() error +pkg io, type ReadSeekCloser interface, Read([]uint8) (int, error) +pkg io, type ReadSeekCloser interface, Seek(int64, int) (int64, error) +pkg io, var Discard Writer +pkg io/fs, const ModeAppend = 1073741824 +pkg io/fs, const ModeAppend FileMode +pkg io/fs, const ModeCharDevice = 2097152 +pkg io/fs, const ModeCharDevice FileMode +pkg io/fs, const ModeDevice = 67108864 +pkg io/fs, const ModeDevice FileMode +pkg io/fs, const ModeDir = 2147483648 +pkg io/fs, const ModeDir FileMode +pkg io/fs, const ModeExclusive = 536870912 +pkg io/fs, const ModeExclusive FileMode +pkg io/fs, const ModeIrregular = 524288 +pkg io/fs, const ModeIrregular FileMode +pkg io/fs, const ModeNamedPipe = 33554432 +pkg io/fs, const ModeNamedPipe FileMode +pkg io/fs, const ModePerm = 511 +pkg io/fs, const ModePerm FileMode +pkg io/fs, const ModeSetgid = 4194304 +pkg io/fs, const ModeSetgid FileMode +pkg io/fs, const ModeSetuid = 8388608 +pkg io/fs, const ModeSetuid FileMode +pkg io/fs, const ModeSocket = 16777216 +pkg io/fs, const ModeSocket FileMode +pkg io/fs, const ModeSticky = 1048576 +pkg io/fs, const ModeSticky FileMode +pkg io/fs, const ModeSymlink = 134217728 +pkg io/fs, const ModeSymlink FileMode +pkg io/fs, const ModeTemporary = 268435456 +pkg io/fs, const ModeTemporary FileMode +pkg io/fs, const ModeType = 2401763328 +pkg io/fs, const ModeType FileMode +pkg io/fs, func Glob(FS, string) ([]string, error) +pkg io/fs, func ReadDir(FS, string) ([]DirEntry, error) +pkg io/fs, func ReadFile(FS, string) ([]uint8, error) +pkg io/fs, func Stat(FS, string) (FileInfo, error) +pkg io/fs, func ValidPath(string) bool +pkg io/fs, method (*PathError) Error() string +pkg io/fs, method (*PathError) Timeout() bool +pkg io/fs, method (*PathError) Unwrap() error +pkg io/fs, method (FileMode) IsDir() bool +pkg io/fs, method (FileMode) IsRegular() bool +pkg io/fs, method (FileMode) Perm() FileMode +pkg io/fs, method (FileMode) String() string +pkg io/fs, method (FileMode) Type() FileMode +pkg io/fs, type DirEntry interface { Info, IsDir, Name, Type } +pkg io/fs, type DirEntry interface, Info() (FileInfo, error) +pkg io/fs, type DirEntry interface, IsDir() bool +pkg io/fs, type DirEntry interface, Name() string +pkg io/fs, type DirEntry interface, Type() FileMode +pkg io/fs, type FS interface { Open } +pkg io/fs, type FS interface, Open(string) (File, error) +pkg io/fs, type File interface { Close, Read, Stat } +pkg io/fs, type File interface, Close() error +pkg io/fs, type File interface, Read([]uint8) (int, error) +pkg io/fs, type File interface, Stat() (FileInfo, error) +pkg io/fs, type FileInfo interface { IsDir, ModTime, Mode, Name, Size, Sys } +pkg io/fs, type FileInfo interface, IsDir() bool +pkg io/fs, type FileInfo interface, ModTime() time.Time +pkg io/fs, type FileInfo interface, Mode() FileMode +pkg io/fs, type FileInfo interface, Name() string +pkg io/fs, type FileInfo interface, Size() int64 +pkg io/fs, type FileInfo interface, Sys() interface{} +pkg io/fs, type FileMode uint32 +pkg io/fs, type GlobFS interface { Glob, Open } +pkg io/fs, type GlobFS interface, Glob(string) ([]string, error) +pkg io/fs, type GlobFS interface, Open(string) (File, error) +pkg io/fs, type PathError struct +pkg io/fs, type PathError struct, Err error +pkg io/fs, type PathError struct, Op string +pkg io/fs, type PathError struct, Path string +pkg io/fs, type ReadDirFS interface { Open, ReadDir } +pkg io/fs, type ReadDirFS interface, Open(string) (File, error) +pkg io/fs, type ReadDirFS interface, ReadDir(string) ([]DirEntry, error) +pkg io/fs, type ReadDirFile interface { Close, Read, ReadDir, Stat } +pkg io/fs, type ReadDirFile interface, Close() error +pkg io/fs, type ReadDirFile interface, Read([]uint8) (int, error) +pkg io/fs, type ReadDirFile interface, ReadDir(int) ([]DirEntry, error) +pkg io/fs, type ReadDirFile interface, Stat() (FileInfo, error) +pkg io/fs, type ReadFileFS interface { Open, ReadFile } +pkg io/fs, type ReadFileFS interface, Open(string) (File, error) +pkg io/fs, type ReadFileFS interface, ReadFile(string) ([]uint8, error) +pkg io/fs, type StatFS interface { Open, Stat } +pkg io/fs, type StatFS interface, Open(string) (File, error) +pkg io/fs, type StatFS interface, Stat(string) (FileInfo, error) +pkg io/fs, var ErrClosed error +pkg io/fs, var ErrExist error +pkg io/fs, var ErrInvalid error +pkg io/fs, var ErrNotExist error +pkg io/fs, var ErrPermission error +pkg log, func Default() *Logger +pkg net, var ErrClosed error +pkg net/http, func FS(fs.FS) FileSystem +pkg net/http, type Transport struct, GetProxyConnectHeader func(context.Context, *url.URL, string) (Header, error) +pkg os, const ModeAppend fs.FileMode +pkg os, const ModeCharDevice fs.FileMode +pkg os, const ModeDevice fs.FileMode +pkg os, const ModeDir fs.FileMode +pkg os, const ModeExclusive fs.FileMode +pkg os, const ModeIrregular fs.FileMode +pkg os, const ModeNamedPipe fs.FileMode +pkg os, const ModePerm fs.FileMode +pkg os, const ModeSetgid fs.FileMode +pkg os, const ModeSetuid fs.FileMode +pkg os, const ModeSocket fs.FileMode +pkg os, const ModeSticky fs.FileMode +pkg os, const ModeSymlink fs.FileMode +pkg os, const ModeTemporary fs.FileMode +pkg os, const ModeType fs.FileMode +pkg os, func Chmod(string, fs.FileMode) error +pkg os, func DirFS(string) fs.FS +pkg os, func Lstat(string) (fs.FileInfo, error) +pkg os, func Mkdir(string, fs.FileMode) error +pkg os, func MkdirAll(string, fs.FileMode) error +pkg os, func OpenFile(string, int, fs.FileMode) (*File, error) +pkg os, func SameFile(fs.FileInfo, fs.FileInfo) bool +pkg os, func Stat(string) (fs.FileInfo, error) +pkg os, method (*File) Chmod(fs.FileMode) error +pkg os, method (*File) ReadDir(int) ([]fs.DirEntry, error) +pkg os, method (*File) Readdir(int) ([]fs.FileInfo, error) +pkg os, method (*File) Stat() (fs.FileInfo, error) +pkg os, type DirEntry = fs.DirEntry +pkg os, type FileInfo = fs.FileInfo +pkg os, type FileMode = fs.FileMode +pkg os, type PathError = fs.PathError +pkg os/signal, func NotifyContext(context.Context, ...os.Signal) (context.Context, context.CancelFunc) +pkg runtime/metrics, const KindBad = 0 +pkg runtime/metrics, const KindBad ValueKind +pkg runtime/metrics, const KindFloat64 = 2 +pkg runtime/metrics, const KindFloat64 ValueKind +pkg runtime/metrics, const KindFloat64Histogram = 3 +pkg runtime/metrics, const KindFloat64Histogram ValueKind +pkg runtime/metrics, const KindUint64 = 1 +pkg runtime/metrics, const KindUint64 ValueKind +pkg runtime/metrics, func All() []Description +pkg runtime/metrics, func Read([]Sample) +pkg runtime/metrics, method (Value) Float64() float64 +pkg runtime/metrics, method (Value) Float64Histogram() *Float64Histogram +pkg runtime/metrics, method (Value) Kind() ValueKind +pkg runtime/metrics, method (Value) Uint64() uint64 +pkg runtime/metrics, type Description struct +pkg runtime/metrics, type Description struct, Cumulative bool +pkg runtime/metrics, type Description struct, Description string +pkg runtime/metrics, type Description struct, Kind ValueKind +pkg runtime/metrics, type Description struct, Name string +pkg runtime/metrics, type Description struct, StopTheWorld bool +pkg runtime/metrics, type Float64Histogram struct +pkg runtime/metrics, type Float64Histogram struct, Buckets []float64 +pkg runtime/metrics, type Float64Histogram struct, Counts []uint64 +pkg runtime/metrics, type Sample struct +pkg runtime/metrics, type Sample struct, Name string +pkg runtime/metrics, type Sample struct, Value Value +pkg runtime/metrics, type Value struct +pkg runtime/metrics, type ValueKind int +pkg syscall (linux-386), func AllThreadsSyscall(uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-386), func AllThreadsSyscall6(uintptr, uintptr, uintptr, uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-386), func Setegid(int) error +pkg syscall (linux-386), func Seteuid(int) error +pkg syscall (linux-386-cgo), func AllThreadsSyscall(uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-386-cgo), func AllThreadsSyscall6(uintptr, uintptr, uintptr, uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-386-cgo), func Setegid(int) error +pkg syscall (linux-386-cgo), func Seteuid(int) error +pkg syscall (linux-amd64), func AllThreadsSyscall(uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-amd64), func AllThreadsSyscall6(uintptr, uintptr, uintptr, uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-amd64), func Setegid(int) error +pkg syscall (linux-amd64), func Seteuid(int) error +pkg syscall (linux-amd64-cgo), func AllThreadsSyscall(uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-amd64-cgo), func AllThreadsSyscall6(uintptr, uintptr, uintptr, uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-amd64-cgo), func Setegid(int) error +pkg syscall (linux-amd64-cgo), func Seteuid(int) error +pkg syscall (linux-arm), func AllThreadsSyscall(uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-arm), func AllThreadsSyscall6(uintptr, uintptr, uintptr, uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-arm), func Setegid(int) error +pkg syscall (linux-arm), func Seteuid(int) error +pkg syscall (linux-arm-cgo), func AllThreadsSyscall(uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-arm-cgo), func AllThreadsSyscall6(uintptr, uintptr, uintptr, uintptr, uintptr, uintptr, uintptr) (uintptr, uintptr, Errno) +pkg syscall (linux-arm-cgo), func Setegid(int) error +pkg syscall (linux-arm-cgo), func Seteuid(int) error +pkg syscall (windows-386), func RtlGenRandom(*uint8, uint32) error +pkg syscall (windows-amd64), func RtlGenRandom(*uint8, uint32) error pkg testing, func Fuzz(func(*F)) FuzzResult pkg testing, func MainStart(testDeps, []InternalTest, []InternalBenchmark, []InternalFuzzTarget, []InternalExample) *M pkg testing, func RunFuzzTargets(func(string, string) (bool, error), []InternalFuzzTarget) bool @@ -49,4 +447,39 @@ pkg testing, type FuzzResult struct, N int pkg testing, type FuzzResult struct, T time.Duration pkg testing, type InternalFuzzTarget struct pkg testing, type InternalFuzzTarget struct, Fn func(*F) -pkg testing, type InternalFuzzTarget struct, Name string \ No newline at end of file +pkg testing, type InternalFuzzTarget struct, Name string +pkg testing/fstest, func TestFS(fs.FS, ...string) error +pkg testing/fstest, method (MapFS) Glob(string) ([]string, error) +pkg testing/fstest, method (MapFS) Open(string) (fs.File, error) +pkg testing/fstest, method (MapFS) ReadDir(string) ([]fs.DirEntry, error) +pkg testing/fstest, method (MapFS) ReadFile(string) ([]uint8, error) +pkg testing/fstest, method (MapFS) Stat(string) (fs.FileInfo, error) +pkg testing/fstest, type MapFS map[string]*MapFile +pkg testing/fstest, type MapFile struct +pkg testing/fstest, type MapFile struct, Data []uint8 +pkg testing/fstest, type MapFile struct, ModTime time.Time +pkg testing/fstest, type MapFile struct, Mode fs.FileMode +pkg testing/fstest, type MapFile struct, Sys interface{} +pkg testing/iotest, func ErrReader(error) io.Reader +pkg testing/iotest, func TestReader(io.Reader, []uint8) error +pkg text/template, func ParseFS(fs.FS, ...string) (*Template, error) +pkg text/template, method (*Template) ParseFS(fs.FS, ...string) (*Template, error) +pkg text/template/parse, const NodeComment = 20 +pkg text/template/parse, const NodeComment NodeType +pkg text/template/parse, const ParseComments = 1 +pkg text/template/parse, const ParseComments Mode +pkg text/template/parse, method (*CommentNode) Copy() Node +pkg text/template/parse, method (*CommentNode) String() string +pkg text/template/parse, method (CommentNode) Position() Pos +pkg text/template/parse, method (CommentNode) Type() NodeType +pkg text/template/parse, type CommentNode struct +pkg text/template/parse, type CommentNode struct, Text string +pkg text/template/parse, type CommentNode struct, embedded NodeType +pkg text/template/parse, type CommentNode struct, embedded Pos +pkg text/template/parse, type Mode uint +pkg text/template/parse, type Tree struct, Mode Mode +pkg unicode, const Version = "13.0.0" +pkg unicode, var Chorasmian *RangeTable +pkg unicode, var Dives_Akuru *RangeTable +pkg unicode, var Khitan_Small_Script *RangeTable +pkg unicode, var Yezidi *RangeTable diff --git a/doc/articles/race_detector.html b/doc/articles/race_detector.html index 63a658f870b..09188c15d51 100644 --- a/doc/articles/race_detector.html +++ b/doc/articles/race_detector.html @@ -418,7 +418,7 @@ close(c) linux/amd64, linux/ppc64le, linux/arm64, freebsd/amd64, netbsd/amd64, darwin/amd64, - and windows/amd64. + darwin/arm64, and windows/amd64.

Runtime Overhead

diff --git a/doc/contribute.html b/doc/contribute.html index 03d02c9d75a..f297477fe03 100644 --- a/doc/contribute.html +++ b/doc/contribute.html @@ -45,8 +45,8 @@ CLA (Contributor License Agreement).
  • Step 2: Configure authentication credentials for the Go Git repository. -Visit go.googlesource.com, click -on the gear icon (top right), then on "Obtain password", and follow the +Visit go.googlesource.com, click +"Generate Password" in the page's top right menu bar, and follow the instructions.
  • @@ -806,10 +806,9 @@ tracker will automatically mark the issue as fixed.

    If the change is a partial step towards the resolution of the issue, -uses the notation "Updates #12345". -This will leave a comment in the issue -linking back to the change in Gerrit, but it will not close the issue -when the change is applied. +write "Updates #12345" instead. +This will leave a comment in the issue linking back to the change in +Gerrit, but it will not close the issue when the change is applied.

    @@ -948,10 +947,18 @@ The Gerrit voting system involves an integer in the range -2 to +2:

  • +

    +At least two maintainers must approve of the change, and at least one +of those maintainers must +2 the change. +The second maintainer may cast a vote of Trust+1, meaning that the +change looks basically OK, but that the maintainer hasn't done the +detailed review required for a +2 vote. +

    +

    Submitting an approved change

    -After the code has been +2'ed, an approver will +After the code has been +2'ed and Trust+1'ed, an approver will apply it to the master branch using the Gerrit user interface. This is called "submitting the change".

    diff --git a/doc/diagnostics.html b/doc/diagnostics.html index 478611c15c0..438cdce45fa 100644 --- a/doc/diagnostics.html +++ b/doc/diagnostics.html @@ -454,6 +454,8 @@ environmental variable is set accordingly.

  • GODEBUG=gctrace=1 prints garbage collector events at each collection, summarizing the amount of memory collected and the length of the pause.
  • +
  • GODEBUG=inittrace=1 prints a summary of execution time and memory allocation +information for completed package initialization work.
  • GODEBUG=schedtrace=X prints scheduling events every X milliseconds.
  • diff --git a/doc/go1.16.html b/doc/go1.16.html index 0ffaecc5a9a..2ff763f9b62 100644 --- a/doc/go1.16.html +++ b/doc/go1.16.html @@ -26,61 +26,395 @@ Do not send CLs removing the interior tags from such phrases.

    Changes to the language

    - TODO + There are no changes to the language.

    Ports

    -

    - TODO +

    Darwin and iOS

    + +

    + Go 1.16 adds support of 64-bit ARM architecture on macOS (also known as + Apple Silicon) with GOOS=darwin, GOARCH=arm64. + Like the darwin/amd64 port, the darwin/arm64 + port supports cgo, internal and external linking, c-archive, + c-shared, and pie build modes, and the race + detector. +

    + +

    + The iOS port, which was previously darwin/arm64, has + been renamed to ios/arm64. GOOS=ios + implies the + darwin build tag, just as GOOS=android + implies the linux build tag. This change should be + transparent to anyone using gomobile to build iOS apps. +

    + +

    + Go 1.16 adds an ios/amd64 port, which targets the iOS + simulator running on AMD64-based macOS. Previously this was + unofficially supported through darwin/amd64 with + the ios build tag set. +

    + +

    + Go 1.16 is the last release that will run on macOS 10.12 Sierra. + Go 1.17 will require macOS 10.13 High Sierra or later. +

    + +

    NetBSD

    + +

    + Go now supports the 64-bit ARM architecture on NetBSD (the + netbsd/arm64 port). +

    + +

    OpenBSD

    + +

    + Go now supports the MIPS64 architecture on OpenBSD + (the openbsd/mips64 port). This port does not yet + support cgo. +

    + +

    386

    + +

    + As announced in the Go 1.15 release notes, + Go 1.16 drops support for x87 mode compilation (GO386=387). + Support for non-SSE2 processors is now available using soft float + mode (GO386=softfloat). + Users running on non-SSE2 processors should replace GO386=387 + with GO386=softfloat. +

    + +

    RISC-V

    + +

    + The linux/riscv64 port now supports cgo and + -buildmode=pie. This release also includes performance + optimizations and code generation improvements for RISC-V.

    Tools

    -

    - TODO +

    Go command

    + +

    Modules

    + +

    + Module-aware mode is enabled by default, regardless of whether a + go.mod file is present in the current working directory or a + parent directory. More precisely, the GO111MODULE environment + variable now defaults to on. To switch to the previous behavior, + set GO111MODULE to auto.

    -

    Go command

    +

    + Build commands like go build and go + test no longer modify go.mod and go.sum + by default. Instead, they report an error if a module requirement or checksum + needs to be added or updated (as if the -mod=readonly flag were + used). Module requirements and sums may be adjusted with go + mod tidy or go get. +

    + +

    + go install now accepts arguments with + version suffixes (for example, go install + example.com/cmd@v1.0.0). This causes go + install to build and install packages in module-aware mode, + ignoring the go.mod file in the current directory or any parent + directory, if there is one. This is useful for installing executables without + affecting the dependencies of the main module. +

    + +

    + go install, with or without a version suffix (as + described above), is now the recommended way to build and install packages in + module mode. go get should be used with the + -d flag to adjust the current module's dependencies without + building packages, and use of go get to build and + install packages is deprecated. In a future release, the -d flag + will always be enabled. +

    retract directives may now be used in a go.mod file to indicate that certain published versions of the module should not be used by other modules. A module author may retract a version after a severe problem is discovered or if the version was published unintentionally.
    - TODO: write and link to section in golang.org/ref/mod
    - TODO: write and link to tutorial or blog post

    -

    - When using go test, a test that +

    + The go mod vendor + and go mod tidy subcommands now accept + the -e flag, which instructs them to proceed despite errors in + resolving missing packages. +

    + +

    + The go command now ignores requirements on module versions + excluded by exclude directives in the main module. Previously, + the go command used the next version higher than an excluded + version, but that version could change over time, resulting in + non-reproducible builds. +

    + +

    Embedding Files

    + +

    + The go command now supports including + static files and file trees as part of the final executable, + using the new //go:embed directive. + See the documentation for the new + embed + package for details. +

    + +

    go test

    + +

    + When using go test, a test that calls os.Exit(0) during execution of a test function will now be considered to fail. This will help catch cases in which a test calls code that calls - os.Exit(0) and thereby stops running all future tests. + os.Exit(0) and thereby stops running all future tests. If a TestMain function calls os.Exit(0) that is still considered to be a passing test.

    +

    + go test reports an error when the -c + or -i flags are used together with unknown flags. Normally, + unknown flags are passed to tests, but when -c or -i + are used, tests are not run. +

    + +

    go get

    + +

    + The go get -insecure flag is + deprecated and will be removed in a future version. This flag permits + fetching from repositories and resolving custom domains using insecure + schemes such as HTTP, and also bypasses module sum validation using the + checksum database. To permit the use of insecure schemes, use the + GOINSECURE environment variable instead. To bypass module + sum validation, use GOPRIVATE or GONOSUMDB. + See go help environment for details. +

    + +

    + go get example.com/mod@patch now + requires that some version of example.com/mod already be + required by the main module. + (However, go get -u=patch continues + to patch even newly-added dependencies.) +

    + +

    GOVCS environment variable

    + +

    + GOVCS is a new environment variable that limits which version + control tools the go command may use to download source code. + This mitigates security issues with tools that are typically used in trusted, + authenticated environments. By default, git and hg + may be used to download code from any repository. svn, + bzr, and fossil may only be used to download code + from repositories with module paths or package paths matching patterns in + the GOPRIVATE environment variable. See + go + help vcs for details. +

    + +

    The all pattern

    + +

    + When the main module's go.mod file + declares go 1.16 or higher, the all + package pattern now matches only those packages that are transitively imported + by a package or test found in the main module. (Packages imported by tests + of packages imported by the main module are no longer included.) This is + the same set of packages retained + by go mod vendor since Go 1.11. +

    + +

    The -toolexec build flag

    + +

    + When the -toolexec build flag is specified to use a program when + invoking toolchain programs like compile or asm, the environment variable + TOOLEXEC_IMPORTPATH is now set to the import path of the package + being built. +

    + +

    The -i build flag

    + +

    + The -i flag accepted by go build, + go install, and go test is + now deprecated. The -i flag instructs the go command + to install packages imported by packages named on the command line. Since + the build cache was introduced in Go 1.10, the -i flag no longer + has a significant effect on build times, and it causes errors when the install + directory is not writable. +

    + +

    The list command

    + +

    + When the -export flag is specified, the BuildID + field is now set to the build ID of the compiled package. This is equivalent + to running go tool buildid on + go list -exported -f {{.Export}, + but without the extra step. +

    + +

    The -overlay flag

    + +

    + The -overlay flag specifies a JSON configuration file containing + a set of file path replacements. The -overlay flag may be used + with all build commands and go mod subcommands. + It is primarily intended to be used by editor tooling such as gopls to + understand the effects of unsaved changes to source files. The config file + maps actual file paths to replacement file paths and the go + command and its builds will run as if the actual file paths exist with the + contents given by the replacement file paths, or don't exist if the replacement + file paths are empty. +

    + +

    Cgo

    + +

    + The cgo tool will no longer try to translate + C struct bitfields into Go struct fields, even if their size can be + represented in Go. The order in which C bitfields appear in memory + is implementation dependent, so in some cases the cgo tool produced + results that were silently incorrect. +

    + +

    Vet

    + +

    New warning for invalid testing.T use in +goroutines

    + +

    + The vet tool now warns about invalid calls to the testing.T + method Fatal from within a goroutine created during the test. + This also warns on calls to Fatalf, FailNow, and + Skip{,f,Now} methods on testing.T tests or + testing.B benchmarks. +

    +

    - TODO + Calls to these methods stop the execution of the created goroutine and not + the Test* or Benchmark* function. So these are + required to be called by the goroutine + running the test or benchmark function. For example: +

    + +
    +func TestFoo(t *testing.T) {
    +    go func() {
    +        if condition() {
    +            t.Fatal("oops") // This exits the inner func instead of TestFoo.
    +        }
    +        ...
    +    }()
    +}
    +
    + +

    + Code calling t.Fatal (or a similar method) from a created + goroutine should be rewritten to signal the test failure using + t.Error and exit the goroutine early using an alternative + method, such as using a return statement. The previous example + could be rewritten as: +

    + +
    +func TestFoo(t *testing.T) {
    +    go func() {
    +        if condition() {
    +            t.Error("oops")
    +            return
    +        }
    +        ...
    +    }()
    +}
    +
    + +

    + The vet tool now warns about amd64 assembly that clobbers the BP + register (the frame pointer) without saving and restoring it, + contrary to the calling convention. Code that doesn't preserve the + BP register must be modified to either not use BP at all or preserve + BP by saving and restoring it. An easy way to preserve BP is to set + the frame size to a nonzero value, which causes the generated + prologue and epilogue to preserve the BP register for you. + See CL 248260 for example + fixes.

    Runtime

    - TODO + The new runtime/metrics package + introduces a stable interface for reading + implementation-defined metrics from the Go runtime. + It supersedes existing functions like + runtime.ReadMemStats + and + debug.GCStats + and is significantly more general and efficient. + See the package documentation for more details. +

    + +

    + Setting the GODEBUG environment variable + to inittrace=1 now causes the runtime to emit a single + line to standard error for each package init, + summarizing its execution time and memory allocation. This trace can + be used to find bottlenecks or regressions in Go startup + performance. + The GODEBUG< + documentation describes the format. +

    + +

    + On Linux, the runtime now defaults to releasing memory to the + operating system promptly (using MADV_DONTNEED), rather + than lazily when the operating system is under memory pressure + (using MADV_FREE). This means process-level memory + statistics like RSS will more accurately reflect the amount of + physical memory being used by Go processes. Systems that are + currently using GODEBUG=madvdontneed=1 to improve + memory monitoring behavior no longer need to set this environment + variable. +

    + +

    + Go 1.16 fixes a discrepancy between the race detector and + the Go memory model. The race detector now + more precisely follows the channel synchronization rules of the + memory model. As a result, the detector may now report races it + previously missed.

    Compiler

    -

    - TODO +

    + The compiler can now inline functions with + non-labeled for loops, method values, and type + switches. The inliner can also detect more indirect calls where + inlining is possible.

    Linker

    -

    +

    This release includes additional improvements to the Go linker, reducing linker resource usage (both time and memory) and improving code robustness/maintainability. These changes form the second half @@ -94,54 +428,74 @@ Do not send CLs removing the interior tags from such phrases. supported architecture/OS combinations (the 1.15 performance improvements were primarily focused on ELF-based OSes and amd64 architectures). For a representative set of - large Go programs, linking is 20-35% faster than 1.15 and requires + large Go programs, linking is 20-25% faster than 1.15 and requires 5-15% less memory on average for linux/amd64, with larger - improvements for other architectures and OSes. + improvements for other architectures and OSes. Most binaries are + also smaller as a result of more aggressive symbol pruning.

    -

    - TODO: update with final numbers later in the release. +

    + On Windows, go build -buildmode=c-shared now generates Windows + ASLR DLLs by default. ASLR can be disabled with --ldflags=-aslr=false.

    Core library

    +

    Embedded Files

    +

    - TODO + The new embed package + provides access to files embedded in the program during compilation + using the new //go:embed directive.

    -

    net

    +

    File Systems

    -

    - The case of I/O on a closed network connection, or I/O on a network - connection that is closed before any of the I/O completes, can now - be detected using the new ErrClosed error. - A typical use would be errors.Is(err, net.ErrClosed). - In earlier releases the only way to reliably detect this case was to - match the string returned by the Error method - with "use of closed network connection". +

    + The new io/fs package + defines an abstraction for read-only trees of files, + the fs.FS interface, + and the standard library packages have + been adapted to make use of the interface as appropriate.

    - -

    text/template/parse

    - -

    - A new CommentNode - was added to the parse tree. The Mode - field in the parse.Tree enables access to it. +

    + On the producer side of the interface, + the new embed.FS type + implements fs.FS, as does + zip.Reader. + The new os.DirFS function + provides an implementation of fs.FS backed by a tree + of operating system files.

    - -

    unicode

    - -

    - The unicode package and associated - support throughout the system has been upgraded from Unicode 12.0.0 to - Unicode 13.0.0, - which adds 5,930 new characters, including four new scripts, and 55 new emoji. - Unicode 13.0.0 also designates plane 3 (U+30000-U+3FFFF) as the tertiary - ideographic plane. +

    + On the consumer side, + the new http.FS + function converts an fs.FS to an + http.Handler. + Also, the html/template + and text/template + packages’ ParseFS + functions and methods read templates from an fs.FS.

    +

    + For testing code that implements fs.FS, + the new testing/fstest + package provides a TestFS + function that checks for and reports common mistakes. + It also provides a simple in-memory file system implementation, + MapFS, + which can be useful for testing code that accepts fs.FS + implementations. +

    + + +

    Minor changes to the library

    @@ -150,9 +504,244 @@ Do not send CLs removing the interior tags from such phrases. in mind.

    -

    - TODO -

    +
    crypto/dsa
    +
    +

    + The crypto/dsa package is now deprecated. + See issue #40337. +

    +
    +
    + +
    crypto/hmac
    +
    +

    + New will now panic if + separate calls to the hash generation function fail to return new values. + Previously, the behavior was undefined and invalid outputs were sometimes + generated. +

    +
    +
    + +
    crypto/tls
    +
    +

    + I/O operations on closing or closed TLS connections can now be detected + using the new net.ErrClosed + error. A typical use would be errors.Is(err, net.ErrClosed). +

    + +

    + A default write deadline is now set in + Conn.Close + before sending the "close notify" alert, in order to prevent blocking + indefinitely. +

    + +

    + The new Conn.HandshakeContext + method allows cancellation of an in-progress handshake. The provided + context is accessible through the new + ClientHelloInfo.Context + and + CertificateRequestInfo.Context methods. Canceling the + context after the handshake has finished has no effect. +

    + +

    + Clients now return a handshake error if the server selects + + an ALPN protocol that was not in + + the list advertised by the client. +

    + +

    + Servers will now prefer other available AEAD cipher suites (such as ChaCha20Poly1305) + over AES-GCM cipher suites if either the client or server doesn't have AES hardware + support, unless both + Config.PreferServerCipherSuites + and Config.CipherSuites + are set. The client is assumed not to have AES hardware support if it does + not signal a preference for AES-GCM cipher suites. +

    + +

    + Config.Clone now + returns nil if the receiver is nil, rather than panicking. +

    +
    +
    + +
    crypto/x509
    +
    +

    + The GODEBUG=x509ignoreCN=0 flag will be removed in Go 1.17. + It enables the legacy behavior of treating the CommonName + field on X.509 certificates as a host name when no Subject Alternative + Names are present. +

    + +

    + ParseCertificate and + CreateCertificate + now enforce string encoding restrictions for the DNSNames, + EmailAddresses, and URIs fields. These fields + can only contain strings with characters within the ASCII range. +

    + +

    + CreateCertificate + now verifies the generated certificate's signature using the signer's + public key. If the signature is invalid, an error is returned, instead of + a malformed certificate. +

    + +

    + A number of additional fields have been added to the + CertificateRequest type. + These fields are now parsed in + ParseCertificateRequest and marshalled in + CreateCertificateRequest. +

    + +

    + DSA signature verification is no longer supported. Note that DSA signature + generation was never supported. + See issue #40337. +

    + +

    + On Windows, Certificate.Verify + will now return all certificate chains that are built by the platform + certificate verifier, instead of just the highest ranked chain. +

    + +

    + The new SystemRootsError.Unwrap + method allows accessing the Err + field through the errors package functions. +

    +
    +
    + +
    encoding/asn1
    +
    +

    + Unmarshal and + UnmarshalWithParams + now return an error instead of panicking when the argument is not + a pointer or is nil. This change matches the behavior of other + encoding packages such as encoding/json. +

    +
    +
    + +
    encoding/json
    +
    +

    + The json struct field tags understood by + Marshal, + Unmarshal, + and related functionality now permit semicolon characters within + a JSON object name for a Go struct field. +

    +
    +
    + +
    encoding/xml
    +
    +

    + The encoder has always taken care to avoid using namespace prefixes + beginning with xml, which are reserved by the XML + specification. + Now, following the specification more closely, that check is + case-insensitive, so that prefixes beginning + with XML, XmL, and so on are also + avoided. +

    +
    +
    + +
    flag
    +
    +

    + The new Func function + allows registering a flag implemented by calling a function, + as a lighter-weight alternative to implementing the + Value interface. +

    +
    +
    + +
    io
    +
    +

    + The package now defines a + ReadSeekCloser interface. +

    +
    +
    + +
    log
    +
    +

    + The new Default function + provides access to the default Logger. +

    +
    +
    + +
    log/syslog
    +
    +

    + The Writer + now uses the local message format + (omitting the host name and using a shorter time stamp) + when logging to custom Unix domain sockets, + matching the format already used for the default log socket. +

    +
    +
    + +
    mime/multipart
    +
    +

    + The Reader's + ReadForm + method no longer rejects form data + when passed the maximum int64 value as a limit. +

    +
    +
    + +
    net
    +
    +

    + The case of I/O on a closed network connection, or I/O on a network + connection that is closed before any of the I/O completes, can now + be detected using the new ErrClosed error. + A typical use would be errors.Is(err, net.ErrClosed). + In earlier releases the only way to reliably detect this case was to + match the string returned by the Error method + with "use of closed network connection". +

    + +

    + In previous Go releases the default TCP listener backlog size on Linux systems, + set by /proc/sys/net/core/somaxconn, was limited to a maximum of 65535. + On Linux kernel version 4.1 and above, the maximum is now 4294967295. +

    + +

    + On Linux, host name lookups no longer use DNS before checking + /etc/hosts when /etc/nsswitch.conf + is missing; this is common on musl-based systems and makes + Go programs match the behavior of C programs on those systems. +

    +
    +
    net/http
    @@ -171,9 +760,208 @@ Do not send CLs removing the interior tags from such phrases.

    - The net/http package now rejects HTTP range requests - of the form "Range": "bytes=--N" where "-N" is a negative suffix length, for - example "Range": "bytes=--2". It now replies with a 416 "Range Not Satisfiable" response. + The net/http package now rejects HTTP range requests + of the form "Range": "bytes=--N" where "-N" is a negative suffix length, for + example "Range": "bytes=--2". It now replies with a 416 "Range Not Satisfiable" response. +

    + +

    + Cookies set with SameSiteDefaultMode + now behave according to the current spec (no attribute is set) instead of + generating a SameSite key without a value. +

    + +

    + The net/http package now passes the + Request context to + tls.Conn.HandshakeContext + when performing TLS handshakes. +

    + +

    + The Client now sends + an explicit Content-Length: 0 + header in PATCH requests with empty bodies, + matching the existing behavior of POST and PUT. +

    + +

    + The ProxyFromEnvironment function + no longer returns the setting of the HTTP_PROXY environment + variable for https:// URLs when HTTPS_PROXY is unset.

    + +
    net/http/httputil
    +
    +

    + The ReverseProxy + now flushes buffered data more aggressively when proxying + streamed responses with unknown body lengths. +

    +
    +
    + +
    net/smtp
    +
    +

    + The Client's + Mail + method now sends the SMTPUTF8 directive to + servers that support it, signaling that addresses are encoded in UTF-8. +

    +
    +
    + +
    os
    +
    +

    + Process.Signal now + returns ErrProcessDone + instead of the unexported errFinished when the process has + already finished. +

    +
    +
    + +
    os/signal
    +
    +

    + The new + NotifyContext + function allows creating contexts that are canceled upon arrival of + specific signals. +

    +
    +
    + +
    path
    +
    +

    + The Match and Glob functions now + return an error if the unmatched part of the pattern has a + syntax error. Previously, the functions returned early on a failed + match, and thus did not report any later syntax error in the + pattern. +

    +
    +
    + +
    path/filepath
    +
    +

    + The Match and Glob functions now + return an error if the unmatched part of the pattern has a + syntax error. Previously, the functions returned early on a failed + match, and thus did not report any later syntax error in the + pattern. +

    +
    +
    + +
    reflect
    +
    +

    + StructTag now allows multiple space-separated keys + in key:value pairs, as in `json xml:"field1"` + (equivalent to `json:"field1" xml:"field1"`). +

    +
    +
    + +
    runtime/debug
    +
    +

    + The runtime.Error values + used when SetPanicOnFault is enabled may now have an + Addr method. If that method exists, it returns the memory + address that triggered the fault. +

    +
    +
    + +
    strconv
    +
    +

    + ParseFloat now uses + the Eisel-Lemire + algorithm, improving performance by up to a factor of 2. This can + also speed up decoding textual formats like encoding/json. +

    +
    +
    + +
    syscall
    +
    +

    + NewCallback + and + NewCallbackCDecl + now correctly support callback functions with multiple + sub-uintptr-sized arguments in a row. This may + require changing uses of these functions to eliminate manual + padding between small arguments. +

    + +

    + SysProcAttr on Windows has a new NoInheritHandles field that disables inheriting handles when creating a new process. +

    + +

    + DLLError on Windows now has an Unwrap function for unwrapping its underlying error. +

    + +

    + On Linux, + Setgid, + Setuid, + and related calls are now implemented. + Previously, they returned an syscall.EOPNOTSUPP error. +

    +
    +
    + +
    text/template
    +
    +

    + Newlines characters are now allowed inside action delimiters, + permitting actions to span multiple lines. +

    +
    +
    + +
    text/template/parse
    +
    +

    + A new CommentNode + was added to the parse tree. The Mode + field in the parse.Tree enables access to it. +

    +
    +
    + +
    time/tzdata
    +
    +

    + The slim timezone data format is now used for the timezone database in + $GOROOT/lib/time/zoneinfo.zip and the embedded copy in this + package. This reduces the size of the timezone database by about 350 KB. +

    +
    +
    + +
    unicode
    +
    +

    + The unicode package and associated + support throughout the system has been upgraded from Unicode 12.0.0 to + Unicode 13.0.0, + which adds 5,930 new characters, including four new scripts, and 55 new emoji. + Unicode 13.0.0 also designates plane 3 (U+30000-U+3FFFF) as the tertiary + ideographic plane. +

    +
    +
    diff --git a/doc/go1.html b/doc/go1.html index 34e305b93cd..939ee24df53 100644 --- a/doc/go1.html +++ b/doc/go1.html @@ -1647,14 +1647,14 @@ c := signal.Incoming() is

    -c := make(chan os.Signal)
    +c := make(chan os.Signal, 1)
     signal.Notify(c) // ask for all signals
     

    but most code should list the specific signals it wants to handle instead:

    -c := make(chan os.Signal)
    +c := make(chan os.Signal, 1)
     signal.Notify(c, syscall.SIGHUP, syscall.SIGQUIT)
     
    diff --git a/doc/go_spec.html b/doc/go_spec.html index 154bdbfeaf9..676407f6f2b 100644 --- a/doc/go_spec.html +++ b/doc/go_spec.html @@ -1,6 +1,6 @@ @@ -3594,23 +3594,33 @@ replaced by its left operand alone.

    +var a [1024]byte
     var s uint = 33
    -var i = 1<<s                  // 1 has type int
    -var j int32 = 1<<s            // 1 has type int32; j == 0
    -var k = uint64(1<<s)          // 1 has type uint64; k == 1<<33
    -var m int = 1.0<<s            // 1.0 has type int; m == 0 if ints are 32bits in size
    -var n = 1.0<<s == j           // 1.0 has type int32; n == true
    -var o = 1<<s == 2<<s          // 1 and 2 have type int; o == true if ints are 32bits in size
    -var p = 1<<s == 1<<33         // illegal if ints are 32bits in size: 1 has type int, but 1<<33 overflows int
    -var u = 1.0<<s                // illegal: 1.0 has type float64, cannot shift
    -var u1 = 1.0<<s != 0          // illegal: 1.0 has type float64, cannot shift
    -var u2 = 1<<s != 1.0          // illegal: 1 has type float64, cannot shift
    -var v float32 = 1<<s          // illegal: 1 has type float32, cannot shift
    -var w int64 = 1.0<<33         // 1.0<<33 is a constant shift expression
    -var x = a[1.0<<s]             // 1.0 has type int; x == a[0] if ints are 32bits in size
    -var a = make([]byte, 1.0<<s)  // 1.0 has type int; len(a) == 0 if ints are 32bits in size
    -
    +// The results of the following examples are given for 64-bit ints. +var i = 1<<s // 1 has type int +var j int32 = 1<<s // 1 has type int32; j == 0 +var k = uint64(1<<s) // 1 has type uint64; k == 1<<33 +var m int = 1.0<<s // 1.0 has type int; m == 1<<33 +var n = 1.0<<s == j // 1.0 has type int; n == true +var o = 1<<s == 2<<s // 1 and 2 have type int; o == false +var p = 1<<s == 1<<33 // 1 has type int; p == true +var u = 1.0<<s // illegal: 1.0 has type float64, cannot shift +var u1 = 1.0<<s != 0 // illegal: 1.0 has type float64, cannot shift +var u2 = 1<<s != 1.0 // illegal: 1 has type float64, cannot shift +var v float32 = 1<<s // illegal: 1 has type float32, cannot shift +var w int64 = 1.0<<33 // 1.0<<33 is a constant shift expression; w == 1<<33 +var x = a[1.0<<s] // panics: 1.0 has type int, but 1<<33 overflows array bounds +var b = make([]byte, 1.0<<s) // 1.0 has type int; len(b) == 1<<33 + +// The results of the following examples are given for 32-bit ints, +// which means the shifts will overflow. +var mm int = 1.0<<s // 1.0 has type int; mm == 0 +var oo = 1<<s == 2<<s // 1 and 2 have type int; oo == true +var pp = 1<<s == 1<<33 // illegal: 1 has type int, but 1<<33 overflows int +var xx = a[1.0<<s] // 1.0 has type int; xx == a[0] +var bb = make([]byte, 1.0<<s) // 1.0 has type int; len(bb) == 0 +

    Operator precedence

    @@ -3646,7 +3656,7 @@ For instance, x / y * z is the same as (x / y) * z. x <= f() ^a >> b f() || g() -x == y+1 && <-chanPtr > 0 +x == y+1 && <-chanInt > 0 diff --git a/doc/install-source.html b/doc/install-source.html index cbf4eac70b0..f0a909263cf 100644 --- a/doc/install-source.html +++ b/doc/install-source.html @@ -119,11 +119,26 @@ The Go toolchain is written in Go. To build it, you need a Go compiler installed The scripts that do the initial build of the tools look for a "go" command in $PATH, so as long as you have Go installed in your system and configured in your $PATH, you are ready to build Go -from source. +from source. Or if you prefer you can set $GOROOT_BOOTSTRAP to the root of a Go installation to use to build the new Go toolchain; $GOROOT_BOOTSTRAP/bin/go should be the go command to use.

    +

    +There are four possible ways to obtain a bootstrap toolchain: +

    + + + +

    +These approaches are detailed below. +

    +

    Bootstrap toolchain from binary release

    @@ -132,30 +147,6 @@ To use a binary release as a bootstrap toolchain, see packaged Go distribution.

    -

    Bootstrap toolchain from source

    - -

    -To build a bootstrap toolchain from source, use -either the git branch release-branch.go1.4 or -go1.4-bootstrap-20171003.tar.gz, -which contains the Go 1.4 source code plus accumulated fixes -to keep the tools running on newer operating systems. -(Go 1.4 was the last distribution in which the toolchain was written in C.) -After unpacking the Go 1.4 source, cd to -the src subdirectory, set CGO_ENABLED=0 in -the environment, and run make.bash (or, -on Windows, make.bat). -

    - -

    -Once the Go 1.4 source has been unpacked into your GOROOT_BOOTSTRAP directory, -you must keep this git clone instance checked out to branch -release-branch.go1.4. Specifically, do not attempt to reuse -this git clone in the later step named "Fetch the repository." The go1.4 -bootstrap toolchain must be able to properly traverse the go1.4 sources -that it assumes are present under this repository root. -

    -

    Bootstrap toolchain from cross-compiled source

    @@ -194,6 +185,36 @@ $ sudo update-alternatives --set go /usr/bin/go-5 $ GOROOT_BOOTSTRAP=/usr ./make.bash +

    Bootstrap toolchain from C source code

    + +

    +To build a bootstrap toolchain from C source code, use +either the git branch release-branch.go1.4 or +go1.4-bootstrap-20171003.tar.gz, +which contains the Go 1.4 source code plus accumulated fixes +to keep the tools running on newer operating systems. +(Go 1.4 was the last distribution in which the toolchain was written in C.) +After unpacking the Go 1.4 source, cd to +the src subdirectory, set CGO_ENABLED=0 in +the environment, and run make.bash (or, +on Windows, make.bat). +

    + +

    +Once the Go 1.4 source has been unpacked into your GOROOT_BOOTSTRAP directory, +you must keep this git clone instance checked out to branch +release-branch.go1.4. Specifically, do not attempt to reuse +this git clone in the later step named "Fetch the repository." The go1.4 +bootstrap toolchain must be able to properly traverse the go1.4 sources +that it assumes are present under this repository root. +

    + +

    +Note that Go 1.4 does not run on all systems that later versions of Go do. +In particular, Go 1.4 does not support current versions of macOS. +On such systems, the bootstrap toolchain must be obtained using one of the other methods. +

    +

    Install Git, if needed

    @@ -507,8 +528,8 @@ These default to the values of $GOHOSTOS and

    Choices for $GOOS are -android, darwin (macOS/iOS), -dragonfly, freebsd, illumos, js, +android, darwin, dragonfly, +freebsd, illumos, ios, js, linux, netbsd, openbsd, plan9, solaris and windows.

    @@ -567,6 +588,9 @@ The valid combinations of $GOOS and $GOARCH are: illumos amd64 +ios arm64 + + js wasm @@ -663,16 +687,13 @@ For example, you should not set $GOHOSTARCH to arm on an x86 system.

    -
  • $GO386 (for 386 only, default is auto-detected -if built on either 386 or amd64, 387 otherwise) +
  • $GO386 (for 386 only, defaults to sse2)

    -This controls the code generated by gc to use either the 387 floating-point unit -(set to 387) or SSE2 instructions (set to sse2) for -floating point computations. +This variable controls how gc implements floating point computations.

  • diff --git a/lib/time/update.bash b/lib/time/update.bash index 683d0cf390f..8eebdf11f4f 100755 --- a/lib/time/update.bash +++ b/lib/time/update.bash @@ -8,8 +8,8 @@ # Consult https://www.iana.org/time-zones for the latest versions. # Versions to use. -CODE=2020a -DATA=2020a +CODE=2020d +DATA=2020d set -e rm -rf work diff --git a/lib/time/zoneinfo.zip b/lib/time/zoneinfo.zip index c4bfd5e5dea..fa143a296da 100644 Binary files a/lib/time/zoneinfo.zip and b/lib/time/zoneinfo.zip differ diff --git a/misc/cgo/errors/badsym_test.go b/misc/cgo/errors/badsym_test.go new file mode 100644 index 00000000000..b2701bf922e --- /dev/null +++ b/misc/cgo/errors/badsym_test.go @@ -0,0 +1,216 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package errorstest + +import ( + "bytes" + "io/ioutil" + "os" + "os/exec" + "path/filepath" + "strings" + "testing" + "unicode" +) + +// A manually modified object file could pass unexpected characters +// into the files generated by cgo. + +const magicInput = "abcdefghijklmnopqrstuvwxyz0123" +const magicReplace = "\n//go:cgo_ldflag \"-badflag\"\n//" + +const cSymbol = "BadSymbol" + magicInput + "Name" +const cDefSource = "int " + cSymbol + " = 1;" +const cRefSource = "extern int " + cSymbol + "; int F() { return " + cSymbol + "; }" + +// goSource is the source code for the trivial Go file we use. +// We will replace TMPDIR with the temporary directory name. +const goSource = ` +package main + +// #cgo LDFLAGS: TMPDIR/cbad.o TMPDIR/cbad.so +// extern int F(); +import "C" + +func main() { + println(C.F()) +} +` + +func TestBadSymbol(t *testing.T) { + dir := t.TempDir() + + mkdir := func(base string) string { + ret := filepath.Join(dir, base) + if err := os.Mkdir(ret, 0755); err != nil { + t.Fatal(err) + } + return ret + } + + cdir := mkdir("c") + godir := mkdir("go") + + makeFile := func(mdir, base, source string) string { + ret := filepath.Join(mdir, base) + if err := ioutil.WriteFile(ret, []byte(source), 0644); err != nil { + t.Fatal(err) + } + return ret + } + + cDefFile := makeFile(cdir, "cdef.c", cDefSource) + cRefFile := makeFile(cdir, "cref.c", cRefSource) + + ccCmd := cCompilerCmd(t) + + cCompile := func(arg, base, src string) string { + out := filepath.Join(cdir, base) + run := append(ccCmd, arg, "-o", out, src) + output, err := exec.Command(run[0], run[1:]...).CombinedOutput() + if err != nil { + t.Log(run) + t.Logf("%s", output) + t.Fatal(err) + } + if err := os.Remove(src); err != nil { + t.Fatal(err) + } + return out + } + + // Build a shared library that defines a symbol whose name + // contains magicInput. + + cShared := cCompile("-shared", "c.so", cDefFile) + + // Build an object file that refers to the symbol whose name + // contains magicInput. + + cObj := cCompile("-c", "c.o", cRefFile) + + // Rewrite the shared library and the object file, replacing + // magicInput with magicReplace. This will have the effect of + // introducing a symbol whose name looks like a cgo command. + // The cgo tool will use that name when it generates the + // _cgo_import.go file, thus smuggling a magic //go:cgo_ldflag + // pragma into a Go file. We used to not check the pragmas in + // _cgo_import.go. + + rewrite := func(from, to string) { + obj, err := ioutil.ReadFile(from) + if err != nil { + t.Fatal(err) + } + + if bytes.Count(obj, []byte(magicInput)) == 0 { + t.Fatalf("%s: did not find magic string", from) + } + + if len(magicInput) != len(magicReplace) { + t.Fatalf("internal test error: different magic lengths: %d != %d", len(magicInput), len(magicReplace)) + } + + obj = bytes.ReplaceAll(obj, []byte(magicInput), []byte(magicReplace)) + + if err := ioutil.WriteFile(to, obj, 0644); err != nil { + t.Fatal(err) + } + } + + cBadShared := filepath.Join(godir, "cbad.so") + rewrite(cShared, cBadShared) + + cBadObj := filepath.Join(godir, "cbad.o") + rewrite(cObj, cBadObj) + + goSourceBadObject := strings.ReplaceAll(goSource, "TMPDIR", godir) + makeFile(godir, "go.go", goSourceBadObject) + + makeFile(godir, "go.mod", "module badsym") + + // Try to build our little package. + cmd := exec.Command("go", "build", "-ldflags=-v") + cmd.Dir = godir + output, err := cmd.CombinedOutput() + + // The build should fail, but we want it to fail because we + // detected the error, not because we passed a bad flag to the + // C linker. + + if err == nil { + t.Errorf("go build succeeded unexpectedly") + } + + t.Logf("%s", output) + + for _, line := range bytes.Split(output, []byte("\n")) { + if bytes.Contains(line, []byte("dynamic symbol")) && bytes.Contains(line, []byte("contains unsupported character")) { + // This is the error from cgo. + continue + } + + // We passed -ldflags=-v to see the external linker invocation, + // which should not include -badflag. + if bytes.Contains(line, []byte("-badflag")) { + t.Error("output should not mention -badflag") + } + + // Also check for compiler errors, just in case. + // GCC says "unrecognized command line option". + // clang says "unknown argument". + if bytes.Contains(line, []byte("unrecognized")) || bytes.Contains(output, []byte("unknown")) { + t.Error("problem should have been caught before invoking C linker") + } + } +} + +func cCompilerCmd(t *testing.T) []string { + cc := []string{goEnv(t, "CC")} + + out := goEnv(t, "GOGCCFLAGS") + quote := '\000' + start := 0 + lastSpace := true + backslash := false + s := string(out) + for i, c := range s { + if quote == '\000' && unicode.IsSpace(c) { + if !lastSpace { + cc = append(cc, s[start:i]) + lastSpace = true + } + } else { + if lastSpace { + start = i + lastSpace = false + } + if quote == '\000' && !backslash && (c == '"' || c == '\'') { + quote = c + backslash = false + } else if !backslash && quote == c { + quote = '\000' + } else if (quote == '\000' || quote == '"') && !backslash && c == '\\' { + backslash = true + } else { + backslash = false + } + } + } + if !lastSpace { + cc = append(cc, s[start:]) + } + return cc +} + +func goEnv(t *testing.T, key string) string { + out, err := exec.Command("go", "env", key).CombinedOutput() + if err != nil { + t.Logf("go env %s\n", key) + t.Logf("%s", out) + t.Fatal(err) + } + return strings.TrimSpace(string(out)) +} diff --git a/misc/cgo/test/callback.go b/misc/cgo/test/callback.go index e7496502931..814888e3ac5 100644 --- a/misc/cgo/test/callback.go +++ b/misc/cgo/test/callback.go @@ -181,7 +181,7 @@ func testCallbackCallers(t *testing.T) { name := []string{ "runtime.cgocallbackg1", "runtime.cgocallbackg", - "runtime.cgocallback_gofunc", + "runtime.cgocallback", "runtime.asmcgocall", "runtime.cgocall", "test._Cfunc_callback", diff --git a/misc/cgo/test/cgo_linux_test.go b/misc/cgo/test/cgo_linux_test.go index 7b56e11a27d..a9746b552ee 100644 --- a/misc/cgo/test/cgo_linux_test.go +++ b/misc/cgo/test/cgo_linux_test.go @@ -15,5 +15,6 @@ func TestSetgid(t *testing.T) { } testSetgid(t) } +func Test1435(t *testing.T) { test1435(t) } func Test6997(t *testing.T) { test6997(t) } func TestBuildID(t *testing.T) { testBuildID(t) } diff --git a/misc/cgo/test/cgo_test.go b/misc/cgo/test/cgo_test.go index b745a4417f1..f7a76d047bd 100644 --- a/misc/cgo/test/cgo_test.go +++ b/misc/cgo/test/cgo_test.go @@ -76,6 +76,8 @@ func TestCheckConst(t *testing.T) { testCheckConst(t) } func TestConst(t *testing.T) { testConst(t) } func TestCthread(t *testing.T) { testCthread(t) } func TestEnum(t *testing.T) { testEnum(t) } +func TestNamedEnum(t *testing.T) { testNamedEnum(t) } +func TestCastToEnum(t *testing.T) { testCastToEnum(t) } func TestErrno(t *testing.T) { testErrno(t) } func TestFpVar(t *testing.T) { testFpVar(t) } func TestHelpers(t *testing.T) { testHelpers(t) } diff --git a/misc/cgo/test/issue1435.go b/misc/cgo/test/issue1435.go new file mode 100644 index 00000000000..a1c7cacde73 --- /dev/null +++ b/misc/cgo/test/issue1435.go @@ -0,0 +1,184 @@ +// Copyright 2019 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build linux,cgo + +package cgotest + +import ( + "fmt" + "io/ioutil" + "strings" + "syscall" + "testing" +) + +// #include +// #include +// #include +// #include +// #include +// +// pthread_t *t = NULL; +// pthread_mutex_t mu; +// int nts = 0; +// int all_done = 0; +// +// static void *aFn(void *vargp) { +// int done = 0; +// while (!done) { +// usleep(100); +// pthread_mutex_lock(&mu); +// done = all_done; +// pthread_mutex_unlock(&mu); +// } +// return NULL; +// } +// +// void trial(int argc) { +// int i; +// nts = argc; +// t = calloc(nts, sizeof(pthread_t)); +// pthread_mutex_init(&mu, NULL); +// for (i = 0; i < nts; i++) { +// pthread_create(&t[i], NULL, aFn, NULL); +// } +// } +// +// void cleanup(void) { +// int i; +// pthread_mutex_lock(&mu); +// all_done = 1; +// pthread_mutex_unlock(&mu); +// for (i = 0; i < nts; i++) { +// pthread_join(t[i], NULL); +// } +// pthread_mutex_destroy(&mu); +// free(t); +// } +import "C" + +// compareStatus is used to confirm the contents of the thread +// specific status files match expectations. +func compareStatus(filter, expect string) error { + expected := filter + expect + pid := syscall.Getpid() + fs, err := ioutil.ReadDir(fmt.Sprintf("/proc/%d/task", pid)) + if err != nil { + return fmt.Errorf("unable to find %d tasks: %v", pid, err) + } + expectedProc := fmt.Sprintf("Pid:\t%d", pid) + foundAThread := false + for _, f := range fs { + tf := fmt.Sprintf("/proc/%s/status", f.Name()) + d, err := ioutil.ReadFile(tf) + if err != nil { + // There are a surprising number of ways this + // can error out on linux. We've seen all of + // the following, so treat any error here as + // equivalent to the "process is gone": + // os.IsNotExist(err), + // "... : no such process", + // "... : bad file descriptor. + continue + } + lines := strings.Split(string(d), "\n") + for _, line := range lines { + // Different kernel vintages pad differently. + line = strings.TrimSpace(line) + if strings.HasPrefix(line, "Pid:\t") { + // On loaded systems, it is possible + // for a TID to be reused really + // quickly. As such, we need to + // validate that the thread status + // info we just read is a task of the + // same process PID as we are + // currently running, and not a + // recently terminated thread + // resurfaced in a different process. + if line != expectedProc { + break + } + // Fall through in the unlikely case + // that filter at some point is + // "Pid:\t". + } + if strings.HasPrefix(line, filter) { + if line != expected { + return fmt.Errorf("%q got:%q want:%q (bad) [pid=%d file:'%s' %v]\n", tf, line, expected, pid, string(d), expectedProc) + } + foundAThread = true + break + } + } + } + if !foundAThread { + return fmt.Errorf("found no thread /proc//status files for process %q", expectedProc) + } + return nil +} + +// test1435 test 9 glibc implemented setuid/gid syscall functions are +// mapped. This test is a slightly more expansive test than that of +// src/syscall/syscall_linux_test.go:TestSetuidEtc() insofar as it +// launches concurrent threads from C code via CGo and validates that +// they are subject to the system calls being tested. For the actual +// Go functionality being tested here, the syscall_linux_test version +// is considered authoritative, but non-trivial improvements to that +// should be mirrored here. +func test1435(t *testing.T) { + if syscall.Getuid() != 0 { + t.Skip("skipping root only test") + } + + // Launch some threads in C. + const cts = 5 + C.trial(cts) + defer C.cleanup() + + vs := []struct { + call string + fn func() error + filter, expect string + }{ + {call: "Setegid(1)", fn: func() error { return syscall.Setegid(1) }, filter: "Gid:", expect: "\t0\t1\t0\t1"}, + {call: "Setegid(0)", fn: func() error { return syscall.Setegid(0) }, filter: "Gid:", expect: "\t0\t0\t0\t0"}, + + {call: "Seteuid(1)", fn: func() error { return syscall.Seteuid(1) }, filter: "Uid:", expect: "\t0\t1\t0\t1"}, + {call: "Setuid(0)", fn: func() error { return syscall.Setuid(0) }, filter: "Uid:", expect: "\t0\t0\t0\t0"}, + + {call: "Setgid(1)", fn: func() error { return syscall.Setgid(1) }, filter: "Gid:", expect: "\t1\t1\t1\t1"}, + {call: "Setgid(0)", fn: func() error { return syscall.Setgid(0) }, filter: "Gid:", expect: "\t0\t0\t0\t0"}, + + {call: "Setgroups([]int{0,1,2,3})", fn: func() error { return syscall.Setgroups([]int{0, 1, 2, 3}) }, filter: "Groups:", expect: "\t0 1 2 3"}, + {call: "Setgroups(nil)", fn: func() error { return syscall.Setgroups(nil) }, filter: "Groups:", expect: ""}, + {call: "Setgroups([]int{0})", fn: func() error { return syscall.Setgroups([]int{0}) }, filter: "Groups:", expect: "\t0"}, + + {call: "Setregid(101,0)", fn: func() error { return syscall.Setregid(101, 0) }, filter: "Gid:", expect: "\t101\t0\t0\t0"}, + {call: "Setregid(0,102)", fn: func() error { return syscall.Setregid(0, 102) }, filter: "Gid:", expect: "\t0\t102\t102\t102"}, + {call: "Setregid(0,0)", fn: func() error { return syscall.Setregid(0, 0) }, filter: "Gid:", expect: "\t0\t0\t0\t0"}, + + {call: "Setreuid(1,0)", fn: func() error { return syscall.Setreuid(1, 0) }, filter: "Uid:", expect: "\t1\t0\t0\t0"}, + {call: "Setreuid(0,2)", fn: func() error { return syscall.Setreuid(0, 2) }, filter: "Uid:", expect: "\t0\t2\t2\t2"}, + {call: "Setreuid(0,0)", fn: func() error { return syscall.Setreuid(0, 0) }, filter: "Uid:", expect: "\t0\t0\t0\t0"}, + + {call: "Setresgid(101,0,102)", fn: func() error { return syscall.Setresgid(101, 0, 102) }, filter: "Gid:", expect: "\t101\t0\t102\t0"}, + {call: "Setresgid(0,102,101)", fn: func() error { return syscall.Setresgid(0, 102, 101) }, filter: "Gid:", expect: "\t0\t102\t101\t102"}, + {call: "Setresgid(0,0,0)", fn: func() error { return syscall.Setresgid(0, 0, 0) }, filter: "Gid:", expect: "\t0\t0\t0\t0"}, + + {call: "Setresuid(1,0,2)", fn: func() error { return syscall.Setresuid(1, 0, 2) }, filter: "Uid:", expect: "\t1\t0\t2\t0"}, + {call: "Setresuid(0,2,1)", fn: func() error { return syscall.Setresuid(0, 2, 1) }, filter: "Uid:", expect: "\t0\t2\t1\t2"}, + {call: "Setresuid(0,0,0)", fn: func() error { return syscall.Setresuid(0, 0, 0) }, filter: "Uid:", expect: "\t0\t0\t0\t0"}, + } + + for i, v := range vs { + if err := v.fn(); err != nil { + t.Errorf("[%d] %q failed: %v", i, v.call, err) + continue + } + if err := compareStatus(v.filter, v.expect); err != nil { + t.Errorf("[%d] %q comparison: %v", i, v.call, err) + } + } +} diff --git a/misc/cgo/test/issue18146.go b/misc/cgo/test/issue18146.go index 196d98f5079..f92d6c7f939 100644 --- a/misc/cgo/test/issue18146.go +++ b/misc/cgo/test/issue18146.go @@ -24,7 +24,7 @@ func test18146(t *testing.T) { t.Skip("skipping in short mode") } - if runtime.GOOS == "darwin" { + if runtime.GOOS == "darwin" || runtime.GOOS == "ios" { t.Skipf("skipping flaky test on %s; see golang.org/issue/18202", runtime.GOOS) } diff --git a/misc/cgo/test/issue4029.c b/misc/cgo/test/issue4029.c index 30646ade021..e79c5a709c8 100644 --- a/misc/cgo/test/issue4029.c +++ b/misc/cgo/test/issue4029.c @@ -3,6 +3,7 @@ // license that can be found in the LICENSE file. // +build !windows,!static +// +build !darwin !internal_pie,!arm64 #include #include diff --git a/misc/cgo/test/issue4029.go b/misc/cgo/test/issue4029.go index 1bf029d760a..b2d131833a9 100644 --- a/misc/cgo/test/issue4029.go +++ b/misc/cgo/test/issue4029.go @@ -3,6 +3,11 @@ // license that can be found in the LICENSE file. // +build !windows,!static +// +build !darwin !internal_pie,!arm64 + +// Excluded in darwin internal linking PIE mode, as dynamic export is not +// supported. +// Excluded in internal linking mode on darwin/arm64, as it is always PIE. package cgotest diff --git a/misc/cgo/test/issue4029w.go b/misc/cgo/test/issue4029w.go index eee33f70101..b969bdd0fe8 100644 --- a/misc/cgo/test/issue4029w.go +++ b/misc/cgo/test/issue4029w.go @@ -2,7 +2,7 @@ // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. -// +build windows static +// +build windows static darwin,internal_pie darwin,arm64 package cgotest diff --git a/misc/cgo/test/issue42495.go b/misc/cgo/test/issue42495.go new file mode 100644 index 00000000000..509a67d9a30 --- /dev/null +++ b/misc/cgo/test/issue42495.go @@ -0,0 +1,15 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package cgotest + +// typedef struct { } T42495A; +// typedef struct { int x[0]; } T42495B; +import "C" + +//export Issue42495A +func Issue42495A(C.T42495A) {} + +//export Issue42495B +func Issue42495B(C.T42495B) {} diff --git a/misc/cgo/test/pkg_test.go b/misc/cgo/test/pkg_test.go index 26c50ad883f..a28ad4ea74f 100644 --- a/misc/cgo/test/pkg_test.go +++ b/misc/cgo/test/pkg_test.go @@ -30,7 +30,7 @@ func TestCrossPackageTests(t *testing.T) { switch runtime.GOOS { case "android": t.Skip("Can't exec cmd/go subprocess on Android.") - case "darwin": + case "darwin", "ios": switch runtime.GOARCH { case "arm64": t.Skip("Can't exec cmd/go subprocess on iOS.") diff --git a/misc/cgo/test/sigaltstack.go b/misc/cgo/test/sigaltstack.go index 8dfa1cb5adb..034cc4b3719 100644 --- a/misc/cgo/test/sigaltstack.go +++ b/misc/cgo/test/sigaltstack.go @@ -62,7 +62,7 @@ import ( func testSigaltstack(t *testing.T) { switch { - case runtime.GOOS == "solaris", runtime.GOOS == "illumos", runtime.GOOS == "darwin" && runtime.GOARCH == "arm64": + case runtime.GOOS == "solaris", runtime.GOOS == "illumos", runtime.GOOS == "ios" && runtime.GOARCH == "arm64": t.Skipf("switching signal stack not implemented on %s/%s", runtime.GOOS, runtime.GOARCH) } diff --git a/misc/cgo/test/test.go b/misc/cgo/test/test.go index 05fa52b3810..65823b1ca0e 100644 --- a/misc/cgo/test/test.go +++ b/misc/cgo/test/test.go @@ -1000,6 +1000,32 @@ func testEnum(t *testing.T) { } } +func testNamedEnum(t *testing.T) { + e := new(C.enum_E) + + *e = C.Enum1 + if *e != 1 { + t.Error("bad enum", C.Enum1) + } + + *e = C.Enum2 + if *e != 2 { + t.Error("bad enum", C.Enum2) + } +} + +func testCastToEnum(t *testing.T) { + e := C.enum_E(C.Enum1) + if e != 1 { + t.Error("bad enum", C.Enum1) + } + + e = C.enum_E(C.Enum2) + if e != 2 { + t.Error("bad enum", C.Enum2) + } +} + func testAtol(t *testing.T) { l := Atol("123") if l != 123 { @@ -1776,7 +1802,7 @@ func test14838(t *testing.T) { var sink C.int func test17065(t *testing.T) { - if runtime.GOOS == "darwin" { + if runtime.GOOS == "darwin" || runtime.GOOS == "ios" { t.Skip("broken on darwin; issue 17065") } for i := range C.ii { diff --git a/misc/cgo/test/testdata/issue41761.go b/misc/cgo/test/testdata/issue41761.go new file mode 100644 index 00000000000..919c7492510 --- /dev/null +++ b/misc/cgo/test/testdata/issue41761.go @@ -0,0 +1,20 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package cgotest + +/* + typedef struct S S; +*/ +import "C" + +import ( + "cgotest/issue41761a" + "testing" +) + +func test41761(t *testing.T) { + var x issue41761a.T + _ = (*C.struct_S)(x.X) +} diff --git a/misc/cgo/test/testdata/issue41761a/a.go b/misc/cgo/test/testdata/issue41761a/a.go new file mode 100644 index 00000000000..ca5c18191eb --- /dev/null +++ b/misc/cgo/test/testdata/issue41761a/a.go @@ -0,0 +1,14 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package issue41761a + +/* + typedef struct S S; +*/ +import "C" + +type T struct { + X *C.S +} diff --git a/misc/cgo/test/testdata/issue9400/asm_riscv64.s b/misc/cgo/test/testdata/issue9400/asm_riscv64.s new file mode 100644 index 00000000000..20fcc0066d6 --- /dev/null +++ b/misc/cgo/test/testdata/issue9400/asm_riscv64.s @@ -0,0 +1,31 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build riscv64 +// +build !gccgo + +#include "textflag.h" + +TEXT ·RewindAndSetgid(SB),NOSPLIT|NOFRAME,$0-0 + // Rewind stack pointer so anything that happens on the stack + // will clobber the test pattern created by the caller + ADD $(1024*8), X2 + + // Ask signaller to setgid + MOV $1, X5 + FENCE + MOVW X5, ·Baton(SB) + FENCE + + // Wait for setgid completion +loop: + FENCE + MOVW ·Baton(SB), X5 + OR X6, X6, X6 // hint that we're in a spin loop + BNE ZERO, X5, loop + FENCE + + // Restore stack + ADD $(-1024*8), X2 + RET diff --git a/misc/cgo/test/testx.go b/misc/cgo/test/testx.go index 7fbc5c64b31..2b2e69ec00f 100644 --- a/misc/cgo/test/testx.go +++ b/misc/cgo/test/testx.go @@ -164,7 +164,7 @@ func Add(x int) { } func testCthread(t *testing.T) { - if runtime.GOOS == "darwin" && runtime.GOARCH == "arm64" { + if (runtime.GOOS == "darwin" || runtime.GOOS == "ios") && runtime.GOARCH == "arm64" { t.Skip("the iOS exec wrapper is unable to properly handle the panic from Add") } sum.i = 0 diff --git a/misc/cgo/testcarchive/carchive_test.go b/misc/cgo/testcarchive/carchive_test.go index b4a046c4bc4..6ed25d8948f 100644 --- a/misc/cgo/testcarchive/carchive_test.go +++ b/misc/cgo/testcarchive/carchive_test.go @@ -118,9 +118,9 @@ func testMain(m *testing.M) int { cc = append(cc, s[start:]) } - if GOOS == "darwin" { + if GOOS == "darwin" || GOOS == "ios" { // For Darwin/ARM. - // TODO(crawshaw): can we do better? + // TODO: do we still need this? cc = append(cc, []string{"-framework", "CoreFoundation", "-framework", "Foundation"}...) } if GOOS == "aix" { @@ -133,7 +133,7 @@ func testMain(m *testing.M) int { libbase = "gccgo_" + libgodir + "_fPIC" } else { switch GOOS { - case "darwin": + case "darwin", "ios": if GOARCH == "arm64" { libbase += "_shared" } @@ -303,7 +303,7 @@ func TestInstall(t *testing.T) { func TestEarlySignalHandler(t *testing.T) { switch GOOS { - case "darwin": + case "darwin", "ios": switch GOARCH { case "arm64": t.Skipf("skipping on %s/%s; see https://golang.org/issue/13701", GOOS, GOARCH) @@ -384,7 +384,7 @@ func TestSignalForwarding(t *testing.T) { expectSignal(t, err, syscall.SIGSEGV) // SIGPIPE is never forwarded on darwin. See golang.org/issue/33384. - if runtime.GOOS != "darwin" { + if runtime.GOOS != "darwin" && runtime.GOOS != "ios" { // Test SIGPIPE forwarding cmd = exec.Command(bin[0], append(bin[1:], "3")...) @@ -485,7 +485,7 @@ func TestSignalForwardingExternal(t *testing.T) { // doesn't work on this platform. func checkSignalForwardingTest(t *testing.T) { switch GOOS { - case "darwin": + case "darwin", "ios": switch GOARCH { case "arm64": t.Skipf("skipping on %s/%s; see https://golang.org/issue/13701", GOOS, GOARCH) @@ -603,7 +603,7 @@ func TestExtar(t *testing.T) { if runtime.Compiler == "gccgo" { t.Skip("skipping -extar test when using gccgo") } - if runtime.GOOS == "darwin" && runtime.GOARCH == "arm64" { + if runtime.GOOS == "ios" { t.Skip("shell scripts are not executable on iOS hosts") } @@ -645,7 +645,7 @@ func TestExtar(t *testing.T) { func TestPIE(t *testing.T) { switch GOOS { - case "windows", "darwin", "plan9": + case "windows", "darwin", "ios", "plan9": t.Skipf("skipping PIE test on %s", GOOS) } @@ -738,7 +738,7 @@ func TestSIGPROF(t *testing.T) { switch GOOS { case "windows", "plan9": t.Skipf("skipping SIGPROF test on %s", GOOS) - case "darwin": + case "darwin", "ios": t.Skipf("skipping SIGPROF test on %s; see https://golang.org/issue/19320", GOOS) } @@ -841,7 +841,7 @@ func TestCompileWithoutShared(t *testing.T) { expectSignal(t, err, syscall.SIGSEGV) // SIGPIPE is never forwarded on darwin. See golang.org/issue/33384. - if runtime.GOOS != "darwin" { + if runtime.GOOS != "darwin" && runtime.GOOS != "ios" { binArgs := append(cmdToRun(exe), "3") t.Log(binArgs) out, err = exec.Command(binArgs[0], binArgs[1:]...).CombinedOutput() diff --git a/misc/cgo/testcshared/cshared_test.go b/misc/cgo/testcshared/cshared_test.go index bd4d341820e..3a4886cf30a 100644 --- a/misc/cgo/testcshared/cshared_test.go +++ b/misc/cgo/testcshared/cshared_test.go @@ -7,6 +7,8 @@ package cshared_test import ( "bytes" "debug/elf" + "debug/pe" + "encoding/binary" "flag" "fmt" "io/ioutil" @@ -98,7 +100,7 @@ func testMain(m *testing.M) int { } switch GOOS { - case "darwin": + case "darwin", "ios": // For Darwin/ARM. // TODO(crawshaw): can we do better? cc = append(cc, []string{"-framework", "CoreFoundation", "-framework", "Foundation"}...) @@ -107,7 +109,7 @@ func testMain(m *testing.M) int { } libgodir := GOOS + "_" + GOARCH switch GOOS { - case "darwin": + case "darwin", "ios": if GOARCH == "arm64" { libgodir += "_shared" } @@ -355,6 +357,101 @@ func TestExportedSymbols(t *testing.T) { } } +func checkNumberOfExportedFunctionsWindows(t *testing.T, exportAllSymbols bool) { + const prog = ` +package main + +import "C" + +//export GoFunc +func GoFunc() { + println(42) +} + +//export GoFunc2 +func GoFunc2() { + println(24) +} + +func main() { +} +` + + tmpdir := t.TempDir() + + srcfile := filepath.Join(tmpdir, "test.go") + objfile := filepath.Join(tmpdir, "test.dll") + if err := ioutil.WriteFile(srcfile, []byte(prog), 0666); err != nil { + t.Fatal(err) + } + argv := []string{"build", "-buildmode=c-shared"} + if exportAllSymbols { + argv = append(argv, "-ldflags", "-extldflags=-Wl,--export-all-symbols") + } + argv = append(argv, "-o", objfile, srcfile) + out, err := exec.Command("go", argv...).CombinedOutput() + if err != nil { + t.Fatalf("build failure: %s\n%s\n", err, string(out)) + } + + f, err := pe.Open(objfile) + if err != nil { + t.Fatalf("pe.Open failed: %v", err) + } + defer f.Close() + section := f.Section(".edata") + if section == nil { + t.Fatalf(".edata section is not present") + } + + // TODO: deduplicate this struct from cmd/link/internal/ld/pe.go + type IMAGE_EXPORT_DIRECTORY struct { + _ [2]uint32 + _ [2]uint16 + _ [2]uint32 + NumberOfFunctions uint32 + NumberOfNames uint32 + _ [3]uint32 + } + var e IMAGE_EXPORT_DIRECTORY + if err := binary.Read(section.Open(), binary.LittleEndian, &e); err != nil { + t.Fatalf("binary.Read failed: %v", err) + } + + // Only the two exported functions and _cgo_dummy_export should be exported + expectedNumber := uint32(3) + + if exportAllSymbols { + if e.NumberOfFunctions <= expectedNumber { + t.Fatalf("missing exported functions: %v", e.NumberOfFunctions) + } + if e.NumberOfNames <= expectedNumber { + t.Fatalf("missing exported names: %v", e.NumberOfNames) + } + } else { + if e.NumberOfFunctions != expectedNumber { + t.Fatalf("got %d exported functions; want %d", e.NumberOfFunctions, expectedNumber) + } + if e.NumberOfNames != expectedNumber { + t.Fatalf("got %d exported names; want %d", e.NumberOfNames, expectedNumber) + } + } +} + +func TestNumberOfExportedFunctions(t *testing.T) { + if GOOS != "windows" { + t.Skip("skipping windows only test") + } + t.Parallel() + + t.Run("OnlyExported", func(t *testing.T) { + checkNumberOfExportedFunctionsWindows(t, false) + }) + t.Run("All", func(t *testing.T) { + checkNumberOfExportedFunctionsWindows(t, true) + }) +} + // test1: shared library can be dynamically loaded and exported symbols are accessible. func TestExportedSymbolsWithDynamicLoad(t *testing.T) { t.Parallel() @@ -407,7 +504,7 @@ func TestUnexportedSymbols(t *testing.T) { adbPush(t, libname) linkFlags := "-Wl,--no-as-needed" - if GOOS == "darwin" { + if GOOS == "darwin" || GOOS == "ios" { linkFlags = "" } @@ -636,7 +733,7 @@ func copyFile(t *testing.T, dst, src string) { func TestGo2C2Go(t *testing.T) { switch GOOS { - case "darwin": + case "darwin", "ios": // Darwin shared libraries don't support the multiple // copies of the runtime package implied by this test. t.Skip("linking c-shared into Go programs not supported on Darwin; issue 29061") diff --git a/misc/cgo/testgodefs/testdata/bitfields.go b/misc/cgo/testgodefs/testdata/bitfields.go new file mode 100644 index 00000000000..6a9724dcd15 --- /dev/null +++ b/misc/cgo/testgodefs/testdata/bitfields.go @@ -0,0 +1,31 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// +// +build ignore + +package main + +// This file tests that we don't generate an incorrect field location +// for a bitfield that appears aligned. + +/* +struct bitfields { + unsigned int B1 : 5; + unsigned int B2 : 1; + unsigned int B3 : 1; + unsigned int B4 : 1; + unsigned int Short1 : 16; // misaligned on 8 bit boundary + unsigned int B5 : 1; + unsigned int B6 : 1; + unsigned int B7 : 1; + unsigned int B8 : 1; + unsigned int B9 : 1; + unsigned int B10 : 3; + unsigned int Short2 : 16; // alignment is OK + unsigned int Short3 : 16; // alignment is OK +}; +*/ +import "C" + +type bitfields C.struct_bitfields diff --git a/misc/cgo/testgodefs/testdata/main.go b/misc/cgo/testgodefs/testdata/main.go index 2e1ad3376a1..4a3f6a701cc 100644 --- a/misc/cgo/testgodefs/testdata/main.go +++ b/misc/cgo/testgodefs/testdata/main.go @@ -4,6 +4,12 @@ package main +import ( + "fmt" + "os" + "reflect" +) + // Test that the struct field in anonunion.go was promoted. var v1 T var v2 = v1.L @@ -23,4 +29,26 @@ var v7 = S{} var _ = issue38649{X: 0} func main() { + pass := true + + // The Go translation of bitfields should not have any of the + // bitfield types. The order in which bitfields are laid out + // in memory is implementation defined, so we can't easily + // know how a bitfield should correspond to a Go type, even if + // it appears to be aligned correctly. + bitfieldType := reflect.TypeOf(bitfields{}) + check := func(name string) { + _, ok := bitfieldType.FieldByName(name) + if ok { + fmt.Fprintf(os.Stderr, "found unexpected bitfields field %s\n", name) + pass = false + } + } + check("Short1") + check("Short2") + check("Short3") + + if !pass { + os.Exit(1) + } } diff --git a/misc/cgo/testgodefs/testgodefs_test.go b/misc/cgo/testgodefs/testgodefs_test.go index e4085f9ca80..4c2312c1c89 100644 --- a/misc/cgo/testgodefs/testgodefs_test.go +++ b/misc/cgo/testgodefs/testgodefs_test.go @@ -19,6 +19,7 @@ import ( // import "C" block. Add more tests here. var filePrefixes = []string{ "anonunion", + "bitfields", "issue8478", "fieldtypedef", "issue37479", diff --git a/misc/cgo/testplugin/plugin_test.go b/misc/cgo/testplugin/plugin_test.go index 2875271c03b..9055dbda044 100644 --- a/misc/cgo/testplugin/plugin_test.go +++ b/misc/cgo/testplugin/plugin_test.go @@ -196,3 +196,17 @@ func TestIssue25756(t *testing.T) { }) } } + +func TestMethod(t *testing.T) { + // Exported symbol's method must be live. + goCmd(t, "build", "-buildmode=plugin", "-o", "plugin.so", "./method/plugin.go") + goCmd(t, "build", "-o", "method.exe", "./method/main.go") + + ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second) + defer cancel() + cmd := exec.CommandContext(ctx, "./method.exe") + out, err := cmd.CombinedOutput() + if err != nil { + t.Fatalf("%s: %v\n%s", strings.Join(cmd.Args, " "), err, out) + } +} diff --git a/misc/cgo/testplugin/testdata/method/main.go b/misc/cgo/testplugin/testdata/method/main.go new file mode 100644 index 00000000000..5e9189b4500 --- /dev/null +++ b/misc/cgo/testplugin/testdata/method/main.go @@ -0,0 +1,26 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Issue 42579: methods of symbols exported from plugin must be live. + +package main + +import ( + "plugin" + "reflect" +) + +func main() { + p, err := plugin.Open("plugin.so") + if err != nil { + panic(err) + } + + x, err := p.Lookup("X") + if err != nil { + panic(err) + } + + reflect.ValueOf(x).Elem().MethodByName("M").Call(nil) +} diff --git a/misc/cgo/testplugin/testdata/method/plugin.go b/misc/cgo/testplugin/testdata/method/plugin.go new file mode 100644 index 00000000000..240edd3bc45 --- /dev/null +++ b/misc/cgo/testplugin/testdata/method/plugin.go @@ -0,0 +1,13 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package main + +func main() {} + +type T int + +func (T) M() { println("M") } + +var X T diff --git a/misc/cgo/testsanitizers/msan_test.go b/misc/cgo/testsanitizers/msan_test.go index 88b90d3d70d..5e2f9759bae 100644 --- a/misc/cgo/testsanitizers/msan_test.go +++ b/misc/cgo/testsanitizers/msan_test.go @@ -28,6 +28,7 @@ func TestMSAN(t *testing.T) { {src: "msan4.go"}, {src: "msan5.go"}, {src: "msan6.go"}, + {src: "msan7.go"}, {src: "msan_fail.go", wantErr: true}, } for _, tc := range cases { diff --git a/misc/cgo/testsanitizers/testdata/msan7.go b/misc/cgo/testsanitizers/testdata/msan7.go new file mode 100644 index 00000000000..2f29fd21b2e --- /dev/null +++ b/misc/cgo/testsanitizers/testdata/msan7.go @@ -0,0 +1,38 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package main + +// Test passing C struct to exported Go function. + +/* +#include +#include + +// T is a C struct with alignment padding after b. +// The padding bytes are not considered initialized by MSAN. +// It is big enough to be passed on stack in C ABI (and least +// on AMD64). +typedef struct { char b; uintptr_t x, y; } T; + +extern void F(T); + +// Use weak as a hack to permit defining a function even though we use export. +void CF(int x) __attribute__ ((weak)); +void CF(int x) { + T *t = malloc(sizeof(T)); + t->b = (char)x; + t->x = x; + t->y = x; + F(*t); +} +*/ +import "C" + +//export F +func F(t C.T) { println(t.b, t.x, t.y) } + +func main() { + C.CF(C.int(0)) +} diff --git a/misc/cgo/testso/so_test.go b/misc/cgo/testso/so_test.go index bdd6bd84685..57f0fd34f78 100644 --- a/misc/cgo/testso/so_test.go +++ b/misc/cgo/testso/so_test.go @@ -21,7 +21,7 @@ func requireTestSOSupported(t *testing.T) { t.Helper() switch runtime.GOARCH { case "arm64": - if runtime.GOOS == "darwin" { + if runtime.GOOS == "darwin" || runtime.GOOS == "ios" { t.Skip("No exec facility on iOS.") } case "ppc64": @@ -74,7 +74,7 @@ func TestSO(t *testing.T) { ext := "so" args := append(gogccflags, "-shared") switch runtime.GOOS { - case "darwin": + case "darwin", "ios": ext = "dylib" args = append(args, "-undefined", "suppress", "-flat_namespace") case "windows": @@ -119,7 +119,7 @@ func TestSO(t *testing.T) { cmd.Env = append(os.Environ(), "GOPATH="+GOPATH) if runtime.GOOS != "windows" { s := "LD_LIBRARY_PATH" - if runtime.GOOS == "darwin" { + if runtime.GOOS == "darwin" || runtime.GOOS == "ios" { s = "DYLD_LIBRARY_PATH" } cmd.Env = append(os.Environ(), s+"=.") diff --git a/misc/cgo/testsovar/so_test.go b/misc/cgo/testsovar/so_test.go index bdd6bd84685..57f0fd34f78 100644 --- a/misc/cgo/testsovar/so_test.go +++ b/misc/cgo/testsovar/so_test.go @@ -21,7 +21,7 @@ func requireTestSOSupported(t *testing.T) { t.Helper() switch runtime.GOARCH { case "arm64": - if runtime.GOOS == "darwin" { + if runtime.GOOS == "darwin" || runtime.GOOS == "ios" { t.Skip("No exec facility on iOS.") } case "ppc64": @@ -74,7 +74,7 @@ func TestSO(t *testing.T) { ext := "so" args := append(gogccflags, "-shared") switch runtime.GOOS { - case "darwin": + case "darwin", "ios": ext = "dylib" args = append(args, "-undefined", "suppress", "-flat_namespace") case "windows": @@ -119,7 +119,7 @@ func TestSO(t *testing.T) { cmd.Env = append(os.Environ(), "GOPATH="+GOPATH) if runtime.GOOS != "windows" { s := "LD_LIBRARY_PATH" - if runtime.GOOS == "darwin" { + if runtime.GOOS == "darwin" || runtime.GOOS == "ios" { s = "DYLD_LIBRARY_PATH" } cmd.Env = append(os.Environ(), s+"=.") diff --git a/misc/ios/README b/misc/ios/README index d7df191414b..433bcdfd8f9 100644 --- a/misc/ios/README +++ b/misc/ios/README @@ -1,13 +1,20 @@ Go on iOS ========= -For details on developing Go for iOS on macOS, see the documentation in the mobile -subrepository: +To run the standard library tests, run all.bash as usual, but with the compiler +set to the clang wrapper that invokes clang for iOS. For example, this command runs + all.bash on the iOS emulator: - https://github.com/golang/mobile + GOOS=ios GOARCH=amd64 CGO_ENABLED=1 CC_FOR_TARGET=$(pwd)/../misc/ios/clangwrap.sh ./all.bash -It is necessary to set up the environment before running tests or programs directly on a -device. +To use the go tool to run individual programs and tests, put $GOROOT/bin into PATH to ensure +the go_ios_$GOARCH_exec wrapper is found. For example, to run the archive/tar tests: + + export PATH=$GOROOT/bin:$PATH + GOOS=ios GOARCH=amd64 CGO_ENABLED=1 go test archive/tar + +The go_ios_exec wrapper uses GOARCH to select the emulator (amd64) or the device (arm64). +However, further setup is required to run tests or programs directly on a device. First make sure you have a valid developer certificate and have setup your device properly to run apps signed by your developer certificate. Then install the libimobiledevice and @@ -29,18 +36,10 @@ which will output something similar to export GOIOS_TEAM_ID=ZZZZZZZZ If you have multiple devices connected, specify the device UDID with the GOIOS_DEVICE_ID -variable. Use `idevice_id -l` to list all available UDIDs. +variable. Use `idevice_id -l` to list all available UDIDs. Then, setting GOARCH to arm64 +will select the device: -Finally, to run the standard library tests, run all.bash as usual, but with the compiler -set to the clang wrapper that invokes clang for iOS. For example, - - GOARCH=arm64 CGO_ENABLED=1 CC_FOR_TARGET=$(pwd)/../misc/ios/clangwrap.sh ./all.bash - -To use the go tool directly to run programs and tests, put $GOROOT/bin into PATH to ensure -the go_darwin_$GOARCH_exec wrapper is found. For example, to run the archive/tar tests - - export PATH=$GOROOT/bin:$PATH - GOARCH=arm64 CGO_ENABLED=1 go test archive/tar + GOOS=ios GOARCH=arm64 CGO_ENABLED=1 CC_FOR_TARGET=$(pwd)/../misc/ios/clangwrap.sh ./all.bash Note that the go_darwin_$GOARCH_exec wrapper uninstalls any existing app identified by the bundle id before installing a new app. If the uninstalled app is the last app by diff --git a/misc/ios/clangwrap.sh b/misc/ios/clangwrap.sh index 5fdbb6db4ac..dca3fcc9043 100755 --- a/misc/ios/clangwrap.sh +++ b/misc/ios/clangwrap.sh @@ -2,17 +2,19 @@ # This uses the latest available iOS SDK, which is recommended. # To select a specific SDK, run 'xcodebuild -showsdks' # to see the available SDKs and replace iphoneos with one of them. -SDK=iphoneos +if [ "$GOARCH" == "arm64" ]; then + SDK=iphoneos + PLATFORM=ios + CLANGARCH="arm64" +else + SDK=iphonesimulator + PLATFORM=ios-simulator + CLANGARCH="x86_64" +fi + SDK_PATH=`xcrun --sdk $SDK --show-sdk-path` export IPHONEOS_DEPLOYMENT_TARGET=5.1 # cmd/cgo doesn't support llvm-gcc-4.2, so we have to use clang. CLANG=`xcrun --sdk $SDK --find clang` -if [ "$GOARCH" == "arm64" ]; then - CLANGARCH="arm64" -else - echo "unknown GOARCH=$GOARCH" >&2 - exit 1 -fi - -exec $CLANG -arch $CLANGARCH -isysroot $SDK_PATH -mios-version-min=10.0 "$@" +exec "$CLANG" -arch $CLANGARCH -isysroot "$SDK_PATH" -m${PLATFORM}-version-min=10.0 "$@" diff --git a/misc/ios/detect.go b/misc/ios/detect.go index 1d47e47c860..d32bcc3202a 100644 --- a/misc/ios/detect.go +++ b/misc/ios/detect.go @@ -6,7 +6,7 @@ // detect attempts to autodetect the correct // values of the environment variables -// used by go_darwin_arm_exec. +// used by go_ios_exec. // detect shells out to ideviceinfo, a third party program that can // be obtained by following the instructions at // https://github.com/libimobiledevice/libimobiledevice. diff --git a/misc/ios/go_darwin_arm_exec.go b/misc/ios/go_ios_exec.go similarity index 81% rename from misc/ios/go_darwin_arm_exec.go rename to misc/ios/go_ios_exec.go index cdf4b07d0aa..0acf1b259c0 100644 --- a/misc/ios/go_darwin_arm_exec.go +++ b/misc/ios/go_ios_exec.go @@ -2,7 +2,7 @@ // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. -// This program can be used as go_darwin_arm_exec by the Go tool. +// This program can be used as go_ios_$GOARCH_exec by the Go tool. // It executes binaries on an iOS device using the XCode toolchain // and the ios-deploy program: https://github.com/phonegap/ios-deploy // @@ -34,6 +34,7 @@ import ( "os/signal" "path/filepath" "runtime" + "strconv" "strings" "syscall" "time" @@ -58,34 +59,16 @@ var lock *os.File func main() { log.SetFlags(0) - log.SetPrefix("go_darwin_arm_exec: ") + log.SetPrefix("go_ios_exec: ") if debug { log.Println(strings.Join(os.Args, " ")) } if len(os.Args) < 2 { - log.Fatal("usage: go_darwin_arm_exec a.out") + log.Fatal("usage: go_ios_exec a.out") } - // e.g. B393DDEB490947F5A463FD074299B6C0AXXXXXXX - devID = getenv("GOIOS_DEV_ID") - - // e.g. Z8B3JBXXXX.org.golang.sample, Z8B3JBXXXX prefix is available at - // https://developer.apple.com/membercenter/index.action#accountSummary as Team ID. - appID = getenv("GOIOS_APP_ID") - - // e.g. Z8B3JBXXXX, available at - // https://developer.apple.com/membercenter/index.action#accountSummary as Team ID. - teamID = getenv("GOIOS_TEAM_ID") - - // Device IDs as listed with ios-deploy -c. - deviceID = os.Getenv("GOIOS_DEVICE_ID") - - parts := strings.SplitN(appID, ".", 2) // For compatibility with the old builders, use a fallback bundle ID bundleID = "golang.gotest" - if len(parts) == 2 { - bundleID = parts[1] - } exitCode, err := runMain() if err != nil { @@ -96,7 +79,7 @@ func main() { func runMain() (int, error) { var err error - tmpdir, err = ioutil.TempDir("", "go_darwin_arm_exec_") + tmpdir, err = ioutil.TempDir("", "go_ios_exec_") if err != nil { return 1, err } @@ -117,7 +100,7 @@ func runMain() (int, error) { // // The lock file is never deleted, to avoid concurrent locks on distinct // files with the same path. - lockName := filepath.Join(os.TempDir(), "go_darwin_arm_exec-"+deviceID+".lock") + lockName := filepath.Join(os.TempDir(), "go_ios_exec-"+deviceID+".lock") lock, err = os.OpenFile(lockName, os.O_CREATE|os.O_RDONLY, 0666) if err != nil { return 1, err @@ -126,28 +109,12 @@ func runMain() (int, error) { return 1, err } - if err := uninstall(bundleID); err != nil { - return 1, err + if goarch := os.Getenv("GOARCH"); goarch == "arm64" { + err = runOnDevice(appdir) + } else { + err = runOnSimulator(appdir) } - - if err := install(appdir); err != nil { - return 1, err - } - - if err := mountDevImage(); err != nil { - return 1, err - } - - // Kill any hanging debug bridges that might take up port 3222. - exec.Command("killall", "idevicedebugserverproxy").Run() - - closer, err := startDebugBridge() if err != nil { - return 1, err - } - defer closer() - - if err := run(appdir, bundleID, os.Args[2:]); err != nil { // If the lldb driver completed with an exit code, use that. if err, ok := err.(*exec.ExitError); ok { if ws, ok := err.Sys().(interface{ ExitStatus() int }); ok { @@ -159,6 +126,62 @@ func runMain() (int, error) { return 0, nil } +func runOnSimulator(appdir string) error { + if err := installSimulator(appdir); err != nil { + return err + } + + return runSimulator(appdir, bundleID, os.Args[2:]) +} + +func runOnDevice(appdir string) error { + // e.g. B393DDEB490947F5A463FD074299B6C0AXXXXXXX + devID = getenv("GOIOS_DEV_ID") + + // e.g. Z8B3JBXXXX.org.golang.sample, Z8B3JBXXXX prefix is available at + // https://developer.apple.com/membercenter/index.action#accountSummary as Team ID. + appID = getenv("GOIOS_APP_ID") + + // e.g. Z8B3JBXXXX, available at + // https://developer.apple.com/membercenter/index.action#accountSummary as Team ID. + teamID = getenv("GOIOS_TEAM_ID") + + // Device IDs as listed with ios-deploy -c. + deviceID = os.Getenv("GOIOS_DEVICE_ID") + + parts := strings.SplitN(appID, ".", 2) + if len(parts) == 2 { + bundleID = parts[1] + } + + if err := signApp(appdir); err != nil { + return err + } + + if err := uninstallDevice(bundleID); err != nil { + return err + } + + if err := installDevice(appdir); err != nil { + return err + } + + if err := mountDevImage(); err != nil { + return err + } + + // Kill any hanging debug bridges that might take up port 3222. + exec.Command("killall", "idevicedebugserverproxy").Run() + + closer, err := startDebugBridge() + if err != nil { + return err + } + defer closer() + + return runDevice(appdir, bundleID, os.Args[2:]) +} + func getenv(envvar string) string { s := os.Getenv(envvar) if s == "" { @@ -191,7 +214,11 @@ func assembleApp(appdir, bin string) error { if err := ioutil.WriteFile(filepath.Join(appdir, "ResourceRules.plist"), []byte(resourceRules), 0744); err != nil { return err } + return nil +} +func signApp(appdir string) error { + entitlementsPath := filepath.Join(tmpdir, "Entitlements.plist") cmd := exec.Command( "codesign", "-f", @@ -421,7 +448,20 @@ func parsePlistDict(dict []byte) (map[string]string, error) { return values, nil } -func uninstall(bundleID string) error { +func installSimulator(appdir string) error { + cmd := exec.Command( + "xcrun", "simctl", "install", + "booted", // Install to the booted simulator. + appdir, + ) + if out, err := cmd.CombinedOutput(); err != nil { + os.Stderr.Write(out) + return fmt.Errorf("xcrun simctl install booted %q: %v", appdir, err) + } + return nil +} + +func uninstallDevice(bundleID string) error { cmd := idevCmd(exec.Command( "ideviceinstaller", "-U", bundleID, @@ -433,7 +473,7 @@ func uninstall(bundleID string) error { return nil } -func install(appdir string) error { +func installDevice(appdir string) error { attempt := 0 for { cmd := idevCmd(exec.Command( @@ -464,15 +504,28 @@ func idevCmd(cmd *exec.Cmd) *exec.Cmd { return cmd } -func run(appdir, bundleID string, args []string) error { - var env []string - for _, e := range os.Environ() { - // Don't override TMPDIR, HOME, GOCACHE on the device. - if strings.HasPrefix(e, "TMPDIR=") || strings.HasPrefix(e, "HOME=") || strings.HasPrefix(e, "GOCACHE=") { - continue - } - env = append(env, e) +func runSimulator(appdir, bundleID string, args []string) error { + cmd := exec.Command( + "xcrun", "simctl", "launch", + "--wait-for-debugger", + "booted", + bundleID, + ) + out, err := cmd.CombinedOutput() + if err != nil { + os.Stderr.Write(out) + return fmt.Errorf("xcrun simctl launch booted %q: %v", bundleID, err) } + var processID int + var ignore string + if _, err := fmt.Sscanf(string(out), "%s %d", &ignore, &processID); err != nil { + return fmt.Errorf("runSimulator: couldn't find processID from `simctl launch`: %v (%q)", err, out) + } + _, err = runLLDB("ios-simulator", appdir, strconv.Itoa(processID), args) + return err +} + +func runDevice(appdir, bundleID string, args []string) error { attempt := 0 for { // The device app path reported by the device might be stale, so retry @@ -487,37 +540,10 @@ func run(appdir, bundleID string, args []string) error { time.Sleep(5 * time.Second) continue } - lldb := exec.Command( - "python", - "-", // Read script from stdin. - appdir, - deviceapp, - ) - lldb.Args = append(lldb.Args, args...) - lldb.Env = env - lldb.Stdin = strings.NewReader(lldbDriver) - lldb.Stdout = os.Stdout - var out bytes.Buffer - lldb.Stderr = io.MultiWriter(&out, os.Stderr) - err = lldb.Start() - if err == nil { - // Forward SIGQUIT to the lldb driver which in turn will forward - // to the running program. - sigs := make(chan os.Signal, 1) - signal.Notify(sigs, syscall.SIGQUIT) - proc := lldb.Process - go func() { - for sig := range sigs { - proc.Signal(sig) - } - }() - err = lldb.Wait() - signal.Stop(sigs) - close(sigs) - } + out, err := runLLDB("remote-ios", appdir, deviceapp, args) // If the program was not started it can be retried without papering over // real test failures. - started := bytes.HasPrefix(out.Bytes(), []byte("lldb: running program")) + started := bytes.HasPrefix(out, []byte("lldb: running program")) if started || err == nil || attempt == 5 { return err } @@ -528,6 +554,47 @@ func run(appdir, bundleID string, args []string) error { } } +func runLLDB(target, appdir, deviceapp string, args []string) ([]byte, error) { + var env []string + for _, e := range os.Environ() { + // Don't override TMPDIR, HOME, GOCACHE on the device. + if strings.HasPrefix(e, "TMPDIR=") || strings.HasPrefix(e, "HOME=") || strings.HasPrefix(e, "GOCACHE=") { + continue + } + env = append(env, e) + } + lldb := exec.Command( + "python", + "-", // Read script from stdin. + target, + appdir, + deviceapp, + ) + lldb.Args = append(lldb.Args, args...) + lldb.Env = env + lldb.Stdin = strings.NewReader(lldbDriver) + lldb.Stdout = os.Stdout + var out bytes.Buffer + lldb.Stderr = io.MultiWriter(&out, os.Stderr) + err := lldb.Start() + if err == nil { + // Forward SIGQUIT to the lldb driver which in turn will forward + // to the running program. + sigs := make(chan os.Signal, 1) + signal.Notify(sigs, syscall.SIGQUIT) + proc := lldb.Process + go func() { + for sig := range sigs { + proc.Signal(sig) + } + }() + err = lldb.Wait() + signal.Stop(sigs) + close(sigs) + } + return out.Bytes(), err +} + func copyLocalDir(dst, src string) error { if err := os.Mkdir(dst, 0755); err != nil { return err @@ -679,6 +746,7 @@ func infoPlist(pkgpath string) string { CFBundleSupportedPlatformsiPhoneOS CFBundleExecutablegotest CFBundleVersion1.0 +CFBundleShortVersionString1.0 CFBundleIdentifier` + bundleID + ` CFBundleResourceSpecificationResourceRules.plist LSRequiresIPhoneOS @@ -739,7 +807,7 @@ import sys import os import signal -exe, device_exe, args = sys.argv[1], sys.argv[2], sys.argv[3:] +platform, exe, device_exe_or_pid, args = sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4:] env = [] for k, v in os.environ.items(): @@ -754,17 +822,21 @@ debugger.SetAsync(True) debugger.SkipLLDBInitFiles(True) err = lldb.SBError() -target = debugger.CreateTarget(exe, None, 'remote-ios', True, err) +target = debugger.CreateTarget(exe, None, platform, True, err) if not target.IsValid() or not err.Success(): sys.stderr.write("lldb: failed to setup up target: %s\n" % (err)) sys.exit(1) -target.modules[0].SetPlatformFileSpec(lldb.SBFileSpec(device_exe)) - listener = debugger.GetListener() -process = target.ConnectRemote(listener, 'connect://localhost:3222', None, err) + +if platform == 'remote-ios': + target.modules[0].SetPlatformFileSpec(lldb.SBFileSpec(device_exe_or_pid)) + process = target.ConnectRemote(listener, 'connect://localhost:3222', None, err) +else: + process = target.AttachToProcessWithID(listener, int(device_exe_or_pid), err) + if not err.Success(): - sys.stderr.write("lldb: failed to connect to remote target: %s\n" % (err)) + sys.stderr.write("lldb: failed to connect to remote target %s: %s\n" % (device_exe_or_pid, err)) sys.exit(1) # Don't stop on signals. @@ -777,6 +849,25 @@ for i in range(0, sigs.GetNumSignals()): event = lldb.SBEvent() running = False prev_handler = None + +def signal_handler(signal, frame): + process.Signal(signal) + +def run_program(): + # Forward SIGQUIT to the program. + prev_handler = signal.signal(signal.SIGQUIT, signal_handler) + # Tell the Go driver that the program is running and should not be retried. + sys.stderr.write("lldb: running program\n") + running = True + # Process is stopped at attach/launch. Let it run. + process.Continue() + +if platform != 'remote-ios': + # For the local emulator the program is ready to run. + # For remote device runs, we need to wait for eStateConnected, + # below. + run_program() + while True: if not listener.WaitForEvent(1, event): continue @@ -800,24 +891,22 @@ while True: signal.signal(signal.SIGQUIT, prev_handler) break elif state == lldb.eStateConnected: - process.RemoteLaunch(args, env, None, None, None, None, 0, False, err) - if not err.Success(): - sys.stderr.write("lldb: failed to launch remote process: %s\n" % (err)) - process.Kill() - debugger.Terminate() - sys.exit(1) - # Forward SIGQUIT to the program. - def signal_handler(signal, frame): - process.Signal(signal) - prev_handler = signal.signal(signal.SIGQUIT, signal_handler) - # Tell the Go driver that the program is running and should not be retried. - sys.stderr.write("lldb: running program\n") - running = True - # Process stops once at the beginning. Continue. - process.Continue() + if platform == 'remote-ios': + process.RemoteLaunch(args, env, None, None, None, None, 0, False, err) + if not err.Success(): + sys.stderr.write("lldb: failed to launch remote process: %s\n" % (err)) + process.Kill() + debugger.Terminate() + sys.exit(1) + run_program() exitStatus = process.GetExitStatus() +exitDesc = process.GetExitDescription() process.Kill() debugger.Terminate() +if exitStatus == 0 and exitDesc is not None: + # Ensure tests fail when killed by a signal. + exitStatus = 123 + sys.exit(exitStatus) ` diff --git a/misc/wasm/wasm_exec.js b/misc/wasm/wasm_exec.js index ef97c4e311d..82041e6bb90 100644 --- a/misc/wasm/wasm_exec.js +++ b/misc/wasm/wasm_exec.js @@ -102,7 +102,7 @@ } } - if (!global.crypto) { + if (!global.crypto && global.require) { const nodeCrypto = require("crypto"); global.crypto = { getRandomValues(b) { @@ -110,6 +110,9 @@ }, }; } + if (!global.crypto) { + throw new Error("global.crypto is not available, polyfill required (getRandomValues only)"); + } if (!global.performance) { global.performance = { @@ -120,13 +123,19 @@ }; } - if (!global.TextEncoder) { + if (!global.TextEncoder && global.require) { global.TextEncoder = require("util").TextEncoder; } + if (!global.TextEncoder) { + throw new Error("global.TextEncoder is not available, polyfill required"); + } - if (!global.TextDecoder) { + if (!global.TextDecoder && global.require) { global.TextDecoder = require("util").TextDecoder; } + if (!global.TextDecoder) { + throw new Error("global.TextDecoder is not available, polyfill required"); + } // End of polyfills for common API. @@ -255,6 +264,7 @@ // func wasmExit(code int32) "runtime.wasmExit": (sp) => { + sp >>>= 0; const code = this.mem.getInt32(sp + 8, true); this.exited = true; delete this._inst; @@ -267,6 +277,7 @@ // func wasmWrite(fd uintptr, p unsafe.Pointer, n int32) "runtime.wasmWrite": (sp) => { + sp >>>= 0; const fd = getInt64(sp + 8); const p = getInt64(sp + 16); const n = this.mem.getInt32(sp + 24, true); @@ -275,16 +286,19 @@ // func resetMemoryDataView() "runtime.resetMemoryDataView": (sp) => { + sp >>>= 0; this.mem = new DataView(this._inst.exports.mem.buffer); }, // func nanotime1() int64 "runtime.nanotime1": (sp) => { + sp >>>= 0; setInt64(sp + 8, (timeOrigin + performance.now()) * 1000000); }, // func walltime1() (sec int64, nsec int32) "runtime.walltime1": (sp) => { + sp >>>= 0; const msec = (new Date).getTime(); setInt64(sp + 8, msec / 1000); this.mem.setInt32(sp + 16, (msec % 1000) * 1000000, true); @@ -292,6 +306,7 @@ // func scheduleTimeoutEvent(delay int64) int32 "runtime.scheduleTimeoutEvent": (sp) => { + sp >>>= 0; const id = this._nextCallbackTimeoutID; this._nextCallbackTimeoutID++; this._scheduledTimeouts.set(id, setTimeout( @@ -311,6 +326,7 @@ // func clearTimeoutEvent(id int32) "runtime.clearTimeoutEvent": (sp) => { + sp >>>= 0; const id = this.mem.getInt32(sp + 8, true); clearTimeout(this._scheduledTimeouts.get(id)); this._scheduledTimeouts.delete(id); @@ -318,11 +334,13 @@ // func getRandomData(r []byte) "runtime.getRandomData": (sp) => { + sp >>>= 0; crypto.getRandomValues(loadSlice(sp + 8)); }, // func finalizeRef(v ref) "syscall/js.finalizeRef": (sp) => { + sp >>>= 0; const id = this.mem.getUint32(sp + 8, true); this._goRefCounts[id]--; if (this._goRefCounts[id] === 0) { @@ -335,44 +353,51 @@ // func stringVal(value string) ref "syscall/js.stringVal": (sp) => { + sp >>>= 0; storeValue(sp + 24, loadString(sp + 8)); }, // func valueGet(v ref, p string) ref "syscall/js.valueGet": (sp) => { + sp >>>= 0; const result = Reflect.get(loadValue(sp + 8), loadString(sp + 16)); - sp = this._inst.exports.getsp(); // see comment above + sp = this._inst.exports.getsp() >>> 0; // see comment above storeValue(sp + 32, result); }, // func valueSet(v ref, p string, x ref) "syscall/js.valueSet": (sp) => { + sp >>>= 0; Reflect.set(loadValue(sp + 8), loadString(sp + 16), loadValue(sp + 32)); }, // func valueDelete(v ref, p string) "syscall/js.valueDelete": (sp) => { + sp >>>= 0; Reflect.deleteProperty(loadValue(sp + 8), loadString(sp + 16)); }, // func valueIndex(v ref, i int) ref "syscall/js.valueIndex": (sp) => { + sp >>>= 0; storeValue(sp + 24, Reflect.get(loadValue(sp + 8), getInt64(sp + 16))); }, // valueSetIndex(v ref, i int, x ref) "syscall/js.valueSetIndex": (sp) => { + sp >>>= 0; Reflect.set(loadValue(sp + 8), getInt64(sp + 16), loadValue(sp + 24)); }, // func valueCall(v ref, m string, args []ref) (ref, bool) "syscall/js.valueCall": (sp) => { + sp >>>= 0; try { const v = loadValue(sp + 8); const m = Reflect.get(v, loadString(sp + 16)); const args = loadSliceOfValues(sp + 32); const result = Reflect.apply(m, v, args); - sp = this._inst.exports.getsp(); // see comment above + sp = this._inst.exports.getsp() >>> 0; // see comment above storeValue(sp + 56, result); this.mem.setUint8(sp + 64, 1); } catch (err) { @@ -383,11 +408,12 @@ // func valueInvoke(v ref, args []ref) (ref, bool) "syscall/js.valueInvoke": (sp) => { + sp >>>= 0; try { const v = loadValue(sp + 8); const args = loadSliceOfValues(sp + 16); const result = Reflect.apply(v, undefined, args); - sp = this._inst.exports.getsp(); // see comment above + sp = this._inst.exports.getsp() >>> 0; // see comment above storeValue(sp + 40, result); this.mem.setUint8(sp + 48, 1); } catch (err) { @@ -398,11 +424,12 @@ // func valueNew(v ref, args []ref) (ref, bool) "syscall/js.valueNew": (sp) => { + sp >>>= 0; try { const v = loadValue(sp + 8); const args = loadSliceOfValues(sp + 16); const result = Reflect.construct(v, args); - sp = this._inst.exports.getsp(); // see comment above + sp = this._inst.exports.getsp() >>> 0; // see comment above storeValue(sp + 40, result); this.mem.setUint8(sp + 48, 1); } catch (err) { @@ -413,11 +440,13 @@ // func valueLength(v ref) int "syscall/js.valueLength": (sp) => { + sp >>>= 0; setInt64(sp + 16, parseInt(loadValue(sp + 8).length)); }, // valuePrepareString(v ref) (ref, int) "syscall/js.valuePrepareString": (sp) => { + sp >>>= 0; const str = encoder.encode(String(loadValue(sp + 8))); storeValue(sp + 16, str); setInt64(sp + 24, str.length); @@ -425,17 +454,20 @@ // valueLoadString(v ref, b []byte) "syscall/js.valueLoadString": (sp) => { + sp >>>= 0; const str = loadValue(sp + 8); loadSlice(sp + 16).set(str); }, // func valueInstanceOf(v ref, t ref) bool "syscall/js.valueInstanceOf": (sp) => { + sp >>>= 0; this.mem.setUint8(sp + 24, (loadValue(sp + 8) instanceof loadValue(sp + 16)) ? 1 : 0); }, // func copyBytesToGo(dst []byte, src ref) (int, bool) "syscall/js.copyBytesToGo": (sp) => { + sp >>>= 0; const dst = loadSlice(sp + 8); const src = loadValue(sp + 32); if (!(src instanceof Uint8Array || src instanceof Uint8ClampedArray)) { @@ -450,6 +482,7 @@ // func copyBytesToJS(dst ref, src []byte) (int, bool) "syscall/js.copyBytesToJS": (sp) => { + sp >>>= 0; const dst = loadValue(sp + 8); const src = loadSlice(sp + 16); if (!(dst instanceof Uint8Array || dst instanceof Uint8ClampedArray)) { @@ -470,6 +503,9 @@ } async run(instance) { + if (!(instance instanceof WebAssembly.Instance)) { + throw new Error("Go.run: WebAssembly.Instance expected"); + } this._inst = instance; this.mem = new DataView(this._inst.exports.mem.buffer); this._values = [ // JS values that Go currently has references to, indexed by reference id diff --git a/src/archive/tar/common.go b/src/archive/tar/common.go index dee9e47e4ae..c667cfc8720 100644 --- a/src/archive/tar/common.go +++ b/src/archive/tar/common.go @@ -13,8 +13,8 @@ package tar import ( "errors" "fmt" + "io/fs" "math" - "os" "path" "reflect" "strconv" @@ -525,12 +525,12 @@ func (h Header) allowedFormats() (format Format, paxHdrs map[string]string, err return format, paxHdrs, err } -// FileInfo returns an os.FileInfo for the Header. -func (h *Header) FileInfo() os.FileInfo { +// FileInfo returns an fs.FileInfo for the Header. +func (h *Header) FileInfo() fs.FileInfo { return headerFileInfo{h} } -// headerFileInfo implements os.FileInfo. +// headerFileInfo implements fs.FileInfo. type headerFileInfo struct { h *Header } @@ -549,57 +549,57 @@ func (fi headerFileInfo) Name() string { } // Mode returns the permission and mode bits for the headerFileInfo. -func (fi headerFileInfo) Mode() (mode os.FileMode) { +func (fi headerFileInfo) Mode() (mode fs.FileMode) { // Set file permission bits. - mode = os.FileMode(fi.h.Mode).Perm() + mode = fs.FileMode(fi.h.Mode).Perm() // Set setuid, setgid and sticky bits. if fi.h.Mode&c_ISUID != 0 { - mode |= os.ModeSetuid + mode |= fs.ModeSetuid } if fi.h.Mode&c_ISGID != 0 { - mode |= os.ModeSetgid + mode |= fs.ModeSetgid } if fi.h.Mode&c_ISVTX != 0 { - mode |= os.ModeSticky + mode |= fs.ModeSticky } // Set file mode bits; clear perm, setuid, setgid, and sticky bits. - switch m := os.FileMode(fi.h.Mode) &^ 07777; m { + switch m := fs.FileMode(fi.h.Mode) &^ 07777; m { case c_ISDIR: - mode |= os.ModeDir + mode |= fs.ModeDir case c_ISFIFO: - mode |= os.ModeNamedPipe + mode |= fs.ModeNamedPipe case c_ISLNK: - mode |= os.ModeSymlink + mode |= fs.ModeSymlink case c_ISBLK: - mode |= os.ModeDevice + mode |= fs.ModeDevice case c_ISCHR: - mode |= os.ModeDevice - mode |= os.ModeCharDevice + mode |= fs.ModeDevice + mode |= fs.ModeCharDevice case c_ISSOCK: - mode |= os.ModeSocket + mode |= fs.ModeSocket } switch fi.h.Typeflag { case TypeSymlink: - mode |= os.ModeSymlink + mode |= fs.ModeSymlink case TypeChar: - mode |= os.ModeDevice - mode |= os.ModeCharDevice + mode |= fs.ModeDevice + mode |= fs.ModeCharDevice case TypeBlock: - mode |= os.ModeDevice + mode |= fs.ModeDevice case TypeDir: - mode |= os.ModeDir + mode |= fs.ModeDir case TypeFifo: - mode |= os.ModeNamedPipe + mode |= fs.ModeNamedPipe } return mode } // sysStat, if non-nil, populates h from system-dependent fields of fi. -var sysStat func(fi os.FileInfo, h *Header) error +var sysStat func(fi fs.FileInfo, h *Header) error const ( // Mode constants from the USTAR spec: @@ -623,10 +623,10 @@ const ( // If fi describes a symlink, FileInfoHeader records link as the link target. // If fi describes a directory, a slash is appended to the name. // -// Since os.FileInfo's Name method only returns the base name of +// Since fs.FileInfo's Name method only returns the base name of // the file it describes, it may be necessary to modify Header.Name // to provide the full path name of the file. -func FileInfoHeader(fi os.FileInfo, link string) (*Header, error) { +func FileInfoHeader(fi fs.FileInfo, link string) (*Header, error) { if fi == nil { return nil, errors.New("archive/tar: FileInfo is nil") } @@ -643,29 +643,29 @@ func FileInfoHeader(fi os.FileInfo, link string) (*Header, error) { case fi.IsDir(): h.Typeflag = TypeDir h.Name += "/" - case fm&os.ModeSymlink != 0: + case fm&fs.ModeSymlink != 0: h.Typeflag = TypeSymlink h.Linkname = link - case fm&os.ModeDevice != 0: - if fm&os.ModeCharDevice != 0 { + case fm&fs.ModeDevice != 0: + if fm&fs.ModeCharDevice != 0 { h.Typeflag = TypeChar } else { h.Typeflag = TypeBlock } - case fm&os.ModeNamedPipe != 0: + case fm&fs.ModeNamedPipe != 0: h.Typeflag = TypeFifo - case fm&os.ModeSocket != 0: + case fm&fs.ModeSocket != 0: return nil, fmt.Errorf("archive/tar: sockets not supported") default: return nil, fmt.Errorf("archive/tar: unknown file mode %v", fm) } - if fm&os.ModeSetuid != 0 { + if fm&fs.ModeSetuid != 0 { h.Mode |= c_ISUID } - if fm&os.ModeSetgid != 0 { + if fm&fs.ModeSetgid != 0 { h.Mode |= c_ISGID } - if fm&os.ModeSticky != 0 { + if fm&fs.ModeSticky != 0 { h.Mode |= c_ISVTX } // If possible, populate additional fields from OS-specific diff --git a/src/archive/tar/reader.go b/src/archive/tar/reader.go index 4f9135b7919..1b1d5b46891 100644 --- a/src/archive/tar/reader.go +++ b/src/archive/tar/reader.go @@ -7,7 +7,6 @@ package tar import ( "bytes" "io" - "io/ioutil" "strconv" "strings" "time" @@ -104,7 +103,7 @@ func (tr *Reader) next() (*Header, error) { continue // This is a meta header affecting the next header case TypeGNULongName, TypeGNULongLink: format.mayOnlyBe(FormatGNU) - realname, err := ioutil.ReadAll(tr) + realname, err := io.ReadAll(tr) if err != nil { return nil, err } @@ -294,7 +293,7 @@ func mergePAX(hdr *Header, paxHdrs map[string]string) (err error) { // parsePAX parses PAX headers. // If an extended header (type 'x') is invalid, ErrHeader is returned func parsePAX(r io.Reader) (map[string]string, error) { - buf, err := ioutil.ReadAll(r) + buf, err := io.ReadAll(r) if err != nil { return nil, err } @@ -850,7 +849,7 @@ func discard(r io.Reader, n int64) error { } } - copySkipped, err := io.CopyN(ioutil.Discard, r, n-seekSkipped) + copySkipped, err := io.CopyN(io.Discard, r, n-seekSkipped) if err == io.EOF && seekSkipped+copySkipped < n { err = io.ErrUnexpectedEOF } diff --git a/src/archive/tar/reader_test.go b/src/archive/tar/reader_test.go index f153b668def..789ddc1bc03 100644 --- a/src/archive/tar/reader_test.go +++ b/src/archive/tar/reader_test.go @@ -10,7 +10,6 @@ import ( "errors" "fmt" "io" - "io/ioutil" "math" "os" "path" @@ -773,7 +772,7 @@ func TestReadTruncation(t *testing.T) { "testdata/pax-path-hdr.tar", "testdata/sparse-formats.tar", } { - buf, err := ioutil.ReadFile(p) + buf, err := os.ReadFile(p) if err != nil { t.Fatalf("unexpected error: %v", err) } @@ -865,7 +864,7 @@ func TestReadTruncation(t *testing.T) { } cnt++ if s2 == "manual" { - if _, err = tr.writeTo(ioutil.Discard); err != nil { + if _, err = tr.writeTo(io.Discard); err != nil { break } } diff --git a/src/archive/tar/stat_unix.go b/src/archive/tar/stat_unix.go index d1576db41d5..581d87dca9d 100644 --- a/src/archive/tar/stat_unix.go +++ b/src/archive/tar/stat_unix.go @@ -7,7 +7,7 @@ package tar import ( - "os" + "io/fs" "os/user" "runtime" "strconv" @@ -23,7 +23,7 @@ func init() { // The downside is that renaming uname or gname by the OS never takes effect. var userMap, groupMap sync.Map // map[int]string -func statUnix(fi os.FileInfo, h *Header) error { +func statUnix(fi fs.FileInfo, h *Header) error { sys, ok := fi.Sys().(*syscall.Stat_t) if !ok { return nil @@ -66,7 +66,7 @@ func statUnix(fi os.FileInfo, h *Header) error { minor := uint32((dev & 0x00000000000000ff) >> 0) minor |= uint32((dev & 0x00000ffffff00000) >> 12) h.Devmajor, h.Devminor = int64(major), int64(minor) - case "darwin": + case "darwin", "ios": // Copied from golang.org/x/sys/unix/dev_darwin.go. major := uint32((dev >> 24) & 0xff) minor := uint32(dev & 0xffffff) diff --git a/src/archive/tar/tar_test.go b/src/archive/tar/tar_test.go index 2676853122a..91b38401b6c 100644 --- a/src/archive/tar/tar_test.go +++ b/src/archive/tar/tar_test.go @@ -10,7 +10,7 @@ import ( "fmt" "internal/testenv" "io" - "io/ioutil" + "io/fs" "math" "os" "path" @@ -262,7 +262,7 @@ func TestFileInfoHeaderDir(t *testing.T) { func TestFileInfoHeaderSymlink(t *testing.T) { testenv.MustHaveSymlink(t) - tmpdir, err := ioutil.TempDir("", "TestFileInfoHeaderSymlink") + tmpdir, err := os.MkdirTemp("", "TestFileInfoHeaderSymlink") if err != nil { t.Fatal(err) } @@ -327,7 +327,7 @@ func TestRoundTrip(t *testing.T) { if !reflect.DeepEqual(rHdr, hdr) { t.Errorf("Header mismatch.\n got %+v\nwant %+v", rHdr, hdr) } - rData, err := ioutil.ReadAll(tr) + rData, err := io.ReadAll(tr) if err != nil { t.Fatalf("Read: %v", err) } @@ -338,7 +338,7 @@ func TestRoundTrip(t *testing.T) { type headerRoundTripTest struct { h *Header - fm os.FileMode + fm fs.FileMode } func TestHeaderRoundTrip(t *testing.T) { @@ -361,7 +361,7 @@ func TestHeaderRoundTrip(t *testing.T) { ModTime: time.Unix(1360600852, 0), Typeflag: TypeSymlink, }, - fm: 0777 | os.ModeSymlink, + fm: 0777 | fs.ModeSymlink, }, { // character device node. h: &Header{ @@ -371,7 +371,7 @@ func TestHeaderRoundTrip(t *testing.T) { ModTime: time.Unix(1360578951, 0), Typeflag: TypeChar, }, - fm: 0666 | os.ModeDevice | os.ModeCharDevice, + fm: 0666 | fs.ModeDevice | fs.ModeCharDevice, }, { // block device node. h: &Header{ @@ -381,7 +381,7 @@ func TestHeaderRoundTrip(t *testing.T) { ModTime: time.Unix(1360578954, 0), Typeflag: TypeBlock, }, - fm: 0660 | os.ModeDevice, + fm: 0660 | fs.ModeDevice, }, { // directory. h: &Header{ @@ -391,7 +391,7 @@ func TestHeaderRoundTrip(t *testing.T) { ModTime: time.Unix(1360601116, 0), Typeflag: TypeDir, }, - fm: 0755 | os.ModeDir, + fm: 0755 | fs.ModeDir, }, { // fifo node. h: &Header{ @@ -401,7 +401,7 @@ func TestHeaderRoundTrip(t *testing.T) { ModTime: time.Unix(1360578949, 0), Typeflag: TypeFifo, }, - fm: 0600 | os.ModeNamedPipe, + fm: 0600 | fs.ModeNamedPipe, }, { // setuid. h: &Header{ @@ -411,7 +411,7 @@ func TestHeaderRoundTrip(t *testing.T) { ModTime: time.Unix(1355405093, 0), Typeflag: TypeReg, }, - fm: 0755 | os.ModeSetuid, + fm: 0755 | fs.ModeSetuid, }, { // setguid. h: &Header{ @@ -421,7 +421,7 @@ func TestHeaderRoundTrip(t *testing.T) { ModTime: time.Unix(1360602346, 0), Typeflag: TypeReg, }, - fm: 0750 | os.ModeSetgid, + fm: 0750 | fs.ModeSetgid, }, { // sticky. h: &Header{ @@ -431,7 +431,7 @@ func TestHeaderRoundTrip(t *testing.T) { ModTime: time.Unix(1360602540, 0), Typeflag: TypeReg, }, - fm: 0600 | os.ModeSticky, + fm: 0600 | fs.ModeSticky, }, { // hard link. h: &Header{ @@ -804,9 +804,9 @@ func Benchmark(b *testing.B) { b.Run(v.label, func(b *testing.B) { b.ReportAllocs() for i := 0; i < b.N; i++ { - // Writing to ioutil.Discard because we want to + // Writing to io.Discard because we want to // test purely the writer code and not bring in disk performance into this. - tw := NewWriter(ioutil.Discard) + tw := NewWriter(io.Discard) for _, file := range v.files { if err := tw.WriteHeader(file.hdr); err != nil { b.Errorf("unexpected WriteHeader error: %v", err) @@ -844,7 +844,7 @@ func Benchmark(b *testing.B) { if _, err := tr.Next(); err != nil { b.Errorf("unexpected Next error: %v", err) } - if _, err := io.Copy(ioutil.Discard, tr); err != nil { + if _, err := io.Copy(io.Discard, tr); err != nil { b.Errorf("unexpected Copy error : %v", err) } } diff --git a/src/archive/tar/writer_test.go b/src/archive/tar/writer_test.go index 30556d27d02..a00f02d8fab 100644 --- a/src/archive/tar/writer_test.go +++ b/src/archive/tar/writer_test.go @@ -9,7 +9,6 @@ import ( "encoding/hex" "errors" "io" - "io/ioutil" "os" "path" "reflect" @@ -520,7 +519,7 @@ func TestWriter(t *testing.T) { } if v.file != "" { - want, err := ioutil.ReadFile(v.file) + want, err := os.ReadFile(v.file) if err != nil { t.Fatalf("ReadFile() = %v, want nil", err) } diff --git a/src/archive/zip/reader.go b/src/archive/zip/reader.go index 13ff9ddcf42..8b4e77875fb 100644 --- a/src/archive/zip/reader.go +++ b/src/archive/zip/reader.go @@ -11,7 +11,12 @@ import ( "hash" "hash/crc32" "io" + "io/fs" "os" + "path" + "sort" + "strings" + "sync" "time" ) @@ -21,18 +26,28 @@ var ( ErrChecksum = errors.New("zip: checksum error") ) +// A Reader serves content from a ZIP archive. type Reader struct { r io.ReaderAt File []*File Comment string decompressors map[uint16]Decompressor + + // fileList is a list of files sorted by ename, + // for use by the Open method. + fileListOnce sync.Once + fileList []fileListEntry } +// A ReadCloser is a Reader that must be closed when no longer needed. type ReadCloser struct { f *os.File Reader } +// A File is a single file in a ZIP archive. +// The file information is in the embedded FileHeader. +// The file content can be accessed by calling Open. type File struct { FileHeader zip *Reader @@ -187,6 +202,10 @@ type checksumReader struct { err error // sticky error } +func (r *checksumReader) Stat() (fs.FileInfo, error) { + return headerFileInfo{&r.f.FileHeader}, nil +} + func (r *checksumReader) Read(b []byte) (n int, err error) { if r.err != nil { return 0, r.err @@ -607,3 +626,173 @@ func (b *readBuf) sub(n int) readBuf { *b = (*b)[n:] return b2 } + +// A fileListEntry is a File and its ename. +// If file == nil, the fileListEntry describes a directory, without metadata. +type fileListEntry struct { + name string + file *File // nil for directories +} + +type fileInfoDirEntry interface { + fs.FileInfo + fs.DirEntry +} + +func (e *fileListEntry) stat() fileInfoDirEntry { + if e.file != nil { + return headerFileInfo{&e.file.FileHeader} + } + return e +} + +// Only used for directories. +func (f *fileListEntry) Name() string { _, elem, _ := split(f.name); return elem } +func (f *fileListEntry) Size() int64 { return 0 } +func (f *fileListEntry) ModTime() time.Time { return time.Time{} } +func (f *fileListEntry) Mode() fs.FileMode { return fs.ModeDir | 0555 } +func (f *fileListEntry) Type() fs.FileMode { return fs.ModeDir } +func (f *fileListEntry) IsDir() bool { return true } +func (f *fileListEntry) Sys() interface{} { return nil } + +func (f *fileListEntry) Info() (fs.FileInfo, error) { return f, nil } + +// toValidName coerces name to be a valid name for fs.FS.Open. +func toValidName(name string) string { + name = strings.ReplaceAll(name, `\`, `/`) + p := path.Clean(name) + if strings.HasPrefix(p, "/") { + p = p[len("/"):] + } + for strings.HasPrefix(name, "../") { + p = p[len("../"):] + } + return p +} + +func (r *Reader) initFileList() { + r.fileListOnce.Do(func() { + dirs := make(map[string]bool) + for _, file := range r.File { + name := toValidName(file.Name) + for dir := path.Dir(name); dir != "."; dir = path.Dir(dir) { + dirs[dir] = true + } + r.fileList = append(r.fileList, fileListEntry{name, file}) + } + for dir := range dirs { + r.fileList = append(r.fileList, fileListEntry{dir + "/", nil}) + } + + sort.Slice(r.fileList, func(i, j int) bool { return fileEntryLess(r.fileList[i].name, r.fileList[j].name) }) + }) +} + +func fileEntryLess(x, y string) bool { + xdir, xelem, _ := split(x) + ydir, yelem, _ := split(y) + return xdir < ydir || xdir == ydir && xelem < yelem +} + +// Open opens the named file in the ZIP archive, +// using the semantics of fs.FS.Open: +// paths are always slash separated, with no +// leading / or ../ elements. +func (r *Reader) Open(name string) (fs.File, error) { + r.initFileList() + + e := r.openLookup(name) + if e == nil || !fs.ValidPath(name) { + return nil, &fs.PathError{Op: "open", Path: name, Err: fs.ErrNotExist} + } + if e.file == nil || strings.HasSuffix(e.file.Name, "/") { + return &openDir{e, r.openReadDir(name), 0}, nil + } + rc, err := e.file.Open() + if err != nil { + return nil, err + } + return rc.(fs.File), nil +} + +func split(name string) (dir, elem string, isDir bool) { + if name[len(name)-1] == '/' { + isDir = true + name = name[:len(name)-1] + } + i := len(name) - 1 + for i >= 0 && name[i] != '/' { + i-- + } + if i < 0 { + return ".", name, isDir + } + return name[:i], name[i+1:], isDir +} + +var dotFile = &fileListEntry{name: "./"} + +func (r *Reader) openLookup(name string) *fileListEntry { + if name == "." { + return dotFile + } + + dir, elem, _ := split(name) + files := r.fileList + i := sort.Search(len(files), func(i int) bool { + idir, ielem, _ := split(files[i].name) + return idir > dir || idir == dir && ielem >= elem + }) + if i < len(files) { + fname := files[i].name + if fname == name || len(fname) == len(name)+1 && fname[len(name)] == '/' && fname[:len(name)] == name { + return &files[i] + } + } + return nil +} + +func (r *Reader) openReadDir(dir string) []fileListEntry { + files := r.fileList + i := sort.Search(len(files), func(i int) bool { + idir, _, _ := split(files[i].name) + return idir >= dir + }) + j := sort.Search(len(files), func(j int) bool { + jdir, _, _ := split(files[j].name) + return jdir > dir + }) + return files[i:j] +} + +type openDir struct { + e *fileListEntry + files []fileListEntry + offset int +} + +func (d *openDir) Close() error { return nil } +func (d *openDir) Stat() (fs.FileInfo, error) { return d.e.stat(), nil } + +func (d *openDir) Read([]byte) (int, error) { + return 0, &fs.PathError{Op: "read", Path: d.e.name, Err: errors.New("is a directory")} +} + +func (d *openDir) ReadDir(count int) ([]fs.DirEntry, error) { + n := len(d.files) - d.offset + if count > 0 && n > count { + n = count + } + if n == 0 { + if count <= 0 { + return nil, nil + } + return nil, io.EOF + } + list := make([]fs.DirEntry, n) + for i := range list { + list[i] = d.files[d.offset+i].stat() + } + d.offset += n + return list, nil +} diff --git a/src/archive/zip/reader_test.go b/src/archive/zip/reader_test.go index adca87a8b38..34e96f7da43 100644 --- a/src/archive/zip/reader_test.go +++ b/src/archive/zip/reader_test.go @@ -10,12 +10,13 @@ import ( "encoding/hex" "internal/obscuretestdata" "io" - "io/ioutil" + "io/fs" "os" "path/filepath" "regexp" "strings" "testing" + "testing/fstest" "time" ) @@ -30,7 +31,7 @@ type ZipTest struct { type ZipTestFile struct { Name string - Mode os.FileMode + Mode fs.FileMode NonUTF8 bool ModTime time.Time Modified time.Time @@ -107,7 +108,7 @@ var tests = []ZipTest{ Name: "symlink", Content: []byte("../target"), Modified: time.Date(2012, 2, 3, 19, 56, 48, 0, timeZone(-2*time.Hour)), - Mode: 0777 | os.ModeSymlink, + Mode: 0777 | fs.ModeSymlink, }, }, }, @@ -149,7 +150,7 @@ var tests = []ZipTest{ Name: "dir/empty/", Content: []byte{}, Modified: time.Date(2011, 12, 8, 10, 8, 6, 0, time.UTC), - Mode: os.ModeDir | 0777, + Mode: fs.ModeDir | 0777, }, { Name: "readonly", @@ -179,7 +180,7 @@ var tests = []ZipTest{ Name: "dir/empty/", Content: []byte{}, Modified: time.Date(2011, 12, 8, 10, 8, 6, 0, timeZone(0)), - Mode: os.ModeDir | 0777, + Mode: fs.ModeDir | 0777, }, { Name: "readonly", @@ -627,7 +628,7 @@ func readTestFile(t *testing.T, zt ZipTest, ft ZipTestFile, f *File) { var c []byte if ft.Content != nil { c = ft.Content - } else if c, err = ioutil.ReadFile("testdata/" + ft.File); err != nil { + } else if c, err = os.ReadFile("testdata/" + ft.File); err != nil { t.Error(err) return } @@ -645,7 +646,7 @@ func readTestFile(t *testing.T, zt ZipTest, ft ZipTestFile, f *File) { } } -func testFileMode(t *testing.T, f *File, want os.FileMode) { +func testFileMode(t *testing.T, f *File, want fs.FileMode) { mode := f.Mode() if want == 0 { t.Errorf("%s mode: got %v, want none", f.Name, mode) @@ -683,7 +684,7 @@ func TestInvalidFiles(t *testing.T) { } func messWith(fileName string, corrupter func(b []byte)) (r io.ReaderAt, size int64) { - data, err := ioutil.ReadFile(filepath.Join("testdata", fileName)) + data, err := os.ReadFile(filepath.Join("testdata", fileName)) if err != nil { panic("Error reading " + fileName + ": " + err.Error()) } @@ -790,17 +791,17 @@ func returnRecursiveZip() (r io.ReaderAt, size int64) { // // func main() { // bigZip := makeZip("big.file", io.LimitReader(zeros{}, 1<<32-1)) -// if err := ioutil.WriteFile("/tmp/big.zip", bigZip, 0666); err != nil { +// if err := os.WriteFile("/tmp/big.zip", bigZip, 0666); err != nil { // log.Fatal(err) // } // // biggerZip := makeZip("big.zip", bytes.NewReader(bigZip)) -// if err := ioutil.WriteFile("/tmp/bigger.zip", biggerZip, 0666); err != nil { +// if err := os.WriteFile("/tmp/bigger.zip", biggerZip, 0666); err != nil { // log.Fatal(err) // } // // biggestZip := makeZip("bigger.zip", bytes.NewReader(biggerZip)) -// if err := ioutil.WriteFile("/tmp/biggest.zip", biggestZip, 0666); err != nil { +// if err := os.WriteFile("/tmp/biggest.zip", biggestZip, 0666); err != nil { // log.Fatal(err) // } // } @@ -928,7 +929,7 @@ func returnBigZipBytes() (r io.ReaderAt, size int64) { if err != nil { panic(err) } - b, err = ioutil.ReadAll(f) + b, err = io.ReadAll(f) if err != nil { panic(err) } @@ -985,7 +986,7 @@ func TestIssue10957(t *testing.T) { continue } if f.UncompressedSize64 < 1e6 { - n, err := io.Copy(ioutil.Discard, r) + n, err := io.Copy(io.Discard, r) if i == 3 && err != io.ErrUnexpectedEOF { t.Errorf("File[3] error = %v; want io.ErrUnexpectedEOF", err) } @@ -1027,7 +1028,7 @@ func TestIssue11146(t *testing.T) { if err != nil { t.Fatal(err) } - _, err = ioutil.ReadAll(r) + _, err = io.ReadAll(r) if err != io.ErrUnexpectedEOF { t.Errorf("File[0] error = %v; want io.ErrUnexpectedEOF", err) } @@ -1070,3 +1071,13 @@ func TestIssue12449(t *testing.T) { t.Errorf("Error reading the archive: %v", err) } } + +func TestFS(t *testing.T) { + z, err := OpenReader("testdata/unix.zip") + if err != nil { + t.Fatal(err) + } + if err := fstest.TestFS(z, "hello", "dir/bar", "dir/empty", "readonly"); err != nil { + t.Fatal(err) + } +} diff --git a/src/archive/zip/register.go b/src/archive/zip/register.go index 51e9c3e4d4b..4389246286f 100644 --- a/src/archive/zip/register.go +++ b/src/archive/zip/register.go @@ -8,7 +8,6 @@ import ( "compress/flate" "errors" "io" - "io/ioutil" "sync" ) @@ -111,7 +110,7 @@ func init() { compressors.Store(Store, Compressor(func(w io.Writer) (io.WriteCloser, error) { return &nopCloser{w}, nil })) compressors.Store(Deflate, Compressor(func(w io.Writer) (io.WriteCloser, error) { return newFlateWriter(w), nil })) - decompressors.Store(Store, Decompressor(ioutil.NopCloser)) + decompressors.Store(Store, Decompressor(io.NopCloser)) decompressors.Store(Deflate, Decompressor(newFlateReader)) } diff --git a/src/archive/zip/struct.go b/src/archive/zip/struct.go index 686e79781ac..4dd29f35fa6 100644 --- a/src/archive/zip/struct.go +++ b/src/archive/zip/struct.go @@ -20,7 +20,7 @@ fields must be used instead. package zip import ( - "os" + "io/fs" "path" "time" ) @@ -137,12 +137,12 @@ type FileHeader struct { ExternalAttrs uint32 // Meaning depends on CreatorVersion } -// FileInfo returns an os.FileInfo for the FileHeader. -func (h *FileHeader) FileInfo() os.FileInfo { +// FileInfo returns an fs.FileInfo for the FileHeader. +func (h *FileHeader) FileInfo() fs.FileInfo { return headerFileInfo{h} } -// headerFileInfo implements os.FileInfo. +// headerFileInfo implements fs.FileInfo. type headerFileInfo struct { fh *FileHeader } @@ -161,17 +161,20 @@ func (fi headerFileInfo) ModTime() time.Time { } return fi.fh.Modified.UTC() } -func (fi headerFileInfo) Mode() os.FileMode { return fi.fh.Mode() } +func (fi headerFileInfo) Mode() fs.FileMode { return fi.fh.Mode() } +func (fi headerFileInfo) Type() fs.FileMode { return fi.fh.Mode().Type() } func (fi headerFileInfo) Sys() interface{} { return fi.fh } +func (fi headerFileInfo) Info() (fs.FileInfo, error) { return fi, nil } + // FileInfoHeader creates a partially-populated FileHeader from an -// os.FileInfo. -// Because os.FileInfo's Name method returns only the base name of +// fs.FileInfo. +// Because fs.FileInfo's Name method returns only the base name of // the file it describes, it may be necessary to modify the Name field // of the returned header to provide the full path name of the file. // If compression is desired, callers should set the FileHeader.Method // field; it is unset by default. -func FileInfoHeader(fi os.FileInfo) (*FileHeader, error) { +func FileInfoHeader(fi fs.FileInfo) (*FileHeader, error) { size := fi.Size() fh := &FileHeader{ Name: fi.Name(), @@ -280,7 +283,7 @@ const ( ) // Mode returns the permission and mode bits for the FileHeader. -func (h *FileHeader) Mode() (mode os.FileMode) { +func (h *FileHeader) Mode() (mode fs.FileMode) { switch h.CreatorVersion >> 8 { case creatorUnix, creatorMacOSX: mode = unixModeToFileMode(h.ExternalAttrs >> 16) @@ -288,18 +291,18 @@ func (h *FileHeader) Mode() (mode os.FileMode) { mode = msdosModeToFileMode(h.ExternalAttrs) } if len(h.Name) > 0 && h.Name[len(h.Name)-1] == '/' { - mode |= os.ModeDir + mode |= fs.ModeDir } return mode } // SetMode changes the permission and mode bits for the FileHeader. -func (h *FileHeader) SetMode(mode os.FileMode) { +func (h *FileHeader) SetMode(mode fs.FileMode) { h.CreatorVersion = h.CreatorVersion&0xff | creatorUnix<<8 h.ExternalAttrs = fileModeToUnixMode(mode) << 16 // set MSDOS attributes too, as the original zip does. - if mode&os.ModeDir != 0 { + if mode&fs.ModeDir != 0 { h.ExternalAttrs |= msdosDir } if mode&0200 == 0 { @@ -312,9 +315,9 @@ func (h *FileHeader) isZip64() bool { return h.CompressedSize64 >= uint32max || h.UncompressedSize64 >= uint32max } -func msdosModeToFileMode(m uint32) (mode os.FileMode) { +func msdosModeToFileMode(m uint32) (mode fs.FileMode) { if m&msdosDir != 0 { - mode = os.ModeDir | 0777 + mode = fs.ModeDir | 0777 } else { mode = 0666 } @@ -324,64 +327,64 @@ func msdosModeToFileMode(m uint32) (mode os.FileMode) { return mode } -func fileModeToUnixMode(mode os.FileMode) uint32 { +func fileModeToUnixMode(mode fs.FileMode) uint32 { var m uint32 - switch mode & os.ModeType { + switch mode & fs.ModeType { default: m = s_IFREG - case os.ModeDir: + case fs.ModeDir: m = s_IFDIR - case os.ModeSymlink: + case fs.ModeSymlink: m = s_IFLNK - case os.ModeNamedPipe: + case fs.ModeNamedPipe: m = s_IFIFO - case os.ModeSocket: + case fs.ModeSocket: m = s_IFSOCK - case os.ModeDevice: - if mode&os.ModeCharDevice != 0 { + case fs.ModeDevice: + if mode&fs.ModeCharDevice != 0 { m = s_IFCHR } else { m = s_IFBLK } } - if mode&os.ModeSetuid != 0 { + if mode&fs.ModeSetuid != 0 { m |= s_ISUID } - if mode&os.ModeSetgid != 0 { + if mode&fs.ModeSetgid != 0 { m |= s_ISGID } - if mode&os.ModeSticky != 0 { + if mode&fs.ModeSticky != 0 { m |= s_ISVTX } return m | uint32(mode&0777) } -func unixModeToFileMode(m uint32) os.FileMode { - mode := os.FileMode(m & 0777) +func unixModeToFileMode(m uint32) fs.FileMode { + mode := fs.FileMode(m & 0777) switch m & s_IFMT { case s_IFBLK: - mode |= os.ModeDevice + mode |= fs.ModeDevice case s_IFCHR: - mode |= os.ModeDevice | os.ModeCharDevice + mode |= fs.ModeDevice | fs.ModeCharDevice case s_IFDIR: - mode |= os.ModeDir + mode |= fs.ModeDir case s_IFIFO: - mode |= os.ModeNamedPipe + mode |= fs.ModeNamedPipe case s_IFLNK: - mode |= os.ModeSymlink + mode |= fs.ModeSymlink case s_IFREG: // nothing to do case s_IFSOCK: - mode |= os.ModeSocket + mode |= fs.ModeSocket } if m&s_ISGID != 0 { - mode |= os.ModeSetgid + mode |= fs.ModeSetgid } if m&s_ISUID != 0 { - mode |= os.ModeSetuid + mode |= fs.ModeSetuid } if m&s_ISVTX != 0 { - mode |= os.ModeSticky + mode |= fs.ModeSticky } return mode } diff --git a/src/archive/zip/writer_test.go b/src/archive/zip/writer_test.go index 1fedfd85e82..5985144e5c2 100644 --- a/src/archive/zip/writer_test.go +++ b/src/archive/zip/writer_test.go @@ -9,7 +9,7 @@ import ( "encoding/binary" "fmt" "io" - "io/ioutil" + "io/fs" "math/rand" "os" "strings" @@ -23,7 +23,7 @@ type WriteTest struct { Name string Data []byte Method uint16 - Mode os.FileMode + Mode fs.FileMode } var writeTests = []WriteTest{ @@ -43,19 +43,19 @@ var writeTests = []WriteTest{ Name: "setuid", Data: []byte("setuid file"), Method: Deflate, - Mode: 0755 | os.ModeSetuid, + Mode: 0755 | fs.ModeSetuid, }, { Name: "setgid", Data: []byte("setgid file"), Method: Deflate, - Mode: 0755 | os.ModeSetgid, + Mode: 0755 | fs.ModeSetgid, }, { Name: "symlink", Data: []byte("../link/target"), Method: Deflate, - Mode: 0755 | os.ModeSymlink, + Mode: 0755 | fs.ModeSymlink, }, } @@ -237,7 +237,7 @@ func TestWriterTime(t *testing.T) { t.Fatalf("unexpected Close error: %v", err) } - want, err := ioutil.ReadFile("testdata/time-go.zip") + want, err := os.ReadFile("testdata/time-go.zip") if err != nil { t.Fatalf("unexpected ReadFile error: %v", err) } @@ -301,7 +301,7 @@ func TestWriterFlush(t *testing.T) { } func TestWriterDir(t *testing.T) { - w := NewWriter(ioutil.Discard) + w := NewWriter(io.Discard) dw, err := w.Create("dir/") if err != nil { t.Fatal(err) @@ -380,7 +380,7 @@ func testReadFile(t *testing.T, f *File, wt *WriteTest) { if err != nil { t.Fatal("opening:", err) } - b, err := ioutil.ReadAll(rc) + b, err := io.ReadAll(rc) if err != nil { t.Fatal("reading:", err) } diff --git a/src/archive/zip/zip_test.go b/src/archive/zip/zip_test.go index b3a7caac7f2..ead9cd3aab9 100644 --- a/src/archive/zip/zip_test.go +++ b/src/archive/zip/zip_test.go @@ -13,7 +13,6 @@ import ( "hash" "internal/testenv" "io" - "io/ioutil" "runtime" "sort" "strings" @@ -620,7 +619,7 @@ func testZip64(t testing.TB, size int64) *rleBuffer { t.Fatal("read:", err) } } - gotEnd, err := ioutil.ReadAll(rc) + gotEnd, err := io.ReadAll(rc) if err != nil { t.Fatal("read end:", err) } diff --git a/src/bufio/bufio_test.go b/src/bufio/bufio_test.go index cb68f3ba239..d7b34bd0d8f 100644 --- a/src/bufio/bufio_test.go +++ b/src/bufio/bufio_test.go @@ -10,7 +10,6 @@ import ( "errors" "fmt" "io" - "io/ioutil" "strings" "testing" "testing/iotest" @@ -147,7 +146,7 @@ func TestReader(t *testing.T) { for i := 0; i < len(texts)-1; i++ { texts[i] = str + "\n" all += texts[i] - str += string(rune(i)%26 + 'a') + str += string(rune(i%26 + 'a')) } texts[len(texts)-1] = all @@ -886,7 +885,7 @@ func TestReadEmptyBuffer(t *testing.T) { func TestLinesAfterRead(t *testing.T) { l := NewReaderSize(bytes.NewReader([]byte("foo")), minReadBufferSize) - _, err := ioutil.ReadAll(l) + _, err := io.ReadAll(l) if err != nil { t.Error(err) return @@ -1130,7 +1129,7 @@ func TestWriterReadFromCounts(t *testing.T) { } } -// A writeCountingDiscard is like ioutil.Discard and counts the number of times +// A writeCountingDiscard is like io.Discard and counts the number of times // Write is called on it. type writeCountingDiscard int @@ -1300,7 +1299,7 @@ func TestReaderReset(t *testing.T) { t.Errorf("buf = %q; want foo", buf) } r.Reset(strings.NewReader("bar bar")) - all, err := ioutil.ReadAll(r) + all, err := io.ReadAll(r) if err != nil { t.Fatal(err) } @@ -1645,13 +1644,13 @@ func BenchmarkReaderWriteToOptimal(b *testing.B) { buf := make([]byte, bufSize) r := bytes.NewReader(buf) srcReader := NewReaderSize(onlyReader{r}, 1<<10) - if _, ok := ioutil.Discard.(io.ReaderFrom); !ok { - b.Fatal("ioutil.Discard doesn't support ReaderFrom") + if _, ok := io.Discard.(io.ReaderFrom); !ok { + b.Fatal("io.Discard doesn't support ReaderFrom") } for i := 0; i < b.N; i++ { r.Seek(0, io.SeekStart) srcReader.Reset(onlyReader{r}) - n, err := srcReader.WriteTo(ioutil.Discard) + n, err := srcReader.WriteTo(io.Discard) if err != nil { b.Fatal(err) } @@ -1722,7 +1721,7 @@ func BenchmarkReaderEmpty(b *testing.B) { str := strings.Repeat("x", 16<<10) for i := 0; i < b.N; i++ { br := NewReader(strings.NewReader(str)) - n, err := io.Copy(ioutil.Discard, br) + n, err := io.Copy(io.Discard, br) if err != nil { b.Fatal(err) } @@ -1737,7 +1736,7 @@ func BenchmarkWriterEmpty(b *testing.B) { str := strings.Repeat("x", 1<<10) bs := []byte(str) for i := 0; i < b.N; i++ { - bw := NewWriter(ioutil.Discard) + bw := NewWriter(io.Discard) bw.Flush() bw.WriteByte('a') bw.Flush() @@ -1752,7 +1751,7 @@ func BenchmarkWriterEmpty(b *testing.B) { func BenchmarkWriterFlush(b *testing.B) { b.ReportAllocs() - bw := NewWriter(ioutil.Discard) + bw := NewWriter(io.Discard) str := strings.Repeat("x", 50) for i := 0; i < b.N; i++ { bw.WriteString(str) diff --git a/src/buildall.bash b/src/buildall.bash index dc67c0630f5..7b3751f42e9 100755 --- a/src/buildall.bash +++ b/src/buildall.bash @@ -3,10 +3,10 @@ # Use of this source code is governed by a BSD-style # license that can be found in the LICENSE file. -# Usage: buildall.sh [-e] [pattern] +# Usage: buildall.bash [-e] [pattern] # # buildall.bash builds the standard library for all Go-supported -# architectures. It is used by the "all-compile" trybot builder, +# architectures. It is used by the "misc-compile" trybot builders, # as a smoke test to quickly flag portability issues. # # Options: @@ -37,12 +37,11 @@ GOROOT="$(cd .. && pwd)" gettargets() { ../bin/go tool dist list | sed -e 's|/|-|' - echo linux-386-387 echo linux-arm-arm5 } selectedtargets() { - gettargets | egrep -v 'android-arm|darwin-arm64' | egrep "$pattern" + gettargets | egrep "$pattern" } # put linux first in the target list to get all the architectures up front. @@ -64,15 +63,11 @@ do echo "### Building $target" export GOOS=$(echo $target | sed 's/-.*//') export GOARCH=$(echo $target | sed 's/.*-//') - unset GO386 GOARM + unset GOARM if [ "$GOARCH" = "arm5" ]; then export GOARCH=arm export GOARM=5 fi - if [ "$GOARCH" = "387" ]; then - export GOARCH=386 - export GO386=387 - fi # Build and vet everything. # cmd/go/internal/work/exec.go enables the same vet flags during go test of std cmd diff --git a/src/bytes/bytes.go b/src/bytes/bytes.go index aa07b9fbc16..ce52649f132 100644 --- a/src/bytes/bytes.go +++ b/src/bytes/bytes.go @@ -227,19 +227,26 @@ func IndexAny(s []byte, chars string) int { continue } r, width = utf8.DecodeRune(s[i:]) - if r == utf8.RuneError { - for _, r = range chars { - if r == utf8.RuneError { + if r != utf8.RuneError { + // r is 2 to 4 bytes + if len(chars) == width { + if chars == string(r) { return i } + continue + } + // Use bytealg.IndexString for performance if available. + if bytealg.MaxLen >= width { + if bytealg.IndexString(chars, string(r)) >= 0 { + return i + } + continue } - continue } - // r is 2 to 4 bytes. Using strings.Index is more reasonable, but as the bytes - // package should not import the strings package, use bytealg.IndexString - // instead. And this does not seem to lose much performance. - if chars == string(r) || bytealg.IndexString(chars, string(r)) >= 0 { - return i + for _, ch := range chars { + if r == ch { + return i + } } } return -1 @@ -304,19 +311,26 @@ func LastIndexAny(s []byte, chars string) int { } r, size := utf8.DecodeLastRune(s[:i]) i -= size - if r == utf8.RuneError { - for _, r = range chars { - if r == utf8.RuneError { + if r != utf8.RuneError { + // r is 2 to 4 bytes + if len(chars) == size { + if chars == string(r) { return i } + continue + } + // Use bytealg.IndexString for performance if available. + if bytealg.MaxLen >= size { + if bytealg.IndexString(chars, string(r)) >= 0 { + return i + } + continue } - continue } - // r is 2 to 4 bytes. Using strings.Index is more reasonable, but as the bytes - // package should not import the strings package, use bytealg.IndexString - // instead. And this does not seem to lose much performance. - if chars == string(r) || bytealg.IndexString(chars, string(r)) >= 0 { - return i + for _, ch := range chars { + if r == ch { + return i + } } } return -1 diff --git a/src/bytes/example_test.go b/src/bytes/example_test.go index 5ba7077c1de..ae93202b570 100644 --- a/src/bytes/example_test.go +++ b/src/bytes/example_test.go @@ -30,6 +30,13 @@ func ExampleBuffer_reader() { // Output: Gophers rule! } +func ExampleBuffer_Bytes() { + buf := bytes.Buffer{} + buf.Write([]byte{'h', 'e', 'l', 'l', 'o', ' ', 'w', 'o', 'r', 'l', 'd'}) + os.Stdout.Write(buf.Bytes()) + // Output: hello world +} + func ExampleBuffer_Grow() { var b bytes.Buffer b.Grow(64) diff --git a/src/bytes/reader_test.go b/src/bytes/reader_test.go index d799e036f0c..8baac5046cb 100644 --- a/src/bytes/reader_test.go +++ b/src/bytes/reader_test.go @@ -8,7 +8,6 @@ import ( . "bytes" "fmt" "io" - "io/ioutil" "sync" "testing" ) @@ -235,7 +234,7 @@ func TestReaderCopyNothing(t *testing.T) { type justWriter struct { io.Writer } - discard := justWriter{ioutil.Discard} // hide ReadFrom + discard := justWriter{io.Discard} // hide ReadFrom var with, withOut nErr with.n, with.err = io.Copy(discard, NewReader(nil)) @@ -248,7 +247,7 @@ func TestReaderCopyNothing(t *testing.T) { // tests that Len is affected by reads, but Size is not. func TestReaderLenSize(t *testing.T) { r := NewReader([]byte("abc")) - io.CopyN(ioutil.Discard, r, 1) + io.CopyN(io.Discard, r, 1) if r.Len() != 2 { t.Errorf("Len = %d; want 2", r.Len()) } @@ -268,7 +267,7 @@ func TestReaderReset(t *testing.T) { if err := r.UnreadRune(); err == nil { t.Errorf("UnreadRune: expected error, got nil") } - buf, err := ioutil.ReadAll(r) + buf, err := io.ReadAll(r) if err != nil { t.Errorf("ReadAll: unexpected error: %v", err) } @@ -314,7 +313,7 @@ func TestReaderZero(t *testing.T) { t.Errorf("UnreadRune: got nil, want error") } - if n, err := (&Reader{}).WriteTo(ioutil.Discard); n != 0 || err != nil { + if n, err := (&Reader{}).WriteTo(io.Discard); n != 0 || err != nil { t.Errorf("WriteTo: got %d, %v; want 0, nil", n, err) } } diff --git a/src/cmd/addr2line/addr2line_test.go b/src/cmd/addr2line/addr2line_test.go index 578d88e432a..992d7ac11e4 100644 --- a/src/cmd/addr2line/addr2line_test.go +++ b/src/cmd/addr2line/addr2line_test.go @@ -8,7 +8,6 @@ import ( "bufio" "bytes" "internal/testenv" - "io/ioutil" "os" "os/exec" "path/filepath" @@ -73,33 +72,41 @@ func testAddr2Line(t *testing.T, exepath, addr string) { if err != nil { t.Fatalf("Stat failed: %v", err) } + // Debug paths are stored slash-separated, so convert to system-native. srcPath = filepath.FromSlash(srcPath) fi2, err := os.Stat(srcPath) - if gorootFinal := os.Getenv("GOROOT_FINAL"); gorootFinal != "" && strings.HasPrefix(srcPath, gorootFinal) { - if os.IsNotExist(err) || (err == nil && !os.SameFile(fi1, fi2)) { - // srcPath has had GOROOT_FINAL substituted for GOROOT, and it doesn't - // match the actual file. GOROOT probably hasn't been moved to its final - // location yet, so try the original location instead. + + // If GOROOT_FINAL is set and srcPath is not the file we expect, perhaps + // srcPath has had GOROOT_FINAL substituted for GOROOT and GOROOT hasn't been + // moved to its final location yet. If so, try the original location instead. + if gorootFinal := os.Getenv("GOROOT_FINAL"); gorootFinal != "" && + (os.IsNotExist(err) || (err == nil && !os.SameFile(fi1, fi2))) { + // srcPath is clean, but GOROOT_FINAL itself might not be. + // (See https://golang.org/issue/41447.) + gorootFinal = filepath.Clean(gorootFinal) + + if strings.HasPrefix(srcPath, gorootFinal) { fi2, err = os.Stat(runtime.GOROOT() + strings.TrimPrefix(srcPath, gorootFinal)) } } + if err != nil { t.Fatalf("Stat failed: %v", err) } if !os.SameFile(fi1, fi2) { t.Fatalf("addr2line_test.go and %s are not same file", srcPath) } - if srcLineNo != "99" { - t.Fatalf("line number = %v; want 99", srcLineNo) + if srcLineNo != "106" { + t.Fatalf("line number = %v; want 106", srcLineNo) } } -// This is line 98. The test depends on that. +// This is line 106. The test depends on that. func TestAddr2Line(t *testing.T) { testenv.MustHaveGoBuild(t) - tmpDir, err := ioutil.TempDir("", "TestAddr2Line") + tmpDir, err := os.MkdirTemp("", "TestAddr2Line") if err != nil { t.Fatal("TempDir failed: ", err) } diff --git a/src/cmd/api/goapi.go b/src/cmd/api/goapi.go index 01b17b88392..ba42812fa63 100644 --- a/src/cmd/api/goapi.go +++ b/src/cmd/api/goapi.go @@ -17,7 +17,6 @@ import ( "go/token" "go/types" "io" - "io/ioutil" "log" "os" "os/exec" @@ -87,7 +86,10 @@ var contexts = []*build.Context{ func contextName(c *build.Context) string { s := c.GOOS + "-" + c.GOARCH if c.CgoEnabled { - return s + "-cgo" + s += "-cgo" + } + if c.Dir != "" { + s += fmt.Sprintf(" [%s]", c.Dir) } return s } @@ -323,15 +325,29 @@ func compareAPI(w io.Writer, features, required, optional, exception []string, a return } +// aliasReplacer applies type aliases to earlier API files, +// to avoid misleading negative results. +// This makes all the references to os.FileInfo in go1.txt +// be read as if they said fs.FileInfo, since os.FileInfo is now an alias. +// If there are many of these, we could do a more general solution, +// but for now the replacer is fine. +var aliasReplacer = strings.NewReplacer( + "os.FileInfo", "fs.FileInfo", + "os.FileMode", "fs.FileMode", + "os.PathError", "fs.PathError", +) + func fileFeatures(filename string) []string { if filename == "" { return nil } - bs, err := ioutil.ReadFile(filename) + bs, err := os.ReadFile(filename) if err != nil { log.Fatalf("Error reading file %s: %v", filename, err) } - lines := strings.Split(string(bs), "\n") + s := string(bs) + s = aliasReplacer.Replace(s) + lines := strings.Split(s, "\n") var nonblank []string for _, line := range lines { line = strings.TrimSpace(line) @@ -478,6 +494,9 @@ func (w *Walker) loadImports() { cmd := exec.Command(goCmd(), "list", "-e", "-deps", "-json", "std") cmd.Env = listEnv(w.context) + if w.context.Dir != "" { + cmd.Dir = w.context.Dir + } out, err := cmd.CombinedOutput() if err != nil { log.Fatalf("loading imports: %v\n%s", err, out) @@ -491,6 +510,7 @@ func (w *Walker) loadImports() { var pkg struct { ImportPath, Dir string ImportMap map[string]string + Standard bool } err := dec.Decode(&pkg) if err == io.EOF { @@ -503,11 +523,13 @@ func (w *Walker) loadImports() { // - Package "unsafe" contains special signatures requiring // extra care when printing them - ignore since it is not // going to change w/o a language change. - // - internal and vendored packages do not contribute to our - // API surface. + // - Internal and vendored packages do not contribute to our + // API surface. (If we are running within the "std" module, + // vendored dependencies appear as themselves instead of + // their "vendor/" standard-library copies.) // - 'go list std' does not include commands, which cannot be // imported anyway. - if ip := pkg.ImportPath; ip != "unsafe" && !strings.HasPrefix(ip, "vendor/") && !internalPkg.MatchString(ip) { + if ip := pkg.ImportPath; pkg.Standard && ip != "unsafe" && !strings.HasPrefix(ip, "vendor/") && !internalPkg.MatchString(ip) { stdPackages = append(stdPackages, ip) } importDir[pkg.ImportPath] = pkg.Dir @@ -847,6 +869,10 @@ func (w *Walker) emitObj(obj types.Object) { func (w *Walker) emitType(obj *types.TypeName) { name := obj.Name() typ := obj.Type() + if obj.IsAlias() { + w.emitf("type %s = %s", name, w.typeString(typ)) + return + } switch typ := typ.Underlying().(type) { case *types.Struct: w.emitStructType(name, typ) diff --git a/src/cmd/api/goapi_test.go b/src/cmd/api/goapi_test.go index eaccc5ceb53..16e0058e5ea 100644 --- a/src/cmd/api/goapi_test.go +++ b/src/cmd/api/goapi_test.go @@ -9,7 +9,6 @@ import ( "flag" "fmt" "go/build" - "io/ioutil" "os" "path/filepath" "sort" @@ -75,7 +74,7 @@ func TestGolden(t *testing.T) { f.Close() } - bs, err := ioutil.ReadFile(goldenFile) + bs, err := os.ReadFile(goldenFile) if err != nil { t.Fatalf("opening golden.txt for package %q: %v", fi.Name(), err) } @@ -216,3 +215,16 @@ func TestIssue29837(t *testing.T) { } } } + +func TestIssue41358(t *testing.T) { + context := new(build.Context) + *context = build.Default + context.Dir = filepath.Join(context.GOROOT, "src") + + w := NewWalker(context, context.Dir) + for _, pkg := range w.stdPackages { + if strings.HasPrefix(pkg, "vendor/") || strings.HasPrefix(pkg, "golang.org/x/") { + t.Fatalf("stdPackages contains unexpected package %s", pkg) + } + } +} diff --git a/src/cmd/asm/internal/arch/arch.go b/src/cmd/asm/internal/arch/arch.go index 2e5d0ff9911..a62e55191e6 100644 --- a/src/cmd/asm/internal/arch/arch.go +++ b/src/cmd/asm/internal/arch/arch.go @@ -535,6 +535,9 @@ func archRISCV64() *Arch { // Standard register names. for i := riscv.REG_X0; i <= riscv.REG_X31; i++ { + if i == riscv.REG_G { + continue + } name := fmt.Sprintf("X%d", i-riscv.REG_X0) register[name] = int16(i) } @@ -571,7 +574,7 @@ func archRISCV64() *Arch { register["S8"] = riscv.REG_S8 register["S9"] = riscv.REG_S9 register["S10"] = riscv.REG_S10 - register["S11"] = riscv.REG_S11 + // Skip S11 as it is the g register. register["T3"] = riscv.REG_T3 register["T4"] = riscv.REG_T4 register["T5"] = riscv.REG_T5 diff --git a/src/cmd/asm/internal/arch/arm64.go b/src/cmd/asm/internal/arch/arm64.go index 3817fcd5c21..e557630ca60 100644 --- a/src/cmd/asm/internal/arch/arm64.go +++ b/src/cmd/asm/internal/arch/arm64.go @@ -75,13 +75,35 @@ func IsARM64STLXR(op obj.As) bool { arm64.ASTXP, arm64.ASTXPW, arm64.ASTLXP, arm64.ASTLXPW: return true } - // atomic instructions + // LDADDx/SWPx/CASx atomic instructions if arm64.IsAtomicInstruction(op) { return true } return false } +// IsARM64TBL reports whether the op (as defined by an arm64.A* +// constant) is one of the TBL-like instructions and one of its +// inputs does not fit into prog.Reg, so require special handling. +func IsARM64TBL(op obj.As) bool { + switch op { + case arm64.AVTBL, arm64.AVMOVQ: + return true + } + return false +} + +// IsARM64CASP reports whether the op (as defined by an arm64.A* +// constant) is one of the CASP-like instructions, and its 2nd +// destination is a register pair that require special handling. +func IsARM64CASP(op obj.As) bool { + switch op { + case arm64.ACASPD, arm64.ACASPW: + return true + } + return false +} + // ARM64Suffix handles the special suffix for the ARM64. // It returns a boolean to indicate success; failure means // cond was unrecognized. @@ -125,13 +147,6 @@ func arm64RegisterNumber(name string, n int16) (int16, bool) { return 0, false } -// IsARM64TBL reports whether the op (as defined by an arm64.A* -// constant) is one of the table lookup instructions that require special -// handling. -func IsARM64TBL(op obj.As) bool { - return op == arm64.AVTBL -} - // ARM64RegisterExtension parses an ARM64 register with extension or arrangement. func ARM64RegisterExtension(a *obj.Addr, ext string, reg, num int16, isAmount, isIndex bool) error { Rnum := (reg & 31) + int16(num<<5) diff --git a/src/cmd/asm/internal/asm/asm.go b/src/cmd/asm/internal/asm/asm.go index 42e217dc233..c4032759bb5 100644 --- a/src/cmd/asm/internal/asm/asm.go +++ b/src/cmd/asm/internal/asm/asm.go @@ -181,7 +181,7 @@ func (p *Parser) asmText(operands [][]lex.Token) { // Argsize set below. }, } - nameAddr.Sym.Func.Text = prog + nameAddr.Sym.Func().Text = prog prog.To.Val = int32(argSize) p.append(prog, "", true) } @@ -622,8 +622,9 @@ func (p *Parser) asmInstruction(op obj.As, cond string, a []obj.Addr) { prog.SetFrom3(a[1]) prog.To = a[2] case sys.ARM64: - // ARM64 instructions with one input and two outputs. - if arch.IsARM64STLXR(op) { + switch { + case arch.IsARM64STLXR(op): + // ARM64 instructions with one input and two outputs. prog.From = a[0] prog.To = a[1] if a[2].Type != obj.TYPE_REG { @@ -631,20 +632,28 @@ func (p *Parser) asmInstruction(op obj.As, cond string, a []obj.Addr) { return } prog.RegTo2 = a[2].Reg - break - } - if arch.IsARM64TBL(op) { + case arch.IsARM64TBL(op): + // one of its inputs does not fit into prog.Reg. prog.From = a[0] - if a[1].Type != obj.TYPE_REGLIST { - p.errorf("%s: expected list; found %s", op, obj.Dconv(prog, &a[1])) - } prog.SetFrom3(a[1]) prog.To = a[2] - break + case arch.IsARM64CASP(op): + prog.From = a[0] + prog.To = a[1] + // both 1st operand and 3rd operand are (Rs, Rs+1) register pair. + // And the register pair must be contiguous. + if (a[0].Type != obj.TYPE_REGREG) || (a[2].Type != obj.TYPE_REGREG) { + p.errorf("invalid addressing modes for 1st or 3rd operand to %s instruction, must be register pair", op) + return + } + // For ARM64 CASP-like instructions, its 2nd destination operand is register pair(Rt, Rt+1) that can + // not fit into prog.RegTo2, so save it to the prog.RestArgs. + prog.SetTo2(a[2]) + default: + prog.From = a[0] + prog.Reg = p.getRegister(prog, op, &a[1]) + prog.To = a[2] } - prog.From = a[0] - prog.Reg = p.getRegister(prog, op, &a[1]) - prog.To = a[2] case sys.I386: prog.From = a[0] prog.SetFrom3(a[1]) @@ -728,7 +737,7 @@ func (p *Parser) asmInstruction(op obj.As, cond string, a []obj.Addr) { } if p.arch.Family == sys.AMD64 { prog.From = a[0] - prog.RestArgs = []obj.Addr{a[1], a[2]} + prog.SetRestArgs([]obj.Addr{a[1], a[2]}) prog.To = a[3] break } @@ -811,13 +820,13 @@ func (p *Parser) asmInstruction(op obj.As, cond string, a []obj.Addr) { } if p.arch.Family == sys.AMD64 { prog.From = a[0] - prog.RestArgs = []obj.Addr{a[1], a[2], a[3]} + prog.SetRestArgs([]obj.Addr{a[1], a[2], a[3]}) prog.To = a[4] break } if p.arch.Family == sys.S390X { prog.From = a[0] - prog.RestArgs = []obj.Addr{a[1], a[2], a[3]} + prog.SetRestArgs([]obj.Addr{a[1], a[2], a[3]}) prog.To = a[4] break } diff --git a/src/cmd/asm/internal/asm/endtoend_test.go b/src/cmd/asm/internal/asm/endtoend_test.go index 0759b7d10f5..7472507caf1 100644 --- a/src/cmd/asm/internal/asm/endtoend_test.go +++ b/src/cmd/asm/internal/asm/endtoend_test.go @@ -31,7 +31,7 @@ func testEndToEnd(t *testing.T, goarch, file string) { architecture, ctxt := setArch(goarch) architecture.Init(ctxt) lexer := lex.NewLexer(input) - parser := NewParser(ctxt, architecture, lexer) + parser := NewParser(ctxt, architecture, lexer, false) pList := new(obj.Plist) var ok bool testOut = new(bytes.Buffer) // The assembler writes test output to this buffer. @@ -257,11 +257,11 @@ func isHexes(s string) bool { return true } -// It would be nice if the error messages began with +// It would be nice if the error messages always began with // the standard file:line: prefix, // but that's not where we are today. // It might be at the beginning but it might be in the middle of the printed instruction. -var fileLineRE = regexp.MustCompile(`(?:^|\()(testdata[/\\][0-9a-z]+\.s:[0-9]+)(?:$|\))`) +var fileLineRE = regexp.MustCompile(`(?:^|\()(testdata[/\\][0-9a-z]+\.s:[0-9]+)(?:$|\)|:)`) // Same as in test/run.go var ( @@ -273,7 +273,7 @@ func testErrors(t *testing.T, goarch, file string) { input := filepath.Join("testdata", file+".s") architecture, ctxt := setArch(goarch) lexer := lex.NewLexer(input) - parser := NewParser(ctxt, architecture, lexer) + parser := NewParser(ctxt, architecture, lexer, false) pList := new(obj.Plist) var ok bool testOut = new(bytes.Buffer) // The assembler writes test output to this buffer. @@ -281,6 +281,7 @@ func testErrors(t *testing.T, goarch, file string) { defer ctxt.Bso.Flush() failed := false var errBuf bytes.Buffer + parser.errorWriter = &errBuf ctxt.DiagFunc = func(format string, args ...interface{}) { failed = true s := fmt.Sprintf(format, args...) @@ -292,7 +293,7 @@ func testErrors(t *testing.T, goarch, file string) { pList.Firstpc, ok = parser.Parse() obj.Flushplist(ctxt, pList, nil, "") if ok && !failed { - t.Errorf("asm: %s had no errors", goarch) + t.Errorf("asm: %s had no errors", file) } errors := map[string]string{} @@ -353,12 +354,7 @@ func testErrors(t *testing.T, goarch, file string) { } func Test386EndToEnd(t *testing.T) { - defer func(old string) { objabi.GO386 = old }(objabi.GO386) - for _, go386 := range []string{"387", "sse2"} { - t.Logf("GO386=%v", go386) - objabi.GO386 = go386 - testEndToEnd(t, "386", "386") - } + testEndToEnd(t, "386", "386") } func TestARMEndToEnd(t *testing.T) { @@ -373,6 +369,10 @@ func TestARMEndToEnd(t *testing.T) { } } +func TestGoBuildErrors(t *testing.T) { + testErrors(t, "amd64", "buildtagerror") +} + func TestARMErrors(t *testing.T) { testErrors(t, "arm", "armerror") } @@ -390,12 +390,7 @@ func TestARM64Errors(t *testing.T) { } func TestAMD64EndToEnd(t *testing.T) { - defer func(old string) { objabi.GOAMD64 = old }(objabi.GOAMD64) - for _, goamd64 := range []string{"normaljumps", "alignedjumps"} { - t.Logf("GOAMD64=%s", goamd64) - objabi.GOAMD64 = goamd64 - testEndToEnd(t, "amd64", "amd64") - } + testEndToEnd(t, "amd64", "amd64") } func Test386Encoder(t *testing.T) { @@ -442,10 +437,6 @@ func TestPPC64EndToEnd(t *testing.T) { testEndToEnd(t, "ppc64", "ppc64") } -func TestPPC64Encoder(t *testing.T) { - testEndToEnd(t, "ppc64", "ppc64enc") -} - func TestRISCVEncoder(t *testing.T) { testEndToEnd(t, "riscv64", "riscvenc") } diff --git a/src/cmd/asm/internal/asm/expr_test.go b/src/cmd/asm/internal/asm/expr_test.go index 1251594349f..e9c92df1f3a 100644 --- a/src/cmd/asm/internal/asm/expr_test.go +++ b/src/cmd/asm/internal/asm/expr_test.go @@ -57,7 +57,7 @@ var exprTests = []exprTest{ } func TestExpr(t *testing.T) { - p := NewParser(nil, nil, nil) // Expression evaluation uses none of these fields of the parser. + p := NewParser(nil, nil, nil, false) // Expression evaluation uses none of these fields of the parser. for i, test := range exprTests { p.start(lex.Tokenize(test.input)) result := int64(p.expr()) @@ -113,7 +113,7 @@ func TestBadExpr(t *testing.T) { } func runBadTest(i int, test badExprTest, t *testing.T) (err error) { - p := NewParser(nil, nil, nil) // Expression evaluation uses none of these fields of the parser. + p := NewParser(nil, nil, nil, false) // Expression evaluation uses none of these fields of the parser. p.start(lex.Tokenize(test.input)) return tryParse(t, func() { p.expr() diff --git a/src/cmd/asm/internal/asm/line_test.go b/src/cmd/asm/internal/asm/line_test.go index 01b058bd956..da857ced3a7 100644 --- a/src/cmd/asm/internal/asm/line_test.go +++ b/src/cmd/asm/internal/asm/line_test.go @@ -39,7 +39,7 @@ func testBadInstParser(t *testing.T, goarch string, tests []badInstTest) { for i, test := range tests { arch, ctxt := setArch(goarch) tokenizer := lex.NewTokenizer("", strings.NewReader(test.input+"\n"), nil) - parser := NewParser(ctxt, arch, tokenizer) + parser := NewParser(ctxt, arch, tokenizer, false) err := tryParse(t, func() { parser.Parse() diff --git a/src/cmd/asm/internal/asm/operand_test.go b/src/cmd/asm/internal/asm/operand_test.go index f187d0b1660..2e83e176b29 100644 --- a/src/cmd/asm/internal/asm/operand_test.go +++ b/src/cmd/asm/internal/asm/operand_test.go @@ -28,7 +28,7 @@ func setArch(goarch string) (*arch.Arch, *obj.Link) { func newParser(goarch string) *Parser { architecture, ctxt := setArch(goarch) - return NewParser(ctxt, architecture, nil) + return NewParser(ctxt, architecture, nil, false) } // tryParse executes parse func in panicOnError=true context. @@ -75,7 +75,12 @@ func testOperandParser(t *testing.T, parser *Parser, tests []operandTest) { parser.start(lex.Tokenize(test.input)) addr := obj.Addr{} parser.operand(&addr) - result := obj.Dconv(&emptyProg, &addr) + var result string + if parser.compilingRuntime { + result = obj.DconvWithABIDetail(&emptyProg, &addr) + } else { + result = obj.Dconv(&emptyProg, &addr) + } if result != test.output { t.Errorf("fail at %s: got %s; expected %s\n", test.input, result, test.output) } @@ -86,6 +91,9 @@ func TestAMD64OperandParser(t *testing.T) { parser := newParser("amd64") testOperandParser(t, parser, amd64OperandTests) testBadOperandParser(t, parser, amd64BadOperandTests) + parser.compilingRuntime = true + testOperandParser(t, parser, amd64RuntimeOperandTests) + testBadOperandParser(t, parser, amd64BadOperandRuntimeTests) } func Test386OperandParser(t *testing.T) { @@ -141,7 +149,7 @@ func TestFuncAddress(t *testing.T) { parser := newParser(sub.arch) for _, test := range sub.tests { parser.start(lex.Tokenize(test.input)) - name, ok := parser.funcAddress() + name, _, ok := parser.funcAddress() isFuncSym := strings.HasSuffix(test.input, "(SB)") && // Ignore static symbols. @@ -298,6 +306,11 @@ var amd64OperandTests = []operandTest{ {"[):[o-FP", ""}, // Issue 12469 - asm hung parsing the o-FP range on non ARM platforms. } +var amd64RuntimeOperandTests = []operandTest{ + {"$bar(SB)", "$bar(SB)"}, + {"$foo(SB)", "$foo(SB)"}, +} + var amd64BadOperandTests = []badOperandTest{ {"[", "register list: expected ']', found EOF"}, {"[4", "register list: bad low register in `[4`"}, @@ -311,6 +324,11 @@ var amd64BadOperandTests = []badOperandTest{ {"[X0-X1-X2]", "register list: expected ']' after `[X0-X1`, found '-'"}, {"[X0,X3]", "register list: expected '-' after `[X0`, found ','"}, {"[X0,X1,X2,X3]", "register list: expected '-' after `[X0`, found ','"}, + {"$foo", "ABI selector only permitted when compiling runtime, reference was to \"foo\""}, +} + +var amd64BadOperandRuntimeTests = []badOperandTest{ + {"$foo", "malformed ABI selector \"bletch\" in reference to \"foo\""}, } var x86OperandTests = []operandTest{ diff --git a/src/cmd/asm/internal/asm/parse.go b/src/cmd/asm/internal/asm/parse.go index 17d40ee4156..154cf9c7a78 100644 --- a/src/cmd/asm/internal/asm/parse.go +++ b/src/cmd/asm/internal/asm/parse.go @@ -25,24 +25,26 @@ import ( ) type Parser struct { - lex lex.TokenReader - lineNum int // Line number in source file. - errorLine int // Line number of last error. - errorCount int // Number of errors. - pc int64 // virtual PC; count of Progs; doesn't advance for GLOBL or DATA. - input []lex.Token - inputPos int - pendingLabels []string // Labels to attach to next instruction. - labels map[string]*obj.Prog - toPatch []Patch - addr []obj.Addr - arch *arch.Arch - ctxt *obj.Link - firstProg *obj.Prog - lastProg *obj.Prog - dataAddr map[string]int64 // Most recent address for DATA for this symbol. - isJump bool // Instruction being assembled is a jump. - errorWriter io.Writer + lex lex.TokenReader + lineNum int // Line number in source file. + errorLine int // Line number of last error. + errorCount int // Number of errors. + sawCode bool // saw code in this file (as opposed to comments and blank lines) + pc int64 // virtual PC; count of Progs; doesn't advance for GLOBL or DATA. + input []lex.Token + inputPos int + pendingLabels []string // Labels to attach to next instruction. + labels map[string]*obj.Prog + toPatch []Patch + addr []obj.Addr + arch *arch.Arch + ctxt *obj.Link + firstProg *obj.Prog + lastProg *obj.Prog + dataAddr map[string]int64 // Most recent address for DATA for this symbol. + isJump bool // Instruction being assembled is a jump. + compilingRuntime bool + errorWriter io.Writer } type Patch struct { @@ -50,14 +52,15 @@ type Patch struct { label string } -func NewParser(ctxt *obj.Link, ar *arch.Arch, lexer lex.TokenReader) *Parser { +func NewParser(ctxt *obj.Link, ar *arch.Arch, lexer lex.TokenReader, compilingRuntime bool) *Parser { return &Parser{ - ctxt: ctxt, - arch: ar, - lex: lexer, - labels: make(map[string]*obj.Prog), - dataAddr: make(map[string]int64), - errorWriter: os.Stderr, + ctxt: ctxt, + arch: ar, + lex: lexer, + labels: make(map[string]*obj.Prog), + dataAddr: make(map[string]int64), + errorWriter: os.Stderr, + compilingRuntime: compilingRuntime, } } @@ -132,6 +135,30 @@ func (p *Parser) ParseSymABIs(w io.Writer) bool { return p.errorCount == 0 } +// nextToken returns the next non-build-comment token from the lexer. +// It reports misplaced //go:build comments but otherwise discards them. +func (p *Parser) nextToken() lex.ScanToken { + for { + tok := p.lex.Next() + if tok == lex.BuildComment { + if p.sawCode { + p.errorf("misplaced //go:build comment") + } + continue + } + if tok != '\n' { + p.sawCode = true + } + if tok == '#' { + // A leftover wisp of a #include/#define/etc, + // to let us know that p.sawCode should be true now. + // Otherwise ignored. + continue + } + return tok + } +} + // line consumes a single assembly line from p.lex of the form // // {label:} WORD[.cond] [ arg {, arg} ] (';' | '\n') @@ -146,7 +173,7 @@ next: // Skip newlines. var tok lex.ScanToken for { - tok = p.lex.Next() + tok = p.nextToken() // We save the line number here so error messages from this instruction // are labeled with this line. Otherwise we complain after we've absorbed // the terminating newline and the line numbers are off by one in errors. @@ -179,11 +206,11 @@ next: items = make([]lex.Token, 0, 3) } for { - tok = p.lex.Next() + tok = p.nextToken() if len(operands) == 0 && len(items) == 0 { if p.arch.InFamily(sys.ARM, sys.ARM64, sys.AMD64, sys.I386) && tok == '.' { // Suffixes: ARM conditionals or x86 modifiers. - tok = p.lex.Next() + tok = p.nextToken() str := p.lex.Text() if tok != scanner.Ident { p.errorf("instruction suffix expected identifier, found %s", str) @@ -285,8 +312,8 @@ func (p *Parser) symDefRef(w io.Writer, word string, operands [][]lex.Token) { // Defines text symbol in operands[0]. if len(operands) > 0 { p.start(operands[0]) - if name, ok := p.funcAddress(); ok { - fmt.Fprintf(w, "def %s ABI0\n", name) + if name, abi, ok := p.funcAddress(); ok { + fmt.Fprintf(w, "def %s %s\n", name, abi) } } return @@ -304,8 +331,8 @@ func (p *Parser) symDefRef(w io.Writer, word string, operands [][]lex.Token) { // Search for symbol references. for _, op := range operands { p.start(op) - if name, ok := p.funcAddress(); ok { - fmt.Fprintf(w, "ref %s ABI0\n", name) + if name, abi, ok := p.funcAddress(); ok { + fmt.Fprintf(w, "ref %s %s\n", name, abi) } } } @@ -740,20 +767,19 @@ func (p *Parser) symbolReference(a *obj.Addr, name string, prefix rune) { case '*': a.Type = obj.TYPE_INDIR } - // Weirdness with statics: Might now have "<>". - isStatic := false - if p.peek() == '<' { - isStatic = true - p.next() - p.get('>') - } + + // Parse optional <> (indicates a static symbol) or + // (selecting text symbol with specific ABI). + doIssueError := true + isStatic, abi := p.symRefAttrs(name, doIssueError) + if p.peek() == '+' || p.peek() == '-' { a.Offset = int64(p.expr()) } if isStatic { a.Sym = p.ctxt.LookupStatic(name) } else { - a.Sym = p.ctxt.Lookup(name) + a.Sym = p.ctxt.LookupABI(name, abi) } if p.peek() == scanner.EOF { if prefix == 0 && p.isJump { @@ -798,12 +824,60 @@ func (p *Parser) setPseudoRegister(addr *obj.Addr, reg string, isStatic bool, pr } } +// symRefAttrs parses an optional function symbol attribute clause for +// the function symbol 'name', logging an error for a malformed +// attribute clause if 'issueError' is true. The return value is a +// (boolean, ABI) pair indicating that the named symbol is either +// static or a particular ABI specification. +// +// The expected form of the attribute clause is: +// +// empty, yielding (false, obj.ABI0) +// "<>", yielding (true, obj.ABI0) +// "" yielding (false, obj.ABI0) +// "" yielding (false, obj.ABIInternal) +// +// Anything else beginning with "<" logs an error if issueError is +// true, otherwise returns (false, obj.ABI0). +// +func (p *Parser) symRefAttrs(name string, issueError bool) (bool, obj.ABI) { + abi := obj.ABI0 + isStatic := false + if p.peek() != '<' { + return isStatic, abi + } + p.next() + tok := p.peek() + if tok == '>' { + isStatic = true + } else if tok == scanner.Ident { + abistr := p.get(scanner.Ident).String() + if !p.compilingRuntime { + if issueError { + p.errorf("ABI selector only permitted when compiling runtime, reference was to %q", name) + } + } else { + theabi, valid := obj.ParseABI(abistr) + if !valid { + if issueError { + p.errorf("malformed ABI selector %q in reference to %q", + abistr, name) + } + } else { + abi = theabi + } + } + } + p.get('>') + return isStatic, abi +} + // funcAddress parses an external function address. This is a // constrained form of the operand syntax that's always SB-based, // non-static, and has at most a simple integer offset: // -// [$|*]sym[+Int](SB) -func (p *Parser) funcAddress() (string, bool) { +// [$|*]sym[][+Int](SB) +func (p *Parser) funcAddress() (string, obj.ABI, bool) { switch p.peek() { case '$', '*': // Skip prefix. @@ -813,25 +887,32 @@ func (p *Parser) funcAddress() (string, bool) { tok := p.next() name := tok.String() if tok.ScanToken != scanner.Ident || p.atStartOfRegister(name) { - return "", false + return "", obj.ABI0, false + } + // Parse optional <> (indicates a static symbol) or + // (selecting text symbol with specific ABI). + noErrMsg := false + isStatic, abi := p.symRefAttrs(name, noErrMsg) + if isStatic { + return "", obj.ABI0, false // This function rejects static symbols. } tok = p.next() if tok.ScanToken == '+' { if p.next().ScanToken != scanner.Int { - return "", false + return "", obj.ABI0, false } tok = p.next() } if tok.ScanToken != '(' { - return "", false + return "", obj.ABI0, false } if reg := p.next(); reg.ScanToken != scanner.Ident || reg.String() != "SB" { - return "", false + return "", obj.ABI0, false } if p.next().ScanToken != ')' || p.peek() != scanner.EOF { - return "", false + return "", obj.ABI0, false } - return name, true + return name, abi, true } // registerIndirect parses the general form of a register indirection. diff --git a/src/cmd/asm/internal/asm/pseudo_test.go b/src/cmd/asm/internal/asm/pseudo_test.go index 100bef91cfd..622ee25ce71 100644 --- a/src/cmd/asm/internal/asm/pseudo_test.go +++ b/src/cmd/asm/internal/asm/pseudo_test.go @@ -37,6 +37,7 @@ func TestErroneous(t *testing.T) { {"TEXT", "$0É:0, 0, $1", "expected end of operand, found É"}, // Issue #12467. {"TEXT", "$:0:(SB, 0, $1", "expected '(', found 0"}, // Issue 12468. {"TEXT", "@B(SB),0,$0", "expected '(', found B"}, // Issue 23580. + {"TEXT", "foo(SB),0", "ABI selector only permitted when compiling runtime, reference was to \"foo\""}, {"FUNCDATA", "", "expect two operands for FUNCDATA"}, {"FUNCDATA", "(SB ", "expect two operands for FUNCDATA"}, {"DATA", "", "expect two operands for DATA"}, diff --git a/src/cmd/asm/internal/asm/testdata/arm64.s b/src/cmd/asm/internal/asm/testdata/arm64.s index f0c716a2b5d..91e3a0ca0a4 100644 --- a/src/cmd/asm/internal/asm/testdata/arm64.s +++ b/src/cmd/asm/internal/asm/testdata/arm64.s @@ -10,14 +10,8 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 -// -// ADD -// -// LTYPE1 imsr ',' spreg ',' reg -// { -// outcode($1, &$2, $4, &$6); -// } -// imsr comes from the old 7a, we only support immediates and registers + +// arithmetic operations ADDW $1, R2, R3 ADDW R1, R2, R3 ADDW R1, ZR, R3 @@ -25,18 +19,29 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 ADD R1, R2, R3 ADD R1, ZR, R3 ADD $1, R2, R3 - ADD $0x000aaa, R2, R3 // ADD $2730, R2, R3 // 43a82a91 - ADD $0x000aaa, R2 // ADD $2730, R2 // 42a82a91 - ADD $0xaaa000, R2, R3 // ADD $11182080, R2, R3 // 43a86a91 - ADD $0xaaa000, R2 // ADD $11182080, R2 // 42a86a91 - ADD $0xaaaaaa, R2, R3 // ADD $11184810, R2, R3 // 43a82a9163a86a91 - ADD $0xaaaaaa, R2 // ADD $11184810, R2 // 42a82a9142a86a91 - SUB $0x000aaa, R2, R3 // SUB $2730, R2, R3 // 43a82ad1 - SUB $0x000aaa, R2 // SUB $2730, R2 // 42a82ad1 - SUB $0xaaa000, R2, R3 // SUB $11182080, R2, R3 // 43a86ad1 - SUB $0xaaa000, R2 // SUB $11182080, R2 // 42a86ad1 - SUB $0xaaaaaa, R2, R3 // SUB $11184810, R2, R3 // 43a82ad163a86ad1 - SUB $0xaaaaaa, R2 // SUB $11184810, R2 // 42a82ad142a86ad1 + ADDW $1, R2 + ADDW R1, R2 + ADD $1, R2 + ADD R1, R2 + ADD R1>>11, R2 + ADD R1<<22, R2 + ADD R1->33, R2 + ADD $0x000aaa, R2, R3 // ADD $2730, R2, R3 // 43a82a91 + ADD $0x000aaa, R2 // ADD $2730, R2 // 42a82a91 + ADD $0xaaa000, R2, R3 // ADD $11182080, R2, R3 // 43a86a91 + ADD $0xaaa000, R2 // ADD $11182080, R2 // 42a86a91 + ADD $0xaaaaaa, R2, R3 // ADD $11184810, R2, R3 // 43a82a9163a86a91 + ADD $0xaaaaaa, R2 // ADD $11184810, R2 // 42a82a9142a86a91 + SUB $0x000aaa, R2, R3 // SUB $2730, R2, R3 // 43a82ad1 + SUB $0x000aaa, R2 // SUB $2730, R2 // 42a82ad1 + SUB $0xaaa000, R2, R3 // SUB $11182080, R2, R3 // 43a86ad1 + SUB $0xaaa000, R2 // SUB $11182080, R2 // 42a86ad1 + SUB $0xaaaaaa, R2, R3 // SUB $11184810, R2, R3 // 43a82ad163a86ad1 + SUB $0xaaaaaa, R2 // SUB $11184810, R2 // 42a82ad142a86ad1 + ADDW $0x60060, R2 // ADDW $393312, R2 // 4280011142804111 + ADD $0x186a0, R2, R5 // ADD $100000, R2, R5 // 45801a91a5604091 + SUB $0xe7791f700, R3, R1 // SUB $62135596800, R3, R1 // 1be09ed23bf2aef2db01c0f261001bcb + ADD $0x3fffffffc000, R5 // ADD $70368744161280, R5 // fb7f72b2a5001b8b ADD R1>>11, R2, R3 ADD R1<<22, R2, R3 ADD R1->33, R2, R3 @@ -59,6 +64,30 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 CMN R1.SXTX<<2, R10 // 5fe921ab CMPW R2.UXTH<<3, R11 // 7f2d226b CMNW R1.SXTB, R9 // 3f81212b + CMPW $0x60060, R2 // CMPW $393312, R2 // 1b0c8052db00a0725f001b6b + CMPW $40960, R0 // 1f284071 + CMPW $27745, R2 // 3b8c8d525f001b6b + CMNW $0x3fffffc0, R2 // CMNW $1073741760, R2 // fb5f1a325f001b2b + CMPW $0xffff0, R1 // CMPW $1048560, R1 // fb3f1c323f001b6b + CMP $0xffffffffffa0, R3 // CMP $281474976710560, R3 // fb0b80921b00e0f27f001beb + CMP $0xf4240, R1 // CMP $1000000, R1 // 1b4888d2fb01a0f23f001beb + CMP $3343198598084851058, R3 // 5bae8ed2db8daef23badcdf2bbcce5f27f001beb + CMP $3, R2 + CMP R1, R2 + CMP R1->11, R2 + CMP R1>>22, R2 + CMP R1<<33, R2 + CMP R22.SXTX, RSP // ffe336eb + CMP $0x22220000, RSP // CMP $572653568, RSP // 5b44a4d2ff633beb + CMPW $0x22220000, RSP // CMPW $572653568, RSP // 5b44a452ff633b6b + CCMN MI, ZR, R1, $4 // e44341ba + // MADD Rn,Rm,Ra,Rd + MADD R1, R2, R3, R4 // 6408019b + // CLS + CLSW R1, R2 + CLS R1, R2 + +// fp/simd instructions. VADDP V1.B16, V2.B16, V3.B16 // 43bc214e VADDP V1.S4, V2.S4, V3.S4 // 43bca14e VADDP V1.D2, V2.D2, V3.D2 // 43bce14e @@ -67,20 +96,6 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 VORR V5.B16, V4.B16, V3.B16 // 831ca54e VADD V16.S4, V5.S4, V9.S4 // a984b04e VEOR V0.B16, V1.B16, V0.B16 // 201c206e - SHA256H V9.S4, V3, V2 // 6240095e - SHA256H2 V9.S4, V4, V3 // 8350095e - SHA256SU0 V8.S4, V7.S4 // 0729285e - SHA256SU1 V6.S4, V5.S4, V7.S4 // a760065e - SHA1SU0 V11.S4, V8.S4, V6.S4 // 06310b5e - SHA1SU1 V5.S4, V1.S4 // a118285e - SHA1C V1.S4, V2, V3 // 4300015e - SHA1H V5, V4 // a408285e - SHA1M V8.S4, V7, V6 // e620085e - SHA1P V11.S4, V10, V9 // 49110b5e - SHA512H V2.D2, V1, V0 // 208062ce - SHA512H2 V4.D2, V3, V2 // 628464ce - SHA512SU0 V9.D2, V8.D2 // 2881c0ce - SHA512SU1 V7.D2, V6.D2, V5.D2 // c58867ce VADDV V0.S4, V0 // 00b8b14e VMOVI $82, V0.B16 // 40e6024f VUADDLV V6.B16, V6 // c638306e @@ -94,10 +109,6 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 VFMLS V1.D2, V12.D2, V1.D2 // 81cde14e VFMLS V1.S2, V12.S2, V1.S2 // 81cda10e VFMLS V1.S4, V12.S4, V1.S4 // 81cda14e - VPMULL V2.D1, V1.D1, V3.Q1 // 23e0e20e - VPMULL2 V2.D2, V1.D2, V4.Q1 // 24e0e24e - VPMULL V2.B8, V1.B8, V3.H8 // 23e0220e - VPMULL2 V2.B16, V1.B16, V4.H8 // 24e0224e VEXT $4, V2.B8, V1.B8, V3.B8 // 2320022e VEXT $8, V2.B16, V1.B16, V3.B16 // 2340026e VRBIT V24.B16, V24.B16 // 185b606e @@ -123,6 +134,14 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 VSRI $8, V1.H8, V2.H8 // 2244186f VSRI $2, V1.B8, V2.B8 // 22440e2f VSRI $2, V1.B16, V2.B16 // 22440e6f + VSLI $7, V2.B16, V3.B16 // 43540f6f + VSLI $15, V3.H4, V4.H4 // 64541f2f + VSLI $31, V5.S4, V6.S4 // a6543f6f + VSLI $63, V7.D2, V8.D2 // e8547f6f + VUSRA $8, V2.B16, V3.B16 // 4314086f + VUSRA $16, V3.H4, V4.H4 // 6414102f + VUSRA $32, V5.S4, V6.S4 // a614206f + VUSRA $64, V7.D2, V8.D2 // e814406f VTBL V22.B16, [V28.B16, V29.B16], V11.B16 // 8b23164e VTBL V18.B8, [V17.B16, V18.B16, V19.B16], V22.B8 // 3642120e VTBL V31.B8, [V14.B16, V15.B16, V16.B16, V17.B16], V15.B8 // cf611f0e @@ -145,108 +164,126 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 VZIP2 V10.D2, V13.D2, V3.D2 // a379ca4e VZIP1 V17.S2, V4.S2, V26.S2 // 9a38910e VZIP2 V25.S2, V14.S2, V25.S2 // d979990e - MOVD (R2)(R6.SXTW), R4 // 44c866f8 - MOVD (R3)(R6), R5 // MOVD (R3)(R6*1), R5 // 656866f8 - MOVD (R2)(R6), R4 // MOVD (R2)(R6*1), R4 // 446866f8 - MOVWU (R19)(R20<<2), R20 // 747a74b8 - MOVD (R2)(R6<<3), R4 // 447866f8 - MOVD (R3)(R7.SXTX<<3), R8 // 68f867f8 - MOVWU (R5)(R4.UXTW), R10 // aa4864b8 - MOVBU (R3)(R9.UXTW), R8 // 68486938 - MOVBU (R5)(R8), R10 // MOVBU (R5)(R8*1), R10 // aa686838 - MOVHU (R2)(R7.SXTW<<1), R11 // 4bd86778 - MOVHU (R1)(R2<<1), R5 // 25786278 - MOVB (R9)(R3.UXTW), R6 // 2649a338 - MOVB (R10)(R6), R15 // MOVB (R10)(R6*1), R15 // 4f69a638 - MOVH (R5)(R7.SXTX<<1), R19 // b3f8a778 - MOVH (R8)(R4<<1), R10 // 0a79a478 - MOVW (R9)(R8.SXTW<<2), R19 // 33d9a8b8 - MOVW (R1)(R4.SXTX), R11 // 2be8a4b8 - MOVW (R1)(R4.SXTX), ZR // 3fe8a4b8 - MOVW (R2)(R5), R12 // MOVW (R2)(R5*1), R12 // 4c68a5b8 - MOVD R5, (R2)(R6<<3) // 457826f8 - MOVD R9, (R6)(R7.SXTX<<3) // c9f827f8 - MOVD ZR, (R6)(R7.SXTX<<3) // dff827f8 - MOVW R8, (R2)(R3.UXTW<<2) // 485823b8 - MOVW R7, (R3)(R4.SXTW) // 67c824b8 - MOVB R4, (R2)(R6.SXTX) // 44e82638 - MOVB R8, (R3)(R9.UXTW) // 68482938 - MOVB R10, (R5)(R8) // MOVB R10, (R5)(R8*1) // aa682838 - MOVH R11, (R2)(R7.SXTW<<1) // 4bd82778 - MOVH R5, (R1)(R2<<1) // 25782278 - MOVH R7, (R2)(R5.SXTX<<1) // 47f82578 - MOVH R8, (R3)(R6.UXTW) // 68482678 - MOVB (R29)(R30<<0), R14 // ae7bbe38 - MOVB (R29)(R30), R14 // MOVB (R29)(R30*1), R14 // ae6bbe38 - MOVB R4, (R2)(R6.SXTX) // 44e82638 + VUXTL V30.B8, V30.H8 // dea7082f + VUXTL V30.H4, V29.S4 // dda7102f + VUXTL V29.S2, V2.D2 // a2a7202f + VUXTL2 V30.H8, V30.S4 // dea7106f + VUXTL2 V29.S4, V2.D2 // a2a7206f + VUXTL2 V30.B16, V2.H8 // c2a7086f + VBIT V21.B16, V25.B16, V4.B16 // 241fb56e + VBSL V23.B16, V3.B16, V7.B16 // 671c776e + VCMTST V2.B8, V29.B8, V2.B8 // a28f220e + VCMTST V2.D2, V23.D2, V3.D2 // e38ee24e + VSUB V2.B8, V30.B8, V30.B8 // de87222e + VUZP1 V0.B8, V30.B8, V1.B8 // c11b000e + VUZP1 V1.B16, V29.B16, V2.B16 // a21b014e + VUZP1 V2.H4, V28.H4, V3.H4 // 831b420e + VUZP1 V3.H8, V27.H8, V4.H8 // 641b434e + VUZP1 V28.S2, V2.S2, V5.S2 // 45189c0e + VUZP1 V29.S4, V1.S4, V6.S4 // 26189d4e + VUZP1 V30.D2, V0.D2, V7.D2 // 0718de4e + VUZP2 V0.D2, V30.D2, V1.D2 // c15bc04e + VUZP2 V30.D2, V0.D2, V29.D2 // 1d58de4e + VUSHLL $0, V30.B8, V30.H8 // dea7082f + VUSHLL $0, V30.H4, V29.S4 // dda7102f + VUSHLL $0, V29.S2, V2.D2 // a2a7202f + VUSHLL2 $0, V30.B16, V2.H8 // c2a7086f + VUSHLL2 $0, V30.H8, V30.S4 // dea7106f + VUSHLL2 $0, V29.S4, V2.D2 // a2a7206f + VUSHLL $7, V30.B8, V30.H8 // dea70f2f + VUSHLL $15, V30.H4, V29.S4 // dda71f2f + VUSHLL2 $31, V30.S4, V2.D2 // c2a73f6f + VBIF V0.B8, V30.B8, V1.B8 // c11fe02e + VBIF V30.B16, V0.B16, V2.B16 // 021cfe6e FMOVS $(4.0), F0 // 0010221e FMOVD $(4.0), F0 // 0010621e FMOVS $(0.265625), F1 // 01302a1e FMOVD $(0.1796875), F2 // 02f0681e FMOVS $(0.96875), F3 // 03f02d1e FMOVD $(28.0), F4 // 0490671e + VUADDW V9.B8, V12.H8, V14.H8 // 8e11292e + VUADDW V13.H4, V10.S4, V11.S4 // 4b116d2e + VUADDW V21.S2, V24.D2, V29.D2 // 1d13b52e + VUADDW2 V9.B16, V12.H8, V14.H8 // 8e11296e + VUADDW2 V13.H8, V20.S4, V30.S4 // 9e126d6e + VUADDW2 V21.S4, V24.D2, V29.D2 // 1d13b56e + FCCMPS LT, F1, F2, $1 // 41b4211e + FMADDS F1, F3, F2, F4 // 440c011f + FMADDD F4, F5, F4, F4 // 8414441f + FMSUBS F13, F21, F13, F19 // b3d50d1f + FMSUBD F11, F7, F15, F31 // ff9d4b1f + FNMADDS F1, F3, F2, F4 // 440c211f + FNMADDD F1, F3, F2, F4 // 440c611f + FNMSUBS F1, F3, F2, F4 // 448c211f + FNMSUBD F1, F3, F2, F4 // 448c611f + FADDS F2, F3, F4 // 6428221e + FADDD F1, F2 // 4228611e + VDUP V19.S[0], V17.S4 // 7106044e - FMOVS (R2)(R6), F4 // FMOVS (R2)(R6*1), F4 // 446866bc - FMOVS (R2)(R6<<2), F4 // 447866bc - FMOVD (R2)(R6), F4 // FMOVD (R2)(R6*1), F4 // 446866fc - FMOVD (R2)(R6<<3), F4 // 447866fc - FMOVS F4, (R2)(R6) // FMOVS F4, (R2)(R6*1) // 446826bc - FMOVS F4, (R2)(R6<<2) // 447826bc - FMOVD F4, (R2)(R6) // FMOVD F4, (R2)(R6*1) // 446826fc - FMOVD F4, (R2)(R6<<3) // 447826fc - CMPW $40960, R0 // 1f284071 - CMPW $27745, R2 // 3b8c8d525f001b6b - CMNW $0x3fffffc0, R2 // CMNW $1073741760, R2 // fb5f1a325f001b2b - CMPW $0xffff0, R1 // CMPW $1048560, R1 // fb3f1c323f001b6b - CMP $0xffffffffffa0, R3 // CMP $281474976710560, R3 // fb0b80921b00e0f27f001beb - CMP $0xf4240, R1 // CMP $1000000, R1 // 1b4888d2fb01a0f23f001beb - ADD $0x186a0, R2, R5 // ADD $100000, R2, R5 // 45801a91a5604091 - SUB $0xe7791f700, R3, R1 // SUB $62135596800, R3, R1 // 1be09ed23bf2aef2db01c0f261001bcb - CMP $3343198598084851058, R3 // 5bae8ed2db8daef23badcdf2bbcce5f27f001beb - ADD $0x3fffffffc000, R5 // ADD $70368744161280, R5 // fb7f72b2a5001b8b -// LTYPE1 imsr ',' spreg ',' -// { -// outcode($1, &$2, $4, &nullgen); -// } -// LTYPE1 imsr ',' reg -// { -// outcode($1, &$2, NREG, &$4); -// } - ADDW $1, R2 - ADDW R1, R2 - ADD $1, R2 - ADD R1, R2 - ADD R1>>11, R2 - ADD R1<<22, R2 - ADD R1->33, R2 - AND R1@>33, R2 +// special + PRFM (R2), PLDL1KEEP // 400080f9 + PRFM 16(R2), PLDL1KEEP // 400880f9 + PRFM 48(R6), PSTL2STRM // d31880f9 + PRFM 8(R12), PLIL3STRM // 8d0580f9 + PRFM (R8), $25 // 190180f9 + PRFM 8(R9), $30 // 3e0580f9 + NOOP // 1f2003d5 + HINT $0 // 1f2003d5 + DMB $1 + SVC + +// encryption + SHA256H V9.S4, V3, V2 // 6240095e + SHA256H2 V9.S4, V4, V3 // 8350095e + SHA256SU0 V8.S4, V7.S4 // 0729285e + SHA256SU1 V6.S4, V5.S4, V7.S4 // a760065e + SHA1SU0 V11.S4, V8.S4, V6.S4 // 06310b5e + SHA1SU1 V5.S4, V1.S4 // a118285e + SHA1C V1.S4, V2, V3 // 4300015e + SHA1H V5, V4 // a408285e + SHA1M V8.S4, V7, V6 // e620085e + SHA1P V11.S4, V10, V9 // 49110b5e + SHA512H V2.D2, V1, V0 // 208062ce + SHA512H2 V4.D2, V3, V2 // 628464ce + SHA512SU0 V9.D2, V8.D2 // 2881c0ce + SHA512SU1 V7.D2, V6.D2, V5.D2 // c58867ce + VRAX1 V26.D2, V29.D2, V30.D2 // be8f7ace + VXAR $63, V27.D2, V21.D2, V26.D2 // bafe9bce + VPMULL V2.D1, V1.D1, V3.Q1 // 23e0e20e + VPMULL2 V2.D2, V1.D2, V4.Q1 // 24e0e24e + VPMULL V2.B8, V1.B8, V3.H8 // 23e0220e + VPMULL2 V2.B16, V1.B16, V4.H8 // 24e0224e + VEOR3 V2.B16, V7.B16, V12.B16, V25.B16 // 990907ce + VBCAX V1.B16, V2.B16, V26.B16, V31.B16 // 5f0722ce + VREV32 V5.B16, V5.B16 // a508206e + VREV64 V2.S2, V3.S2 // 4308a00e + VREV64 V2.S4, V3.S4 // 4308a04e // logical ops +// // make sure constants get encoded into an instruction when it could - AND $(1<<63), R1 // AND $-9223372036854775808, R1 // 21004192 - AND $(1<<63-1), R1 // AND $9223372036854775807, R1 // 21f84092 - ORR $(1<<63), R1 // ORR $-9223372036854775808, R1 // 210041b2 - ORR $(1<<63-1), R1 // ORR $9223372036854775807, R1 // 21f840b2 - EOR $(1<<63), R1 // EOR $-9223372036854775808, R1 // 210041d2 - EOR $(1<<63-1), R1 // EOR $9223372036854775807, R1 // 21f840d2 - - ANDW $0x3ff00000, R2 // ANDW $1072693248, R2 // 42240c12 - BICW $0x3ff00000, R2 // BICW $1072693248, R2 // 42540212 - ORRW $0x3ff00000, R2 // ORRW $1072693248, R2 // 42240c32 - ORNW $0x3ff00000, R2 // ORNW $1072693248, R2 // 42540232 - EORW $0x3ff00000, R2 // EORW $1072693248, R2 // 42240c52 - EONW $0x3ff00000, R2 // EONW $1072693248, R2 // 42540252 - - AND $0x22220000, R3, R4 // AND $572653568, R3, R4 // 5b44a4d264001b8a - ORR $0x22220000, R3, R4 // ORR $572653568, R3, R4 // 5b44a4d264001baa - EOR $0x22220000, R3, R4 // EOR $572653568, R3, R4 // 5b44a4d264001bca - BIC $0x22220000, R3, R4 // BIC $572653568, R3, R4 // 5b44a4d264003b8a - ORN $0x22220000, R3, R4 // ORN $572653568, R3, R4 // 5b44a4d264003baa - EON $0x22220000, R3, R4 // EON $572653568, R3, R4 // 5b44a4d264003bca - ANDS $0x22220000, R3, R4 // ANDS $572653568, R3, R4 // 5b44a4d264001bea - BICS $0x22220000, R3, R4 // BICS $572653568, R3, R4 // 5b44a4d264003bea - + AND R1@>33, R2 + AND $(1<<63), R1 // AND $-9223372036854775808, R1 // 21004192 + AND $(1<<63-1), R1 // AND $9223372036854775807, R1 // 21f84092 + ORR $(1<<63), R1 // ORR $-9223372036854775808, R1 // 210041b2 + ORR $(1<<63-1), R1 // ORR $9223372036854775807, R1 // 21f840b2 + EOR $(1<<63), R1 // EOR $-9223372036854775808, R1 // 210041d2 + EOR $(1<<63-1), R1 // EOR $9223372036854775807, R1 // 21f840d2 + ANDW $0x3ff00000, R2 // ANDW $1072693248, R2 // 42240c12 + BICW $0x3ff00000, R2 // BICW $1072693248, R2 // 42540212 + ORRW $0x3ff00000, R2 // ORRW $1072693248, R2 // 42240c32 + ORNW $0x3ff00000, R2 // ORNW $1072693248, R2 // 42540232 + EORW $0x3ff00000, R2 // EORW $1072693248, R2 // 42240c52 + EONW $0x3ff00000, R2 // EONW $1072693248, R2 // 42540252 + AND $0x22220000, R3, R4 // AND $572653568, R3, R4 // 5b44a4d264001b8a + ORR $0x22220000, R3, R4 // ORR $572653568, R3, R4 // 5b44a4d264001baa + EOR $0x22220000, R3, R4 // EOR $572653568, R3, R4 // 5b44a4d264001bca + BIC $0x22220000, R3, R4 // BIC $572653568, R3, R4 // 5b44a4d264003b8a + ORN $0x22220000, R3, R4 // ORN $572653568, R3, R4 // 5b44a4d264003baa + EON $0x22220000, R3, R4 // EON $572653568, R3, R4 // 5b44a4d264003bca + ANDS $0x22220000, R3, R4 // ANDS $572653568, R3, R4 // 5b44a4d264001bea + BICS $0x22220000, R3, R4 // BICS $572653568, R3, R4 // 5b44a4d264003bea EOR $0xe03fffffffffffff, R20, R22 // EOR $-2287828610704211969, R20, R22 // 96e243d2 TSTW $0x600000006, R1 // TSTW $25769803782, R1 // 3f041f72 TST $0x4900000049, R0 // TST $313532612681, R0 // 3b0980d23b09c0f21f001bea @@ -275,19 +312,22 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 EONW $0x6006000060060, R5 // EONW $1689262177517664, R5 // 1b0c8052db00a072a5003b4a ORNW $0x6006000060060, R5 // ORNW $1689262177517664, R5 // 1b0c8052db00a072a5003b2a BICSW $0x6006000060060, R5 // BICSW $1689262177517664, R5 // 1b0c8052db00a072a5003b6a - ADDW $0x60060, R2 // ADDW $393312, R2 // 4280011142804111 - CMPW $0x60060, R2 // CMPW $393312, R2 // 1b0c8052db00a0725f001b6b - // TODO: this could have better encoding - ANDW $-1, R10 // 1b0080124a011b0a - - AND $8, R0, RSP // 1f007d92 - ORR $8, R0, RSP // 1f007db2 - EOR $8, R0, RSP // 1f007dd2 - BIC $8, R0, RSP // 1ff87c92 - ORN $8, R0, RSP // 1ff87cb2 - EON $8, R0, RSP // 1ff87cd2 + ANDW $-1, R10 // 1b0080124a011b0a + AND $8, R0, RSP // 1f007d92 + ORR $8, R0, RSP // 1f007db2 + EOR $8, R0, RSP // 1f007dd2 + BIC $8, R0, RSP // 1ff87c92 + ORN $8, R0, RSP // 1ff87cb2 + EON $8, R0, RSP // 1ff87cd2 + TST $15, R2 // 5f0c40f2 + TST R1, R2 // 5f0001ea + TST R1->11, R2 // 5f2c81ea + TST R1>>22, R2 // 5f5841ea + TST R1<<33, R2 // 5f8401ea + TST $0x22220000, R3 // TST $572653568, R3 // 5b44a4d27f001bea +// move an immediate to a Rn. MOVD $0x3fffffffc000, R0 // MOVD $70368744161280, R0 // e07f72b2 MOVW $1000000, R4 // 04488852e401a072 MOVW $0xaaaa0000, R1 // MOVW $2863267840, R1 // 4155b552 @@ -305,37 +345,39 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 MOVD $0x1111ffff1111aaaa, R1 // MOVD $1230045644216969898, R1 // a1aa8a922122a2f22122e2f2 MOVD $0, R1 // 010080d2 MOVD $-1, R1 // 01008092 - MOVD $0x210000, R0 // MOVD $2162688, R0 // 2004a0d2 - MOVD $0xffffffffffffaaaa, R1 // MOVD $-21846, R1 // a1aa8a92 + MOVD $0x210000, R0 // MOVD $2162688, R0 // 2004a0d2 + MOVD $0xffffffffffffaaaa, R1 // MOVD $-21846, R1 // a1aa8a92 + MOVW $1, ZR + MOVW $1, R1 + MOVD $1, ZR + MOVD $1, R1 + MOVK $1, R1 -// -// CLS -// -// LTYPE2 imsr ',' reg -// { -// outcode($1, &$2, NREG, &$4); -// } - CLSW R1, R2 - CLS R1, R2 +// move a large constant to a Vd. + VMOVS $0x80402010, V11 // VMOVS $2151686160, V11 + VMOVD $0x8040201008040201, V20 // VMOVD $-9205322385119247871, V20 + VMOVQ $0x7040201008040201, $0x8040201008040201, V10 // VMOVQ $8088500183983456769, $-9205322385119247871, V10 + VMOVQ $0x8040201008040202, $0x7040201008040201, V20 // VMOVQ $-9205322385119247870, $8088500183983456769, V20 -// -// MOV -// -// LTYPE3 addr ',' addr -// { -// outcode($1, &$2, NREG, &$4); -// } +// mov(to/from sp) + MOVD $0x1002(RSP), R1 // MOVD $4098(RSP), R1 // fb074091610b0091 + MOVD $0x1708(RSP), RSP // MOVD $5896(RSP), RSP // fb0740917f231c91 + MOVD $0x2001(R7), R1 // MOVD $8193(R7), R1 // fb08409161070091 + MOVD $0xffffff(R7), R1 // MOVD $16777215(R7), R1 // fbfc7f9161ff3f91 + MOVD $-0x1(R7), R1 // MOVD $-1(R7), R1 // e10400d1 + MOVD $-0x30(R7), R1 // MOVD $-48(R7), R1 // e1c000d1 + MOVD $-0x708(R7), R1 // MOVD $-1800(R7), R1 // e1201cd1 + MOVD $-0x2000(RSP), R1 // MOVD $-8192(RSP), R1 // e10b40d1 + MOVD $-0x10000(RSP), RSP // MOVD $-65536(RSP), RSP // ff4340d1 MOVW R1, R2 MOVW ZR, R1 MOVW R1, ZR - MOVW $1, ZR - MOVW $1, R1 - MOVW ZR, (R1) MOVD R1, R2 MOVD ZR, R1 - MOVD $1, ZR - MOVD $1, R1 - MOVD ZR, (R1) + +// store and load +// +// LD1/ST1 VLD1 (R8), [V1.B16, V2.B16] // 01a1404c VLD1.P (R3), [V31.H8, V0.H8] // 7fa4df4c VLD1.P (R8)(R20), [V21.B16, V22.B16] // VLD1.P (R8)(R20*1), [V21.B16,V22.B16] // 15a1d44c @@ -393,34 +435,21 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 VST4 [V22.D2, V23.D2, V24.D2, V25.D2], (R3) // 760c004c VST4.P [V14.D2, V15.D2, V16.D2, V17.D2], 64(R15) // ee0d9f4c VST4.P [V24.B8, V25.B8, V26.B8, V27.B8], (R3)(R23) // VST4.P [V24.B8, V25.B8, V26.B8, V27.B8], (R3)(R23*1) // 7800970c - FMOVS F20, (R0) // 140000bd + +// pre/post-indexed FMOVS.P F20, 4(R0) // 144400bc FMOVS.W F20, 4(R0) // 144c00bc - FMOVS (R0), F20 // 140040bd + FMOVD.P F20, 8(R1) // 348400fc + FMOVQ.P F13, 11(R10) // 4db5803c + FMOVQ.W F15, 11(R20) // 8fbe803c + FMOVS.P 8(R0), F20 // 148440bc FMOVS.W 8(R0), F20 // 148c40bc - FMOVD F20, (R2) // 540000fd - FMOVD.P F20, 8(R1) // 348400fc FMOVD.W 8(R1), F20 // 348c40fc - PRFM (R2), PLDL1KEEP // 400080f9 - PRFM 16(R2), PLDL1KEEP // 400880f9 - PRFM 48(R6), PSTL2STRM // d31880f9 - PRFM 8(R12), PLIL3STRM // 8d0580f9 - PRFM (R8), $25 // 190180f9 - PRFM 8(R9), $30 // 3e0580f9 + FMOVQ.P 11(R10), F13 // 4db5c03c + FMOVQ.W 11(R20), F15 // 8fbec03c - // small offset fits into instructions - MOVB 1(R1), R2 // 22048039 - MOVH 1(R1), R2 // 22108078 - MOVH 2(R1), R2 // 22048079 - MOVW 1(R1), R2 // 221080b8 - MOVW 4(R1), R2 // 220480b9 - MOVD 1(R1), R2 // 221040f8 - MOVD 8(R1), R2 // 220440f9 - FMOVS 1(R1), F2 // 221040bc - FMOVS 4(R1), F2 // 220440bd - FMOVD 1(R1), F2 // 221040fc - FMOVD 8(R1), F2 // 220440fd +// small offset fits into instructions MOVB R1, 1(R2) // 41040039 MOVH R1, 1(R2) // 41100078 MOVH R1, 2(R2) // 41040079 @@ -428,18 +457,37 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 MOVW R1, 4(R2) // 410400b9 MOVD R1, 1(R2) // 411000f8 MOVD R1, 8(R2) // 410400f9 + MOVD ZR, (R1) + MOVW ZR, (R1) FMOVS F1, 1(R2) // 411000bc FMOVS F1, 4(R2) // 410400bd + FMOVS F20, (R0) // 140000bd FMOVD F1, 1(R2) // 411000fc FMOVD F1, 8(R2) // 410400fd + FMOVD F20, (R2) // 540000fd + FMOVQ F0, 32(R5)// a008803d + FMOVQ F10, 65520(R10) // 4afdbf3d + FMOVQ F11, 64(RSP) // eb13803d + FMOVQ F11, 8(R20) // 8b82803c + FMOVQ F11, 4(R20) // 8b42803c - // large aligned offset, use two instructions - MOVB 0x1001(R1), R2 // MOVB 4097(R1), R2 // 3b04409162078039 - MOVH 0x2002(R1), R2 // MOVH 8194(R1), R2 // 3b08409162078079 - MOVW 0x4004(R1), R2 // MOVW 16388(R1), R2 // 3b104091620780b9 - MOVD 0x8008(R1), R2 // MOVD 32776(R1), R2 // 3b204091620740f9 - FMOVS 0x4004(R1), F2 // FMOVS 16388(R1), F2 // 3b104091620740bd - FMOVD 0x8008(R1), F2 // FMOVD 32776(R1), F2 // 3b204091620740fd + MOVB 1(R1), R2 // 22048039 + MOVH 1(R1), R2 // 22108078 + MOVH 2(R1), R2 // 22048079 + MOVW 1(R1), R2 // 221080b8 + MOVW 4(R1), R2 // 220480b9 + MOVD 1(R1), R2 // 221040f8 + MOVD 8(R1), R2 // 220440f9 + FMOVS (R0), F20 // 140040bd + FMOVS 1(R1), F2 // 221040bc + FMOVS 4(R1), F2 // 220440bd + FMOVD 1(R1), F2 // 221040fc + FMOVD 8(R1), F2 // 220440fd + FMOVQ 32(R5), F2 // a208c03d + FMOVQ 65520(R10), F10 // 4afdff3d + FMOVQ 64(RSP), F11 // eb13c03d + +// large aligned offset, use two instructions(add+ldr/store). MOVB R1, 0x1001(R2) // MOVB R1, 4097(R2) // 5b04409161070039 MOVH R1, 0x2002(R2) // MOVH R1, 8194(R2) // 5b08409161070079 MOVW R1, 0x4004(R2) // MOVW R1, 16388(R2) // 5b104091610700b9 @@ -447,15 +495,16 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 FMOVS F1, 0x4004(R2) // FMOVS F1, 16388(R2) // 5b104091610700bd FMOVD F1, 0x8008(R2) // FMOVD F1, 32776(R2) // 5b204091610700fd - // very large or unaligned offset uses constant pool - // the encoding cannot be checked as the address of the constant pool is unknown. - // here we only test that they can be assembled. - MOVB 0x44332211(R1), R2 // MOVB 1144201745(R1), R2 - MOVH 0x44332211(R1), R2 // MOVH 1144201745(R1), R2 - MOVW 0x44332211(R1), R2 // MOVW 1144201745(R1), R2 - MOVD 0x44332211(R1), R2 // MOVD 1144201745(R1), R2 - FMOVS 0x44332211(R1), F2 // FMOVS 1144201745(R1), F2 - FMOVD 0x44332211(R1), F2 // FMOVD 1144201745(R1), F2 + MOVB 0x1001(R1), R2 // MOVB 4097(R1), R2 // 3b04409162078039 + MOVH 0x2002(R1), R2 // MOVH 8194(R1), R2 // 3b08409162078079 + MOVW 0x4004(R1), R2 // MOVW 16388(R1), R2 // 3b104091620780b9 + MOVD 0x8008(R1), R2 // MOVD 32776(R1), R2 // 3b204091620740f9 + FMOVS 0x4004(R1), F2 // FMOVS 16388(R1), F2 // 3b104091620740bd + FMOVD 0x8008(R1), F2 // FMOVD 32776(R1), F2 // 3b204091620740fd + +// very large or unaligned offset uses constant pool. +// the encoding cannot be checked as the address of the constant pool is unknown. +// here we only test that they can be assembled. MOVB R1, 0x44332211(R2) // MOVB R1, 1144201745(R2) MOVH R1, 0x44332211(R2) // MOVH R1, 1144201745(R2) MOVW R1, 0x44332211(R2) // MOVW R1, 1144201745(R2) @@ -463,14 +512,59 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 FMOVS F1, 0x44332211(R2) // FMOVS F1, 1144201745(R2) FMOVD F1, 0x44332211(R2) // FMOVD F1, 1144201745(R2) -// -// MOVK -// -// LMOVK imm ',' reg -// { -// outcode($1, &$2, NREG, &$4); -// } - MOVK $1, R1 + MOVB 0x44332211(R1), R2 // MOVB 1144201745(R1), R2 + MOVH 0x44332211(R1), R2 // MOVH 1144201745(R1), R2 + MOVW 0x44332211(R1), R2 // MOVW 1144201745(R1), R2 + MOVD 0x44332211(R1), R2 // MOVD 1144201745(R1), R2 + FMOVS 0x44332211(R1), F2 // FMOVS 1144201745(R1), F2 + FMOVD 0x44332211(R1), F2 // FMOVD 1144201745(R1), F2 + +// shifted or extended register offset. + MOVD (R2)(R6.SXTW), R4 // 44c866f8 + MOVD (R3)(R6), R5 // MOVD (R3)(R6*1), R5 // 656866f8 + MOVD (R2)(R6), R4 // MOVD (R2)(R6*1), R4 // 446866f8 + MOVWU (R19)(R20<<2), R20 // 747a74b8 + MOVD (R2)(R6<<3), R4 // 447866f8 + MOVD (R3)(R7.SXTX<<3), R8 // 68f867f8 + MOVWU (R5)(R4.UXTW), R10 // aa4864b8 + MOVBU (R3)(R9.UXTW), R8 // 68486938 + MOVBU (R5)(R8), R10 // MOVBU (R5)(R8*1), R10 // aa686838 + MOVHU (R2)(R7.SXTW<<1), R11 // 4bd86778 + MOVHU (R1)(R2<<1), R5 // 25786278 + MOVB (R9)(R3.UXTW), R6 // 2649a338 + MOVB (R10)(R6), R15 // MOVB (R10)(R6*1), R15 // 4f69a638 + MOVB (R29)(R30<<0), R14 // ae7bbe38 + MOVB (R29)(R30), R14 // MOVB (R29)(R30*1), R14 // ae6bbe38 + MOVH (R5)(R7.SXTX<<1), R19 // b3f8a778 + MOVH (R8)(R4<<1), R10 // 0a79a478 + MOVW (R9)(R8.SXTW<<2), R19 // 33d9a8b8 + MOVW (R1)(R4.SXTX), R11 // 2be8a4b8 + MOVW (R1)(R4.SXTX), ZR // 3fe8a4b8 + MOVW (R2)(R5), R12 // MOVW (R2)(R5*1), R12 // 4c68a5b8 + FMOVS (R2)(R6), F4 // FMOVS (R2)(R6*1), F4 // 446866bc + FMOVS (R2)(R6<<2), F4 // 447866bc + FMOVD (R2)(R6), F4 // FMOVD (R2)(R6*1), F4 // 446866fc + FMOVD (R2)(R6<<3), F4 // 447866fc + + MOVD R5, (R2)(R6<<3) // 457826f8 + MOVD R9, (R6)(R7.SXTX<<3) // c9f827f8 + MOVD ZR, (R6)(R7.SXTX<<3) // dff827f8 + MOVW R8, (R2)(R3.UXTW<<2) // 485823b8 + MOVW R7, (R3)(R4.SXTW) // 67c824b8 + MOVB R4, (R2)(R6.SXTX) // 44e82638 + MOVB R8, (R3)(R9.UXTW) // 68482938 + MOVB R10, (R5)(R8) // MOVB R10, (R5)(R8*1) // aa682838 + MOVH R11, (R2)(R7.SXTW<<1) // 4bd82778 + MOVH R5, (R1)(R2<<1) // 25782278 + MOVH R7, (R2)(R5.SXTX<<1) // 47f82578 + MOVH R8, (R3)(R6.UXTW) // 68482678 + MOVB R4, (R2)(R6.SXTX) // 44e82638 + FMOVS F4, (R2)(R6) // FMOVS F4, (R2)(R6*1) // 446826bc + FMOVS F4, (R2)(R6<<2) // 447826bc + FMOVD F4, (R2)(R6) // FMOVD F4, (R2)(R6*1) // 446826fc + FMOVD F4, (R2)(R6<<3) // 447826fc + +// vmov VMOV V8.S[1], R1 // 013d0c0e VMOV V0.D[0], R11 // 0b3c084e VMOV V0.D[1], R11 // 0b3c184e @@ -485,205 +579,28 @@ TEXT foo(SB), DUPOK|NOSPLIT, $-8 VMOV V9.H[0], V12.H[1] // 2c05066e VMOV V8.B[0], V12.B[1] // 0c05036e VMOV V8.B[7], V4.B[8] // 043d116e - VREV32 V5.B16, V5.B16 // a508206e - VREV64 V2.S2, V3.S2 // 4308a00e - VREV64 V2.S4, V3.S4 // 4308a04e - VDUP V19.S[0], V17.S4 // 7106044e -// -// B/BL -// -// LTYPE4 comma rel -// { -// outcode($1, &nullgen, NREG, &$3); -// } - BL 1(PC) // CALL 1(PC) -// LTYPE4 comma nireg -// { -// outcode($1, &nullgen, NREG, &$3); -// } - BL (R2) // CALL (R2) - BL foo(SB) // CALL foo(SB) - BL bar<>(SB) // CALL bar<>(SB) -// -// BEQ -// -// LTYPE5 comma rel -// { -// outcode($1, &nullgen, NREG, &$3); -// } - BEQ 1(PC) -// -// SVC -// -// LTYPE6 -// { -// outcode($1, &nullgen, NREG, &nullgen); -// } - SVC - -// -// CMP -// -// LTYPE7 imsr ',' spreg comma -// { -// outcode($1, &$2, $4, &nullgen); -// } - CMP $3, R2 - CMP R1, R2 - CMP R1->11, R2 - CMP R1>>22, R2 - CMP R1<<33, R2 - CMP R22.SXTX, RSP // ffe336eb - - CMP $0x22220000, RSP // CMP $572653568, RSP // 5b44a4d2ff633beb - CMPW $0x22220000, RSP // CMPW $572653568, RSP // 5b44a452ff633b6b - -// TST - TST $15, R2 // 5f0c40f2 - TST R1, R2 // 5f0001ea - TST R1->11, R2 // 5f2c81ea - TST R1>>22, R2 // 5f5841ea - TST R1<<33, R2 // 5f8401ea - TST $0x22220000, R3 // TST $572653568, R3 // 5b44a4d27f001bea - -// // CBZ -// -// LTYPE8 reg ',' rel -// { -// outcode($1, &$2, NREG, &$4); -// } again: CBZ R1, again // CBZ R1 -// -// CSET -// -// LTYPER cond ',' reg -// { -// outcode($1, &$2, NREG, &$4); -// } - CSET GT, R1 // e1d79f9a - CSETW HI, R2 // e2979f1a -// -// CSEL/CSINC/CSNEG/CSINV -// -// LTYPES cond ',' reg ',' reg ',' reg -// { -// outgcode($1, &$2, $6.reg, &$4, &$8); -// } +// conditional operations + CSET GT, R1 // e1d79f9a + CSETW HI, R2 // e2979f1a CSEL LT, R1, R2, ZR // 3fb0829a CSELW LT, R2, R3, R4 // 44b0831a CSINC GT, R1, ZR, R3 // 23c49f9a CSNEG MI, R1, R2, R3 // 234482da CSINV CS, R1, R2, R3 // CSINV HS, R1, R2, R3 // 232082da CSINVW MI, R2, ZR, R2 // 42409f5a - -// LTYPES cond ',' reg ',' reg -// { -// outcode($1, &$2, $4.reg, &$6); -// } CINC EQ, R4, R9 // 8914849a CINCW PL, R2, ZR // 5f44821a CINV PL, R11, R22 // 76418bda CINVW LS, R7, R13 // ed80875a CNEG LS, R13, R7 // a7858dda CNEGW EQ, R8, R13 // 0d15885a -// -// CCMN -// -// LTYPEU cond ',' imsr ',' reg ',' imm comma -// { -// outgcode($1, &$2, $6.reg, &$4, &$8); -// } - CCMN MI, ZR, R1, $4 // e44341ba -// -// FADDD -// -// LTYPEK frcon ',' freg -// { -// outcode($1, &$2, NREG, &$4); -// } -// FADDD $0.5, F1 // FADDD $(0.5), F1 - FADDD F1, F2 - -// LTYPEK frcon ',' freg ',' freg -// { -// outcode($1, &$2, $4.reg, &$6); -// } -// FADDD $0.7, F1, F2 // FADDD $(0.69999999999999996), F1, F2 - FADDD F1, F2, F3 - -// -// FCMP -// -// LTYPEL frcon ',' freg comma -// { -// outcode($1, &$2, $4.reg, &nullgen); -// } -// FCMP $0.2, F1 -// FCMP F1, F2 - -// -// FCCMP -// -// LTYPEF cond ',' freg ',' freg ',' imm comma -// { -// outgcode($1, &$2, $6.reg, &$4, &$8); -// } - FCCMPS LT, F1, F2, $1 // 41b4211e - -// -// FMULA -// -// LTYPE9 freg ',' freg ',' freg ',' freg comma -// { -// outgcode($1, &$2, $4.reg, &$6, &$8); -// } -// FMULA F1, F2, F3, F4 - -// -// FCSEL -// -// LFCSEL cond ',' freg ',' freg ',' freg -// { -// outgcode($1, &$2, $6.reg, &$4, &$8); -// } -// -// MADD Rn,Rm,Ra,Rd -// -// LTYPEM reg ',' reg ',' sreg ',' reg -// { -// outgcode($1, &$2, $6, &$4, &$8); -// } -// MADD R1, R2, R3, R4 - - FMADDS F1, F3, F2, F4 // 440c011f - FMADDD F4, F5, F4, F4 // 8414441f - FMSUBS F13, F21, F13, F19 // b3d50d1f - FMSUBD F11, F7, F15, F31 // ff9d4b1f - FNMADDS F1, F3, F2, F4 // 440c211f - FNMADDD F1, F3, F2, F4 // 440c611f - FNMSUBS F1, F3, F2, F4 // 448c211f - FNMSUBD F1, F3, F2, F4 // 448c611f - -// DMB, HINT -// -// LDMB imm -// { -// outcode($1, &$2, NREG, &nullgen); -// } - DMB $1 - -// -// STXR -// -// LSTXR reg ',' addr ',' reg -// { -// outcode($1, &$2, &$4, &$6); -// } +// atomic ops LDARB (R25), R2 // 22ffdf08 LDARH (R5), R7 // a7fcdf48 LDAXPW (R10), (R20, R16) // 54c17f88 @@ -764,38 +681,38 @@ again: LDADDLH R5, (RSP), R7 // e7036578 LDADDLB R5, (R6), R7 // c7006538 LDADDLB R5, (RSP), R7 // e7036538 - LDANDAD R5, (R6), R7 // c710a5f8 - LDANDAD R5, (RSP), R7 // e713a5f8 - LDANDAW R5, (R6), R7 // c710a5b8 - LDANDAW R5, (RSP), R7 // e713a5b8 - LDANDAH R5, (R6), R7 // c710a578 - LDANDAH R5, (RSP), R7 // e713a578 - LDANDAB R5, (R6), R7 // c710a538 - LDANDAB R5, (RSP), R7 // e713a538 - LDANDALD R5, (R6), R7 // c710e5f8 - LDANDALD R5, (RSP), R7 // e713e5f8 - LDANDALW R5, (R6), R7 // c710e5b8 - LDANDALW R5, (RSP), R7 // e713e5b8 - LDANDALH R5, (R6), R7 // c710e578 - LDANDALH R5, (RSP), R7 // e713e578 - LDANDALB R5, (R6), R7 // c710e538 - LDANDALB R5, (RSP), R7 // e713e538 - LDANDD R5, (R6), R7 // c71025f8 - LDANDD R5, (RSP), R7 // e71325f8 - LDANDW R5, (R6), R7 // c71025b8 - LDANDW R5, (RSP), R7 // e71325b8 - LDANDH R5, (R6), R7 // c7102578 - LDANDH R5, (RSP), R7 // e7132578 - LDANDB R5, (R6), R7 // c7102538 - LDANDB R5, (RSP), R7 // e7132538 - LDANDLD R5, (R6), R7 // c71065f8 - LDANDLD R5, (RSP), R7 // e71365f8 - LDANDLW R5, (R6), R7 // c71065b8 - LDANDLW R5, (RSP), R7 // e71365b8 - LDANDLH R5, (R6), R7 // c7106578 - LDANDLH R5, (RSP), R7 // e7136578 - LDANDLB R5, (R6), R7 // c7106538 - LDANDLB R5, (RSP), R7 // e7136538 + LDCLRAD R5, (R6), R7 // c710a5f8 + LDCLRAD R5, (RSP), R7 // e713a5f8 + LDCLRAW R5, (R6), R7 // c710a5b8 + LDCLRAW R5, (RSP), R7 // e713a5b8 + LDCLRAH R5, (R6), R7 // c710a578 + LDCLRAH R5, (RSP), R7 // e713a578 + LDCLRAB R5, (R6), R7 // c710a538 + LDCLRAB R5, (RSP), R7 // e713a538 + LDCLRALD R5, (R6), R7 // c710e5f8 + LDCLRALD R5, (RSP), R7 // e713e5f8 + LDCLRALW R5, (R6), R7 // c710e5b8 + LDCLRALW R5, (RSP), R7 // e713e5b8 + LDCLRALH R5, (R6), R7 // c710e578 + LDCLRALH R5, (RSP), R7 // e713e578 + LDCLRALB R5, (R6), R7 // c710e538 + LDCLRALB R5, (RSP), R7 // e713e538 + LDCLRD R5, (R6), R7 // c71025f8 + LDCLRD R5, (RSP), R7 // e71325f8 + LDCLRW R5, (R6), R7 // c71025b8 + LDCLRW R5, (RSP), R7 // e71325b8 + LDCLRH R5, (R6), R7 // c7102578 + LDCLRH R5, (RSP), R7 // e7132578 + LDCLRB R5, (R6), R7 // c7102538 + LDCLRB R5, (RSP), R7 // e7132538 + LDCLRLD R5, (R6), R7 // c71065f8 + LDCLRLD R5, (RSP), R7 // e71365f8 + LDCLRLW R5, (R6), R7 // c71065b8 + LDCLRLW R5, (RSP), R7 // e71365b8 + LDCLRLH R5, (R6), R7 // c7106578 + LDCLRLH R5, (RSP), R7 // e7136578 + LDCLRLB R5, (R6), R7 // c7106538 + LDCLRLB R5, (RSP), R7 // e7136538 LDEORAD R5, (R6), R7 // c720a5f8 LDEORAD R5, (RSP), R7 // e723a5f8 LDEORAW R5, (R6), R7 // c720a5b8 @@ -860,21 +777,36 @@ again: LDORLH R5, (RSP), R7 // e7336578 LDORLB R5, (R6), R7 // c7306538 LDORLB R5, (RSP), R7 // e7336538 + CASD R1, (R2), ZR // 5f7ca1c8 + CASW R1, (RSP), ZR // ff7fa188 + CASB ZR, (R5), R3 // a37cbf08 + CASH R3, (RSP), ZR // ff7fa348 + CASW R5, (R7), R6 // e67ca588 + CASLD ZR, (RSP), R8 // e8ffbfc8 + CASLW R9, (R10), ZR // 5ffda988 + CASAD R7, (R11), R15 // 6f7de7c8 + CASAW R10, (RSP), R19 // f37fea88 + CASALD R5, (R6), R7 // c7fce5c8 + CASALD R5, (RSP), R7 // e7ffe5c8 + CASALW R5, (R6), R7 // c7fce588 + CASALW R5, (RSP), R7 // e7ffe588 + CASALH ZR, (R5), R8 // a8fcff48 + CASALB R8, (R9), ZR // 3ffde808 + CASPD (R30, ZR), (RSP), (R8, R9) // e87f3e48 + CASPW (R6, R7), (R8), (R4, R5) // 047d2608 + CASPD (R2, R3), (R2), (R8, R9) // 487c2248 + // RET -// -// LTYPEA comma -// { -// outcode($1, &nullgen, NREG, &nullgen); -// } - BEQ 2(PC) RET RET foo(SB) -// More B/BL cases, and canonical names JMP, CALL. - - BEQ 2(PC) - B foo(SB) // JMP foo(SB) - BL foo(SB) // CALL foo(SB) +// B/BL/B.cond cases, and canonical names JMP, CALL. + BL 1(PC) // CALL 1(PC) + BL (R2) // CALL (R2) + BL foo(SB) // CALL foo(SB) + BL bar<>(SB) // CALL bar<>(SB) + B foo(SB) // JMP foo(SB) + BEQ 1(PC) BEQ 2(PC) TBZ $1, R1, 2(PC) TBNZ $2, R2, 2(PC) @@ -1049,8 +981,6 @@ again: FSTPS (F3, F4), 1024(RSP) // fb0310916313002d FSTPS (F3, F4), x(SB) FSTPS (F3, F4), x+8(SB) - NOOP // 1f2003d5 - HINT $0 // 1f2003d5 // System Register MSR $1, SPSel // bf4100d5 @@ -1612,11 +1542,4 @@ again: MSR R13, ZCR_EL1 // 0d1218d5 MRS ZCR_EL1, R23 // 171238d5 MSR R17, ZCR_EL1 // 111218d5 - -// END -// -// LTYPEE comma -// { -// outcode($1, &nullgen, NREG, &nullgen); -// } END diff --git a/src/cmd/asm/internal/asm/testdata/arm64error.s b/src/cmd/asm/internal/asm/testdata/arm64error.s index 9f377817a90..e579f20836a 100644 --- a/src/cmd/asm/internal/asm/testdata/arm64error.s +++ b/src/cmd/asm/internal/asm/testdata/arm64error.s @@ -87,13 +87,13 @@ TEXT errors(SB),$0 VLD1.P 32(R1), [V8.S4, V9.S4, V10.S4] // ERROR "invalid post-increment offset" VLD1.P 48(R1), [V7.S4, V8.S4, V9.S4, V10.S4] // ERROR "invalid post-increment offset" VPMULL V1.D1, V2.H4, V3.Q1 // ERROR "invalid arrangement" - VPMULL V1.H4, V2.H4, V3.Q1 // ERROR "invalid arrangement" - VPMULL V1.D2, V2.D2, V3.Q1 // ERROR "invalid arrangement" - VPMULL V1.B16, V2.B16, V3.H8 // ERROR "invalid arrangement" + VPMULL V1.H4, V2.H4, V3.Q1 // ERROR "operand mismatch" + VPMULL V1.D2, V2.D2, V3.Q1 // ERROR "operand mismatch" + VPMULL V1.B16, V2.B16, V3.H8 // ERROR "operand mismatch" VPMULL2 V1.D2, V2.H4, V3.Q1 // ERROR "invalid arrangement" - VPMULL2 V1.H4, V2.H4, V3.Q1 // ERROR "invalid arrangement" - VPMULL2 V1.D1, V2.D1, V3.Q1 // ERROR "invalid arrangement" - VPMULL2 V1.B8, V2.B8, V3.H8 // ERROR "invalid arrangement" + VPMULL2 V1.H4, V2.H4, V3.Q1 // ERROR "operand mismatch" + VPMULL2 V1.D1, V2.D1, V3.Q1 // ERROR "operand mismatch" + VPMULL2 V1.B8, V2.B8, V3.H8 // ERROR "operand mismatch" VEXT $8, V1.B16, V2.B8, V2.B16 // ERROR "invalid arrangement" VEXT $8, V1.H8, V2.H8, V2.H8 // ERROR "invalid arrangement" VRBIT V1.B16, V2.B8 // ERROR "invalid arrangement" @@ -123,14 +123,14 @@ TEXT errors(SB),$0 LDADDLW R5, (R6), ZR // ERROR "illegal destination register" LDADDLH R5, (R6), ZR // ERROR "illegal destination register" LDADDLB R5, (R6), ZR // ERROR "illegal destination register" - LDANDD R5, (R6), ZR // ERROR "illegal destination register" - LDANDW R5, (R6), ZR // ERROR "illegal destination register" - LDANDH R5, (R6), ZR // ERROR "illegal destination register" - LDANDB R5, (R6), ZR // ERROR "illegal destination register" - LDANDLD R5, (R6), ZR // ERROR "illegal destination register" - LDANDLW R5, (R6), ZR // ERROR "illegal destination register" - LDANDLH R5, (R6), ZR // ERROR "illegal destination register" - LDANDLB R5, (R6), ZR // ERROR "illegal destination register" + LDCLRD R5, (R6), ZR // ERROR "illegal destination register" + LDCLRW R5, (R6), ZR // ERROR "illegal destination register" + LDCLRH R5, (R6), ZR // ERROR "illegal destination register" + LDCLRB R5, (R6), ZR // ERROR "illegal destination register" + LDCLRLD R5, (R6), ZR // ERROR "illegal destination register" + LDCLRLW R5, (R6), ZR // ERROR "illegal destination register" + LDCLRLH R5, (R6), ZR // ERROR "illegal destination register" + LDCLRLB R5, (R6), ZR // ERROR "illegal destination register" LDEORD R5, (R6), ZR // ERROR "illegal destination register" LDEORW R5, (R6), ZR // ERROR "illegal destination register" LDEORH R5, (R6), ZR // ERROR "illegal destination register" @@ -163,22 +163,22 @@ TEXT errors(SB),$0 LDADDLW R5, (R6), RSP // ERROR "illegal destination register" LDADDLH R5, (R6), RSP // ERROR "illegal destination register" LDADDLB R5, (R6), RSP // ERROR "illegal destination register" - LDANDAD R5, (R6), RSP // ERROR "illegal destination register" - LDANDAW R5, (R6), RSP // ERROR "illegal destination register" - LDANDAH R5, (R6), RSP // ERROR "illegal destination register" - LDANDAB R5, (R6), RSP // ERROR "illegal destination register" - LDANDALD R5, (R6), RSP // ERROR "illegal destination register" - LDANDALW R5, (R6), RSP // ERROR "illegal destination register" - LDANDALH R5, (R6), RSP // ERROR "illegal destination register" - LDANDALB R5, (R6), RSP // ERROR "illegal destination register" - LDANDD R5, (R6), RSP // ERROR "illegal destination register" - LDANDW R5, (R6), RSP // ERROR "illegal destination register" - LDANDH R5, (R6), RSP // ERROR "illegal destination register" - LDANDB R5, (R6), RSP // ERROR "illegal destination register" - LDANDLD R5, (R6), RSP // ERROR "illegal destination register" - LDANDLW R5, (R6), RSP // ERROR "illegal destination register" - LDANDLH R5, (R6), RSP // ERROR "illegal destination register" - LDANDLB R5, (R6), RSP // ERROR "illegal destination register" + LDCLRAD R5, (R6), RSP // ERROR "illegal destination register" + LDCLRAW R5, (R6), RSP // ERROR "illegal destination register" + LDCLRAH R5, (R6), RSP // ERROR "illegal destination register" + LDCLRAB R5, (R6), RSP // ERROR "illegal destination register" + LDCLRALD R5, (R6), RSP // ERROR "illegal destination register" + LDCLRALW R5, (R6), RSP // ERROR "illegal destination register" + LDCLRALH R5, (R6), RSP // ERROR "illegal destination register" + LDCLRALB R5, (R6), RSP // ERROR "illegal destination register" + LDCLRD R5, (R6), RSP // ERROR "illegal destination register" + LDCLRW R5, (R6), RSP // ERROR "illegal destination register" + LDCLRH R5, (R6), RSP // ERROR "illegal destination register" + LDCLRB R5, (R6), RSP // ERROR "illegal destination register" + LDCLRLD R5, (R6), RSP // ERROR "illegal destination register" + LDCLRLW R5, (R6), RSP // ERROR "illegal destination register" + LDCLRLH R5, (R6), RSP // ERROR "illegal destination register" + LDCLRLB R5, (R6), RSP // ERROR "illegal destination register" LDEORAD R5, (R6), RSP // ERROR "illegal destination register" LDEORAW R5, (R6), RSP // ERROR "illegal destination register" LDEORAH R5, (R6), RSP // ERROR "illegal destination register" @@ -340,4 +340,25 @@ TEXT errors(SB),$0 MRS PMSWINC_EL0, R3 // ERROR "system register is not readable" MRS OSLAR_EL1, R3 // ERROR "system register is not readable" VLD3R.P 24(R15), [V15.H4,V16.H4,V17.H4] // ERROR "invalid post-increment offset" + VBIT V1.H4, V12.H4, V3.H4 // ERROR "invalid arrangement" + VBSL V1.D2, V12.D2, V3.D2 // ERROR "invalid arrangement" + VUXTL V30.D2, V30.H8 // ERROR "operand mismatch" + VUXTL2 V20.B8, V21.H8 // ERROR "operand mismatch" + VUXTL V3.D2, V4.B8 // ERROR "operand mismatch" + VUZP1 V0.B8, V30.B8, V1.B16 // ERROR "operand mismatch" + VUZP2 V0.Q1, V30.Q1, V1.Q1 // ERROR "invalid arrangement" + VUSHLL $0, V30.D2, V30.H8 // ERROR "operand mismatch" + VUSHLL2 $0, V20.B8, V21.H8 // ERROR "operand mismatch" + VUSHLL $8, V30.B8, V30.H8 // ERROR "shift amount out of range" + VUSHLL2 $32, V30.S4, V2.D2 // ERROR "shift amount out of range" + VBIF V0.B8, V1.B8, V2.B16 // ERROR "operand mismatch" + VBIF V0.D2, V1.D2, V2.D2 // ERROR "invalid arrangement" + VUADDW V9.B8, V12.H8, V14.B8 // ERROR "invalid arrangement" + VUADDW2 V9.B8, V12.S4, V14.S4 // ERROR "operand mismatch" + VSLI $64, V7.D2, V8.D2 // ERROR "shift out of range" + VUSRA $0, V7.D2, V8.D2 // ERROR "shift out of range" + CASPD (R3, R4), (R2), (R8, R9) // ERROR "source register pair must start from even register" + CASPD (R2, R3), (R2), (R9, R10) // ERROR "destination register pair must start from even register" + CASPD (R2, R4), (R2), (R8, R9) // ERROR "source register pair must be contiguous" + CASPD (R2, R3), (R2), (R8, R10) // ERROR "destination register pair must be contiguous" RET diff --git a/src/cmd/asm/internal/asm/testdata/buildtagerror.s b/src/cmd/asm/internal/asm/testdata/buildtagerror.s new file mode 100644 index 00000000000..5a2d65b9784 --- /dev/null +++ b/src/cmd/asm/internal/asm/testdata/buildtagerror.s @@ -0,0 +1,8 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +#define X 1 + +//go:build x // ERROR "misplaced //go:build comment" + diff --git a/src/cmd/asm/internal/asm/testdata/ppc64.s b/src/cmd/asm/internal/asm/testdata/ppc64.s index ba64d84a352..8f6eb14f73c 100644 --- a/src/cmd/asm/internal/asm/testdata/ppc64.s +++ b/src/cmd/asm/internal/asm/testdata/ppc64.s @@ -2,1311 +2,719 @@ // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. -// This input was created by taking the instruction productions in -// the old assembler's (9a's) grammar and hand-writing complete -// instructions for each rule, to guarantee we cover the same space. +// This contains the majority of valid opcode combinations +// available in cmd/internal/obj/ppc64/asm9.go with +// their valid instruction encodings. #include "../../../../../runtime/textflag.h" -TEXT foo(SB),DUPOK|NOSPLIT,$0 +TEXT asmtest(SB),DUPOK|NOSPLIT,$0 + // move constants + MOVD $1, R3 // 38600001 + MOVD $-1, R4 // 3880ffff + MOVD $65535, R5 // 6005ffff + MOVD $65536, R6 // 64060001 + MOVD $-32767, R5 // 38a08001 + MOVD $-32768, R6 // 38c08000 + MOVD $1234567, R5 // 6405001260a5d687 + MOVW $1, R3 // 38600001 + MOVW $-1, R4 // 3880ffff + MOVW $65535, R5 // 6005ffff + MOVW $65536, R6 // 64060001 + MOVW $-32767, R5 // 38a08001 + MOVW $-32768, R6 // 38c08000 + MOVW $1234567, R5 // 6405001260a5d687 + MOVD 8(R3), R4 // e8830008 + MOVD (R3)(R4), R5 // 7ca4182a + MOVW 4(R3), R4 // e8830006 + MOVW (R3)(R4), R5 // 7ca41aaa + MOVWZ 4(R3), R4 // 80830004 + MOVWZ (R3)(R4), R5 // 7ca4182e + MOVH 4(R3), R4 // a8830004 + MOVH (R3)(R4), R5 // 7ca41aae + MOVHZ 2(R3), R4 // a0830002 + MOVHZ (R3)(R4), R5 // 7ca41a2e + MOVB 1(R3), R4 // 888300017c840774 + MOVB (R3)(R4), R5 // 7ca418ae7ca50774 + MOVBZ 1(R3), R4 // 88830001 + MOVBZ (R3)(R4), R5 // 7ca418ae + MOVDBR (R3)(R4), R5 // 7ca41c28 + MOVWBR (R3)(R4), R5 // 7ca41c2c + MOVHBR (R3)(R4), R5 // 7ca41e2c + + MOVDU 8(R3), R4 // e8830009 + MOVDU (R3)(R4), R5 // 7ca4186a + MOVWU (R3)(R4), R5 // 7ca41aea + MOVWZU 4(R3), R4 // 84830004 + MOVWZU (R3)(R4), R5 // 7ca4186e + MOVHU 2(R3), R4 // ac830002 + MOVHU (R3)(R4), R5 // 7ca41aee + MOVHZU 2(R3), R4 // a4830002 + MOVHZU (R3)(R4), R5 // 7ca41a6e + MOVBU 1(R3), R4 // 8c8300017c840774 + MOVBU (R3)(R4), R5 // 7ca418ee7ca50774 + MOVBZU 1(R3), R4 // 8c830001 + MOVBZU (R3)(R4), R5 // 7ca418ee + + MOVD R4, 8(R3) // f8830008 + MOVD R5, (R3)(R4) // 7ca4192a + MOVW R4, 4(R3) // 90830004 + MOVW R5, (R3)(R4) // 7ca4192e + MOVH R4, 2(R3) // b0830002 + MOVH R5, (R3)(R4) // 7ca41b2e + MOVB R4, 1(R3) // 98830001 + MOVB R5, (R3)(R4) // 7ca419ae + MOVDBR R5, (R3)(R4) // 7ca41d28 + MOVWBR R5, (R3)(R4) // 7ca41d2c + MOVHBR R5, (R3)(R4) // 7ca41f2c + + MOVDU R4, 8(R3) // f8830009 + MOVDU R5, (R3)(R4) // 7ca4196a + MOVWU R4, 4(R3) // 94830004 + MOVWU R5, (R3)(R4) // 7ca4196e + MOVHU R4, 2(R3) // b4830002 + MOVHU R5, (R3)(R4) // 7ca41b6e + MOVBU R4, 1(R3) // 9c830001 + MOVBU R5, (R3)(R4) // 7ca419ee + + ADD $1, R3 // 38630001 + ADD $1, R3, R4 // 38830001 + ADD $-1, R4 // 3884ffff + ADD $-1, R4, R5 // 38a4ffff + ADD $65535, R5 // 601fffff7cbf2a14 + ADD $65535, R5, R6 // 601fffff7cdf2a14 + ADD $65536, R6 // 3cc60001 + ADD $65536, R6, R7 // 3ce60001 + ADD $-32767, R5 // 38a58001 + ADD $-32767, R5, R4 // 38858001 + ADD $-32768, R6 // 38c68000 + ADD $-32768, R6, R5 // 38a68000 + ADD $1234567, R5 // 641f001263ffd6877cbf2a14 + ADD $1234567, R5, R6 // 641f001263ffd6877cdf2a14 + ADDEX R3, R5, $3, R6 // 7cc32f54 + ADDIS $8, R3 // 3c630008 + ADDIS $1000, R3, R4 // 3c8303e8 + + ANDCC $1, R3 // 70630001 + ANDCC $1, R3, R4 // 70640001 + ANDCC $-1, R4 // 3be0ffff7fe42039 + ANDCC $-1, R4, R5 // 3be0ffff7fe52039 + ANDCC $65535, R5 // 70a5ffff + ANDCC $65535, R5, R6 // 70a6ffff + ANDCC $65536, R6 // 74c60001 + ANDCC $65536, R6, R7 // 74c70001 + ANDCC $-32767, R5 // 3be080017fe52839 + ANDCC $-32767, R5, R4 // 3be080017fe42839 + ANDCC $-32768, R6 // 3be080007fe63039 + ANDCC $-32768, R5, R6 // 3be080007fe62839 + ANDCC $1234567, R5 // 641f001263ffd6877fe52839 + ANDCC $1234567, R5, R6 // 641f001263ffd6877fe62839 + ANDISCC $1, R3 // 74630001 + ANDISCC $1000, R3, R4 // 746403e8 + + OR $1, R3 // 60630001 + OR $1, R3, R4 // 60640001 + OR $-1, R4 // 3be0ffff7fe42378 + OR $-1, R4, R5 // 3be0ffff7fe52378 + OR $65535, R5 // 60a5ffff + OR $65535, R5, R6 // 60a6ffff + OR $65536, R6 // 64c60001 + OR $65536, R6, R7 // 64c70001 + OR $-32767, R5 // 3be080017fe52b78 + OR $-32767, R5, R6 // 3be080017fe62b78 + OR $-32768, R6 // 3be080007fe63378 + OR $-32768, R6, R7 // 3be080007fe73378 + OR $1234567, R5 // 641f001263ffd6877fe52b78 + OR $1234567, R5, R3 // 641f001263ffd6877fe32b78 + ORIS $255, R3, R4 + + XOR $1, R3 // 68630001 + XOR $1, R3, R4 // 68640001 + XOR $-1, R4 // 3be0ffff7fe42278 + XOR $-1, R4, R5 // 3be0ffff7fe52278 + XOR $65535, R5 // 68a5ffff + XOR $65535, R5, R6 // 68a6ffff + XOR $65536, R6 // 6cc60001 + XOR $65536, R6, R7 // 6cc70001 + XOR $-32767, R5 // 3be080017fe52a78 + XOR $-32767, R5, R6 // 3be080017fe62a78 + XOR $-32768, R6 // 3be080007fe63278 + XOR $-32768, R6, R7 // 3be080007fe73278 + XOR $1234567, R5 // 641f001263ffd6877fe52a78 + XOR $1234567, R5, R3 // 641f001263ffd6877fe32a78 + XORIS $15, R3, R4 + + // TODO: the order of CR operands don't match + CMP R3, R4 // 7c232000 + CMPU R3, R4 // 7c232040 + CMPW R3, R4 // 7c032000 + CMPWU R3, R4 // 7c032040 + CMPB R3,R4,R4 // 7c6423f8 + CMPEQB R3,R4,CR6 // 7f0321c0 + + // TODO: constants for ADDC? + ADD R3, R4 // 7c841a14 + ADD R3, R4, R5 // 7ca41a14 + ADDC R3, R4 // 7c841814 + ADDC R3, R4, R5 // 7ca41814 + ADDE R3, R4 // 7c841914 + ADDECC R3, R4 // 7c841915 + ADDEV R3, R4 // 7c841d14 + ADDEVCC R3, R4 // 7c841d15 + ADDV R3, R4 // 7c841e14 + ADDVCC R3, R4 // 7c841e15 + ADDCCC R3, R4, R5 // 7ca41815 + ADDME R3, R4 // 7c8301d4 + ADDMECC R3, R4 // 7c8301d5 + ADDMEV R3, R4 // 7c8305d4 + ADDMEVCC R3, R4 // 7c8305d5 + ADDCV R3, R4 // 7c841c14 + ADDCVCC R3, R4 // 7c841c15 + ADDZE R3, R4 // 7c830194 + ADDZECC R3, R4 // 7c830195 + ADDZEV R3, R4 // 7c830594 + ADDZEVCC R3, R4 // 7c830595 + SUBME R3, R4 // 7c8301d0 + SUBMECC R3, R4 // 7c8301d1 + SUBMEV R3, R4 // 7c8305d0 + SUBZE R3, R4 // 7c830190 + SUBZECC R3, R4 // 7c830191 + SUBZEV R3, R4 // 7c830590 + SUBZEVCC R3, R4 // 7c830591 + + AND R3, R4 // 7c841838 + AND R3, R4, R5 // 7c851838 + ANDN R3, R4, R5 // 7c851878 + ANDCC R3, R4, R5 // 7c851839 + OR R3, R4 // 7c841b78 + OR R3, R4, R5 // 7c851b78 + ORN R3, R4, R5 // 7c851b38 + ORCC R3, R4, R5 // 7c851b79 + XOR R3, R4 // 7c841a78 + XOR R3, R4, R5 // 7c851a78 + XORCC R3, R4, R5 // 7c851a79 + NAND R3, R4, R5 // 7c851bb8 + NANDCC R3, R4, R5 // 7c851bb9 + EQV R3, R4, R5 // 7c851a38 + EQVCC R3, R4, R5 // 7c851a39 + NOR R3, R4, R5 // 7c8518f8 + NORCC R3, R4, R5 // 7c8518f9 + + SUB R3, R4 // 7c832050 + SUB R3, R4, R5 // 7ca32050 + SUBC R3, R4 // 7c832010 + SUBC R3, R4, R5 // 7ca32010 + + MULLW R3, R4 // 7c8419d6 + MULLW R3, R4, R5 // 7ca419d6 + MULLW $10, R3 // 1c63000a + MULLW $10000000, R3 // 641f009863ff96807c7f19d6 + + MULLWCC R3, R4, R5 // 7ca419d7 + MULHW R3, R4, R5 // 7ca41896 + + MULHWU R3, R4, R5 // 7ca41816 + MULLD R3, R4 // 7c8419d2 + MULLD R4, R4, R5 // 7ca421d2 + MULLD $20, R4 // 1c840014 + MULLD $200000000, R4 // 641f0beb63ffc2007c9f21d2 + + MULLDCC R3, R4, R5 // 7ca419d3 + MULHD R3, R4, R5 // 7ca41892 + MULHDCC R3, R4, R5 // 7ca41893 + + MULLWV R3, R4 // 7c841dd6 + MULLWV R3, R4, R5 // 7ca41dd6 + MULLWVCC R3, R4, R5 // 7ca41dd7 + MULHWUCC R3, R4, R5 // 7ca41817 + MULLDV R3, R4, R5 // 7ca41dd2 + MULLDVCC R3, R4, R5 // 7ca41dd3 + + DIVD R3,R4 // 7c841bd2 + DIVD R3, R4, R5 // 7ca41bd2 + DIVDCC R3,R4, R5 // 7ca41bd3 + DIVDU R3, R4, R5 // 7ca41b92 + DIVDV R3, R4, R5 // 7ca41fd2 + DIVDUCC R3, R4, R5 // 7ca41b93 + DIVDVCC R3, R4, R5 // 7ca41fd3 + DIVDUV R3, R4, R5 // 7ca41f92 + DIVDUVCC R3, R4, R5 // 7ca41f93 + DIVDE R3, R4, R5 // 7ca41b52 + DIVDECC R3, R4, R5 // 7ca41b53 + DIVDEU R3, R4, R5 // 7ca41b12 + DIVDEUCC R3, R4, R5 // 7ca41b13 + + REM R3, R4, R5 // 7fe41bd67fff19d67cbf2050 + REMU R3, R4, R5 // 7fe41b967fff19d67bff00287cbf2050 + REMD R3, R4, R5 // 7fe41bd27fff19d27cbf2050 + REMDU R3, R4, R5 // 7fe41b927fff19d27cbf2050 + + MADDHD R3,R4,R5,R6 // 10c32170 + MADDHDU R3,R4,R5,R6 // 10c32171 + + MODUD R3, R4, R5 // 7ca41a12 + MODUW R3, R4, R5 // 7ca41a16 + MODSD R3, R4, R5 // 7ca41e12 + MODSW R3, R4, R5 // 7ca41e16 + + SLW $8, R3, R4 // 5464402e + SLW R3, R4, R5 // 7c851830 + SLWCC R3, R4 // 7c841831 + SLD $16, R3, R4 // 786483e4 + SLD R3, R4, R5 // 7c851836 + SLDCC R3, R4 // 7c841837 + + SRW $8, R3, R4 // 5464c23e + SRW R3, R4, R5 // 7c851c30 + SRWCC R3, R4 // 7c841c31 + SRAW $8, R3, R4 // 7c644670 + SRAW R3, R4, R5 // 7c851e30 + SRAWCC R3, R4 // 7c841e31 + SRD $16, R3, R4 // 78648402 + SRD R3, R4, R5 // 7c851c36 + SRDCC R3, R4 // 7c841c37 + SRAD $16, R3, R4 // 7c648674 + SRAD R3, R4, R5 // 7c851e34 + SRDCC R3, R4 // 7c841c37 + ROTLW $16, R3, R4 // 5464803e + ROTLW R3, R4, R5 // 5c85183e + EXTSWSLI $3, R4, R5 // 7c851ef4 + RLWMI $7, R3, $65535, R6 // 50663c3e + RLWMICC $7, R3, $65535, R6 // 50663c3f + RLWNM $3, R4, $7, R6 // 54861f7e + RLWNM R3, R4, $7, R6 // 5c861f7e + RLWNMCC $3, R4, $7, R6 // 54861f7f + RLWNMCC R3, R4, $7, R6 // 5c861f7f + RLDMI $0, R4, $7, R6 // 7886076c + RLDMICC $0, R4, $7, R6 // 7886076d + RLDIMI $0, R4, $7, R6 // 788601cc + RLDIMICC $0, R4, $7, R6 // 788601cd + RLDC $0, R4, $15, R6 // 78860728 + RLDCCC $0, R4, $15, R6 // 78860729 + RLDCL $0, R4, $7, R6 // 78860770 + RLDCLCC $0, R4, $15, R6 // 78860721 + RLDCR $0, R4, $-16, R6 // 788606f2 + RLDCRCC $0, R4, $-16, R6 // 788606f3 + RLDICL $0, R4, $15, R6 // 788603c0 + RLDICLCC $0, R4, $15, R6 // 788603c1 + RLDICR $0, R4, $15, R6 // 788603c4 + RLDICRCC $0, R4, $15, R6 // 788603c5 + RLDIC $0, R4, $15, R6 // 788603c8 + RLDICCC $0, R4, $15, R6 // 788603c9 + CLRLSLWI $16, R5, $8, R4 // 54a4422e + CLRLSLDI $24, R4, $2, R3 // 78831588 + + BEQ 0(PC) // 41820000 + BEQ CR1,0(PC) // 41860000 + BGE 0(PC) // 40800000 + BGE CR2,0(PC) // 40880000 + BGT 4(PC) // 41810010 + BGT CR3,4(PC) // 418d0010 + BLE 0(PC) // 40810000 + BLE CR4,0(PC) // 40910000 + BLT 0(PC) // 41800000 + BLT CR5,0(PC) // 41940000 + BNE 0(PC) // 40820000 + BLT CR6,0(PC) // 41980000 + JMP 8(PC) // 48000010 -//inst: -// -// load ints and bytes -// -// LMOVW rreg ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW R1, R2 - -// LMOVW addr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW foo<>+4(SB), R2 - MOVW 16(R1), R2 - -// LMOVW regaddr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW (R1), R2 - MOVW (R1+R2), R3 // MOVW (R1)(R2*1), R3 - -// LMOVB rreg ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW R1, R2 - -// LMOVB addr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVB foo<>+3(SB), R2 - MOVB 16(R1), R2 - -// LMOVB regaddr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVB (R1), R2 - MOVB (R1+R2), R3 // MOVB (R1)(R2*1), R3 - -// -// load floats -// -// LFMOV addr ',' freg -// { -// outcode(int($1), &$2, 0, &$4); -// } - FMOVD foo<>+4(SB), F2 - FMOVD 16(R1), F2 - -// LFMOV regaddr ',' freg -// { -// outcode(int($1), &$2, 0, &$4); -// } - FMOVD (R1), F2 - -// LFMOV fimm ',' freg -// { -// outcode(int($1), &$2, 0, &$4); -// } - FMOVD $0.1, F2 // FMOVD $(0.10000000000000001), F2 - -// LFMOV freg ',' freg -// { -// outcode(int($1), &$2, 0, &$4); -// } - FMOVD F1, F2 - -// LFMOV freg ',' addr -// { -// outcode(int($1), &$2, 0, &$4); -// } - FMOVD F2, foo<>+4(SB) - FMOVD F2, 16(R1) - -// LFMOV freg ',' regaddr -// { -// outcode(int($1), &$2, 0, &$4); -// } - FMOVD F2, (R1) - -// -// store ints and bytes -// -// LMOVW rreg ',' addr -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW R1, foo<>+3(SB) - MOVW R1, 16(R2) - -// LMOVW rreg ',' regaddr -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW R1, (R1) - MOVW R1, (R2+R3) // MOVW R1, (R2)(R3*1) - -// LMOVB rreg ',' addr -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVB R1, foo<>+3(SB) - MOVB R1, 16(R2) - -// LMOVB rreg ',' regaddr -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVB R1, (R1) - MOVB R1, (R2+R3) // MOVB R1, (R2)(R3*1) -// -// store floats -// -// LMOVW freg ',' addr -// { -// outcode(int($1), &$2, 0, &$4); -// } - FMOVD F1, foo<>+4(SB) - FMOVD F1, 16(R2) - -// LMOVW freg ',' regaddr -// { -// outcode(int($1), &$2, 0, &$4); -// } - FMOVD F1, (R1) - -// -// floating point status -// -// LMOVW fpscr ',' freg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVFL FPSCR, F1 - -// LMOVW freg ',' fpscr -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVFL F1, FPSCR - -// LMOVW freg ',' imm ',' fpscr -// { -// outgcode(int($1), &$2, 0, &$4, &$6); -// } - MOVFL F1, $4, FPSCR - -// LMOVW fpscr ',' creg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVFL FPSCR, CR0 - -// LMTFSB imm ',' con -// { -// outcode(int($1), &$2, int($4), &nullgen); -// } -//TODO 9a doesn't work MTFSB0 $4, 4 - -// -// field moves (mtcrf) -// -// LMOVW rreg ',' imm ',' lcr -// { -// outgcode(int($1), &$2, 0, &$4, &$6); -// } -// TODO 9a doesn't work MOVFL R1,$4,CR - -// LMOVW rreg ',' creg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW R1, CR1 - -// LMOVW rreg ',' lcr -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW R1, CR - -// -// integer operations -// logical instructions -// shift instructions -// unary instructions -// -// LADDW rreg ',' sreg ',' rreg -// { -// outcode(int($1), &$2, int($4), &$6); -// } - ADD R1, R2, R3 - -// LADDW imm ',' sreg ',' rreg -// { -// outcode(int($1), &$2, int($4), &$6); -// } - ADD $1, R2, R3 - -// LADDW rreg ',' imm ',' rreg -// { -// outgcode(int($1), &$2, 0, &$4, &$6); -// } -//TODO 9a trouble ADD R1, $2, R3 maybe swap rreg and imm - -// LADDW rreg ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - ADD R1, R2 - -// LADDW imm ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - ADD $4, R1 - -// LLOGW rreg ',' sreg ',' rreg -// { -// outcode(int($1), &$2, int($4), &$6); -// } - ADDE R1, R2, R3 - -// LLOGW rreg ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - ADDE R1, R2 - -// LSHW rreg ',' sreg ',' rreg -// { -// outcode(int($1), &$2, int($4), &$6); -// } - SLW R1, R2, R3 - -// LSHW rreg ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - SLW R1, R2 - -// LSHW imm ',' sreg ',' rreg -// { -// outcode(int($1), &$2, int($4), &$6); -// } - SLW $4, R1, R2 - -// LSHW imm ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - SLW $4, R1 - -// LABS rreg ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - SLW $4, R1 - -// LABS rreg -// { -// outcode(int($1), &$2, 0, &$2); -// } - SUBME R1 // SUBME R1, R1 - -// -// multiply-accumulate -// -// LMA rreg ',' sreg ',' rreg -// { -// outcode(int($1), &$2, int($4), &$6); -// } -//TODO this instruction is undefined in lex.go LMA R1, R2, R3 NOT SUPPORTED (called MAC) - -// -// move immediate: macro for cau+or, addi, addis, and other combinations -// -// LMOVW imm ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW $1, R1 - -// LMOVW ximm ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW $1, R1 - MOVW $foo(SB), R1 - -// condition register operations -// -// LCROP cbit ',' cbit -// { -// outcode(int($1), &$2, int($4.Reg), &$4); -// } -//TODO 9a trouble CREQV 1, 2 delete? liblink encodes like a divide (maybe wrong too) - -// LCROP cbit ',' con ',' cbit -// { -// outcode(int($1), &$2, int($4), &$6); -// } -//TODO 9a trouble CREQV 1, 2, 3 - -// -// condition register moves -// move from machine state register -// -// LMOVW creg ',' creg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVFL CR0, CR1 - -// LMOVW psr ',' creg // TODO: should psr should be fpscr -// { -// outcode(int($1), &$2, 0, &$4); -// } -//TODO 9a trouble MOVW FPSCR, CR1 - -// LMOVW lcr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW CR, R1 - -// LMOVW psr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW SPR(0), R1 - MOVW SPR(7), R1 - -// LMOVW xlreg ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW LR, R1 - MOVW CTR, R1 - -// LMOVW rreg ',' xlreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW R1, LR - MOVW R1, CTR - -// LMOVW creg ',' psr // TODO doesn't exist -// { -// outcode(int($1), &$2, 0, &$4); -// } -//TODO 9a trouble MOVW CR1, SPR(7) - -// LMOVW rreg ',' psr -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVW R1, SPR(7) - -// -// branch, branch conditional -// branch conditional register -// branch conditional to count register -// -// LBRA rel -// { -// outcode(int($1), &nullgen, 0, &$2); -// } - BEQ CR1, 2(PC) -label0: - BR 1(PC) // JMP 1(PC) - BEQ CR1, 2(PC) - BR label0+0 // JMP 62 - -// LBRA addr -// { -// outcode(int($1), &nullgen, 0, &$2); -// } - BEQ CR1, 2(PC) - BR LR // JMP LR - BEQ CR1, 2(PC) -// BR 0(R1) // TODO should work - BEQ CR1, 2(PC) - BR foo+0(SB) // JMP foo(SB) - -// LBRA '(' xlreg ')' -// { -// outcode(int($1), &nullgen, 0, &$3); -// } - BEQ CR1, 2(PC) - BR (CTR) // JMP CTR - -// LBRA ',' rel // asm doesn't support the leading comma -// { -// outcode(int($1), &nullgen, 0, &$3); -// } -// LBRA ',' addr // asm doesn't support the leading comma -// { -// outcode(int($1), &nullgen, 0, &$3); -// } -// LBRA ',' '(' xlreg ')' // asm doesn't support the leading comma -// { -// outcode(int($1), &nullgen, 0, &$4); -// } -// LBRA creg ',' rel -// { -// outcode(int($1), &$2, 0, &$4); -// } -label1: - BEQ CR1, 1(PC) - BEQ CR1, label1 // BEQ CR1, 72 - -// LBRA creg ',' addr // TODO DOES NOT WORK in 9a -// { -// outcode(int($1), &$2, 0, &$4); -// } - -// LBRA creg ',' '(' xlreg ')' // TODO DOES NOT WORK in 9a -// { -// outcode(int($1), &$2, 0, &$5); -// } - -// LBRA con ',' rel // TODO DOES NOT WORK in 9a -// { -// outcode(int($1), &nullgen, int($2), &$4); -// } - -// LBRA con ',' addr // TODO DOES NOT WORK in 9a -// { -// outcode(int($1), &nullgen, int($2), &$4); -// } - -// LBRA con ',' '(' xlreg ')' -// { -// outcode(int($1), &nullgen, int($2), &$5); -// } -// BC 4, (CTR) // TODO - should work - -// LBRA con ',' con ',' rel -// { -// var g obj.Addr -// g = nullgen; -// g.Type = obj.TYPE_CONST; -// g.Offset = $2; -// outcode(int($1), &g, int(REG_R0+$4), &$6); -// } -// BC 3, 4, label1 // TODO - should work - -// LBRA con ',' con ',' addr // TODO mystery -// { -// var g obj.Addr -// g = nullgen; -// g.Type = obj.TYPE_CONST; -// g.Offset = $2; -// outcode(int($1), &g, int(REG_R0+$4), &$6); -// } -//TODO 9a trouble BC 3, 3, 4(R1) - -// LBRA con ',' con ',' '(' xlreg ')' -// { -// var g obj.Addr -// g = nullgen; -// g.Type = obj.TYPE_CONST; -// g.Offset = $2; -// outcode(int($1), &g, int(REG_R0+$4), &$7); -// } - BC 3, 3, (LR) // BC $3, R3, LR - -// -// conditional trap // TODO NOT DEFINED -// TODO these instructions are not in lex.go -// -// LTRAP rreg ',' sreg -// { -// outcode(int($1), &$2, int($4), &nullgen); -// } -// LTRAP imm ',' sreg -// { -// outcode(int($1), &$2, int($4), &nullgen); -// } -// LTRAP rreg comma -// { -// outcode(int($1), &$2, 0, &nullgen); -// } -// LTRAP comma -// { -// outcode(int($1), &nullgen, 0, &nullgen); -// } - -// -// floating point operate -// -// LFCONV freg ',' freg -// { -// outcode(int($1), &$2, 0, &$4); -// } - FABS F1, F2 - -// LFADD freg ',' freg -// { -// outcode(int($1), &$2, 0, &$4); -// } - FADD F1, F2 - -// LFADD freg ',' freg ',' freg -// { -// outcode(int($1), &$2, int($4.Reg), &$6); -// } - FADD F1, F2, F3 - -// LFMA freg ',' freg ',' freg ',' freg -// { -// outgcode(int($1), &$2, int($4.Reg), &$6, &$8); -// } - FMADD F1, F2, F3, F4 - -// LFCMP freg ',' freg -// { -// outcode(int($1), &$2, 0, &$4); -// } - FCMPU F1, F2 - -// LFCMP freg ',' freg ',' creg -// { -// outcode(int($1), &$2, int($6.Reg), &$4); -// } -// FCMPU F1, F2, CR0 - -// FTDIV FRA, FRB, BF produces -// ftdiv BF, FRA, FRB - FTDIV F1,F2,$7 - -// FTSQRT FRB, BF produces -// ftsqrt BF, FRB - FTSQRT F2,$7 - -// FCFID -// FCFIDS - - FCFID F2,F3 - FCFIDCC F3,F3 - FCFIDS F2,F3 - FCFIDSCC F2,F3 - -// -// CMP -// -// LCMP rreg ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - CMP R1, R2 - -// LCMP rreg ',' imm -// { -// outcode(int($1), &$2, 0, &$4); -// } - CMP R1, $4 - -// LCMP rreg ',' rreg ',' creg -// { -// outcode(int($1), &$2, int($6.Reg), &$4); -// } - CMP R1, R2, CR0 // CMP R1, CR0, R2 - -// LCMP rreg ',' imm ',' creg -// { -// outcode(int($1), &$2, int($6.Reg), &$4); -// } - CMP R1, $4, CR0 // CMP R1, CR0, $4 - -// CMPB RS,RB,RA produces -// cmpb RA,RS,RB - CMPB R2,R2,R1 - -// CMPEQB RA,RB,BF produces -// cmpeqb BF,RA,RB - CMPEQB R1, R2, CR0 - -// -// rotate extended mnemonics map onto other shift instructions -// - - ROTL $12,R2,R3 - ROTL R2,R3,R4 - ROTLW $9,R2,R3 - ROTLW R2,R3,R4 - -// -// rotate and mask -// -// LRLWM imm ',' rreg ',' imm ',' rreg -// { -// outgcode(int($1), &$2, int($4.Reg), &$6, &$8); -// } - RLDC $4, R1, $16, R2 - -// LRLWM imm ',' rreg ',' mask ',' rreg -// { -// outgcode(int($1), &$2, int($4.Reg), &$6, &$8); -// } - RLDC $26, R1, 4, 5, R2 // RLDC $26, R1, $201326592, R2 - -// LRLWM rreg ',' rreg ',' imm ',' rreg -// { -// outgcode(int($1), &$2, int($4.Reg), &$6, &$8); -// } - RLDCL R1, R2, $7, R3 - -// LRLWM rreg ',' rreg ',' mask ',' rreg -// { -// outgcode(int($1), &$2, int($4.Reg), &$6, &$8); -// } - RLWMI R1, R2, 4, 5, R3 // RLWMI R1, R2, $201326592, R3 - - -// opcodes added with constant shift counts, not masks - - RLDICR $3, R2, $24, R4 - - RLDICL $1, R2, $61, R6 - - RLDIMI $7, R2, $52, R7 - -// opcodes for right and left shifts, const and reg shift counts - - SLD $4, R3, R4 - SLD R2, R3, R4 - SLW $4, R3, R4 - SLW R2, R3, R4 - SRD $8, R3, R4 - SRD R2, R3, R4 - SRW $8, R3, R4 - SRW R2, R3, R4 - -// -// load/store multiple -// -// LMOVMW addr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } -// MOVMW foo+0(SB), R2 // TODO TLS broke this! - MOVMW 4(R1), R2 - -// LMOVMW rreg ',' addr -// { -// outcode(int($1), &$2, 0, &$4); -// } -// MOVMW R1, foo+0(SB) // TODO TLS broke this! - MOVMW R1, 4(R2) - -// -// various indexed load/store -// indexed unary (eg, cache clear) -// -// LXLD regaddr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - LSW (R1), R2 - LSW (R1+R2), R3 // LSW (R1)(R2*1), R3 - -// LXLD regaddr ',' imm ',' rreg -// { -// outgcode(int($1), &$2, 0, &$4, &$6); -// } - LSW (R1), $1, R2 - LSW (R1+R2), $1, R3 // LSW (R1)(R2*1), $1, R3 - -// LXST rreg ',' regaddr -// { -// outcode(int($1), &$2, 0, &$4); -// } - STSW R1, (R2) - STSW R1, (R2+R3) // STSW R1, (R2)(R3*1) - -// LXST rreg ',' imm ',' regaddr -// { -// outgcode(int($1), &$2, 0, &$4, &$6); -// } - STSW R1, $1, (R2) - STSW R1, $1, (R2+R3) // STSW R1, $1, (R2)(R3*1) - -// LXMV regaddr ',' rreg -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVHBR (R1), R2 - MOVHBR (R1+R2), R3 // MOVHBR (R1)(R2*1), R3 - -// LXMV rreg ',' regaddr -// { -// outcode(int($1), &$2, 0, &$4); -// } - MOVHBR R1, (R2) - MOVHBR R1, (R2+R3) // MOVHBR R1, (R2)(R3*1) - -// LXOP regaddr -// { -// outcode(int($1), &$2, 0, &nullgen); -// } - DCBF (R1) - DCBF (R1+R2) // DCBF (R1)(R2*1) - DCBF (R1), $1 - DCBF (R1)(R2*1), $1 - DCBT (R1), $1 - DCBT (R1)(R2*1), $1 - -// LDMX (RB)(RA*1),RT produces -// ldmx RT,RA,RB - LDMX (R2)(R1*1), R3 - -// Population count, X-form -// RS,RA produces -// RA,RS - POPCNTD R1,R2 - POPCNTW R1,R2 - POPCNTB R1,R2 - -// Copysign - FCPSGN F1,F2,F3 - -// Random number generator, X-form -// DARN L,RT produces -// darn RT,L - DARN $1, R1 - -// Copy/Paste facility -// RB,RA produces -// RA,RB - COPY R2,R1 - PASTECC R2,R1 - -// Modulo signed/unsigned double/word X-form -// RA,RB,RT produces -// RT,RA,RB - MODUD R3,R4,R5 - MODUW R3,R4,R5 - MODSD R3,R4,R5 - MODSW R3,R4,R5 - -// VMX instructions - -// Described as: -// , -// produces -// - -// Vector load, VX-form -// (RB)(RA*1),VRT produces -// VRT,RA,RB - LVEBX (R1)(R2*1), V0 - LVEHX (R3)(R4*1), V1 - LVEWX (R5)(R6*1), V2 - LVX (R7)(R8*1), V3 - LVXL (R9)(R10*1), V4 - LVSL (R11)(R12*1), V5 - LVSR (R14)(R15*1), V6 - -// Vector store, VX-form -// VRT,(RB)(RA*1) produces -// VRT,RA,RB - STVEBX V31, (R1)(R2*1) - STVEHX V30, (R2)(R3*1) - STVEWX V29, (R4)(R5*1) - STVX V28, (R6)(R7*1) - STVXL V27, (R9)(R9*1) - -// Vector AND, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VAND V10, V9, V8 - VANDC V15, V14, V13 - VNAND V19, V18, V17 - -// Vector OR, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VOR V26, V25, V24 - VORC V23, V22, V21 - VNOR V20, V19, V18 - VXOR V17, V16, V15 - VEQV V14, V13, V12 - -// Vector ADD, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VADDUBM V3, V2, V1 - VADDUHM V3, V2, V1 - VADDUWM V3, V2, V1 - VADDUDM V3, V2, V1 - VADDUQM V3, V2, V1 - VADDCUQ V3, V2, V1 - VADDCUW V3, V2, V1 - VADDUBS V3, V2, V1 - VADDUHS V3, V2, V1 - VADDUWS V3, V2, V1 - VADDSBS V3, V2, V1 - VADDSHS V3, V2, V1 - VADDSWS V3, V2, V1 - -// Vector ADD extended, VA-form -// VRA,VRB,VRC,VRT produces -// VRT,VRA,VRB,VRC - VADDEUQM V4, V3, V2, V1 - VADDECUQ V4, V3, V2, V1 - -// Vector multiply, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VMULESB V2, V3, V1 - VMULOSB V2, V3, V1 - VMULEUB V2, V3, V1 - VMULOUB V2, V3, V1 - VMULESH V2, V3, V1 - VMULOSH V2, V3, V1 - VMULEUH V2, V3, V1 - VMULOUH V2, V3, V1 - VMULESW V2, V3, V1 - VMULOSW V2, V3, V1 - VMULEUW V2, V3, V1 - VMULOUW V2, V3, V1 - VMULUWM V2, V3, V1 - -// Vector polynomial multiply-sum, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VPMSUMB V2, V3, V1 - VPMSUMH V2, V3, V1 - VPMSUMW V2, V3, V1 - VPMSUMD V2, V3, V1 - -// Vector multiply-sum, VA-form -// VRA, VRB, VRC, VRT produces -// VRT, VRA, VRB, VRC - VMSUMUDM V4, V3, V2, V1 - -// Vector SUB, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VSUBUBM V3, V2, V1 - VSUBUHM V3, V2, V1 - VSUBUWM V3, V2, V1 - VSUBUDM V3, V2, V1 - VSUBUQM V3, V2, V1 - VSUBCUQ V3, V2, V1 - VSUBCUW V3, V2, V1 - VSUBUBS V3, V2, V1 - VSUBUHS V3, V2, V1 - VSUBUWS V3, V2, V1 - VSUBSBS V3, V2, V1 - VSUBSHS V3, V2, V1 - VSUBSWS V3, V2, V1 - -// Vector SUB extended, VA-form -// VRA,VRB,VRC,VRT produces -// VRT,VRA,VRB,VRC - VSUBEUQM V4, V3, V2, V1 - VSUBECUQ V4, V3, V2, V1 - -// Vector rotate, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VRLB V2, V1, V0 - VRLH V2, V1, V0 - VRLW V2, V1, V0 - VRLD V2, V1, V0 - -// Vector shift, VX-form -// VRA,VRB,VRT -// VRT,VRA,VRB - VSLB V2, V1, V0 - VSLH V2, V1, V0 - VSLW V2, V1, V0 - VSL V2, V1, V0 - VSLO V2, V1, V0 - VSRB V2, V1, V0 - VSRH V2, V1, V0 - VSRW V2, V1, V0 - VSR V2, V1, V0 - VSRO V2, V1, V0 - VSLD V2, V1, V0 - VSRD V2, V1, V0 - VSRAB V2, V1, V0 - VSRAH V2, V1, V0 - VSRAW V2, V1, V0 - VSRAD V2, V1, V0 - -// Vector shift by octect immediate, VA-form with SHB 4-bit field -// SHB,VRA,VRB,VRT produces -// VRT,VRA,VRB,SHB - VSLDOI $4, V2, V1, V0 - -// Vector merge odd and even word -// VRA,VRB,VRT produces -// VRT,VRA,VRB - - VMRGOW V4,V5,V6 - VMRGEW V4,V5,V6 - -// Vector count, VX-form -// VRB,VRT produces -// VRT,VRB - VCLZB V4, V5 - VCLZH V4, V5 - VCLZW V4, V5 - VCLZD V4, V5 - VPOPCNTB V4, V5 - VPOPCNTH V4, V5 - VPOPCNTW V4, V5 - VPOPCNTD V4, V5 - -// Vector compare, VC-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB -// * Note: 'CC' suffix denotes Rc=1 -// i.e. vcmpequb. v3,v1,v2 equals VCMPEQUBCC V1,V2,V3 - VCMPEQUB V3, V2, V1 - VCMPEQUBCC V3, V2, V1 - VCMPEQUH V3, V2, V1 - VCMPEQUHCC V3, V2, V1 - VCMPEQUW V3, V2, V1 - VCMPEQUWCC V3, V2, V1 - VCMPEQUD V3, V2, V1 - VCMPEQUDCC V3, V2, V1 - VCMPGTUB V3, V2, V1 - VCMPGTUBCC V3, V2, V1 - VCMPGTUH V3, V2, V1 - VCMPGTUHCC V3, V2, V1 - VCMPGTUW V3, V2, V1 - VCMPGTUWCC V3, V2, V1 - VCMPGTUD V3, V2, V1 - VCMPGTUDCC V3, V2, V1 - VCMPGTSB V3, V2, V1 - VCMPGTSBCC V3, V2, V1 - VCMPGTSH V3, V2, V1 - VCMPGTSHCC V3, V2, V1 - VCMPGTSW V3, V2, V1 - VCMPGTSWCC V3, V2, V1 - VCMPGTSD V3, V2, V1 - VCMPGTSDCC V3, V2, V1 - VCMPNEZB V3, V2, V1 - VCMPNEZBCC V3, V2, V1 - VCMPNEB V3, V2, V1 - VCMPNEBCC V3, V2, V1 - VCMPNEH V3, V2, V1 - VCMPNEHCC V3, V2, V1 - VCMPNEW V3, V2, V1 - VCMPNEWCC V3, V2, V1 - -// Vector permute, VA-form -// VRA,VRB,VRC,VRT produces -// VRT,VRA,VRB,VRC - VPERM V3, V2, V1, V0 - VPERMXOR V3, V2, V1, V0 - VPERMR V3, V2, V1, V0 - -// Vector bit permute, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VBPERMQ V3,V1,V2 - VBPERMD V3,V1,V2 - -// Vector select, VA-form -// VRA,VRB,VRC,VRT produces -// VRT,VRA,VRB,VRC - VSEL V3, V2, V1, V0 - -// Vector splat, VX-form with 4-bit UIM field -// UIM,VRB,VRT produces -// VRT,VRB,UIM - VSPLTB $15, V1, V0 - VSPLTH $7, V1, V0 - VSPLTW $3, V1, V0 - -// Vector splat immediate signed, VX-form with 5-bit SIM field -// SIM,VRT produces -// VRT,SIM - VSPLTISB $31, V4 - VSPLTISH $31, V4 - VSPLTISW $31, V4 - -// Vector AES cipher, VX-form -// VRA,VRB,VRT produces -// VRT,VRA,VRB - VCIPHER V3, V2, V1 - VCIPHERLAST V3, V2, V1 - VNCIPHER V3, V2, V1 - VNCIPHERLAST V3, V2, V1 - -// Vector AES subbytes, VX-form -// VRA,VRT produces -// VRT,VRA - VSBOX V2, V1 - -// Vector SHA, VX-form with ST bit field and 4-bit SIX field -// SIX,VRA,ST,VRT produces -// VRT,VRA,ST,SIX - VSHASIGMAW $15, V1, $1, V0 - VSHASIGMAD $15, V1, $1, V0 - -// VSX instructions -// Described as: -// , -// produces -// - -// VSX load, XX1-form -// (RB)(RA*1),XT produces -// XT,RA,RB - LXVD2X (R1)(R2*1), VS0 - LXVW4X (R1)(R2*1), VS0 - LXVH8X (R1)(R2*1), VS0 - LXVB16X (R1)(R2*1), VS0 - LXVDSX (R1)(R2*1), VS0 - LXSDX (R1)(R2*1), VS0 - LXSIWAX (R1)(R2*1), VS0 - LXSIWZX (R1)(R2*1), VS0 - -// VSX load with length X-form (also left-justified) - LXVL R3,R4, VS0 - LXVLL R3,R4, VS0 - LXVX R3,R4, VS0 -// VSX load, DQ-form -// DQ(RA), XS produces -// XS, DQ(RA) - LXV 32752(R1), VS0 - -// VSX store, XX1-form -// XS,(RB)(RA*1) produces -// XS,RA,RB - STXVD2X VS63, (R1)(R2*1) - STXVW4X VS63, (R1)(R2*1) - STXVH8X VS63, (R1)(R2*1) - STXVB16X VS63, (R1)(R2*1) - STXSDX VS63, (R1)(R2*1) - STXSIWX VS63, (R1)(R2*1) - -// VSX store, DQ-form -// DQ(RA), XS produces -// XS, DQ(RA) - STXV VS63, -32752(R1) - -// VSX store with length, X-form (also left-justified) - STXVL VS0, R3,R4 - STXVLL VS0, R3,R4 - STXVX VS0, R3,R4 - -// VSX move from VSR, XX1-form -// XS,RA produces -// RA,XS -// Extended mnemonics accept VMX and FP registers as sources - MFVSRD VS0, R1 - MFVSRWZ VS33, R1 - MFVSRLD VS63, R1 - MFVRD V0, R1 - MFFPRD F0, R1 - -// VSX move to VSR, XX1-form -// RA,XT produces -// XT,RA -// Extended mnemonics accept VMX and FP registers as targets - MTVSRD R1, VS0 - MTVSRWA R1, VS31 - MTVSRWZ R1, VS63 - MTVSRDD R1, R2, VS0 - MTVSRWS R1, VS32 - MTVRD R1, V13 - MTFPRD R1, F24 - -// VSX AND, XX3-form -// XA,XB,XT produces -// XT,XA,XB - XXLAND VS0,VS1,VS32 - XXLANDC VS0,VS1,VS32 - XXLEQV VS0,VS1,VS32 - XXLNAND VS0,VS1,VS32 - -// VSX OR, XX3-form -// XA,XB,XT produces -// XT,XA,XB - XXLORC VS0,VS1,VS32 - XXLNOR VS0,VS1,VS32 - XXLORQ VS0,VS1,VS32 - XXLXOR VS0,VS1,VS32 - XXLOR VS0,VS1,VS32 - -// VSX select, XX4-form -// XA,XB,XC,XT produces -// XT,XA,XB,XC - XXSEL VS0,VS1,VS3,VS32 - -// VSX merge, XX3-form -// XA,XB,XT produces -// XT,XA,XB - XXMRGHW VS0,VS1,VS32 - XXMRGLW VS0,VS1,VS32 - -// VSX splat, XX2-form -// XB,UIM,XT produces -// XT,XB,UIM - XXSPLTW VS0,$3,VS32 - XXSPLTIB $26,VS0 - -// VSX permute, XX3-form -// XA,XB,XT produces -// XT,XA,XB - XXPERM VS0,VS1,VS32 - -// VSX permute, XX3-form -// XA,XB,DM,XT produces -// XT,XA,XB,DM - XXPERMDI VS0,VS1,$3,VS32 - -// VSX shift, XX3-form -// XA,XB,SHW,XT produces -// XT,XA,XB,SHW - XXSLDWI VS0,VS1,$3,VS32 - -// VSX byte-reverse XX2-form -// XB,XT produces -// XT,XB - XXBRQ VS0,VS1 - XXBRD VS0,VS1 - XXBRW VS0,VS1 - XXBRH VS0,VS1 - -// VSX scalar FP-FP conversion, XX2-form -// XB,XT produces -// XT,XB - XSCVDPSP VS0,VS32 - XSCVSPDP VS0,VS32 - XSCVDPSPN VS0,VS32 - XSCVSPDPN VS0,VS32 - -// VSX vector FP-FP conversion, XX2-form -// XB,XT produces -// XT,XB - XVCVDPSP VS0,VS32 - XVCVSPDP VS0,VS32 - -// VSX scalar FP-integer conversion, XX2-form -// XB,XT produces -// XT,XB - XSCVDPSXDS VS0,VS32 - XSCVDPSXWS VS0,VS32 - XSCVDPUXDS VS0,VS32 - XSCVDPUXWS VS0,VS32 - -// VSX scalar integer-FP conversion, XX2-form -// XB,XT produces -// XT,XB - XSCVSXDDP VS0,VS32 - XSCVUXDDP VS0,VS32 - XSCVSXDSP VS0,VS32 - XSCVUXDSP VS0,VS32 - -// VSX vector FP-integer conversion, XX2-form -// XB,XT produces -// XT,XB - XVCVDPSXDS VS0,VS32 - XVCVDPSXWS VS0,VS32 - XVCVDPUXDS VS0,VS32 - XVCVDPUXWS VS0,VS32 - XVCVSPSXDS VS0,VS32 - XVCVSPSXWS VS0,VS32 - XVCVSPUXDS VS0,VS32 - XVCVSPUXWS VS0,VS32 - -// VSX scalar integer-FP conversion, XX2-form -// XB,XT produces -// XT,XB - XVCVSXDDP VS0,VS32 - XVCVSXWDP VS0,VS32 - XVCVUXDDP VS0,VS32 - XVCVUXWDP VS0,VS32 - XVCVSXDSP VS0,VS32 - XVCVSXWSP VS0,VS32 - XVCVUXDSP VS0,VS32 - XVCVUXWSP VS0,VS32 - -// Multiply-Add High Doubleword -// RA,RB,RC,RT produces -// RT,RA,RB,RC - MADDHD R1,R2,R3,R4 - MADDHDU R1,R2,R3,R4 - -// Add Extended using alternate carry bit -// ADDEX RA,RB,CY,RT produces -// addex RT, RA, RB, CY - ADDEX R1, R2, $0, R3 - -// Immediate-shifted operations -// ADDIS SI, RA, RT produces -// addis RT, RA, SI - ADDIS $8, R3, R4 - ADDIS $-1, R3, R4 - -// ANDISCC UI, RS, RA produces -// andis. RA, RS, UI - ANDISCC $7, R4, R5 - -// ORIS UI, RS, RA produces -// oris RA, RS, UI - ORIS $4, R2, R3 - -// XORIS UI, RS, RA produces -// xoris RA, RS, UI - XORIS $1, R1, R2 - -// -// NOP -// -// LNOP comma // asm doesn't support the trailing comma. -// { -// outcode(int($1), &nullgen, 0, &nullgen); -// } NOP - -// LNOP rreg comma // asm doesn't support the trailing comma. -// { -// outcode(int($1), &$2, 0, &nullgen); -// } NOP R2 + NOP F2 + NOP $4 -// LNOP freg comma // asm doesn't support the trailing comma. -// { -// outcode(int($1), &$2, 0, &nullgen); -// } - NOP F2 + CRAND CR1, CR2, CR3 // 4c620a02 + CRANDN CR1, CR2, CR3 // 4c620902 + CREQV CR1, CR2, CR3 // 4c620a42 + CRNAND CR1, CR2, CR3 // 4c6209c2 + CRNOR CR1, CR2, CR3 // 4c620842 + CROR CR1, CR2, CR3 // 4c620b82 + CRORN CR1, CR2, CR3 // 4c620b42 + CRXOR CR1, CR2, CR3 // 4c620982 -// LNOP ',' rreg // asm doesn't support the leading comma. -// { -// outcode(int($1), &nullgen, 0, &$3); -// } - NOP R2 + ISEL $1, R3, R4, R5 // 7ca3205e + ISEL $0, R3, R4, R5 // 7ca3201e + ISEL $2, R3, R4, R5 // 7ca3209e + ISEL $3, R3, R4, R5 // 7ca320de + ISEL $4, R3, R4, R5 // 7ca3211e + POPCNTB R3, R4 // 7c6400f4 + POPCNTW R3, R4 // 7c6402f4 + POPCNTD R3, R4 // 7c6403f4 -// LNOP ',' freg // asm doesn't support the leading comma. -// { -// outcode(int($1), &nullgen, 0, &$3); -// } - NOP F2 + PASTECC R3, R4 // 7c23270d + COPY R3, R4 // 7c23260c -// LNOP imm // SYSCALL $num: load $num to R0 before syscall and restore R0 to 0 afterwards. -// { -// outcode(int($1), &$2, 0, &nullgen); -// } - NOP $4 + // load-and-reserve + LBAR (R4)(R3*1),$1,R5 // 7ca32069 + LBAR (R4),$0,R5 // 7ca02068 + LBAR (R3),R5 // 7ca01868 + LHAR (R4)(R3*1),$1,R5 // 7ca320e9 + LHAR (R4),$0,R5 // 7ca020e8 + LHAR (R3),R5 // 7ca018e8 + LWAR (R4)(R3*1),$1,R5 // 7ca32029 + LWAR (R4),$0,R5 // 7ca02028 + LWAR (R3),R5 // 7ca01828 + LDAR (R4)(R3*1),$1,R5 // 7ca320a9 + LDAR (R4),$0,R5 // 7ca020a8 + LDAR (R3),R5 // 7ca018a8 + + STBCCC R3, (R4)(R5) // 7c65256d + STWCCC R3, (R4)(R5) // 7c65212d + STDCCC R3, (R4)(R5) // 7c6521ad + STHCCC R3, (R4)(R5) + STSW R3, (R4)(R5) + + SYNC // 7c0004ac + ISYNC // 4c00012c + LWSYNC // 7c2004ac + + DARN $1, R5 // 7ca105e6 + + DCBF (R3)(R4) // 7c0418ac + DCBI (R3)(R4) // 7c041bac + DCBST (R3)(R4) // 7c04186c + DCBZ (R3)(R4) // 7c041fec + DCBT (R3)(R4) // 7c041a2c + ICBI (R3)(R4) // 7c041fac + + // float constants + FMOVD $(0.0), F1 // f0210cd0 + FMOVD $(-0.0), F1 // f0210cd0fc200850 + + FMOVD 8(R3), F1 // c8230008 + FMOVD (R3)(R4), F1 // 7c241cae + FMOVDU 8(R3), F1 // cc230008 + FMOVDU (R3)(R4), F1 // 7c241cee + FMOVS 4(R3), F1 // c0230004 + FMOVS (R3)(R4), F1 // 7c241c2e + FMOVSU 4(R3), F1 // c4230004 + FMOVSU (R3)(R4), F1 // 7c241c6e + + FMOVD F1, 8(R3) // d8230008 + FMOVD F1, (R3)(R4) // 7c241dae + FMOVDU F1, 8(R3) // dc230008 + FMOVDU F1, (R3)(R4) // 7c241dee + FMOVS F1, 4(R3) // d0230004 + FMOVS F1, (R3)(R4) // 7c241d2e + FMOVSU F1, 4(R3) // d4230004 + FMOVSU F1, (R3)(R4) // 7c241d6e + FADD F1, F2 // fc42082a + FADD F1, F2, F3 // fc62082a + FADDCC F1, F2, F3 // fc62082b + FADDS F1, F2 // ec42082a + FADDS F1, F2, F3 // ec62082a + FADDSCC F1, F2, F3 // ec62082b + FSUB F1, F2 // fc420828 + FSUB F1, F2, F3 // fc620828 + FSUBCC F1, F2, F3 // fc620829 + FSUBS F1, F2 // ec420828 + FSUBS F1, F2, F3 // ec620828 + FSUBCC F1, F2, F3 // fc620829 + FMUL F1, F2 // fc420072 + FMUL F1, F2, F3 // fc620072 + FMULCC F1, F2, F3 // fc620073 + FMULS F1, F2 // ec420072 + FMULS F1, F2, F3 // ec620072 + FMULSCC F1, F2, F3 // ec620073 + FDIV F1, F2 // fc420824 + FDIV F1, F2, F3 // fc620824 + FDIVCC F1, F2, F3 // fc620825 + FDIVS F1, F2 // ec420824 + FDIVS F1, F2, F3 // ec620824 + FDIVSCC F1, F2, F3 // ec620825 + FMADD F1, F2, F3, F4 // fc8110fa + FMADDCC F1, F2, F3, F4 // fc8110fb + FMADDS F1, F2, F3, F4 // ec8110fa + FMADDSCC F1, F2, F3, F4 // ec8110fb + FMSUB F1, F2, F3, F4 // fc8110f8 + FMSUBCC F1, F2, F3, F4 // fc8110f9 + FMSUBS F1, F2, F3, F4 // ec8110f8 + FMSUBSCC F1, F2, F3, F4 // ec8110f9 + FNMADD F1, F2, F3, F4 // fc8110fe + FNMADDCC F1, F2, F3, F4 // fc8110ff + FNMADDS F1, F2, F3, F4 // ec8110fe + FNMADDSCC F1, F2, F3, F4 // ec8110ff + FNMSUB F1, F2, F3, F4 // fc8110fc + FNMSUBCC F1, F2, F3, F4 // fc8110fd + FNMSUBS F1, F2, F3, F4 // ec8110fc + FNMSUBSCC F1, F2, F3, F4 // ec8110fd + FSEL F1, F2, F3, F4 // fc8110ee + FSELCC F1, F2, F3, F4 // fc8110ef + FABS F1, F2 // fc400a10 + FABSCC F1, F2 // fc400a11 + FNEG F1, F2 // fc400850 + FABSCC F1, F2 // fc400a11 + FRSP F1, F2 // fc400818 + FRSPCC F1, F2 // fc400819 + FCTIW F1, F2 // fc40081c + FCTIWCC F1, F2 // fc40081d + FCTIWZ F1, F2 // fc40081e + FCTIWZCC F1, F2 // fc40081f + FCTID F1, F2 // fc400e5c + FCTIDCC F1, F2 // fc400e5d + FCTIDZ F1, F2 // fc400e5e + FCTIDZCC F1, F2 // fc400e5f + FCFID F1, F2 // fc400e9c + FCFIDCC F1, F2 // fc400e9d + FCFIDU F1, F2 // fc400f9c + FCFIDUCC F1, F2 // fc400f9d + FCFIDS F1, F2 // ec400e9c + FCFIDSCC F1, F2 // ec400e9d + FRES F1, F2 // ec400830 + FRESCC F1, F2 // ec400831 + FRIM F1, F2 // fc400bd0 + FRIMCC F1, F2 // fc400bd1 + FRIP F1, F2 // fc400b90 + FRIPCC F1, F2 // fc400b91 + FRIZ F1, F2 // fc400b50 + FRIZCC F1, F2 // fc400b51 + FRIN F1, F2 // fc400b10 + FRINCC F1, F2 // fc400b11 + FRSQRTE F1, F2 // fc400834 + FRSQRTECC F1, F2 // fc400835 + FSQRT F1, F2 // fc40082c + FSQRTCC F1, F2 // fc40082d + FSQRTS F1, F2 // ec40082c + FSQRTSCC F1, F2 // ec40082d + FCPSGN F1, F2 // fc420810 + FCPSGNCC F1, F2 // fc420811 + FCMPO F1, F2 // fc011040 + FCMPU F1, F2 // fc011000 + LVX (R3)(R4), V1 // 7c2418ce + LVXL (R3)(R4), V1 // 7c241ace + LVSL (R3)(R4), V1 // 7c24180c + LVSR (R3)(R4), V1 // 7c24184c + LVEBX (R3)(R4), V1 // 7c24180e + LVEHX (R3)(R4), V1 // 7c24184e + LVEWX (R3)(R4), V1 // 7c24188e + STVX V1, (R3)(R4) // 7c2419ce + STVXL V1, (R3)(R4) // 7c241bce + STVEBX V1, (R3)(R4) // 7c24190e + STVEHX V1, (R3)(R4) // 7c24194e + STVEWX V1, (R3)(R4) // 7c24198e + + VAND V1, V2, V3 // 10611404 + VANDC V1, V2, V3 // 10611444 + VNAND V1, V2, V3 // 10611584 + VOR V1, V2, V3 // 10611484 + VORC V1, V2, V3 // 10611544 + VXOR V1, V2, V3 // 106114c4 + VNOR V1, V2, V3 // 10611504 + VEQV V1, V2, V3 // 10611684 + VADDUBM V1, V2, V3 // 10611000 + VADDUHM V1, V2, V3 // 10611040 + VADDUWM V1, V2, V3 // 10611080 + VADDUDM V1, V2, V3 // 106110c0 + VADDUQM V1, V2, V3 // 10611100 + VADDCUQ V1, V2, V3 // 10611140 + VADDCUW V1, V2, V3 // 10611180 + VADDUBS V1, V2, V3 // 10611200 + VADDUHS V1, V2, V3 // 10611240 + VADDUWS V1, V2, V3 // 10611280 + VSUBUBM V1, V2, V3 // 10611400 + VSUBUHM V1, V2, V3 // 10611440 + VSUBUWM V1, V2, V3 // 10611480 + VSUBUDM V1, V2, V3 // 106114c0 + VSUBUQM V1, V2, V3 // 10611500 + VSUBCUQ V1, V2, V3 // 10611540 + VSUBCUW V1, V2, V3 // 10611580 + VSUBUBS V1, V2, V3 // 10611600 + VSUBUHS V1, V2, V3 // 10611640 + VSUBUWS V1, V2, V3 // 10611680 + VSUBSBS V1, V2, V3 // 10611700 + VSUBSHS V1, V2, V3 // 10611740 + VSUBSWS V1, V2, V3 // 10611780 + VSUBEUQM V1, V2, V3, V4 // 108110fe + VSUBECUQ V1, V2, V3, V4 // 108110ff + VMULESB V1, V2, V3 // 10611308 + VMULOSB V1, V2, V3 // 10611108 + VMULEUB V1, V2, V3 // 10611208 + VMULOUB V1, V2, V3 // 10611008 + VMULESH V1, V2, V3 // 10611348 + VMULOSH V1, V2, V3 // 10611148 + VMULEUH V1, V2, V3 // 10611248 + VMULOUH V1, V2, V3 // 10611048 + VMULESH V1, V2, V3 // 10611348 + VMULOSW V1, V2, V3 // 10611188 + VMULEUW V1, V2, V3 // 10611288 + VMULOUW V1, V2, V3 // 10611088 + VMULUWM V1, V2, V3 // 10611089 + VPMSUMB V1, V2, V3 // 10611408 + VPMSUMH V1, V2, V3 // 10611448 + VPMSUMW V1, V2, V3 // 10611488 + VPMSUMD V1, V2, V3 // 106114c8 + VMSUMUDM V1, V2, V3, V4 // 108110e3 + VRLB V1, V2, V3 // 10611004 + VRLH V1, V2, V3 // 10611044 + VRLW V1, V2, V3 // 10611084 + VRLD V1, V2, V3 // 106110c4 + VSLB V1, V2, V3 // 10611104 + VSLH V1, V2, V3 // 10611144 + VSLW V1, V2, V3 // 10611184 + VSL V1, V2, V3 // 106111c4 + VSLO V1, V2, V3 // 1061140c + VSRB V1, V2, V3 // 10611204 + VSRH V1, V2, V3 // 10611244 + VSRW V1, V2, V3 // 10611284 + VSR V1, V2, V3 // 106112c4 + VSRO V1, V2, V3 // 1061144c + VSLD V1, V2, V3 // 106115c4 + VSRAB V1, V2, V3 // 10611304 + VSRAH V1, V2, V3 // 10611344 + VSRAW V1, V2, V3 // 10611384 + VSRAD V1, V2, V3 // 106113c4 + VSLDOI $3, V1, V2, V3 // 106110ec + VCLZB V1, V2 // 10400f02 + VCLZH V1, V2 // 10400f42 + VCLZW V1, V2 // 10400f82 + VCLZD V1, V2 // 10400fc2 + VPOPCNTB V1, V2 // 10400f03 + VPOPCNTH V1, V2 // 10400f43 + VPOPCNTW V1, V2 // 10400f83 + VPOPCNTD V1, V2 // 10400fc3 + VCMPEQUB V1, V2, V3 // 10611006 + VCMPEQUBCC V1, V2, V3 // 10611406 + VCMPEQUH V1, V2, V3 // 10611046 + VCMPEQUHCC V1, V2, V3 // 10611446 + VCMPEQUW V1, V2, V3 // 10611086 + VCMPEQUWCC V1, V2, V3 // 10611486 + VCMPEQUD V1, V2, V3 // 106110c7 + VCMPEQUDCC V1, V2, V3 // 106114c7 + VCMPGTUB V1, V2, V3 // 10611206 + VCMPGTUBCC V1, V2, V3 // 10611606 + VCMPGTUH V1, V2, V3 // 10611246 + VCMPGTUHCC V1, V2, V3 // 10611646 + VCMPGTUW V1, V2, V3 // 10611286 + VCMPGTUWCC V1, V2, V3 // 10611686 + VCMPGTUD V1, V2, V3 // 106112c7 + VCMPGTUDCC V1, V2, V3 // 106116c7 + VCMPGTSB V1, V2, V3 // 10611306 + VCMPGTSBCC V1, V2, V3 // 10611706 + VCMPGTSH V1, V2, V3 // 10611346 + VCMPGTSHCC V1, V2, V3 // 10611746 + VCMPGTSW V1, V2, V3 // 10611386 + VCMPGTSWCC V1, V2, V3 // 10611786 + VCMPGTSD V1, V2, V3 // 106113c7 + VCMPGTSDCC V1, V2, V3 // 106117c7 + VCMPNEZB V1, V2, V3 // 10611107 + VCMPNEZBCC V1, V2, V3 // 10611507 + VCMPNEB V1, V2, V3 // 10611007 + VCMPNEBCC V1, V2, V3 // 10611407 + VCMPNEH V1, V2, V3 // 10611047 + VCMPNEHCC V1, V2, V3 // 10611447 + VCMPNEW V1, V2, V3 // 10611087 + VCMPNEWCC V1, V2, V3 // 10611487 + VPERM V1, V2, V3, V4 // 108110eb + VPERMR V1, V2, V3, V4 // 108110fb + VPERMXOR V1, V2, V3, V4 // 108110ed + VBPERMQ V1, V2, V3 // 1061154c + VBPERMD V1, V2, V3 // 106115cc + VSEL V1, V2, V3, V4 // 108110ea + VSPLTB $1, V1, V2 // 10410a0c + VSPLTH $1, V1, V2 // 10410a4c + VSPLTW $1, V1, V2 // 10410a8c + VSPLTISB $1, V1 // 1021030c + VSPLTISW $1, V1 // 1021038c + VSPLTISH $1, V1 // 1021034c + VCIPHER V1, V2, V3 // 10611508 + VCIPHERLAST V1, V2, V3 // 10611509 + VNCIPHER V1, V2, V3 // 10611548 + VNCIPHERLAST V1, V2, V3 // 10611549 + VSBOX V1, V2 // 104105c8 + VSHASIGMAW $1, V1, $15, V2 // 10418e82 + VSHASIGMAD $2, V1, $15, V2 // 104196c2 + + LXVD2X (R3)(R4), VS1 // 7c241e98 + LXVDSX (R3)(R4), VS1 // 7c241a98 + LXVH8X (R3)(R4), VS1 // 7c241e58 + LXVB16X (R3)(R4), VS1 // 7c241ed8 + LXVW4X (R3)(R4), VS1 // 7c241e18 + LXV 16(R3), VS1 // f4230011 + LXVL R3, R4, VS1 // 7c23221a + LXVLL R3, R4, VS1 // 7c23225a + LXVX R3, R4, VS1 // 7c232218 + LXSDX (R3)(R4), VS1 // 7c241c98 + STXVD2X VS1, (R3)(R4) // 7c241f98 + STXV VS1,16(R3) // f4230015 + STXVL VS1, R3, R4 // 7c23231a + STXVLL VS1, R3, R4 // 7c23235a + STXVX VS1, R3, R4 // 7c232318 + STXVB16X VS1, (R4)(R5) // 7c2527d8 + STXVH8X VS1, (R4)(R5) // 7c252758 + + STXSDX VS1, (R3)(R4) // 7c241d98 + LXSIWAX (R3)(R4), VS1 // 7c241898 + STXSIWX VS1, (R3)(R4) // 7c241918 + MFVSRD VS1, R3 // 7c230066 + MTFPRD R3, F0 // 7c030166 + MFVRD V0, R3 // 7c030067 + MFVSRLD VS63,R4 // 7fe40267 + MFVSRWZ VS33,R4 // 7c2400e7 + MTVSRD R3, VS1 // 7c230166 + MTVRD R3, V13 // 7da30167 + MTVSRWA R4, VS31 // 7fe401a6 + MTVSRWS R4, VS32 // 7c040327 + MTVSRWZ R4, VS63 // 7fe401e7 + XXBRD VS0, VS1 // f037076c + XXBRW VS1, VS2 // f04f0f6c + XXBRH VS2, VS3 // f067176c + XXLAND VS1, VS2, VS3 // f0611410 + XXLANDC VS1, VS2, VS3 // f0611450 + XXLEQV VS0, VS1, VS2 // f0400dd0 + XXLNAND VS0, VS1, VS2 // f0400d90 + XXLNOR VS0, VS1, VS32 // f0000d11 + XXLOR VS1, VS2, VS3 // f0611490 + XXLORC VS1, VS2, VS3 // f0611550 + XXLORQ VS1, VS2, VS3 // f0611490 + XXLXOR VS1, VS2, VS3 // f06114d0 + XXSEL VS1, VS2, VS3, VS4 // f08110f0 + XXMRGHW VS1, VS2, VS3 // f0611090 + XXMRGLW VS1, VS2, VS3 // f0611190 + XXSPLTW VS1, $1, VS2 // f0410a90 + XXPERM VS1, VS2, VS3 // f06110d0 + XXSLDWI VS1, VS2, $1, VS3 // f0611110 + XSCVDPSP VS1, VS2 // f0400c24 + XVCVDPSP VS1, VS2 // f0400e24 + XSCVSXDDP VS1, VS2 // f0400de0 + XVCVDPSXDS VS1, VS2 // f0400f60 + XVCVSXDDP VS1, VS2 // f0400fe0 + XSCVDPSPN VS1,VS32 // f0000c2d + XSCVDPSP VS1,VS32 // f0000c25 + XSCVDPSXDS VS1,VS32 // f0000d61 + XSCVDPSXWS VS1,VS32 // f0000961 + XSCVDPUXDS VS1,VS32 // f0000d21 + XSCVDPUXWS VS1,VS32 // f0000921 + XSCVSPDPN VS1,VS32 // f0000d2d + XSCVSPDP VS1,VS32 // f0000d25 + XSCVSXDDP VS1,VS32 // f0000de1 + XSCVSXDSP VS1,VS32 // f0000ce1 + XSCVUXDDP VS1,VS32 // f0000da1 + XSCVUXDSP VS1,VS32 // f0000ca1 + XVCVDPSP VS1,VS32 // f0000e25 + XVCVDPSXDS VS1,VS32 // f0000f61 + XVCVDPSXWS VS1,VS32 // f0000b61 + XVCVDPUXDS VS1,VS32 // f0000f21 + XVCVDPUXWS VS1,VS32 // f0000b21 + XVCVSPDP VS1,VS32 // f0000f25 + XVCVSPSXDS VS1,VS32 // f0000e61 + XVCVSPSXWS VS1,VS32 // f0000a61 + XVCVSPUXDS VS1,VS32 // f0000e21 + XVCVSPUXWS VS1,VS32 // f0000a21 + XVCVSXDDP VS1,VS32 // f0000fe1 + XVCVSXDSP VS1,VS32 // f0000ee1 + XVCVSXWDP VS1,VS32 // f0000be1 + XVCVSXWSP VS1,VS32 // f0000ae1 + XVCVUXDDP VS1,VS32 // f0000fa1 + XVCVUXDSP VS1,VS32 // f0000ea1 + XVCVUXWDP VS1,VS32 // f0000ba1 + XVCVUXWSP VS1,VS32 // f0000aa1 + + MOVD R3, LR // 7c6803a6 + MOVD R3, CTR // 7c6903a6 + MOVD R3, XER // 7c6103a6 + MOVD LR, R3 // 7c6802a6 + MOVD CTR, R3 // 7c6902a6 + MOVD XER, R3 // 7c6102a6 + MOVFL CR3, CR1 // 4c8c0000 -// RET -// -// LRETRN comma // asm doesn't support the trailing comma. -// { -// outcode(int($1), &nullgen, 0, &nullgen); -// } - BEQ 2(PC) RET - -// More BR/BL cases, and canonical names JMP, CALL. - - BEQ 2(PC) - BR foo(SB) // JMP foo(SB) - BL foo(SB) // CALL foo(SB) - BEQ 2(PC) - JMP foo(SB) - CALL foo(SB) - RET foo(SB) - -// load-and-reserve -// L*AR (RB)(RA*1),EH,RT produces -// l*arx RT,RA,RB,EH -// -// Extended forms also accepted. Assumes RA=0, EH=0: -// L*AR (RB),RT -// L*AR (RB),EH,RT - LBAR (R4)(R3*1), $1, R5 - LBAR (R4), $0, R5 - LBAR (R3), R5 - LHAR (R4)(R3*1), $1, R5 - LHAR (R4), $0, R5 - LHAR (R3), R5 - LWAR (R4)(R3*1), $1, R5 - LWAR (R4), $0, R5 - LWAR (R3), R5 - LDAR (R4)(R3*1), $1, R5 - LDAR (R4), $0, R5 - LDAR (R3), R5 - -// END -// -// LEND comma // asm doesn't support the trailing comma. -// { -// outcode(int($1), &nullgen, 0, &nullgen); -// } - END diff --git a/src/cmd/asm/internal/asm/testdata/ppc64enc.s b/src/cmd/asm/internal/asm/testdata/ppc64enc.s deleted file mode 100644 index 10a05ec402b..00000000000 --- a/src/cmd/asm/internal/asm/testdata/ppc64enc.s +++ /dev/null @@ -1,633 +0,0 @@ -// Copyright 2018 The Go Authors. All rights reserved. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// Initial set of opcode combinations based on -// improvements to processing of constant -// operands. - -// Full set will be added at a later date. - -#include "../../../../../runtime/textflag.h" - -TEXT asmtest(SB),DUPOK|NOSPLIT,$0 - // move constants - MOVD $1, R3 // 38600001 - MOVD $-1, R4 // 3880ffff - MOVD $65535, R5 // 6005ffff - MOVD $65536, R6 // 64060001 - MOVD $-32767, R5 // 38a08001 - MOVD $-32768, R6 // 38c08000 - MOVD $1234567, R5 // 6405001260a5d687 - MOVW $1, R3 // 38600001 - MOVW $-1, R4 // 3880ffff - MOVW $65535, R5 // 6005ffff - MOVW $65536, R6 // 64060001 - MOVW $-32767, R5 // 38a08001 - MOVW $-32768, R6 // 38c08000 - MOVW $1234567, R5 // 6405001260a5d687 - MOVD 8(R3), R4 // e8830008 - MOVD (R3)(R4), R5 // 7ca4182a - MOVW 4(R3), R4 // e8830006 - MOVW (R3)(R4), R5 // 7ca41aaa - MOVWZ 4(R3), R4 // 80830004 - MOVWZ (R3)(R4), R5 // 7ca4182e - MOVH 4(R3), R4 // a8830004 - MOVH (R3)(R4), R5 // 7ca41aae - MOVHZ 2(R3), R4 // a0830002 - MOVHZ (R3)(R4), R5 // 7ca41a2e - MOVB 1(R3), R4 // 888300017c840774 - MOVB (R3)(R4), R5 // 7ca418ae7ca50774 - MOVBZ 1(R3), R4 // 88830001 - MOVBZ (R3)(R4), R5 // 7ca418ae - MOVDBR (R3)(R4), R5 // 7ca41c28 - MOVWBR (R3)(R4), R5 // 7ca41c2c - MOVHBR (R3)(R4), R5 // 7ca41e2c - - MOVDU 8(R3), R4 // e8830009 - MOVDU (R3)(R4), R5 // 7ca4186a - MOVWU (R3)(R4), R5 // 7ca41aea - MOVWZU 4(R3), R4 // 84830004 - MOVWZU (R3)(R4), R5 // 7ca4186e - MOVHU 2(R3), R4 // ac830002 - MOVHU (R3)(R4), R5 // 7ca41aee - MOVHZU 2(R3), R4 // a4830002 - MOVHZU (R3)(R4), R5 // 7ca41a6e - MOVBU 1(R3), R4 // 8c8300017c840774 - MOVBU (R3)(R4), R5 // 7ca418ee7ca50774 - MOVBZU 1(R3), R4 // 8c830001 - MOVBZU (R3)(R4), R5 // 7ca418ee - - MOVD R4, 8(R3) // f8830008 - MOVD R5, (R3)(R4) // 7ca4192a - MOVW R4, 4(R3) // 90830004 - MOVW R5, (R3)(R4) // 7ca4192e - MOVH R4, 2(R3) // b0830002 - MOVH R5, (R3)(R4) // 7ca41b2e - MOVB R4, 1(R3) // 98830001 - MOVB R5, (R3)(R4) // 7ca419ae - MOVDBR R5, (R3)(R4) // 7ca41d28 - MOVWBR R5, (R3)(R4) // 7ca41d2c - MOVHBR R5, (R3)(R4) // 7ca41f2c - - MOVDU R4, 8(R3) // f8830009 - MOVDU R5, (R3)(R4) // 7ca4196a - MOVWU R4, 4(R3) // 94830004 - MOVWU R5, (R3)(R4) // 7ca4196e - MOVHU R4, 2(R3) // b4830002 - MOVHU R5, (R3)(R4) // 7ca41b6e - MOVBU R4, 1(R3) // 9c830001 - MOVBU R5, (R3)(R4) // 7ca419ee - - ADD $1, R3 // 38630001 - ADD $1, R3, R4 // 38830001 - ADD $-1, R4 // 3884ffff - ADD $-1, R4, R5 // 38a4ffff - ADD $65535, R5 // 601fffff7cbf2a14 - ADD $65535, R5, R6 // 601fffff7cdf2a14 - ADD $65536, R6 // 3cc60001 - ADD $65536, R6, R7 // 3ce60001 - ADD $-32767, R5 // 38a58001 - ADD $-32767, R5, R4 // 38858001 - ADD $-32768, R6 // 38c68000 - ADD $-32768, R6, R5 // 38a68000 - ADD $1234567, R5 // 641f001263ffd6877cbf2a14 - ADD $1234567, R5, R6 // 641f001263ffd6877cdf2a14 - ADDIS $8, R3 // 3c630008 - ADDIS $1000, R3, R4 // 3c8303e8 - - ANDCC $1, R3 // 70630001 - ANDCC $1, R3, R4 // 70640001 - ANDCC $-1, R4 // 3be0ffff7fe42039 - ANDCC $-1, R4, R5 // 3be0ffff7fe52039 - ANDCC $65535, R5 // 70a5ffff - ANDCC $65535, R5, R6 // 70a6ffff - ANDCC $65536, R6 // 74c60001 - ANDCC $65536, R6, R7 // 74c70001 - ANDCC $-32767, R5 // 3be080017fe52839 - ANDCC $-32767, R5, R4 // 3be080017fe42839 - ANDCC $-32768, R6 // 3be080007fe63039 - ANDCC $-32768, R5, R6 // 3be080007fe62839 - ANDCC $1234567, R5 // 641f001263ffd6877fe52839 - ANDCC $1234567, R5, R6 // 641f001263ffd6877fe62839 - ANDISCC $1, R3 // 74630001 - ANDISCC $1000, R3, R4 // 746403e8 - - OR $1, R3 // 60630001 - OR $1, R3, R4 // 60640001 - OR $-1, R4 // 3be0ffff7fe42378 - OR $-1, R4, R5 // 3be0ffff7fe52378 - OR $65535, R5 // 60a5ffff - OR $65535, R5, R6 // 60a6ffff - OR $65536, R6 // 64c60001 - OR $65536, R6, R7 // 64c70001 - OR $-32767, R5 // 3be080017fe52b78 - OR $-32767, R5, R6 // 3be080017fe62b78 - OR $-32768, R6 // 3be080007fe63378 - OR $-32768, R6, R7 // 3be080007fe73378 - OR $1234567, R5 // 641f001263ffd6877fe52b78 - OR $1234567, R5, R3 // 641f001263ffd6877fe32b78 - - XOR $1, R3 // 68630001 - XOR $1, R3, R4 // 68640001 - XOR $-1, R4 // 3be0ffff7fe42278 - XOR $-1, R4, R5 // 3be0ffff7fe52278 - XOR $65535, R5 // 68a5ffff - XOR $65535, R5, R6 // 68a6ffff - XOR $65536, R6 // 6cc60001 - XOR $65536, R6, R7 // 6cc70001 - XOR $-32767, R5 // 3be080017fe52a78 - XOR $-32767, R5, R6 // 3be080017fe62a78 - XOR $-32768, R6 // 3be080007fe63278 - XOR $-32768, R6, R7 // 3be080007fe73278 - XOR $1234567, R5 // 641f001263ffd6877fe52a78 - XOR $1234567, R5, R3 // 641f001263ffd6877fe32a78 - - // TODO: the order of CR operands don't match - CMP R3, R4 // 7c232000 - CMPU R3, R4 // 7c232040 - CMPW R3, R4 // 7c032000 - CMPWU R3, R4 // 7c032040 - - // TODO: constants for ADDC? - ADD R3, R4 // 7c841a14 - ADD R3, R4, R5 // 7ca41a14 - ADDC R3, R4 // 7c841814 - ADDC R3, R4, R5 // 7ca41814 - ADDE R3, R4 // 7c841914 - ADDECC R3, R4 // 7c841915 - ADDEV R3, R4 // 7c841d14 - ADDEVCC R3, R4 // 7c841d15 - ADDV R3, R4 // 7c841e14 - ADDVCC R3, R4 // 7c841e15 - ADDCCC R3, R4, R5 // 7ca41815 - ADDME R3, R4 // 7c8301d4 - ADDMECC R3, R4 // 7c8301d5 - ADDMEV R3, R4 // 7c8305d4 - ADDMEVCC R3, R4 // 7c8305d5 - ADDCV R3, R4 // 7c841c14 - ADDCVCC R3, R4 // 7c841c15 - ADDZE R3, R4 // 7c830194 - ADDZECC R3, R4 // 7c830195 - ADDZEV R3, R4 // 7c830594 - ADDZEVCC R3, R4 // 7c830595 - SUBME R3, R4 // 7c8301d0 - SUBMECC R3, R4 // 7c8301d1 - SUBMEV R3, R4 // 7c8305d0 - SUBZE R3, R4 // 7c830190 - SUBZECC R3, R4 // 7c830191 - SUBZEV R3, R4 // 7c830590 - SUBZEVCC R3, R4 // 7c830591 - - AND R3, R4 // 7c841838 - AND R3, R4, R5 // 7c851838 - ANDN R3, R4, R5 // 7c851878 - ANDCC R3, R4, R5 // 7c851839 - OR R3, R4 // 7c841b78 - OR R3, R4, R5 // 7c851b78 - ORN R3, R4, R5 // 7c851b38 - ORCC R3, R4, R5 // 7c851b79 - XOR R3, R4 // 7c841a78 - XOR R3, R4, R5 // 7c851a78 - XORCC R3, R4, R5 // 7c851a79 - NAND R3, R4, R5 // 7c851bb8 - NANDCC R3, R4, R5 // 7c851bb9 - EQV R3, R4, R5 // 7c851a38 - EQVCC R3, R4, R5 // 7c851a39 - NOR R3, R4, R5 // 7c8518f8 - NORCC R3, R4, R5 // 7c8518f9 - - SUB R3, R4 // 7c832050 - SUB R3, R4, R5 // 7ca32050 - SUBC R3, R4 // 7c832010 - SUBC R3, R4, R5 // 7ca32010 - - MULLW R3, R4 // 7c8419d6 - MULLW R3, R4, R5 // 7ca419d6 - MULLWCC R3, R4, R5 // 7ca419d7 - MULHW R3, R4, R5 // 7ca41896 - - MULHWU R3, R4, R5 // 7ca41816 - MULLD R3, R4 // 7c8419d2 - MULLD R4, R4, R5 // 7ca421d2 - MULLDCC R3, R4, R5 // 7ca419d3 - MULHD R3, R4, R5 // 7ca41892 - MULHDCC R3, R4, R5 // 7ca41893 - - MULLWV R3, R4 // 7c841dd6 - MULLWV R3, R4, R5 // 7ca41dd6 - MULLWVCC R3, R4, R5 // 7ca41dd7 - MULHWUCC R3, R4, R5 // 7ca41817 - MULLDV R3, R4, R5 // 7ca41dd2 - MULLDVCC R3, R4, R5 // 7ca41dd3 - - DIVD R3,R4 // 7c841bd2 - DIVD R3, R4, R5 // 7ca41bd2 - DIVDCC R3,R4, R5 // 7ca41bd3 - DIVDU R3, R4, R5 // 7ca41b92 - DIVDV R3, R4, R5 // 7ca41fd2 - DIVDUCC R3, R4, R5 // 7ca41b93 - DIVDVCC R3, R4, R5 // 7ca41fd3 - DIVDUV R3, R4, R5 // 7ca41f92 - DIVDUVCC R3, R4, R5 // 7ca41f93 - DIVDE R3, R4, R5 // 7ca41b52 - DIVDECC R3, R4, R5 // 7ca41b53 - DIVDEU R3, R4, R5 // 7ca41b12 - DIVDEUCC R3, R4, R5 // 7ca41b13 - - REM R3, R4, R5 // 7fe41bd67fff19d67cbf2050 - REMU R3, R4, R5 // 7fe41b967fff19d67bff00287cbf2050 - REMD R3, R4, R5 // 7fe41bd27fff19d27cbf2050 - REMDU R3, R4, R5 // 7fe41b927fff19d27cbf2050 - - MODUD R3, R4, R5 // 7ca41a12 - MODUW R3, R4, R5 // 7ca41a16 - MODSD R3, R4, R5 // 7ca41e12 - MODSW R3, R4, R5 // 7ca41e16 - - SLW $8, R3, R4 // 5464402e - SLW R3, R4, R5 // 7c851830 - SLWCC R3, R4 // 7c841831 - SLD $16, R3, R4 // 786483e4 - SLD R3, R4, R5 // 7c851836 - SLDCC R3, R4 // 7c841837 - - SRW $8, R3, R4 // 5464c23e - SRW R3, R4, R5 // 7c851c30 - SRWCC R3, R4 // 7c841c31 - SRAW $8, R3, R4 // 7c644670 - SRAW R3, R4, R5 // 7c851e30 - SRAWCC R3, R4 // 7c841e31 - SRD $16, R3, R4 // 78648402 - SRD R3, R4, R5 // 7c851c36 - SRDCC R3, R4 // 7c841c37 - SRAD $16, R3, R4 // 7c648674 - SRAD R3, R4, R5 // 7c851e34 - SRDCC R3, R4 // 7c841c37 - ROTLW $16, R3, R4 // 5464803e - ROTLW R3, R4, R5 // 5c85183e - RLWMI $7, R3, $65535, R6 // 50663c3e - RLWMICC $7, R3, $65535, R6 // 50663c3f - RLWNM $3, R4, $7, R6 // 54861f7e - RLWNMCC $3, R4, $7, R6 // 54861f7f - RLDMI $0, R4, $7, R6 // 7886076c - RLDMICC $0, R4, $7, R6 // 7886076d - RLDIMI $0, R4, $7, R6 // 788601cc - RLDIMICC $0, R4, $7, R6 // 788601cd - RLDC $0, R4, $15, R6 // 78860728 - RLDCCC $0, R4, $15, R6 // 78860729 - RLDCL $0, R4, $7, R6 // 78860770 - RLDCLCC $0, R4, $15, R6 // 78860721 - RLDCR $0, R4, $-16, R6 // 788606f2 - RLDCRCC $0, R4, $-16, R6 // 788606f3 - RLDICL $0, R4, $15, R6 // 788603c0 - RLDICLCC $0, R4, $15, R6 // 788603c1 - RLDICR $0, R4, $15, R6 // 788603c4 - RLDICRCC $0, R4, $15, R6 // 788603c5 - - BEQ 0(PC) // 41820000 - BGE 0(PC) // 40800000 - BGT 4(PC) // 41810030 - BLE 0(PC) // 40810000 - BLT 0(PC) // 41800000 - BNE 0(PC) // 40820000 - JMP 8(PC) // 48000020 - - CRAND CR1, CR2, CR3 // 4c620a02 - CRANDN CR1, CR2, CR3 // 4c620902 - CREQV CR1, CR2, CR3 // 4c620a42 - CRNAND CR1, CR2, CR3 // 4c6209c2 - CRNOR CR1, CR2, CR3 // 4c620842 - CROR CR1, CR2, CR3 // 4c620b82 - CRORN CR1, CR2, CR3 // 4c620b42 - CRXOR CR1, CR2, CR3 // 4c620982 - - ISEL $1, R3, R4, R5 // 7ca3205e - ISEL $0, R3, R4, R5 // 7ca3201e - ISEL $2, R3, R4, R5 // 7ca3209e - ISEL $3, R3, R4, R5 // 7ca320de - ISEL $4, R3, R4, R5 // 7ca3211e - POPCNTB R3, R4 // 7c6400f4 - POPCNTW R3, R4 // 7c6402f4 - POPCNTD R3, R4 // 7c6403f4 - - PASTECC R3, R4 // 7c23270d - COPY R3, R4 // 7c23260c - - // load-and-reserve - LBAR (R4)(R3*1),$1,R5 // 7ca32069 - LBAR (R4),$0,R5 // 7ca02068 - LBAR (R3),R5 // 7ca01868 - LHAR (R4)(R3*1),$1,R5 // 7ca320e9 - LHAR (R4),$0,R5 // 7ca020e8 - LHAR (R3),R5 // 7ca018e8 - LWAR (R4)(R3*1),$1,R5 // 7ca32029 - LWAR (R4),$0,R5 // 7ca02028 - LWAR (R3),R5 // 7ca01828 - LDAR (R4)(R3*1),$1,R5 // 7ca320a9 - LDAR (R4),$0,R5 // 7ca020a8 - LDAR (R3),R5 // 7ca018a8 - - STBCCC R3, (R4)(R5) // 7c65256d - STWCCC R3, (R4)(R5) // 7c65212d - STDCCC R3, (R4)(R5) // 7c6521ad - STHCCC R3, (R4)(R5) - - SYNC // 7c0004ac - ISYNC // 4c00012c - LWSYNC // 7c2004ac - - DCBF (R3)(R4) // 7c0418ac - DCBI (R3)(R4) // 7c041bac - DCBST (R3)(R4) // 7c04186c - DCBZ (R3)(R4) // 7c041fec - DCBT (R3)(R4) // 7c041a2c - ICBI (R3)(R4) // 7c041fac - - // float constants - FMOVD $(0.0), F1 // f0210cd0 - FMOVD $(-0.0), F1 // f0210cd0fc200850 - - FMOVD 8(R3), F1 // c8230008 - FMOVD (R3)(R4), F1 // 7c241cae - FMOVDU 8(R3), F1 // cc230008 - FMOVDU (R3)(R4), F1 // 7c241cee - FMOVS 4(R3), F1 // c0230004 - FMOVS (R3)(R4), F1 // 7c241c2e - FMOVSU 4(R3), F1 // c4230004 - FMOVSU (R3)(R4), F1 // 7c241c6e - - FMOVD F1, 8(R3) // d8230008 - FMOVD F1, (R3)(R4) // 7c241dae - FMOVDU F1, 8(R3) // dc230008 - FMOVDU F1, (R3)(R4) // 7c241dee - FMOVS F1, 4(R3) // d0230004 - FMOVS F1, (R3)(R4) // 7c241d2e - FMOVSU F1, 4(R3) // d4230004 - FMOVSU F1, (R3)(R4) // 7c241d6e - FADD F1, F2 // fc42082a - FADD F1, F2, F3 // fc62082a - FADDCC F1, F2, F3 // fc62082b - FADDS F1, F2 // ec42082a - FADDS F1, F2, F3 // ec62082a - FADDSCC F1, F2, F3 // ec62082b - FSUB F1, F2 // fc420828 - FSUB F1, F2, F3 // fc620828 - FSUBCC F1, F2, F3 // fc620829 - FSUBS F1, F2 // ec420828 - FSUBS F1, F2, F3 // ec620828 - FSUBCC F1, F2, F3 // fc620829 - FMUL F1, F2 // fc420072 - FMUL F1, F2, F3 // fc620072 - FMULCC F1, F2, F3 // fc620073 - FMULS F1, F2 // ec420072 - FMULS F1, F2, F3 // ec620072 - FMULSCC F1, F2, F3 // ec620073 - FDIV F1, F2 // fc420824 - FDIV F1, F2, F3 // fc620824 - FDIVCC F1, F2, F3 // fc620825 - FDIVS F1, F2 // ec420824 - FDIVS F1, F2, F3 // ec620824 - FDIVSCC F1, F2, F3 // ec620825 - FMADD F1, F2, F3, F4 // fc8110fa - FMADDCC F1, F2, F3, F4 // fc8110fb - FMADDS F1, F2, F3, F4 // ec8110fa - FMADDSCC F1, F2, F3, F4 // ec8110fb - FMSUB F1, F2, F3, F4 // fc8110f8 - FMSUBCC F1, F2, F3, F4 // fc8110f9 - FMSUBS F1, F2, F3, F4 // ec8110f8 - FMSUBSCC F1, F2, F3, F4 // ec8110f9 - FNMADD F1, F2, F3, F4 // fc8110fe - FNMADDCC F1, F2, F3, F4 // fc8110ff - FNMADDS F1, F2, F3, F4 // ec8110fe - FNMADDSCC F1, F2, F3, F4 // ec8110ff - FNMSUB F1, F2, F3, F4 // fc8110fc - FNMSUBCC F1, F2, F3, F4 // fc8110fd - FNMSUBS F1, F2, F3, F4 // ec8110fc - FNMSUBSCC F1, F2, F3, F4 // ec8110fd - FSEL F1, F2, F3, F4 // fc8110ee - FSELCC F1, F2, F3, F4 // fc8110ef - FABS F1, F2 // fc400a10 - FABSCC F1, F2 // fc400a11 - FNEG F1, F2 // fc400850 - FABSCC F1, F2 // fc400a11 - FRSP F1, F2 // fc400818 - FRSPCC F1, F2 // fc400819 - FCTIW F1, F2 // fc40081c - FCTIWCC F1, F2 // fc40081d - FCTIWZ F1, F2 // fc40081e - FCTIWZCC F1, F2 // fc40081f - FCTID F1, F2 // fc400e5c - FCTIDCC F1, F2 // fc400e5d - FCTIDZ F1, F2 // fc400e5e - FCTIDZCC F1, F2 // fc400e5f - FCFID F1, F2 // fc400e9c - FCFIDCC F1, F2 // fc400e9d - FCFIDU F1, F2 // fc400f9c - FCFIDUCC F1, F2 // fc400f9d - FCFIDS F1, F2 // ec400e9c - FCFIDSCC F1, F2 // ec400e9d - FRES F1, F2 // ec400830 - FRESCC F1, F2 // ec400831 - FRIM F1, F2 // fc400bd0 - FRIMCC F1, F2 // fc400bd1 - FRIP F1, F2 // fc400b90 - FRIPCC F1, F2 // fc400b91 - FRIZ F1, F2 // fc400b50 - FRIZCC F1, F2 // fc400b51 - FRIN F1, F2 // fc400b10 - FRINCC F1, F2 // fc400b11 - FRSQRTE F1, F2 // fc400834 - FRSQRTECC F1, F2 // fc400835 - FSQRT F1, F2 // fc40082c - FSQRTCC F1, F2 // fc40082d - FSQRTS F1, F2 // ec40082c - FSQRTSCC F1, F2 // ec40082d - FCPSGN F1, F2 // fc420810 - FCPSGNCC F1, F2 // fc420811 - FCMPO F1, F2 // fc011040 - FCMPU F1, F2 // fc011000 - LVX (R3)(R4), V1 // 7c2418ce - LVXL (R3)(R4), V1 // 7c241ace - LVSL (R3)(R4), V1 // 7c24180c - LVSR (R3)(R4), V1 // 7c24184c - LVEBX (R3)(R4), V1 // 7c24180e - LVEHX (R3)(R4), V1 // 7c24184e - LVEWX (R3)(R4), V1 // 7c24188e - STVX V1, (R3)(R4) // 7c2419ce - STVXL V1, (R3)(R4) // 7c241bce - STVEBX V1, (R3)(R4) // 7c24190e - STVEHX V1, (R3)(R4) // 7c24194e - STVEWX V1, (R3)(R4) // 7c24198e - - VAND V1, V2, V3 // 10611404 - VANDC V1, V2, V3 // 10611444 - VNAND V1, V2, V3 // 10611584 - VOR V1, V2, V3 // 10611484 - VORC V1, V2, V3 // 10611544 - VXOR V1, V2, V3 // 106114c4 - VNOR V1, V2, V3 // 10611504 - VEQV V1, V2, V3 // 10611684 - VADDUBM V1, V2, V3 // 10611000 - VADDUHM V1, V2, V3 // 10611040 - VADDUWM V1, V2, V3 // 10611080 - VADDUDM V1, V2, V3 // 106110c0 - VADDUQM V1, V2, V3 // 10611100 - VADDCUQ V1, V2, V3 // 10611140 - VADDCUW V1, V2, V3 // 10611180 - VADDUBS V1, V2, V3 // 10611200 - VADDUHS V1, V2, V3 // 10611240 - VADDUWS V1, V2, V3 // 10611280 - VSUBUBM V1, V2, V3 // 10611400 - VSUBUHM V1, V2, V3 // 10611440 - VSUBUWM V1, V2, V3 // 10611480 - VSUBUDM V1, V2, V3 // 106114c0 - VSUBUQM V1, V2, V3 // 10611500 - VSUBCUQ V1, V2, V3 // 10611540 - VSUBCUW V1, V2, V3 // 10611580 - VSUBUBS V1, V2, V3 // 10611600 - VSUBUHS V1, V2, V3 // 10611640 - VSUBUWS V1, V2, V3 // 10611680 - VSUBSBS V1, V2, V3 // 10611700 - VSUBSHS V1, V2, V3 // 10611740 - VSUBSWS V1, V2, V3 // 10611780 - VSUBEUQM V1, V2, V3, V4 // 108110fe - VSUBECUQ V1, V2, V3, V4 // 108110ff - VMULESB V1, V2, V3 // 10611308 - VMULOSB V1, V2, V3 // 10611108 - VMULEUB V1, V2, V3 // 10611208 - VMULOUB V1, V2, V3 // 10611008 - VMULESH V1, V2, V3 // 10611348 - VMULOSH V1, V2, V3 // 10611148 - VMULEUH V1, V2, V3 // 10611248 - VMULOUH V1, V2, V3 // 10611048 - VMULESH V1, V2, V3 // 10611348 - VMULOSW V1, V2, V3 // 10611188 - VMULEUW V1, V2, V3 // 10611288 - VMULOUW V1, V2, V3 // 10611088 - VMULUWM V1, V2, V3 // 10611089 - VPMSUMB V1, V2, V3 // 10611408 - VPMSUMH V1, V2, V3 // 10611448 - VPMSUMW V1, V2, V3 // 10611488 - VPMSUMD V1, V2, V3 // 106114c8 - VMSUMUDM V1, V2, V3, V4 // 108110e3 - VRLB V1, V2, V3 // 10611004 - VRLH V1, V2, V3 // 10611044 - VRLW V1, V2, V3 // 10611084 - VRLD V1, V2, V3 // 106110c4 - VSLB V1, V2, V3 // 10611104 - VSLH V1, V2, V3 // 10611144 - VSLW V1, V2, V3 // 10611184 - VSL V1, V2, V3 // 106111c4 - VSLO V1, V2, V3 // 1061140c - VSRB V1, V2, V3 // 10611204 - VSRH V1, V2, V3 // 10611244 - VSRW V1, V2, V3 // 10611284 - VSR V1, V2, V3 // 106112c4 - VSRO V1, V2, V3 // 1061144c - VSLD V1, V2, V3 // 106115c4 - VSRAB V1, V2, V3 // 10611304 - VSRAH V1, V2, V3 // 10611344 - VSRAW V1, V2, V3 // 10611384 - VSRAD V1, V2, V3 // 106113c4 - VSLDOI $3, V1, V2, V3 // 106110ec - VCLZB V1, V2 // 10400f02 - VCLZH V1, V2 // 10400f42 - VCLZW V1, V2 // 10400f82 - VCLZD V1, V2 // 10400fc2 - VPOPCNTB V1, V2 // 10400f03 - VPOPCNTH V1, V2 // 10400f43 - VPOPCNTW V1, V2 // 10400f83 - VPOPCNTD V1, V2 // 10400fc3 - VCMPEQUB V1, V2, V3 // 10611006 - VCMPEQUBCC V1, V2, V3 // 10611406 - VCMPEQUH V1, V2, V3 // 10611046 - VCMPEQUHCC V1, V2, V3 // 10611446 - VCMPEQUW V1, V2, V3 // 10611086 - VCMPEQUWCC V1, V2, V3 // 10611486 - VCMPEQUD V1, V2, V3 // 106110c7 - VCMPEQUDCC V1, V2, V3 // 106114c7 - VCMPGTUB V1, V2, V3 // 10611206 - VCMPGTUBCC V1, V2, V3 // 10611606 - VCMPGTUH V1, V2, V3 // 10611246 - VCMPGTUHCC V1, V2, V3 // 10611646 - VCMPGTUW V1, V2, V3 // 10611286 - VCMPGTUWCC V1, V2, V3 // 10611686 - VCMPGTUD V1, V2, V3 // 106112c7 - VCMPGTUDCC V1, V2, V3 // 106116c7 - VCMPGTSB V1, V2, V3 // 10611306 - VCMPGTSBCC V1, V2, V3 // 10611706 - VCMPGTSH V1, V2, V3 // 10611346 - VCMPGTSHCC V1, V2, V3 // 10611746 - VCMPGTSW V1, V2, V3 // 10611386 - VCMPGTSWCC V1, V2, V3 // 10611786 - VCMPGTSD V1, V2, V3 // 106113c7 - VCMPGTSDCC V1, V2, V3 // 106117c7 - VCMPNEZB V1, V2, V3 // 10611107 - VCMPNEZBCC V1, V2, V3 // 10611507 - VCMPNEB V1, V2, V3 // 10611007 - VCMPNEBCC V1, V2, V3 // 10611407 - VCMPNEH V1, V2, V3 // 10611047 - VCMPNEHCC V1, V2, V3 // 10611447 - VCMPNEW V1, V2, V3 // 10611087 - VCMPNEWCC V1, V2, V3 // 10611487 - VPERM V1, V2, V3, V4 // 108110eb - VPERMR V1, V2, V3, V4 // 108110fb - VPERMXOR V1, V2, V3, V4 // 108110ed - VBPERMQ V1, V2, V3 // 1061154c - VBPERMD V1, V2, V3 // 106115cc - VSEL V1, V2, V3, V4 // 108110ea - VSPLTB $1, V1, V2 // 10410a0c - VSPLTH $1, V1, V2 // 10410a4c - VSPLTW $1, V1, V2 // 10410a8c - VSPLTISB $1, V1 // 1021030c - VSPLTISW $1, V1 // 1021038c - VSPLTISH $1, V1 // 1021034c - VCIPHER V1, V2, V3 // 10611508 - VCIPHERLAST V1, V2, V3 // 10611509 - VNCIPHER V1, V2, V3 // 10611548 - VNCIPHERLAST V1, V2, V3 // 10611549 - VSBOX V1, V2 // 104105c8 - VSHASIGMAW $1, V1, $15, V2 // 10418e82 - VSHASIGMAD $2, V1, $15, V2 // 104196c2 - - LXVD2X (R3)(R4), VS1 // 7c241e98 - LXV 16(R3), VS1 // f4230011 - LXVL R3, R4, VS1 // 7c23221a - LXVLL R3, R4, VS1 // 7c23225a - LXVX R3, R4, VS1 // 7c232218 - LXSDX (R3)(R4), VS1 // 7c241c98 - STXVD2X VS1, (R3)(R4) // 7c241f98 - STXV VS1,16(R3) // f4230015 - STXVL VS1, R3, R4 // 7c23231a - STXVLL VS1, R3, R4 // 7c23235a - STXVX VS1, R3, R4 // 7c232318 - STXSDX VS1, (R3)(R4) // 7c241d98 - LXSIWAX (R3)(R4), VS1 // 7c241898 - STXSIWX VS1, (R3)(R4) // 7c241918 - MFVSRD VS1, R3 // 7c230066 - MTVSRD R3, VS1 // 7c230166 - XXLAND VS1, VS2, VS3 // f0611410 - XXLOR VS1, VS2, VS3 // f0611490 - XXLORC VS1, VS2, VS3 // f0611550 - XXLXOR VS1, VS2, VS3 // f06114d0 - XXSEL VS1, VS2, VS3, VS4 // f08110f0 - XXMRGHW VS1, VS2, VS3 // f0611090 - XXSPLTW VS1, $1, VS2 // f0410a90 - XXPERM VS1, VS2, VS3 // f06110d0 - XXSLDWI VS1, VS2, $1, VS3 // f0611110 - XSCVDPSP VS1, VS2 // f0400c24 - XVCVDPSP VS1, VS2 // f0400e24 - XSCVSXDDP VS1, VS2 // f0400de0 - XVCVDPSXDS VS1, VS2 // f0400f60 - XVCVSXDDP VS1, VS2 // f0400fe0 - - MOVD R3, LR // 7c6803a6 - MOVD R3, CTR // 7c6903a6 - MOVD R3, XER // 7c6103a6 - MOVD LR, R3 // 7c6802a6 - MOVD CTR, R3 // 7c6902a6 - MOVD XER, R3 // 7c6102a6 - MOVFL CR3, CR1 // 4c8c0000 - - RET diff --git a/src/cmd/asm/internal/asm/testdata/riscvenc.s b/src/cmd/asm/internal/asm/testdata/riscvenc.s index 8d301f2dd5e..9a49d96ca0a 100644 --- a/src/cmd/asm/internal/asm/testdata/riscvenc.s +++ b/src/cmd/asm/internal/asm/testdata/riscvenc.s @@ -297,6 +297,13 @@ start: MOVW X5, (X6) // 23205300 MOVW X5, 4(X6) // 23225300 + MOVB X5, X6 // 1393820313538343 + MOVH X5, X6 // 1393020313530343 + MOVW X5, X6 // 1b830200 + MOVBU X5, X6 // 13f3f20f + MOVHU X5, X6 // 1393020313530303 + MOVWU X5, X6 // 1393020213530302 + MOVF 4(X5), F0 // 07a04200 MOVF F0, 4(X5) // 27a20200 MOVF F0, F1 // d3000020 @@ -318,7 +325,7 @@ start: // These jumps can get printed as jumps to 2 because they go to the // second instruction in the function (the first instruction is an // invisible stack pointer adjustment). - JMP start // JMP 2 // 6ff01fc5 + JMP start // JMP 2 // 6ff09fc2 JMP (X5) // 67800200 JMP 4(X5) // 67804200 @@ -331,16 +338,16 @@ start: JMP asmtest(SB) // 970f0000 // Branch pseudo-instructions - BEQZ X5, start // BEQZ X5, 2 // e38a02c2 - BGEZ X5, start // BGEZ X5, 2 // e3d802c2 - BGT X5, X6, start // BGT X5, X6, 2 // e3c662c2 - BGTU X5, X6, start // BGTU X5, X6, 2 // e3e462c2 - BGTZ X5, start // BGTZ X5, 2 // e34250c2 - BLE X5, X6, start // BLE X5, X6, 2 // e3d062c2 - BLEU X5, X6, start // BLEU X5, X6, 2 // e3fe62c0 - BLEZ X5, start // BLEZ X5, 2 // e35c50c0 - BLTZ X5, start // BLTZ X5, 2 // e3ca02c0 - BNEZ X5, start // BNEZ X5, 2 // e39802c0 + BEQZ X5, start // BEQZ X5, 2 // e38602c0 + BGEZ X5, start // BGEZ X5, 2 // e3d402c0 + BGT X5, X6, start // BGT X5, X6, 2 // e34253c0 + BGTU X5, X6, start // BGTU X5, X6, 2 // e36053c0 + BGTZ X5, start // BGTZ X5, 2 // e34e50be + BLE X5, X6, start // BLE X5, X6, 2 // e35c53be + BLEU X5, X6, start // BLEU X5, X6, 2 // e37a53be + BLEZ X5, start // BLEZ X5, 2 // e35850be + BLTZ X5, start // BLTZ X5, 2 // e3c602be + BNEZ X5, start // BNEZ X5, 2 // e39402be // Set pseudo-instructions SEQZ X15, X15 // 93b71700 diff --git a/src/cmd/asm/internal/asm/testdata/s390x.s b/src/cmd/asm/internal/asm/testdata/s390x.s index 03b84cfa62d..7c5d26be33f 100644 --- a/src/cmd/asm/internal/asm/testdata/s390x.s +++ b/src/cmd/asm/internal/asm/testdata/s390x.s @@ -412,6 +412,8 @@ TEXT main·foo(SB),DUPOK|NOSPLIT,$16-0 // TEXT main.foo(SB), DUPOK|NOSPLIT, $16- UNDEF // 00000000 NOPH // 0700 + SYNC // 07e0 + // vector add and sub instructions VAB V3, V4, V4 // e743400000f3 VAH V3, V4, V4 // e743400010f3 diff --git a/src/cmd/asm/internal/flags/flags.go b/src/cmd/asm/internal/flags/flags.go index 1df9df9563e..1335860315a 100644 --- a/src/cmd/asm/internal/flags/flags.go +++ b/src/cmd/asm/internal/flags/flags.go @@ -15,27 +15,30 @@ import ( ) var ( - Debug = flag.Bool("debug", false, "dump instructions as they are parsed") - OutputFile = flag.String("o", "", "output file; default foo.o for /a/b/c/foo.s as first argument") - PrintOut = flag.Bool("S", false, "print assembly and machine code") - TrimPath = flag.String("trimpath", "", "remove prefix from recorded source file paths") - Shared = flag.Bool("shared", false, "generate code that can be linked into a shared library") - Dynlink = flag.Bool("dynlink", false, "support references to Go symbols defined in other shared libraries") - AllErrors = flag.Bool("e", false, "no limit on number of errors reported") - SymABIs = flag.Bool("gensymabis", false, "write symbol ABI information to output file, don't assemble") - Importpath = flag.String("p", "", "set expected package import to path") - Spectre = flag.String("spectre", "", "enable spectre mitigations in `list` (all, ret)") + Debug = flag.Bool("debug", false, "dump instructions as they are parsed") + OutputFile = flag.String("o", "", "output file; default foo.o for /a/b/c/foo.s as first argument") + TrimPath = flag.String("trimpath", "", "remove prefix from recorded source file paths") + Shared = flag.Bool("shared", false, "generate code that can be linked into a shared library") + Dynlink = flag.Bool("dynlink", false, "support references to Go symbols defined in other shared libraries") + Linkshared = flag.Bool("linkshared", false, "generate code that will be linked against Go shared libraries") + AllErrors = flag.Bool("e", false, "no limit on number of errors reported") + SymABIs = flag.Bool("gensymabis", false, "write symbol ABI information to output file, don't assemble") + Importpath = flag.String("p", "", "set expected package import to path") + Spectre = flag.String("spectre", "", "enable spectre mitigations in `list` (all, ret)") + CompilingRuntime = flag.Bool("compiling-runtime", false, "source to be compiled is part of the Go runtime") ) var ( - D MultiFlag - I MultiFlag + D MultiFlag + I MultiFlag + PrintOut int ) func init() { flag.Var(&D, "D", "predefined symbol with optional simple value -D=identifier=value; can be set multiple times") flag.Var(&I, "I", "include directory; can be set multiple times") objabi.AddVersionFlag() // -V + objabi.Flagcount("S", "print assembly and machine code", &PrintOut) } // MultiFlag allows setting a value multiple times to collect a list, as in -I=dir1 -I=dir2. diff --git a/src/cmd/asm/internal/lex/input.go b/src/cmd/asm/internal/lex/input.go index a43953b5154..da4ebe6d6e3 100644 --- a/src/cmd/asm/internal/lex/input.go +++ b/src/cmd/asm/internal/lex/input.go @@ -109,6 +109,9 @@ func (in *Input) Next() ScanToken { in.Error("'#' must be first item on line") } in.beginningOfLine = in.hash() + in.text = "#" + return '#' + case scanner.Ident: // Is it a macro name? name := in.Stack.Text() diff --git a/src/cmd/asm/internal/lex/lex.go b/src/cmd/asm/internal/lex/lex.go index f1f7da79112..7cd41a55a9b 100644 --- a/src/cmd/asm/internal/lex/lex.go +++ b/src/cmd/asm/internal/lex/lex.go @@ -22,11 +22,13 @@ type ScanToken rune const ( // Asm defines some two-character lexemes. We make up // a rune/ScanToken value for them - ugly but simple. - LSH ScanToken = -1000 - iota // << Left shift. - RSH // >> Logical right shift. - ARR // -> Used on ARM for shift type 3, arithmetic right shift. - ROT // @> Used on ARM for shift type 4, rotate right. - macroName // name of macro that should not be expanded + LSH ScanToken = -1000 - iota // << Left shift. + RSH // >> Logical right shift. + ARR // -> Used on ARM for shift type 3, arithmetic right shift. + ROT // @> Used on ARM for shift type 4, rotate right. + Include // included file started here + BuildComment // //go:build or +build comment + macroName // name of macro that should not be expanded ) // IsRegisterShift reports whether the token is one of the ARM register shift operators. diff --git a/src/cmd/asm/internal/lex/lex_test.go b/src/cmd/asm/internal/lex/lex_test.go index f606ffe07b0..51679d2fbc7 100644 --- a/src/cmd/asm/internal/lex/lex_test.go +++ b/src/cmd/asm/internal/lex/lex_test.go @@ -281,6 +281,9 @@ func drain(input *Input) string { if tok == scanner.EOF { return buf.String() } + if tok == '#' { + continue + } if buf.Len() > 0 { buf.WriteByte('.') } diff --git a/src/cmd/asm/internal/lex/tokenizer.go b/src/cmd/asm/internal/lex/tokenizer.go index aef9ea86368..861a2d421d5 100644 --- a/src/cmd/asm/internal/lex/tokenizer.go +++ b/src/cmd/asm/internal/lex/tokenizer.go @@ -107,10 +107,13 @@ func (t *Tokenizer) Next() ScanToken { if t.tok != scanner.Comment { break } - length := strings.Count(s.TokenText(), "\n") - t.line += length - // TODO: If we ever have //go: comments in assembly, will need to keep them here. - // For now, just discard all comments. + text := s.TokenText() + t.line += strings.Count(text, "\n") + // TODO: Use constraint.IsGoBuild once it exists. + if strings.HasPrefix(text, "//go:build") { + t.tok = BuildComment + break + } } switch t.tok { case '\n': diff --git a/src/cmd/asm/main.go b/src/cmd/asm/main.go index a6eb44de734..31636e30458 100644 --- a/src/cmd/asm/main.go +++ b/src/cmd/asm/main.go @@ -35,10 +35,9 @@ func main() { flags.Parse() ctxt := obj.Linknew(architecture.LinkArch) - if *flags.PrintOut { - ctxt.Debugasm = 1 - } + ctxt.Debugasm = flags.PrintOut ctxt.Flag_dynlink = *flags.Dynlink + ctxt.Flag_linkshared = *flags.Linkshared ctxt.Flag_shared = *flags.Shared || *flags.Dynlink ctxt.IsAsm = true ctxt.Pkgpath = *flags.Importpath @@ -76,7 +75,8 @@ func main() { var failedFile string for _, f := range flag.Args() { lexer := lex.NewLexer(f) - parser := asm.NewParser(ctxt, architecture, lexer) + parser := asm.NewParser(ctxt, architecture, lexer, + *flags.CompilingRuntime) ctxt.DiagFunc = func(format string, args ...interface{}) { diag = true log.Printf(format, args...) diff --git a/src/cmd/buildid/buildid.go b/src/cmd/buildid/buildid.go index 1c7b228c987..8e02a7ae109 100644 --- a/src/cmd/buildid/buildid.go +++ b/src/cmd/buildid/buildid.go @@ -22,21 +22,6 @@ func usage() { var wflag = flag.Bool("w", false, "write build ID") -// taken from cmd/go/internal/work/buildid.go -func hashToString(h [32]byte) string { - const b64 = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-_" - const chunks = 5 - var dst [chunks * 4]byte - for i := 0; i < chunks; i++ { - v := uint32(h[3*i])<<16 | uint32(h[3*i+1])<<8 | uint32(h[3*i+2]) - dst[4*i+0] = b64[(v>>18)&0x3F] - dst[4*i+1] = b64[(v>>12)&0x3F] - dst[4*i+2] = b64[(v>>6)&0x3F] - dst[4*i+3] = b64[v&0x3F] - } - return string(dst[:]) -} - func main() { log.SetPrefix("buildid: ") log.SetFlags(0) @@ -63,12 +48,12 @@ func main() { log.Fatal(err) } matches, hash, err := buildid.FindAndHash(f, id, 0) + f.Close() if err != nil { log.Fatal(err) } - f.Close() - newID := id[:strings.LastIndex(id, "/")] + "/" + hashToString(hash) + newID := id[:strings.LastIndex(id, "/")] + "/" + buildid.HashToString(hash) if len(newID) != len(id) { log.Fatalf("%s: build ID length mismatch %q vs %q", file, id, newID) } @@ -77,7 +62,7 @@ func main() { return } - f, err = os.OpenFile(file, os.O_WRONLY, 0) + f, err = os.OpenFile(file, os.O_RDWR, 0) if err != nil { log.Fatal(err) } diff --git a/src/cmd/cgo/ast.go b/src/cmd/cgo/ast.go index 54d6bc2559d..a073407a961 100644 --- a/src/cmd/cgo/ast.go +++ b/src/cmd/cgo/ast.go @@ -13,7 +13,6 @@ import ( "go/scanner" "go/token" "os" - "path/filepath" "strings" ) @@ -44,14 +43,7 @@ func sourceLine(n ast.Node) int { // attached to the import "C" comment, a list of references to C.xxx, // a list of exported functions, and the actual AST, to be rewritten and // printed. -func (f *File) ParseGo(name string, src []byte) { - // Create absolute path for file, so that it will be used in error - // messages and recorded in debug line number information. - // This matches the rest of the toolchain. See golang.org/issue/5122. - if aname, err := filepath.Abs(name); err == nil { - name = aname - } - +func (f *File) ParseGo(abspath string, src []byte) { // Two different parses: once with comments, once without. // The printer is not good enough at printing comments in the // right place when we start editing the AST behind its back, @@ -60,8 +52,8 @@ func (f *File) ParseGo(name string, src []byte) { // and reprinting. // In cgo mode, we ignore ast2 and just apply edits directly // the text behind ast1. In godefs mode we modify and print ast2. - ast1 := parse(name, src, parser.ParseComments) - ast2 := parse(name, src, 0) + ast1 := parse(abspath, src, parser.ParseComments) + ast2 := parse(abspath, src, 0) f.Package = ast1.Name.Name f.Name = make(map[string]*Name) @@ -88,7 +80,7 @@ func (f *File) ParseGo(name string, src []byte) { cg = d.Doc } if cg != nil { - f.Preamble += fmt.Sprintf("#line %d %q\n", sourceLine(cg), name) + f.Preamble += fmt.Sprintf("#line %d %q\n", sourceLine(cg), abspath) f.Preamble += commentText(cg) + "\n" f.Preamble += "#line 1 \"cgo-generated-wrapper\"\n" } diff --git a/src/cmd/cgo/doc.go b/src/cmd/cgo/doc.go index b3f371b08c4..e782c866ac7 100644 --- a/src/cmd/cgo/doc.go +++ b/src/cmd/cgo/doc.go @@ -721,7 +721,7 @@ linkage to the desired libraries. The main function is provided by _cgo_main.c: int main() { return 0; } - void crosscall2(void(*fn)(void*, int, uintptr_t), void *a, int c, uintptr_t ctxt) { } + void crosscall2(void(*fn)(void*), void *a, int c, uintptr_t ctxt) { } uintptr_t _cgo_wait_runtime_init_done(void) { return 0; } void _cgo_release_context(uintptr_t ctxt) { } char* _cgo_topofstack(void) { return (char*)0; } diff --git a/src/cmd/cgo/gcc.go b/src/cmd/cgo/gcc.go index 9179b5490e2..111a309eb5c 100644 --- a/src/cmd/cgo/gcc.go +++ b/src/cmd/cgo/gcc.go @@ -298,7 +298,7 @@ func (p *Package) guessKinds(f *File) []*Name { continue } - if goos == "darwin" && strings.HasSuffix(n.C, "Ref") { + if (goos == "darwin" || goos == "ios") && strings.HasSuffix(n.C, "Ref") { // For FooRef, find out if FooGetTypeID exists. s := n.C[:len(n.C)-3] + "GetTypeID" n := &Name{Go: s, C: s} @@ -2448,6 +2448,18 @@ func (c *typeConv) loadType(dtype dwarf.Type, pos token.Pos, parent string) *Typ tt := *t tt.C = &TypeRepr{"%s %s", []interface{}{dt.Kind, tag}} tt.Go = c.Ident("struct{}") + if dt.Kind == "struct" { + // We don't know what the representation of this struct is, so don't let + // anyone allocate one on the Go side. As a side effect of this annotation, + // pointers to this type will not be considered pointers in Go. They won't + // get writebarrier-ed or adjusted during a stack copy. This should handle + // all the cases badPointerTypedef used to handle, but hopefully will + // continue to work going forward without any more need for cgo changes. + tt.NotInHeap = true + // TODO: we should probably do the same for unions. Unions can't live + // on the Go heap, right? It currently doesn't work for unions because + // they are defined as a type alias for struct{}, not a defined type. + } typedef[name.Name] = &tt break } @@ -2518,6 +2530,7 @@ func (c *typeConv) loadType(dtype dwarf.Type, pos token.Pos, parent string) *Typ } t.Go = name t.BadPointer = sub.BadPointer + t.NotInHeap = sub.NotInHeap if unionWithPointer[sub.Go] { unionWithPointer[t.Go] = true } @@ -2528,6 +2541,7 @@ func (c *typeConv) loadType(dtype dwarf.Type, pos token.Pos, parent string) *Typ tt := *t tt.Go = sub.Go tt.BadPointer = sub.BadPointer + tt.NotInHeap = sub.NotInHeap typedef[name.Name] = &tt } @@ -2831,21 +2845,11 @@ func (c *typeConv) Struct(dt *dwarf.StructType, pos token.Pos) (expr *ast.Struct tgo := t.Go size := t.Size talign := t.Align - if f.BitSize > 0 { - switch f.BitSize { - case 8, 16, 32, 64: - default: - continue - } - size = f.BitSize / 8 - name := tgo.(*ast.Ident).String() - if strings.HasPrefix(name, "int") { - name = "int" - } else { - name = "uint" - } - tgo = ast.NewIdent(name + fmt.Sprint(f.BitSize)) - talign = size + if f.BitOffset > 0 || f.BitSize > 0 { + // The layout of bitfields is implementation defined, + // so we don't know how they correspond to Go fields + // even if they are aligned at byte boundaries. + continue } if talign > 0 && f.ByteOffset%talign != 0 { @@ -3036,6 +3040,7 @@ func (c *typeConv) anonymousStructTypedef(dt *dwarf.TypedefType) bool { // non-pointers in this type. // TODO: Currently our best solution is to find these manually and list them as // they come up. A better solution is desired. +// Note: DEPRECATED. There is now a better solution. Search for NotInHeap in this file. func (c *typeConv) badPointerTypedef(dt *dwarf.TypedefType) bool { if c.badCFType(dt) { return true @@ -3070,7 +3075,7 @@ func (c *typeConv) badCFType(dt *dwarf.TypedefType) bool { // We identify the correct set of types as those ending in Ref and for which // there exists a corresponding GetTypeID function. // See comment below for details about the bad pointers. - if goos != "darwin" { + if goos != "darwin" && goos != "ios" { return false } s := dt.Name diff --git a/src/cmd/cgo/godefs.go b/src/cmd/cgo/godefs.go index b4fd9c5a6e3..c0d59aee01d 100644 --- a/src/cmd/cgo/godefs.go +++ b/src/cmd/cgo/godefs.go @@ -16,7 +16,7 @@ import ( ) // godefs returns the output for -godefs mode. -func (p *Package) godefs(f *File, srcfile string) string { +func (p *Package) godefs(f *File) string { var buf bytes.Buffer fmt.Fprintf(&buf, "// Code generated by cmd/cgo -godefs; DO NOT EDIT.\n") diff --git a/src/cmd/cgo/main.go b/src/cmd/cgo/main.go index 5a7bb3f87bc..c1116e28ecd 100644 --- a/src/cmd/cgo/main.go +++ b/src/cmd/cgo/main.go @@ -151,7 +151,8 @@ type Type struct { Go ast.Expr EnumValues map[string]int64 Typedef string - BadPointer bool + BadPointer bool // this pointer type should be represented as a uintptr (deprecated) + NotInHeap bool // this type should have a go:notinheap annotation } // A FuncType collects information about a function type in both the C and Go worlds. @@ -169,35 +170,51 @@ func usage() { var ptrSizeMap = map[string]int64{ "386": 4, + "alpha": 8, "amd64": 8, "arm": 4, "arm64": 8, + "m68k": 4, "mips": 4, "mipsle": 4, "mips64": 8, "mips64le": 8, + "nios2": 4, + "ppc": 4, "ppc64": 8, "ppc64le": 8, + "riscv": 4, "riscv64": 8, "s390": 4, "s390x": 8, + "sh": 4, + "shbe": 4, + "sparc": 4, "sparc64": 8, } var intSizeMap = map[string]int64{ "386": 4, + "alpha": 8, "amd64": 8, "arm": 4, "arm64": 8, + "m68k": 4, "mips": 4, "mipsle": 4, "mips64": 8, "mips64le": 8, + "nios2": 4, + "ppc": 4, "ppc64": 8, "ppc64le": 8, + "riscv": 4, "riscv64": 8, "s390": 4, "s390x": 8, + "sh": 4, + "shbe": 4, + "sparc": 4, "sparc64": 8, } @@ -223,10 +240,11 @@ var exportHeader = flag.String("exportheader", "", "where to write export header var gccgo = flag.Bool("gccgo", false, "generate files for use with gccgo") var gccgoprefix = flag.String("gccgoprefix", "", "-fgo-prefix option used with gccgo") var gccgopkgpath = flag.String("gccgopkgpath", "", "-fgo-pkgpath option used with gccgo") -var gccgoMangleCheckDone bool -var gccgoNewmanglingInEffect bool +var gccgoMangler func(string) string var importRuntimeCgo = flag.Bool("import_runtime_cgo", true, "import runtime/cgo in generated code") var importSyscall = flag.Bool("import_syscall", true, "import syscall in generated code") +var trimpath = flag.String("trimpath", "", "applies supplied rewrites or trims prefixes to recorded source file paths") + var goarch, goos string func main() { @@ -306,6 +324,13 @@ func main() { input = filepath.Join(*srcDir, input) } + // Create absolute path for file, so that it will be used in error + // messages and recorded in debug line number information. + // This matches the rest of the toolchain. See golang.org/issue/5122. + if aname, err := filepath.Abs(input); err == nil { + input = aname + } + b, err := ioutil.ReadFile(input) if err != nil { fatalf("%s", err) @@ -314,6 +339,10 @@ func main() { fatalf("%s", err) } + // Apply trimpath to the file path. The path won't be read from after this point. + input, _ = objabi.ApplyRewrites(input, *trimpath) + goFiles[i] = input + f := new(File) f.Edit = edit.NewBuffer(b) f.ParseGo(input, b) @@ -351,7 +380,7 @@ func main() { p.PackagePath = f.Package p.Record(f) if *godefs { - os.Stdout.WriteString(p.godefs(f, input)) + os.Stdout.WriteString(p.godefs(f)) } else { p.writeOutput(f, input) } diff --git a/src/cmd/cgo/out.go b/src/cmd/cgo/out.go index 50d2811f1b9..11c53facf8f 100644 --- a/src/cmd/cgo/out.go +++ b/src/cmd/cgo/out.go @@ -6,6 +6,7 @@ package main import ( "bytes" + "cmd/internal/pkgpath" "debug/elf" "debug/macho" "debug/pe" @@ -15,7 +16,6 @@ import ( "go/token" "internal/xcoff" "io" - "io/ioutil" "os" "os/exec" "path/filepath" @@ -59,14 +59,14 @@ func (p *Package) writeDefs() { // Write C main file for using gcc to resolve imports. fmt.Fprintf(fm, "int main() { return 0; }\n") if *importRuntimeCgo { - fmt.Fprintf(fm, "void crosscall2(void(*fn)(void*, int, __SIZE_TYPE__), void *a, int c, __SIZE_TYPE__ ctxt) { }\n") + fmt.Fprintf(fm, "void crosscall2(void(*fn)(void*), void *a, int c, __SIZE_TYPE__ ctxt) { }\n") fmt.Fprintf(fm, "__SIZE_TYPE__ _cgo_wait_runtime_init_done(void) { return 0; }\n") fmt.Fprintf(fm, "void _cgo_release_context(__SIZE_TYPE__ ctxt) { }\n") fmt.Fprintf(fm, "char* _cgo_topofstack(void) { return (char*)0; }\n") } else { // If we're not importing runtime/cgo, we *are* runtime/cgo, // which provides these functions. We just need a prototype. - fmt.Fprintf(fm, "void crosscall2(void(*fn)(void*, int, __SIZE_TYPE__), void *a, int c, __SIZE_TYPE__ ctxt);\n") + fmt.Fprintf(fm, "void crosscall2(void(*fn)(void*), void *a, int c, __SIZE_TYPE__ ctxt);\n") fmt.Fprintf(fm, "__SIZE_TYPE__ _cgo_wait_runtime_init_done(void);\n") fmt.Fprintf(fm, "void _cgo_release_context(__SIZE_TYPE__);\n") } @@ -108,6 +108,9 @@ func (p *Package) writeDefs() { sort.Strings(typedefNames) for _, name := range typedefNames { def := typedef[name] + if def.NotInHeap { + fmt.Fprintf(fgo2, "//go:notinheap\n") + } fmt.Fprintf(fgo2, "type %s ", name) // We don't have source info for these types, so write them out without source info. // Otherwise types would look like: @@ -183,7 +186,7 @@ func (p *Package) writeDefs() { panic(fmt.Errorf("invalid var kind %q", n.Kind)) } if *gccgo { - fmt.Fprintf(fc, `extern void *%s __asm__("%s.%s");`, n.Mangle, gccgoSymbolPrefix, n.Mangle) + fmt.Fprintf(fc, `extern void *%s __asm__("%s.%s");`, n.Mangle, gccgoSymbolPrefix, gccgoToSymbol(n.Mangle)) fmt.Fprintf(&gccgoInit, "\t%s = &%s;\n", n.Mangle, n.C) fmt.Fprintf(fc, "\n") } @@ -334,6 +337,8 @@ func dynimport(obj string) { if s.Version != "" { targ += "#" + s.Version } + checkImportSymName(s.Name) + checkImportSymName(targ) fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", s.Name, targ, s.Library) } lib, _ := f.ImportedLibraries() @@ -349,6 +354,7 @@ func dynimport(obj string) { if len(s) > 0 && s[0] == '_' { s = s[1:] } + checkImportSymName(s) fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", s, s, "") } lib, _ := f.ImportedLibraries() @@ -363,6 +369,8 @@ func dynimport(obj string) { for _, s := range sym { ss := strings.Split(s, ":") name := strings.Split(ss[0], "@")[0] + checkImportSymName(name) + checkImportSymName(ss[0]) fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", name, ss[0], strings.ToLower(ss[1])) } return @@ -380,6 +388,7 @@ func dynimport(obj string) { // Go symbols. continue } + checkImportSymName(s.Name) fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", s.Name, s.Name, s.Library) } lib, err := f.ImportedLibraries() @@ -395,6 +404,23 @@ func dynimport(obj string) { fatalf("cannot parse %s as ELF, Mach-O, PE or XCOFF", obj) } +// checkImportSymName checks a symbol name we are going to emit as part +// of a //go:cgo_import_dynamic pragma. These names come from object +// files, so they may be corrupt. We are going to emit them unquoted, +// so while they don't need to be valid symbol names (and in some cases, +// involving symbol versions, they won't be) they must contain only +// graphic characters and must not contain Go comments. +func checkImportSymName(s string) { + for _, c := range s { + if !unicode.IsGraphic(c) || unicode.IsSpace(c) { + fatalf("dynamic symbol %q contains unsupported character", s) + } + } + if strings.Index(s, "//") >= 0 || strings.Index(s, "/*") >= 0 { + fatalf("dynamic symbol %q contains Go comment") + } +} + // Construct a gcc struct matching the gc argument frame. // Assumes that in gcc, char is 1 byte, short 2 bytes, int 4 bytes, long long 8 bytes. // These assumptions are checked by the gccProlog. @@ -849,7 +875,7 @@ func (p *Package) writeExports(fgo2, fm, fgcc, fgcch io.Writer) { fmt.Fprintf(fgcc, "#pragma GCC diagnostic ignored \"-Wpragmas\"\n") fmt.Fprintf(fgcc, "#pragma GCC diagnostic ignored \"-Waddress-of-packed-member\"\n") - fmt.Fprintf(fgcc, "extern void crosscall2(void (*fn)(void *, int, __SIZE_TYPE__), void *, int, __SIZE_TYPE__);\n") + fmt.Fprintf(fgcc, "extern void crosscall2(void (*fn)(void *), void *, int, __SIZE_TYPE__);\n") fmt.Fprintf(fgcc, "extern __SIZE_TYPE__ _cgo_wait_runtime_init_done(void);\n") fmt.Fprintf(fgcc, "extern void _cgo_release_context(__SIZE_TYPE__);\n\n") fmt.Fprintf(fgcc, "extern char* _cgo_topofstack(void);") @@ -859,59 +885,48 @@ func (p *Package) writeExports(fgo2, fm, fgcc, fgcch io.Writer) { for _, exp := range p.ExpFunc { fn := exp.Func - // Construct a gcc struct matching the gc argument and - // result frame. The gcc struct will be compiled with - // __attribute__((packed)) so all padding must be accounted - // for explicitly. + // Construct a struct that will be used to communicate + // arguments from C to Go. The C and Go definitions + // just have to agree. The gcc struct will be compiled + // with __attribute__((packed)) so all padding must be + // accounted for explicitly. ctype := "struct {\n" + gotype := new(bytes.Buffer) + fmt.Fprintf(gotype, "struct {\n") off := int64(0) npad := 0 - if fn.Recv != nil { - t := p.cgoType(fn.Recv.List[0].Type) - ctype += fmt.Sprintf("\t\t%s recv;\n", t.C) + argField := func(typ ast.Expr, namePat string, args ...interface{}) { + name := fmt.Sprintf(namePat, args...) + t := p.cgoType(typ) + if off%t.Align != 0 { + pad := t.Align - off%t.Align + ctype += fmt.Sprintf("\t\tchar __pad%d[%d];\n", npad, pad) + off += pad + npad++ + } + ctype += fmt.Sprintf("\t\t%s %s;\n", t.C, name) + fmt.Fprintf(gotype, "\t\t%s ", name) + noSourceConf.Fprint(gotype, fset, typ) + fmt.Fprintf(gotype, "\n") off += t.Size } + if fn.Recv != nil { + argField(fn.Recv.List[0].Type, "recv") + } fntype := fn.Type forFieldList(fntype.Params, func(i int, aname string, atype ast.Expr) { - t := p.cgoType(atype) - if off%t.Align != 0 { - pad := t.Align - off%t.Align - ctype += fmt.Sprintf("\t\tchar __pad%d[%d];\n", npad, pad) - off += pad - npad++ - } - ctype += fmt.Sprintf("\t\t%s p%d;\n", t.C, i) - off += t.Size + argField(atype, "p%d", i) }) - if off%p.PtrSize != 0 { - pad := p.PtrSize - off%p.PtrSize - ctype += fmt.Sprintf("\t\tchar __pad%d[%d];\n", npad, pad) - off += pad - npad++ - } forFieldList(fntype.Results, func(i int, aname string, atype ast.Expr) { - t := p.cgoType(atype) - if off%t.Align != 0 { - pad := t.Align - off%t.Align - ctype += fmt.Sprintf("\t\tchar __pad%d[%d];\n", npad, pad) - off += pad - npad++ - } - ctype += fmt.Sprintf("\t\t%s r%d;\n", t.C, i) - off += t.Size + argField(atype, "r%d", i) }) - if off%p.PtrSize != 0 { - pad := p.PtrSize - off%p.PtrSize - ctype += fmt.Sprintf("\t\tchar __pad%d[%d];\n", npad, pad) - off += pad - npad++ - } if ctype == "struct {\n" { ctype += "\t\tchar unused;\n" // avoid empty struct } ctype += "\t}" + fmt.Fprintf(gotype, "\t}") // Get the return type of the wrapper function // compiled by gcc. @@ -936,7 +951,11 @@ func (p *Package) writeExports(fgo2, fm, fgcc, fgcch io.Writer) { } // Build the wrapper function compiled by gcc. - s := fmt.Sprintf("%s %s(", gccResult, exp.ExpName) + gccExport := "" + if goos == "windows" { + gccExport = "__declspec(dllexport)" + } + s := fmt.Sprintf("%s %s %s(", gccExport, gccResult, exp.ExpName) if fn.Recv != nil { s += p.cgoType(fn.Recv.List[0].Type).C.String() s += " recv" @@ -958,12 +977,23 @@ func (p *Package) writeExports(fgo2, fm, fgcc, fgcch io.Writer) { } fmt.Fprintf(fgcch, "extern %s;\n", s) - fmt.Fprintf(fgcc, "extern void _cgoexp%s_%s(void *, int, __SIZE_TYPE__);\n", cPrefix, exp.ExpName) + fmt.Fprintf(fgcc, "extern void _cgoexp%s_%s(void *);\n", cPrefix, exp.ExpName) fmt.Fprintf(fgcc, "\nCGO_NO_SANITIZE_THREAD") fmt.Fprintf(fgcc, "\n%s\n", s) fmt.Fprintf(fgcc, "{\n") fmt.Fprintf(fgcc, "\t__SIZE_TYPE__ _cgo_ctxt = _cgo_wait_runtime_init_done();\n") - fmt.Fprintf(fgcc, "\t%s %v _cgo_a;\n", ctype, p.packedAttribute()) + // The results part of the argument structure must be + // initialized to 0 so the write barriers generated by + // the assignments to these fields in Go are safe. + // + // We use a local static variable to get the zeroed + // value of the argument type. This avoids including + // string.h for memset, and is also robust to C++ + // types with constructors. Both GCC and LLVM optimize + // this into just zeroing _cgo_a. + fmt.Fprintf(fgcc, "\ttypedef %s %v _cgo_argtype;\n", ctype, p.packedAttribute()) + fmt.Fprintf(fgcc, "\tstatic _cgo_argtype _cgo_zero;\n") + fmt.Fprintf(fgcc, "\t_cgo_argtype _cgo_a = _cgo_zero;\n") if gccResult != "void" && (len(fntype.Results.List) > 1 || len(fntype.Results.List[0].Names) > 1) { fmt.Fprintf(fgcc, "\t%s r;\n", gccResult) } @@ -992,82 +1022,28 @@ func (p *Package) writeExports(fgo2, fm, fgcc, fgcch io.Writer) { fmt.Fprintf(fgcc, "}\n") // Build the wrapper function compiled by cmd/compile. - goname := "_cgoexpwrap" + cPrefix + "_" - if fn.Recv != nil { - goname += fn.Recv.List[0].Names[0].Name + "_" - } - goname += exp.Func.Name.Name + // This unpacks the argument struct above and calls the Go function. fmt.Fprintf(fgo2, "//go:cgo_export_dynamic %s\n", exp.ExpName) fmt.Fprintf(fgo2, "//go:linkname _cgoexp%s_%s _cgoexp%s_%s\n", cPrefix, exp.ExpName, cPrefix, exp.ExpName) fmt.Fprintf(fgo2, "//go:cgo_export_static _cgoexp%s_%s\n", cPrefix, exp.ExpName) - fmt.Fprintf(fgo2, "//go:nosplit\n") // no split stack, so no use of m or g - fmt.Fprintf(fgo2, "//go:norace\n") // must not have race detector calls inserted - fmt.Fprintf(fgo2, "func _cgoexp%s_%s(a unsafe.Pointer, n int32, ctxt uintptr) {\n", cPrefix, exp.ExpName) - fmt.Fprintf(fgo2, "\tfn := %s\n", goname) - // The indirect here is converting from a Go function pointer to a C function pointer. - fmt.Fprintf(fgo2, "\t_cgo_runtime_cgocallback(**(**unsafe.Pointer)(unsafe.Pointer(&fn)), a, uintptr(n), ctxt);\n") - fmt.Fprintf(fgo2, "}\n") + fmt.Fprintf(fgo2, "func _cgoexp%s_%s(a *%s) {\n", cPrefix, exp.ExpName, gotype) fmt.Fprintf(fm, "int _cgoexp%s_%s;\n", cPrefix, exp.ExpName) - // This code uses printer.Fprint, not conf.Fprint, - // because we don't want //line comments in the middle - // of the function types. - fmt.Fprintf(fgo2, "\n") - fmt.Fprintf(fgo2, "func %s(", goname) - comma := false - if fn.Recv != nil { - fmt.Fprintf(fgo2, "recv ") - printer.Fprint(fgo2, fset, fn.Recv.List[0].Type) - comma = true - } - forFieldList(fntype.Params, - func(i int, aname string, atype ast.Expr) { - if comma { - fmt.Fprintf(fgo2, ", ") - } - fmt.Fprintf(fgo2, "p%d ", i) - printer.Fprint(fgo2, fset, atype) - comma = true - }) - fmt.Fprintf(fgo2, ")") if gccResult != "void" { - fmt.Fprint(fgo2, " (") + // Write results back to frame. + fmt.Fprintf(fgo2, "\t") forFieldList(fntype.Results, func(i int, aname string, atype ast.Expr) { if i > 0 { - fmt.Fprint(fgo2, ", ") + fmt.Fprintf(fgo2, ", ") } - fmt.Fprintf(fgo2, "r%d ", i) - printer.Fprint(fgo2, fset, atype) + fmt.Fprintf(fgo2, "a.r%d", i) }) - fmt.Fprint(fgo2, ")") - } - fmt.Fprint(fgo2, " {\n") - if gccResult == "void" { - fmt.Fprint(fgo2, "\t") - } else { - // Verify that any results don't contain any - // Go pointers. - addedDefer := false - forFieldList(fntype.Results, - func(i int, aname string, atype ast.Expr) { - if !p.hasPointer(nil, atype, false) { - return - } - if !addedDefer { - fmt.Fprint(fgo2, "\tdefer func() {\n") - addedDefer = true - } - fmt.Fprintf(fgo2, "\t\t_cgoCheckResult(r%d)\n", i) - }) - if addedDefer { - fmt.Fprint(fgo2, "\t}()\n") - } - fmt.Fprint(fgo2, "\treturn ") + fmt.Fprintf(fgo2, " = ") } if fn.Recv != nil { - fmt.Fprintf(fgo2, "recv.") + fmt.Fprintf(fgo2, "a.recv.") } fmt.Fprintf(fgo2, "%s(", exp.Func.Name) forFieldList(fntype.Params, @@ -1075,9 +1051,20 @@ func (p *Package) writeExports(fgo2, fm, fgcc, fgcch io.Writer) { if i > 0 { fmt.Fprint(fgo2, ", ") } - fmt.Fprintf(fgo2, "p%d", i) + fmt.Fprintf(fgo2, "a.p%d", i) }) fmt.Fprint(fgo2, ")\n") + if gccResult != "void" { + // Verify that any results don't contain any + // Go pointers. + forFieldList(fntype.Results, + func(i int, aname string, atype ast.Expr) { + if !p.hasPointer(nil, atype, false) { + return + } + fmt.Fprintf(fgo2, "\t_cgoCheckResult(a.r%d)\n", i) + }) + } fmt.Fprint(fgo2, "}\n") } @@ -1161,7 +1148,7 @@ func (p *Package) writeGccgoExports(fgo2, fm, fgcc, fgcch io.Writer) { // will not be able to link against it from the C // code. goName := "Cgoexp_" + exp.ExpName - fmt.Fprintf(fgcc, `extern %s %s %s __asm__("%s.%s");`, cRet, goName, cParams, gccgoSymbolPrefix, goName) + fmt.Fprintf(fgcc, `extern %s %s %s __asm__("%s.%s");`, cRet, goName, cParams, gccgoSymbolPrefix, gccgoToSymbol(goName)) fmt.Fprint(fgcc, "\n") fmt.Fprint(fgcc, "\nCGO_NO_SANITIZE_THREAD\n") @@ -1195,7 +1182,7 @@ func (p *Package) writeGccgoExports(fgo2, fm, fgcc, fgcch io.Writer) { fmt.Fprint(fgcc, "}\n") // Dummy declaration for _cgo_main.c - fmt.Fprintf(fm, `char %s[1] __asm__("%s.%s");`, goName, gccgoSymbolPrefix, goName) + fmt.Fprintf(fm, `char %s[1] __asm__("%s.%s");`, goName, gccgoSymbolPrefix, gccgoToSymbol(goName)) fmt.Fprint(fm, "\n") // For gccgo we use a wrapper function in Go, in order @@ -1279,112 +1266,23 @@ func (p *Package) writeExportHeader(fgcch io.Writer) { fmt.Fprintf(fgcch, "%s\n", p.gccExportHeaderProlog()) } -// gccgoUsesNewMangling reports whether gccgo uses the new collision-free -// packagepath mangling scheme (see determineGccgoManglingScheme for more -// info). -func gccgoUsesNewMangling() bool { - if !gccgoMangleCheckDone { - gccgoNewmanglingInEffect = determineGccgoManglingScheme() - gccgoMangleCheckDone = true - } - return gccgoNewmanglingInEffect -} - -const mangleCheckCode = ` -package läufer -func Run(x int) int { - return 1 -} -` - -// determineGccgoManglingScheme performs a runtime test to see which -// flavor of packagepath mangling gccgo is using. Older versions of -// gccgo use a simple mangling scheme where there can be collisions -// between packages whose paths are different but mangle to the same -// string. More recent versions of gccgo use a new mangler that avoids -// these collisions. Return value is whether gccgo uses the new mangling. -func determineGccgoManglingScheme() bool { - - // Emit a small Go file for gccgo to compile. - filepat := "*_gccgo_manglecheck.go" - var f *os.File - var err error - if f, err = ioutil.TempFile(*objDir, filepat); err != nil { - fatalf("%v", err) - } - gofilename := f.Name() - defer os.Remove(gofilename) - - if err = ioutil.WriteFile(gofilename, []byte(mangleCheckCode), 0666); err != nil { - fatalf("%v", err) - } - - // Compile with gccgo, capturing generated assembly. - gccgocmd := os.Getenv("GCCGO") - if gccgocmd == "" { - gpath, gerr := exec.LookPath("gccgo") - if gerr != nil { - fatalf("unable to locate gccgo: %v", gerr) +// gccgoToSymbol converts a name to a mangled symbol for gccgo. +func gccgoToSymbol(ppath string) string { + if gccgoMangler == nil { + var err error + cmd := os.Getenv("GCCGO") + if cmd == "" { + cmd, err = exec.LookPath("gccgo") + if err != nil { + fatalf("unable to locate gccgo: %v", err) + } } - gccgocmd = gpath - } - cmd := exec.Command(gccgocmd, "-S", "-o", "-", gofilename) - buf, cerr := cmd.CombinedOutput() - if cerr != nil { - fatalf("%s", cerr) - } - - // New mangling: expect go.l..u00e4ufer.Run - // Old mangling: expect go.l__ufer.Run - return regexp.MustCompile(`go\.l\.\.u00e4ufer\.Run`).Match(buf) -} - -// gccgoPkgpathToSymbolNew converts a package path to a gccgo-style -// package symbol. -func gccgoPkgpathToSymbolNew(ppath string) string { - bsl := []byte{} - changed := false - for _, c := range []byte(ppath) { - switch { - case 'A' <= c && c <= 'Z', 'a' <= c && c <= 'z', - '0' <= c && c <= '9', c == '_': - bsl = append(bsl, c) - case c == '.': - bsl = append(bsl, ".x2e"...) - default: - changed = true - encbytes := []byte(fmt.Sprintf("..z%02x", c)) - bsl = append(bsl, encbytes...) + gccgoMangler, err = pkgpath.ToSymbolFunc(cmd, *objDir) + if err != nil { + fatalf("%v", err) } } - if !changed { - return ppath - } - return string(bsl) -} - -// gccgoPkgpathToSymbolOld converts a package path to a gccgo-style -// package symbol using the older mangling scheme. -func gccgoPkgpathToSymbolOld(ppath string) string { - clean := func(r rune) rune { - switch { - case 'A' <= r && r <= 'Z', 'a' <= r && r <= 'z', - '0' <= r && r <= '9': - return r - } - return '_' - } - return strings.Map(clean, ppath) -} - -// gccgoPkgpathToSymbol converts a package path to a mangled packagepath -// symbol. -func gccgoPkgpathToSymbol(ppath string) string { - if gccgoUsesNewMangling() { - return gccgoPkgpathToSymbolNew(ppath) - } else { - return gccgoPkgpathToSymbolOld(ppath) - } + return gccgoMangler(ppath) } // Return the package prefix when using gccgo. @@ -1394,12 +1292,12 @@ func (p *Package) gccgoSymbolPrefix() string { } if *gccgopkgpath != "" { - return gccgoPkgpathToSymbol(*gccgopkgpath) + return gccgoToSymbol(*gccgopkgpath) } if *gccgoprefix == "" && p.PackageName == "main" { return "main" } - prefix := gccgoPkgpathToSymbol(*gccgoprefix) + prefix := gccgoToSymbol(*gccgoprefix) if prefix == "" { prefix = "go" } @@ -1663,9 +1561,6 @@ const goProlog = ` //go:linkname _cgo_runtime_cgocall runtime.cgocall func _cgo_runtime_cgocall(unsafe.Pointer, uintptr) int32 -//go:linkname _cgo_runtime_cgocallback runtime.cgocallback -func _cgo_runtime_cgocallback(unsafe.Pointer, unsafe.Pointer, uintptr, uintptr) - //go:linkname _cgoCheckPointer runtime.cgoCheckPointer func _cgoCheckPointer(interface{}, interface{}) @@ -1791,8 +1686,12 @@ void _cgoPREFIX_Cfunc__Cmalloc(void *v) { ` func (p *Package) cPrologGccgo() string { - return strings.Replace(strings.Replace(cPrologGccgo, "PREFIX", cPrefix, -1), - "GCCGOSYMBOLPREF", p.gccgoSymbolPrefix(), -1) + r := strings.NewReplacer( + "PREFIX", cPrefix, + "GCCGOSYMBOLPREF", p.gccgoSymbolPrefix(), + "_cgoCheckPointer", gccgoToSymbol("_cgoCheckPointer"), + "_cgoCheckResult", gccgoToSymbol("_cgoCheckResult")) + return r.Replace(cPrologGccgo) } const cPrologGccgo = ` diff --git a/src/cmd/compile/fmt_test.go b/src/cmd/compile/fmt_test.go index e372259c788..6625ccf5e24 100644 --- a/src/cmd/compile/fmt_test.go +++ b/src/cmd/compile/fmt_test.go @@ -52,6 +52,7 @@ import ( "go/types" "internal/testenv" "io" + "io/fs" "io/ioutil" "log" "os" @@ -89,7 +90,7 @@ func TestFormats(t *testing.T) { testenv.MustHaveGoBuild(t) // more restrictive than necessary, but that's ok // process all directories - filepath.Walk(".", func(path string, info os.FileInfo, err error) error { + filepath.WalkDir(".", func(path string, info fs.DirEntry, err error) error { if info.IsDir() { if info.Name() == "testdata" { return filepath.SkipDir diff --git a/src/cmd/compile/fmtmap_test.go b/src/cmd/compile/fmtmap_test.go index 179c60187fe..0811df7f7b8 100644 --- a/src/cmd/compile/fmtmap_test.go +++ b/src/cmd/compile/fmtmap_test.go @@ -105,10 +105,8 @@ var knownFormats = map[string]string{ "cmd/compile/internal/ssa.GCNode %v": "", "cmd/compile/internal/ssa.ID %d": "", "cmd/compile/internal/ssa.ID %v": "", - "cmd/compile/internal/ssa.LocPair %s": "", "cmd/compile/internal/ssa.LocalSlot %s": "", "cmd/compile/internal/ssa.LocalSlot %v": "", - "cmd/compile/internal/ssa.Location %T": "", "cmd/compile/internal/ssa.Location %s": "", "cmd/compile/internal/ssa.Op %s": "", "cmd/compile/internal/ssa.Op %v": "", diff --git a/src/cmd/compile/internal/amd64/ssa.go b/src/cmd/compile/internal/amd64/ssa.go index 4ac877986cd..5ff05a0edd1 100644 --- a/src/cmd/compile/internal/amd64/ssa.go +++ b/src/cmd/compile/internal/amd64/ssa.go @@ -42,10 +42,11 @@ func ssaMarkMoves(s *gc.SSAGenState, b *ssa.Block) { // loadByType returns the load instruction of the given type. func loadByType(t *types.Type) obj.As { // Avoid partial register write - if !t.IsFloat() && t.Size() <= 2 { - if t.Size() == 1 { + if !t.IsFloat() { + switch t.Size() { + case 1: return x86.AMOVBLZX - } else { + case 2: return x86.AMOVWLZX } } @@ -75,7 +76,7 @@ func storeByType(t *types.Type) obj.As { return x86.AMOVQ } } - panic("bad store type") + panic(fmt.Sprintf("bad store type %v", t)) } // moveByType returns the reg->reg move instruction of the given type. @@ -100,7 +101,7 @@ func moveByType(t *types.Type) obj.As { case 16: return x86.AMOVUPS // int128s are in SSE registers default: - panic(fmt.Sprintf("bad int register width %d:%s", t.Size(), t)) + panic(fmt.Sprintf("bad int register width %d:%v", t.Size(), t)) } } } @@ -1070,7 +1071,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p := s.Prog(v.Op.Asm()) val := v.AuxInt // 0 means math.RoundToEven, 1 Floor, 2 Ceil, 3 Trunc - if val != 0 && val != 1 && val != 2 && val != 3 { + if val < 0 || val > 3 { v.Fatalf("Invalid rounding mode") } p.From.Offset = val @@ -1210,7 +1211,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p = s.Prog(x86.ASETEQ) p.To.Type = obj.TYPE_REG p.To.Reg = v.Reg0() - case ssa.OpAMD64ANDBlock, ssa.OpAMD64ORBlock: + case ssa.OpAMD64ANDBlock, ssa.OpAMD64ANDLlock, ssa.OpAMD64ORBlock, ssa.OpAMD64ORLlock: s.Prog(x86.ALOCK) p := s.Prog(v.Op.Asm()) p.From.Type = obj.TYPE_REG diff --git a/src/cmd/compile/internal/arm64/ggen.go b/src/cmd/compile/internal/arm64/ggen.go index f698919e9b2..f3fec03854f 100644 --- a/src/cmd/compile/internal/arm64/ggen.go +++ b/src/cmd/compile/internal/arm64/ggen.go @@ -11,7 +11,7 @@ import ( "cmd/internal/objabi" ) -var darwin = objabi.GOOS == "darwin" +var darwin = objabi.GOOS == "darwin" || objabi.GOOS == "ios" func padframe(frame int64) int64 { // arm64 requires that the frame size (not counting saved FP&LR) diff --git a/src/cmd/compile/internal/arm64/ssa.go b/src/cmd/compile/internal/arm64/ssa.go index 1d6ea6b9d8e..22b28a9308b 100644 --- a/src/cmd/compile/internal/arm64/ssa.go +++ b/src/cmd/compile/internal/arm64/ssa.go @@ -581,6 +581,24 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p2.From.Reg = arm64.REGTMP p2.To.Type = obj.TYPE_BRANCH gc.Patch(p2, p) + case ssa.OpARM64LoweredAtomicExchange64Variant, + ssa.OpARM64LoweredAtomicExchange32Variant: + swap := arm64.ASWPALD + if v.Op == ssa.OpARM64LoweredAtomicExchange32Variant { + swap = arm64.ASWPALW + } + r0 := v.Args[0].Reg() + r1 := v.Args[1].Reg() + out := v.Reg0() + + // SWPALD Rarg1, (Rarg0), Rout + p := s.Prog(swap) + p.From.Type = obj.TYPE_REG + p.From.Reg = r1 + p.To.Type = obj.TYPE_MEM + p.To.Reg = r0 + p.RegTo2 = out + case ssa.OpARM64LoweredAtomicAdd64, ssa.OpARM64LoweredAtomicAdd32: // LDAXR (Rarg0), Rout @@ -687,16 +705,74 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p5.To.Type = obj.TYPE_REG p5.To.Reg = out gc.Patch(p2, p5) + case ssa.OpARM64LoweredAtomicCas64Variant, + ssa.OpARM64LoweredAtomicCas32Variant: + // Rarg0: ptr + // Rarg1: old + // Rarg2: new + // MOV Rarg1, Rtmp + // CASAL Rtmp, (Rarg0), Rarg2 + // CMP Rarg1, Rtmp + // CSET EQ, Rout + cas := arm64.ACASALD + cmp := arm64.ACMP + mov := arm64.AMOVD + if v.Op == ssa.OpARM64LoweredAtomicCas32Variant { + cas = arm64.ACASALW + cmp = arm64.ACMPW + mov = arm64.AMOVW + } + r0 := v.Args[0].Reg() + r1 := v.Args[1].Reg() + r2 := v.Args[2].Reg() + out := v.Reg0() + + // MOV Rarg1, Rtmp + p := s.Prog(mov) + p.From.Type = obj.TYPE_REG + p.From.Reg = r1 + p.To.Type = obj.TYPE_REG + p.To.Reg = arm64.REGTMP + + // CASAL Rtmp, (Rarg0), Rarg2 + p1 := s.Prog(cas) + p1.From.Type = obj.TYPE_REG + p1.From.Reg = arm64.REGTMP + p1.To.Type = obj.TYPE_MEM + p1.To.Reg = r0 + p1.RegTo2 = r2 + + // CMP Rarg1, Rtmp + p2 := s.Prog(cmp) + p2.From.Type = obj.TYPE_REG + p2.From.Reg = r1 + p2.Reg = arm64.REGTMP + + // CSET EQ, Rout + p3 := s.Prog(arm64.ACSET) + p3.From.Type = obj.TYPE_REG + p3.From.Reg = arm64.COND_EQ + p3.To.Type = obj.TYPE_REG + p3.To.Reg = out + case ssa.OpARM64LoweredAtomicAnd8, - ssa.OpARM64LoweredAtomicOr8: - // LDAXRB (Rarg0), Rout + ssa.OpARM64LoweredAtomicAnd32, + ssa.OpARM64LoweredAtomicOr8, + ssa.OpARM64LoweredAtomicOr32: + // LDAXRB/LDAXRW (Rarg0), Rout // AND/OR Rarg1, Rout - // STLXRB Rout, (Rarg0), Rtmp + // STLXRB/STLXRB Rout, (Rarg0), Rtmp // CBNZ Rtmp, -3(PC) + ld := arm64.ALDAXRB + st := arm64.ASTLXRB + if v.Op == ssa.OpARM64LoweredAtomicAnd32 || v.Op == ssa.OpARM64LoweredAtomicOr32 { + ld = arm64.ALDAXRW + st = arm64.ASTLXRW + } r0 := v.Args[0].Reg() r1 := v.Args[1].Reg() out := v.Reg0() - p := s.Prog(arm64.ALDAXRB) + p := s.Prog(ld) p.From.Type = obj.TYPE_MEM p.From.Reg = r0 p.To.Type = obj.TYPE_REG @@ -706,7 +782,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p1.From.Reg = r1 p1.To.Type = obj.TYPE_REG p1.To.Reg = out - p2 := s.Prog(arm64.ASTLXRB) + p2 := s.Prog(st) p2.From.Type = obj.TYPE_REG p2.From.Reg = out p2.To.Type = obj.TYPE_MEM @@ -717,6 +793,63 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p3.From.Reg = arm64.REGTMP p3.To.Type = obj.TYPE_BRANCH gc.Patch(p3, p) + case ssa.OpARM64LoweredAtomicAnd8Variant, + ssa.OpARM64LoweredAtomicAnd32Variant: + atomic_clear := arm64.ALDCLRALW + if v.Op == ssa.OpARM64LoweredAtomicAnd8Variant { + atomic_clear = arm64.ALDCLRALB + } + r0 := v.Args[0].Reg() + r1 := v.Args[1].Reg() + out := v.Reg0() + + // MNV Rarg1 Rtemp + p := s.Prog(arm64.AMVN) + p.From.Type = obj.TYPE_REG + p.From.Reg = r1 + p.To.Type = obj.TYPE_REG + p.To.Reg = arm64.REGTMP + + // LDCLRALW Rtemp, (Rarg0), Rout + p1 := s.Prog(atomic_clear) + p1.From.Type = obj.TYPE_REG + p1.From.Reg = arm64.REGTMP + p1.To.Type = obj.TYPE_MEM + p1.To.Reg = r0 + p1.RegTo2 = out + + // AND Rarg1, Rout + p2 := s.Prog(arm64.AAND) + p2.From.Type = obj.TYPE_REG + p2.From.Reg = r1 + p2.To.Type = obj.TYPE_REG + p2.To.Reg = out + + case ssa.OpARM64LoweredAtomicOr8Variant, + ssa.OpARM64LoweredAtomicOr32Variant: + atomic_or := arm64.ALDORALW + if v.Op == ssa.OpARM64LoweredAtomicOr8Variant { + atomic_or = arm64.ALDORALB + } + r0 := v.Args[0].Reg() + r1 := v.Args[1].Reg() + out := v.Reg0() + + // LDORALW Rarg1, (Rarg0), Rout + p := s.Prog(atomic_or) + p.From.Type = obj.TYPE_REG + p.From.Reg = r1 + p.To.Type = obj.TYPE_MEM + p.To.Reg = r0 + p.RegTo2 = out + + // ORR Rarg1, Rout + p2 := s.Prog(arm64.AORR) + p2.From.Type = obj.TYPE_REG + p2.From.Reg = r1 + p2.To.Type = obj.TYPE_REG + p2.To.Reg = out + case ssa.OpARM64MOVBreg, ssa.OpARM64MOVBUreg, ssa.OpARM64MOVHreg, diff --git a/src/cmd/compile/internal/gc/alg.go b/src/cmd/compile/internal/gc/alg.go index c9d71ea00b4..2f7fa27bb91 100644 --- a/src/cmd/compile/internal/gc/alg.go +++ b/src/cmd/compile/internal/gc/alg.go @@ -282,7 +282,7 @@ func genhash(t *types.Type) *obj.LSym { } sym := typesymprefix(".hash", t) - if Debug['r'] != 0 { + if Debug.r != 0 { fmt.Printf("genhash %v %v %v\n", closure, sym, t) } @@ -374,7 +374,7 @@ func genhash(t *types.Type) *obj.LSym { r.List.Append(nh) fn.Nbody.Append(r) - if Debug['r'] != 0 { + if Debug.r != 0 { dumplist("genhash body", fn.Nbody) } @@ -392,7 +392,7 @@ func genhash(t *types.Type) *obj.LSym { } fn.Func.SetNilCheckDisabled(true) - funccompile(fn) + xtop = append(xtop, fn) // Build closure. It doesn't close over any variables, so // it contains just the function pointer. @@ -509,7 +509,7 @@ func geneq(t *types.Type) *obj.LSym { return closure } sym := typesymprefix(".eq", t) - if Debug['r'] != 0 { + if Debug.r != 0 { fmt.Printf("geneq %v\n", t) } @@ -529,6 +529,10 @@ func geneq(t *types.Type) *obj.LSym { fn := dclfunc(sym, tfn) np := asNode(tfn.Type.Params().Field(0).Nname) nq := asNode(tfn.Type.Params().Field(1).Nname) + nr := asNode(tfn.Type.Results().Field(0).Nname) + + // Label to jump to if an equality test fails. + neq := autolabel(".neq") // We reach here only for types that have equality but // cannot be handled by the standard algorithms, @@ -555,13 +559,13 @@ func geneq(t *types.Type) *obj.LSym { // for i := 0; i < nelem; i++ { // if eq(p[i], q[i]) { // } else { - // return + // goto neq // } // } // // TODO(josharian): consider doing some loop unrolling // for larger nelem as well, processing a few elements at a time in a loop. - checkAll := func(unroll int64, eq func(pi, qi *Node) *Node) { + checkAll := func(unroll int64, last bool, eq func(pi, qi *Node) *Node) { // checkIdx generates a node to check for equality at index i. checkIdx := func(i *Node) *Node { // pi := p[i] @@ -576,37 +580,38 @@ func geneq(t *types.Type) *obj.LSym { } if nelem <= unroll { - // Generate a series of checks. - var cond *Node - for i := int64(0); i < nelem; i++ { - c := nodintconst(i) - check := checkIdx(c) - if cond == nil { - cond = check - continue - } - cond = nod(OANDAND, cond, check) + if last { + // Do last comparison in a different manner. + nelem-- + } + // Generate a series of checks. + for i := int64(0); i < nelem; i++ { + // if check {} else { goto neq } + nif := nod(OIF, checkIdx(nodintconst(i)), nil) + nif.Rlist.Append(nodSym(OGOTO, nil, neq)) + fn.Nbody.Append(nif) + } + if last { + fn.Nbody.Append(nod(OAS, nr, checkIdx(nodintconst(nelem)))) + } + } else { + // Generate a for loop. + // for i := 0; i < nelem; i++ + i := temp(types.Types[TINT]) + init := nod(OAS, i, nodintconst(0)) + cond := nod(OLT, i, nodintconst(nelem)) + post := nod(OAS, i, nod(OADD, i, nodintconst(1))) + loop := nod(OFOR, cond, post) + loop.Ninit.Append(init) + // if eq(pi, qi) {} else { goto neq } + nif := nod(OIF, checkIdx(i), nil) + nif.Rlist.Append(nodSym(OGOTO, nil, neq)) + loop.Nbody.Append(nif) + fn.Nbody.Append(loop) + if last { + fn.Nbody.Append(nod(OAS, nr, nodbool(true))) } - nif := nod(OIF, cond, nil) - nif.Rlist.Append(nod(ORETURN, nil, nil)) - fn.Nbody.Append(nif) - return } - - // Generate a for loop. - // for i := 0; i < nelem; i++ - i := temp(types.Types[TINT]) - init := nod(OAS, i, nodintconst(0)) - cond := nod(OLT, i, nodintconst(nelem)) - post := nod(OAS, i, nod(OADD, i, nodintconst(1))) - loop := nod(OFOR, cond, post) - loop.Ninit.Append(init) - // if eq(pi, qi) {} else { return } - check := checkIdx(i) - nif := nod(OIF, check, nil) - nif.Rlist.Append(nod(ORETURN, nil, nil)) - loop.Nbody.Append(nif) - fn.Nbody.Append(loop) } switch t.Elem().Etype { @@ -614,32 +619,28 @@ func geneq(t *types.Type) *obj.LSym { // Do two loops. First, check that all the lengths match (cheap). // Second, check that all the contents match (expensive). // TODO: when the array size is small, unroll the length match checks. - checkAll(3, func(pi, qi *Node) *Node { + checkAll(3, false, func(pi, qi *Node) *Node { // Compare lengths. eqlen, _ := eqstring(pi, qi) return eqlen }) - checkAll(1, func(pi, qi *Node) *Node { + checkAll(1, true, func(pi, qi *Node) *Node { // Compare contents. _, eqmem := eqstring(pi, qi) return eqmem }) case TFLOAT32, TFLOAT64: - checkAll(2, func(pi, qi *Node) *Node { + checkAll(2, true, func(pi, qi *Node) *Node { // p[i] == q[i] return nod(OEQ, pi, qi) }) // TODO: pick apart structs, do them piecemeal too default: - checkAll(1, func(pi, qi *Node) *Node { + checkAll(1, true, func(pi, qi *Node) *Node { // p[i] == q[i] return nod(OEQ, pi, qi) }) } - // return true - ret := nod(ORETURN, nil, nil) - ret.List.Append(nodbool(true)) - fn.Nbody.Append(ret) case TSTRUCT: // Build a list of conditions to satisfy. @@ -717,22 +718,42 @@ func geneq(t *types.Type) *obj.LSym { flatConds = append(flatConds, c...) } - var cond *Node if len(flatConds) == 0 { - cond = nodbool(true) + fn.Nbody.Append(nod(OAS, nr, nodbool(true))) } else { - cond = flatConds[0] - for _, c := range flatConds[1:] { - cond = nod(OANDAND, cond, c) + for _, c := range flatConds[:len(flatConds)-1] { + // if cond {} else { goto neq } + n := nod(OIF, c, nil) + n.Rlist.Append(nodSym(OGOTO, nil, neq)) + fn.Nbody.Append(n) } + fn.Nbody.Append(nod(OAS, nr, flatConds[len(flatConds)-1])) } - - ret := nod(ORETURN, nil, nil) - ret.List.Append(cond) - fn.Nbody.Append(ret) } - if Debug['r'] != 0 { + // ret: + // return + ret := autolabel(".ret") + fn.Nbody.Append(nodSym(OLABEL, nil, ret)) + fn.Nbody.Append(nod(ORETURN, nil, nil)) + + // neq: + // r = false + // return (or goto ret) + fn.Nbody.Append(nodSym(OLABEL, nil, neq)) + fn.Nbody.Append(nod(OAS, nr, nodbool(false))) + if EqCanPanic(t) || hasCall(fn) { + // Epilogue is large, so share it with the equal case. + fn.Nbody.Append(nodSym(OGOTO, nil, ret)) + } else { + // Epilogue is small, so don't bother sharing. + fn.Nbody.Append(nod(ORETURN, nil, nil)) + } + // TODO(khr): the epilogue size detection condition above isn't perfect. + // We should really do a generic CL that shares epilogues across + // the board. See #24936. + + if Debug.r != 0 { dumplist("geneq body", fn.Nbody) } @@ -754,7 +775,7 @@ func geneq(t *types.Type) *obj.LSym { // neither of which can be nil, and our comparisons // are shallow. fn.Func.SetNilCheckDisabled(true) - funccompile(fn) + xtop = append(xtop, fn) // Generate a closure which points at the function we just generated. dsymptr(closure, 0, sym.Linksym(), 0) @@ -762,6 +783,39 @@ func geneq(t *types.Type) *obj.LSym { return closure } +func hasCall(n *Node) bool { + if n.Op == OCALL || n.Op == OCALLFUNC { + return true + } + if n.Left != nil && hasCall(n.Left) { + return true + } + if n.Right != nil && hasCall(n.Right) { + return true + } + for _, x := range n.Ninit.Slice() { + if hasCall(x) { + return true + } + } + for _, x := range n.Nbody.Slice() { + if hasCall(x) { + return true + } + } + for _, x := range n.List.Slice() { + if hasCall(x) { + return true + } + } + for _, x := range n.Rlist.Slice() { + if hasCall(x) { + return true + } + } + return false +} + // eqfield returns the node // p.field == q.field func eqfield(p *Node, q *Node, field *types.Sym) *Node { diff --git a/src/cmd/compile/internal/gc/align.go b/src/cmd/compile/internal/gc/align.go index ab578ee8c7e..a3a0c8fce82 100644 --- a/src/cmd/compile/internal/gc/align.go +++ b/src/cmd/compile/internal/gc/align.go @@ -5,7 +5,9 @@ package gc import ( + "bytes" "cmd/compile/internal/types" + "fmt" "sort" ) @@ -84,7 +86,7 @@ func expandiface(t *types.Type) { sort.Sort(methcmp(methods)) if int64(len(methods)) >= thearch.MAXWIDTH/int64(Widthptr) { - yyerror("interface too large") + yyerrorl(typePos(t), "interface too large") } for i, m := range methods { m.Offset = int64(i) * int64(Widthptr) @@ -148,7 +150,7 @@ func widstruct(errtype *types.Type, t *types.Type, o int64, flag int) int64 { maxwidth = 1<<31 - 1 } if o >= maxwidth { - yyerror("type %L too large", errtype) + yyerrorl(typePos(errtype), "type %L too large", errtype) o = 8 // small but nonzero } } @@ -173,6 +175,91 @@ func widstruct(errtype *types.Type, t *types.Type, o int64, flag int) int64 { return o } +// findTypeLoop searches for an invalid type declaration loop involving +// type t and reports whether one is found. If so, path contains the +// loop. +// +// path points to a slice used for tracking the sequence of types +// visited. Using a pointer to a slice allows the slice capacity to +// grow and limit reallocations. +func findTypeLoop(t *types.Type, path *[]*types.Type) bool { + // We implement a simple DFS loop-finding algorithm. This + // could be faster, but type cycles are rare. + + if t.Sym != nil { + // Declared type. Check for loops and otherwise + // recurse on the type expression used in the type + // declaration. + + for i, x := range *path { + if x == t { + *path = (*path)[i:] + return true + } + } + + *path = append(*path, t) + if p := asNode(t.Nod).Name.Param; p != nil && findTypeLoop(p.Ntype.Type, path) { + return true + } + *path = (*path)[:len(*path)-1] + } else { + // Anonymous type. Recurse on contained types. + + switch t.Etype { + case TARRAY: + if findTypeLoop(t.Elem(), path) { + return true + } + case TSTRUCT: + for _, f := range t.Fields().Slice() { + if findTypeLoop(f.Type, path) { + return true + } + } + case TINTER: + for _, m := range t.Methods().Slice() { + if m.Type.IsInterface() { // embedded interface + if findTypeLoop(m.Type, path) { + return true + } + } + } + } + } + + return false +} + +func reportTypeLoop(t *types.Type) { + if t.Broke() { + return + } + + var l []*types.Type + if !findTypeLoop(t, &l) { + Fatalf("failed to find type loop for: %v", t) + } + + // Rotate loop so that the earliest type declaration is first. + i := 0 + for j, t := range l[1:] { + if typePos(t).Before(typePos(l[i])) { + i = j + 1 + } + } + l = append(l[i:], l[:i]...) + + var msg bytes.Buffer + fmt.Fprintf(&msg, "invalid recursive type %v\n", l[0]) + for _, t := range l { + fmt.Fprintf(&msg, "\t%v: %v refers to\n", linestr(typePos(t)), t) + t.SetBroke(true) + } + fmt.Fprintf(&msg, "\t%v: %v", linestr(typePos(l[0])), l[0]) + yyerrorl(typePos(l[0]), msg.String()) +} + // dowidth calculates and stores the size and alignment for t. // If sizeCalculationDisabled is set, and the size/alignment // have not already been calculated, it calls Fatal. @@ -192,11 +279,7 @@ func dowidth(t *types.Type) { } if t.Width == -2 { - if !t.Broke() { - t.SetBroke(true) - yyerrorl(asNode(t.Nod).Pos, "invalid recursive type %v", t) - } - + reportTypeLoop(t) t.Width = 0 t.Align = 1 return @@ -298,7 +381,7 @@ func dowidth(t *types.Type) { t1 := t.ChanArgs() dowidth(t1) // just in case if t1.Elem().Width >= 1<<16 { - yyerror("channel element type too large (>64kB)") + yyerrorl(typePos(t1), "channel element type too large (>64kB)") } w = 1 // anything will do @@ -308,10 +391,7 @@ func dowidth(t *types.Type) { checkwidth(t.Key()) case TFORW: // should have been filled in - if !t.Broke() { - t.SetBroke(true) - yyerror("invalid recursive type %v", t) - } + reportTypeLoop(t) w = 1 // anything will do case TANY: @@ -334,7 +414,7 @@ func dowidth(t *types.Type) { if t.Elem().Width != 0 { cap := (uint64(thearch.MAXWIDTH) - 1) / uint64(t.Elem().Width) if uint64(t.NumElem()) > cap { - yyerror("type %L larger than address space", t) + yyerrorl(typePos(t), "type %L larger than address space", t) } } w = t.NumElem() * t.Elem().Width @@ -376,7 +456,7 @@ func dowidth(t *types.Type) { } if Widthptr == 4 && w != int64(int32(w)) { - yyerror("type %v too large", t) + yyerrorl(typePos(t), "type %v too large", t) } t.Width = w diff --git a/src/cmd/compile/internal/gc/bench_test.go b/src/cmd/compile/internal/gc/bench_test.go index 09aaf428c34..8c4288128f2 100644 --- a/src/cmd/compile/internal/gc/bench_test.go +++ b/src/cmd/compile/internal/gc/bench_test.go @@ -7,6 +7,7 @@ package gc import "testing" var globl int64 +var globl32 int32 func BenchmarkLoadAdd(b *testing.B) { x := make([]int64, 1024) @@ -20,6 +21,18 @@ func BenchmarkLoadAdd(b *testing.B) { } } +// Added for ppc64 extswsli on power9 +func BenchmarkExtShift(b *testing.B) { + x := make([]int32, 1024) + for i := 0; i < b.N; i++ { + var s int64 + for i := range x { + s ^= int64(x[i]+32) * 8 + } + globl = s + } +} + func BenchmarkModify(b *testing.B) { a := make([]int64, 1024) v := globl @@ -30,6 +43,17 @@ func BenchmarkModify(b *testing.B) { } } +func BenchmarkMullImm(b *testing.B) { + x := make([]int32, 1024) + for i := 0; i < b.N; i++ { + var s int32 + for i := range x { + s += x[i] * 100 + } + globl32 = s + } +} + func BenchmarkConstModify(b *testing.B) { a := make([]int64, 1024) for i := 0; i < b.N; i++ { diff --git a/src/cmd/compile/internal/gc/bexport.go b/src/cmd/compile/internal/gc/bexport.go index 5ced66c0da8..10f21f86df5 100644 --- a/src/cmd/compile/internal/gc/bexport.go +++ b/src/cmd/compile/internal/gc/bexport.go @@ -81,11 +81,6 @@ func (p *exporter) markType(t *types.Type) { } } -// deltaNewFile is a magic line delta offset indicating a new file. -// We use -64 because it is rare; see issue 20080 and CL 41619. -// -64 is the smallest int that fits in a single byte as a varint. -const deltaNewFile = -64 - // ---------------------------------------------------------------------------- // Export format @@ -126,30 +121,6 @@ const ( aliasTag ) -// untype returns the "pseudo" untyped type for a Ctype (import/export use only). -// (we can't use a pre-initialized array because we must be sure all types are -// set up) -func untype(ctype Ctype) *types.Type { - switch ctype { - case CTINT: - return types.Idealint - case CTRUNE: - return types.Idealrune - case CTFLT: - return types.Idealfloat - case CTCPLX: - return types.Idealcomplex - case CTSTR: - return types.Idealstring - case CTBOOL: - return types.Idealbool - case CTNIL: - return types.Types[TNIL] - } - Fatalf("exporter: unknown Ctype") - return nil -} - var predecl []*types.Type // initialized lazily func predeclared() []*types.Type { @@ -184,13 +155,13 @@ func predeclared() []*types.Type { types.Errortype, // untyped types - untype(CTBOOL), - untype(CTINT), - untype(CTRUNE), - untype(CTFLT), - untype(CTCPLX), - untype(CTSTR), - untype(CTNIL), + types.UntypedBool, + types.UntypedInt, + types.UntypedRune, + types.UntypedFloat, + types.UntypedComplex, + types.UntypedString, + types.Types[TNIL], // package unsafe types.Types[TUNSAFEPTR], diff --git a/src/cmd/compile/internal/gc/builtin.go b/src/cmd/compile/internal/gc/builtin.go index 861ffaaa5bd..e04f23e2294 100644 --- a/src/cmd/compile/internal/gc/builtin.go +++ b/src/cmd/compile/internal/gc/builtin.go @@ -44,6 +44,7 @@ var runtimeDecls = [...]struct { {"printcomplex", funcTag, 27}, {"printstring", funcTag, 29}, {"printpointer", funcTag, 30}, + {"printuintptr", funcTag, 31}, {"printiface", funcTag, 30}, {"printeface", funcTag, 30}, {"printslice", funcTag, 30}, @@ -51,20 +52,19 @@ var runtimeDecls = [...]struct { {"printsp", funcTag, 9}, {"printlock", funcTag, 9}, {"printunlock", funcTag, 9}, - {"concatstring2", funcTag, 33}, - {"concatstring3", funcTag, 34}, - {"concatstring4", funcTag, 35}, - {"concatstring5", funcTag, 36}, - {"concatstrings", funcTag, 38}, - {"cmpstring", funcTag, 39}, - {"intstring", funcTag, 42}, - {"slicebytetostring", funcTag, 43}, - {"slicebytetostringtmp", funcTag, 44}, - {"slicerunetostring", funcTag, 47}, - {"stringtoslicebyte", funcTag, 49}, - {"stringtoslicerune", funcTag, 52}, - {"slicecopy", funcTag, 53}, - {"slicestringcopy", funcTag, 54}, + {"concatstring2", funcTag, 34}, + {"concatstring3", funcTag, 35}, + {"concatstring4", funcTag, 36}, + {"concatstring5", funcTag, 37}, + {"concatstrings", funcTag, 39}, + {"cmpstring", funcTag, 40}, + {"intstring", funcTag, 43}, + {"slicebytetostring", funcTag, 44}, + {"slicebytetostringtmp", funcTag, 45}, + {"slicerunetostring", funcTag, 48}, + {"stringtoslicebyte", funcTag, 50}, + {"stringtoslicerune", funcTag, 53}, + {"slicecopy", funcTag, 54}, {"decoderune", funcTag, 55}, {"countrunes", funcTag, 56}, {"convI2I", funcTag, 57}, @@ -175,15 +175,16 @@ var runtimeDecls = [...]struct { {"uint64tofloat64", funcTag, 118}, {"uint32tofloat64", funcTag, 119}, {"complex128div", funcTag, 120}, - {"racefuncenter", funcTag, 121}, + {"racefuncenter", funcTag, 31}, {"racefuncenterfp", funcTag, 9}, {"racefuncexit", funcTag, 9}, - {"raceread", funcTag, 121}, - {"racewrite", funcTag, 121}, - {"racereadrange", funcTag, 122}, - {"racewriterange", funcTag, 122}, - {"msanread", funcTag, 122}, - {"msanwrite", funcTag, 122}, + {"raceread", funcTag, 31}, + {"racewrite", funcTag, 31}, + {"racereadrange", funcTag, 121}, + {"racewriterange", funcTag, 121}, + {"msanread", funcTag, 121}, + {"msanwrite", funcTag, 121}, + {"msanmove", funcTag, 122}, {"checkptrAlignment", funcTag, 123}, {"checkptrArithmetic", funcTag, 125}, {"libfuzzerTraceCmp1", funcTag, 127}, @@ -234,31 +235,31 @@ func runtimeTypes() []*types.Type { typs[28] = types.Types[TSTRING] typs[29] = functype(nil, []*Node{anonfield(typs[28])}, nil) typs[30] = functype(nil, []*Node{anonfield(typs[2])}, nil) - typs[31] = types.NewArray(typs[0], 32) - typs[32] = types.NewPtr(typs[31]) - typs[33] = functype(nil, []*Node{anonfield(typs[32]), anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[28])}) - typs[34] = functype(nil, []*Node{anonfield(typs[32]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[28])}) - typs[35] = functype(nil, []*Node{anonfield(typs[32]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[28])}) - typs[36] = functype(nil, []*Node{anonfield(typs[32]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[28])}) - typs[37] = types.NewSlice(typs[28]) - typs[38] = functype(nil, []*Node{anonfield(typs[32]), anonfield(typs[37])}, []*Node{anonfield(typs[28])}) - typs[39] = functype(nil, []*Node{anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[15])}) - typs[40] = types.NewArray(typs[0], 4) - typs[41] = types.NewPtr(typs[40]) - typs[42] = functype(nil, []*Node{anonfield(typs[41]), anonfield(typs[22])}, []*Node{anonfield(typs[28])}) - typs[43] = functype(nil, []*Node{anonfield(typs[32]), anonfield(typs[1]), anonfield(typs[15])}, []*Node{anonfield(typs[28])}) - typs[44] = functype(nil, []*Node{anonfield(typs[1]), anonfield(typs[15])}, []*Node{anonfield(typs[28])}) - typs[45] = types.Runetype - typs[46] = types.NewSlice(typs[45]) - typs[47] = functype(nil, []*Node{anonfield(typs[32]), anonfield(typs[46])}, []*Node{anonfield(typs[28])}) - typs[48] = types.NewSlice(typs[0]) - typs[49] = functype(nil, []*Node{anonfield(typs[32]), anonfield(typs[28])}, []*Node{anonfield(typs[48])}) - typs[50] = types.NewArray(typs[45], 32) - typs[51] = types.NewPtr(typs[50]) - typs[52] = functype(nil, []*Node{anonfield(typs[51]), anonfield(typs[28])}, []*Node{anonfield(typs[46])}) - typs[53] = functype(nil, []*Node{anonfield(typs[3]), anonfield(typs[15]), anonfield(typs[3]), anonfield(typs[15]), anonfield(typs[5])}, []*Node{anonfield(typs[15])}) - typs[54] = functype(nil, []*Node{anonfield(typs[1]), anonfield(typs[15]), anonfield(typs[28])}, []*Node{anonfield(typs[15])}) - typs[55] = functype(nil, []*Node{anonfield(typs[28]), anonfield(typs[15])}, []*Node{anonfield(typs[45]), anonfield(typs[15])}) + typs[31] = functype(nil, []*Node{anonfield(typs[5])}, nil) + typs[32] = types.NewArray(typs[0], 32) + typs[33] = types.NewPtr(typs[32]) + typs[34] = functype(nil, []*Node{anonfield(typs[33]), anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[28])}) + typs[35] = functype(nil, []*Node{anonfield(typs[33]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[28])}) + typs[36] = functype(nil, []*Node{anonfield(typs[33]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[28])}) + typs[37] = functype(nil, []*Node{anonfield(typs[33]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[28])}) + typs[38] = types.NewSlice(typs[28]) + typs[39] = functype(nil, []*Node{anonfield(typs[33]), anonfield(typs[38])}, []*Node{anonfield(typs[28])}) + typs[40] = functype(nil, []*Node{anonfield(typs[28]), anonfield(typs[28])}, []*Node{anonfield(typs[15])}) + typs[41] = types.NewArray(typs[0], 4) + typs[42] = types.NewPtr(typs[41]) + typs[43] = functype(nil, []*Node{anonfield(typs[42]), anonfield(typs[22])}, []*Node{anonfield(typs[28])}) + typs[44] = functype(nil, []*Node{anonfield(typs[33]), anonfield(typs[1]), anonfield(typs[15])}, []*Node{anonfield(typs[28])}) + typs[45] = functype(nil, []*Node{anonfield(typs[1]), anonfield(typs[15])}, []*Node{anonfield(typs[28])}) + typs[46] = types.Runetype + typs[47] = types.NewSlice(typs[46]) + typs[48] = functype(nil, []*Node{anonfield(typs[33]), anonfield(typs[47])}, []*Node{anonfield(typs[28])}) + typs[49] = types.NewSlice(typs[0]) + typs[50] = functype(nil, []*Node{anonfield(typs[33]), anonfield(typs[28])}, []*Node{anonfield(typs[49])}) + typs[51] = types.NewArray(typs[46], 32) + typs[52] = types.NewPtr(typs[51]) + typs[53] = functype(nil, []*Node{anonfield(typs[52]), anonfield(typs[28])}, []*Node{anonfield(typs[47])}) + typs[54] = functype(nil, []*Node{anonfield(typs[3]), anonfield(typs[15]), anonfield(typs[3]), anonfield(typs[15]), anonfield(typs[5])}, []*Node{anonfield(typs[15])}) + typs[55] = functype(nil, []*Node{anonfield(typs[28]), anonfield(typs[15])}, []*Node{anonfield(typs[46]), anonfield(typs[15])}) typs[56] = functype(nil, []*Node{anonfield(typs[28])}, []*Node{anonfield(typs[15])}) typs[57] = functype(nil, []*Node{anonfield(typs[1]), anonfield(typs[2])}, []*Node{anonfield(typs[2])}) typs[58] = functype(nil, []*Node{anonfield(typs[2])}, []*Node{anonfield(typs[7])}) @@ -324,8 +325,8 @@ func runtimeTypes() []*types.Type { typs[118] = functype(nil, []*Node{anonfield(typs[24])}, []*Node{anonfield(typs[20])}) typs[119] = functype(nil, []*Node{anonfield(typs[65])}, []*Node{anonfield(typs[20])}) typs[120] = functype(nil, []*Node{anonfield(typs[26]), anonfield(typs[26])}, []*Node{anonfield(typs[26])}) - typs[121] = functype(nil, []*Node{anonfield(typs[5])}, nil) - typs[122] = functype(nil, []*Node{anonfield(typs[5]), anonfield(typs[5])}, nil) + typs[121] = functype(nil, []*Node{anonfield(typs[5]), anonfield(typs[5])}, nil) + typs[122] = functype(nil, []*Node{anonfield(typs[5]), anonfield(typs[5]), anonfield(typs[5])}, nil) typs[123] = functype(nil, []*Node{anonfield(typs[7]), anonfield(typs[1]), anonfield(typs[5])}, nil) typs[124] = types.NewSlice(typs[7]) typs[125] = functype(nil, []*Node{anonfield(typs[7]), anonfield(typs[124])}, nil) diff --git a/src/cmd/compile/internal/gc/builtin/runtime.go b/src/cmd/compile/internal/gc/builtin/runtime.go index 635da80f7c4..acb69c7b282 100644 --- a/src/cmd/compile/internal/gc/builtin/runtime.go +++ b/src/cmd/compile/internal/gc/builtin/runtime.go @@ -54,6 +54,7 @@ func printuint(uint64) func printcomplex(complex128) func printstring(string) func printpointer(any) +func printuintptr(uintptr) func printiface(any) func printeface(any) func printslice(any) @@ -75,8 +76,7 @@ func slicebytetostringtmp(ptr *byte, n int) string func slicerunetostring(*[32]byte, []rune) string func stringtoslicebyte(*[32]byte, string) []byte func stringtoslicerune(*[32]rune, string) []rune -func slicecopy(toPtr *any, toLen int, frPtr *any, frLen int, wid uintptr) int -func slicestringcopy(toPtr *byte, toLen int, fr string) int +func slicecopy(toPtr *any, toLen int, fromPtr *any, fromLen int, wid uintptr) int func decoderune(string, int) (retv rune, retk int) func countrunes(string) int @@ -237,6 +237,7 @@ func racewriterange(addr, size uintptr) // memory sanitizer func msanread(addr, size uintptr) func msanwrite(addr, size uintptr) +func msanmove(dst, src, size uintptr) func checkptrAlignment(unsafe.Pointer, *byte, uintptr) func checkptrArithmetic(unsafe.Pointer, []unsafe.Pointer) diff --git a/src/cmd/compile/internal/gc/closure.go b/src/cmd/compile/internal/gc/closure.go index 250be38e5bf..bd350f696e8 100644 --- a/src/cmd/compile/internal/gc/closure.go +++ b/src/cmd/compile/internal/gc/closure.go @@ -71,6 +71,10 @@ func (p *noder) funcLit(expr *syntax.FuncLit) *Node { return clo } +// typecheckclosure typechecks an OCLOSURE node. It also creates the named +// function associated with the closure. +// TODO: This creation of the named function should probably really be done in a +// separate pass from type-checking. func typecheckclosure(clo *Node, top int) { xfunc := clo.Func.Closure // Set current associated iota value, so iota can be used inside @@ -198,7 +202,7 @@ func capturevars(xfunc *Node) { outer = nod(OADDR, outer, nil) } - if Debug['m'] > 1 { + if Debug.m > 1 { var name *types.Sym if v.Name.Curfn != nil && v.Name.Curfn.Func.Nname != nil { name = v.Name.Curfn.Func.Nname.Sym @@ -434,6 +438,8 @@ func typecheckpartialcall(fn *Node, sym *types.Sym) { fn.Type = xfunc.Type } +// makepartialcall returns a DCLFUNC node representing the wrapper function (*-fm) needed +// for partial calls. func makepartialcall(fn *Node, t0 *types.Type, meth *types.Sym) *Node { rcvrtype := fn.Left.Type sym := methodSymSuffix(rcvrtype, meth, "-fm") @@ -500,6 +506,10 @@ func makepartialcall(fn *Node, t0 *types.Type, meth *types.Sym) *Node { funcbody() xfunc = typecheck(xfunc, ctxStmt) + // Need to typecheck the body of the just-generated wrapper. + // typecheckslice() requires that Curfn is set when processing an ORETURN. + Curfn = xfunc + typecheckslice(xfunc.Nbody.Slice(), ctxStmt) sym.Def = asTypesNode(xfunc) xtop = append(xtop, xfunc) Curfn = savecurfn diff --git a/src/cmd/compile/internal/gc/const.go b/src/cmd/compile/internal/gc/const.go index fe73df9d574..b92c8d66b5a 100644 --- a/src/cmd/compile/internal/gc/const.go +++ b/src/cmd/compile/internal/gc/const.go @@ -44,7 +44,7 @@ func (v Val) Ctype() Ctype { Fatalf("unexpected Ctype for %T", v.U) panic("unreachable") case nil: - return 0 + return CTxxx case *NilVal: return CTNIL case bool: @@ -114,16 +114,16 @@ func (v Val) Interface() interface{} { type NilVal struct{} -// Int64 returns n as an int64. +// Int64Val returns n as an int64. // n must be an integer or rune constant. -func (n *Node) Int64() int64 { +func (n *Node) Int64Val() int64 { if !Isconst(n, CTINT) { - Fatalf("Int64(%v)", n) + Fatalf("Int64Val(%v)", n) } return n.Val().U.(*Mpint).Int64() } -// CanInt64 reports whether it is safe to call Int64() on n. +// CanInt64 reports whether it is safe to call Int64Val() on n. func (n *Node) CanInt64() bool { if !Isconst(n, CTINT) { return false @@ -131,18 +131,27 @@ func (n *Node) CanInt64() bool { // if the value inside n cannot be represented as an int64, the // return value of Int64 is undefined - return n.Val().U.(*Mpint).CmpInt64(n.Int64()) == 0 + return n.Val().U.(*Mpint).CmpInt64(n.Int64Val()) == 0 } -// Bool returns n as a bool. +// BoolVal returns n as a bool. // n must be a boolean constant. -func (n *Node) Bool() bool { +func (n *Node) BoolVal() bool { if !Isconst(n, CTBOOL) { - Fatalf("Bool(%v)", n) + Fatalf("BoolVal(%v)", n) } return n.Val().U.(bool) } +// StringVal returns the value of a literal string Node as a string. +// n must be a string constant. +func (n *Node) StringVal() string { + if !Isconst(n, CTSTR) { + Fatalf("StringVal(%v)", n) + } + return n.Val().U.(string) +} + // truncate float literal fv to 32-bit or 64-bit precision // according to type; return truncated value. func truncfltlit(oldv *Mpflt, t *types.Type) *Mpflt { @@ -261,7 +270,7 @@ func convlit1(n *Node, t *types.Type, explicit bool, context func() string) *Nod } if t == nil || !okforconst[t.Etype] { - t = defaultType(idealkind(n)) + t = defaultType(n.Type) } switch n.Op { @@ -612,7 +621,7 @@ func evconst(n *Node) { var strs []string i2 := i1 for i2 < len(s) && Isconst(s[i2], CTSTR) { - strs = append(strs, strlit(s[i2])) + strs = append(strs, s[i2].StringVal()) i2++ } @@ -635,7 +644,7 @@ func evconst(n *Node) { switch nl.Type.Etype { case TSTRING: if Isconst(nl, CTSTR) { - setintconst(n, int64(len(strlit(nl)))) + setintconst(n, int64(len(nl.StringVal()))) } case TARRAY: if !hascallchan(nl) { @@ -838,10 +847,6 @@ Outer: return Val{} } u.Quo(y) - case OMOD, OOR, OAND, OANDNOT, OXOR: - // TODO(mdempsky): Move to typecheck; see #31060. - yyerror("invalid operation: operator %v not defined on untyped float", op) - return Val{} default: break Outer } @@ -867,10 +872,6 @@ Outer: yyerror("complex division by zero") return Val{} } - case OMOD, OOR, OAND, OANDNOT, OXOR: - // TODO(mdempsky): Move to typecheck; see #31060. - yyerror("invalid operation: operator %v not defined on untyped complex", op) - return Val{} default: break Outer } @@ -932,15 +933,6 @@ func unaryOp(op Op, x Val, t *types.Type) Val { } u.Xor(x) return Val{U: u} - - case CTFLT: - // TODO(mdempsky): Move to typecheck; see #31060. - yyerror("invalid operation: operator %v not defined on untyped float", op) - return Val{} - case CTCPLX: - // TODO(mdempsky): Move to typecheck; see #31060. - yyerror("invalid operation: operator %v not defined on untyped complex", op) - return Val{} } case ONOT: @@ -994,10 +986,8 @@ func setconst(n *Node, v Val) { Xoffset: BADWIDTH, } n.SetVal(v) - if n.Type.IsUntyped() { - // TODO(mdempsky): Make typecheck responsible for setting - // the correct untyped type. - n.Type = idealType(v.Ctype()) + if vt := idealType(v.Ctype()); n.Type.IsUntyped() && n.Type != vt { + Fatalf("untyped type mismatch, have: %v, want: %v", n.Type, vt) } // Check range. @@ -1038,17 +1028,17 @@ func nodlit(v Val) *Node { func idealType(ct Ctype) *types.Type { switch ct { case CTSTR: - return types.Idealstring + return types.UntypedString case CTBOOL: - return types.Idealbool + return types.UntypedBool case CTINT: - return types.Idealint + return types.UntypedInt case CTRUNE: - return types.Idealrune + return types.UntypedRune case CTFLT: - return types.Idealfloat + return types.UntypedFloat case CTCPLX: - return types.Idealcomplex + return types.UntypedComplex case CTNIL: return types.Types[TNIL] } @@ -1056,67 +1046,6 @@ func idealType(ct Ctype) *types.Type { return nil } -// idealkind returns a constant kind like consttype -// but for an arbitrary "ideal" (untyped constant) expression. -func idealkind(n *Node) Ctype { - if n == nil || !n.Type.IsUntyped() { - return CTxxx - } - - switch n.Op { - default: - return CTxxx - - case OLITERAL: - return n.Val().Ctype() - - // numeric kinds. - case OADD, - OAND, - OANDNOT, - OBITNOT, - ODIV, - ONEG, - OMOD, - OMUL, - OSUB, - OXOR, - OOR, - OPLUS: - k1 := idealkind(n.Left) - k2 := idealkind(n.Right) - if k1 > k2 { - return k1 - } else { - return k2 - } - - case OREAL, OIMAG: - return CTFLT - - case OCOMPLEX: - return CTCPLX - - case OADDSTR: - return CTSTR - - case OANDAND, - OEQ, - OGE, - OGT, - OLE, - OLT, - ONE, - ONOT, - OOROR: - return CTBOOL - - // shifts (beware!). - case OLSH, ORSH: - return idealkind(n.Left) - } -} - // defaultlit on both nodes simultaneously; // if they're both ideal going in they better // get the same type going out. @@ -1152,41 +1081,63 @@ func defaultlit2(l *Node, r *Node, force bool) (*Node, *Node) { return l, r } - k := idealkind(l) - if rk := idealkind(r); rk > k { - k = rk - } - t := defaultType(k) + t := defaultType(mixUntyped(l.Type, r.Type)) l = convlit(l, t) r = convlit(r, t) return l, r } -func defaultType(k Ctype) *types.Type { - switch k { - case CTBOOL: +func ctype(t *types.Type) Ctype { + switch t { + case types.UntypedBool: + return CTBOOL + case types.UntypedString: + return CTSTR + case types.UntypedInt: + return CTINT + case types.UntypedRune: + return CTRUNE + case types.UntypedFloat: + return CTFLT + case types.UntypedComplex: + return CTCPLX + } + Fatalf("bad type %v", t) + panic("unreachable") +} + +func mixUntyped(t1, t2 *types.Type) *types.Type { + t := t1 + if ctype(t2) > ctype(t1) { + t = t2 + } + return t +} + +func defaultType(t *types.Type) *types.Type { + if !t.IsUntyped() || t.Etype == TNIL { + return t + } + + switch t { + case types.UntypedBool: return types.Types[TBOOL] - case CTSTR: + case types.UntypedString: return types.Types[TSTRING] - case CTINT: + case types.UntypedInt: return types.Types[TINT] - case CTRUNE: + case types.UntypedRune: return types.Runetype - case CTFLT: + case types.UntypedFloat: return types.Types[TFLOAT64] - case CTCPLX: + case types.UntypedComplex: return types.Types[TCOMPLEX128] } - Fatalf("bad idealkind: %v", k) + + Fatalf("bad type %v", t) return nil } -// strlit returns the value of a literal string Node as a string. -func strlit(n *Node) string { - return n.Val().U.(string) -} - -// TODO(gri) smallintconst is only used in one place - can we used indexconst? func smallintconst(n *Node) bool { if n.Op == OLITERAL && Isconst(n, CTINT) && n.Type != nil { switch simtype[n.Type.Etype] { diff --git a/src/cmd/compile/internal/gc/dcl.go b/src/cmd/compile/internal/gc/dcl.go index 69eb13f6076..6e90eb4d65b 100644 --- a/src/cmd/compile/internal/gc/dcl.go +++ b/src/cmd/compile/internal/gc/dcl.go @@ -257,6 +257,8 @@ func symfield(s *types.Sym, typ *types.Type) *Node { // oldname returns the Node that declares symbol s in the current scope. // If no such Node currently exists, an ONONAME Node is returned instead. +// Automatically creates a new closure variable if the referenced symbol was +// declared in a different (containing) function. func oldname(s *types.Sym) *Node { n := asNode(s.Def) if n == nil { @@ -283,7 +285,7 @@ func oldname(s *types.Sym) *Node { c.Name.Defn = n // Link into list of active closure variables. - // Popped from list in func closurebody. + // Popped from list in func funcLit. c.Name.Param.Outer = n.Name.Param.Innermost n.Name.Param.Innermost = c @@ -382,14 +384,11 @@ func ifacedcl(n *Node) { // returns in auto-declaration context. func funchdr(n *Node) { // change the declaration context from extern to auto - if Curfn == nil && dclcontext != PEXTERN { - Fatalf("funchdr: dclcontext = %d", dclcontext) - } - - dclcontext = PAUTO - types.Markdcl() - funcstack = append(funcstack, Curfn) + funcStack = append(funcStack, funcStackEnt{Curfn, dclcontext}) Curfn = n + dclcontext = PAUTO + + types.Markdcl() if n.Func.Nname != nil { funcargs(n.Func.Nname.Name.Param.Ntype) @@ -497,21 +496,22 @@ func funcarg2(f *types.Field, ctxt Class) { declare(n, ctxt) } -var funcstack []*Node // stack of previous values of Curfn +var funcStack []funcStackEnt // stack of previous values of Curfn/dclcontext + +type funcStackEnt struct { + curfn *Node + dclcontext Class +} // finish the body. // called in auto-declaration context. // returns in extern-declaration context. func funcbody() { - // change the declaration context from auto to extern - if dclcontext != PAUTO { - Fatalf("funcbody: unexpected dclcontext %d", dclcontext) - } + // change the declaration context from auto to previous context types.Popdcl() - funcstack, Curfn = funcstack[:len(funcstack)-1], funcstack[len(funcstack)-1] - if Curfn == nil { - dclcontext = PEXTERN - } + var e funcStackEnt + funcStack, e = funcStack[:len(funcStack)-1], funcStack[len(funcStack)-1] + Curfn, dclcontext = e.curfn, e.dclcontext } // structs, functions, and methods. diff --git a/src/cmd/compile/internal/gc/dwinl.go b/src/cmd/compile/internal/gc/dwinl.go index 27e2cbcd982..bb5ae61cbb3 100644 --- a/src/cmd/compile/internal/gc/dwinl.go +++ b/src/cmd/compile/internal/gc/dwinl.go @@ -8,6 +8,7 @@ import ( "cmd/internal/dwarf" "cmd/internal/obj" "cmd/internal/src" + "fmt" "strings" ) @@ -34,7 +35,7 @@ func assembleInlines(fnsym *obj.LSym, dwVars []*dwarf.Var) dwarf.InlCalls { // Walk progs to build up the InlCalls data structure var prevpos src.XPos - for p := fnsym.Func.Text; p != nil; p = p.Link { + for p := fnsym.Func().Text; p != nil; p = p.Link { if p.Pos == prevpos { continue } @@ -150,7 +151,7 @@ func assembleInlines(fnsym *obj.LSym, dwVars []*dwarf.Var) dwarf.InlCalls { start := int64(-1) curii := -1 var prevp *obj.Prog - for p := fnsym.Func.Text; p != nil; prevp, p = p, p.Link { + for p := fnsym.Func().Text; p != nil; prevp, p = p, p.Link { if prevp != nil && p.Pos == prevp.Pos { continue } @@ -170,12 +171,32 @@ func assembleInlines(fnsym *obj.LSym, dwVars []*dwarf.Var) dwarf.InlCalls { addRange(inlcalls.Calls, start, fnsym.Size, curii, imap) } + // Issue 33188: if II foo is a child of II bar, then ensure that + // bar's ranges include the ranges of foo (the loop above will produce + // disjoint ranges). + for k, c := range inlcalls.Calls { + if c.Root { + unifyCallRanges(inlcalls, k) + } + } + // Debugging if Debug_gendwarfinl != 0 { dumpInlCalls(inlcalls) dumpInlVars(dwVars) } + // Perform a consistency check on inlined routine PC ranges + // produced by unifyCallRanges above. In particular, complain in + // cases where you have A -> B -> C (e.g. C is inlined into B, and + // B is inlined into A) and the ranges for B are not enclosed + // within the ranges for A, or C within B. + for k, c := range inlcalls.Calls { + if c.Root { + checkInlCall(fnsym.Name, inlcalls, fnsym.Size, k, -1) + } + } + return inlcalls } @@ -355,3 +376,74 @@ func dumpInlVars(dwvars []*dwarf.Var) { Ctxt.Logf("V%d: %s CI:%d II:%d IA:%d %s\n", i, dwv.Name, dwv.ChildIndex, dwv.InlIndex-1, ia, typ) } } + +func rangesContains(par []dwarf.Range, rng dwarf.Range) (bool, string) { + for _, r := range par { + if rng.Start >= r.Start && rng.End <= r.End { + return true, "" + } + } + msg := fmt.Sprintf("range [%d,%d) not contained in {", rng.Start, rng.End) + for _, r := range par { + msg += fmt.Sprintf(" [%d,%d)", r.Start, r.End) + } + msg += " }" + return false, msg +} + +func rangesContainsAll(parent, child []dwarf.Range) (bool, string) { + for _, r := range child { + c, m := rangesContains(parent, r) + if !c { + return false, m + } + } + return true, "" +} + +// checkInlCall verifies that the PC ranges for inline info 'idx' are +// enclosed/contained within the ranges of its parent inline (or if +// this is a root/toplevel inline, checks that the ranges fall within +// the extent of the top level function). A panic is issued if a +// malformed range is found. +func checkInlCall(funcName string, inlCalls dwarf.InlCalls, funcSize int64, idx, parentIdx int) { + + // Callee + ic := inlCalls.Calls[idx] + callee := Ctxt.InlTree.InlinedFunction(ic.InlIndex).Name + calleeRanges := ic.Ranges + + // Caller + caller := funcName + parentRanges := []dwarf.Range{dwarf.Range{Start: int64(0), End: funcSize}} + if parentIdx != -1 { + pic := inlCalls.Calls[parentIdx] + caller = Ctxt.InlTree.InlinedFunction(pic.InlIndex).Name + parentRanges = pic.Ranges + } + + // Callee ranges contained in caller ranges? + c, m := rangesContainsAll(parentRanges, calleeRanges) + if !c { + Fatalf("** malformed inlined routine range in %s: caller %s callee %s II=%d %s\n", funcName, caller, callee, idx, m) + } + + // Now visit kids + for _, k := range ic.Children { + checkInlCall(funcName, inlCalls, funcSize, k, idx) + } +} + +// unifyCallRanges ensures that the ranges for a given inline +// transitively include all of the ranges for its child inlines. +func unifyCallRanges(inlcalls dwarf.InlCalls, idx int) { + ic := &inlcalls.Calls[idx] + for _, childIdx := range ic.Children { + // First make sure child ranges are unified. + unifyCallRanges(inlcalls, childIdx) + + // Then merge child ranges into ranges for this inline. + cic := inlcalls.Calls[childIdx] + ic.Ranges = dwarf.MergeRanges(ic.Ranges, cic.Ranges) + } +} diff --git a/src/cmd/compile/internal/gc/embed.go b/src/cmd/compile/internal/gc/embed.go new file mode 100644 index 00000000000..103949c1f9f --- /dev/null +++ b/src/cmd/compile/internal/gc/embed.go @@ -0,0 +1,273 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package gc + +import ( + "cmd/compile/internal/syntax" + "cmd/compile/internal/types" + "cmd/internal/obj" + "encoding/json" + "io/ioutil" + "log" + "path" + "sort" + "strconv" + "strings" +) + +var embedlist []*Node + +var embedCfg struct { + Patterns map[string][]string + Files map[string]string +} + +func readEmbedCfg(file string) { + data, err := ioutil.ReadFile(file) + if err != nil { + log.Fatalf("-embedcfg: %v", err) + } + if err := json.Unmarshal(data, &embedCfg); err != nil { + log.Fatalf("%s: %v", file, err) + } + if embedCfg.Patterns == nil { + log.Fatalf("%s: invalid embedcfg: missing Patterns", file) + } + if embedCfg.Files == nil { + log.Fatalf("%s: invalid embedcfg: missing Files", file) + } +} + +const ( + embedUnknown = iota + embedBytes + embedString + embedFiles +) + +var numLocalEmbed int + +func varEmbed(p *noder, names []*Node, typ *Node, exprs []*Node, embeds []PragmaEmbed) (newExprs []*Node) { + haveEmbed := false + for _, decl := range p.file.DeclList { + imp, ok := decl.(*syntax.ImportDecl) + if !ok { + // imports always come first + break + } + path, _ := strconv.Unquote(imp.Path.Value) + if path == "embed" { + haveEmbed = true + break + } + } + + pos := embeds[0].Pos + if !haveEmbed { + p.yyerrorpos(pos, "invalid go:embed: missing import \"embed\"") + return exprs + } + if embedCfg.Patterns == nil { + p.yyerrorpos(pos, "invalid go:embed: build system did not supply embed configuration") + return exprs + } + if len(names) > 1 { + p.yyerrorpos(pos, "go:embed cannot apply to multiple vars") + return exprs + } + if len(exprs) > 0 { + p.yyerrorpos(pos, "go:embed cannot apply to var with initializer") + return exprs + } + if typ == nil { + // Should not happen, since len(exprs) == 0 now. + p.yyerrorpos(pos, "go:embed cannot apply to var without type") + return exprs + } + + kind := embedKindApprox(typ) + if kind == embedUnknown { + p.yyerrorpos(pos, "go:embed cannot apply to var of type %v", typ) + return exprs + } + + // Build list of files to store. + have := make(map[string]bool) + var list []string + for _, e := range embeds { + for _, pattern := range e.Patterns { + files, ok := embedCfg.Patterns[pattern] + if !ok { + p.yyerrorpos(e.Pos, "invalid go:embed: build system did not map pattern: %s", pattern) + } + for _, file := range files { + if embedCfg.Files[file] == "" { + p.yyerrorpos(e.Pos, "invalid go:embed: build system did not map file: %s", file) + continue + } + if !have[file] { + have[file] = true + list = append(list, file) + } + if kind == embedFiles { + for dir := path.Dir(file); dir != "." && !have[dir]; dir = path.Dir(dir) { + have[dir] = true + list = append(list, dir+"/") + } + } + } + } + } + sort.Slice(list, func(i, j int) bool { + return embedFileLess(list[i], list[j]) + }) + + if kind == embedString || kind == embedBytes { + if len(list) > 1 { + p.yyerrorpos(pos, "invalid go:embed: multiple files for type %v", typ) + return exprs + } + } + + v := names[0] + if dclcontext != PEXTERN { + numLocalEmbed++ + v = newnamel(v.Pos, lookupN("embed.", numLocalEmbed)) + v.Sym.Def = asTypesNode(v) + v.Name.Param.Ntype = typ + v.SetClass(PEXTERN) + externdcl = append(externdcl, v) + exprs = []*Node{v} + } + + v.Name.Param.SetEmbedFiles(list) + embedlist = append(embedlist, v) + return exprs +} + +// embedKindApprox determines the kind of embedding variable, approximately. +// The match is approximate because we haven't done scope resolution yet and +// can't tell whether "string" and "byte" really mean "string" and "byte". +// The result must be confirmed later, after type checking, using embedKind. +func embedKindApprox(typ *Node) int { + if typ.Sym != nil && typ.Sym.Name == "FS" && (typ.Sym.Pkg.Path == "embed" || (typ.Sym.Pkg == localpkg && myimportpath == "embed")) { + return embedFiles + } + // These are not guaranteed to match only string and []byte - + // maybe the local package has redefined one of those words. + // But it's the best we can do now during the noder. + // The stricter check happens later, in initEmbed calling embedKind. + if typ.Sym != nil && typ.Sym.Name == "string" && typ.Sym.Pkg == localpkg { + return embedString + } + if typ.Op == OTARRAY && typ.Left == nil && typ.Right.Sym != nil && typ.Right.Sym.Name == "byte" && typ.Right.Sym.Pkg == localpkg { + return embedBytes + } + return embedUnknown +} + +// embedKind determines the kind of embedding variable. +func embedKind(typ *types.Type) int { + if typ.Sym != nil && typ.Sym.Name == "FS" && (typ.Sym.Pkg.Path == "embed" || (typ.Sym.Pkg == localpkg && myimportpath == "embed")) { + return embedFiles + } + if typ == types.Types[TSTRING] { + return embedString + } + if typ.Sym == nil && typ.IsSlice() && typ.Elem() == types.Bytetype { + return embedBytes + } + return embedUnknown +} + +func embedFileNameSplit(name string) (dir, elem string, isDir bool) { + if name[len(name)-1] == '/' { + isDir = true + name = name[:len(name)-1] + } + i := len(name) - 1 + for i >= 0 && name[i] != '/' { + i-- + } + if i < 0 { + return ".", name, isDir + } + return name[:i], name[i+1:], isDir +} + +// embedFileLess implements the sort order for a list of embedded files. +// See the comment inside ../../../../embed/embed.go's Files struct for rationale. +func embedFileLess(x, y string) bool { + xdir, xelem, _ := embedFileNameSplit(x) + ydir, yelem, _ := embedFileNameSplit(y) + return xdir < ydir || xdir == ydir && xelem < yelem +} + +func dumpembeds() { + for _, v := range embedlist { + initEmbed(v) + } +} + +// initEmbed emits the init data for a //go:embed variable, +// which is either a string, a []byte, or an embed.FS. +func initEmbed(v *Node) { + files := v.Name.Param.EmbedFiles() + switch kind := embedKind(v.Type); kind { + case embedUnknown: + yyerrorl(v.Pos, "go:embed cannot apply to var of type %v", v.Type) + + case embedString, embedBytes: + file := files[0] + fsym, size, err := fileStringSym(v.Pos, embedCfg.Files[file], kind == embedString, nil) + if err != nil { + yyerrorl(v.Pos, "embed %s: %v", file, err) + } + sym := v.Sym.Linksym() + off := 0 + off = dsymptr(sym, off, fsym, 0) // data string + off = duintptr(sym, off, uint64(size)) // len + if kind == embedBytes { + duintptr(sym, off, uint64(size)) // cap for slice + } + + case embedFiles: + slicedata := Ctxt.Lookup(`"".` + v.Sym.Name + `.files`) + off := 0 + // []files pointed at by Files + off = dsymptr(slicedata, off, slicedata, 3*Widthptr) // []file, pointing just past slice + off = duintptr(slicedata, off, uint64(len(files))) + off = duintptr(slicedata, off, uint64(len(files))) + + // embed/embed.go type file is: + // name string + // data string + // hash [16]byte + // Emit one of these per file in the set. + const hashSize = 16 + hash := make([]byte, hashSize) + for _, file := range files { + off = dsymptr(slicedata, off, stringsym(v.Pos, file), 0) // file string + off = duintptr(slicedata, off, uint64(len(file))) + if strings.HasSuffix(file, "/") { + // entry for directory - no data + off = duintptr(slicedata, off, 0) + off = duintptr(slicedata, off, 0) + off += hashSize + } else { + fsym, size, err := fileStringSym(v.Pos, embedCfg.Files[file], true, hash) + if err != nil { + yyerrorl(v.Pos, "embed %s: %v", file, err) + } + off = dsymptr(slicedata, off, fsym, 0) // data string + off = duintptr(slicedata, off, uint64(size)) + off = int(slicedata.WriteBytes(Ctxt, int64(off), hash)) + } + } + ggloblsym(slicedata, int32(off), obj.RODATA|obj.LOCAL) + sym := v.Sym.Linksym() + dsymptr(sym, 0, slicedata, 0) + } +} diff --git a/src/cmd/compile/internal/gc/esc.go b/src/cmd/compile/internal/gc/esc.go index 4b843aba35b..6f328ab5ea9 100644 --- a/src/cmd/compile/internal/gc/esc.go +++ b/src/cmd/compile/internal/gc/esc.go @@ -169,36 +169,47 @@ func mayAffectMemory(n *Node) bool { } } -func mustHeapAlloc(n *Node) bool { +// heapAllocReason returns the reason the given Node must be heap +// allocated, or the empty string if it doesn't. +func heapAllocReason(n *Node) string { if n.Type == nil { - return false + return "" } // Parameters are always passed via the stack. if n.Op == ONAME && (n.Class() == PPARAM || n.Class() == PPARAMOUT) { - return false + return "" } if n.Type.Width > maxStackVarSize { - return true + return "too large for stack" } if (n.Op == ONEW || n.Op == OPTRLIT) && n.Type.Elem().Width >= maxImplicitStackVarSize { - return true + return "too large for stack" } if n.Op == OCLOSURE && closureType(n).Size() >= maxImplicitStackVarSize { - return true + return "too large for stack" } if n.Op == OCALLPART && partialCallType(n).Size() >= maxImplicitStackVarSize { - return true + return "too large for stack" } - if n.Op == OMAKESLICE && !isSmallMakeSlice(n) { - return true + if n.Op == OMAKESLICE { + r := n.Right + if r == nil { + r = n.Left + } + if !smallintconst(r) { + return "non-constant size" + } + if t := n.Type; t.Elem().Width != 0 && r.Int64Val() >= maxImplicitStackVarSize/t.Elem().Width { + return "too large for stack" + } } - return false + return "" } // addrescapes tags node n as having had its address taken @@ -271,7 +282,7 @@ func addrescapes(n *Node) { // moveToHeap records the parameter or local variable n as moved to the heap. func moveToHeap(n *Node) { - if Debug['r'] != 0 { + if Debug.r != 0 { Dump("MOVE", n) } if compiling_runtime { @@ -348,7 +359,7 @@ func moveToHeap(n *Node) { n.Xoffset = 0 n.Name.Param.Heapaddr = heapaddr n.Esc = EscHeap - if Debug['m'] != 0 { + if Debug.m != 0 { Warnl(n.Pos, "moved to heap: %v", n) } } @@ -377,8 +388,8 @@ func (e *Escape) paramTag(fn *Node, narg int, f *types.Field) string { // This really doesn't have much to do with escape analysis per se, // but we are reusing the ability to annotate an individual function // argument and pass those annotations along to importing code. - if f.Type.Etype == TUINTPTR { - if Debug['m'] != 0 { + if f.Type.IsUintptr() { + if Debug.m != 0 { Warnl(f.Pos, "assuming %v is unsafe uintptr", name()) } return unsafeUintptrTag @@ -393,11 +404,11 @@ func (e *Escape) paramTag(fn *Node, narg int, f *types.Field) string { // External functions are assumed unsafe, unless // //go:noescape is given before the declaration. if fn.Func.Pragma&Noescape != 0 { - if Debug['m'] != 0 && f.Sym != nil { + if Debug.m != 0 && f.Sym != nil { Warnl(f.Pos, "%v does not escape", name()) } } else { - if Debug['m'] != 0 && f.Sym != nil { + if Debug.m != 0 && f.Sym != nil { Warnl(f.Pos, "leaking param: %v", name()) } esc.AddHeap(0) @@ -407,15 +418,15 @@ func (e *Escape) paramTag(fn *Node, narg int, f *types.Field) string { } if fn.Func.Pragma&UintptrEscapes != 0 { - if f.Type.Etype == TUINTPTR { - if Debug['m'] != 0 { + if f.Type.IsUintptr() { + if Debug.m != 0 { Warnl(f.Pos, "marking %v as escaping uintptr", name()) } return uintptrEscapesTag } - if f.IsDDD() && f.Type.Elem().Etype == TUINTPTR { + if f.IsDDD() && f.Type.Elem().IsUintptr() { // final argument is ...uintptr. - if Debug['m'] != 0 { + if Debug.m != 0 { Warnl(f.Pos, "marking %v as escaping ...uintptr", name()) } return uintptrEscapesTag @@ -437,7 +448,7 @@ func (e *Escape) paramTag(fn *Node, narg int, f *types.Field) string { esc := loc.paramEsc esc.Optimize() - if Debug['m'] != 0 && !loc.escapes { + if Debug.m != 0 && !loc.escapes { if esc.Empty() { Warnl(f.Pos, "%v does not escape", name()) } diff --git a/src/cmd/compile/internal/gc/escape.go b/src/cmd/compile/internal/gc/escape.go index d5cca4a38bd..618bdf78e25 100644 --- a/src/cmd/compile/internal/gc/escape.go +++ b/src/cmd/compile/internal/gc/escape.go @@ -170,7 +170,7 @@ func (e *Escape) initFunc(fn *Node) { Fatalf("unexpected node: %v", fn) } fn.Esc = EscFuncPlanned - if Debug['m'] > 3 { + if Debug.m > 3 { Dump("escAnalyze", fn) } @@ -247,7 +247,7 @@ func (e *Escape) stmt(n *Node) { lineno = lno }() - if Debug['m'] > 2 { + if Debug.m > 2 { fmt.Printf("%v:[%d] %v stmt: %v\n", linestr(lineno), e.loopDepth, funcSym(e.curfn), n) } @@ -275,11 +275,11 @@ func (e *Escape) stmt(n *Node) { case OLABEL: switch asNode(n.Sym.Label) { case &nonlooping: - if Debug['m'] > 2 { + if Debug.m > 2 { fmt.Printf("%v:%v non-looping label\n", linestr(lineno), n) } case &looping: - if Debug['m'] > 2 { + if Debug.m > 2 { fmt.Printf("%v: %v looping label\n", linestr(lineno), n) } e.loopDepth++ @@ -485,7 +485,7 @@ func (e *Escape) exprSkipInit(k EscHole, n *Node) { e.discard(max) case OCONV, OCONVNOP: - if checkPtr(e.curfn, 2) && n.Type.Etype == TUNSAFEPTR && n.Left.Type.IsPtr() { + if checkPtr(e.curfn, 2) && n.Type.IsUnsafePtr() && n.Left.Type.IsPtr() { // When -d=checkptr=2 is enabled, treat // conversions to unsafe.Pointer as an // escaping operation. This allows better @@ -493,7 +493,7 @@ func (e *Escape) exprSkipInit(k EscHole, n *Node) { // easily detect object boundaries on the heap // than the stack. e.assignHeap(n.Left, "conversion to unsafe.Pointer", n) - } else if n.Type.Etype == TUNSAFEPTR && n.Left.Type.Etype == TUINTPTR { + } else if n.Type.IsUnsafePtr() && n.Left.Type.IsUintptr() { e.unsafeValue(k, n.Left) } else { e.expr(k, n.Left) @@ -625,7 +625,7 @@ func (e *Escape) unsafeValue(k EscHole, n *Node) { switch n.Op { case OCONV, OCONVNOP: - if n.Left.Type.Etype == TUNSAFEPTR { + if n.Left.Type.IsUnsafePtr() { e.expr(k, n.Left) } else { e.discard(n.Left) @@ -717,7 +717,7 @@ func (e *Escape) addrs(l Nodes) []EscHole { func (e *Escape) assign(dst, src *Node, why string, where *Node) { // Filter out some no-op assignments for escape analysis. ignore := dst != nil && src != nil && isSelfAssign(dst, src) - if ignore && Debug['m'] != 0 { + if ignore && Debug.m != 0 { Warnl(where.Pos, "%v ignoring self-assignment in %S", funcSym(e.curfn), where) } @@ -771,10 +771,11 @@ func (e *Escape) call(ks []EscHole, call, where *Node) { var fn *Node switch call.Op { case OCALLFUNC: - if call.Left.Op == ONAME && call.Left.Class() == PFUNC { - fn = call.Left - } else if call.Left.Op == OCLOSURE { - fn = call.Left.Func.Closure.Func.Nname + switch v := staticValue(call.Left); { + case v.Op == ONAME && v.Class() == PFUNC: + fn = v + case v.Op == OCLOSURE: + fn = v.Func.Closure.Func.Nname } case OCALLMETH: fn = asNode(call.Left.Type.FuncType().Nname) @@ -930,7 +931,7 @@ func (k EscHole) note(where *Node, why string) EscHole { if where == nil || why == "" { Fatalf("note: missing where/why") } - if Debug['m'] >= 2 || logopt.Enabled() { + if Debug.m >= 2 || logopt.Enabled() { k.notes = &EscNote{ next: k.notes, where: where, @@ -1030,7 +1031,7 @@ func (e *Escape) newLoc(n *Node, transient bool) *EscLocation { Fatalf("e.curfn isn't set") } if n != nil && n.Type != nil && n.Type.NotInHeap() { - yyerrorl(n.Pos, "%v is go:notinheap; stack allocation disallowed", n.Type) + yyerrorl(n.Pos, "%v is incomplete (or unallocatable); stack allocation disallowed", n.Type) } n = canonicalNode(n) @@ -1051,11 +1052,7 @@ func (e *Escape) newLoc(n *Node, transient bool) *EscLocation { } n.SetOpt(loc) - if mustHeapAlloc(n) { - why := "too large for stack" - if n.Op == OMAKESLICE && (!Isconst(n.Left, CTINT) || !Isconst(n.Right, CTINT)) { - why = "non-constant size" - } + if why := heapAllocReason(n); why != "" { e.flow(e.heapHole().addr(n, why), loc) } } @@ -1080,9 +1077,9 @@ func (e *Escape) flow(k EscHole, src *EscLocation) { return } if dst.escapes && k.derefs < 0 { // dst = &src - if Debug['m'] >= 2 || logopt.Enabled() { + if Debug.m >= 2 || logopt.Enabled() { pos := linestr(src.n.Pos) - if Debug['m'] >= 2 { + if Debug.m >= 2 { fmt.Printf("%s: %v escapes to heap:\n", pos, src.n) } explanation := e.explainFlow(pos, dst, src, k.derefs, k.notes, []*logopt.LoggedOpt{}) @@ -1182,8 +1179,8 @@ func (e *Escape) walkOne(root *EscLocation, walkgen uint32, enqueue func(*EscLoc // that value flow for tagging the function // later. if l.isName(PPARAM) { - if (logopt.Enabled() || Debug['m'] >= 2) && !l.escapes { - if Debug['m'] >= 2 { + if (logopt.Enabled() || Debug.m >= 2) && !l.escapes { + if Debug.m >= 2 { fmt.Printf("%s: parameter %v leaks to %s with derefs=%d:\n", linestr(l.n.Pos), l.n, e.explainLoc(root), base) } explanation := e.explainPath(root, l) @@ -1199,8 +1196,8 @@ func (e *Escape) walkOne(root *EscLocation, walkgen uint32, enqueue func(*EscLoc // outlives it, then l needs to be heap // allocated. if addressOf && !l.escapes { - if logopt.Enabled() || Debug['m'] >= 2 { - if Debug['m'] >= 2 { + if logopt.Enabled() || Debug.m >= 2 { + if Debug.m >= 2 { fmt.Printf("%s: %v escapes to heap:\n", linestr(l.n.Pos), l.n) } explanation := e.explainPath(root, l) @@ -1238,7 +1235,7 @@ func (e *Escape) explainPath(root, src *EscLocation) []*logopt.LoggedOpt { for { // Prevent infinite loop. if visited[src] { - if Debug['m'] >= 2 { + if Debug.m >= 2 { fmt.Printf("%s: warning: truncated explanation due to assignment cycle; see golang.org/issue/35518\n", pos) } break @@ -1266,7 +1263,7 @@ func (e *Escape) explainFlow(pos string, dst, srcloc *EscLocation, derefs int, n if derefs >= 0 { ops = strings.Repeat("*", derefs) } - print := Debug['m'] >= 2 + print := Debug.m >= 2 flow := fmt.Sprintf(" flow: %s = %s%v:", e.explainLoc(dst), ops, e.explainLoc(srcloc)) if print { @@ -1420,7 +1417,7 @@ func (e *Escape) finish(fns []*Node) { if loc.escapes { if n.Op != ONAME { - if Debug['m'] != 0 { + if Debug.m != 0 { Warnl(n.Pos, "%S escapes to heap", n) } if logopt.Enabled() { @@ -1430,7 +1427,7 @@ func (e *Escape) finish(fns []*Node) { n.Esc = EscHeap addrescapes(n) } else { - if Debug['m'] != 0 && n.Op != ONAME { + if Debug.m != 0 && n.Op != ONAME { Warnl(n.Pos, "%S does not escape", n) } n.Esc = EscNone diff --git a/src/cmd/compile/internal/gc/export.go b/src/cmd/compile/internal/gc/export.go index 44bea2b1fd4..c6917e0f810 100644 --- a/src/cmd/compile/internal/gc/export.go +++ b/src/cmd/compile/internal/gc/export.go @@ -31,7 +31,7 @@ func exportsym(n *Node) { } n.Sym.SetOnExportList(true) - if Debug['E'] != 0 { + if Debug.E != 0 { fmt.Printf("export symbol %v\n", n.Sym) } @@ -96,7 +96,7 @@ func importsym(ipkg *types.Pkg, s *types.Sym, op Op) *Node { return n } -// pkgtype returns the named type declared by symbol s. +// importtype returns the named type declared by symbol s. // If no such type has been declared yet, a forward declaration is returned. // ipkg is the package being imported func importtype(ipkg *types.Pkg, pos src.XPos, s *types.Sym) *types.Type { @@ -150,7 +150,7 @@ func importconst(ipkg *types.Pkg, pos src.XPos, s *types.Sym, t *types.Type, val n.SetVal(val) - if Debug['E'] != 0 { + if Debug.E != 0 { fmt.Printf("import const %v %L = %v\n", s, t, val) } } @@ -166,7 +166,7 @@ func importfunc(ipkg *types.Pkg, pos src.XPos, s *types.Sym, t *types.Type) { n.Func = new(Func) t.SetNname(asTypesNode(n)) - if Debug['E'] != 0 { + if Debug.E != 0 { fmt.Printf("import func %v%S\n", s, t) } } @@ -179,7 +179,7 @@ func importvar(ipkg *types.Pkg, pos src.XPos, s *types.Sym, t *types.Type) { return } - if Debug['E'] != 0 { + if Debug.E != 0 { fmt.Printf("import var %v %L\n", s, t) } } @@ -192,7 +192,7 @@ func importalias(ipkg *types.Pkg, pos src.XPos, s *types.Sym, t *types.Type) { return } - if Debug['E'] != 0 { + if Debug.E != 0 { fmt.Printf("import type %v = %L\n", s, t) } } diff --git a/src/cmd/compile/internal/gc/float_test.go b/src/cmd/compile/internal/gc/float_test.go index 6ae363be229..c619d257050 100644 --- a/src/cmd/compile/internal/gc/float_test.go +++ b/src/cmd/compile/internal/gc/float_test.go @@ -6,17 +6,9 @@ package gc import ( "math" - "os" - "runtime" "testing" ) -// For GO386=387, make sure fucomi* opcodes are not used -// for comparison operations. -// Note that this test will fail only on a Pentium MMX -// processor (with GOARCH=386 GO386=387), as it just runs -// some code and looks for an unimplemented instruction fault. - //go:noinline func compare1(a, b float64) bool { return a < b @@ -137,9 +129,6 @@ func TestFloatCompareFolded(t *testing.T) { } } -// For GO386=387, make sure fucomi* opcodes are not used -// for float->int conversions. - //go:noinline func cvt1(a float64) uint64 { return uint64(a) @@ -370,14 +359,6 @@ func TestFloat32StoreToLoadConstantFold(t *testing.T) { // are not converted to quiet NaN (qNaN) values during compilation. // See issue #27193 for more information. - // TODO: this method for detecting 387 won't work if the compiler has been - // built using GOARCH=386 GO386=387 and either the target is a different - // architecture or the GO386=387 environment variable is not set when the - // test is run. - if runtime.GOARCH == "386" && os.Getenv("GO386") == "387" { - t.Skip("signaling NaNs are not propagated on 387 (issue #27516)") - } - // signaling NaNs { const nan = uint32(0x7f800001) // sNaN diff --git a/src/cmd/compile/internal/gc/fmt.go b/src/cmd/compile/internal/gc/fmt.go index 866cd0a7145..f92f5d0e884 100644 --- a/src/cmd/compile/internal/gc/fmt.go +++ b/src/cmd/compile/internal/gc/fmt.go @@ -419,10 +419,19 @@ func (n *Node) format(s fmt.State, verb rune, mode fmtMode) { func (n *Node) jconv(s fmt.State, flag FmtFlag) { c := flag & FmtShort + // Useful to see which nodes in a Node Dump/dumplist are actually identical + if Debug_dumpptrs != 0 { + fmt.Fprintf(s, " p(%p)", n) + } if c == 0 && n.Name != nil && n.Name.Vargen != 0 { fmt.Fprintf(s, " g(%d)", n.Name.Vargen) } + if Debug_dumpptrs != 0 && c == 0 && n.Name != nil && n.Name.Defn != nil { + // Useful to see where Defn is set and what node it points to + fmt.Fprintf(s, " defn(%p)", n.Name.Defn) + } + if n.Pos.IsKnown() { pfx := "" switch n.Pos.IsStmt() { @@ -492,6 +501,15 @@ func (n *Node) jconv(s fmt.State, flag FmtFlag) { if n.Name.Assigned() { fmt.Fprint(s, " assigned") } + if n.Name.IsClosureVar() { + fmt.Fprint(s, " closurevar") + } + if n.Name.Captured() { + fmt.Fprint(s, " captured") + } + if n.Name.IsOutputParamHeapAddr() { + fmt.Fprint(s, " outputparamheapaddr") + } } if n.Bounded() { fmt.Fprint(s, " bounded") @@ -711,6 +729,17 @@ func tconv2(b *bytes.Buffer, t *types.Type, flag FmtFlag, mode fmtMode, visited return } + if t.Etype == types.TRESULTS { + tys := t.Extra.(*types.Results).Types + for i, et := range tys { + if i > 0 { + b.WriteByte(',') + } + b.WriteString(et.String()) + } + return + } + flag, mode = flag.update(mode) if mode == FTypeIdName { flag |= FmtUnsigned @@ -762,17 +791,17 @@ func tconv2(b *bytes.Buffer, t *types.Type, flag FmtFlag, mode fmtMode, visited if int(t.Etype) < len(basicnames) && basicnames[t.Etype] != "" { var name string switch t { - case types.Idealbool: + case types.UntypedBool: name = "untyped bool" - case types.Idealstring: + case types.UntypedString: name = "untyped string" - case types.Idealint: + case types.UntypedInt: name = "untyped int" - case types.Idealrune: + case types.UntypedRune: name = "untyped rune" - case types.Idealfloat: + case types.UntypedFloat: name = "untyped float" - case types.Idealcomplex: + case types.UntypedComplex: name = "untyped complex" default: name = basicnames[t.Etype] @@ -781,6 +810,13 @@ func tconv2(b *bytes.Buffer, t *types.Type, flag FmtFlag, mode fmtMode, visited return } + if mode == FDbg { + b.WriteString(t.Etype.String()) + b.WriteByte('-') + tconv2(b, t, flag, FErr, visited) + return + } + // At this point, we might call tconv2 recursively. Add the current type to the visited list so we don't // try to print it recursively. // We record the offset in the result buffer where the type's text starts. This offset serves as a reference @@ -794,12 +830,6 @@ func tconv2(b *bytes.Buffer, t *types.Type, flag FmtFlag, mode fmtMode, visited visited[t] = b.Len() defer delete(visited, t) - if mode == FDbg { - b.WriteString(t.Etype.String()) - b.WriteByte('-') - tconv2(b, t, flag, FErr, visited) - return - } switch t.Etype { case TPTR: b.WriteByte('*') @@ -1322,7 +1352,7 @@ func (n *Node) exprfmt(s fmt.State, prec int, mode fmtMode) { n.Orig.exprfmt(s, prec, mode) return } - if n.Type != nil && n.Type.Etype != TIDEAL && n.Type.Etype != TNIL && n.Type != types.Idealbool && n.Type != types.Idealstring { + if n.Type != nil && !n.Type.IsUntyped() { // Need parens when type begins with what might // be misinterpreted as a unary operator: * or <-. if n.Type.IsPtr() || (n.Type.IsChan() && n.Type.ChanDir() == types.Crecv) { @@ -1407,7 +1437,7 @@ func (n *Node) exprfmt(s fmt.State, prec int, mode fmtMode) { return } if n.Right != nil { - mode.Fprintf(s, "%v literal", n.Right) + mode.Fprintf(s, "%v{%s}", n.Right, ellipsisIf(n.List.Len() != 0)) return } @@ -1421,7 +1451,7 @@ func (n *Node) exprfmt(s fmt.State, prec int, mode fmtMode) { case OSTRUCTLIT, OARRAYLIT, OSLICELIT, OMAPLIT: if mode == FErr { - mode.Fprintf(s, "%v literal", n.Type) + mode.Fprintf(s, "%v{%s}", n.Type, ellipsisIf(n.List.Len() != 0)) return } mode.Fprintf(s, "(%v{ %.v })", n.Type, n.List) @@ -1698,6 +1728,9 @@ func (n *Node) nodedump(s fmt.State, flag FmtFlag, mode fmtMode) { } } + if n.Op == OCLOSURE && n.Func.Closure != nil && n.Func.Closure.Func.Nname.Sym != nil { + mode.Fprintf(s, " fnName %v", n.Func.Closure.Func.Nname.Sym) + } if n.Sym != nil && n.Op != ONAME { mode.Fprintf(s, " %v", n.Sym) } @@ -1713,6 +1746,16 @@ func (n *Node) nodedump(s fmt.State, flag FmtFlag, mode fmtMode) { if n.Right != nil { mode.Fprintf(s, "%v", n.Right) } + if n.Func != nil && n.Func.Closure != nil && n.Func.Closure.Nbody.Len() != 0 { + indent(s) + // The function associated with a closure + mode.Fprintf(s, "%v-clofunc%v", n.Op, n.Func.Closure) + } + if n.Func != nil && n.Func.Dcl != nil && len(n.Func.Dcl) != 0 { + indent(s) + // The dcls for a func or closure + mode.Fprintf(s, "%v-dcl%v", n.Op, asNodes(n.Func.Dcl)) + } if n.List.Len() != 0 { indent(s) mode.Fprintf(s, "%v-list%v", n.Op, n.List) @@ -1934,3 +1977,10 @@ func indent(s fmt.State) { fmt.Fprint(s, ". ") } } + +func ellipsisIf(b bool) string { + if b { + return "..." + } + return "" +} diff --git a/src/cmd/compile/internal/gc/go.go b/src/cmd/compile/internal/gc/go.go index 9079ce2afc4..274930bd156 100644 --- a/src/cmd/compile/internal/gc/go.go +++ b/src/cmd/compile/internal/gc/go.go @@ -61,12 +61,12 @@ type Class uint8 //go:generate stringer -type=Class const ( Pxxx Class = iota // no class; used during ssa conversion to indicate pseudo-variables - PEXTERN // global variable + PEXTERN // global variables PAUTO // local variables - PAUTOHEAP // local variable or parameter moved to heap + PAUTOHEAP // local variables or parameters moved to heap PPARAM // input arguments PPARAMOUT // output results - PFUNC // global function + PFUNC // global functions // Careful: Class is stored in three bits in Node.flags. _ = uint((1 << 3) - iota) // static assert for iota <= (1 << 3) @@ -116,7 +116,15 @@ var decldepth int32 var nolocalimports bool -var Debug [256]int +// gc debug flags +type DebugFlags struct { + P, B, C, E, + K, L, N, S, + W, e, h, j, + l, m, r, w int +} + +var Debug DebugFlags var debugstr string @@ -259,7 +267,6 @@ type Arch struct { REGSP int MAXWIDTH int64 - Use387 bool // should 386 backend use 387 FP instructions instead of sse2. SoftFloat bool PadFrame func(int64) int64 @@ -302,6 +309,7 @@ var ( growslice, msanread, msanwrite, + msanmove, newobject, newproc, panicdivide, @@ -328,10 +336,6 @@ var ( BoundsCheckFunc [ssa.BoundsKindCount]*obj.LSym ExtendCheckFunc [ssa.BoundsKindCount]*obj.LSym - // GO386=387 - ControlWord64trunc, - ControlWord32 *obj.LSym - // Wasm WasmMove, WasmZero, diff --git a/src/cmd/compile/internal/gc/gsubr.go b/src/cmd/compile/internal/gc/gsubr.go index 480d411f495..d599a383e7b 100644 --- a/src/cmd/compile/internal/gc/gsubr.go +++ b/src/cmd/compile/internal/gc/gsubr.go @@ -70,12 +70,8 @@ func newProgs(fn *Node, worker int) *Progs { pp.pos = fn.Pos pp.settext(fn) // PCDATA tables implicitly start with index -1. - pp.prevLive = LivenessIndex{-1, -1, false} - if go115ReduceLiveness { - pp.nextLive = pp.prevLive - } else { - pp.nextLive = LivenessInvalid - } + pp.prevLive = LivenessIndex{-1, false} + pp.nextLive = pp.prevLive return pp } @@ -120,31 +116,15 @@ func (pp *Progs) Prog(as obj.As) *obj.Prog { Addrconst(&p.From, objabi.PCDATA_StackMapIndex) Addrconst(&p.To, int64(idx)) } - if !go115ReduceLiveness { + if pp.nextLive.isUnsafePoint != pp.prevLive.isUnsafePoint { + // Emit unsafe-point marker. + pp.prevLive.isUnsafePoint = pp.nextLive.isUnsafePoint + p := pp.Prog(obj.APCDATA) + Addrconst(&p.From, objabi.PCDATA_UnsafePoint) if pp.nextLive.isUnsafePoint { - // Unsafe points are encoded as a special value in the - // register map. - pp.nextLive.regMapIndex = objabi.PCDATA_RegMapUnsafe - } - if pp.nextLive.regMapIndex != pp.prevLive.regMapIndex { - // Emit register map index change. - idx := pp.nextLive.regMapIndex - pp.prevLive.regMapIndex = idx - p := pp.Prog(obj.APCDATA) - Addrconst(&p.From, objabi.PCDATA_RegMapIndex) - Addrconst(&p.To, int64(idx)) - } - } else { - if pp.nextLive.isUnsafePoint != pp.prevLive.isUnsafePoint { - // Emit unsafe-point marker. - pp.prevLive.isUnsafePoint = pp.nextLive.isUnsafePoint - p := pp.Prog(obj.APCDATA) - Addrconst(&p.From, objabi.PCDATA_UnsafePoint) - if pp.nextLive.isUnsafePoint { - Addrconst(&p.To, objabi.PCDATA_UnsafePointUnsafe) - } else { - Addrconst(&p.To, objabi.PCDATA_UnsafePointSafe) - } + Addrconst(&p.To, objabi.PCDATA_UnsafePointUnsafe) + } else { + Addrconst(&p.To, objabi.PCDATA_UnsafePointSafe) } } @@ -153,7 +133,7 @@ func (pp *Progs) Prog(as obj.As) *obj.Prog { pp.clearp(pp.next) p.Link = pp.next - if !pp.pos.IsKnown() && Debug['K'] != 0 { + if !pp.pos.IsKnown() && Debug.K != 0 { Warn("prog: unknown position (line 0)") } @@ -199,7 +179,7 @@ func (pp *Progs) settext(fn *Node) { ptxt := pp.Prog(obj.ATEXT) pp.Text = ptxt - fn.Func.lsym.Func.Text = ptxt + fn.Func.lsym.Func().Text = ptxt ptxt.From.Type = obj.TYPE_MEM ptxt.From.Name = obj.NAME_EXTERN ptxt.From.Sym = fn.Func.lsym @@ -322,6 +302,12 @@ func ggloblnod(nam *Node) { if nam.Name.LibfuzzerExtraCounter() { s.Type = objabi.SLIBFUZZER_EXTRA_COUNTER } + if nam.Sym.Linkname != "" { + // Make sure linkname'd symbol is non-package. When a symbol is + // both imported and linkname'd, s.Pkg may not set to "_" in + // types.Sym.Linksym because LSym already exists. Set it here. + s.Pkg = "_" + } } func ggloblsym(s *obj.LSym, width int32, flags int16) { diff --git a/src/cmd/compile/internal/gc/iexport.go b/src/cmd/compile/internal/gc/iexport.go index b3f50b63af5..1f53d8ca7dc 100644 --- a/src/cmd/compile/internal/gc/iexport.go +++ b/src/cmd/compile/internal/gc/iexport.go @@ -751,11 +751,11 @@ func (w *exportWriter) param(f *types.Field) { func constTypeOf(typ *types.Type) Ctype { switch typ { - case types.Idealint, types.Idealrune: + case types.UntypedInt, types.UntypedRune: return CTINT - case types.Idealfloat: + case types.UntypedFloat: return CTFLT - case types.Idealcomplex: + case types.UntypedComplex: return CTCPLX } @@ -780,8 +780,8 @@ func constTypeOf(typ *types.Type) Ctype { } func (w *exportWriter) value(typ *types.Type, v Val) { - if typ.IsUntyped() { - typ = untype(v.Ctype()) + if vt := idealType(v.Ctype()); typ.IsUntyped() && typ != vt { + Fatalf("exporter: untyped type mismatch, have: %v, want: %v", typ, vt) } w.typ(typ) @@ -1017,6 +1017,8 @@ func (w *exportWriter) symIdx(s *types.Sym) { } func (w *exportWriter) typeExt(t *types.Type) { + // Export whether this type is marked notinheap. + w.bool(t.NotInHeap()) // For type T, export the index of type descriptor symbols of T and *T. if i, ok := typeSymIdx[t]; ok { w.int64(i[0]) @@ -1136,13 +1138,10 @@ func (w *exportWriter) stmt(n *Node) { w.pos(n.Pos) w.stmtList(n.Ninit) w.exprsOrNil(n.Left, nil) - w.stmtList(n.List) + w.caseList(n) - case OCASE: - w.op(OCASE) - w.pos(n.Pos) - w.stmtList(n.List) - w.stmtList(n.Nbody) + // case OCASE: + // handled by caseList case OFALL: w.op(OFALL) @@ -1166,6 +1165,24 @@ func (w *exportWriter) stmt(n *Node) { } } +func (w *exportWriter) caseList(sw *Node) { + namedTypeSwitch := sw.Op == OSWITCH && sw.Left != nil && sw.Left.Op == OTYPESW && sw.Left.Left != nil + + cases := sw.List.Slice() + w.uint64(uint64(len(cases))) + for _, cas := range cases { + if cas.Op != OCASE { + Fatalf("expected OCASE, got %v", cas) + } + w.pos(cas.Pos) + w.stmtList(cas.List) + if namedTypeSwitch { + w.localName(cas.Rlist.First()) + } + w.stmtList(cas.Nbody) + } +} + func (w *exportWriter) exprList(list Nodes) { for _, n := range list.Slice() { w.expr(n) @@ -1230,6 +1247,19 @@ func (w *exportWriter) expr(n *Node) { w.op(OTYPE) w.typ(n.Type) + case OTYPESW: + w.op(OTYPESW) + w.pos(n.Pos) + var s *types.Sym + if n.Left != nil { + if n.Left.Op != ONONAME { + Fatalf("expected ONONAME, got %v", n.Left) + } + s = n.Left.Sym + } + w.localIdent(s, 0) // declared pseudo-variable, if any + w.exprsOrNil(n.Right, nil) + // case OTARRAY, OTMAP, OTCHAN, OTSTRUCT, OTINTER, OTFUNC: // should have been resolved by typechecking - handled by default case @@ -1264,8 +1294,13 @@ func (w *exportWriter) expr(n *Node) { // case OSTRUCTKEY: // unreachable - handled in case OSTRUCTLIT by elemList - // case OCALLPART: - // unimplemented - handled by default case + case OCALLPART: + // An OCALLPART is an OXDOT before type checking. + w.op(OXDOT) + w.pos(n.Pos) + w.expr(n.Left) + // Right node should be ONAME + w.selector(n.Right.Sym) case OXDOT, ODOT, ODOTPTR, ODOTINTER, ODOTMETH: w.op(OXDOT) diff --git a/src/cmd/compile/internal/gc/iimport.go b/src/cmd/compile/internal/gc/iimport.go index 4169222c14f..c0114d0e53b 100644 --- a/src/cmd/compile/internal/gc/iimport.go +++ b/src/cmd/compile/internal/gc/iimport.go @@ -375,7 +375,7 @@ func (p *importReader) value() (typ *types.Type, v Val) { v.U = p.string() case CTINT: x := new(Mpint) - x.Rune = typ == types.Idealrune + x.Rune = typ == types.UntypedRune p.mpint(&x.Val, typ) v.U = x case CTFLT: @@ -596,7 +596,6 @@ func (r *importReader) typ1() *types.Type { // Ensure we expand the interface in the frontend (#25055). checkwidth(t) - return t } } @@ -711,6 +710,7 @@ func (r *importReader) symIdx(s *types.Sym) { } func (r *importReader) typeExt(t *types.Type) { + t.SetNotInHeap(r.bool()) i, pi := r.int64(), r.int64() if i != -1 && pi != -1 { typeSymIdx[t] = [2]int64{i, pi} @@ -742,8 +742,8 @@ func (r *importReader) doInline(n *Node) { importlist = append(importlist, n) - if Debug['E'] > 0 && Debug['m'] > 2 { - if Debug['m'] > 3 { + if Debug.E > 0 && Debug.m > 2 { + if Debug.m > 3 { fmt.Printf("inl body for %v %#v: %+v\n", n, n.Type, asNodes(n.Func.Inl.Body)) } else { fmt.Printf("inl body for %v %#v: %v\n", n, n.Type, asNodes(n.Func.Inl.Body)) @@ -784,6 +784,28 @@ func (r *importReader) stmtList() []*Node { return list } +func (r *importReader) caseList(sw *Node) []*Node { + namedTypeSwitch := sw.Op == OSWITCH && sw.Left != nil && sw.Left.Op == OTYPESW && sw.Left.Left != nil + + cases := make([]*Node, r.uint64()) + for i := range cases { + cas := nodl(r.pos(), OCASE, nil, nil) + cas.List.Set(r.stmtList()) + if namedTypeSwitch { + // Note: per-case variables will have distinct, dotted + // names after import. That's okay: swt.go only needs + // Sym for diagnostics anyway. + caseVar := newnamel(cas.Pos, r.ident()) + declare(caseVar, dclcontext) + cas.Rlist.Set1(caseVar) + caseVar.Name.Defn = sw.Left + } + cas.Nbody.Set(r.stmtList()) + cases[i] = cas + } + return cases +} + func (r *importReader) exprList() []*Node { var list []*Node for { @@ -831,6 +853,14 @@ func (r *importReader) node() *Node { case OTYPE: return typenod(r.typ()) + case OTYPESW: + n := nodl(r.pos(), OTYPESW, nil, nil) + if s := r.ident(); s != nil { + n.Left = npos(n.Pos, newnoname(s)) + } + n.Right, _ = r.exprsOrNil() + return n + // case OTARRAY, OTMAP, OTCHAN, OTSTRUCT, OTINTER, OTFUNC: // unreachable - should have been resolved by typechecking @@ -866,7 +896,7 @@ func (r *importReader) node() *Node { // unreachable - handled in case OSTRUCTLIT by elemList // case OCALLPART: - // unimplemented + // unreachable - mapped to case OXDOT below by exporter // case OXDOT, ODOT, ODOTPTR, ODOTINTER, ODOTMETH: // unreachable - mapped to case OXDOT below by exporter @@ -1025,16 +1055,11 @@ func (r *importReader) node() *Node { n := nodl(r.pos(), op, nil, nil) n.Ninit.Set(r.stmtList()) n.Left, _ = r.exprsOrNil() - n.List.Set(r.stmtList()) + n.List.Set(r.caseList(n)) return n - case OCASE: - n := nodl(r.pos(), OCASE, nil, nil) - n.List.Set(r.exprList()) - // TODO(gri) eventually we must declare variables for type switch - // statements (type switch statements are not yet exported) - n.Nbody.Set(r.stmtList()) - return n + // case OCASE: + // handled by caseList case OFALL: n := nodl(r.pos(), OFALL, nil, nil) diff --git a/src/cmd/compile/internal/gc/init.go b/src/cmd/compile/internal/gc/init.go index 94cbcf98467..ec9cc4bddc5 100644 --- a/src/cmd/compile/internal/gc/init.go +++ b/src/cmd/compile/internal/gc/init.go @@ -59,7 +59,7 @@ func fninit(n []*Node) { Curfn = fn typecheckslice(nf, ctxStmt) Curfn = nil - funccompile(fn) + xtop = append(xtop, fn) fns = append(fns, initializers.Linksym()) } if dummyInitFn.Func.Dcl != nil { @@ -68,16 +68,14 @@ func fninit(n []*Node) { // something's weird if we get here. Fatalf("dummyInitFn still has declarations") } + dummyInitFn = nil // Record user init functions. for i := 0; i < renameinitgen; i++ { s := lookupN("init.", i) fn := asNode(s.Def).Name.Defn // Skip init functions with empty bodies. - // noder.go doesn't allow external init functions, and - // order.go has already removed any OEMPTY nodes, so - // checking Len() == 0 is sufficient here. - if fn.Nbody.Len() == 0 { + if fn.Nbody.Len() == 1 && fn.Nbody.First().Op == OEMPTY { continue } fns = append(fns, s.Linksym()) diff --git a/src/cmd/compile/internal/gc/inl.go b/src/cmd/compile/internal/gc/inl.go index fa5b3ec698b..419056985fc 100644 --- a/src/cmd/compile/internal/gc/inl.go +++ b/src/cmd/compile/internal/gc/inl.go @@ -7,7 +7,7 @@ // saves a copy of the body. Then inlcalls walks each function body to // expand calls to inlinable functions. // -// The debug['l'] flag controls the aggressiveness. Note that main() swaps level 0 and 1, +// The Debug.l flag controls the aggressiveness. Note that main() swaps level 0 and 1, // making 1 the default and -l disable. Additional levels (beyond -l) may be buggy and // are not supported. // 0: disabled @@ -21,7 +21,7 @@ // The -d typcheckinl flag enables early typechecking of all imported bodies, // which is useful to flush out bugs. // -// The debug['m'] flag enables diagnostic output. a single -m is useful for verifying +// The Debug.m flag enables diagnostic output. a single -m is useful for verifying // which calls get inlined or not, more is for debugging, and may go away at any point. package gc @@ -85,7 +85,7 @@ func typecheckinl(fn *Node) { return // typecheckinl on local function } - if Debug['m'] > 2 || Debug_export != 0 { + if Debug.m > 2 || Debug_export != 0 { fmt.Printf("typecheck import [%v] %L { %#v }\n", fn.Sym, fn, asNodes(fn.Func.Inl.Body)) } @@ -94,10 +94,11 @@ func typecheckinl(fn *Node) { typecheckslice(fn.Func.Inl.Body, ctxStmt) Curfn = savefn - // During typechecking, declarations are added to - // Curfn.Func.Dcl. Move them to Inl.Dcl for consistency with - // how local functions behave. (Append because typecheckinl - // may be called multiple times.) + // During expandInline (which imports fn.Func.Inl.Body), + // declarations are added to fn.Func.Dcl by funcHdr(). Move them + // to fn.Func.Inl.Dcl for consistency with how local functions + // behave. (Append because typecheckinl may be called multiple + // times.) fn.Func.Inl.Dcl = append(fn.Func.Inl.Dcl, fn.Func.Dcl...) fn.Func.Dcl = nil @@ -116,10 +117,10 @@ func caninl(fn *Node) { } var reason string // reason, if any, that the function was not inlined - if Debug['m'] > 1 || logopt.Enabled() { + if Debug.m > 1 || logopt.Enabled() { defer func() { if reason != "" { - if Debug['m'] > 1 { + if Debug.m > 1 { fmt.Printf("%v: cannot inline %v: %s\n", fn.Line(), fn.Func.Nname, reason) } if logopt.Enabled() { @@ -187,7 +188,7 @@ func caninl(fn *Node) { defer n.Func.SetInlinabilityChecked(true) cc := int32(inlineExtraCallCost) - if Debug['l'] == 4 { + if Debug.l == 4 { cc = 1 // this appears to yield better performance than 0. } @@ -224,9 +225,9 @@ func caninl(fn *Node) { // this is so export can find the body of a method fn.Type.FuncType().Nname = asTypesNode(n) - if Debug['m'] > 1 { + if Debug.m > 1 { fmt.Printf("%v: can inline %#v with cost %d as: %#v { %#v }\n", fn.Line(), n, inlineMaxBudget-visitor.budget, fn.Type, asNodes(n.Func.Inl.Body)) - } else if Debug['m'] != 0 { + } else if Debug.m != 0 { fmt.Printf("%v: can inline %v\n", fn.Line(), n) } if logopt.Enabled() { @@ -257,21 +258,39 @@ func inlFlood(n *Node) { typecheckinl(n) + // Recursively identify all referenced functions for + // reexport. We want to include even non-called functions, + // because after inlining they might be callable. inspectList(asNodes(n.Func.Inl.Body), func(n *Node) bool { switch n.Op { case ONAME: - // Mark any referenced global variables or - // functions for reexport. Skip methods, - // because they're reexported alongside their - // receiver type. - if n.Class() == PEXTERN || n.Class() == PFUNC && !n.isMethodExpression() { + switch n.Class() { + case PFUNC: + if n.isMethodExpression() { + inlFlood(asNode(n.Type.Nname())) + } else { + inlFlood(n) + exportsym(n) + } + case PEXTERN: exportsym(n) } - case OCALLFUNC, OCALLMETH: - // Recursively flood any functions called by - // this one. - inlFlood(asNode(n.Left.Type.Nname())) + case ODOTMETH: + fn := asNode(n.Type.Nname()) + inlFlood(fn) + + case OCALLPART: + // Okay, because we don't yet inline indirect + // calls to method values. + case OCLOSURE: + // If the closure is inlinable, we'll need to + // flood it too. But today we don't support + // inlining functions that contain closures. + // + // When we do, we'll probably want: + // inlFlood(n.Func.Closure.Func.Nname) + Fatalf("unexpected closure in inlinable function") } return true }) @@ -325,18 +344,10 @@ func (v *hairyVisitor) visit(n *Node) bool { break } - if fn := n.Left.Func; fn != nil && fn.Inl != nil { - v.budget -= fn.Inl.Cost + if fn := inlCallee(n.Left); fn != nil && fn.Func.Inl != nil { + v.budget -= fn.Func.Inl.Cost break } - if n.Left.isMethodExpression() { - if d := asNode(n.Left.Sym.Def); d != nil && d.Func.Inl != nil { - v.budget -= d.Func.Inl.Cost - break - } - } - // TODO(mdempsky): Budget for OCLOSURE calls if we - // ever allow that. See #15561 and #23093. // Call cost for non-leaf inlining. v.budget -= v.extraCallCost @@ -383,16 +394,11 @@ func (v *hairyVisitor) visit(n *Node) bool { return true case OCLOSURE, - OCALLPART, ORANGE, - OFOR, - OFORUNTIL, OSELECT, - OTYPESW, OGO, ODEFER, ODCLTYPE, // can't print yet - OBREAK, ORETJMP: v.reason = "unhandled op " + n.Op.String() return true @@ -400,10 +406,23 @@ func (v *hairyVisitor) visit(n *Node) bool { case OAPPEND: v.budget -= inlineExtraAppendCost - case ODCLCONST, OEMPTY, OFALL, OLABEL: + case ODCLCONST, OEMPTY, OFALL: // These nodes don't produce code; omit from inlining budget. return false + case OLABEL: + // TODO(mdempsky): Add support for inlining labeled control statements. + if n.labeledControl() != nil { + v.reason = "labeled control" + return true + } + + case OBREAK, OCONTINUE: + if n.Sym != nil { + // Should have short-circuited due to labeledControl above. + Fatalf("unexpected labeled break/continue: %v", n) + } + case OIF: if Isconst(n.Left, CTBOOL) { // This if and the condition cost nothing. @@ -421,7 +440,7 @@ func (v *hairyVisitor) visit(n *Node) bool { v.budget-- // When debugging, don't stop early, to get full cost of inlining this function - if v.budget < 0 && Debug['m'] < 2 && !logopt.Enabled() { + if v.budget < 0 && Debug.m < 2 && !logopt.Enabled() { return true } @@ -430,9 +449,9 @@ func (v *hairyVisitor) visit(n *Node) bool { v.visitList(n.Ninit) || v.visitList(n.Nbody) } -// Inlcopy and inlcopylist recursively copy the body of a function. -// Any name-like node of non-local class is marked for re-export by adding it to -// the exportlist. +// inlcopylist (together with inlcopy) recursively copies a list of nodes, except +// that it keeps the same ONAME, OTYPE, and OLITERAL nodes. It is used for copying +// the body and dcls of an inlineable function. func inlcopylist(ll []*Node) []*Node { s := make([]*Node, 0, len(ll)) for _, n := range ll { @@ -452,7 +471,7 @@ func inlcopy(n *Node) *Node { } m := n.copy() - if m.Func != nil { + if n.Op != OCALLPART && m.Func != nil { Fatalf("unexpected Func: %v", m) } m.Left = inlcopy(n.Left) @@ -570,13 +589,11 @@ func inlnode(n *Node, maxCost int32, inlMap map[*Node]bool) *Node { } switch n.Op { - // inhibit inlining of their argument case ODEFER, OGO: switch n.Left.Op { case OCALLFUNC, OCALLMETH: n.Left.SetNoInline(true) } - return n // TODO do them here (or earlier), // so escape analysis can avoid more heapmoves. @@ -666,60 +683,18 @@ func inlnode(n *Node, maxCost int32, inlMap map[*Node]bool) *Node { switch n.Op { case OCALLFUNC: - if Debug['m'] > 3 { + if Debug.m > 3 { fmt.Printf("%v:call to func %+v\n", n.Line(), n.Left) } - if n.Left.Func != nil && n.Left.Func.Inl != nil && !isIntrinsicCall(n) { // normal case - n = mkinlcall(n, n.Left, maxCost, inlMap) - } else if n.Left.isMethodExpression() && asNode(n.Left.Sym.Def) != nil { - n = mkinlcall(n, asNode(n.Left.Sym.Def), maxCost, inlMap) - } else if n.Left.Op == OCLOSURE { - if f := inlinableClosure(n.Left); f != nil { - n = mkinlcall(n, f, maxCost, inlMap) - } - } else if n.Left.Op == ONAME && n.Left.Name != nil && n.Left.Name.Defn != nil { - if d := n.Left.Name.Defn; d.Op == OAS && d.Right.Op == OCLOSURE { - if f := inlinableClosure(d.Right); f != nil { - // NB: this check is necessary to prevent indirect re-assignment of the variable - // having the address taken after the invocation or only used for reads is actually fine - // but we have no easy way to distinguish the safe cases - if d.Left.Name.Addrtaken() { - if Debug['m'] > 1 { - fmt.Printf("%v: cannot inline escaping closure variable %v\n", n.Line(), n.Left) - } - if logopt.Enabled() { - logopt.LogOpt(n.Pos, "cannotInlineCall", "inline", Curfn.funcname(), - fmt.Sprintf("%v cannot be inlined (escaping closure variable)", n.Left)) - } - break - } - - // ensure the variable is never re-assigned - if unsafe, a := reassigned(n.Left); unsafe { - if Debug['m'] > 1 { - if a != nil { - fmt.Printf("%v: cannot inline re-assigned closure variable at %v: %v\n", n.Line(), a.Line(), a) - if logopt.Enabled() { - logopt.LogOpt(n.Pos, "cannotInlineCall", "inline", Curfn.funcname(), - fmt.Sprintf("%v cannot be inlined (re-assigned closure variable)", a)) - } - } else { - fmt.Printf("%v: cannot inline global closure variable %v\n", n.Line(), n.Left) - if logopt.Enabled() { - logopt.LogOpt(n.Pos, "cannotInlineCall", "inline", Curfn.funcname(), - fmt.Sprintf("%v cannot be inlined (global closure variable)", n.Left)) - } - } - } - break - } - n = mkinlcall(n, f, maxCost, inlMap) - } - } + if isIntrinsicCall(n) { + break + } + if fn := inlCallee(n.Left); fn != nil && fn.Func.Inl != nil { + n = mkinlcall(n, fn, maxCost, inlMap) } case OCALLMETH: - if Debug['m'] > 3 { + if Debug.m > 3 { fmt.Printf("%v:call to meth %L\n", n.Line(), n.Left.Right) } @@ -739,16 +714,85 @@ func inlnode(n *Node, maxCost int32, inlMap map[*Node]bool) *Node { return n } -// inlinableClosure takes an OCLOSURE node and follows linkage to the matching ONAME with -// the inlinable body. Returns nil if the function is not inlinable. -func inlinableClosure(n *Node) *Node { - c := n.Func.Closure - caninl(c) - f := c.Func.Nname - if f == nil || f.Func.Inl == nil { +// inlCallee takes a function-typed expression and returns the underlying function ONAME +// that it refers to if statically known. Otherwise, it returns nil. +func inlCallee(fn *Node) *Node { + fn = staticValue(fn) + switch { + case fn.Op == ONAME && fn.Class() == PFUNC: + if fn.isMethodExpression() { + n := asNode(fn.Type.Nname()) + // Check that receiver type matches fn.Left. + // TODO(mdempsky): Handle implicit dereference + // of pointer receiver argument? + if n == nil || !types.Identical(n.Type.Recv().Type, fn.Left.Type) { + return nil + } + return n + } + return fn + case fn.Op == OCLOSURE: + c := fn.Func.Closure + caninl(c) + return c.Func.Nname + } + return nil +} + +func staticValue(n *Node) *Node { + for { + if n.Op == OCONVNOP { + n = n.Left + continue + } + + n1 := staticValue1(n) + if n1 == nil { + return n + } + n = n1 + } +} + +// staticValue1 implements a simple SSA-like optimization. If n is a local variable +// that is initialized and never reassigned, staticValue1 returns the initializer +// expression. Otherwise, it returns nil. +func staticValue1(n *Node) *Node { + if n.Op != ONAME || n.Class() != PAUTO || n.Name.Addrtaken() { return nil } - return f + + defn := n.Name.Defn + if defn == nil { + return nil + } + + var rhs *Node +FindRHS: + switch defn.Op { + case OAS: + rhs = defn.Right + case OAS2: + for i, lhs := range defn.List.Slice() { + if lhs == n { + rhs = defn.Rlist.Index(i) + break FindRHS + } + } + Fatalf("%v missing from LHS of %v", n, defn) + default: + return nil + } + if rhs == nil { + Fatalf("RHS is nil: %v", defn) + } + + unsafe, _ := reassigned(n) + if unsafe { + return nil + } + + return rhs } // reassigned takes an ONAME node, walks the function in which it is defined, and returns a boolean @@ -792,14 +836,12 @@ func (v *reassignVisitor) visit(n *Node) *Node { if n.Left == v.name && n != v.name.Name.Defn { return n } - return nil case OAS2, OAS2FUNC, OAS2MAPR, OAS2DOTTYPE: for _, p := range n.List.Slice() { if p == v.name && n != v.name.Name.Defn { return n } } - return nil } if a := v.visit(n.Left); a != nil { return a @@ -831,24 +873,27 @@ func (v *reassignVisitor) visitList(l Nodes) *Node { return nil } -func tinlvar(t *types.Field, inlvars map[*Node]*Node) *Node { - if n := asNode(t.Nname); n != nil && !n.isBlank() { - inlvar := inlvars[n] - if inlvar == nil { - Fatalf("missing inlvar for %v\n", n) - } - return inlvar +func inlParam(t *types.Field, as *Node, inlvars map[*Node]*Node) *Node { + n := asNode(t.Nname) + if n == nil || n.isBlank() { + return nblank } - return typecheck(nblank, ctxExpr|ctxAssign) + inlvar := inlvars[n] + if inlvar == nil { + Fatalf("missing inlvar for %v", n) + } + as.Ninit.Append(nod(ODCL, inlvar, nil)) + inlvar.Name.Defn = as + return inlvar } var inlgen int -// If n is a call, and fn is a function with an inlinable body, -// return an OINLCALL. -// On return ninit has the parameter assignments, the nbody is the -// inlined function body and list, rlist contain the input, output +// If n is a call node (OCALLFUNC or OCALLMETH), and fn is an ONAME node for a +// function with an inlinable body, return an OINLCALL node that can replace n. +// The returned node's Ninit has the parameter assignments, the Nbody is the +// inlined function body, and (List, Rlist) contain the (input, output) // parameters. // The result of mkinlcall MUST be assigned back to n, e.g. // n.Left = mkinlcall(n.Left, fn, isddd) @@ -889,7 +934,7 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { } if inlMap[fn] { - if Debug['m'] > 1 { + if Debug.m > 1 { fmt.Printf("%v: cannot inline %v into %v: repeated recursive cycle\n", n.Line(), fn, Curfn.funcname()) } return n @@ -903,12 +948,12 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { } // We have a function node, and it has an inlineable body. - if Debug['m'] > 1 { + if Debug.m > 1 { fmt.Printf("%v: inlining call to %v %#v { %#v }\n", n.Line(), fn.Sym, fn.Type, asNodes(fn.Func.Inl.Body)) - } else if Debug['m'] != 0 { + } else if Debug.m != 0 { fmt.Printf("%v: inlining call to %v\n", n.Line(), fn) } - if Debug['m'] > 2 { + if Debug.m > 2 { fmt.Printf("%v: Before inlining: %+v\n", n.Line(), n) } @@ -918,6 +963,21 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { ninit := n.Ninit + // For normal function calls, the function callee expression + // may contain side effects (e.g., added by addinit during + // inlconv2expr or inlconv2list). Make sure to preserve these, + // if necessary (#42703). + if n.Op == OCALLFUNC { + callee := n.Left + for callee.Op == OCONVNOP { + ninit.AppendNodes(&callee.Ninit) + callee = callee.Left + } + if callee.Op != ONAME && callee.Op != OCLOSURE { + Fatalf("unexpected callee expression: %v", callee) + } + } + // Make temp names to use instead of the originals. inlvars := make(map[*Node]*Node) @@ -970,14 +1030,15 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { continue } if ln.isParamStackCopy() { // ignore the on-stack copy of a parameter that moved to the heap - continue - } - inlvars[ln] = typecheck(inlvar(ln), ctxExpr) - if ln.Class() == PPARAM || ln.Name.Param.Stackcopy != nil && ln.Name.Param.Stackcopy.Class() == PPARAM { - ninit.Append(nod(ODCL, inlvars[ln], nil)) + // TODO(mdempsky): Remove once I'm confident + // this never actually happens. We currently + // perform inlining before escape analysis, so + // nothing should have moved to the heap yet. + Fatalf("impossible: %v", ln) } + inlf := typecheck(inlvar(ln), ctxExpr) + inlvars[ln] = inlf if genDwarfInline > 0 { - inlf := inlvars[ln] if ln.Class() == PPARAM { inlf.Name.SetInlFormal(true) } else { @@ -988,15 +1049,28 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { } } + nreturns := 0 + inspectList(asNodes(fn.Func.Inl.Body), func(n *Node) bool { + if n != nil && n.Op == ORETURN { + nreturns++ + } + return true + }) + + // We can delay declaring+initializing result parameters if: + // (1) there's only one "return" statement in the inlined + // function, and (2) the result parameters aren't named. + delayretvars := nreturns == 1 + // temporaries for return values. var retvars []*Node for i, t := range fn.Type.Results().Fields().Slice() { var m *Node - mpos := t.Pos - if n := asNode(t.Nname); n != nil && !n.isBlank() { + if n := asNode(t.Nname); n != nil && !n.isBlank() && !strings.HasPrefix(n.Sym.Name, "~r") { m = inlvar(n) m = typecheck(m, ctxExpr) inlvars[n] = m + delayretvars = false // found a named result parameter } else { // anonymous return values, synthesize names for use in assignment that replaces return m = retvar(t, i) @@ -1008,67 +1082,52 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { // were not part of the original callee. if !strings.HasPrefix(m.Sym.Name, "~R") { m.Name.SetInlFormal(true) - m.Pos = mpos + m.Pos = t.Pos inlfvars = append(inlfvars, m) } } - ninit.Append(nod(ODCL, m, nil)) retvars = append(retvars, m) } // Assign arguments to the parameters' temp names. as := nod(OAS2, nil, nil) - as.Rlist.Set(n.List.Slice()) + as.SetColas(true) + if n.Op == OCALLMETH { + if n.Left.Left == nil { + Fatalf("method call without receiver: %+v", n) + } + as.Rlist.Append(n.Left.Left) + } + as.Rlist.Append(n.List.Slice()...) // For non-dotted calls to variadic functions, we assign the // variadic parameter's temp name separately. var vas *Node - if fn.IsMethod() { - rcv := fn.Type.Recv() - - if n.Left.Op == ODOTMETH { - // For x.M(...), assign x directly to the - // receiver parameter. - if n.Left.Left == nil { - Fatalf("method call without receiver: %+v", n) - } - ras := nod(OAS, tinlvar(rcv, inlvars), n.Left.Left) - ras = typecheck(ras, ctxStmt) - ninit.Append(ras) - } else { - // For T.M(...), add the receiver parameter to - // as.List, so it's assigned by the normal - // arguments. - if as.Rlist.Len() == 0 { - Fatalf("non-method call to method without first arg: %+v", n) - } - as.List.Append(tinlvar(rcv, inlvars)) - } + if recv := fn.Type.Recv(); recv != nil { + as.List.Append(inlParam(recv, as, inlvars)) } - for _, param := range fn.Type.Params().Fields().Slice() { // For ordinary parameters or variadic parameters in // dotted calls, just add the variable to the // assignment list, and we're done. if !param.IsDDD() || n.IsDDD() { - as.List.Append(tinlvar(param, inlvars)) + as.List.Append(inlParam(param, as, inlvars)) continue } // Otherwise, we need to collect the remaining values // to pass as a slice. - numvals := n.List.Len() - x := as.List.Len() - for as.List.Len() < numvals { + for as.List.Len() < as.Rlist.Len() { as.List.Append(argvar(param.Type, as.List.Len())) } varargs := as.List.Slice()[x:] - vas = nod(OAS, tinlvar(param, inlvars), nil) + vas = nod(OAS, nil, nil) + vas.Left = inlParam(param, vas, inlvars) if len(varargs) == 0 { vas.Right = nodnil() vas.Right.Type = param.Type @@ -1088,11 +1147,14 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { ninit.Append(vas) } - // Zero the return parameters. - for _, n := range retvars { - ras := nod(OAS, n, nil) - ras = typecheck(ras, ctxStmt) - ninit.Append(ras) + if !delayretvars { + // Zero the return parameters. + for _, n := range retvars { + ninit.Append(nod(ODCL, n, nil)) + ras := nod(OAS, n, nil) + ras = typecheck(ras, ctxStmt) + ninit.Append(ras) + } } retlabel := autolabel(".i") @@ -1123,11 +1185,12 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { } subst := inlsubst{ - retlabel: retlabel, - retvars: retvars, - inlvars: inlvars, - bases: make(map[*src.PosBase]*src.PosBase), - newInlIndex: newIndex, + retlabel: retlabel, + retvars: retvars, + delayretvars: delayretvars, + inlvars: inlvars, + bases: make(map[*src.PosBase]*src.PosBase), + newInlIndex: newIndex, } body := subst.list(asNodes(fn.Func.Inl.Body)) @@ -1165,7 +1228,7 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { } } - if Debug['m'] > 2 { + if Debug.m > 2 { fmt.Printf("%v: After inlining %+v\n\n", call.Line(), call) } @@ -1176,7 +1239,7 @@ func mkinlcall(n, fn *Node, maxCost int32, inlMap map[*Node]bool) *Node { // PAUTO's in the calling functions, and link them off of the // PPARAM's, PAUTOS and PPARAMOUTs of the called function. func inlvar(var_ *Node) *Node { - if Debug['m'] > 3 { + if Debug.m > 3 { fmt.Printf("inlvar %+v\n", var_) } @@ -1223,6 +1286,10 @@ type inlsubst struct { // Temporary result variables. retvars []*Node + // Whether result variables should be initialized at the + // "return" statement. + delayretvars bool + inlvars map[*Node]*Node // bases maps from original PosBase to PosBase with an extra @@ -1255,13 +1322,13 @@ func (subst *inlsubst) node(n *Node) *Node { switch n.Op { case ONAME: if inlvar := subst.inlvars[n]; inlvar != nil { // These will be set during inlnode - if Debug['m'] > 2 { + if Debug.m > 2 { fmt.Printf("substituting name %+v -> %+v\n", n, inlvar) } return inlvar } - if Debug['m'] > 2 { + if Debug.m > 2 { fmt.Printf("not substituting name %+v\n", n) } return n @@ -1291,6 +1358,14 @@ func (subst *inlsubst) node(n *Node) *Node { as.List.Append(n) } as.Rlist.Set(subst.list(n.List)) + + if subst.delayretvars { + for _, n := range as.List.Slice() { + as.Ninit.Append(nod(ODCL, n, nil)) + n.Name.Defn = as + } + } + as = typecheck(as, ctxStmt) m.Ninit.Append(as) } @@ -1353,3 +1428,68 @@ func pruneUnusedAutos(ll []*Node, vis *hairyVisitor) []*Node { } return s } + +// devirtualize replaces interface method calls within fn with direct +// concrete-type method calls where applicable. +func devirtualize(fn *Node) { + Curfn = fn + inspectList(fn.Nbody, func(n *Node) bool { + if n.Op == OCALLINTER { + devirtualizeCall(n) + } + return true + }) +} + +func devirtualizeCall(call *Node) { + recv := staticValue(call.Left.Left) + if recv.Op != OCONVIFACE { + return + } + + typ := recv.Left.Type + if typ.IsInterface() { + return + } + + x := nodl(call.Left.Pos, ODOTTYPE, call.Left.Left, nil) + x.Type = typ + x = nodlSym(call.Left.Pos, OXDOT, x, call.Left.Sym) + x = typecheck(x, ctxExpr|ctxCallee) + switch x.Op { + case ODOTMETH: + if Debug.m != 0 { + Warnl(call.Pos, "devirtualizing %v to %v", call.Left, typ) + } + call.Op = OCALLMETH + call.Left = x + case ODOTINTER: + // Promoted method from embedded interface-typed field (#42279). + if Debug.m != 0 { + Warnl(call.Pos, "partially devirtualizing %v to %v", call.Left, typ) + } + call.Op = OCALLINTER + call.Left = x + default: + // TODO(mdempsky): Turn back into Fatalf after more testing. + if Debug.m != 0 { + Warnl(call.Pos, "failed to devirtualize %v (%v)", x, x.Op) + } + return + } + + // Duplicated logic from typecheck for function call return + // value types. + // + // Receiver parameter size may have changed; need to update + // call.Type to get correct stack offsets for result + // parameters. + checkwidth(x.Type) + switch ft := x.Type; ft.NumResults() { + case 0: + case 1: + call.Type = ft.Results().Field(0).Type + default: + call.Type = ft.Results() + } +} diff --git a/src/cmd/compile/internal/gc/inl_test.go b/src/cmd/compile/internal/gc/inl_test.go index 9d3b8c59fd6..02735e50fb7 100644 --- a/src/cmd/compile/internal/gc/inl_test.go +++ b/src/cmd/compile/internal/gc/inl_test.go @@ -51,6 +51,7 @@ func TestIntendedInlining(t *testing.T) { "funcPC", "getArgInfoFast", "getm", + "getMCache", "isDirectIface", "itabHashFunc", "noescape", @@ -83,7 +84,7 @@ func TestIntendedInlining(t *testing.T) { "puintptr.ptr", "spanOf", "spanOfUnchecked", - //"(*gcWork).putFast", // TODO(austin): For debugging #27993 + "(*gcWork).putFast", "(*gcWork).tryGetFast", "(*guintptr).set", "(*markBits).advance", @@ -115,6 +116,7 @@ func TestIntendedInlining(t *testing.T) { "byLiteral.Len", "byLiteral.Less", "byLiteral.Swap", + "(*dictDecoder).tryWriteCopy", }, "encoding/base64": { "assemble32", diff --git a/src/cmd/compile/internal/gc/lex.go b/src/cmd/compile/internal/gc/lex.go index 1a344c65662..7cce371408a 100644 --- a/src/cmd/compile/internal/gc/lex.go +++ b/src/cmd/compile/internal/gc/lex.go @@ -48,8 +48,11 @@ const ( Nowritebarrierrec // error on write barrier in this or recursive callees Yeswritebarrierrec // cancels Nowritebarrierrec in this function and callees - // Runtime-only type pragmas + // Runtime and cgo type pragmas NotInHeap // values of this type must not be heap allocated + + // Go command pragmas + GoBuildPragma ) const ( @@ -71,6 +74,8 @@ const ( func pragmaFlag(verb string) PragmaFlag { switch verb { + case "go:build": + return GoBuildPragma case "go:nointerface": if objabi.Fieldtrack_enabled != 0 { return Nointerface diff --git a/src/cmd/compile/internal/gc/main.go b/src/cmd/compile/internal/gc/main.go index eedfc4bb259..a6963a3d66e 100644 --- a/src/cmd/compile/internal/gc/main.go +++ b/src/cmd/compile/internal/gc/main.go @@ -34,8 +34,6 @@ import ( "strings" ) -var imported_unsafe bool - var ( buildid string spectre string @@ -48,6 +46,7 @@ var ( Debug_closure int Debug_compilelater int debug_dclstack int + Debug_dumpptrs int Debug_libfuzzer int Debug_panic int Debug_slice int @@ -77,6 +76,7 @@ var debugtab = []struct { {"compilelater", "compile functions as late as possible", &Debug_compilelater}, {"disablenil", "disable nil checks", &disable_checknil}, {"dclstack", "run internal dclstack check", &debug_dclstack}, + {"dumpptrs", "show Node pointer values in Dump/dumplist output", &Debug_dumpptrs}, {"gcprog", "print dump of GC programs", &Debug_gcprog}, {"libfuzzer", "coverage instrumentation for libfuzzer", &Debug_libfuzzer}, {"nil", "print information about nil checks", &Debug_checknil}, @@ -91,6 +91,7 @@ var debugtab = []struct { {"dwarfinl", "print information about DWARF inlined function creation", &Debug_gendwarfinl}, {"softfloat", "force compiler to emit soft-float code", &Debug_softfloat}, {"defer", "print information about defer compilation", &Debug_defer}, + {"fieldtrack", "enable fieldtracking", &objabi.Fieldtrack_enabled}, } const debugHelpHeader = `usage: -d arg[,arg]* and arg is [=] @@ -132,7 +133,7 @@ func hidePanic() { // supportsDynlink reports whether or not the code generator for the given // architecture supports the -shared and -dynlink flags. func supportsDynlink(arch *sys.Arch) bool { - return arch.InFamily(sys.AMD64, sys.ARM, sys.ARM64, sys.I386, sys.PPC64, sys.S390X) + return arch.InFamily(sys.AMD64, sys.ARM, sys.ARM64, sys.I386, sys.PPC64, sys.RISCV64, sys.S390X) } // timing data for compiler phases @@ -211,18 +212,27 @@ func Main(archInit func(*Arch)) { flag.BoolVar(&compiling_runtime, "+", false, "compiling runtime") flag.BoolVar(&compiling_std, "std", false, "compiling standard library") - objabi.Flagcount("%", "debug non-static initializers", &Debug['%']) - objabi.Flagcount("B", "disable bounds checking", &Debug['B']) - objabi.Flagcount("C", "disable printing of columns in error messages", &Debug['C']) // TODO(gri) remove eventually flag.StringVar(&localimport, "D", "", "set relative `path` for local imports") - objabi.Flagcount("E", "debug symbol export", &Debug['E']) + + objabi.Flagcount("%", "debug non-static initializers", &Debug.P) + objabi.Flagcount("B", "disable bounds checking", &Debug.B) + objabi.Flagcount("C", "disable printing of columns in error messages", &Debug.C) + objabi.Flagcount("E", "debug symbol export", &Debug.E) + objabi.Flagcount("K", "debug missing line numbers", &Debug.K) + objabi.Flagcount("L", "show full file names in error messages", &Debug.L) + objabi.Flagcount("N", "disable optimizations", &Debug.N) + objabi.Flagcount("S", "print assembly listing", &Debug.S) + objabi.Flagcount("W", "debug parse tree after type checking", &Debug.W) + objabi.Flagcount("e", "no limit on number of errors reported", &Debug.e) + objabi.Flagcount("h", "halt on error", &Debug.h) + objabi.Flagcount("j", "debug runtime-initialized variables", &Debug.j) + objabi.Flagcount("l", "disable inlining", &Debug.l) + objabi.Flagcount("m", "print optimization decisions", &Debug.m) + objabi.Flagcount("r", "debug generated wrappers", &Debug.r) + objabi.Flagcount("w", "debug type checking", &Debug.w) + objabi.Flagfn1("I", "add `directory` to import search path", addidir) - objabi.Flagcount("K", "debug missing line numbers", &Debug['K']) - objabi.Flagcount("L", "show full file names in error messages", &Debug['L']) - objabi.Flagcount("N", "disable optimizations", &Debug['N']) - objabi.Flagcount("S", "print assembly listing", &Debug['S']) objabi.AddVersionFlag() // -V - objabi.Flagcount("W", "debug parse tree after type checking", &Debug['W']) flag.StringVar(&asmhdr, "asmhdr", "", "write assembly header to `file`") flag.StringVar(&buildid, "buildid", "", "record `id` as the build id in the export metadata") flag.IntVar(&nBackendWorkers, "c", 1, "concurrency during compilation, 1 means no concurrency") @@ -231,17 +241,13 @@ func Main(archInit func(*Arch)) { flag.BoolVar(&flagDWARF, "dwarf", !Wasm, "generate DWARF symbols") flag.BoolVar(&Ctxt.Flag_locationlists, "dwarflocationlists", true, "add location lists to DWARF in optimized mode") flag.IntVar(&genDwarfInline, "gendwarfinl", 2, "generate DWARF inline info records") - objabi.Flagcount("e", "no limit on number of errors reported", &Debug['e']) - objabi.Flagcount("h", "halt on error", &Debug['h']) + objabi.Flagfn1("embedcfg", "read go:embed configuration from `file`", readEmbedCfg) objabi.Flagfn1("importmap", "add `definition` of the form source=actual to import map", addImportMap) objabi.Flagfn1("importcfg", "read import configuration from `file`", readImportCfg) flag.StringVar(&flag_installsuffix, "installsuffix", "", "set pkg directory `suffix`") - objabi.Flagcount("j", "debug runtime-initialized variables", &Debug['j']) - objabi.Flagcount("l", "disable inlining", &Debug['l']) flag.StringVar(&flag_lang, "lang", "", "release to compile for") flag.StringVar(&linkobj, "linkobj", "", "write linker-specific object to `file`") objabi.Flagcount("live", "debug liveness analysis", &debuglive) - objabi.Flagcount("m", "print optimization decisions", &Debug['m']) if sys.MSanSupported(objabi.GOOS, objabi.GOARCH) { flag.BoolVar(&flag_msan, "msan", false, "build code compatible with C/C++ memory sanitizer") } @@ -249,7 +255,6 @@ func Main(archInit func(*Arch)) { flag.StringVar(&outfile, "o", "", "write output to `file`") flag.StringVar(&myimportpath, "p", "", "set expected package import `path`") flag.BoolVar(&writearchive, "pack", false, "write to file.a instead of file.o") - objabi.Flagcount("r", "debug generated wrappers", &Debug['r']) if sys.RaceDetectorSupported(objabi.GOOS, objabi.GOARCH) { flag.BoolVar(&flag_race, "race", false, "enable race detector") } @@ -259,7 +264,6 @@ func Main(archInit func(*Arch)) { } flag.StringVar(&pathPrefix, "trimpath", "", "remove `prefix` from recorded source file paths") flag.BoolVar(&Debug_vlog, "v", false, "increase debug verbosity") - objabi.Flagcount("w", "debug type checking", &Debug['w']) flag.BoolVar(&use_writebarrier, "wb", true, "enable write barrier") var flag_shared bool var flag_dynlink bool @@ -325,9 +329,9 @@ func Main(archInit func(*Arch)) { Ctxt.Flag_shared = flag_dynlink || flag_shared Ctxt.Flag_dynlink = flag_dynlink - Ctxt.Flag_optimize = Debug['N'] == 0 + Ctxt.Flag_optimize = Debug.N == 0 - Ctxt.Debugasm = Debug['S'] + Ctxt.Debugasm = Debug.S Ctxt.Debugvlog = Debug_vlog if flagDWARF { Ctxt.DebugInfo = debuginfo @@ -399,7 +403,7 @@ func Main(archInit func(*Arch)) { instrumenting = true } - if compiling_runtime && Debug['N'] != 0 { + if compiling_runtime && Debug.N != 0 { log.Fatal("cannot disable optimizations while compiling runtime") } if nBackendWorkers < 1 { @@ -504,11 +508,11 @@ func Main(archInit func(*Arch)) { } // enable inlining. for now: - // default: inlining on. (debug['l'] == 1) - // -l: inlining off (debug['l'] == 0) - // -l=2, -l=3: inlining on again, with extra debugging (debug['l'] > 1) - if Debug['l'] <= 1 { - Debug['l'] = 1 - Debug['l'] + // default: inlining on. (Debug.l == 1) + // -l: inlining off (Debug.l == 0) + // -l=2, -l=3: inlining on again, with extra debugging (Debug.l > 1) + if Debug.l <= 1 { + Debug.l = 1 - Debug.l } if jsonLogOpt != "" { // parse version,destination from json logging optimization. @@ -516,6 +520,7 @@ func Main(archInit func(*Arch)) { } ssaDump = os.Getenv("GOSSAFUNC") + ssaDir = os.Getenv("GOSSADIR") if ssaDump != "" { if strings.HasSuffix(ssaDump, "+") { ssaDump = ssaDump[:len(ssaDump)-1] @@ -594,7 +599,7 @@ func Main(archInit func(*Arch)) { timings.Start("fe", "typecheck", "top1") for i := 0; i < len(xtop); i++ { n := xtop[i] - if op := n.Op; op != ODCL && op != OAS && op != OAS2 && (op != ODCLTYPE || !n.Left.Name.Param.Alias) { + if op := n.Op; op != ODCL && op != OAS && op != OAS2 && (op != ODCLTYPE || !n.Left.Name.Param.Alias()) { xtop[i] = typecheck(n, ctxStmt) } } @@ -606,7 +611,7 @@ func Main(archInit func(*Arch)) { timings.Start("fe", "typecheck", "top2") for i := 0; i < len(xtop); i++ { n := xtop[i] - if op := n.Op; op == ODCL || op == OAS || op == OAS2 || op == ODCLTYPE && n.Left.Name.Param.Alias { + if op := n.Op; op == ODCL || op == OAS || op == OAS2 || op == ODCLTYPE && n.Left.Name.Param.Alias() { xtop[i] = typecheck(n, ctxStmt) } } @@ -617,7 +622,7 @@ func Main(archInit func(*Arch)) { var fcount int64 for i := 0; i < len(xtop); i++ { n := xtop[i] - if op := n.Op; op == ODCLFUNC || op == OCLOSURE { + if n.Op == ODCLFUNC { Curfn = n decldepth = 1 saveerrors() @@ -642,6 +647,8 @@ func Main(archInit func(*Arch)) { errorexit() } + fninit(xtop) + // Phase 4: Decide how to capture closed variables. // This needs to run before escape analysis, // because variables captured by value do not escape. @@ -663,7 +670,7 @@ func Main(archInit func(*Arch)) { // Phase 5: Inlining timings.Start("fe", "inlining") if Debug_typecheckinl != 0 { - // Typecheck imported function bodies if debug['l'] > 1, + // Typecheck imported function bodies if Debug.l > 1, // otherwise lazily when used or re-exported. for _, n := range importlist { if n.Func.Inl != nil { @@ -677,7 +684,7 @@ func Main(archInit func(*Arch)) { } } - if Debug['l'] != 0 { + if Debug.l != 0 { // Find functions that can be inlined and clone them before walk expands them. visitBottomUp(xtop, func(list []*Node, recursive bool) { numfns := numNonClosures(list) @@ -688,7 +695,7 @@ func Main(archInit func(*Arch)) { // across more than one function. caninl(n) } else { - if Debug['m'] > 1 { + if Debug.m > 1 { fmt.Printf("%v: cannot inline %v: recursive\n", n.Line(), n.Func.Nname) } } @@ -697,6 +704,13 @@ func Main(archInit func(*Arch)) { }) } + for _, n := range xtop { + if n.Op == ODCLFUNC { + devirtualize(n) + } + } + Curfn = nil + // Phase 6: Escape analysis. // Required for moving heap allocations onto stack, // which in turn is required by the closure implementation, @@ -751,10 +765,6 @@ func Main(archInit func(*Arch)) { } timings.AddEvent(fcount, "funcs") - if nsavederrors+nerrors == 0 { - fninit(xtop) - } - compileFunctions() if nowritebarrierrecCheck != nil { @@ -809,6 +819,9 @@ func Main(archInit func(*Arch)) { } } + if len(funcStack) != 0 { + Fatalf("funcStack is non-empty: %v", len(funcStack)) + } if len(compilequeue) != 0 { Fatalf("%d uncompiled functions", len(compilequeue)) } @@ -966,9 +979,10 @@ func readSymABIs(file, myimportpath string) { if len(parts) != 3 { log.Fatalf(`%s:%d: invalid symabi: syntax is "%s sym abi"`, file, lineNum, parts[0]) } - sym, abi := parts[1], parts[2] - if abi != "ABI0" { // Only supported external ABI right now - log.Fatalf(`%s:%d: invalid symabi: unknown abi "%s"`, file, lineNum, abi) + sym, abistr := parts[1], parts[2] + abi, valid := obj.ParseABI(abistr) + if !valid { + log.Fatalf(`%s:%d: invalid symabi: unknown abi "%s"`, file, lineNum, abistr) } // If the symbol is already prefixed with @@ -981,9 +995,9 @@ func readSymABIs(file, myimportpath string) { // Record for later. if parts[0] == "def" { - symabiDefs[sym] = obj.ABI0 + symabiDefs[sym] = abi } else { - symabiRefs[sym] = obj.ABI0 + symabiRefs[sym] = abi } default: log.Fatalf(`%s:%d: invalid symabi type "%s"`, file, lineNum, parts[0]) @@ -1172,7 +1186,6 @@ func importfile(f *Val) *types.Pkg { } if path_ == "unsafe" { - imported_unsafe = true return unsafepkg } @@ -1405,29 +1418,34 @@ func IsAlias(sym *types.Sym) bool { return sym.Def != nil && asNode(sym.Def).Sym != sym } -// By default, assume any debug flags are incompatible with concurrent compilation. -// A few are safe and potentially in common use for normal compiles, though; mark them as such here. -var concurrentFlagOK = [256]bool{ - 'B': true, // disabled bounds checking - 'C': true, // disable printing of columns in error messages - 'e': true, // no limit on errors; errors all come from non-concurrent code - 'I': true, // add `directory` to import search path - 'N': true, // disable optimizations - 'l': true, // disable inlining - 'w': true, // all printing happens before compilation - 'W': true, // all printing happens before compilation - 'S': true, // printing disassembly happens at the end (but see concurrentBackendAllowed below) +// By default, assume any debug flags are incompatible with concurrent +// compilation. A few are safe and potentially in common use for +// normal compiles, though; return true for those. +func concurrentFlagOk() bool { + // Report whether any debug flag that would prevent concurrent + // compilation is set, by zeroing out the allowed ones and then + // checking if the resulting struct is zero. + d := Debug + d.B = 0 // disable bounds checking + d.C = 0 // disable printing of columns in error messages + d.e = 0 // no limit on errors; errors all come from non-concurrent code + d.N = 0 // disable optimizations + d.l = 0 // disable inlining + d.w = 0 // all printing happens before compilation + d.W = 0 // all printing happens before compilation + d.S = 0 // printing disassembly happens at the end (but see concurrentBackendAllowed below) + + return d == DebugFlags{} } func concurrentBackendAllowed() bool { - for i, x := range &Debug { - if x != 0 && !concurrentFlagOK[i] { - return false - } + if !concurrentFlagOk() { + return false } - // Debug['S'] by itself is ok, because all printing occurs + + // Debug.S by itself is ok, because all printing occurs // while writing the object file, and that is non-concurrent. - // Adding Debug_vlog, however, causes Debug['S'] to also print + // Adding Debug_vlog, however, causes Debug.S to also print // while flushing the plist, which happens concurrently. if Debug_vlog || debugstr != "" || debuglive > 0 { return false diff --git a/src/cmd/compile/internal/gc/noder.go b/src/cmd/compile/internal/gc/noder.go index 5dce533e4b7..67d24ef0bc7 100644 --- a/src/cmd/compile/internal/gc/noder.go +++ b/src/cmd/compile/internal/gc/noder.go @@ -11,6 +11,7 @@ import ( "runtime" "strconv" "strings" + "unicode" "unicode/utf8" "cmd/compile/internal/syntax" @@ -90,7 +91,11 @@ func (p *noder) makeSrcPosBase(b0 *syntax.PosBase) *src.PosBase { } else { // line directive base p0 := b0.Pos() - p1 := src.MakePos(p.makeSrcPosBase(p0.Base()), p0.Line(), p0.Col()) + p0b := p0.Base() + if p0b == b0 { + panic("infinite recursion in makeSrcPosBase") + } + p1 := src.MakePos(p.makeSrcPosBase(p0b), p0.Line(), p0.Col()) b1 = src.NewLinePragmaBase(p1, fn, fileh(fn), b0.Line(), b0.Col()) } p.basemap[b0] = b1 @@ -130,11 +135,13 @@ type noder struct { base *src.PosBase } - file *syntax.File - linknames []linkname - pragcgobuf [][]string - err chan syntax.Error - scope ScopeID + file *syntax.File + linknames []linkname + pragcgobuf [][]string + err chan syntax.Error + scope ScopeID + importedUnsafe bool + importedEmbed bool // scopeVars is a stack tracking the number of variables declared in the // current function at the moment each open scope was opened. @@ -236,19 +243,21 @@ type linkname struct { func (p *noder) node() { types.Block = 1 - imported_unsafe = false + p.importedUnsafe = false + p.importedEmbed = false p.setlineno(p.file.PkgName) mkpackage(p.file.PkgName.Value) if pragma, ok := p.file.Pragma.(*Pragma); ok { + pragma.Flag &^= GoBuildPragma p.checkUnused(pragma) } xtop = append(xtop, p.decls(p.file.DeclList)...) for _, n := range p.linknames { - if !imported_unsafe { + if !p.importedUnsafe { p.yyerrorpos(n.pos, "//go:linkname only allowed in Go files that import \"unsafe\"") continue } @@ -323,7 +332,6 @@ func (p *noder) importDecl(imp *syntax.ImportDecl) { val := p.basicLit(imp.Path) ipkg := importfile(&val) - if ipkg == nil { if nerrors == 0 { Fatalf("phase error in import") @@ -331,6 +339,13 @@ func (p *noder) importDecl(imp *syntax.ImportDecl) { return } + if ipkg == unsafepkg { + p.importedUnsafe = true + } + if ipkg.Path == "embed" { + p.importedEmbed = true + } + ipkg.Direct = true var my *types.Sym @@ -372,6 +387,20 @@ func (p *noder) varDecl(decl *syntax.VarDecl) []*Node { } if pragma, ok := decl.Pragma.(*Pragma); ok { + if len(pragma.Embeds) > 0 { + if !p.importedEmbed { + // This check can't be done when building the list pragma.Embeds + // because that list is created before the noder starts walking over the file, + // so at that point it hasn't seen the imports. + // We're left to check now, just before applying the //go:embed lines. + for _, e := range pragma.Embeds { + p.yyerrorpos(e.Pos, "//go:embed only allowed in Go files that import \"embed\"") + } + } else { + exprs = varEmbed(p, names, typ, exprs, pragma.Embeds) + } + pragma.Embeds = nil + } p.checkUnused(pragma) } @@ -454,17 +483,17 @@ func (p *noder) typeDecl(decl *syntax.TypeDecl) *Node { param := n.Name.Param param.Ntype = typ - param.Alias = decl.Alias + param.SetAlias(decl.Alias) if pragma, ok := decl.Pragma.(*Pragma); ok { if !decl.Alias { - param.Pragma = pragma.Flag & TypePragmas + param.SetPragma(pragma.Flag & TypePragmas) pragma.Flag &^= TypePragmas } p.checkUnused(pragma) } nod := p.nod(decl, ODCLTYPE, n, nil) - if param.Alias && !langSupported(1, 9, localpkg) { + if param.Alias() && !langSupported(1, 9, localpkg) { yyerrorl(nod.Pos, "type aliases only supported as of -lang=go1.9") } return nod @@ -773,7 +802,7 @@ func (p *noder) sum(x syntax.Expr) *Node { n := p.expr(x) if Isconst(n, CTSTR) && n.Sym == nil { nstr = n - chunks = append(chunks, strlit(nstr)) + chunks = append(chunks, nstr.StringVal()) } for i := len(adds) - 1; i >= 0; i-- { @@ -783,12 +812,12 @@ func (p *noder) sum(x syntax.Expr) *Node { if Isconst(r, CTSTR) && r.Sym == nil { if nstr != nil { // Collapse r into nstr instead of adding to n. - chunks = append(chunks, strlit(r)) + chunks = append(chunks, r.StringVal()) continue } nstr = r - chunks = append(chunks, strlit(nstr)) + chunks = append(chunks, nstr.StringVal()) } else { if len(chunks) > 1 { nstr.SetVal(Val{U: strings.Join(chunks, "")}) @@ -1437,11 +1466,6 @@ func (p *noder) mkname(name *syntax.Name) *Node { return mkname(p.name(name)) } -func (p *noder) newname(name *syntax.Name) *Node { - // TODO(mdempsky): Set line number? - return newname(p.name(name)) -} - func (p *noder) wrapname(n syntax.Node, x *Node) *Node { // These nodes do not carry line numbers. // Introduce a wrapper node to give them the correct line. @@ -1497,13 +1521,15 @@ var allowedStdPragmas = map[string]bool{ "go:cgo_import_dynamic": true, "go:cgo_ldflag": true, "go:cgo_dynamic_linker": true, + "go:embed": true, "go:generate": true, } // *Pragma is the value stored in a syntax.Pragma during parsing. type Pragma struct { - Flag PragmaFlag // collected bits - Pos []PragmaPos // position of each individual flag + Flag PragmaFlag // collected bits + Pos []PragmaPos // position of each individual flag + Embeds []PragmaEmbed } type PragmaPos struct { @@ -1511,12 +1537,22 @@ type PragmaPos struct { Pos syntax.Pos } +type PragmaEmbed struct { + Pos syntax.Pos + Patterns []string +} + func (p *noder) checkUnused(pragma *Pragma) { for _, pos := range pragma.Pos { if pos.Flag&pragma.Flag != 0 { p.yyerrorpos(pos.Pos, "misplaced compiler directive") } } + if len(pragma.Embeds) > 0 { + for _, e := range pragma.Embeds { + p.yyerrorpos(e.Pos, "misplaced go:embed directive") + } + } } func (p *noder) checkUnusedDuringParse(pragma *Pragma) { @@ -1525,6 +1561,11 @@ func (p *noder) checkUnusedDuringParse(pragma *Pragma) { p.error(syntax.Error{Pos: pos.Pos, Msg: "misplaced compiler directive"}) } } + if len(pragma.Embeds) > 0 { + for _, e := range pragma.Embeds { + p.error(syntax.Error{Pos: e.Pos, Msg: "misplaced go:embed directive"}) + } + } } // pragma is called concurrently if files are parsed concurrently. @@ -1569,6 +1610,17 @@ func (p *noder) pragma(pos syntax.Pos, blankLine bool, text string, old syntax.P } p.linknames = append(p.linknames, linkname{pos, f[1], target}) + case text == "go:embed", strings.HasPrefix(text, "go:embed "): + args, err := parseGoEmbed(text[len("go:embed"):]) + if err != nil { + p.error(syntax.Error{Pos: pos, Msg: err.Error()}) + } + if len(args) == 0 { + p.error(syntax.Error{Pos: pos, Msg: "usage: //go:embed pattern..."}) + break + } + pragma.Embeds = append(pragma.Embeds, PragmaEmbed{pos, args}) + case strings.HasPrefix(text, "go:cgo_import_dynamic "): // This is permitted for general use because Solaris // code relies on it in golang.org/x/sys/unix and others. @@ -1641,3 +1693,64 @@ func mkname(sym *types.Sym) *Node { } return n } + +// parseGoEmbed parses the text following "//go:embed" to extract the glob patterns. +// It accepts unquoted space-separated patterns as well as double-quoted and back-quoted Go strings. +// go/build/read.go also processes these strings and contains similar logic. +func parseGoEmbed(args string) ([]string, error) { + var list []string + for args = strings.TrimSpace(args); args != ""; args = strings.TrimSpace(args) { + var path string + Switch: + switch args[0] { + default: + i := len(args) + for j, c := range args { + if unicode.IsSpace(c) { + i = j + break + } + } + path = args[:i] + args = args[i:] + + case '`': + i := strings.Index(args[1:], "`") + if i < 0 { + return nil, fmt.Errorf("invalid quoted string in //go:embed: %s", args) + } + path = args[1 : 1+i] + args = args[1+i+1:] + + case '"': + i := 1 + for ; i < len(args); i++ { + if args[i] == '\\' { + i++ + continue + } + if args[i] == '"' { + q, err := strconv.Unquote(args[:i+1]) + if err != nil { + return nil, fmt.Errorf("invalid quoted string in //go:embed: %s", args[:i+1]) + } + path = q + args = args[i+1:] + break Switch + } + } + if i >= len(args) { + return nil, fmt.Errorf("invalid quoted string in //go:embed: %s", args) + } + } + + if args != "" { + r, _ := utf8.DecodeRuneInString(args) + if !unicode.IsSpace(r) { + return nil, fmt.Errorf("invalid quoted string in //go:embed: %s", args) + } + } + list = append(list, path) + } + return list, nil +} diff --git a/src/cmd/compile/internal/gc/obj.go b/src/cmd/compile/internal/gc/obj.go index af5037c5a8b..32aa7c5bb1a 100644 --- a/src/cmd/compile/internal/gc/obj.go +++ b/src/cmd/compile/internal/gc/obj.go @@ -14,6 +14,8 @@ import ( "encoding/json" "fmt" "io" + "io/ioutil" + "os" "sort" "strconv" ) @@ -113,14 +115,19 @@ func dumpCompilerObj(bout *bio.Writer) { func dumpdata() { externs := len(externdcl) + xtops := len(xtop) dumpglobls() addptabs() + exportlistLen := len(exportlist) addsignats(externdcl) dumpsignats() dumptabs() + ptabsLen := len(ptabs) + itabsLen := len(itabs) dumpimportstrings() dumpbasictypes() + dumpembeds() // Calls to dumpsignats can generate functions, // like method wrappers and hash and equality routines. @@ -129,9 +136,19 @@ func dumpdata() { // number of types in a finite amount of code. // In the typical case, we loop 0 or 1 times. // It was not until issue 24761 that we found any code that required a loop at all. - for len(compilequeue) > 0 { + for { + for i := xtops; i < len(xtop); i++ { + n := xtop[i] + if n.Op == ODCLFUNC { + funccompile(n) + } + } + xtops = len(xtop) compileFunctions() dumpsignats() + if xtops == len(xtop) { + break + } } // Dump extra globals. @@ -149,6 +166,16 @@ func dumpdata() { } addGCLocals() + + if exportlistLen != len(exportlist) { + Fatalf("exportlist changed after compile functions loop") + } + if ptabsLen != len(ptabs) { + Fatalf("ptabs changed after compile functions loop") + } + if itabsLen != len(itabs) { + Fatalf("itabs changed after compile functions loop") + } } func dumpLinkerObj(bout *bio.Writer) { @@ -248,7 +275,7 @@ func dumpGlobalConst(n *Node) { default: return } - Ctxt.DwarfIntConst(myimportpath, n.Sym.Name, typesymname(t), n.Int64()) + Ctxt.DwarfIntConst(myimportpath, n.Sym.Name, typesymname(t), n.Int64Val()) } func dumpglobls() { @@ -281,20 +308,21 @@ func dumpglobls() { // global symbols can't be declared during parallel compilation. func addGCLocals() { for _, s := range Ctxt.Text { - if s.Func == nil { + fn := s.Func() + if fn == nil { continue } - for _, gcsym := range []*obj.LSym{s.Func.GCArgs, s.Func.GCLocals, s.Func.GCRegs} { + for _, gcsym := range []*obj.LSym{fn.GCArgs, fn.GCLocals} { if gcsym != nil && !gcsym.OnList() { ggloblsym(gcsym, int32(len(gcsym.P)), obj.RODATA|obj.DUPOK) } } - if x := s.Func.StackObjects; x != nil { + if x := fn.StackObjects; x != nil { attr := int16(obj.RODATA) ggloblsym(x, int32(len(x.P)), attr) x.Set(obj.AttrStatic, true) } - if x := s.Func.OpenCodedDeferInfo; x != nil { + if x := fn.OpenCodedDeferInfo; x != nil { ggloblsym(x, int32(len(x.P)), obj.RODATA|obj.DUPOK) } } @@ -333,28 +361,31 @@ func dbvec(s *obj.LSym, off int, bv bvec) int { return off } +const ( + stringSymPrefix = "go.string." + stringSymPattern = ".gostring.%d.%x" +) + +// stringsym returns a symbol containing the string s. +// The symbol contains the string data, not a string header. func stringsym(pos src.XPos, s string) (data *obj.LSym) { var symname string if len(s) > 100 { // Huge strings are hashed to avoid long names in object files. // Indulge in some paranoia by writing the length of s, too, // as protection against length extension attacks. + // Same pattern is known to fileStringSym below. h := sha256.New() io.WriteString(h, s) - symname = fmt.Sprintf(".gostring.%d.%x", len(s), h.Sum(nil)) + symname = fmt.Sprintf(stringSymPattern, len(s), h.Sum(nil)) } else { // Small strings get named directly by their contents. symname = strconv.Quote(s) } - const prefix = "go.string." - symdataname := prefix + symname - - symdata := Ctxt.Lookup(symdataname) - + symdata := Ctxt.Lookup(stringSymPrefix + symname) if !symdata.OnList() { - // string data - off := dsname(symdata, 0, s, pos, "string") + off := dstringdata(symdata, 0, s, pos, "string") ggloblsym(symdata, int32(off), obj.DUPOK|obj.RODATA|obj.LOCAL) symdata.Set(obj.AttrContentAddressable, true) } @@ -362,26 +393,122 @@ func stringsym(pos src.XPos, s string) (data *obj.LSym) { return symdata } -var slicebytes_gen int +// fileStringSym returns a symbol for the contents and the size of file. +// If readonly is true, the symbol shares storage with any literal string +// or other file with the same content and is placed in a read-only section. +// If readonly is false, the symbol is a read-write copy separate from any other, +// for use as the backing store of a []byte. +// The content hash of file is copied into hash. (If hash is nil, nothing is copied.) +// The returned symbol contains the data itself, not a string header. +func fileStringSym(pos src.XPos, file string, readonly bool, hash []byte) (*obj.LSym, int64, error) { + f, err := os.Open(file) + if err != nil { + return nil, 0, err + } + defer f.Close() + info, err := f.Stat() + if err != nil { + return nil, 0, err + } + if !info.Mode().IsRegular() { + return nil, 0, fmt.Errorf("not a regular file") + } + size := info.Size() + if size <= 1*1024 { + data, err := ioutil.ReadAll(f) + if err != nil { + return nil, 0, err + } + if int64(len(data)) != size { + return nil, 0, fmt.Errorf("file changed between reads") + } + var sym *obj.LSym + if readonly { + sym = stringsym(pos, string(data)) + } else { + sym = slicedata(pos, string(data)).Sym.Linksym() + } + if len(hash) > 0 { + sum := sha256.Sum256(data) + copy(hash, sum[:]) + } + return sym, size, nil + } + if size > 2e9 { + // ggloblsym takes an int32, + // and probably the rest of the toolchain + // can't handle such big symbols either. + // See golang.org/issue/9862. + return nil, 0, fmt.Errorf("file too large") + } -func slicebytes(nam *Node, s string) { - slicebytes_gen++ - symname := fmt.Sprintf(".gobytes.%d", slicebytes_gen) + // File is too big to read and keep in memory. + // Compute hash if needed for read-only content hashing or if the caller wants it. + var sum []byte + if readonly || len(hash) > 0 { + h := sha256.New() + n, err := io.Copy(h, f) + if err != nil { + return nil, 0, err + } + if n != size { + return nil, 0, fmt.Errorf("file changed between reads") + } + sum = h.Sum(nil) + copy(hash, sum) + } + + var symdata *obj.LSym + if readonly { + symname := fmt.Sprintf(stringSymPattern, size, sum) + symdata = Ctxt.Lookup(stringSymPrefix + symname) + if !symdata.OnList() { + info := symdata.NewFileInfo() + info.Name = file + info.Size = size + ggloblsym(symdata, int32(size), obj.DUPOK|obj.RODATA|obj.LOCAL) + // Note: AttrContentAddressable cannot be set here, + // because the content-addressable-handling code + // does not know about file symbols. + } + } else { + // Emit a zero-length data symbol + // and then fix up length and content to use file. + symdata = slicedata(pos, "").Sym.Linksym() + symdata.Size = size + symdata.Type = objabi.SNOPTRDATA + info := symdata.NewFileInfo() + info.Name = file + info.Size = size + } + + return symdata, size, nil +} + +var slicedataGen int + +func slicedata(pos src.XPos, s string) *Node { + slicedataGen++ + symname := fmt.Sprintf(".gobytes.%d", slicedataGen) sym := localpkg.Lookup(symname) symnode := newname(sym) sym.Def = asTypesNode(symnode) lsym := sym.Linksym() - off := dsname(lsym, 0, s, nam.Pos, "slice") + off := dstringdata(lsym, 0, s, pos, "slice") ggloblsym(lsym, int32(off), obj.NOPTR|obj.LOCAL) + return symnode +} + +func slicebytes(nam *Node, s string) { if nam.Op != ONAME { Fatalf("slicebytes %v", nam) } - slicesym(nam, symnode, int64(len(s))) + slicesym(nam, slicedata(nam.Pos, s), int64(len(s))) } -func dsname(s *obj.LSym, off int, t string, pos src.XPos, what string) int { +func dstringdata(s *obj.LSym, off int, t string, pos src.XPos, what string) int { // Objects that are too large will cause the data section to overflow right away, // causing a cryptic error message by the linker. Check for oversize objects here // and provide a useful error message instead. diff --git a/src/cmd/compile/internal/gc/order.go b/src/cmd/compile/internal/gc/order.go index aa91160e5c9..30e1535c09c 100644 --- a/src/cmd/compile/internal/gc/order.go +++ b/src/cmd/compile/internal/gc/order.go @@ -50,7 +50,7 @@ type Order struct { // Order rewrites fn.Nbody to apply the ordering constraints // described in the comment at the top of the file. func order(fn *Node) { - if Debug['W'] > 1 { + if Debug.W > 1 { s := fmt.Sprintf("\nbefore order %v", fn.Func.Nname.Sym) dumplist(s, fn.Nbody) } @@ -288,20 +288,13 @@ func (o *Order) popTemp(mark ordermarker) { o.temp = o.temp[:mark] } -// cleanTempNoPop emits VARKILL and if needed VARLIVE instructions -// to *out for each temporary above the mark on the temporary stack. +// cleanTempNoPop emits VARKILL instructions to *out +// for each temporary above the mark on the temporary stack. // It does not pop the temporaries from the stack. func (o *Order) cleanTempNoPop(mark ordermarker) []*Node { var out []*Node for i := len(o.temp) - 1; i >= int(mark); i-- { n := o.temp[i] - if n.Name.Keepalive() { - n.Name.SetKeepalive(false) - n.Name.SetAddrtaken(true) // ensure SSA keeps the n variable - live := nod(OVARLIVE, n, nil) - live = typecheck(live, ctxStmt) - out = append(out, live) - } kill := nod(OVARKILL, n, nil) kill = typecheck(kill, ctxStmt) out = append(out, kill) @@ -330,12 +323,7 @@ func (o *Order) stmtList(l Nodes) { // and rewrites it to: // m = OMAKESLICECOPY([]T, x, s); nil func orderMakeSliceCopy(s []*Node) { - const go115makeslicecopy = true - if !go115makeslicecopy { - return - } - - if Debug['N'] != 0 || instrumenting { + if Debug.N != 0 || instrumenting { return } @@ -500,8 +488,9 @@ func (o *Order) call(n *Node) { // still alive when we pop the temp stack. if arg.Op == OCONVNOP && arg.Left.Type.IsUnsafePtr() { x := o.copyExpr(arg.Left, arg.Left.Type, false) - x.Name.SetKeepalive(true) arg.Left = x + x.Name.SetAddrtaken(true) // ensure SSA keeps the x variable + n.Nbody.Append(typecheck(nod(OVARLIVE, x, nil), ctxStmt)) } } @@ -902,7 +891,7 @@ func (o *Order) stmt(n *Node) { // c is always evaluated; x and ok are only evaluated when assigned. r.Right.Left = o.expr(r.Right.Left, nil) - if r.Right.Left.Op != ONAME { + if !r.Right.Left.IsAutoTmp() { r.Right.Left = o.copyExpr(r.Right.Left, r.Right.Left.Type, false) } @@ -1108,7 +1097,7 @@ func (o *Order) expr(n, lhs *Node) *Node { haslit := false for _, n1 := range n.List.Slice() { hasbyte = hasbyte || n1.Op == OBYTES2STR - haslit = haslit || n1.Op == OLITERAL && len(strlit(n1)) != 0 + haslit = haslit || n1.Op == OLITERAL && len(n1.StringVal()) != 0 } if haslit && hasbyte { @@ -1280,7 +1269,7 @@ func (o *Order) expr(n, lhs *Node) *Node { var t *types.Type switch n.Op { case OSLICELIT: - t = types.NewArray(n.Type.Elem(), n.Right.Int64()) + t = types.NewArray(n.Type.Elem(), n.Right.Int64Val()) case OCALLPART: t = partialCallType(n) } diff --git a/src/cmd/compile/internal/gc/pgen.go b/src/cmd/compile/internal/gc/pgen.go index 74262595b05..353f4b08c9d 100644 --- a/src/cmd/compile/internal/gc/pgen.go +++ b/src/cmd/compile/internal/gc/pgen.go @@ -231,6 +231,11 @@ func compile(fn *Node) { return } + // Set up the function's LSym early to avoid data races with the assemblers. + // Do this before walk, as walk needs the LSym to set attributes/relocations + // (e.g. in markTypeUsedInInterface). + fn.Func.initLSym(true) + walk(fn) if nerrors != 0 { return @@ -250,9 +255,6 @@ func compile(fn *Node) { return } - // Set up the function's LSym early to avoid data races with the assemblers. - fn.Func.initLSym(true) - // Make sure type syms are declared for all types that might // be types of stack objects. We need to do this here // because symbols must be allocated before the parallel @@ -264,8 +266,8 @@ func compile(fn *Node) { dtypesym(n.Type) // Also make sure we allocate a linker symbol // for the stack object data, for the same reason. - if fn.Func.lsym.Func.StackObjects == nil { - fn.Func.lsym.Func.StackObjects = Ctxt.Lookup(fn.Func.lsym.Name + ".stkobj") + if fn.Func.lsym.Func().StackObjects == nil { + fn.Func.lsym.Func().StackObjects = Ctxt.Lookup(fn.Func.lsym.Name + ".stkobj") } } } @@ -413,7 +415,7 @@ func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.S case PAUTO: if !n.Name.Used() { // Text == nil -> generating abstract function - if fnsym.Func.Text != nil { + if fnsym.Func().Text != nil { Fatalf("debuginfo unused node (AllocFrame should truncate fn.Func.Dcl)") } continue @@ -423,7 +425,7 @@ func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.S continue } apdecls = append(apdecls, n) - fnsym.Func.RecordAutoType(ngotype(n).Linksym()) + fnsym.Func().RecordAutoType(ngotype(n).Linksym()) } decls, dwarfVars := createDwarfVars(fnsym, fn.Func, apdecls) @@ -433,7 +435,7 @@ func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.S // the function symbol to insure that the type included in DWARF // processing during linking. typesyms := []*obj.LSym{} - for t, _ := range fnsym.Func.Autot { + for t, _ := range fnsym.Func().Autot { typesyms = append(typesyms, t) } sort.Sort(obj.BySymName(typesyms)) @@ -442,7 +444,7 @@ func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.S r.Sym = sym r.Type = objabi.R_USETYPE } - fnsym.Func.Autot = nil + fnsym.Func().Autot = nil var varScopes []ScopeID for _, decl := range decls { @@ -520,7 +522,7 @@ func createSimpleVar(fnsym *obj.LSym, n *Node) *dwarf.Var { } typename := dwarf.InfoPrefix + typesymname(n.Type) - delete(fnsym.Func.Autot, ngotype(n).Linksym()) + delete(fnsym.Func().Autot, ngotype(n).Linksym()) inlIndex := 0 if genDwarfInline > 1 { if n.Name.InlFormal() || n.Name.InlLocal() { @@ -665,7 +667,7 @@ func createDwarfVars(fnsym *obj.LSym, fn *Func, apDecls []*Node) ([]*Node, []*dw ChildIndex: -1, }) // Record go type of to insure that it gets emitted by the linker. - fnsym.Func.RecordAutoType(ngotype(n).Linksym()) + fnsym.Func().RecordAutoType(ngotype(n).Linksym()) } return decls, vars @@ -729,7 +731,7 @@ func createComplexVar(fnsym *obj.LSym, fn *Func, varID ssa.VarID) *dwarf.Var { } gotype := ngotype(n).Linksym() - delete(fnsym.Func.Autot, gotype) + delete(fnsym.Func().Autot, gotype) typename := dwarf.InfoPrefix + gotype.Name[len("type."):] inlIndex := 0 if genDwarfInline > 1 { diff --git a/src/cmd/compile/internal/gc/plive.go b/src/cmd/compile/internal/gc/plive.go index 8976ed657ab..a48173e0d65 100644 --- a/src/cmd/compile/internal/gc/plive.go +++ b/src/cmd/compile/internal/gc/plive.go @@ -24,16 +24,6 @@ import ( "strings" ) -// go115ReduceLiveness disables register maps and only produces stack -// maps at call sites. -// -// In Go 1.15, we changed debug call injection to use conservative -// scanning instead of precise pointer maps, so these are no longer -// necessary. -// -// Keep in sync with runtime/preempt.go:go115ReduceLiveness. -const go115ReduceLiveness = true - // OpVarDef is an annotation for the liveness analysis, marking a place // where a complete initialization (definition) of a variable begins. // Since the liveness analysis can see initialization of single-word @@ -96,15 +86,15 @@ type BlockEffects struct { // // uevar: upward exposed variables (used before set in block) // varkill: killed variables (set in block) - uevar varRegVec - varkill varRegVec + uevar bvec + varkill bvec // Computed during Liveness.solve using control flow information: // // livein: variables live at block entry // liveout: variables live at block exit - livein varRegVec - liveout varRegVec + livein bvec + liveout bvec } // A collection of global state used by liveness analysis. @@ -128,16 +118,14 @@ type Liveness struct { // current Block during Liveness.epilogue. Indexed in Value // order for that block. Additionally, for the entry block // livevars[0] is the entry bitmap. Liveness.compact moves - // these to stackMaps and regMaps. - livevars []varRegVec + // these to stackMaps. + livevars []bvec // livenessMap maps from safe points (i.e., CALLs) to their // liveness map indexes. livenessMap LivenessMap stackMapSet bvecSet stackMaps []bvec - regMapSet map[liveRegMask]int - regMaps []liveRegMask cache progeffectscache } @@ -158,7 +146,7 @@ func (m *LivenessMap) reset() { delete(m.vals, k) } } - m.deferreturn = LivenessInvalid + m.deferreturn = LivenessDontCare } func (m *LivenessMap) set(v *ssa.Value, i LivenessIndex) { @@ -166,27 +154,17 @@ func (m *LivenessMap) set(v *ssa.Value, i LivenessIndex) { } func (m LivenessMap) Get(v *ssa.Value) LivenessIndex { - if !go115ReduceLiveness { - // All safe-points are in the map, so if v isn't in - // the map, it's an unsafe-point. - if idx, ok := m.vals[v.ID]; ok { - return idx - } - return LivenessInvalid - } - // If v isn't in the map, then it's a "don't care" and not an // unsafe-point. if idx, ok := m.vals[v.ID]; ok { return idx } - return LivenessIndex{StackMapDontCare, StackMapDontCare, false} + return LivenessIndex{StackMapDontCare, false} } // LivenessIndex stores the liveness map information for a Value. type LivenessIndex struct { stackMapIndex int - regMapIndex int // only for !go115ReduceLiveness // isUnsafePoint indicates that this is an unsafe-point. // @@ -197,8 +175,10 @@ type LivenessIndex struct { isUnsafePoint bool } -// LivenessInvalid indicates an unsafe point with no stack map. -var LivenessInvalid = LivenessIndex{StackMapDontCare, StackMapDontCare, true} // only for !go115ReduceLiveness +// LivenessDontCare indicates that the liveness information doesn't +// matter. Currently it is used in deferreturn liveness when we don't +// actually need it. It should never be emitted to the PCDATA stream. +var LivenessDontCare = LivenessIndex{StackMapDontCare, true} // StackMapDontCare indicates that the stack map index at a Value // doesn't matter. @@ -212,46 +192,12 @@ func (idx LivenessIndex) StackMapValid() bool { return idx.stackMapIndex != StackMapDontCare } -func (idx LivenessIndex) RegMapValid() bool { - return idx.regMapIndex != StackMapDontCare -} - type progeffectscache struct { retuevar []int32 tailuevar []int32 initialized bool } -// varRegVec contains liveness bitmaps for variables and registers. -type varRegVec struct { - vars bvec - regs liveRegMask -} - -func (v *varRegVec) Eq(v2 varRegVec) bool { - return v.vars.Eq(v2.vars) && v.regs == v2.regs -} - -func (v *varRegVec) Copy(v2 varRegVec) { - v.vars.Copy(v2.vars) - v.regs = v2.regs -} - -func (v *varRegVec) Clear() { - v.vars.Clear() - v.regs = 0 -} - -func (v *varRegVec) Or(v1, v2 varRegVec) { - v.vars.Or(v1.vars, v2.vars) - v.regs = v1.regs | v2.regs -} - -func (v *varRegVec) AndNot(v1, v2 varRegVec) { - v.vars.AndNot(v1.vars, v2.vars) - v.regs = v1.regs &^ v2.regs -} - // livenessShouldTrack reports whether the liveness analysis // should track the variable n. // We don't care about variables that have no pointers, @@ -400,110 +346,6 @@ func affectedNode(v *ssa.Value) (*Node, ssa.SymEffect) { } } -// regEffects returns the registers affected by v. -func (lv *Liveness) regEffects(v *ssa.Value) (uevar, kill liveRegMask) { - if go115ReduceLiveness { - return 0, 0 - } - if v.Op == ssa.OpPhi { - // All phi node arguments must come from the same - // register and the result must also go to that - // register, so there's no overall effect. - return 0, 0 - } - addLocs := func(mask liveRegMask, v *ssa.Value, ptrOnly bool) liveRegMask { - if int(v.ID) >= len(lv.f.RegAlloc) { - // v has no allocated registers. - return mask - } - loc := lv.f.RegAlloc[v.ID] - if loc == nil { - // v has no allocated registers. - return mask - } - if v.Op == ssa.OpGetG { - // GetG represents the G register, which is a - // pointer, but not a valid GC register. The - // current G is always reachable, so it's okay - // to ignore this register. - return mask - } - - // Collect registers and types from v's location. - var regs [2]*ssa.Register - nreg := 0 - switch loc := loc.(type) { - case ssa.LocalSlot: - return mask - case *ssa.Register: - if ptrOnly && !v.Type.HasPointers() { - return mask - } - regs[0] = loc - nreg = 1 - case ssa.LocPair: - // The value will have TTUPLE type, and the - // children are nil or *ssa.Register. - if v.Type.Etype != types.TTUPLE { - v.Fatalf("location pair %s has non-tuple type %v", loc, v.Type) - } - for i, loc1 := range &loc { - if loc1 == nil { - continue - } - if ptrOnly && !v.Type.FieldType(i).HasPointers() { - continue - } - regs[nreg] = loc1.(*ssa.Register) - nreg++ - } - default: - v.Fatalf("weird RegAlloc location: %s (%T)", loc, loc) - } - - // Add register locations to vars. - for _, reg := range regs[:nreg] { - if reg.GCNum() == -1 { - if ptrOnly { - v.Fatalf("pointer in non-pointer register %v", reg) - } else { - continue - } - } - mask |= 1 << uint(reg.GCNum()) - } - return mask - } - - // v clobbers all registers it writes to (whether or not the - // write is pointer-typed). - kill = addLocs(0, v, false) - for _, arg := range v.Args { - // v uses all registers is reads from, but we only - // care about marking those containing pointers. - uevar = addLocs(uevar, arg, true) - } - return uevar, kill -} - -type liveRegMask uint32 // only if !go115ReduceLiveness - -func (m liveRegMask) niceString(config *ssa.Config) string { - if m == 0 { - return "" - } - str := "" - for i, reg := range config.GCRegMap { - if m&(1<= f.NumBlocks() { lv.be = lc.be[:f.NumBlocks()] } - lv.livenessMap = LivenessMap{vals: lc.livenessMap.vals, deferreturn: LivenessInvalid} + lv.livenessMap = LivenessMap{vals: lc.livenessMap.vals, deferreturn: LivenessDontCare} lc.livenessMap.vals = nil } if lv.be == nil { @@ -546,10 +386,10 @@ func newliveness(fn *Node, f *ssa.Func, vars []*Node, idx map[*Node]int32, stkpt for _, b := range f.Blocks { be := lv.blockEffects(b) - be.uevar = varRegVec{vars: bulk.next()} - be.varkill = varRegVec{vars: bulk.next()} - be.livein = varRegVec{vars: bulk.next()} - be.liveout = varRegVec{vars: bulk.next()} + be.uevar = bulk.next() + be.varkill = bulk.next() + be.livein = bulk.next() + be.liveout = bulk.next() } lv.livenessMap.reset() @@ -637,20 +477,6 @@ func onebitwalktype1(t *types.Type, off int64, bv bvec) { } } -// usedRegs returns the maximum width of the live register map. -func (lv *Liveness) usedRegs() int32 { - var any liveRegMask - for _, live := range lv.regMaps { - any |= live - } - i := int32(0) - for any != 0 { - any >>= 1 - i++ - } - return i -} - // Generates live pointer value maps for arguments and local variables. The // this argument and the in arguments are always assumed live. The vars // argument is a slice of *Nodes. @@ -851,31 +677,16 @@ func (lv *Liveness) markUnsafePoints() { // particular, call Values can have a stack map in case the callee // grows the stack, but not themselves be a safe-point. func (lv *Liveness) hasStackMap(v *ssa.Value) bool { - // The runtime only has safe-points in function prologues, so - // we only need stack maps at call sites. go:nosplit functions - // are similar. - if go115ReduceLiveness || compiling_runtime || lv.f.NoSplit { - if !v.Op.IsCall() { - return false - } - // typedmemclr and typedmemmove are write barriers and - // deeply non-preemptible. They are unsafe points and - // hence should not have liveness maps. - if sym, _ := v.Aux.(*obj.LSym); sym == typedmemclr || sym == typedmemmove { - return false - } - return true - } - - switch v.Op { - case ssa.OpInitMem, ssa.OpArg, ssa.OpSP, ssa.OpSB, - ssa.OpSelect0, ssa.OpSelect1, ssa.OpGetG, - ssa.OpVarDef, ssa.OpVarLive, ssa.OpKeepAlive, - ssa.OpPhi: - // These don't produce code (see genssa). + if !v.Op.IsCall() { return false } - return !lv.unsafePoints.Get(int32(v.ID)) + // typedmemclr and typedmemmove are write barriers and + // deeply non-preemptible. They are unsafe points and + // hence should not have liveness maps. + if sym, ok := v.Aux.(*ssa.AuxCall); ok && (sym.Fn == typedmemclr || sym.Fn == typedmemmove) { + return false + } + return true } // Initializes the sets for solving the live variables. Visits all the @@ -891,17 +702,13 @@ func (lv *Liveness) prologue() { // effects with the each prog effects. for j := len(b.Values) - 1; j >= 0; j-- { pos, e := lv.valueEffects(b.Values[j]) - regUevar, regKill := lv.regEffects(b.Values[j]) if e&varkill != 0 { - be.varkill.vars.Set(pos) - be.uevar.vars.Unset(pos) + be.varkill.Set(pos) + be.uevar.Unset(pos) } - be.varkill.regs |= regKill - be.uevar.regs &^= regKill if e&uevar != 0 { - be.uevar.vars.Set(pos) + be.uevar.Set(pos) } - be.uevar.regs |= regUevar } } } @@ -911,8 +718,8 @@ func (lv *Liveness) solve() { // These temporary bitvectors exist to avoid successive allocations and // frees within the loop. nvars := int32(len(lv.vars)) - newlivein := varRegVec{vars: bvalloc(nvars)} - newliveout := varRegVec{vars: bvalloc(nvars)} + newlivein := bvalloc(nvars) + newliveout := bvalloc(nvars) // Walk blocks in postorder ordering. This improves convergence. po := lv.f.Postorder() @@ -930,11 +737,11 @@ func (lv *Liveness) solve() { switch b.Kind { case ssa.BlockRet: for _, pos := range lv.cache.retuevar { - newliveout.vars.Set(pos) + newliveout.Set(pos) } case ssa.BlockRetJmp: for _, pos := range lv.cache.tailuevar { - newliveout.vars.Set(pos) + newliveout.Set(pos) } case ssa.BlockExit: // panic exit - nothing to do @@ -969,7 +776,7 @@ func (lv *Liveness) solve() { // variables at each safe point locations. func (lv *Liveness) epilogue() { nvars := int32(len(lv.vars)) - liveout := varRegVec{vars: bvalloc(nvars)} + liveout := bvalloc(nvars) livedefer := bvalloc(nvars) // always-live variables // If there is a defer (that could recover), then all output @@ -1025,12 +832,11 @@ func (lv *Liveness) epilogue() { { // Reserve an entry for function entry. live := bvalloc(nvars) - lv.livevars = append(lv.livevars, varRegVec{vars: live}) + lv.livevars = append(lv.livevars, live) } for _, b := range lv.f.Blocks { be := lv.blockEffects(b) - firstBitmapIndex := len(lv.livevars) // Walk forward through the basic block instructions and // allocate liveness maps for those instructions that need them. @@ -1040,7 +846,7 @@ func (lv *Liveness) epilogue() { } live := bvalloc(nvars) - lv.livevars = append(lv.livevars, varRegVec{vars: live}) + lv.livevars = append(lv.livevars, live) } // walk backward, construct maps at each safe point @@ -1056,21 +862,18 @@ func (lv *Liveness) epilogue() { live := &lv.livevars[index] live.Or(*live, liveout) - live.vars.Or(live.vars, livedefer) // only for non-entry safe points + live.Or(*live, livedefer) // only for non-entry safe points index-- } // Update liveness information. pos, e := lv.valueEffects(v) - regUevar, regKill := lv.regEffects(v) if e&varkill != 0 { - liveout.vars.Unset(pos) + liveout.Unset(pos) } - liveout.regs &^= regKill if e&uevar != 0 { - liveout.vars.Set(pos) + liveout.Set(pos) } - liveout.regs |= regUevar } if b == lv.f.Entry { @@ -1080,7 +883,7 @@ func (lv *Liveness) epilogue() { // Check to make sure only input variables are live. for i, n := range lv.vars { - if !liveout.vars.Get(int32(i)) { + if !liveout.Get(int32(i)) { continue } if n.Class() == PPARAM { @@ -1094,32 +897,16 @@ func (lv *Liveness) epilogue() { live.Or(*live, liveout) } - // Check that no registers are live across calls. - // For closure calls, the CALLclosure is the last use - // of the context register, so it's dead after the call. - index = int32(firstBitmapIndex) - for _, v := range b.Values { - if lv.hasStackMap(v) { - live := lv.livevars[index] - if v.Op.IsCall() && live.regs != 0 { - lv.printDebug() - v.Fatalf("%v register %s recorded as live at call", lv.fn.Func.Nname, live.regs.niceString(lv.f.Config)) - } - index++ - } - } - // The liveness maps for this block are now complete. Compact them. lv.compact(b) } // If we have an open-coded deferreturn call, make a liveness map for it. if lv.fn.Func.OpenCodedDeferDisallowed() { - lv.livenessMap.deferreturn = LivenessInvalid + lv.livenessMap.deferreturn = LivenessDontCare } else { lv.livenessMap.deferreturn = LivenessIndex{ stackMapIndex: lv.stackMapSet.add(livedefer), - regMapIndex: 0, // entry regMap, containing no live registers isUnsafePoint: false, } } @@ -1136,20 +923,10 @@ func (lv *Liveness) epilogue() { lv.f.Fatalf("%v %L recorded as live on entry", lv.fn.Func.Nname, n) } } - if !go115ReduceLiveness { - // Check that no registers are live at function entry. - // The context register, if any, comes from a - // LoweredGetClosurePtr operation first thing in the function, - // so it doesn't appear live at entry. - if regs := lv.regMaps[0]; regs != 0 { - lv.printDebug() - lv.f.Fatalf("%v register %s recorded as live on entry", lv.fn.Func.Nname, regs.niceString(lv.f.Config)) - } - } } // Compact coalesces identical bitmaps from lv.livevars into the sets -// lv.stackMapSet and lv.regMaps. +// lv.stackMapSet. // // Compact clears lv.livevars. // @@ -1165,45 +942,23 @@ func (lv *Liveness) epilogue() { // PCDATA tables cost about 100k. So for now we keep using a single index for // both bitmap lists. func (lv *Liveness) compact(b *ssa.Block) { - add := func(live varRegVec, isUnsafePoint bool) LivenessIndex { // only if !go115ReduceLiveness - // Deduplicate the stack map. - stackIndex := lv.stackMapSet.add(live.vars) - // Deduplicate the register map. - regIndex, ok := lv.regMapSet[live.regs] - if !ok { - regIndex = len(lv.regMapSet) - lv.regMapSet[live.regs] = regIndex - lv.regMaps = append(lv.regMaps, live.regs) - } - return LivenessIndex{stackIndex, regIndex, isUnsafePoint} - } pos := 0 if b == lv.f.Entry { // Handle entry stack map. - if !go115ReduceLiveness { - add(lv.livevars[0], false) - } else { - lv.stackMapSet.add(lv.livevars[0].vars) - } + lv.stackMapSet.add(lv.livevars[0]) pos++ } for _, v := range b.Values { - if go115ReduceLiveness { - hasStackMap := lv.hasStackMap(v) - isUnsafePoint := lv.allUnsafe || lv.unsafePoints.Get(int32(v.ID)) - idx := LivenessIndex{StackMapDontCare, StackMapDontCare, isUnsafePoint} - if hasStackMap { - idx.stackMapIndex = lv.stackMapSet.add(lv.livevars[pos].vars) - pos++ - } - if hasStackMap || isUnsafePoint { - lv.livenessMap.set(v, idx) - } - } else if lv.hasStackMap(v) { - isUnsafePoint := lv.allUnsafe || lv.unsafePoints.Get(int32(v.ID)) - lv.livenessMap.set(v, add(lv.livevars[pos], isUnsafePoint)) + hasStackMap := lv.hasStackMap(v) + isUnsafePoint := lv.allUnsafe || lv.unsafePoints.Get(int32(v.ID)) + idx := LivenessIndex{StackMapDontCare, isUnsafePoint} + if hasStackMap { + idx.stackMapIndex = lv.stackMapSet.add(lv.livevars[pos]) pos++ } + if hasStackMap || isUnsafePoint { + lv.livenessMap.set(v, idx) + } } // Reset livevars. @@ -1231,8 +986,8 @@ func (lv *Liveness) showlive(v *ssa.Value, live bvec) { s := "live at " if v == nil { s += fmt.Sprintf("entry to %s:", lv.fn.funcname()) - } else if sym, ok := v.Aux.(*obj.LSym); ok { - fn := sym.Name + } else if sym, ok := v.Aux.(*ssa.AuxCall); ok && sym.Fn != nil { + fn := sym.Fn.Name if pos := strings.Index(fn, "."); pos >= 0 { fn = fn[pos+1:] } @@ -1250,8 +1005,8 @@ func (lv *Liveness) showlive(v *ssa.Value, live bvec) { Warnl(pos, s) } -func (lv *Liveness) printbvec(printed bool, name string, live varRegVec) bool { - if live.vars.IsEmpty() && live.regs == 0 { +func (lv *Liveness) printbvec(printed bool, name string, live bvec) bool { + if live.IsEmpty() { return printed } @@ -1264,19 +1019,18 @@ func (lv *Liveness) printbvec(printed bool, name string, live varRegVec) bool { comma := "" for i, n := range lv.vars { - if !live.vars.Get(int32(i)) { + if !live.Get(int32(i)) { continue } fmt.Printf("%s%s", comma, n.Sym.Name) comma = "," } - fmt.Printf("%s%s", comma, live.regs.niceString(lv.f.Config)) return true } -// printeffect is like printbvec, but for valueEffects and regEffects. -func (lv *Liveness) printeffect(printed bool, name string, pos int32, x bool, regMask liveRegMask) bool { - if !x && regMask == 0 { +// printeffect is like printbvec, but for valueEffects. +func (lv *Liveness) printeffect(printed bool, name string, pos int32, x bool) bool { + if !x { return printed } if !printed { @@ -1288,15 +1042,7 @@ func (lv *Liveness) printeffect(printed bool, name string, pos int32, x bool, re if x { fmt.Printf("%s", lv.vars[pos].Sym.Name) } - for j, reg := range lv.f.Config.GCRegMap { - if regMask&(1< 32 { - // Our uint32 conversion below won't work. - Fatalf("GP registers overflow uint32") - } - - if regs.n > 0 { - for _, live := range lv.regMaps { - regs.Clear() - regs.b[0] = uint32(live) - roff = dbvec(®sSymTmp, roff, regs) - } - } - } - // Give these LSyms content-addressable names, // so that they can be de-duplicated. // This provides significant binary size savings. @@ -1502,11 +1219,7 @@ func (lv *Liveness) emit() (argsSym, liveSym, regsSym *obj.LSym) { lsym.Set(obj.AttrContentAddressable, true) }) } - if !go115ReduceLiveness { - return makeSym(&argsSymTmp), makeSym(&liveSymTmp), makeSym(®sSymTmp) - } - // TODO(go115ReduceLiveness): Remove regsSym result - return makeSym(&argsSymTmp), makeSym(&liveSymTmp), nil + return makeSym(&argsSymTmp), makeSym(&liveSymTmp) } // Entry pointer for liveness analysis. Solves for the liveness of @@ -1552,27 +1265,20 @@ func liveness(e *ssafn, f *ssa.Func, pp *Progs) LivenessMap { // Emit the live pointer map data structures ls := e.curfn.Func.lsym - ls.Func.GCArgs, ls.Func.GCLocals, ls.Func.GCRegs = lv.emit() + fninfo := ls.Func() + fninfo.GCArgs, fninfo.GCLocals = lv.emit() p := pp.Prog(obj.AFUNCDATA) Addrconst(&p.From, objabi.FUNCDATA_ArgsPointerMaps) p.To.Type = obj.TYPE_MEM p.To.Name = obj.NAME_EXTERN - p.To.Sym = ls.Func.GCArgs + p.To.Sym = fninfo.GCArgs p = pp.Prog(obj.AFUNCDATA) Addrconst(&p.From, objabi.FUNCDATA_LocalsPointerMaps) p.To.Type = obj.TYPE_MEM p.To.Name = obj.NAME_EXTERN - p.To.Sym = ls.Func.GCLocals - - if !go115ReduceLiveness { - p = pp.Prog(obj.AFUNCDATA) - Addrconst(&p.From, objabi.FUNCDATA_RegPointerMaps) - p.To.Type = obj.TYPE_MEM - p.To.Name = obj.NAME_EXTERN - p.To.Sym = ls.Func.GCRegs - } + p.To.Sym = fninfo.GCLocals return lv.livenessMap } diff --git a/src/cmd/compile/internal/gc/range.go b/src/cmd/compile/internal/gc/range.go index 5434b0167aa..1b4d765d423 100644 --- a/src/cmd/compile/internal/gc/range.go +++ b/src/cmd/compile/internal/gc/range.go @@ -112,12 +112,13 @@ func typecheckrangeExpr(n *Node) { v2 = nil } - var why string if v1 != nil { if v1.Name != nil && v1.Name.Defn == n { v1.Type = t1 - } else if v1.Type != nil && assignop(t1, v1.Type, &why) == 0 { - yyerrorl(n.Pos, "cannot assign type %v to %L in range%s", t1, v1, why) + } else if v1.Type != nil { + if op, why := assignop(t1, v1.Type); op == OXXX { + yyerrorl(n.Pos, "cannot assign type %v to %L in range%s", t1, v1, why) + } } checkassign(n, v1) } @@ -125,8 +126,10 @@ func typecheckrangeExpr(n *Node) { if v2 != nil { if v2.Name != nil && v2.Name.Defn == n { v2.Type = t2 - } else if v2.Type != nil && assignop(t2, v2.Type, &why) == 0 { - yyerrorl(n.Pos, "cannot assign type %v to %L in range%s", t2, v2, why) + } else if v2.Type != nil { + if op, why := assignop(t2, v2.Type); op == OXXX { + yyerrorl(n.Pos, "cannot assign type %v to %L in range%s", t2, v2, why) + } } checkassign(n, v2) } @@ -463,7 +466,7 @@ func walkrange(n *Node) *Node { // // where == for keys of map m is reflexive. func isMapClear(n *Node) bool { - if Debug['N'] != 0 || instrumenting { + if Debug.N != 0 || instrumenting { return false } @@ -530,7 +533,7 @@ func mapClear(m *Node) *Node { // // Parameters are as in walkrange: "for v1, v2 = range a". func arrayClear(n, v1, v2, a *Node) bool { - if Debug['N'] != 0 || instrumenting { + if Debug.N != 0 || instrumenting { return false } diff --git a/src/cmd/compile/internal/gc/reflect.go b/src/cmd/compile/internal/gc/reflect.go index 49b2a0ed49a..9401eba7a50 100644 --- a/src/cmd/compile/internal/gc/reflect.go +++ b/src/cmd/compile/internal/gc/reflect.go @@ -61,8 +61,9 @@ const ( MAXELEMSIZE = 128 ) -func structfieldSize() int { return 3 * Widthptr } // Sizeof(runtime.structfield{}) -func imethodSize() int { return 4 + 4 } // Sizeof(runtime.imethod{}) +func structfieldSize() int { return 3 * Widthptr } // Sizeof(runtime.structfield{}) +func imethodSize() int { return 4 + 4 } // Sizeof(runtime.imethod{}) +func commonSize() int { return 4*Widthptr + 8 + 8 } // Sizeof(runtime._type{}) func uncommonSize(t *types.Type) int { // Sizeof(runtime.uncommontype{}) if t.Sym == nil && len(methods(t)) == 0 { @@ -510,6 +511,7 @@ func dimportpath(p *types.Pkg) { s := Ctxt.Lookup("type..importpath." + p.Prefix + ".") ot := dnameData(s, 0, str, "", nil, false) ggloblsym(s, int32(ot), obj.DUPOK|obj.RODATA) + s.Set(obj.AttrContentAddressable, true) p.Pathsym = s } @@ -637,6 +639,7 @@ func dname(name, tag string, pkg *types.Pkg, exported bool) *obj.LSym { } ot := dnameData(s, 0, name, tag, pkg, exported) ggloblsym(s, int32(ot), obj.DUPOK|obj.RODATA) + s.Set(obj.AttrContentAddressable, true) return s } @@ -1422,6 +1425,20 @@ func dtypesym(t *types.Type) *obj.LSym { return lsym } +// ifaceMethodOffset returns the offset of the i-th method in the interface +// type descriptor, ityp. +func ifaceMethodOffset(ityp *types.Type, i int64) int64 { + // interface type descriptor layout is struct { + // _type // commonSize + // pkgpath // 1 word + // []imethod // 3 words (pointing to [...]imethod below) + // uncommontype // uncommonSize + // [...]imethod + // } + // The size of imethod is 8. + return int64(commonSize()+4*Widthptr+uncommonSize(ityp)) + i*8 +} + // for each itabEntry, gather the methods on // the concrete type that implement the interface func peekitabs() { @@ -1574,8 +1591,12 @@ func dumptabs() { // typ typeOff // pointer to symbol // } nsym := dname(p.s.Name, "", nil, true) + tsym := dtypesym(p.t) ot = dsymptrOff(s, ot, nsym) - ot = dsymptrOff(s, ot, dtypesym(p.t)) + ot = dsymptrOff(s, ot, tsym) + // Plugin exports symbols as interfaces. Mark their types + // as UsedInIface. + tsym.Set(obj.AttrUsedInIface, true) } ggloblsym(s, int32(ot), int16(obj.RODATA)) diff --git a/src/cmd/compile/internal/gc/scc.go b/src/cmd/compile/internal/gc/scc.go index 60e0a9b8b51..5c7935aa876 100644 --- a/src/cmd/compile/internal/gc/scc.go +++ b/src/cmd/compile/internal/gc/scc.go @@ -75,8 +75,19 @@ func (v *bottomUpVisitor) visit(n *Node) uint32 { inspectList(n.Nbody, func(n *Node) bool { switch n.Op { - case OCALLFUNC, OCALLMETH: - fn := asNode(n.Left.Type.Nname()) + case ONAME: + if n.Class() == PFUNC { + if n.isMethodExpression() { + n = asNode(n.Type.Nname()) + } + if n != nil && n.Name.Defn != nil { + if m := v.visit(n.Name.Defn); m < min { + min = m + } + } + } + case ODOTMETH: + fn := asNode(n.Type.Nname()) if fn != nil && fn.Op == ONAME && fn.Class() == PFUNC && fn.Name.Defn != nil { if m := v.visit(fn.Name.Defn); m < min { min = m diff --git a/src/cmd/compile/internal/gc/scope.go b/src/cmd/compile/internal/gc/scope.go index d7239d56930..e66b859e100 100644 --- a/src/cmd/compile/internal/gc/scope.go +++ b/src/cmd/compile/internal/gc/scope.go @@ -62,9 +62,9 @@ func scopePCs(fnsym *obj.LSym, marks []Mark, dwarfScopes []dwarf.Scope) { if len(marks) == 0 { return } - p0 := fnsym.Func.Text + p0 := fnsym.Func().Text scope := findScope(marks, p0.Pos) - for p := fnsym.Func.Text; p != nil; p = p.Link { + for p := p0; p != nil; p = p.Link { if p.Pos == p0.Pos { continue } diff --git a/src/cmd/compile/internal/gc/sinit.go b/src/cmd/compile/internal/gc/sinit.go index 71ed5584612..212fcc022db 100644 --- a/src/cmd/compile/internal/gc/sinit.go +++ b/src/cmd/compile/internal/gc/sinit.go @@ -39,7 +39,7 @@ func (s *InitSchedule) append(n *Node) { // staticInit adds an initialization statement n to the schedule. func (s *InitSchedule) staticInit(n *Node) { if !s.tryStaticInit(n) { - if Debug['%'] != 0 { + if Debug.P != 0 { Dump("nonstatic", n) } s.append(n) @@ -128,7 +128,7 @@ func (s *InitSchedule) staticcopy(l *Node, r *Node) bool { case OSLICELIT: // copy slice a := s.inittemps[r] - slicesym(l, a, r.Right.Int64()) + slicesym(l, a, r.Right.Int64Val()) return true case OARRAYLIT, OSTRUCTLIT: @@ -205,7 +205,7 @@ func (s *InitSchedule) staticassign(l *Node, r *Node) bool { case OSTR2BYTES: if l.Class() == PEXTERN && r.Left.Op == OLITERAL { - sval := strlit(r.Left) + sval := r.Left.StringVal() slicebytes(l, sval) return true } @@ -213,7 +213,7 @@ func (s *InitSchedule) staticassign(l *Node, r *Node) bool { case OSLICELIT: s.initplan(r) // Init slice. - bound := r.Right.Int64() + bound := r.Right.Int64Val() ta := types.NewArray(r.Type.Elem(), bound) ta.SetNoalg(true) a := staticname(ta) @@ -278,7 +278,7 @@ func (s *InitSchedule) staticassign(l *Node, r *Node) bool { return Isconst(val, CTNIL) } - markTypeUsedInInterface(val.Type) + markTypeUsedInInterface(val.Type, l.Sym.Linksym()) var itab *Node if l.Type.IsEmptyInterface() { @@ -375,11 +375,6 @@ func readonlystaticname(t *types.Type) *Node { return n } -func isLiteral(n *Node) bool { - // Treat nils as zeros rather than literals. - return n.Op == OLITERAL && n.Val().Ctype() != CTNIL -} - func (n *Node) isSimpleName() bool { return n.Op == ONAME && n.Class() != PAUTOHEAP && n.Class() != PEXTERN } @@ -404,7 +399,7 @@ const ( func getdyn(n *Node, top bool) initGenType { switch n.Op { default: - if isLiteral(n) { + if n.isGoConst() { return initConst } return initDynamic @@ -413,7 +408,7 @@ func getdyn(n *Node, top bool) initGenType { if !top { return initDynamic } - if n.Right.Int64()/4 > int64(n.List.Len()) { + if n.Right.Int64Val()/4 > int64(n.List.Len()) { // <25% of entries have explicit values. // Very rough estimation, it takes 4 bytes of instructions // to initialize 1 byte of result. So don't use a static @@ -559,7 +554,7 @@ func fixedlit(ctxt initContext, kind initKind, n *Node, var_ *Node, init *Nodes) continue } - islit := isLiteral(value) + islit := value.isGoConst() if (kind == initKindStatic && !islit) || (kind == initKindDynamic && islit) { continue } @@ -589,12 +584,12 @@ func isSmallSliceLit(n *Node) bool { r := n.Right - return smallintconst(r) && (n.Type.Elem().Width == 0 || r.Int64() <= smallArrayBytes/n.Type.Elem().Width) + return smallintconst(r) && (n.Type.Elem().Width == 0 || r.Int64Val() <= smallArrayBytes/n.Type.Elem().Width) } func slicelit(ctxt initContext, n *Node, var_ *Node, init *Nodes) { // make an array type corresponding the number of elements we have - t := types.NewArray(n.Type.Elem(), n.Right.Int64()) + t := types.NewArray(n.Type.Elem(), n.Right.Int64Val()) dowidth(t) if ctxt == inNonInitFunction { @@ -732,7 +727,7 @@ func slicelit(ctxt initContext, n *Node, var_ *Node, init *Nodes) { continue } - if vstat != nil && isLiteral(value) { // already set by copy from static value + if vstat != nil && value.isGoConst() { // already set by copy from static value continue } @@ -993,7 +988,7 @@ func oaslit(n *Node, init *Nodes) bool { func getlit(lit *Node) int { if smallintconst(lit) { - return int(lit.Int64()) + return int(lit.Int64Val()) } return -1 } diff --git a/src/cmd/compile/internal/gc/ssa.go b/src/cmd/compile/internal/gc/ssa.go index 52083d999e5..5b74754b532 100644 --- a/src/cmd/compile/internal/gc/ssa.go +++ b/src/cmd/compile/internal/gc/ssa.go @@ -9,6 +9,7 @@ import ( "fmt" "html" "os" + "path/filepath" "sort" "bufio" @@ -26,6 +27,7 @@ var ssaConfig *ssa.Config var ssaCaches []ssa.Cache var ssaDump string // early copy of $GOSSAFUNC; the func name to dump output for +var ssaDir string // optional destination for ssa dump file var ssaDumpStdout bool // whether to dump to stdout var ssaDumpCFG string // generate CFGs for these phases const ssaDumpFile = "ssa.html" @@ -48,21 +50,16 @@ func initssaconfig() { // Caching is disabled in the backend, so generating these here avoids allocations. _ = types.NewPtr(types.Types[TINTER]) // *interface{} _ = types.NewPtr(types.NewPtr(types.Types[TSTRING])) // **string - _ = types.NewPtr(types.NewPtr(types.Idealstring)) // **string _ = types.NewPtr(types.NewSlice(types.Types[TINTER])) // *[]interface{} _ = types.NewPtr(types.NewPtr(types.Bytetype)) // **byte _ = types.NewPtr(types.NewSlice(types.Bytetype)) // *[]byte _ = types.NewPtr(types.NewSlice(types.Types[TSTRING])) // *[]string - _ = types.NewPtr(types.NewSlice(types.Idealstring)) // *[]string _ = types.NewPtr(types.NewPtr(types.NewPtr(types.Types[TUINT8]))) // ***uint8 _ = types.NewPtr(types.Types[TINT16]) // *int16 _ = types.NewPtr(types.Types[TINT64]) // *int64 _ = types.NewPtr(types.Errortype) // *error types.NewPtrCacheEnabled = false - ssaConfig = ssa.NewConfig(thearch.LinkArch.Name, *types_, Ctxt, Debug['N'] == 0) - if thearch.LinkArch.Name == "386" { - ssaConfig.Set387(thearch.Use387) - } + ssaConfig = ssa.NewConfig(thearch.LinkArch.Name, *types_, Ctxt, Debug.N == 0) ssaConfig.SoftFloat = thearch.SoftFloat ssaConfig.Race = flag_race ssaCaches = make([]ssa.Cache, nBackendWorkers) @@ -75,13 +72,14 @@ func initssaconfig() { deferproc = sysfunc("deferproc") deferprocStack = sysfunc("deferprocStack") Deferreturn = sysfunc("deferreturn") - Duffcopy = sysvar("duffcopy") // asm func with special ABI - Duffzero = sysvar("duffzero") // asm func with special ABI - gcWriteBarrier = sysvar("gcWriteBarrier") // asm func with special ABI + Duffcopy = sysfunc("duffcopy") + Duffzero = sysfunc("duffzero") + gcWriteBarrier = sysfunc("gcWriteBarrier") goschedguarded = sysfunc("goschedguarded") growslice = sysfunc("growslice") msanread = sysfunc("msanread") msanwrite = sysfunc("msanwrite") + msanmove = sysfunc("msanmove") newobject = sysfunc("newobject") newproc = sysfunc("newproc") panicdivide = sysfunc("panicdivide") @@ -108,51 +106,51 @@ func initssaconfig() { // asm funcs with special ABI if thearch.LinkArch.Name == "amd64" { GCWriteBarrierReg = map[int16]*obj.LSym{ - x86.REG_AX: sysvar("gcWriteBarrier"), - x86.REG_CX: sysvar("gcWriteBarrierCX"), - x86.REG_DX: sysvar("gcWriteBarrierDX"), - x86.REG_BX: sysvar("gcWriteBarrierBX"), - x86.REG_BP: sysvar("gcWriteBarrierBP"), - x86.REG_SI: sysvar("gcWriteBarrierSI"), - x86.REG_R8: sysvar("gcWriteBarrierR8"), - x86.REG_R9: sysvar("gcWriteBarrierR9"), + x86.REG_AX: sysfunc("gcWriteBarrier"), + x86.REG_CX: sysfunc("gcWriteBarrierCX"), + x86.REG_DX: sysfunc("gcWriteBarrierDX"), + x86.REG_BX: sysfunc("gcWriteBarrierBX"), + x86.REG_BP: sysfunc("gcWriteBarrierBP"), + x86.REG_SI: sysfunc("gcWriteBarrierSI"), + x86.REG_R8: sysfunc("gcWriteBarrierR8"), + x86.REG_R9: sysfunc("gcWriteBarrierR9"), } } if thearch.LinkArch.Family == sys.Wasm { - BoundsCheckFunc[ssa.BoundsIndex] = sysvar("goPanicIndex") - BoundsCheckFunc[ssa.BoundsIndexU] = sysvar("goPanicIndexU") - BoundsCheckFunc[ssa.BoundsSliceAlen] = sysvar("goPanicSliceAlen") - BoundsCheckFunc[ssa.BoundsSliceAlenU] = sysvar("goPanicSliceAlenU") - BoundsCheckFunc[ssa.BoundsSliceAcap] = sysvar("goPanicSliceAcap") - BoundsCheckFunc[ssa.BoundsSliceAcapU] = sysvar("goPanicSliceAcapU") - BoundsCheckFunc[ssa.BoundsSliceB] = sysvar("goPanicSliceB") - BoundsCheckFunc[ssa.BoundsSliceBU] = sysvar("goPanicSliceBU") - BoundsCheckFunc[ssa.BoundsSlice3Alen] = sysvar("goPanicSlice3Alen") - BoundsCheckFunc[ssa.BoundsSlice3AlenU] = sysvar("goPanicSlice3AlenU") - BoundsCheckFunc[ssa.BoundsSlice3Acap] = sysvar("goPanicSlice3Acap") - BoundsCheckFunc[ssa.BoundsSlice3AcapU] = sysvar("goPanicSlice3AcapU") - BoundsCheckFunc[ssa.BoundsSlice3B] = sysvar("goPanicSlice3B") - BoundsCheckFunc[ssa.BoundsSlice3BU] = sysvar("goPanicSlice3BU") - BoundsCheckFunc[ssa.BoundsSlice3C] = sysvar("goPanicSlice3C") - BoundsCheckFunc[ssa.BoundsSlice3CU] = sysvar("goPanicSlice3CU") + BoundsCheckFunc[ssa.BoundsIndex] = sysfunc("goPanicIndex") + BoundsCheckFunc[ssa.BoundsIndexU] = sysfunc("goPanicIndexU") + BoundsCheckFunc[ssa.BoundsSliceAlen] = sysfunc("goPanicSliceAlen") + BoundsCheckFunc[ssa.BoundsSliceAlenU] = sysfunc("goPanicSliceAlenU") + BoundsCheckFunc[ssa.BoundsSliceAcap] = sysfunc("goPanicSliceAcap") + BoundsCheckFunc[ssa.BoundsSliceAcapU] = sysfunc("goPanicSliceAcapU") + BoundsCheckFunc[ssa.BoundsSliceB] = sysfunc("goPanicSliceB") + BoundsCheckFunc[ssa.BoundsSliceBU] = sysfunc("goPanicSliceBU") + BoundsCheckFunc[ssa.BoundsSlice3Alen] = sysfunc("goPanicSlice3Alen") + BoundsCheckFunc[ssa.BoundsSlice3AlenU] = sysfunc("goPanicSlice3AlenU") + BoundsCheckFunc[ssa.BoundsSlice3Acap] = sysfunc("goPanicSlice3Acap") + BoundsCheckFunc[ssa.BoundsSlice3AcapU] = sysfunc("goPanicSlice3AcapU") + BoundsCheckFunc[ssa.BoundsSlice3B] = sysfunc("goPanicSlice3B") + BoundsCheckFunc[ssa.BoundsSlice3BU] = sysfunc("goPanicSlice3BU") + BoundsCheckFunc[ssa.BoundsSlice3C] = sysfunc("goPanicSlice3C") + BoundsCheckFunc[ssa.BoundsSlice3CU] = sysfunc("goPanicSlice3CU") } else { - BoundsCheckFunc[ssa.BoundsIndex] = sysvar("panicIndex") - BoundsCheckFunc[ssa.BoundsIndexU] = sysvar("panicIndexU") - BoundsCheckFunc[ssa.BoundsSliceAlen] = sysvar("panicSliceAlen") - BoundsCheckFunc[ssa.BoundsSliceAlenU] = sysvar("panicSliceAlenU") - BoundsCheckFunc[ssa.BoundsSliceAcap] = sysvar("panicSliceAcap") - BoundsCheckFunc[ssa.BoundsSliceAcapU] = sysvar("panicSliceAcapU") - BoundsCheckFunc[ssa.BoundsSliceB] = sysvar("panicSliceB") - BoundsCheckFunc[ssa.BoundsSliceBU] = sysvar("panicSliceBU") - BoundsCheckFunc[ssa.BoundsSlice3Alen] = sysvar("panicSlice3Alen") - BoundsCheckFunc[ssa.BoundsSlice3AlenU] = sysvar("panicSlice3AlenU") - BoundsCheckFunc[ssa.BoundsSlice3Acap] = sysvar("panicSlice3Acap") - BoundsCheckFunc[ssa.BoundsSlice3AcapU] = sysvar("panicSlice3AcapU") - BoundsCheckFunc[ssa.BoundsSlice3B] = sysvar("panicSlice3B") - BoundsCheckFunc[ssa.BoundsSlice3BU] = sysvar("panicSlice3BU") - BoundsCheckFunc[ssa.BoundsSlice3C] = sysvar("panicSlice3C") - BoundsCheckFunc[ssa.BoundsSlice3CU] = sysvar("panicSlice3CU") + BoundsCheckFunc[ssa.BoundsIndex] = sysfunc("panicIndex") + BoundsCheckFunc[ssa.BoundsIndexU] = sysfunc("panicIndexU") + BoundsCheckFunc[ssa.BoundsSliceAlen] = sysfunc("panicSliceAlen") + BoundsCheckFunc[ssa.BoundsSliceAlenU] = sysfunc("panicSliceAlenU") + BoundsCheckFunc[ssa.BoundsSliceAcap] = sysfunc("panicSliceAcap") + BoundsCheckFunc[ssa.BoundsSliceAcapU] = sysfunc("panicSliceAcapU") + BoundsCheckFunc[ssa.BoundsSliceB] = sysfunc("panicSliceB") + BoundsCheckFunc[ssa.BoundsSliceBU] = sysfunc("panicSliceBU") + BoundsCheckFunc[ssa.BoundsSlice3Alen] = sysfunc("panicSlice3Alen") + BoundsCheckFunc[ssa.BoundsSlice3AlenU] = sysfunc("panicSlice3AlenU") + BoundsCheckFunc[ssa.BoundsSlice3Acap] = sysfunc("panicSlice3Acap") + BoundsCheckFunc[ssa.BoundsSlice3AcapU] = sysfunc("panicSlice3AcapU") + BoundsCheckFunc[ssa.BoundsSlice3B] = sysfunc("panicSlice3B") + BoundsCheckFunc[ssa.BoundsSlice3BU] = sysfunc("panicSlice3BU") + BoundsCheckFunc[ssa.BoundsSlice3C] = sysfunc("panicSlice3C") + BoundsCheckFunc[ssa.BoundsSlice3CU] = sysfunc("panicSlice3CU") } if thearch.LinkArch.PtrSize == 4 { ExtendCheckFunc[ssa.BoundsIndex] = sysvar("panicExtendIndex") @@ -173,10 +171,6 @@ func initssaconfig() { ExtendCheckFunc[ssa.BoundsSlice3CU] = sysvar("panicExtendSlice3CU") } - // GO386=387 runtime definitions - ControlWord64trunc = sysvar("controlWord64trunc") // uint16 - ControlWord32 = sysvar("controlWord32") // uint16 - // Wasm (all asm funcs with special ABIs) WasmMove = sysvar("wasmMove") WasmZero = sysvar("wasmZero") @@ -247,7 +241,7 @@ func dvarint(x *obj.LSym, off int, v int64) int { // - Offset of where argument should be placed in the args frame when making call func (s *state) emitOpenDeferInfo() { x := Ctxt.Lookup(s.curfn.Func.lsym.Name + ".opendefer") - s.curfn.Func.lsym.Func.OpenCodedDeferInfo = x + s.curfn.Func.lsym.Func().OpenCodedDeferInfo = x off := 0 // Compute maxargsize (max size of arguments for all defers) @@ -346,7 +340,13 @@ func buildssa(fn *Node, worker int) *ssa.Func { s.f.Entry.Pos = fn.Pos if printssa { - s.f.HTMLWriter = ssa.NewHTMLWriter(ssaDumpFile, s.f, ssaDumpCFG) + ssaDF := ssaDumpFile + if ssaDir != "" { + ssaDF = filepath.Join(ssaDir, myimportpath+"."+name+".html") + ssaD := filepath.Dir(ssaDF) + os.MkdirAll(ssaD, 0755) + } + s.f.HTMLWriter = ssa.NewHTMLWriter(ssaDF, s.f, ssaDumpCFG) // TODO: generate and print a mapping from nodes to values and blocks dumpSourcesColumn(s.f.HTMLWriter, fn) s.f.HTMLWriter.WriteAST("AST", astBuf) @@ -358,7 +358,7 @@ func buildssa(fn *Node, worker int) *ssa.Func { s.fwdVars = map[*Node]*ssa.Value{} s.startmem = s.entryNewValue0(ssa.OpInitMem, types.TypeMem) - s.hasOpenDefers = Debug['N'] == 0 && s.hasdefer && !s.curfn.Func.OpenCodedDeferDisallowed() + s.hasOpenDefers = Debug.N == 0 && s.hasdefer && !s.curfn.Func.OpenCodedDeferDisallowed() switch { case s.hasOpenDefers && (Ctxt.Flag_shared || Ctxt.Flag_dynlink) && thearch.LinkArch.Name == "386": // Don't support open-coded defers for 386 ONLY when using shared @@ -410,11 +410,17 @@ func buildssa(fn *Node, worker int) *ssa.Func { // Generate addresses of local declarations s.decladdrs = map[*Node]*ssa.Value{} + var args []ssa.Param + var results []ssa.Param for _, n := range fn.Func.Dcl { switch n.Class() { - case PPARAM, PPARAMOUT: + case PPARAM: s.decladdrs[n] = s.entryNewValue2A(ssa.OpLocalAddr, types.NewPtr(n.Type), n, s.sp, s.startmem) - if n.Class() == PPARAMOUT && s.canSSA(n) { + args = append(args, ssa.Param{Type: n.Type, Offset: int32(n.Xoffset)}) + case PPARAMOUT: + s.decladdrs[n] = s.entryNewValue2A(ssa.OpLocalAddr, types.NewPtr(n.Type), n, s.sp, s.startmem) + results = append(results, ssa.Param{Type: n.Type, Offset: int32(n.Xoffset)}) + if s.canSSA(n) { // Save ssa-able PPARAMOUT variables so we can // store them back to the stack at the end of // the function. @@ -651,6 +657,8 @@ type state struct { lastDeferExit *ssa.Block // Entry block of last defer exit code we generated lastDeferFinalBlock *ssa.Block // Final block of last defer exit code we generated lastDeferCount int // Number of defers encountered at that point + + prevCall *ssa.Value // the previous call; use this to tie results to the call op. } type funcLine struct { @@ -740,7 +748,7 @@ func (s *state) pushLine(line src.XPos) { // the frontend may emit node with line number missing, // use the parent line number in this case. line = s.peekPos() - if Debug['K'] != 0 { + if Debug.K != 0 { Warn("buildssa: unknown position (line 0)") } } else { @@ -805,6 +813,11 @@ func (s *state) newValue2(op ssa.Op, t *types.Type, arg0, arg1 *ssa.Value) *ssa. return s.curBlock.NewValue2(s.peekPos(), op, t, arg0, arg1) } +// newValue2A adds a new value with two arguments and an aux value to the current block. +func (s *state) newValue2A(op ssa.Op, t *types.Type, aux interface{}, arg0, arg1 *ssa.Value) *ssa.Value { + return s.curBlock.NewValue2A(s.peekPos(), op, t, aux, arg0, arg1) +} + // newValue2Apos adds a new value with two arguments and an aux value to the current block. // isStmt determines whether the created values may be a statement or not // (i.e., false means never, yes means maybe). @@ -954,7 +967,45 @@ func (s *state) newValueOrSfCall2(op ssa.Op, t *types.Type, arg0, arg1 *ssa.Valu return s.newValue2(op, t, arg0, arg1) } -func (s *state) instrument(t *types.Type, addr *ssa.Value, wr bool) { +type instrumentKind uint8 + +const ( + instrumentRead = iota + instrumentWrite + instrumentMove +) + +func (s *state) instrument(t *types.Type, addr *ssa.Value, kind instrumentKind) { + s.instrument2(t, addr, nil, kind) +} + +// instrumentFields instruments a read/write operation on addr. +// If it is instrumenting for MSAN and t is a struct type, it instruments +// operation for each field, instead of for the whole struct. +func (s *state) instrumentFields(t *types.Type, addr *ssa.Value, kind instrumentKind) { + if !flag_msan || !t.IsStruct() { + s.instrument(t, addr, kind) + return + } + for _, f := range t.Fields().Slice() { + if f.Sym.IsBlank() { + continue + } + offptr := s.newValue1I(ssa.OpOffPtr, types.NewPtr(f.Type), f.Offset, addr) + s.instrumentFields(f.Type, offptr, kind) + } +} + +func (s *state) instrumentMove(t *types.Type, dst, src *ssa.Value) { + if flag_msan { + s.instrument2(t, dst, src, instrumentMove) + } else { + s.instrument(t, src, instrumentRead) + s.instrument(t, dst, instrumentWrite) + } +} + +func (s *state) instrument2(t *types.Type, addr, addr2 *ssa.Value, kind instrumentKind) { if !s.curfn.Func.InstrumentBody() { return } @@ -971,33 +1022,54 @@ func (s *state) instrument(t *types.Type, addr *ssa.Value, wr bool) { var fn *obj.LSym needWidth := false + if addr2 != nil && kind != instrumentMove { + panic("instrument2: non-nil addr2 for non-move instrumentation") + } + if flag_msan { - fn = msanread - if wr { + switch kind { + case instrumentRead: + fn = msanread + case instrumentWrite: fn = msanwrite + case instrumentMove: + fn = msanmove + default: + panic("unreachable") } needWidth = true } else if flag_race && t.NumComponents(types.CountBlankFields) > 1 { // for composite objects we have to write every address // because a write might happen to any subobject. // composites with only one element don't have subobjects, though. - fn = racereadrange - if wr { + switch kind { + case instrumentRead: + fn = racereadrange + case instrumentWrite: fn = racewriterange + default: + panic("unreachable") } needWidth = true } else if flag_race { // for non-composite objects we can write just the start // address, as any write must write the first byte. - fn = raceread - if wr { + switch kind { + case instrumentRead: + fn = raceread + case instrumentWrite: fn = racewrite + default: + panic("unreachable") } } else { panic("unreachable") } args := []*ssa.Value{addr} + if addr2 != nil { + args = append(args, addr2) + } if needWidth { args = append(args, s.constInt(types.Types[TUINTPTR], w)) } @@ -1005,7 +1077,7 @@ func (s *state) instrument(t *types.Type, addr *ssa.Value, wr bool) { } func (s *state) load(t *types.Type, src *ssa.Value) *ssa.Value { - s.instrument(t, src, false) + s.instrumentFields(t, src, instrumentRead) return s.rawLoad(t, src) } @@ -1018,15 +1090,14 @@ func (s *state) store(t *types.Type, dst, val *ssa.Value) { } func (s *state) zero(t *types.Type, dst *ssa.Value) { - s.instrument(t, dst, true) + s.instrument(t, dst, instrumentWrite) store := s.newValue2I(ssa.OpZero, types.TypeMem, t.Size(), dst, s.mem()) store.Aux = t s.vars[&memVar] = store } func (s *state) move(t *types.Type, dst, src *ssa.Value) { - s.instrument(t, src, false) - s.instrument(t, dst, true) + s.instrumentMove(t, dst, src) store := s.newValue3I(ssa.OpMove, types.TypeMem, t.Size(), dst, src, s.mem()) store.Aux = t s.vars[&memVar] = store @@ -1071,7 +1142,7 @@ func (s *state) stmt(n *Node) { fallthrough case OCALLMETH, OCALLINTER: - s.call(n, callNormal) + s.callResult(n, callNormal) if n.Op == OCALLFUNC && n.Left.Op == ONAME && n.Left.Class() == PFUNC { if fn := n.Left.Sym.Name; compiling_runtime && fn == "throw" || n.Left.Sym.Pkg == Runtimepkg && (fn == "throwinit" || fn == "gopanic" || fn == "panicwrap" || fn == "block" || fn == "panicmakeslicelen" || fn == "panicmakeslicecap") { @@ -1103,10 +1174,10 @@ func (s *state) stmt(n *Node) { if n.Esc == EscNever { d = callDeferStack } - s.call(n.Left, d) + s.callResult(n.Left, d) } case OGO: - s.call(n.Left, callGo) + s.callResult(n.Left, callGo) case OAS2DOTTYPE: res, resok := s.dottype(n.Right, true) @@ -1208,7 +1279,7 @@ func (s *state) stmt(n *Node) { // Check whether we're writing the result of an append back to the same slice. // If so, we handle it specially to avoid write barriers on the fast // (non-growth) path. - if !samesafeexpr(n.Left, rhs.List.First()) || Debug['N'] != 0 { + if !samesafeexpr(n.Left, rhs.List.First()) || Debug.N != 0 { break } // If the slice can be SSA'd, it'll be on the stack, @@ -1265,7 +1336,7 @@ func (s *state) stmt(n *Node) { // We're assigning a slicing operation back to its source. // Don't write back fields we aren't changing. See issue #14855. i, j, k := rhs.SliceBounds() - if i != nil && (i.Op == OLITERAL && i.Val().Ctype() == CTINT && i.Int64() == 0) { + if i != nil && (i.Op == OLITERAL && i.Val().Ctype() == CTINT && i.Int64Val() == 0) { // [0:...] is the same as [:...] i = nil } @@ -1295,7 +1366,7 @@ func (s *state) stmt(n *Node) { case OIF: if Isconst(n.Left, CTBOOL) { s.stmtList(n.Left.Ninit) - if n.Left.Bool() { + if n.Left.BoolVal() { s.stmtList(n.Nbody) } else { s.stmtList(n.Rlist) @@ -2113,7 +2184,7 @@ func (s *state) expr(n *Node) *ssa.Value { } // unsafe.Pointer <--> *T - if to.Etype == TUNSAFEPTR && from.IsPtrShaped() || from.Etype == TUNSAFEPTR && to.IsPtrShaped() { + if to.IsUnsafePtr() && from.IsPtrShaped() || from.IsUnsafePtr() && to.IsPtrShaped() { return v } @@ -2466,6 +2537,11 @@ func (s *state) expr(n *Node) *ssa.Value { a := s.expr(n.Left) b := s.expr(n.Right) return s.newValue2(s.ssaOp(n.Op, n.Type), a.Type, a, b) + case OANDNOT: + a := s.expr(n.Left) + b := s.expr(n.Right) + b = s.newValue1(s.ssaOp(OBITNOT, b.Type), b.Type, b) + return s.newValue2(s.ssaOp(OAND, n.Type), a.Type, a, b) case OLSH, ORSH: a := s.expr(n.Left) b := s.expr(n.Right) @@ -2549,8 +2625,23 @@ func (s *state) expr(n *Node) *ssa.Value { return s.addr(n.Left) case ORESULT: - addr := s.constOffPtrSP(types.NewPtr(n.Type), n.Xoffset) - return s.load(n.Type, addr) + if s.prevCall == nil || s.prevCall.Op != ssa.OpStaticLECall && s.prevCall.Op != ssa.OpInterLECall && s.prevCall.Op != ssa.OpClosureLECall { + // Do the old thing + addr := s.constOffPtrSP(types.NewPtr(n.Type), n.Xoffset) + return s.rawLoad(n.Type, addr) + } + which := s.prevCall.Aux.(*ssa.AuxCall).ResultForOffset(n.Xoffset) + if which == -1 { + // Do the old thing // TODO: Panic instead. + addr := s.constOffPtrSP(types.NewPtr(n.Type), n.Xoffset) + return s.rawLoad(n.Type, addr) + } + if canSSAType(n.Type) { + return s.newValue1I(ssa.OpSelectN, n.Type, which, s.prevCall) + } else { + addr := s.newValue1I(ssa.OpSelectNAddr, types.NewPtr(n.Type), which, s.prevCall) + return s.rawLoad(n.Type, addr) + } case ODEREF: p := s.exprPtr(n.Left, n.Bounded(), n.Pos) @@ -2589,7 +2680,7 @@ func (s *state) expr(n *Node) *ssa.Value { // Replace "abc"[1] with 'b'. // Delayed until now because "abc"[1] is not an ideal constant. // See test/fixedbugs/issue11370.go. - return s.newValue0I(ssa.OpConst8, types.Types[TUINT8], int64(int8(strlit(n.Left)[n.Right.Int64()]))) + return s.newValue0I(ssa.OpConst8, types.Types[TUINT8], int64(int8(n.Left.StringVal()[n.Right.Int64Val()]))) } a := s.expr(n.Left) i := s.expr(n.Right) @@ -2598,7 +2689,7 @@ func (s *state) expr(n *Node) *ssa.Value { ptrtyp := s.f.Config.Types.BytePtr ptr := s.newValue1(ssa.OpStringPtr, ptrtyp, a) if Isconst(n.Right, CTINT) { - ptr = s.newValue1I(ssa.OpOffPtr, ptrtyp, n.Right.Int64(), ptr) + ptr = s.newValue1I(ssa.OpOffPtr, ptrtyp, n.Right.Int64Val(), ptr) } else { ptr = s.newValue2(ssa.OpAddPtr, ptrtyp, ptr, i) } @@ -2710,8 +2801,7 @@ func (s *state) expr(n *Node) *ssa.Value { fallthrough case OCALLINTER, OCALLMETH: - a := s.call(n, callNormal) - return s.load(n.Type, a) + return s.callResult(n, callNormal) case OGETG: return s.newValue1(ssa.OpGetG, n.Type, s.mem()) @@ -3369,6 +3459,13 @@ func init() { return s.newValue1(ssa.OpSelect0, types.Types[TUINT32], v) }, sys.PPC64, sys.S390X) + addF("runtime/internal/atomic", "LoadAcq64", + func(s *state, n *Node, args []*ssa.Value) *ssa.Value { + v := s.newValue2(ssa.OpAtomicLoadAcq64, types.NewTuple(types.Types[TUINT64], types.TypeMem), args[0], s.mem()) + s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) + return s.newValue1(ssa.OpSelect0, types.Types[TUINT64], v) + }, + sys.PPC64) addF("runtime/internal/atomic", "Loadp", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { v := s.newValue2(ssa.OpAtomicLoadPtr, types.NewTuple(s.f.Config.Types.BytePtr, types.TypeMem), args[0], s.mem()) @@ -3407,6 +3504,12 @@ func init() { return nil }, sys.PPC64, sys.S390X) + addF("runtime/internal/atomic", "StoreRel64", + func(s *state, n *Node, args []*ssa.Value) *ssa.Value { + s.vars[&memVar] = s.newValue3(ssa.OpAtomicStoreRel64, types.TypeMem, args[0], args[1], s.mem()) + return nil + }, + sys.PPC64) addF("runtime/internal/atomic", "Xchg", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { @@ -3414,14 +3517,64 @@ func init() { s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) return s.newValue1(ssa.OpSelect0, types.Types[TUINT32], v) }, - sys.AMD64, sys.ARM64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) + sys.AMD64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) addF("runtime/internal/atomic", "Xchg64", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { v := s.newValue3(ssa.OpAtomicExchange64, types.NewTuple(types.Types[TUINT64], types.TypeMem), args[0], args[1], s.mem()) s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) return s.newValue1(ssa.OpSelect0, types.Types[TUINT64], v) }, - sys.AMD64, sys.ARM64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) + sys.AMD64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) + + type atomicOpEmitter func(s *state, n *Node, args []*ssa.Value, op ssa.Op, typ types.EType) + + makeAtomicGuardedIntrinsicARM64 := func(op0, op1 ssa.Op, typ, rtyp types.EType, emit atomicOpEmitter) intrinsicBuilder { + + return func(s *state, n *Node, args []*ssa.Value) *ssa.Value { + // Target Atomic feature is identified by dynamic detection + addr := s.entryNewValue1A(ssa.OpAddr, types.Types[TBOOL].PtrTo(), arm64HasATOMICS, s.sb) + v := s.load(types.Types[TBOOL], addr) + b := s.endBlock() + b.Kind = ssa.BlockIf + b.SetControl(v) + bTrue := s.f.NewBlock(ssa.BlockPlain) + bFalse := s.f.NewBlock(ssa.BlockPlain) + bEnd := s.f.NewBlock(ssa.BlockPlain) + b.AddEdgeTo(bTrue) + b.AddEdgeTo(bFalse) + b.Likely = ssa.BranchLikely + + // We have atomic instructions - use it directly. + s.startBlock(bTrue) + emit(s, n, args, op1, typ) + s.endBlock().AddEdgeTo(bEnd) + + // Use original instruction sequence. + s.startBlock(bFalse) + emit(s, n, args, op0, typ) + s.endBlock().AddEdgeTo(bEnd) + + // Merge results. + s.startBlock(bEnd) + if rtyp == TNIL { + return nil + } else { + return s.variable(n, types.Types[rtyp]) + } + } + } + + atomicXchgXaddEmitterARM64 := func(s *state, n *Node, args []*ssa.Value, op ssa.Op, typ types.EType) { + v := s.newValue3(op, types.NewTuple(types.Types[typ], types.TypeMem), args[0], args[1], s.mem()) + s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) + s.vars[n] = s.newValue1(ssa.OpSelect0, types.Types[typ], v) + } + addF("runtime/internal/atomic", "Xchg", + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicExchange32, ssa.OpAtomicExchange32Variant, TUINT32, TUINT32, atomicXchgXaddEmitterARM64), + sys.ARM64) + addF("runtime/internal/atomic", "Xchg64", + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicExchange64, ssa.OpAtomicExchange64Variant, TUINT64, TUINT64, atomicXchgXaddEmitterARM64), + sys.ARM64) addF("runtime/internal/atomic", "Xadd", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { @@ -3438,46 +3591,11 @@ func init() { }, sys.AMD64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) - makeXaddARM64 := func(op0 ssa.Op, op1 ssa.Op, ty types.EType) func(s *state, n *Node, args []*ssa.Value) *ssa.Value { - return func(s *state, n *Node, args []*ssa.Value) *ssa.Value { - // Target Atomic feature is identified by dynamic detection - addr := s.entryNewValue1A(ssa.OpAddr, types.Types[TBOOL].PtrTo(), arm64HasATOMICS, s.sb) - v := s.load(types.Types[TBOOL], addr) - b := s.endBlock() - b.Kind = ssa.BlockIf - b.SetControl(v) - bTrue := s.f.NewBlock(ssa.BlockPlain) - bFalse := s.f.NewBlock(ssa.BlockPlain) - bEnd := s.f.NewBlock(ssa.BlockPlain) - b.AddEdgeTo(bTrue) - b.AddEdgeTo(bFalse) - b.Likely = ssa.BranchUnlikely // most machines don't have Atomics nowadays - - // We have atomic instructions - use it directly. - s.startBlock(bTrue) - v0 := s.newValue3(op1, types.NewTuple(types.Types[ty], types.TypeMem), args[0], args[1], s.mem()) - s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v0) - s.vars[n] = s.newValue1(ssa.OpSelect0, types.Types[ty], v0) - s.endBlock().AddEdgeTo(bEnd) - - // Use original instruction sequence. - s.startBlock(bFalse) - v1 := s.newValue3(op0, types.NewTuple(types.Types[ty], types.TypeMem), args[0], args[1], s.mem()) - s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v1) - s.vars[n] = s.newValue1(ssa.OpSelect0, types.Types[ty], v1) - s.endBlock().AddEdgeTo(bEnd) - - // Merge results. - s.startBlock(bEnd) - return s.variable(n, types.Types[ty]) - } - } - addF("runtime/internal/atomic", "Xadd", - makeXaddARM64(ssa.OpAtomicAdd32, ssa.OpAtomicAdd32Variant, TUINT32), + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicAdd32, ssa.OpAtomicAdd32Variant, TUINT32, TUINT32, atomicXchgXaddEmitterARM64), sys.ARM64) addF("runtime/internal/atomic", "Xadd64", - makeXaddARM64(ssa.OpAtomicAdd64, ssa.OpAtomicAdd64Variant, TUINT64), + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicAdd64, ssa.OpAtomicAdd64Variant, TUINT64, TUINT64, atomicXchgXaddEmitterARM64), sys.ARM64) addF("runtime/internal/atomic", "Cas", @@ -3486,14 +3604,14 @@ func init() { s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) return s.newValue1(ssa.OpSelect0, types.Types[TBOOL], v) }, - sys.AMD64, sys.ARM64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) + sys.AMD64, sys.MIPS, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) addF("runtime/internal/atomic", "Cas64", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { v := s.newValue4(ssa.OpAtomicCompareAndSwap64, types.NewTuple(types.Types[TBOOL], types.TypeMem), args[0], args[1], args[2], s.mem()) s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) return s.newValue1(ssa.OpSelect0, types.Types[TBOOL], v) }, - sys.AMD64, sys.ARM64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) + sys.AMD64, sys.MIPS64, sys.PPC64, sys.RISCV64, sys.S390X) addF("runtime/internal/atomic", "CasRel", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { v := s.newValue4(ssa.OpAtomicCompareAndSwap32, types.NewTuple(types.Types[TBOOL], types.TypeMem), args[0], args[1], args[2], s.mem()) @@ -3502,18 +3620,60 @@ func init() { }, sys.PPC64) + atomicCasEmitterARM64 := func(s *state, n *Node, args []*ssa.Value, op ssa.Op, typ types.EType) { + v := s.newValue4(op, types.NewTuple(types.Types[TBOOL], types.TypeMem), args[0], args[1], args[2], s.mem()) + s.vars[&memVar] = s.newValue1(ssa.OpSelect1, types.TypeMem, v) + s.vars[n] = s.newValue1(ssa.OpSelect0, types.Types[typ], v) + } + + addF("runtime/internal/atomic", "Cas", + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicCompareAndSwap32, ssa.OpAtomicCompareAndSwap32Variant, TUINT32, TBOOL, atomicCasEmitterARM64), + sys.ARM64) + addF("runtime/internal/atomic", "Cas64", + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicCompareAndSwap64, ssa.OpAtomicCompareAndSwap64Variant, TUINT64, TBOOL, atomicCasEmitterARM64), + sys.ARM64) + addF("runtime/internal/atomic", "And8", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { s.vars[&memVar] = s.newValue3(ssa.OpAtomicAnd8, types.TypeMem, args[0], args[1], s.mem()) return nil }, - sys.AMD64, sys.ARM64, sys.MIPS, sys.PPC64, sys.S390X) + sys.AMD64, sys.MIPS, sys.PPC64, sys.S390X) + addF("runtime/internal/atomic", "And", + func(s *state, n *Node, args []*ssa.Value) *ssa.Value { + s.vars[&memVar] = s.newValue3(ssa.OpAtomicAnd32, types.TypeMem, args[0], args[1], s.mem()) + return nil + }, + sys.AMD64, sys.MIPS, sys.PPC64, sys.S390X) addF("runtime/internal/atomic", "Or8", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { s.vars[&memVar] = s.newValue3(ssa.OpAtomicOr8, types.TypeMem, args[0], args[1], s.mem()) return nil }, sys.AMD64, sys.ARM64, sys.MIPS, sys.PPC64, sys.S390X) + addF("runtime/internal/atomic", "Or", + func(s *state, n *Node, args []*ssa.Value) *ssa.Value { + s.vars[&memVar] = s.newValue3(ssa.OpAtomicOr32, types.TypeMem, args[0], args[1], s.mem()) + return nil + }, + sys.AMD64, sys.MIPS, sys.PPC64, sys.S390X) + + atomicAndOrEmitterARM64 := func(s *state, n *Node, args []*ssa.Value, op ssa.Op, typ types.EType) { + s.vars[&memVar] = s.newValue3(op, types.TypeMem, args[0], args[1], s.mem()) + } + + addF("runtime/internal/atomic", "And8", + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicAnd8, ssa.OpAtomicAnd8Variant, TNIL, TNIL, atomicAndOrEmitterARM64), + sys.ARM64) + addF("runtime/internal/atomic", "And", + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicAnd32, ssa.OpAtomicAnd32Variant, TNIL, TNIL, atomicAndOrEmitterARM64), + sys.ARM64) + addF("runtime/internal/atomic", "Or8", + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicOr8, ssa.OpAtomicOr8Variant, TNIL, TNIL, atomicAndOrEmitterARM64), + sys.ARM64) + addF("runtime/internal/atomic", "Or", + makeAtomicGuardedIntrinsicARM64(ssa.OpAtomicOr32, ssa.OpAtomicOr32Variant, TNIL, TNIL, atomicAndOrEmitterARM64), + sys.ARM64) alias("runtime/internal/atomic", "Loadint64", "runtime/internal/atomic", "Load64", all...) alias("runtime/internal/atomic", "Xaddint64", "runtime/internal/atomic", "Xadd64", all...) @@ -3522,9 +3682,19 @@ func init() { alias("runtime/internal/atomic", "Loaduintptr", "runtime/internal/atomic", "Load", p4...) alias("runtime/internal/atomic", "Loaduintptr", "runtime/internal/atomic", "Load64", p8...) alias("runtime/internal/atomic", "LoadAcq", "runtime/internal/atomic", "Load", lwatomics...) + alias("runtime/internal/atomic", "LoadAcq64", "runtime/internal/atomic", "Load64", lwatomics...) + alias("runtime/internal/atomic", "LoadAcquintptr", "runtime/internal/atomic", "LoadAcq", p4...) + alias("sync", "runtime_LoadAcquintptr", "runtime/internal/atomic", "LoadAcq", p4...) // linknamed + alias("runtime/internal/atomic", "LoadAcquintptr", "runtime/internal/atomic", "LoadAcq64", p8...) + alias("sync", "runtime_LoadAcquintptr", "runtime/internal/atomic", "LoadAcq64", p8...) // linknamed alias("runtime/internal/atomic", "Storeuintptr", "runtime/internal/atomic", "Store", p4...) alias("runtime/internal/atomic", "Storeuintptr", "runtime/internal/atomic", "Store64", p8...) alias("runtime/internal/atomic", "StoreRel", "runtime/internal/atomic", "Store", lwatomics...) + alias("runtime/internal/atomic", "StoreRel64", "runtime/internal/atomic", "Store64", lwatomics...) + alias("runtime/internal/atomic", "StoreReluintptr", "runtime/internal/atomic", "StoreRel", p4...) + alias("sync", "runtime_StoreReluintptr", "runtime/internal/atomic", "StoreRel", p4...) // linknamed + alias("runtime/internal/atomic", "StoreReluintptr", "runtime/internal/atomic", "StoreRel64", p8...) + alias("sync", "runtime_StoreReluintptr", "runtime/internal/atomic", "StoreRel64", p8...) // linknamed alias("runtime/internal/atomic", "Xchguintptr", "runtime/internal/atomic", "Xchg", p4...) alias("runtime/internal/atomic", "Xchguintptr", "runtime/internal/atomic", "Xchg64", p8...) alias("runtime/internal/atomic", "Xadduintptr", "runtime/internal/atomic", "Xadd", p4...) @@ -3584,8 +3754,7 @@ func init() { addF("math", "FMA", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { if !s.config.UseFMA { - a := s.call(n, callNormal) - s.vars[n] = s.load(types.Types[TFLOAT64], a) + s.vars[n] = s.callResult(n, callNormal) // types.Types[TFLOAT64] return s.variable(n, types.Types[TFLOAT64]) } v := s.entryNewValue0A(ssa.OpHasCPUFeature, types.Types[TBOOL], x86HasFMA) @@ -3606,8 +3775,7 @@ func init() { // Call the pure Go version. s.startBlock(bFalse) - a := s.call(n, callNormal) - s.vars[n] = s.load(types.Types[TFLOAT64], a) + s.vars[n] = s.callResult(n, callNormal) // types.Types[TFLOAT64] s.endBlock().AddEdgeTo(bEnd) // Merge results. @@ -3618,8 +3786,7 @@ func init() { addF("math", "FMA", func(s *state, n *Node, args []*ssa.Value) *ssa.Value { if !s.config.UseFMA { - a := s.call(n, callNormal) - s.vars[n] = s.load(types.Types[TFLOAT64], a) + s.vars[n] = s.callResult(n, callNormal) // types.Types[TFLOAT64] return s.variable(n, types.Types[TFLOAT64]) } addr := s.entryNewValue1A(ssa.OpAddr, types.Types[TBOOL].PtrTo(), armHasVFPv4, s.sb) @@ -3641,8 +3808,7 @@ func init() { // Call the pure Go version. s.startBlock(bFalse) - a := s.call(n, callNormal) - s.vars[n] = s.load(types.Types[TFLOAT64], a) + s.vars[n] = s.callResult(n, callNormal) // types.Types[TFLOAT64] s.endBlock().AddEdgeTo(bEnd) // Merge results. @@ -3671,8 +3837,7 @@ func init() { // Call the pure Go version. s.startBlock(bFalse) - a := s.call(n, callNormal) - s.vars[n] = s.load(types.Types[TFLOAT64], a) + s.vars[n] = s.callResult(n, callNormal) // types.Types[TFLOAT64] s.endBlock().AddEdgeTo(bEnd) // Merge results. @@ -3882,8 +4047,7 @@ func init() { // Call the pure Go version. s.startBlock(bFalse) - a := s.call(n, callNormal) - s.vars[n] = s.load(types.Types[TINT], a) + s.vars[n] = s.callResult(n, callNormal) // types.Types[TINT] s.endBlock().AddEdgeTo(bEnd) // Merge results. @@ -4006,11 +4170,6 @@ func init() { return s.newValue2(ssa.OpMul64uhilo, types.NewTuple(types.Types[TUINT64], types.Types[TUINT64]), args[0], args[1]) }, sys.ArchAMD64, sys.ArchARM64, sys.ArchPPC64LE, sys.ArchPPC64, sys.ArchS390X) - add("math/big", "divWW", - func(s *state, n *Node, args []*ssa.Value) *ssa.Value { - return s.newValue3(ssa.OpDiv128u, types.NewTuple(types.Types[TUINT64], types.Types[TUINT64]), args[0], args[1], args[2]) - }, - sys.ArchAMD64) } // findIntrinsic returns a function which builds the SSA equivalent of the @@ -4240,6 +4399,7 @@ func (s *state) openDeferExit() { s.lastDeferExit = deferExit s.lastDeferCount = len(s.openDefers) zeroval := s.constInt8(types.Types[TUINT8], 0) + testLateExpansion := ssa.LateCallExpansionEnabledWithin(s.f) // Test for and run defers in reverse order for i := len(s.openDefers) - 1; i >= 0; i-- { r := s.openDefers[i] @@ -4275,21 +4435,40 @@ func (s *state) openDeferExit() { argStart := Ctxt.FixedFrameSize() fn := r.n.Left stksize := fn.Type.ArgWidth() + var ACArgs []ssa.Param + var ACResults []ssa.Param + var callArgs []*ssa.Value if r.rcvr != nil { // rcvr in case of OCALLINTER v := s.load(r.rcvr.Type.Elem(), r.rcvr) addr := s.constOffPtrSP(s.f.Config.Types.UintptrPtr, argStart) - s.store(types.Types[TUINTPTR], addr, v) + ACArgs = append(ACArgs, ssa.Param{Type: types.Types[TUINTPTR], Offset: int32(argStart)}) + if testLateExpansion { + callArgs = append(callArgs, v) + } else { + s.store(types.Types[TUINTPTR], addr, v) + } } for j, argAddrVal := range r.argVals { f := getParam(r.n, j) pt := types.NewPtr(f.Type) - addr := s.constOffPtrSP(pt, argStart+f.Offset) - if !canSSAType(f.Type) { - s.move(f.Type, addr, argAddrVal) + ACArgs = append(ACArgs, ssa.Param{Type: f.Type, Offset: int32(argStart + f.Offset)}) + if testLateExpansion { + var a *ssa.Value + if !canSSAType(f.Type) { + a = s.newValue2(ssa.OpDereference, f.Type, argAddrVal, s.mem()) + } else { + a = s.load(f.Type, argAddrVal) + } + callArgs = append(callArgs, a) } else { - argVal := s.load(f.Type, argAddrVal) - s.storeType(f.Type, addr, argVal, 0, false) + addr := s.constOffPtrSP(pt, argStart+f.Offset) + if !canSSAType(f.Type) { + s.move(f.Type, addr, argAddrVal) + } else { + argVal := s.load(f.Type, argAddrVal) + s.storeType(f.Type, addr, argVal, 0, false) + } } } var call *ssa.Value @@ -4297,13 +4476,31 @@ func (s *state) openDeferExit() { v := s.load(r.closure.Type.Elem(), r.closure) s.maybeNilCheckClosure(v, callDefer) codeptr := s.rawLoad(types.Types[TUINTPTR], v) - call = s.newValue3(ssa.OpClosureCall, types.TypeMem, codeptr, v, s.mem()) + aux := ssa.ClosureAuxCall(ACArgs, ACResults) + if testLateExpansion { + callArgs = append(callArgs, s.mem()) + call = s.newValue2A(ssa.OpClosureLECall, aux.LateExpansionResultType(), aux, codeptr, v) + call.AddArgs(callArgs...) + } else { + call = s.newValue3A(ssa.OpClosureCall, types.TypeMem, aux, codeptr, v, s.mem()) + } } else { - // Do a static call if the original call was a static function or method - call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, fn.Sym.Linksym(), s.mem()) + aux := ssa.StaticAuxCall(fn.Sym.Linksym(), ACArgs, ACResults) + if testLateExpansion { + callArgs = append(callArgs, s.mem()) + call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux) + call.AddArgs(callArgs...) + } else { + // Do a static call if the original call was a static function or method + call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem()) + } } call.AuxInt = stksize - s.vars[&memVar] = call + if testLateExpansion { + s.vars[&memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call) + } else { + s.vars[&memVar] = call + } // Make sure that the stack slots with pointers are kept live // through the call (which is a pre-emption point). Also, we will // use the first call of the last defer exit to compute liveness @@ -4327,16 +4524,40 @@ func (s *state) openDeferExit() { } } +func (s *state) callResult(n *Node, k callKind) *ssa.Value { + return s.call(n, k, false) +} + +func (s *state) callAddr(n *Node, k callKind) *ssa.Value { + return s.call(n, k, true) +} + // Calls the function n using the specified call type. // Returns the address of the return value (or nil if none). -func (s *state) call(n *Node, k callKind) *ssa.Value { +func (s *state) call(n *Node, k callKind, returnResultAddr bool) *ssa.Value { + s.prevCall = nil var sym *types.Sym // target symbol (if static) var closure *ssa.Value // ptr to closure to run (if dynamic) var codeptr *ssa.Value // ptr to target code (if dynamic) var rcvr *ssa.Value // receiver to set fn := n.Left + var ACArgs []ssa.Param + var ACResults []ssa.Param + var callArgs []*ssa.Value + res := n.Left.Type.Results() + if k == callNormal { + nf := res.NumFields() + for i := 0; i < nf; i++ { + fp := res.Field(i) + ACResults = append(ACResults, ssa.Param{Type: fp.Type, Offset: int32(fp.Offset + Ctxt.FixedFrameSize())}) + } + } + + testLateExpansion := false + switch n.Op { case OCALLFUNC: + testLateExpansion = k != callDeferStack && ssa.LateCallExpansionEnabledWithin(s.f) if k == callNormal && fn.Op == ONAME && fn.Class() == PFUNC { sym = fn.Sym break @@ -4351,6 +4572,7 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { if fn.Op != ODOTMETH { s.Fatalf("OCALLMETH: n.Left not an ODOTMETH: %v", fn) } + testLateExpansion = k != callDeferStack && ssa.LateCallExpansionEnabledWithin(s.f) if k == callNormal { sym = fn.Sym break @@ -4362,6 +4584,7 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { if fn.Op != ODOTINTER { s.Fatalf("OCALLINTER: n.Left not an ODOTINTER: %v", fn.Op) } + testLateExpansion = k != callDeferStack && ssa.LateCallExpansionEnabledWithin(s.f) var iclosure *ssa.Value iclosure, rcvr = s.getClosureAndRcvr(fn) if k == callNormal { @@ -4380,6 +4603,7 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { var call *ssa.Value if k == callDeferStack { + testLateExpansion = ssa.LateCallExpansionEnabledWithin(s.f) // Make a defer struct d on the stack. t := deferstruct(stksize) d := tempAt(n.Pos, s.curfn, t) @@ -4430,12 +4654,21 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { } // Call runtime.deferprocStack with pointer to _defer record. - arg0 := s.constOffPtrSP(types.Types[TUINTPTR], Ctxt.FixedFrameSize()) - s.store(types.Types[TUINTPTR], arg0, addr) - call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, deferprocStack, s.mem()) + ACArgs = append(ACArgs, ssa.Param{Type: types.Types[TUINTPTR], Offset: int32(Ctxt.FixedFrameSize())}) + aux := ssa.StaticAuxCall(deferprocStack, ACArgs, ACResults) + if testLateExpansion { + callArgs = append(callArgs, addr, s.mem()) + call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux) + call.AddArgs(callArgs...) + } else { + arg0 := s.constOffPtrSP(types.Types[TUINTPTR], Ctxt.FixedFrameSize()) + s.store(types.Types[TUINTPTR], arg0, addr) + call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem()) + } if stksize < int64(Widthptr) { // We need room for both the call to deferprocStack and the call to // the deferred function. + // TODO Revisit this if/when we pass args in registers. stksize = int64(Widthptr) } call.AuxInt = stksize @@ -4447,10 +4680,20 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { if k != callNormal { // Write argsize and closure (args to newproc/deferproc). argsize := s.constInt32(types.Types[TUINT32], int32(stksize)) - addr := s.constOffPtrSP(s.f.Config.Types.UInt32Ptr, argStart) - s.store(types.Types[TUINT32], addr, argsize) - addr = s.constOffPtrSP(s.f.Config.Types.UintptrPtr, argStart+int64(Widthptr)) - s.store(types.Types[TUINTPTR], addr, closure) + ACArgs = append(ACArgs, ssa.Param{Type: types.Types[TUINT32], Offset: int32(argStart)}) + if testLateExpansion { + callArgs = append(callArgs, argsize) + } else { + addr := s.constOffPtrSP(s.f.Config.Types.UInt32Ptr, argStart) + s.store(types.Types[TUINT32], addr, argsize) + } + ACArgs = append(ACArgs, ssa.Param{Type: types.Types[TUINTPTR], Offset: int32(argStart) + int32(Widthptr)}) + if testLateExpansion { + callArgs = append(callArgs, closure) + } else { + addr := s.constOffPtrSP(s.f.Config.Types.UintptrPtr, argStart+int64(Widthptr)) + s.store(types.Types[TUINTPTR], addr, closure) + } stksize += 2 * int64(Widthptr) argStart += 2 * int64(Widthptr) } @@ -4458,7 +4701,12 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { // Set receiver (for interface calls). if rcvr != nil { addr := s.constOffPtrSP(s.f.Config.Types.UintptrPtr, argStart) - s.store(types.Types[TUINTPTR], addr, rcvr) + ACArgs = append(ACArgs, ssa.Param{Type: types.Types[TUINTPTR], Offset: int32(argStart)}) + if testLateExpansion { + callArgs = append(callArgs, rcvr) + } else { + s.store(types.Types[TUINTPTR], addr, rcvr) + } } // Write args. @@ -4466,20 +4714,38 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { args := n.Rlist.Slice() if n.Op == OCALLMETH { f := t.Recv() - s.storeArg(args[0], f.Type, argStart+f.Offset) + ACArg, arg := s.putArg(args[0], f.Type, argStart+f.Offset, testLateExpansion) + ACArgs = append(ACArgs, ACArg) + callArgs = append(callArgs, arg) args = args[1:] } for i, n := range args { f := t.Params().Field(i) - s.storeArg(n, f.Type, argStart+f.Offset) + ACArg, arg := s.putArg(n, f.Type, argStart+f.Offset, testLateExpansion) + ACArgs = append(ACArgs, ACArg) + callArgs = append(callArgs, arg) } + callArgs = append(callArgs, s.mem()) + // call target switch { case k == callDefer: - call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, deferproc, s.mem()) + aux := ssa.StaticAuxCall(deferproc, ACArgs, ACResults) + if testLateExpansion { + call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux) + call.AddArgs(callArgs...) + } else { + call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem()) + } case k == callGo: - call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, newproc, s.mem()) + aux := ssa.StaticAuxCall(newproc, ACArgs, ACResults) + if testLateExpansion { + call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux) + call.AddArgs(callArgs...) + } else { + call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem()) + } case closure != nil: // rawLoad because loading the code pointer from a // closure is always safe, but IsSanitizerSafeAddr @@ -4487,17 +4753,42 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { // critical that we not clobber any arguments already // stored onto the stack. codeptr = s.rawLoad(types.Types[TUINTPTR], closure) - call = s.newValue3(ssa.OpClosureCall, types.TypeMem, codeptr, closure, s.mem()) + if testLateExpansion { + aux := ssa.ClosureAuxCall(ACArgs, ACResults) + call = s.newValue2A(ssa.OpClosureLECall, aux.LateExpansionResultType(), aux, codeptr, closure) + call.AddArgs(callArgs...) + } else { + call = s.newValue3A(ssa.OpClosureCall, types.TypeMem, ssa.ClosureAuxCall(ACArgs, ACResults), codeptr, closure, s.mem()) + } case codeptr != nil: - call = s.newValue2(ssa.OpInterCall, types.TypeMem, codeptr, s.mem()) + if testLateExpansion { + aux := ssa.InterfaceAuxCall(ACArgs, ACResults) + call = s.newValue1A(ssa.OpInterLECall, aux.LateExpansionResultType(), aux, codeptr) + call.AddArgs(callArgs...) + } else { + call = s.newValue2A(ssa.OpInterCall, types.TypeMem, ssa.InterfaceAuxCall(ACArgs, ACResults), codeptr, s.mem()) + } case sym != nil: - call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, sym.Linksym(), s.mem()) + if testLateExpansion { + aux := ssa.StaticAuxCall(sym.Linksym(), ACArgs, ACResults) + call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux) + call.AddArgs(callArgs...) + } else { + call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, ssa.StaticAuxCall(sym.Linksym(), ACArgs, ACResults), s.mem()) + } default: s.Fatalf("bad call type %v %v", n.Op, n) } call.AuxInt = stksize // Call operations carry the argsize of the callee along with them } - s.vars[&memVar] = call + if testLateExpansion { + s.prevCall = call + s.vars[&memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call) + } else { + s.vars[&memVar] = call + } + // Insert OVARLIVE nodes + s.stmtList(n.Nbody) // Finish block for defers if k == callDefer || k == callDeferStack { @@ -4515,13 +4806,23 @@ func (s *state) call(n *Node, k callKind) *ssa.Value { s.startBlock(bNext) } - res := n.Left.Type.Results() if res.NumFields() == 0 || k != callNormal { // call has no return value. Continue with the next statement. return nil } fp := res.Field(0) - return s.constOffPtrSP(types.NewPtr(fp.Type), fp.Offset+Ctxt.FixedFrameSize()) + if returnResultAddr { + pt := types.NewPtr(fp.Type) + if testLateExpansion { + return s.newValue1I(ssa.OpSelectNAddr, pt, 0, call) + } + return s.constOffPtrSP(pt, fp.Offset+Ctxt.FixedFrameSize()) + } + + if testLateExpansion { + return s.newValue1I(ssa.OpSelectN, fp.Type, 0, call) + } + return s.load(n.Type, s.constOffPtrSP(types.NewPtr(fp.Type), fp.Offset+Ctxt.FixedFrameSize())) } // maybeNilCheckClosure checks if a nil check of a closure is needed in some @@ -4558,7 +4859,7 @@ func (s *state) getClosureAndRcvr(fn *Node) (*ssa.Value, *ssa.Value) { s.nilCheck(itab) itabidx := fn.Xoffset + 2*int64(Widthptr) + 8 // offset of fun field in runtime.itab closure := s.newValue1I(ssa.OpOffPtr, s.f.Config.Types.UintptrPtr, itabidx, itab) - rcvr := s.newValue1(ssa.OpIData, types.Types[TUINTPTR], i) + rcvr := s.newValue1(ssa.OpIData, s.f.Config.Types.BytePtr, i) return closure, rcvr } @@ -4619,7 +4920,17 @@ func (s *state) addr(n *Node) *ssa.Value { } case ORESULT: // load return from callee - return s.constOffPtrSP(t, n.Xoffset) + if s.prevCall == nil || s.prevCall.Op != ssa.OpStaticLECall && s.prevCall.Op != ssa.OpInterLECall && s.prevCall.Op != ssa.OpClosureLECall { + return s.constOffPtrSP(t, n.Xoffset) + } + which := s.prevCall.Aux.(*ssa.AuxCall).ResultForOffset(n.Xoffset) + if which == -1 { + // Do the old thing // TODO: Panic instead. + return s.constOffPtrSP(t, n.Xoffset) + } + x := s.newValue1I(ssa.OpSelectNAddr, t, which, s.prevCall) + return x + case OINDEX: if n.Left.Type.IsSlice() { a := s.expr(n.Left) @@ -4650,7 +4961,7 @@ func (s *state) addr(n *Node) *ssa.Value { addr := s.addr(n.Left) return s.newValue1(ssa.OpCopy, t, addr) // ensure that addr has the right type case OCALLFUNC, OCALLINTER, OCALLMETH: - return s.call(n, callNormal) + return s.callAddr(n, callNormal) case ODOTTYPE: v, _ := s.dottype(n, false) if v.Op != ssa.OpLoad { @@ -4669,7 +4980,7 @@ func (s *state) addr(n *Node) *ssa.Value { // canSSA reports whether n is SSA-able. // n must be an ONAME (or an ODOT sequence with an ONAME base). func (s *state) canSSA(n *Node) bool { - if Debug['N'] != 0 { + if Debug.N != 0 { return false } for n.Op == ODOT || (n.Op == OINDEX && n.Left.Type.IsArray()) { @@ -4708,7 +5019,7 @@ func (s *state) canSSA(n *Node) bool { if n.Class() == PPARAM && n.Sym != nil && n.Sym.Name == ".this" { // wrappers generated by genwrapper need to update // the .this pointer in place. - // TODO: treat as a PPARMOUT? + // TODO: treat as a PPARAMOUT? return false } return canSSAType(n.Type) @@ -4780,7 +5091,7 @@ func (s *state) nilCheck(ptr *ssa.Value) { func (s *state) boundsCheck(idx, len *ssa.Value, kind ssa.BoundsKind, bounded bool) *ssa.Value { idx = s.extendIndex(idx, len, kind, bounded) - if bounded || Debug['B'] != 0 { + if bounded || Debug.B != 0 { // If bounded or bounds checking is flag-disabled, then no check necessary, // just return the extended index. // @@ -4909,21 +5220,49 @@ func (s *state) intDivide(n *Node, a, b *ssa.Value) *ssa.Value { // The call is added to the end of the current block. // If returns is false, the block is marked as an exit block. func (s *state) rtcall(fn *obj.LSym, returns bool, results []*types.Type, args ...*ssa.Value) []*ssa.Value { + s.prevCall = nil // Write args to the stack off := Ctxt.FixedFrameSize() + testLateExpansion := ssa.LateCallExpansionEnabledWithin(s.f) + var ACArgs []ssa.Param + var ACResults []ssa.Param + var callArgs []*ssa.Value + for _, arg := range args { t := arg.Type off = Rnd(off, t.Alignment()) - ptr := s.constOffPtrSP(t.PtrTo(), off) size := t.Size() - s.store(t, ptr, arg) + ACArgs = append(ACArgs, ssa.Param{Type: t, Offset: int32(off)}) + if testLateExpansion { + callArgs = append(callArgs, arg) + } else { + ptr := s.constOffPtrSP(t.PtrTo(), off) + s.store(t, ptr, arg) + } off += size } off = Rnd(off, int64(Widthreg)) + // Accumulate results types and offsets + offR := off + for _, t := range results { + offR = Rnd(offR, t.Alignment()) + ACResults = append(ACResults, ssa.Param{Type: t, Offset: int32(offR)}) + offR += t.Size() + } + // Issue call - call := s.newValue1A(ssa.OpStaticCall, types.TypeMem, fn, s.mem()) - s.vars[&memVar] = call + var call *ssa.Value + aux := ssa.StaticAuxCall(fn, ACArgs, ACResults) + if testLateExpansion { + callArgs = append(callArgs, s.mem()) + call = s.newValue0A(ssa.OpStaticLECall, aux.LateExpansionResultType(), aux) + call.AddArgs(callArgs...) + s.vars[&memVar] = s.newValue1I(ssa.OpSelectN, types.TypeMem, int64(len(ACResults)), call) + } else { + call = s.newValue1A(ssa.OpStaticCall, types.TypeMem, aux, s.mem()) + s.vars[&memVar] = call + } if !returns { // Finish block @@ -4939,11 +5278,24 @@ func (s *state) rtcall(fn *obj.LSym, returns bool, results []*types.Type, args . // Load results res := make([]*ssa.Value, len(results)) - for i, t := range results { - off = Rnd(off, t.Alignment()) - ptr := s.constOffPtrSP(types.NewPtr(t), off) - res[i] = s.load(t, ptr) - off += t.Size() + if testLateExpansion { + for i, t := range results { + off = Rnd(off, t.Alignment()) + if canSSAType(t) { + res[i] = s.newValue1I(ssa.OpSelectN, t, int64(i), call) + } else { + addr := s.newValue1I(ssa.OpSelectNAddr, types.NewPtr(t), int64(i), call) + res[i] = s.rawLoad(t, addr) + } + off += t.Size() + } + } else { + for i, t := range results { + off = Rnd(off, t.Alignment()) + ptr := s.constOffPtrSP(types.NewPtr(t), off) + res[i] = s.load(t, ptr) + off += t.Size() + } } off = Rnd(off, int64(Widthptr)) @@ -4955,7 +5307,7 @@ func (s *state) rtcall(fn *obj.LSym, returns bool, results []*types.Type, args . // do *left = right for type t. func (s *state) storeType(t *types.Type, left, right *ssa.Value, skip skipMask, leftIsStmt bool) { - s.instrument(t, left, true) + s.instrument(t, left, instrumentWrite) if skip == 0 && (!t.HasPointers() || ssa.IsStackAddr(left)) { // Known to not have write barrier. Store the whole type. @@ -4980,7 +5332,10 @@ func (s *state) storeTypeScalars(t *types.Type, left, right *ssa.Value, skip ski case t.IsBoolean() || t.IsInteger() || t.IsFloat() || t.IsComplex(): s.store(t, left, right) case t.IsPtrShaped(): - // no scalar fields. + if t.IsPtr() && t.Elem().NotInHeap() { + s.store(t, left, right) // see issue 42032 + } + // otherwise, no scalar fields. case t.IsString(): if skip&skipLen != 0 { return @@ -5024,6 +5379,9 @@ func (s *state) storeTypeScalars(t *types.Type, left, right *ssa.Value, skip ski func (s *state) storeTypePtrs(t *types.Type, left, right *ssa.Value) { switch { case t.IsPtrShaped(): + if t.IsPtr() && t.Elem().NotInHeap() { + break // see issue 42032 + } s.store(t, left, right) case t.IsString(): ptr := s.newValue1(ssa.OpStringPtr, s.f.Config.Types.BytePtr, right) @@ -5057,8 +5415,21 @@ func (s *state) storeTypePtrs(t *types.Type, left, right *ssa.Value) { } } -func (s *state) storeArg(n *Node, t *types.Type, off int64) { - s.storeArgWithBase(n, t, s.sp, off) +// putArg evaluates n for the purpose of passing it as an argument to a function and returns the corresponding Param for the call. +// If forLateExpandedCall is true, it returns the argument value to pass to the call operation. +// If forLateExpandedCall is false, then the value is stored at the specified stack offset, and the returned value is nil. +func (s *state) putArg(n *Node, t *types.Type, off int64, forLateExpandedCall bool) (ssa.Param, *ssa.Value) { + var a *ssa.Value + if forLateExpandedCall { + if !canSSAType(t) { + a = s.newValue2(ssa.OpDereference, t, s.addr(n), s.mem()) + } else { + a = s.expr(n) + } + } else { + s.storeArgWithBase(n, t, s.sp, off) + } + return ssa.Param{Type: t, Offset: int32(off)}, a } func (s *state) storeArgWithBase(n *Node, t *types.Type, base *ssa.Value, off int64) { @@ -5554,7 +5925,7 @@ func (s *state) dottype(n *Node, commaok bool) (res, resok *ssa.Value) { // Load type out of itab, build interface with existing idata. off := s.newValue1I(ssa.OpOffPtr, byteptr, int64(Widthptr), itab) typ := s.load(byteptr, off) - idata := s.newValue1(ssa.OpIData, n.Type, iface) + idata := s.newValue1(ssa.OpIData, byteptr, iface) res = s.newValue2(ssa.OpIMake, n.Type, typ, idata) return } @@ -5576,7 +5947,7 @@ func (s *state) dottype(n *Node, commaok bool) (res, resok *ssa.Value) { bOk.AddEdgeTo(bEnd) bFail.AddEdgeTo(bEnd) s.startBlock(bEnd) - idata := s.newValue1(ssa.OpIData, n.Type, iface) + idata := s.newValue1(ssa.OpIData, byteptr, iface) res = s.newValue2(ssa.OpIMake, n.Type, s.variable(&typVar, byteptr), idata) resok = cond delete(s.vars, &typVar) @@ -5787,9 +6158,7 @@ type SSAGenState struct { // bstart remembers where each block starts (indexed by block ID) bstart []*obj.Prog - // 387 port: maps from SSE registers (REG_X?) to 387 registers (REG_F?) - SSEto387 map[int16]int16 - // Some architectures require a 64-bit temporary for FP-related register shuffling. Examples include x86-387, PPC, and Sparc V8. + // Some architectures require a 64-bit temporary for FP-related register shuffling. Examples include PPC and Sparc V8. ScratchFpMem *Node maxarg int64 // largest frame size for arguments to calls made by the function @@ -5895,7 +6264,7 @@ func emitStackObjects(e *ssafn, pp *Progs) { // Populate the stack object data. // Format must match runtime/stack.go:stackObjectRecord. - x := e.curfn.Func.lsym.Func.StackObjects + x := e.curfn.Func.lsym.Func().StackObjects off := 0 off = duintptr(x, off, uint64(len(vars))) for _, v := range vars { @@ -5932,7 +6301,7 @@ func genssa(f *ssa.Func, pp *Progs) { s.livenessMap = liveness(e, f, pp) emitStackObjects(e, pp) - openDeferInfo := e.curfn.Func.lsym.Func.OpenCodedDeferInfo + openDeferInfo := e.curfn.Func.lsym.Func().OpenCodedDeferInfo if openDeferInfo != nil { // This function uses open-coded defers -- write out the funcdata // info that we computed at the end of genssa. @@ -5956,10 +6325,6 @@ func genssa(f *ssa.Func, pp *Progs) { progToBlock[s.pp.next] = f.Blocks[0] } - if thearch.Use387 { - s.SSEto387 = map[int16]int16{} - } - s.ScratchFpMem = e.scratchFpMem if Ctxt.Flag_locationlists { @@ -6004,7 +6369,7 @@ func genssa(f *ssa.Func, pp *Progs) { // instruction. We won't use the actual liveness map on a // control instruction. Just mark it something that is // preemptible, unless this function is "all unsafe". - s.pp.nextLive = LivenessIndex{-1, -1, allUnsafe(f)} + s.pp.nextLive = LivenessIndex{-1, allUnsafe(f)} // Emit values in block thearch.SSAMarkMoves(&s, b) @@ -6082,7 +6447,7 @@ func genssa(f *ssa.Func, pp *Progs) { } // Emit control flow instructions for block var next *ssa.Block - if i < len(f.Blocks)-1 && Debug['N'] == 0 { + if i < len(f.Blocks)-1 && Debug.N == 0 { // If -N, leave next==nil so every block with successors // ends in a JMP (except call blocks - plive doesn't like // select{send,recv} followed by a JMP call). Helps keep @@ -6141,7 +6506,7 @@ func genssa(f *ssa.Func, pp *Progs) { // some of the inline marks. // Use this instruction instead. p.Pos = p.Pos.WithIsStmt() // promote position to a statement - pp.curfn.Func.lsym.Func.AddInlMark(p, inlMarks[m]) + pp.curfn.Func.lsym.Func().AddInlMark(p, inlMarks[m]) // Make the inline mark a real nop, so it doesn't generate any code. m.As = obj.ANOP m.Pos = src.NoXPos @@ -6153,7 +6518,7 @@ func genssa(f *ssa.Func, pp *Progs) { // Any unmatched inline marks now need to be added to the inlining tree (and will generate a nop instruction). for _, p := range inlMarkList { if p.As != obj.ANOP { - pp.curfn.Func.lsym.Func.AddInlMark(p, inlMarks[p]) + pp.curfn.Func.lsym.Func().AddInlMark(p, inlMarks[p]) } } } @@ -6353,6 +6718,9 @@ func AddAux2(a *obj.Addr, v *ssa.Value, offset int64) { } // Add symbol's offset from its base register. switch n := v.Aux.(type) { + case *ssa.AuxCall: + a.Name = obj.NAME_EXTERN + a.Sym = n.Fn case *obj.LSym: a.Name = obj.NAME_EXTERN a.Sym = n @@ -6387,7 +6755,7 @@ func (s *state) extendIndex(idx, len *ssa.Value, kind ssa.BoundsKind, bounded bo } else { lo = s.newValue1(ssa.OpInt64Lo, types.Types[TUINT], idx) } - if bounded || Debug['B'] != 0 { + if bounded || Debug.B != 0 { return lo } bNext := s.f.NewBlock(ssa.BlockPlain) @@ -6539,10 +6907,10 @@ func (s *SSAGenState) Call(v *ssa.Value) *obj.Prog { } else { p.Pos = v.Pos.WithNotStmt() } - if sym, ok := v.Aux.(*obj.LSym); ok { + if sym, ok := v.Aux.(*ssa.AuxCall); ok && sym.Fn != nil { p.To.Type = obj.TYPE_MEM p.To.Name = obj.NAME_EXTERN - p.To.Sym = sym + p.To.Sym = sym.Fn } else { // TODO(mdempsky): Can these differences be eliminated? switch thearch.LinkArch.Family { @@ -6565,12 +6933,14 @@ func (s *SSAGenState) PrepareCall(v *ssa.Value) { idx := s.livenessMap.Get(v) if !idx.StackMapValid() { // See Liveness.hasStackMap. - if sym, _ := v.Aux.(*obj.LSym); !(sym == typedmemclr || sym == typedmemmove) { + if sym, ok := v.Aux.(*ssa.AuxCall); !ok || !(sym.Fn == typedmemclr || sym.Fn == typedmemmove) { Fatalf("missing stack map index for %v", v.LongString()) } } - if sym, _ := v.Aux.(*obj.LSym); sym == Deferreturn { + call, ok := v.Aux.(*ssa.AuxCall) + + if ok && call.Fn == Deferreturn { // Deferred calls will appear to be returning to // the CALL deferreturn(SB) that we are about to emit. // However, the stack trace code will show the line @@ -6582,11 +6952,11 @@ func (s *SSAGenState) PrepareCall(v *ssa.Value) { thearch.Ginsnopdefer(s.pp) } - if sym, ok := v.Aux.(*obj.LSym); ok { + if ok { // Record call graph information for nowritebarrierrec // analysis. if nowritebarrierrecCheck != nil { - nowritebarrierrecCheck.recordCall(s.pp.curfn, sym, v.Pos) + nowritebarrierrecCheck.recordCall(s.pp.curfn, call.Fn, v.Pos) } } @@ -6659,56 +7029,38 @@ func (e *ssafn) Auto(pos src.XPos, t *types.Type) ssa.GCNode { } func (e *ssafn) SplitString(name ssa.LocalSlot) (ssa.LocalSlot, ssa.LocalSlot) { - n := name.N.(*Node) ptrType := types.NewPtr(types.Types[TUINT8]) lenType := types.Types[TINT] - if n.Class() == PAUTO && !n.Name.Addrtaken() { - // Split this string up into two separate variables. - p := e.splitSlot(&name, ".ptr", 0, ptrType) - l := e.splitSlot(&name, ".len", ptrType.Size(), lenType) - return p, l - } - // Return the two parts of the larger variable. - return ssa.LocalSlot{N: n, Type: ptrType, Off: name.Off}, ssa.LocalSlot{N: n, Type: lenType, Off: name.Off + int64(Widthptr)} + // Split this string up into two separate variables. + p := e.SplitSlot(&name, ".ptr", 0, ptrType) + l := e.SplitSlot(&name, ".len", ptrType.Size(), lenType) + return p, l } func (e *ssafn) SplitInterface(name ssa.LocalSlot) (ssa.LocalSlot, ssa.LocalSlot) { n := name.N.(*Node) u := types.Types[TUINTPTR] t := types.NewPtr(types.Types[TUINT8]) - if n.Class() == PAUTO && !n.Name.Addrtaken() { - // Split this interface up into two separate variables. - f := ".itab" - if n.Type.IsEmptyInterface() { - f = ".type" - } - c := e.splitSlot(&name, f, 0, u) // see comment in plive.go:onebitwalktype1. - d := e.splitSlot(&name, ".data", u.Size(), t) - return c, d + // Split this interface up into two separate variables. + f := ".itab" + if n.Type.IsEmptyInterface() { + f = ".type" } - // Return the two parts of the larger variable. - return ssa.LocalSlot{N: n, Type: u, Off: name.Off}, ssa.LocalSlot{N: n, Type: t, Off: name.Off + int64(Widthptr)} + c := e.SplitSlot(&name, f, 0, u) // see comment in plive.go:onebitwalktype1. + d := e.SplitSlot(&name, ".data", u.Size(), t) + return c, d } func (e *ssafn) SplitSlice(name ssa.LocalSlot) (ssa.LocalSlot, ssa.LocalSlot, ssa.LocalSlot) { - n := name.N.(*Node) ptrType := types.NewPtr(name.Type.Elem()) lenType := types.Types[TINT] - if n.Class() == PAUTO && !n.Name.Addrtaken() { - // Split this slice up into three separate variables. - p := e.splitSlot(&name, ".ptr", 0, ptrType) - l := e.splitSlot(&name, ".len", ptrType.Size(), lenType) - c := e.splitSlot(&name, ".cap", ptrType.Size()+lenType.Size(), lenType) - return p, l, c - } - // Return the three parts of the larger variable. - return ssa.LocalSlot{N: n, Type: ptrType, Off: name.Off}, - ssa.LocalSlot{N: n, Type: lenType, Off: name.Off + int64(Widthptr)}, - ssa.LocalSlot{N: n, Type: lenType, Off: name.Off + int64(2*Widthptr)} + p := e.SplitSlot(&name, ".ptr", 0, ptrType) + l := e.SplitSlot(&name, ".len", ptrType.Size(), lenType) + c := e.SplitSlot(&name, ".cap", ptrType.Size()+lenType.Size(), lenType) + return p, l, c } func (e *ssafn) SplitComplex(name ssa.LocalSlot) (ssa.LocalSlot, ssa.LocalSlot) { - n := name.N.(*Node) s := name.Type.Size() / 2 var t *types.Type if s == 8 { @@ -6716,53 +7068,30 @@ func (e *ssafn) SplitComplex(name ssa.LocalSlot) (ssa.LocalSlot, ssa.LocalSlot) } else { t = types.Types[TFLOAT32] } - if n.Class() == PAUTO && !n.Name.Addrtaken() { - // Split this complex up into two separate variables. - r := e.splitSlot(&name, ".real", 0, t) - i := e.splitSlot(&name, ".imag", t.Size(), t) - return r, i - } - // Return the two parts of the larger variable. - return ssa.LocalSlot{N: n, Type: t, Off: name.Off}, ssa.LocalSlot{N: n, Type: t, Off: name.Off + s} + r := e.SplitSlot(&name, ".real", 0, t) + i := e.SplitSlot(&name, ".imag", t.Size(), t) + return r, i } func (e *ssafn) SplitInt64(name ssa.LocalSlot) (ssa.LocalSlot, ssa.LocalSlot) { - n := name.N.(*Node) var t *types.Type if name.Type.IsSigned() { t = types.Types[TINT32] } else { t = types.Types[TUINT32] } - if n.Class() == PAUTO && !n.Name.Addrtaken() { - // Split this int64 up into two separate variables. - if thearch.LinkArch.ByteOrder == binary.BigEndian { - return e.splitSlot(&name, ".hi", 0, t), e.splitSlot(&name, ".lo", t.Size(), types.Types[TUINT32]) - } - return e.splitSlot(&name, ".hi", t.Size(), t), e.splitSlot(&name, ".lo", 0, types.Types[TUINT32]) - } - // Return the two parts of the larger variable. if thearch.LinkArch.ByteOrder == binary.BigEndian { - return ssa.LocalSlot{N: n, Type: t, Off: name.Off}, ssa.LocalSlot{N: n, Type: types.Types[TUINT32], Off: name.Off + 4} + return e.SplitSlot(&name, ".hi", 0, t), e.SplitSlot(&name, ".lo", t.Size(), types.Types[TUINT32]) } - return ssa.LocalSlot{N: n, Type: t, Off: name.Off + 4}, ssa.LocalSlot{N: n, Type: types.Types[TUINT32], Off: name.Off} + return e.SplitSlot(&name, ".hi", t.Size(), t), e.SplitSlot(&name, ".lo", 0, types.Types[TUINT32]) } func (e *ssafn) SplitStruct(name ssa.LocalSlot, i int) ssa.LocalSlot { - n := name.N.(*Node) st := name.Type - ft := st.FieldType(i) - var offset int64 - for f := 0; f < i; f++ { - offset += st.FieldType(f).Size() - } - if n.Class() == PAUTO && !n.Name.Addrtaken() { - // Note: the _ field may appear several times. But - // have no fear, identically-named but distinct Autos are - // ok, albeit maybe confusing for a debugger. - return e.splitSlot(&name, "."+st.FieldName(i), offset, ft) - } - return ssa.LocalSlot{N: n, Type: ft, Off: name.Off + st.FieldOff(i)} + // Note: the _ field may appear several times. But + // have no fear, identically-named but distinct Autos are + // ok, albeit maybe confusing for a debugger. + return e.SplitSlot(&name, "."+st.FieldName(i), st.FieldOff(i), st.FieldType(i)) } func (e *ssafn) SplitArray(name ssa.LocalSlot) ssa.LocalSlot { @@ -6772,19 +7101,23 @@ func (e *ssafn) SplitArray(name ssa.LocalSlot) ssa.LocalSlot { e.Fatalf(n.Pos, "bad array size") } et := at.Elem() - if n.Class() == PAUTO && !n.Name.Addrtaken() { - return e.splitSlot(&name, "[0]", 0, et) - } - return ssa.LocalSlot{N: n, Type: et, Off: name.Off} + return e.SplitSlot(&name, "[0]", 0, et) } func (e *ssafn) DerefItab(it *obj.LSym, offset int64) *obj.LSym { return itabsym(it, offset) } -// splitSlot returns a slot representing the data of parent starting at offset. -func (e *ssafn) splitSlot(parent *ssa.LocalSlot, suffix string, offset int64, t *types.Type) ssa.LocalSlot { - s := &types.Sym{Name: parent.N.(*Node).Sym.Name + suffix, Pkg: localpkg} +// SplitSlot returns a slot representing the data of parent starting at offset. +func (e *ssafn) SplitSlot(parent *ssa.LocalSlot, suffix string, offset int64, t *types.Type) ssa.LocalSlot { + node := parent.N.(*Node) + + if node.Class() != PAUTO || node.Name.Addrtaken() { + // addressed things and non-autos retain their parents (i.e., cannot truly be split) + return ssa.LocalSlot{N: node, Type: t, Off: parent.Off + offset} + } + + s := &types.Sym{Name: node.Sym.Name + suffix, Pkg: localpkg} n := &Node{ Name: new(Name), diff --git a/src/cmd/compile/internal/gc/subr.go b/src/cmd/compile/internal/gc/subr.go index 8fa3fca50f5..defefd76b34 100644 --- a/src/cmd/compile/internal/gc/subr.go +++ b/src/cmd/compile/internal/gc/subr.go @@ -96,7 +96,7 @@ func flusherrors() { } func hcrash() { - if Debug['h'] != 0 { + if Debug.h != 0 { flusherrors() if outfile != "" { os.Remove(outfile) @@ -107,7 +107,7 @@ func hcrash() { } func linestr(pos src.XPos) string { - return Ctxt.OutermostPos(pos).Format(Debug['C'] == 0, Debug['L'] == 1) + return Ctxt.OutermostPos(pos).Format(Debug.C == 0, Debug.L == 1) } // lasterror keeps track of the most recently issued error. @@ -153,7 +153,7 @@ func yyerrorl(pos src.XPos, format string, args ...interface{}) { hcrash() nerrors++ - if nsavederrors+nerrors >= 10 && Debug['e'] == 0 { + if nsavederrors+nerrors >= 10 && Debug.e == 0 { flusherrors() fmt.Printf("%v: too many errors\n", linestr(pos)) errorexit() @@ -175,7 +175,7 @@ func Warn(fmt_ string, args ...interface{}) { func Warnl(line src.XPos, fmt_ string, args ...interface{}) { adderr(line, fmt_, args...) - if Debug['m'] != 0 { + if Debug.m != 0 { flusherrors() } } @@ -222,7 +222,7 @@ func hasUniquePos(n *Node) bool { } if !n.Pos.IsKnown() { - if Debug['K'] != 0 { + if Debug.K != 0 { Warn("setlineno: unknown position (line 0)") } return false @@ -348,7 +348,7 @@ func newname(s *types.Sym) *Node { return n } -// newname returns a new ONAME Node associated with symbol s at position pos. +// newnamel returns a new ONAME Node associated with symbol s at position pos. // The caller is responsible for setting n.Name.Curfn. func newnamel(pos src.XPos, s *types.Sym) *Node { if s == nil { @@ -546,22 +546,19 @@ func methtype(t *types.Type) *types.Type { // Is type src assignment compatible to type dst? // If so, return op code to use in conversion. -// If not, return OXXX. -func assignop(src, dst *types.Type, why *string) Op { - if why != nil { - *why = "" - } - +// If not, return OXXX. In this case, the string return parameter may +// hold a reason why. In all other cases, it'll be the empty string. +func assignop(src, dst *types.Type) (Op, string) { if src == dst { - return OCONVNOP + return OCONVNOP, "" } if src == nil || dst == nil || src.Etype == TFORW || dst.Etype == TFORW || src.Orig == nil || dst.Orig == nil { - return OXXX + return OXXX, "" } // 1. src type is identical to dst. if types.Identical(src, dst) { - return OCONVNOP + return OCONVNOP, "" } // 2. src and dst have identical underlying types @@ -575,13 +572,13 @@ func assignop(src, dst *types.Type, why *string) Op { if src.IsEmptyInterface() { // Conversion between two empty interfaces // requires no code. - return OCONVNOP + return OCONVNOP, "" } if (src.Sym == nil || dst.Sym == nil) && !src.IsInterface() { // Conversion between two types, at least one unnamed, // needs no conversion. The exception is nonempty interfaces // which need to have their itab updated. - return OCONVNOP + return OCONVNOP, "" } } @@ -590,49 +587,47 @@ func assignop(src, dst *types.Type, why *string) Op { var missing, have *types.Field var ptr int if implements(src, dst, &missing, &have, &ptr) { - return OCONVIFACE + return OCONVIFACE, "" } // we'll have complained about this method anyway, suppress spurious messages. if have != nil && have.Sym == missing.Sym && (have.Type.Broke() || missing.Type.Broke()) { - return OCONVIFACE + return OCONVIFACE, "" } - if why != nil { - if isptrto(src, TINTER) { - *why = fmt.Sprintf(":\n\t%v is pointer to interface, not interface", src) - } else if have != nil && have.Sym == missing.Sym && have.Nointerface() { - *why = fmt.Sprintf(":\n\t%v does not implement %v (%v method is marked 'nointerface')", src, dst, missing.Sym) - } else if have != nil && have.Sym == missing.Sym { - *why = fmt.Sprintf(":\n\t%v does not implement %v (wrong type for %v method)\n"+ - "\t\thave %v%0S\n\t\twant %v%0S", src, dst, missing.Sym, have.Sym, have.Type, missing.Sym, missing.Type) - } else if ptr != 0 { - *why = fmt.Sprintf(":\n\t%v does not implement %v (%v method has pointer receiver)", src, dst, missing.Sym) - } else if have != nil { - *why = fmt.Sprintf(":\n\t%v does not implement %v (missing %v method)\n"+ - "\t\thave %v%0S\n\t\twant %v%0S", src, dst, missing.Sym, have.Sym, have.Type, missing.Sym, missing.Type) - } else { - *why = fmt.Sprintf(":\n\t%v does not implement %v (missing %v method)", src, dst, missing.Sym) - } + var why string + if isptrto(src, TINTER) { + why = fmt.Sprintf(":\n\t%v is pointer to interface, not interface", src) + } else if have != nil && have.Sym == missing.Sym && have.Nointerface() { + why = fmt.Sprintf(":\n\t%v does not implement %v (%v method is marked 'nointerface')", src, dst, missing.Sym) + } else if have != nil && have.Sym == missing.Sym { + why = fmt.Sprintf(":\n\t%v does not implement %v (wrong type for %v method)\n"+ + "\t\thave %v%0S\n\t\twant %v%0S", src, dst, missing.Sym, have.Sym, have.Type, missing.Sym, missing.Type) + } else if ptr != 0 { + why = fmt.Sprintf(":\n\t%v does not implement %v (%v method has pointer receiver)", src, dst, missing.Sym) + } else if have != nil { + why = fmt.Sprintf(":\n\t%v does not implement %v (missing %v method)\n"+ + "\t\thave %v%0S\n\t\twant %v%0S", src, dst, missing.Sym, have.Sym, have.Type, missing.Sym, missing.Type) + } else { + why = fmt.Sprintf(":\n\t%v does not implement %v (missing %v method)", src, dst, missing.Sym) } - return OXXX + return OXXX, why } if isptrto(dst, TINTER) { - if why != nil { - *why = fmt.Sprintf(":\n\t%v is pointer to interface, not interface", dst) - } - return OXXX + why := fmt.Sprintf(":\n\t%v is pointer to interface, not interface", dst) + return OXXX, why } if src.IsInterface() && dst.Etype != TBLANK { var missing, have *types.Field var ptr int - if why != nil && implements(dst, src, &missing, &have, &ptr) { - *why = ": need type assertion" + var why string + if implements(dst, src, &missing, &have, &ptr) { + why = ": need type assertion" } - return OXXX + return OXXX, why } // 4. src is a bidirectional channel value, dst is a channel type, @@ -640,7 +635,7 @@ func assignop(src, dst *types.Type, why *string) Op { // either src or dst is not a named type. if src.IsChan() && src.ChanDir() == types.Cboth && dst.IsChan() { if types.Identical(src.Elem(), dst.Elem()) && (src.Sym == nil || dst.Sym == nil) { - return OCONVNOP + return OCONVNOP, "" } } @@ -653,7 +648,7 @@ func assignop(src, dst *types.Type, why *string) Op { TCHAN, TINTER, TSLICE: - return OCONVNOP + return OCONVNOP, "" } } @@ -661,26 +656,23 @@ func assignop(src, dst *types.Type, why *string) Op { // 7. Any typed value can be assigned to the blank identifier. if dst.Etype == TBLANK { - return OCONVNOP + return OCONVNOP, "" } - return OXXX + return OXXX, "" } // Can we convert a value of type src to a value of type dst? // If so, return op code to use in conversion (maybe OCONVNOP). -// If not, return OXXX. +// If not, return OXXX. In this case, the string return parameter may +// hold a reason why. In all other cases, it'll be the empty string. // srcConstant indicates whether the value of type src is a constant. -func convertop(srcConstant bool, src, dst *types.Type, why *string) Op { - if why != nil { - *why = "" - } - +func convertop(srcConstant bool, src, dst *types.Type) (Op, string) { if src == dst { - return OCONVNOP + return OCONVNOP, "" } if src == nil || dst == nil { - return OXXX + return OXXX, "" } // Conversions from regular to go:notinheap are not allowed @@ -688,23 +680,19 @@ func convertop(srcConstant bool, src, dst *types.Type, why *string) Op { // rules. // (a) Disallow (*T) to (*U) where T is go:notinheap but U isn't. if src.IsPtr() && dst.IsPtr() && dst.Elem().NotInHeap() && !src.Elem().NotInHeap() { - if why != nil { - *why = fmt.Sprintf(":\n\t%v is go:notinheap, but %v is not", dst.Elem(), src.Elem()) - } - return OXXX + why := fmt.Sprintf(":\n\t%v is incomplete (or unallocatable), but %v is not", dst.Elem(), src.Elem()) + return OXXX, why } // (b) Disallow string to []T where T is go:notinheap. if src.IsString() && dst.IsSlice() && dst.Elem().NotInHeap() && (dst.Elem().Etype == types.Bytetype.Etype || dst.Elem().Etype == types.Runetype.Etype) { - if why != nil { - *why = fmt.Sprintf(":\n\t%v is go:notinheap", dst.Elem()) - } - return OXXX + why := fmt.Sprintf(":\n\t%v is incomplete (or unallocatable)", dst.Elem()) + return OXXX, why } // 1. src can be assigned to dst. - op := assignop(src, dst, why) + op, why := assignop(src, dst) if op != OXXX { - return op + return op, why } // The rules for interfaces are no different in conversions @@ -712,60 +700,57 @@ func convertop(srcConstant bool, src, dst *types.Type, why *string) Op { // with the good message from assignop. // Otherwise clear the error. if src.IsInterface() || dst.IsInterface() { - return OXXX - } - if why != nil { - *why = "" + return OXXX, why } // 2. Ignoring struct tags, src and dst have identical underlying types. if types.IdenticalIgnoreTags(src.Orig, dst.Orig) { - return OCONVNOP + return OCONVNOP, "" } // 3. src and dst are unnamed pointer types and, ignoring struct tags, // their base types have identical underlying types. if src.IsPtr() && dst.IsPtr() && src.Sym == nil && dst.Sym == nil { if types.IdenticalIgnoreTags(src.Elem().Orig, dst.Elem().Orig) { - return OCONVNOP + return OCONVNOP, "" } } // 4. src and dst are both integer or floating point types. if (src.IsInteger() || src.IsFloat()) && (dst.IsInteger() || dst.IsFloat()) { if simtype[src.Etype] == simtype[dst.Etype] { - return OCONVNOP + return OCONVNOP, "" } - return OCONV + return OCONV, "" } // 5. src and dst are both complex types. if src.IsComplex() && dst.IsComplex() { if simtype[src.Etype] == simtype[dst.Etype] { - return OCONVNOP + return OCONVNOP, "" } - return OCONV + return OCONV, "" } // Special case for constant conversions: any numeric // conversion is potentially okay. We'll validate further // within evconst. See #38117. if srcConstant && (src.IsInteger() || src.IsFloat() || src.IsComplex()) && (dst.IsInteger() || dst.IsFloat() || dst.IsComplex()) { - return OCONV + return OCONV, "" } // 6. src is an integer or has type []byte or []rune // and dst is a string type. if src.IsInteger() && dst.IsString() { - return ORUNESTR + return ORUNESTR, "" } if src.IsSlice() && dst.IsString() { if src.Elem().Etype == types.Bytetype.Etype { - return OBYTES2STR + return OBYTES2STR, "" } if src.Elem().Etype == types.Runetype.Etype { - return ORUNES2STR + return ORUNES2STR, "" } } @@ -773,21 +758,21 @@ func convertop(srcConstant bool, src, dst *types.Type, why *string) Op { // String to slice. if src.IsString() && dst.IsSlice() { if dst.Elem().Etype == types.Bytetype.Etype { - return OSTR2BYTES + return OSTR2BYTES, "" } if dst.Elem().Etype == types.Runetype.Etype { - return OSTR2RUNES + return OSTR2RUNES, "" } } // 8. src is a pointer or uintptr and dst is unsafe.Pointer. - if (src.IsPtr() || src.Etype == TUINTPTR) && dst.Etype == TUNSAFEPTR { - return OCONVNOP + if (src.IsPtr() || src.IsUintptr()) && dst.IsUnsafePtr() { + return OCONVNOP, "" } // 9. src is unsafe.Pointer and dst is a pointer or uintptr. - if src.Etype == TUNSAFEPTR && (dst.IsPtr() || dst.Etype == TUINTPTR) { - return OCONVNOP + if src.IsUnsafePtr() && (dst.IsPtr() || dst.IsUintptr()) { + return OCONVNOP, "" } // src is map and dst is a pointer to corresponding hmap. @@ -795,10 +780,10 @@ func convertop(srcConstant bool, src, dst *types.Type, why *string) Op { // go gc maps are implemented as a pointer to a hmap struct. if src.Etype == TMAP && dst.IsPtr() && src.MapType().Hmap == dst.Elem() { - return OCONVNOP + return OCONVNOP, "" } - return OXXX + return OXXX, "" } func assignconv(n *Node, t *types.Type, context string) *Node { @@ -825,7 +810,7 @@ func assignconvfn(n *Node, t *types.Type, context func() string) *Node { // Convert ideal bool from comparison to plain bool // if the next step is non-bool (like interface{}). - if n.Type == types.Idealbool && !t.IsBoolean() { + if n.Type == types.UntypedBool && !t.IsBoolean() { if n.Op == ONAME || n.Op == OLITERAL { r := nod(OCONVNOP, n, nil) r.Type = types.Types[TBOOL] @@ -839,8 +824,7 @@ func assignconvfn(n *Node, t *types.Type, context func() string) *Node { return n } - var why string - op := assignop(n.Type, t, &why) + op, why := assignop(n.Type, t) if op == OXXX { yyerror("cannot use %L as type %v in %s%s", n, t, context(), why) op = OCONV @@ -928,16 +912,20 @@ func (o Op) IsSlice3() bool { return false } -// slicePtrLen extracts the pointer and length from a slice. +// backingArrayPtrLen extracts the pointer and length from a slice or string. // This constructs two nodes referring to n, so n must be a cheapexpr. -func (n *Node) slicePtrLen() (ptr, len *Node) { +func (n *Node) backingArrayPtrLen() (ptr, len *Node) { var init Nodes c := cheapexpr(n, &init) if c != n || init.Len() != 0 { - Fatalf("slicePtrLen not cheap: %v", n) + Fatalf("backingArrayPtrLen not cheap: %v", n) } ptr = nod(OSPTR, n, nil) - ptr.Type = n.Type.Elem().PtrTo() + if n.Type.IsString() { + ptr.Type = types.Types[TUINT8].PtrTo() + } else { + ptr.Type = n.Type.Elem().PtrTo() + } len = nod(OLEN, n, nil) len.Type = types.Types[TINT] return ptr, len @@ -1036,25 +1024,24 @@ func calcHasCall(n *Node) bool { return false } -func badtype(op Op, tl *types.Type, tr *types.Type) { - fmt_ := "" +func badtype(op Op, tl, tr *types.Type) { + var s string if tl != nil { - fmt_ += fmt.Sprintf("\n\t%v", tl) + s += fmt.Sprintf("\n\t%v", tl) } if tr != nil { - fmt_ += fmt.Sprintf("\n\t%v", tr) + s += fmt.Sprintf("\n\t%v", tr) } // common mistake: *struct and *interface. if tl != nil && tr != nil && tl.IsPtr() && tr.IsPtr() { if tl.Elem().IsStruct() && tr.Elem().IsInterface() { - fmt_ += "\n\t(*struct vs *interface)" + s += "\n\t(*struct vs *interface)" } else if tl.Elem().IsInterface() && tr.Elem().IsStruct() { - fmt_ += "\n\t(*interface vs *struct)" + s += "\n\t(*interface vs *struct)" } } - s := fmt_ yyerror("illegal types for operand: %v%s", op, s) } @@ -1519,7 +1506,7 @@ func structargs(tl *types.Type, mustname bool) []*Node { // method - M func (t T)(), a TFIELD type struct // newnam - the eventual mangled name of this function func genwrapper(rcvr *types.Type, method *types.Field, newnam *types.Sym) { - if false && Debug['r'] != 0 { + if false && Debug.r != 0 { fmt.Printf("genwrapper rcvrtype=%v method=%v newnam=%v\n", rcvr, method, newnam) } @@ -1592,7 +1579,7 @@ func genwrapper(rcvr *types.Type, method *types.Field, newnam *types.Sym) { fn.Nbody.Append(call) } - if false && Debug['r'] != 0 { + if false && Debug.r != 0 { dumplist("genwrapper body", fn.Nbody) } @@ -1615,7 +1602,7 @@ func genwrapper(rcvr *types.Type, method *types.Field, newnam *types.Sym) { escapeFuncs([]*Node{fn}, false) Curfn = nil - funccompile(fn) + xtop = append(xtop, fn) } func paramNnames(ft *types.Type) []*Node { @@ -1733,7 +1720,7 @@ func implements(t, iface *types.Type, m, samename **types.Field, ptr *int) bool // the method does not exist for value types. rcvr := tm.Type.Recv().Type if rcvr.IsPtr() && !t0.IsPtr() && !followptr && !isifacemethod(tm.Type) { - if false && Debug['r'] != 0 { + if false && Debug.r != 0 { yyerror("interface pointer mismatch") } @@ -1867,8 +1854,10 @@ func isdirectiface(t *types.Type) bool { } switch t.Etype { - case TPTR, - TCHAN, + case TPTR: + // Pointers to notinheap types must be stored indirectly. See issue 42076. + return !t.Elem().NotInHeap() + case TCHAN, TMAP, TFUNC, TUNSAFEPTR: @@ -1917,3 +1906,13 @@ func ifaceData(pos src.XPos, n *Node, t *types.Type) *Node { ind.SetBounded(true) return ind } + +// typePos returns the position associated with t. +// This is where t was declared or where it appeared as a type expression. +func typePos(t *types.Type) src.XPos { + n := asNode(t.Nod) + if n == nil || !n.Pos.IsKnown() { + Fatalf("bad type: %v", t) + } + return n.Pos +} diff --git a/src/cmd/compile/internal/gc/swt.go b/src/cmd/compile/internal/gc/swt.go index 138b0acc537..8d9fbe300e8 100644 --- a/src/cmd/compile/internal/gc/swt.go +++ b/src/cmd/compile/internal/gc/swt.go @@ -189,16 +189,19 @@ func typecheckExprSwitch(n *Node) { continue } - switch { - case nilonly != "" && !n1.isNil(): + if nilonly != "" && !n1.isNil() { yyerrorl(ncase.Pos, "invalid case %v in switch (can only compare %s %v to nil)", n1, nilonly, n.Left) - case t.IsInterface() && !n1.Type.IsInterface() && !IsComparable(n1.Type): + } else if t.IsInterface() && !n1.Type.IsInterface() && !IsComparable(n1.Type) { yyerrorl(ncase.Pos, "invalid case %L in switch (incomparable type)", n1) - case assignop(n1.Type, t, nil) == 0 && assignop(t, n1.Type, nil) == 0: - if n.Left != nil { - yyerrorl(ncase.Pos, "invalid case %v in switch on %v (mismatched types %v and %v)", n1, n.Left, n1.Type, t) - } else { - yyerrorl(ncase.Pos, "invalid case %v in switch (mismatched types %v and bool)", n1, n1.Type) + } else { + op1, _ := assignop(n1.Type, t) + op2, _ := assignop(t, n1.Type) + if op1 == OXXX && op2 == OXXX { + if n.Left != nil { + yyerrorl(ncase.Pos, "invalid case %v in switch on %v (mismatched types %v and %v)", n1, n.Left, n1.Type, t) + } else { + yyerrorl(ncase.Pos, "invalid case %v in switch (mismatched types %v and bool)", n1, n1.Type) + } } } @@ -358,8 +361,8 @@ func (s *exprSwitch) flush() { // all we need here is consistency. We respect this // sorting below. sort.Slice(cc, func(i, j int) bool { - si := strlit(cc[i].lo) - sj := strlit(cc[j].lo) + si := cc[i].lo.StringVal() + sj := cc[j].lo.StringVal() if len(si) != len(sj) { return len(si) < len(sj) } @@ -368,7 +371,7 @@ func (s *exprSwitch) flush() { // runLen returns the string length associated with a // particular run of exprClauses. - runLen := func(run []exprClause) int64 { return int64(len(strlit(run[0].lo))) } + runLen := func(run []exprClause) int64 { return int64(len(run[0].lo.StringVal())) } // Collapse runs of consecutive strings with the same length. var runs [][]exprClause @@ -405,7 +408,7 @@ func (s *exprSwitch) flush() { merged := cc[:1] for _, c := range cc[1:] { last := &merged[len(merged)-1] - if last.jmp == c.jmp && last.hi.Int64()+1 == c.lo.Int64() { + if last.jmp == c.jmp && last.hi.Int64Val()+1 == c.lo.Int64Val() { last.hi = c.lo } else { merged = append(merged, c) @@ -440,7 +443,7 @@ func (c *exprClause) test(exprname *Node) *Node { // Optimize "switch true { ...}" and "switch false { ... }". if Isconst(exprname, CTBOOL) && !c.lo.Type.IsInterface() { - if exprname.Val().U.(bool) { + if exprname.BoolVal() { return c.lo } else { return nodl(c.pos, ONOT, c.lo, nil) diff --git a/src/cmd/compile/internal/gc/syntax.go b/src/cmd/compile/internal/gc/syntax.go index 47e5e59156c..43358333b82 100644 --- a/src/cmd/compile/internal/gc/syntax.go +++ b/src/cmd/compile/internal/gc/syntax.go @@ -142,7 +142,7 @@ const ( _, _ // second nodeInitorder bit _, nodeHasBreak _, nodeNoInline // used internally by inliner to indicate that a function call should not be inlined; set for OCALLFUNC and OCALLMETH only - _, nodeImplicit // implicit OADDR or ODEREF; ++/-- statement represented as OASOP; or ANDNOT lowered to OAND + _, nodeImplicit // implicit OADDR or ODEREF; ++/-- statement represented as OASOP _, nodeIsDDD // is the argument variadic _, nodeDiag // already printed error about this _, nodeColas // OAS resulting from := @@ -247,7 +247,7 @@ func (n *Node) Val() Val { // SetVal sets the Val for the node, which must not have been used with SetOpt. func (n *Node) SetVal(v Val) { if n.HasOpt() { - Debug['h'] = 1 + Debug.h = 1 Dump("have Opt", n) Fatalf("have Opt") } @@ -270,7 +270,7 @@ func (n *Node) SetOpt(x interface{}) { return } if n.HasVal() { - Debug['h'] = 1 + Debug.h = 1 Dump("have Val", n) Fatalf("have Val") } @@ -344,14 +344,22 @@ func (n *Node) CanBeAnSSASym() { // Name holds Node fields used only by named nodes (ONAME, OTYPE, OPACK, OLABEL, some OLITERAL). type Name struct { - Pack *Node // real package for import . names - Pkg *types.Pkg // pkg for OPACK nodes - Defn *Node // initializing assignment - Curfn *Node // function for local variables - Param *Param // additional fields for ONAME, OTYPE - Decldepth int32 // declaration loop depth, increased for every loop or label - Vargen int32 // unique name for ONAME within a function. Function outputs are numbered starting at one. - flags bitset16 + Pack *Node // real package for import . names + Pkg *types.Pkg // pkg for OPACK nodes + // For a local variable (not param) or extern, the initializing assignment (OAS or OAS2). + // For a closure var, the ONAME node of the outer captured variable + Defn *Node + // The ODCLFUNC node (for a static function/method or a closure) in which + // local variable or param is declared. + Curfn *Node + Param *Param // additional fields for ONAME, OTYPE + Decldepth int32 // declaration loop depth, increased for every loop or label + // Unique number for ONAME nodes within a function. Function outputs + // (results) are numbered starting at one, followed by function inputs + // (parameters), and then local variables. Vargen is used to distinguish + // local variables/params with the same name. + Vargen int32 + flags bitset16 } const ( @@ -359,7 +367,6 @@ const ( nameReadonly nameByval // is the variable captured by value or by reference nameNeedzero // if it contains pointers, needs to be zeroed on function entry - nameKeepalive // mark value live across unknown assembly call nameAutoTemp // is the variable a temporary (implies no dwarf info. reset if escapes to heap) nameUsed // for variable declared and not used error nameIsClosureVar // PAUTOHEAP closure pseudo-variable; original at n.Name.Defn @@ -376,7 +383,6 @@ func (n *Name) Captured() bool { return n.flags&nameCaptured != 0 } func (n *Name) Readonly() bool { return n.flags&nameReadonly != 0 } func (n *Name) Byval() bool { return n.flags&nameByval != 0 } func (n *Name) Needzero() bool { return n.flags&nameNeedzero != 0 } -func (n *Name) Keepalive() bool { return n.flags&nameKeepalive != 0 } func (n *Name) AutoTemp() bool { return n.flags&nameAutoTemp != 0 } func (n *Name) Used() bool { return n.flags&nameUsed != 0 } func (n *Name) IsClosureVar() bool { return n.flags&nameIsClosureVar != 0 } @@ -392,7 +398,6 @@ func (n *Name) SetCaptured(b bool) { n.flags.set(nameCaptured, b) } func (n *Name) SetReadonly(b bool) { n.flags.set(nameReadonly, b) } func (n *Name) SetByval(b bool) { n.flags.set(nameByval, b) } func (n *Name) SetNeedzero(b bool) { n.flags.set(nameNeedzero, b) } -func (n *Name) SetKeepalive(b bool) { n.flags.set(nameKeepalive, b) } func (n *Name) SetAutoTemp(b bool) { n.flags.set(nameAutoTemp, b) } func (n *Name) SetUsed(b bool) { n.flags.set(nameUsed, b) } func (n *Name) SetIsClosureVar(b bool) { n.flags.set(nameIsClosureVar, b) } @@ -463,14 +468,14 @@ type Param struct { // x1 := xN.Defn // x1.Innermost = xN.Outer // - // We leave xN.Innermost set so that we can still get to the original + // We leave x1.Innermost set so that we can still get to the original // variable quickly. Not shown here, but once we're // done parsing a function and no longer need xN.Outer for the - // lexical x reference links as described above, closurebody + // lexical x reference links as described above, funcLit // recomputes xN.Outer as the semantic x reference link tree, // even filling in x in intermediate closures that might not // have mentioned it along the way to inner closures that did. - // See closurebody for details. + // See funcLit for details. // // During the eventual compilation, then, for closure variables we have: // @@ -483,11 +488,87 @@ type Param struct { Innermost *Node Outer *Node - // OTYPE - // - // TODO: Should Func pragmas also be stored on the Name? - Pragma PragmaFlag - Alias bool // node is alias for Ntype (only used when type-checking ODCLTYPE) + // OTYPE & ONAME //go:embed info, + // sharing storage to reduce gc.Param size. + // Extra is nil, or else *Extra is a *paramType or an *embedFileList. + Extra *interface{} +} + +type paramType struct { + flag PragmaFlag + alias bool +} + +type embedFileList []string + +// Pragma returns the PragmaFlag for p, which must be for an OTYPE. +func (p *Param) Pragma() PragmaFlag { + if p.Extra == nil { + return 0 + } + return (*p.Extra).(*paramType).flag +} + +// SetPragma sets the PragmaFlag for p, which must be for an OTYPE. +func (p *Param) SetPragma(flag PragmaFlag) { + if p.Extra == nil { + if flag == 0 { + return + } + p.Extra = new(interface{}) + *p.Extra = ¶mType{flag: flag} + return + } + (*p.Extra).(*paramType).flag = flag +} + +// Alias reports whether p, which must be for an OTYPE, is a type alias. +func (p *Param) Alias() bool { + if p.Extra == nil { + return false + } + t, ok := (*p.Extra).(*paramType) + if !ok { + return false + } + return t.alias +} + +// SetAlias sets whether p, which must be for an OTYPE, is a type alias. +func (p *Param) SetAlias(alias bool) { + if p.Extra == nil { + if !alias { + return + } + p.Extra = new(interface{}) + *p.Extra = ¶mType{alias: alias} + return + } + (*p.Extra).(*paramType).alias = alias +} + +// EmbedFiles returns the list of embedded files for p, +// which must be for an ONAME var. +func (p *Param) EmbedFiles() []string { + if p.Extra == nil { + return nil + } + return *(*p.Extra).(*embedFileList) +} + +// SetEmbedFiles sets the list of embedded files for p, +// which must be for an ONAME var. +func (p *Param) SetEmbedFiles(list []string) { + if p.Extra == nil { + if len(list) == 0 { + return + } + f := embedFileList(list) + p.Extra = new(interface{}) + *p.Extra = &f + return + } + *(*p.Extra).(*embedFileList) = list } // Functions @@ -535,10 +616,16 @@ type Param struct { // Func holds Node fields used only with function-like nodes. type Func struct { Shortname *types.Sym - Enter Nodes // for example, allocate and initialize memory for escaping parameters - Exit Nodes - Cvars Nodes // closure params - Dcl []*Node // autodcl for this func/closure + // Extra entry code for the function. For example, allocate and initialize + // memory for escaping parameters. However, just for OCLOSURE, Enter is a + // list of ONAME nodes of captured variables + Enter Nodes + Exit Nodes + // ONAME nodes for closure params, each should have closurevar set + Cvars Nodes + // ONAME nodes for all params/locals for this func/closure, does NOT + // include closurevars until transformclosure runs. + Dcl []*Node // Parents records the parent scope of each scope within a // function. The root scope (0) has no parent, so the i'th @@ -557,8 +644,8 @@ type Func struct { DebugInfo *ssa.FuncDebug Ntype *Node // signature Top int // top context (ctxCallee, etc) - Closure *Node // OCLOSURE <-> ODCLFUNC - Nname *Node + Closure *Node // OCLOSURE <-> ODCLFUNC (see header comment above) + Nname *Node // The ONAME node associated with an ODCLFUNC (both have same Type) lsym *obj.LSym Inl *Inline @@ -607,6 +694,8 @@ const ( funcWrapper // is method wrapper funcNeedctxt // function uses context register (has closure variables) funcReflectMethod // function calls reflect.Type.Method or MethodByName + // true if closure inside a function; false if a simple function or a + // closure in a global variable initialization funcIsHiddenClosure funcHasDefer // contains a defer statement funcNilCheckDisabled // disable nil checks when compiling this function @@ -658,8 +747,10 @@ const ( OXXX Op = iota // names - ONAME // var or func name - ONONAME // unnamed arg or return value: f(int, string) (int, error) { etc } + ONAME // var or func name + // Unnamed arg or return value: f(int, string) (int, error) { etc } + // Also used for a qualified package identifier that hasn't been resolved yet. + ONONAME OTYPE // type name OPACK // import OLITERAL // literal @@ -679,19 +770,24 @@ const ( OSTR2BYTES // Type(Left) (Type is []byte, Left is a string) OSTR2BYTESTMP // Type(Left) (Type is []byte, Left is a string, ephemeral) OSTR2RUNES // Type(Left) (Type is []rune, Left is a string) - OAS // Left = Right or (if Colas=true) Left := Right - OAS2 // List = Rlist (x, y, z = a, b, c) - OAS2DOTTYPE // List = Right (x, ok = I.(int)) - OAS2FUNC // List = Right (x, y = f()) - OAS2MAPR // List = Right (x, ok = m["foo"]) - OAS2RECV // List = Right (x, ok = <-c) - OASOP // Left Etype= Right (x += y) - OCALL // Left(List) (function call, method call or type conversion) + // Left = Right or (if Colas=true) Left := Right + // If Colas, then Ninit includes a DCL node for Left. + OAS + // List = Rlist (x, y, z = a, b, c) or (if Colas=true) List := Rlist + // If Colas, then Ninit includes DCL nodes for List + OAS2 + OAS2DOTTYPE // List = Right (x, ok = I.(int)) + OAS2FUNC // List = Right (x, y = f()) + OAS2MAPR // List = Right (x, ok = m["foo"]) + OAS2RECV // List = Right (x, ok = <-c) + OASOP // Left Etype= Right (x += y) + OCALL // Left(List) (function call, method call or type conversion) // OCALLFUNC, OCALLMETH, and OCALLINTER have the same structure. // Prior to walk, they are: Left(List), where List is all regular arguments. // After walk, List is a series of assignments to temporaries, // and Rlist is an updated set of arguments. + // Nbody is all OVARLIVE nodes that are attached to OCALLxxx. // TODO(josharian/khr): Use Ninit instead of List for the assignments to temporaries. See CL 114797. OCALLFUNC // Left(List/Rlist) (function call f(args)) OCALLMETH // Left(List/Rlist) (direct method call x.Method(args)) @@ -699,7 +795,7 @@ const ( OCALLPART // Left.Right (method expression x.Method, not called) OCAP // cap(Left) OCLOSE // close(Left) - OCLOSURE // func Type { Body } (func literal) + OCLOSURE // func Type { Func.Closure.Nbody } (func literal) OCOMPLIT // Right{List} (composite literal, not yet lowered to specific form) OMAPLIT // Type{List} (composite literal, Type is map) OSTRUCTLIT // Type{List} (composite literal, Type is struct) @@ -718,7 +814,7 @@ const ( ODCLCONST // const pi = 3.14 ODCLTYPE // type Int int or type Int = int - ODELETE // delete(Left, Right) + ODELETE // delete(List) ODOT // Left.Sym (Left is of struct type) ODOTPTR // Left.Sym (Left is of pointer to struct type) ODOTMETH // Left.Sym (Left is non-interface, Right is method name) @@ -789,9 +885,14 @@ const ( OSIZEOF // unsafe.Sizeof(Left) // statements - OBLOCK // { List } (block of code) - OBREAK // break [Sym] - OCASE // case List: Nbody (List==nil means default) + OBLOCK // { List } (block of code) + OBREAK // break [Sym] + // OCASE: case List: Nbody (List==nil means default) + // For OTYPESW, List is a OTYPE node for the specified type (or OLITERAL + // for nil), and, if a type-switch variable is specified, Rlist is an + // ONAME for the version of the type-switch variable with the specified + // type. + OCASE OCONTINUE // continue [Sym] ODEFER // defer Left (Left must be call) OEMPTY // no-op (empty statement) @@ -815,15 +916,19 @@ const ( ORETURN // return List OSELECT // select { List } (List is list of OCASE) OSWITCH // switch Ninit; Left { List } (List is a list of OCASE) - OTYPESW // Left = Right.(type) (appears as .Left of OSWITCH) + // OTYPESW: Left := Right.(type) (appears as .Left of OSWITCH) + // Left is nil if there is no type-switch variable + OTYPESW // types OTCHAN // chan int OTMAP // map[string]int OTSTRUCT // struct{} OTINTER // interface{} - OTFUNC // func() - OTARRAY // []int, [8]int, [N]int or [...]int + // OTFUNC: func() - Left is receiver field, List is list of param fields, Rlist is + // list of result fields. + OTFUNC + OTARRAY // []int, [8]int, [N]int or [...]int // misc ODDD // func f(args ...int) or f(l...) or var a = [...]int{0, 1, 2}. diff --git a/src/cmd/compile/internal/gc/typecheck.go b/src/cmd/compile/internal/gc/typecheck.go index dec4b96fc4c..c0b05035f03 100644 --- a/src/cmd/compile/internal/gc/typecheck.go +++ b/src/cmd/compile/internal/gc/typecheck.go @@ -6,7 +6,6 @@ package gc import ( "cmd/compile/internal/types" - "cmd/internal/objabi" "fmt" "strings" ) @@ -151,8 +150,8 @@ var _typekind = []string{ } func typekind(t *types.Type) string { - if t.IsSlice() { - return "slice" + if t.IsUntyped() { + return fmt.Sprintf("%v", t) } et := t.Etype if int(et) < len(_typekind) { @@ -257,12 +256,12 @@ func typecheck(n *Node, top int) (res *Node) { // are substituted. cycle := cycleFor(n) for _, n1 := range cycle { - if n1.Name != nil && !n1.Name.Param.Alias { + if n1.Name != nil && !n1.Name.Param.Alias() { // Cycle is ok. But if n is an alias type and doesn't // have a type yet, we have a recursive type declaration // with aliases that we can't handle properly yet. // Report an error rather than crashing later. - if n.Name != nil && n.Name.Param.Alias && n.Type == nil { + if n.Name != nil && n.Name.Param.Alias() && n.Type == nil { lineno = n.Pos Fatalf("cannot handle alias type declaration (issue #25838): %v", n) } @@ -361,7 +360,7 @@ func typecheck1(n *Node, top int) (res *Node) { ok |= ctxExpr if n.Type == nil && n.Val().Ctype() == CTSTR { - n.Type = types.Idealstring + n.Type = types.UntypedString } case ONONAME: @@ -471,10 +470,10 @@ func typecheck1(n *Node, top int) (res *Node) { return n } if l.Type.NotInHeap() { - yyerror("go:notinheap map key not allowed") + yyerror("incomplete (or unallocatable) map key not allowed") } if r.Type.NotInHeap() { - yyerror("go:notinheap map value not allowed") + yyerror("incomplete (or unallocatable) map value not allowed") } setTypeNode(n, types.NewMap(l.Type, r.Type)) @@ -491,7 +490,7 @@ func typecheck1(n *Node, top int) (res *Node) { return n } if l.Type.NotInHeap() { - yyerror("chan of go:notinheap type not allowed") + yyerror("chan of incomplete (or unallocatable) type not allowed") } setTypeNode(n, types.NewChan(l.Type, n.TChanDir())) @@ -623,10 +622,29 @@ func typecheck1(n *Node, top int) (res *Node) { // no defaultlit for left // the outer context gives the type n.Type = l.Type + if (l.Type == types.UntypedFloat || l.Type == types.UntypedComplex) && r.Op == OLITERAL { + n.Type = types.UntypedInt + } break } + // For "x == x && len(s)", it's better to report that "len(s)" (type int) + // can't be used with "&&" than to report that "x == x" (type untyped bool) + // can't be converted to int (see issue #41500). + if n.Op == OANDAND || n.Op == OOROR { + if !n.Left.Type.IsBoolean() { + yyerror("invalid operation: %v (operator %v not defined on %s)", n, n.Op, typekind(n.Left.Type)) + n.Type = nil + return n + } + if !n.Right.Type.IsBoolean() { + yyerror("invalid operation: %v (operator %v not defined on %s)", n, n.Op, typekind(n.Right.Type)) + n.Type = nil + return n + } + } + // ideal mixed with non-ideal l, r = defaultlit2(l, r, false) @@ -655,8 +673,8 @@ func typecheck1(n *Node, top int) (res *Node) { // The conversion allocates, so only do it if the concrete type is huge. converted := false if r.Type.Etype != TBLANK { - aop = assignop(l.Type, r.Type, nil) - if aop != 0 { + aop, _ = assignop(l.Type, r.Type) + if aop != OXXX { if r.Type.IsInterface() && !l.Type.IsInterface() && !IsComparable(l.Type) { yyerror("invalid operation: %v (operator %v not defined on %s)", n, op, typekind(l.Type)) n.Type = nil @@ -677,8 +695,8 @@ func typecheck1(n *Node, top int) (res *Node) { } if !converted && l.Type.Etype != TBLANK { - aop = assignop(r.Type, l.Type, nil) - if aop != 0 { + aop, _ = assignop(r.Type, l.Type) + if aop != OXXX { if l.Type.IsInterface() && !r.Type.IsInterface() && !IsComparable(r.Type) { yyerror("invalid operation: %v (operator %v not defined on %s)", n, op, typekind(r.Type)) n.Type = nil @@ -713,7 +731,10 @@ func typecheck1(n *Node, top int) (res *Node) { } } - if !okfor[op][et] { + if t.Etype == TIDEAL { + t = mixUntyped(l.Type, r.Type) + } + if dt := defaultType(t); !okfor[op][dt.Etype] { yyerror("invalid operation: %v (operator %v not defined on %s)", n, op, typekind(t)) n.Type = nil return n @@ -753,17 +774,9 @@ func typecheck1(n *Node, top int) (res *Node) { } } - t = l.Type if iscmp[n.Op] { - // TIDEAL includes complex constant, but only OEQ and ONE are defined for complex, - // so check that the n.op is available for complex here before doing evconst. - if !okfor[n.Op][TCOMPLEX128] && (Isconst(l, CTCPLX) || Isconst(r, CTCPLX)) { - yyerror("invalid operation: %v (operator %v not defined on untyped complex)", n, n.Op) - n.Type = nil - return n - } evconst(n) - t = types.Idealbool + t = types.UntypedBool if n.Op != OLITERAL { l, r = defaultlit2(l, r, true) n.Left = l @@ -808,8 +821,8 @@ func typecheck1(n *Node, top int) (res *Node) { n.Type = nil return n } - if !okfor[n.Op][t.Etype] { - yyerror("invalid operation: %v %v", n.Op, t) + if !okfor[n.Op][defaultType(t).Etype] { + yyerror("invalid operation: %v (operator %v not defined on %s)", n, n.Op, typekind(t)) n.Type = nil return n } @@ -1032,13 +1045,13 @@ func typecheck1(n *Node, top int) (res *Node) { } if !n.Bounded() && Isconst(n.Right, CTINT) { - x := n.Right.Int64() + x := n.Right.Int64Val() if x < 0 { yyerror("invalid %s index %v (index must be non-negative)", why, n.Right) } else if t.IsArray() && x >= t.NumElem() { yyerror("invalid array index %v (out of bounds for %d-element array)", n.Right, t.NumElem()) - } else if Isconst(n.Left, CTSTR) && x >= int64(len(strlit(n.Left))) { - yyerror("invalid string index %v (out of bounds for %d-byte string)", n.Right, len(strlit(n.Left))) + } else if Isconst(n.Left, CTSTR) && x >= int64(len(n.Left.StringVal())) { + yyerror("invalid string index %v (out of bounds for %d-byte string)", n.Right, len(n.Left.StringVal())) } else if n.Right.Val().U.(*Mpint).Cmp(maxintval[TINT]) > 0 { yyerror("invalid %s index %v (index too large)", why, n.Right) } @@ -1134,11 +1147,11 @@ func typecheck1(n *Node, top int) (res *Node) { l = defaultlit(l, types.Types[TINT]) c = defaultlit(c, types.Types[TINT]) - if Isconst(l, CTINT) && l.Int64() < 0 { + if Isconst(l, CTINT) && l.Int64Val() < 0 { Fatalf("len for OSLICEHEADER must be non-negative") } - if Isconst(c, CTINT) && c.Int64() < 0 { + if Isconst(c, CTINT) && c.Int64Val() < 0 { Fatalf("cap for OSLICEHEADER must be non-negative") } @@ -1187,7 +1200,7 @@ func typecheck1(n *Node, top int) (res *Node) { if n.Left.Val().U.(*Mpint).Cmp(maxintval[TINT]) > 0 { Fatalf("len for OMAKESLICECOPY too large") } - if n.Left.Int64() < 0 { + if n.Left.Int64Val() < 0 { Fatalf("len for OMAKESLICECOPY must be non-negative") } } @@ -1444,7 +1457,7 @@ func typecheck1(n *Node, top int) (res *Node) { // Determine result type. switch t.Etype { case TIDEAL: - n.Type = types.Idealfloat + n.Type = types.UntypedFloat case TCOMPLEX64: n.Type = types.Types[TFLOAT32] case TCOMPLEX128: @@ -1490,7 +1503,7 @@ func typecheck1(n *Node, top int) (res *Node) { return n case TIDEAL: - t = types.Idealcomplex + t = types.UntypedComplex case TFLOAT32: t = types.Types[TCOMPLEX64] @@ -1677,8 +1690,8 @@ func typecheck1(n *Node, top int) (res *Node) { return n } var why string - n.Op = convertop(n.Left.Op == OLITERAL, t, n.Type, &why) - if n.Op == 0 { + n.Op, why = convertop(n.Left.Op == OLITERAL, t, n.Type) + if n.Op == OXXX { if !n.Diag() && !n.Type.Broke() && !n.Left.Diag() { yyerror("cannot convert %L to type %v%s", n.Left, n.Type, why) n.SetDiag(true) @@ -1756,7 +1769,7 @@ func typecheck1(n *Node, top int) (res *Node) { n.Type = nil return n } - if !checkmake(t, "len", l) || r != nil && !checkmake(t, "cap", r) { + if !checkmake(t, "len", &l) || r != nil && !checkmake(t, "cap", &r) { n.Type = nil return n } @@ -1780,7 +1793,7 @@ func typecheck1(n *Node, top int) (res *Node) { n.Type = nil return n } - if !checkmake(t, "size", l) { + if !checkmake(t, "size", &l) { n.Type = nil return n } @@ -1801,7 +1814,7 @@ func typecheck1(n *Node, top int) (res *Node) { n.Type = nil return n } - if !checkmake(t, "buffer", l) { + if !checkmake(t, "buffer", &l) { n.Type = nil return n } @@ -2051,11 +2064,6 @@ func typecheck1(n *Node, top int) (res *Node) { n.Type = nil return n - case OCASE: - ok |= ctxStmt - typecheckslice(n.List.Slice(), ctxExpr) - typecheckslice(n.Nbody.Slice(), ctxStmt) - case ODCLFUNC: ok |= ctxStmt typecheckfunc(n) @@ -2068,12 +2076,6 @@ func typecheck1(n *Node, top int) (res *Node) { ok |= ctxStmt n.Left = typecheck(n.Left, ctxType) checkwidth(n.Left.Type) - if n.Left.Type != nil && n.Left.Type.NotInHeap() && n.Left.Name.Param.Pragma&NotInHeap == 0 { - // The type contains go:notinheap types, so it - // must be marked as such (alternatively, we - // could silently propagate go:notinheap). - yyerror("type %v must be go:notinheap", n.Left.Type) - } } t := n.Type @@ -2179,14 +2181,14 @@ func checksliceindex(l *Node, r *Node, tp *types.Type) bool { } if r.Op == OLITERAL { - if r.Int64() < 0 { + if r.Int64Val() < 0 { yyerror("invalid slice index %v (index must be non-negative)", r) return false - } else if tp != nil && tp.NumElem() >= 0 && r.Int64() > tp.NumElem() { + } else if tp != nil && tp.NumElem() >= 0 && r.Int64Val() > tp.NumElem() { yyerror("invalid slice index %v (out of bounds for %d-element array)", r, tp.NumElem()) return false - } else if Isconst(l, CTSTR) && r.Int64() > int64(len(strlit(l))) { - yyerror("invalid slice index %v (out of bounds for %d-byte string)", r, len(strlit(l))) + } else if Isconst(l, CTSTR) && r.Int64Val() > int64(len(l.StringVal())) { + yyerror("invalid slice index %v (out of bounds for %d-byte string)", r, len(l.StringVal())) return false } else if r.Val().U.(*Mpint).Cmp(maxintval[TINT]) > 0 { yyerror("invalid slice index %v (index too large)", r) @@ -2439,15 +2441,6 @@ func derefall(t *types.Type) *types.Type { return t } -type typeSymKey struct { - t *types.Type - s *types.Sym -} - -// dotField maps (*types.Type, *types.Sym) pairs to the corresponding struct field (*types.Type with Etype==TFIELD). -// It is a cache for use during usefield in walk.go, only enabled when field tracking. -var dotField = map[typeSymKey]*types.Field{} - func lookdot(n *Node, t *types.Type, dostrcmp int) *types.Field { s := n.Sym @@ -2478,9 +2471,6 @@ func lookdot(n *Node, t *types.Type, dostrcmp int) *types.Field { } n.Xoffset = f1.Offset n.Type = f1.Type - if objabi.Fieldtrack_enabled > 0 { - dotField[typeSymKey{t.Orig, s}] = f1 - } if t.IsInterface() { if n.Left.Type.IsPtr() { n.Left = nod(ODEREF, n.Left, nil) // implicitstar @@ -2489,6 +2479,8 @@ func lookdot(n *Node, t *types.Type, dostrcmp int) *types.Field { } n.Op = ODOTINTER + } else { + n.SetOpt(f1) } return f1 @@ -2508,7 +2500,7 @@ func lookdot(n *Node, t *types.Type, dostrcmp int) *types.Field { n.Left = nod(OADDR, n.Left, nil) n.Left.SetImplicit(true) n.Left = typecheck(n.Left, ctxType|ctxExpr) - } else if tt.IsPtr() && !rcvr.IsPtr() && types.Identical(tt.Elem(), rcvr) { + } else if tt.IsPtr() && (!rcvr.IsPtr() || rcvr.IsPtr() && rcvr.Elem().NotInHeap()) && types.Identical(tt.Elem(), rcvr) { n.Left = nod(ODEREF, n.Left, nil) n.Left.SetImplicit(true) n.Left = typecheck(n.Left, ctxType|ctxExpr) @@ -2667,7 +2659,7 @@ func typecheckaste(op Op, call *Node, isddd bool, tstruct *types.Type, nl Nodes, return notenough: - if n == nil || !n.Diag() { + if n == nil || (!n.Diag() && n.Type != nil) { details := errorDetails(nl, tstruct, isddd) if call != nil { // call is the expression being called, not the overall call. @@ -2708,17 +2700,17 @@ func errorDetails(nl Nodes, tstruct *types.Type, isddd bool) string { return "" } } - return fmt.Sprintf("\n\thave %s\n\twant %v", nl.retsigerr(isddd), tstruct) + return fmt.Sprintf("\n\thave %s\n\twant %v", nl.sigerr(isddd), tstruct) } // sigrepr is a type's representation to the outside world, // in string representations of return signatures // e.g in error messages about wrong arguments to return. -func sigrepr(t *types.Type) string { +func sigrepr(t *types.Type, isddd bool) string { switch t { - case types.Idealstring: + case types.UntypedString: return "string" - case types.Idealbool: + case types.UntypedBool: return "bool" } @@ -2729,26 +2721,29 @@ func sigrepr(t *types.Type) string { return "number" } + // Turn []T... argument to ...T for clearer error message. + if isddd { + if !t.IsSlice() { + Fatalf("bad type for ... argument: %v", t) + } + return "..." + t.Elem().String() + } return t.String() } -// retsigerr returns the signature of the types -// at the respective return call site of a function. -func (nl Nodes) retsigerr(isddd bool) string { +// sigerr returns the signature of the types at the call or return. +func (nl Nodes) sigerr(isddd bool) string { if nl.Len() < 1 { return "()" } var typeStrings []string - for _, n := range nl.Slice() { - typeStrings = append(typeStrings, sigrepr(n.Type)) + for i, n := range nl.Slice() { + isdddArg := isddd && i == nl.Len()-1 + typeStrings = append(typeStrings, sigrepr(n.Type, isdddArg)) } - ddd := "" - if isddd { - ddd = "..." - } - return fmt.Sprintf("(%s%s)", strings.Join(typeStrings, ", "), ddd) + return fmt.Sprintf("(%s)", strings.Join(typeStrings, ", ")) } // type check composite @@ -3135,9 +3130,14 @@ func checkassign(stmt *Node, n *Node) { return } - if n.Op == ODOT && n.Left.Op == OINDEXMAP { + switch { + case n.Op == ODOT && n.Left.Op == OINDEXMAP: yyerror("cannot assign to struct field %v in map", n) - } else { + case (n.Op == OINDEX && n.Left.Type.IsString()) || n.Op == OSLICESTR: + yyerror("cannot assign to %v (strings are immutable)", n) + case n.Op == OLITERAL && n.Sym != nil && n.isGoConst(): + yyerror("cannot assign to %v (declared const)", n) + default: yyerror("cannot assign to %v", n) } n.Type = nil @@ -3251,9 +3251,7 @@ func typecheckas(n *Node) { } func checkassignto(src *types.Type, dst *Node) { - var why string - - if assignop(src, dst.Type, &why) == 0 { + if op, why := assignop(src, dst.Type); op == OXXX { yyerror("cannot assign %v to %L in multiple assignment%s", src, dst, why) return } @@ -3434,9 +3432,8 @@ func stringtoruneslit(n *Node) *Node { } var l []*Node - s := strlit(n.Left) i := 0 - for _, r := range s { + for _, r := range n.Left.StringVal() { l = append(l, nod(OKEY, nodintconst(int64(i)), nodintconst(int64(r)))) i++ } @@ -3491,7 +3488,7 @@ func setUnderlying(t, underlying *types.Type) { } // Propagate go:notinheap pragma from the Name to the Type. - if n.Name != nil && n.Name.Param != nil && n.Name.Param.Pragma&NotInHeap != 0 { + if n.Name != nil && n.Name.Param != nil && n.Name.Param.Pragma()&NotInHeap != 0 { t.SetNotInHeap(true) } @@ -3663,7 +3660,7 @@ func typecheckdef(n *Node) { n.Name.Defn = typecheck(n.Name.Defn, ctxStmt) // fills in n.Type case OTYPE: - if p := n.Name.Param; p.Alias { + if p := n.Name.Param; p.Alias() { // Type alias declaration: Simply use the rhs type - no need // to create a new type. // If we have a syntax error, p.Ntype may be nil. @@ -3713,7 +3710,8 @@ ret: n.SetWalkdef(1) } -func checkmake(t *types.Type, arg string, n *Node) bool { +func checkmake(t *types.Type, arg string, np **Node) bool { + n := *np if !n.Type.IsInteger() && n.Type.Etype != TIDEAL { yyerror("non-integer %s argument in make(%v) - %v", arg, t, n.Type) return false @@ -3723,12 +3721,12 @@ func checkmake(t *types.Type, arg string, n *Node) bool { // to avoid redundant "constant NNN overflows int" errors. switch consttype(n) { case CTINT, CTRUNE, CTFLT, CTCPLX: - n.SetVal(toint(n.Val())) - if n.Val().U.(*Mpint).CmpInt64(0) < 0 { + v := toint(n.Val()).U.(*Mpint) + if v.CmpInt64(0) < 0 { yyerror("negative %s argument in make(%v)", arg, t) return false } - if n.Val().U.(*Mpint).Cmp(maxintval[TINT]) > 0 { + if v.Cmp(maxintval[TINT]) > 0 { yyerror("%s argument too large in make(%v)", arg, t) return false } @@ -3740,6 +3738,7 @@ func checkmake(t *types.Type, arg string, n *Node) bool { // for instance, indexlit might be called here and incorporate some // of the bounds checks done for make. n = defaultlit(n, types.Types[TINT]) + *np = n return true } @@ -3886,7 +3885,7 @@ func deadcodefn(fn *Node) { return } case OFOR: - if !Isconst(n.Left, CTBOOL) || n.Left.Bool() { + if !Isconst(n.Left, CTBOOL) || n.Left.BoolVal() { return } default: @@ -3916,7 +3915,7 @@ func deadcodeslice(nn Nodes) { n.Left = deadcodeexpr(n.Left) if Isconst(n.Left, CTBOOL) { var body Nodes - if n.Left.Bool() { + if n.Left.BoolVal() { n.Rlist = Nodes{} body = n.Nbody } else { @@ -3959,7 +3958,7 @@ func deadcodeexpr(n *Node) *Node { n.Left = deadcodeexpr(n.Left) n.Right = deadcodeexpr(n.Right) if Isconst(n.Left, CTBOOL) { - if n.Left.Bool() { + if n.Left.BoolVal() { return n.Right // true && x => x } else { return n.Left // false && x => false @@ -3969,7 +3968,7 @@ func deadcodeexpr(n *Node) *Node { n.Left = deadcodeexpr(n.Left) n.Right = deadcodeexpr(n.Right) if Isconst(n.Left, CTBOOL) { - if n.Left.Bool() { + if n.Left.BoolVal() { return n.Left // true || x => true } else { return n.Right // false || x => x diff --git a/src/cmd/compile/internal/gc/universe.go b/src/cmd/compile/internal/gc/universe.go index 04861c8dd42..ff8cabd8e38 100644 --- a/src/cmd/compile/internal/gc/universe.go +++ b/src/cmd/compile/internal/gc/universe.go @@ -123,21 +123,21 @@ func lexinit() { asNode(s2.Def).SetSubOp(s.op) } - types.Idealstring = types.New(TSTRING) - types.Idealbool = types.New(TBOOL) + types.UntypedString = types.New(TSTRING) + types.UntypedBool = types.New(TBOOL) types.Types[TANY] = types.New(TANY) s := builtinpkg.Lookup("true") s.Def = asTypesNode(nodbool(true)) asNode(s.Def).Sym = lookup("true") asNode(s.Def).Name = new(Name) - asNode(s.Def).Type = types.Idealbool + asNode(s.Def).Type = types.UntypedBool s = builtinpkg.Lookup("false") s.Def = asTypesNode(nodbool(false)) asNode(s.Def).Sym = lookup("false") asNode(s.Def).Name = new(Name) - asNode(s.Def).Type = types.Idealbool + asNode(s.Def).Type = types.UntypedBool s = lookup("_") s.Block = -100 @@ -351,7 +351,7 @@ func typeinit() { sizeofString = Rnd(sliceLenOffset+int64(Widthptr), int64(Widthptr)) dowidth(types.Types[TSTRING]) - dowidth(types.Idealstring) + dowidth(types.UntypedString) } func makeErrorInterface() *types.Type { diff --git a/src/cmd/compile/internal/gc/walk.go b/src/cmd/compile/internal/gc/walk.go index 0158af8700f..a7b6e7fcb3c 100644 --- a/src/cmd/compile/internal/gc/walk.go +++ b/src/cmd/compile/internal/gc/walk.go @@ -21,7 +21,7 @@ const zeroValSize = 1024 // must match value of runtime/map.go:maxZero func walk(fn *Node) { Curfn = fn - if Debug['W'] != 0 { + if Debug.W != 0 { s := fmt.Sprintf("\nbefore walk %v", Curfn.Func.Nname.Sym) dumplist(s, Curfn.Nbody) } @@ -63,14 +63,14 @@ func walk(fn *Node) { return } walkstmtlist(Curfn.Nbody.Slice()) - if Debug['W'] != 0 { + if Debug.W != 0 { s := fmt.Sprintf("after walk %v", Curfn.Func.Nname.Sym) dumplist(s, Curfn.Nbody) } zeroResults() heapmoves() - if Debug['W'] != 0 && Curfn.Func.Enter.Len() > 0 { + if Debug.W != 0 && Curfn.Func.Enter.Len() > 0 { s := fmt.Sprintf("enter %v", Curfn.Func.Nname.Sym) dumplist(s, Curfn.Func.Enter) } @@ -231,6 +231,13 @@ func walkstmt(n *Node) *Node { case OCOPY: n.Left = copyany(n.Left, &n.Ninit, true) + case OCALLFUNC, OCALLMETH, OCALLINTER: + if n.Left.Nbody.Len() > 0 { + n.Left = wrapCall(n.Left, &n.Ninit) + } else { + n.Left = walkexpr(n.Left, &n.Ninit) + } + default: n.Left = walkexpr(n.Left, &n.Ninit) } @@ -329,19 +336,6 @@ func walkstmt(n *Node) *Node { return n } -func isSmallMakeSlice(n *Node) bool { - if n.Op != OMAKESLICE { - return false - } - r := n.Right - if r == nil { - r = n.Left - } - t := n.Type - - return smallintconst(r) && (t.Elem().Width == 0 || r.Int64() < maxImplicitStackVarSize/t.Elem().Width) -} - // walk the whole tree of the body of an // expression or simple statement. // the types expressions are calculated. @@ -442,7 +436,7 @@ func walkexpr(n *Node, init *Nodes) *Node { lno := setlineno(n) - if Debug['w'] > 1 { + if Debug.w > 1 { Dump("before walk expr", n) } @@ -480,7 +474,7 @@ opswitch: ODEREF, OSPTR, OITAB, OIDATA, OADDR: n.Left = walkexpr(n.Left, init) - case OEFACE, OAND, OSUB, OMUL, OADD, OOR, OXOR, OLSH, ORSH: + case OEFACE, OAND, OANDNOT, OSUB, OMUL, OADD, OOR, OXOR, OLSH, ORSH: n.Left = walkexpr(n.Left, init) n.Right = walkexpr(n.Right, init) @@ -558,6 +552,7 @@ opswitch: case OCALLINTER, OCALLFUNC, OCALLMETH: if n.Op == OCALLINTER { usemethod(n) + markUsedIfaceMethod(n) } if n.Op == OCALLFUNC && n.Left.Op == OCLOSURE { @@ -641,7 +636,7 @@ opswitch: // x = append(...) r := n.Right if r.Type.Elem().NotInHeap() { - yyerror("%v is go:notinheap; heap allocation disallowed", r.Type.Elem()) + yyerror("%v can't be allocated in Go; it is incomplete (or unallocatable)", r.Type.Elem()) } switch { case isAppendOfMake(r): @@ -798,8 +793,8 @@ opswitch: fromType := n.Left.Type toType := n.Type - if !fromType.IsInterface() { - markTypeUsedInInterface(fromType) + if !fromType.IsInterface() && !Curfn.Func.Nname.isBlank() { // skip unnamed functions (func _()) + markTypeUsedInInterface(fromType, Curfn.Func.lsym) } // typeword generates the type word of the interface value. @@ -954,11 +949,11 @@ opswitch: case OCONV, OCONVNOP: n.Left = walkexpr(n.Left, init) if n.Op == OCONVNOP && checkPtr(Curfn, 1) { - if n.Type.IsPtr() && n.Left.Type.Etype == TUNSAFEPTR { // unsafe.Pointer to *T + if n.Type.IsPtr() && n.Left.Type.IsUnsafePtr() { // unsafe.Pointer to *T n = walkCheckPtrAlignment(n, init, nil) break } - if n.Type.Etype == TUNSAFEPTR && n.Left.Type.Etype == TUINTPTR { // uintptr to unsafe.Pointer + if n.Type.IsUnsafePtr() && n.Left.Type.IsUintptr() { // uintptr to unsafe.Pointer n = walkCheckPtrArithmetic(n, init) break } @@ -970,14 +965,6 @@ opswitch: fn := basicnames[param] + "to" + basicnames[result] n = conv(mkcall(fn, types.Types[result], init, conv(n.Left, types.Types[param])), n.Type) - case OANDNOT: - n.Left = walkexpr(n.Left, init) - n.Op = OAND - n.SetImplicit(true) // for walkCheckPtrArithmetic - n.Right = nod(OBITNOT, n.Right, nil) - n.Right = typecheck(n.Right, ctxExpr) - n.Right = walkexpr(n.Right, init) - case ODIV, OMOD: n.Left = walkexpr(n.Left, init) n.Right = walkexpr(n.Right, init) @@ -1002,11 +989,11 @@ opswitch: // runtime calls late in SSA processing. if Widthreg < 8 && (et == TINT64 || et == TUINT64) { if n.Right.Op == OLITERAL { - // Leave div/mod by constant powers of 2. + // Leave div/mod by constant powers of 2 or small 16-bit constants. // The SSA backend will handle those. switch et { case TINT64: - c := n.Right.Int64() + c := n.Right.Int64Val() if c < 0 { c = -c } @@ -1014,7 +1001,10 @@ opswitch: break opswitch } case TUINT64: - c := uint64(n.Right.Int64()) + c := uint64(n.Right.Int64Val()) + if c < 1<<16 { + break opswitch + } if c != 0 && c&(c-1) == 0 { break opswitch } @@ -1054,15 +1044,15 @@ opswitch: } if t.IsArray() { n.SetBounded(bounded(r, t.NumElem())) - if Debug['m'] != 0 && n.Bounded() && !Isconst(n.Right, CTINT) { + if Debug.m != 0 && n.Bounded() && !Isconst(n.Right, CTINT) { Warn("index bounds check elided") } if smallintconst(n.Right) && !n.Bounded() { yyerror("index out of bounds") } } else if Isconst(n.Left, CTSTR) { - n.SetBounded(bounded(r, int64(len(strlit(n.Left))))) - if Debug['m'] != 0 && n.Bounded() && !Isconst(n.Right, CTINT) { + n.SetBounded(bounded(r, int64(len(n.Left.StringVal())))) + if Debug.m != 0 && n.Bounded() && !Isconst(n.Right, CTINT) { Warn("index bounds check elided") } if smallintconst(n.Right) && !n.Bounded() { @@ -1123,7 +1113,7 @@ opswitch: n.List.SetSecond(walkexpr(n.List.Second(), init)) case OSLICE, OSLICEARR, OSLICESTR, OSLICE3, OSLICE3ARR: - checkSlice := checkPtr(Curfn, 1) && n.Op == OSLICE3ARR && n.Left.Op == OCONVNOP && n.Left.Left.Type.Etype == TUNSAFEPTR + checkSlice := checkPtr(Curfn, 1) && n.Op == OSLICE3ARR && n.Left.Op == OCONVNOP && n.Left.Left.Type.IsUnsafePtr() if checkSlice { n.Left.Left = walkexpr(n.Left.Left, init) } else { @@ -1157,7 +1147,7 @@ opswitch: case ONEW: if n.Type.Elem().NotInHeap() { - yyerror("%v is go:notinheap; heap allocation disallowed", n.Type.Elem()) + yyerror("%v can't be allocated in Go; it is incomplete (or unallocatable)", n.Type.Elem()) } if n.Esc == EscNone { if n.Type.Elem().Width >= maxImplicitStackVarSize { @@ -1328,11 +1318,11 @@ opswitch: } t := n.Type if t.Elem().NotInHeap() { - yyerror("%v is go:notinheap; heap allocation disallowed", t.Elem()) + yyerror("%v can't be allocated in Go; it is incomplete (or unallocatable)", t.Elem()) } if n.Esc == EscNone { - if !isSmallMakeSlice(n) { - Fatalf("non-small OMAKESLICE with EscNone: %v", n) + if why := heapAllocReason(n); why != "" { + Fatalf("%v has EscNone, but %v", n, why) } // var arr [r]T // n = arr[:l] @@ -1405,7 +1395,7 @@ opswitch: t := n.Type if t.Elem().NotInHeap() { - yyerror("%v is go:notinheap; heap allocation disallowed", t.Elem()) + yyerror("%v can't be allocated in Go; it is incomplete (or unallocatable)", t.Elem()) } length := conv(n.Left, types.Types[TINT]) @@ -1477,7 +1467,7 @@ opswitch: } else { // slicebytetostring(*[32]byte, ptr *byte, n int) string n.Left = cheapexpr(n.Left, init) - ptr, len := n.Left.slicePtrLen() + ptr, len := n.Left.backingArrayPtrLen() n = mkcall("slicebytetostring", n.Type, init, a, ptr, len) } @@ -1490,13 +1480,13 @@ opswitch: } // slicebytetostringtmp(ptr *byte, n int) string n.Left = cheapexpr(n.Left, init) - ptr, len := n.Left.slicePtrLen() + ptr, len := n.Left.backingArrayPtrLen() n = mkcall("slicebytetostringtmp", n.Type, init, ptr, len) case OSTR2BYTES: s := n.Left if Isconst(s, CTSTR) { - sc := strlit(s) + sc := s.StringVal() // Allocate a [n]byte of the right size. t := types.NewArray(types.Types[TUINT8], int64(len(sc))) @@ -1604,7 +1594,7 @@ opswitch: updateHasCall(n) - if Debug['w'] != 0 && n != nil { + if Debug.w != 0 && n != nil { Dump("after walk expr", n) } @@ -1614,8 +1604,27 @@ opswitch: // markTypeUsedInInterface marks that type t is converted to an interface. // This information is used in the linker in dead method elimination. -func markTypeUsedInInterface(t *types.Type) { - typenamesym(t).Linksym().Set(obj.AttrUsedInIface, true) +func markTypeUsedInInterface(t *types.Type, from *obj.LSym) { + tsym := typenamesym(t).Linksym() + // Emit a marker relocation. The linker will know the type is converted + // to an interface if "from" is reachable. + r := obj.Addrel(from) + r.Sym = tsym + r.Type = objabi.R_USEIFACE +} + +// markUsedIfaceMethod marks that an interface method is used in the current +// function. n is OCALLINTER node. +func markUsedIfaceMethod(n *Node) { + ityp := n.Left.Left.Type + tsym := typenamesym(ityp).Linksym() + r := obj.Addrel(Curfn.Func.lsym) + r.Sym = tsym + // n.Left.Xoffset is the method index * Widthptr (the offset of code pointer + // in itab). + midx := n.Left.Xoffset / int64(Widthptr) + r.Add = ifaceMethodOffset(ityp, midx) + r.Type = objabi.R_USEIFACEMETHOD } // rtconvfn returns the parameter and result types that will be used by a @@ -1905,7 +1914,7 @@ func walkprint(nn *Node, init *Nodes) *Node { for i := 0; i < len(s); { var strs []string for i < len(s) && Isconst(s[i], CTSTR) { - strs = append(strs, strlit(s[i])) + strs = append(strs, s[i].StringVal()) i++ } if len(strs) > 0 { @@ -1951,7 +1960,17 @@ func walkprint(nn *Node, init *Nodes) *Node { on = syslook("printiface") } on = substArgTypes(on, n.Type) // any-1 - case TPTR, TCHAN, TMAP, TFUNC, TUNSAFEPTR: + case TPTR: + if n.Type.Elem().NotInHeap() { + on = syslook("printuintptr") + n = nod(OCONV, n, nil) + n.Type = types.Types[TUNSAFEPTR] + n = nod(OCONV, n, nil) + n.Type = types.Types[TUINTPTR] + break + } + fallthrough + case TCHAN, TMAP, TFUNC, TUNSAFEPTR: on = syslook("printpointer") on = substArgTypes(on, n.Type) // any-1 case TSLICE: @@ -1974,7 +1993,7 @@ func walkprint(nn *Node, init *Nodes) *Node { case TSTRING: cs := "" if Isconst(n, CTSTR) { - cs = strlit(n) + cs = n.StringVal() } switch cs { case " ": @@ -2143,7 +2162,7 @@ func reorder3(all []*Node) []*Node { // The result of reorder3save MUST be assigned back to n, e.g. // n.Left = reorder3save(n.Left, all, i, early) func reorder3save(n *Node, all []*Node, i int, early *[]*Node) *Node { - if !aliased(n, all, i) { + if !aliased(n, all[:i]) { return n } @@ -2175,73 +2194,75 @@ func outervalue(n *Node) *Node { } } -// Is it possible that the computation of n might be -// affected by writes in as up to but not including the ith element? -func aliased(n *Node, all []*Node, i int) bool { - if n == nil { +// Is it possible that the computation of r might be +// affected by assignments in all? +func aliased(r *Node, all []*Node) bool { + if r == nil { return false } // Treat all fields of a struct as referring to the whole struct. // We could do better but we would have to keep track of the fields. - for n.Op == ODOT { - n = n.Left + for r.Op == ODOT { + r = r.Left } // Look for obvious aliasing: a variable being assigned // during the all list and appearing in n. - // Also record whether there are any writes to main memory. - // Also record whether there are any writes to variables - // whose addresses have been taken. + // Also record whether there are any writes to addressable + // memory (either main memory or variables whose addresses + // have been taken). memwrite := false - varwrite := false - for _, an := range all[:i] { - a := outervalue(an.Left) - - for a.Op == ODOT { - a = a.Left + for _, as := range all { + // We can ignore assignments to blank. + if as.Left.isBlank() { + continue } - if a.Op != ONAME { + l := outervalue(as.Left) + if l.Op != ONAME { memwrite = true continue } - switch n.Class() { + switch l.Class() { default: - varwrite = true + Fatalf("unexpected class: %v, %v", l, l.Class()) + + case PAUTOHEAP, PEXTERN: + memwrite = true continue case PAUTO, PPARAM, PPARAMOUT: - if n.Name.Addrtaken() { - varwrite = true + if l.Name.Addrtaken() { + memwrite = true continue } - if vmatch2(a, n) { - // Direct hit. + if vmatch2(l, r) { + // Direct hit: l appears in r. return true } } } - // The variables being written do not appear in n. - // However, n might refer to computed addresses + // The variables being written do not appear in r. + // However, r might refer to computed addresses // that are being written. // If no computed addresses are affected by the writes, no aliasing. - if !memwrite && !varwrite { + if !memwrite { return false } - // If n does not refer to computed addresses - // (that is, if n only refers to variables whose addresses + // If r does not refer to computed addresses + // (that is, if r only refers to variables whose addresses // have not been taken), no aliasing. - if varexpr(n) { + if varexpr(r) { return false } - // Otherwise, both the writes and n refer to computed memory addresses. + // Otherwise, both the writes and r refer to computed memory addresses. // Assume that they might conflict. return true } @@ -2629,7 +2650,7 @@ func addstr(n *Node, init *Nodes) *Node { sz := int64(0) for _, n1 := range n.List.Slice() { if n1.Op == OLITERAL { - sz += int64(len(strlit(n1))) + sz += int64(len(n1.StringVal())) } } @@ -2757,36 +2778,25 @@ func appendslice(n *Node, init *Nodes) *Node { // instantiate typedslicecopy(typ *type, dstPtr *any, dstLen int, srcPtr *any, srcLen int) int fn := syslook("typedslicecopy") fn = substArgTypes(fn, l1.Type.Elem(), l2.Type.Elem()) - ptr1, len1 := nptr1.slicePtrLen() - ptr2, len2 := nptr2.slicePtrLen() + ptr1, len1 := nptr1.backingArrayPtrLen() + ptr2, len2 := nptr2.backingArrayPtrLen() ncopy = mkcall1(fn, types.Types[TINT], &nodes, typename(elemtype), ptr1, len1, ptr2, len2) - } else if instrumenting && !compiling_runtime { - // rely on runtime to instrument copy. - // copy(s[len(l1):], l2) + // rely on runtime to instrument: + // copy(s[len(l1):], l2) + // l2 can be a slice or string. nptr1 := nod(OSLICE, s, nil) nptr1.Type = s.Type nptr1.SetSliceBounds(nod(OLEN, l1, nil), nil, nil) nptr1 = cheapexpr(nptr1, &nodes) - nptr2 := l2 - if l2.Type.IsString() { - // instantiate func slicestringcopy(toPtr *byte, toLen int, fr string) int - fn := syslook("slicestringcopy") - ptr, len := nptr1.slicePtrLen() - str := nod(OCONVNOP, nptr2, nil) - str.Type = types.Types[TSTRING] - ncopy = mkcall1(fn, types.Types[TINT], &nodes, ptr, len, str) - } else { - // instantiate func slicecopy(to any, fr any, wid uintptr) int - fn := syslook("slicecopy") - fn = substArgTypes(fn, l1.Type.Elem(), l2.Type.Elem()) - ptr1, len1 := nptr1.slicePtrLen() - ptr2, len2 := nptr2.slicePtrLen() - ncopy = mkcall1(fn, types.Types[TINT], &nodes, ptr1, len1, ptr2, len2, nodintconst(elemtype.Width)) - } + ptr1, len1 := nptr1.backingArrayPtrLen() + ptr2, len2 := nptr2.backingArrayPtrLen() + fn := syslook("slicecopy") + fn = substArgTypes(fn, ptr1.Type.Elem(), ptr2.Type.Elem()) + ncopy = mkcall1(fn, types.Types[TINT], &nodes, ptr1, len1, ptr2, len2, nodintconst(elemtype.Width)) } else { // memmove(&s[len(l1)], &l2[0], len(l2)*sizeof(T)) nptr1 := nod(OINDEX, s, nod(OLEN, l1, nil)) @@ -2814,7 +2824,7 @@ func appendslice(n *Node, init *Nodes) *Node { // isAppendOfMake reports whether n is of the form append(x , make([]T, y)...). // isAppendOfMake assumes n has already been typechecked. func isAppendOfMake(n *Node) bool { - if Debug['N'] != 0 || instrumenting { + if Debug.N != 0 || instrumenting { return false } @@ -3085,28 +3095,25 @@ func copyany(n *Node, init *Nodes, runtimecall bool) *Node { Curfn.Func.setWBPos(n.Pos) fn := writebarrierfn("typedslicecopy", n.Left.Type.Elem(), n.Right.Type.Elem()) n.Left = cheapexpr(n.Left, init) - ptrL, lenL := n.Left.slicePtrLen() + ptrL, lenL := n.Left.backingArrayPtrLen() n.Right = cheapexpr(n.Right, init) - ptrR, lenR := n.Right.slicePtrLen() + ptrR, lenR := n.Right.backingArrayPtrLen() return mkcall1(fn, n.Type, init, typename(n.Left.Type.Elem()), ptrL, lenL, ptrR, lenR) } if runtimecall { - if n.Right.Type.IsString() { - fn := syslook("slicestringcopy") - n.Left = cheapexpr(n.Left, init) - ptr, len := n.Left.slicePtrLen() - str := nod(OCONVNOP, n.Right, nil) - str.Type = types.Types[TSTRING] - return mkcall1(fn, n.Type, init, ptr, len, str) - } + // rely on runtime to instrument: + // copy(n.Left, n.Right) + // n.Right can be a slice or string. + + n.Left = cheapexpr(n.Left, init) + ptrL, lenL := n.Left.backingArrayPtrLen() + n.Right = cheapexpr(n.Right, init) + ptrR, lenR := n.Right.backingArrayPtrLen() fn := syslook("slicecopy") - fn = substArgTypes(fn, n.Left.Type.Elem(), n.Right.Type.Elem()) - n.Left = cheapexpr(n.Left, init) - ptrL, lenL := n.Left.slicePtrLen() - n.Right = cheapexpr(n.Right, init) - ptrR, lenR := n.Right.slicePtrLen() + fn = substArgTypes(fn, ptrL.Type.Elem(), ptrR.Type.Elem()) + return mkcall1(fn, n.Type, init, ptrL, lenL, ptrR, lenR, nodintconst(n.Left.Type.Elem().Width)) } @@ -3437,7 +3444,7 @@ func walkcompare(n *Node, init *Nodes) *Node { func tracecmpArg(n *Node, t *types.Type, init *Nodes) *Node { // Ugly hack to avoid "constant -1 overflows uintptr" errors, etc. - if n.Op == OLITERAL && n.Type.IsSigned() && n.Int64() < 0 { + if n.Op == OLITERAL && n.Type.IsSigned() && n.Int64Val() < 0 { n = copyexpr(n, n.Type, init) } @@ -3507,7 +3514,7 @@ func walkcompareString(n *Node, init *Nodes) *Node { // Length-only checks are ok, though. maxRewriteLen = 0 } - if s := strlit(cs); len(s) <= maxRewriteLen { + if s := cs.StringVal(); len(s) <= maxRewriteLen { if len(s) > 0 { ncs = safeexpr(ncs, init) } @@ -3602,26 +3609,32 @@ func bounded(n *Node, max int64) bool { bits := int32(8 * n.Type.Width) if smallintconst(n) { - v := n.Int64() + v := n.Int64Val() return 0 <= v && v < max } switch n.Op { - case OAND: + case OAND, OANDNOT: v := int64(-1) - if smallintconst(n.Left) { - v = n.Left.Int64() - } else if smallintconst(n.Right) { - v = n.Right.Int64() + switch { + case smallintconst(n.Left): + v = n.Left.Int64Val() + case smallintconst(n.Right): + v = n.Right.Int64Val() + if n.Op == OANDNOT { + v = ^v + if !sign { + v &= 1< 0 && v >= 2 { bits-- v >>= 1 @@ -3638,7 +3651,7 @@ func bounded(n *Node, max int64) bool { case ORSH: if !sign && smallintconst(n.Right) { - v := n.Right.Int64() + v := n.Right.Int64Val() if v > int64(bits) { return true } @@ -3694,6 +3707,8 @@ func usemethod(n *Node) { // Also need to check for reflect package itself (see Issue #38515). if s := res0.Type.Sym; s != nil && s.Name == "Method" && isReflectPkg(s.Pkg) { Curfn.Func.SetReflectMethod(true) + // The LSym is initialized at this point. We need to set the attribute on the LSym. + Curfn.Func.lsym.Set(obj.AttrReflectMethod, true) } } @@ -3719,10 +3734,13 @@ func usefield(n *Node) { if t.IsPtr() { t = t.Elem() } - field := dotField[typeSymKey{t.Orig, n.Sym}] + field := n.Opt().(*types.Field) if field == nil { Fatalf("usefield %v %v without paramfld", n.Left.Type, n.Sym) } + if field.Sym != n.Sym || field.Offset != n.Xoffset { + Fatalf("field inconsistency: %v,%v != %v,%v", field.Sym, field.Offset, n.Sym, n.Xoffset) + } if !strings.Contains(field.Note, "go:\"track\"") { return } @@ -3857,6 +3875,14 @@ func candiscard(n *Node) bool { // builtin(a1, a2, a3) // }(x, y, z) // for print, println, and delete. +// +// Rewrite +// go f(x, y, uintptr(unsafe.Pointer(z))) +// into +// go func(a1, a2, a3) { +// builtin(a1, a2, uintptr(a3)) +// }(x, y, unsafe.Pointer(z)) +// for function contains unsafe-uintptr arguments. var wrapCall_prgen int @@ -3868,9 +3894,27 @@ func wrapCall(n *Node, init *Nodes) *Node { init.AppendNodes(&n.Ninit) } + isBuiltinCall := n.Op != OCALLFUNC && n.Op != OCALLMETH && n.Op != OCALLINTER + + // Turn f(a, b, []T{c, d, e}...) back into f(a, b, c, d, e). + if !isBuiltinCall && n.IsDDD() { + last := n.List.Len() - 1 + if va := n.List.Index(last); va.Op == OSLICELIT { + n.List.Set(append(n.List.Slice()[:last], va.List.Slice()...)) + n.SetIsDDD(false) + } + } + + // origArgs keeps track of what argument is uintptr-unsafe/unsafe-uintptr conversion. + origArgs := make([]*Node, n.List.Len()) t := nod(OTFUNC, nil, nil) for i, arg := range n.List.Slice() { s := lookupN("a", i) + if !isBuiltinCall && arg.Op == OCONVNOP && arg.Type.IsUintptr() && arg.Left.Type.IsUnsafePtr() { + origArgs[i] = arg + arg = arg.Left + n.List.SetIndex(i, arg) + } t.List.Append(symfield(s, arg.Type)) } @@ -3878,10 +3922,23 @@ func wrapCall(n *Node, init *Nodes) *Node { sym := lookupN("wrap·", wrapCall_prgen) fn := dclfunc(sym, t) - a := nod(n.Op, nil, nil) - a.List.Set(paramNnames(t.Type)) - a = typecheck(a, ctxStmt) - fn.Nbody.Set1(a) + args := paramNnames(t.Type) + for i, origArg := range origArgs { + if origArg == nil { + continue + } + arg := nod(origArg.Op, args[i], nil) + arg.Type = origArg.Type + args[i] = arg + } + call := nod(n.Op, nil, nil) + if !isBuiltinCall { + call.Op = OCALL + call.Left = n.Left + call.SetIsDDD(n.IsDDD()) + } + call.List.Set(args) + fn.Nbody.Set1(call) funcbody() @@ -3889,12 +3946,12 @@ func wrapCall(n *Node, init *Nodes) *Node { typecheckslice(fn.Nbody.Slice(), ctxStmt) xtop = append(xtop, fn) - a = nod(OCALL, nil, nil) - a.Left = fn.Func.Nname - a.List.Set(n.List.Slice()) - a = typecheck(a, ctxStmt) - a = walkexpr(a, init) - return a + call = nod(OCALL, nil, nil) + call.Left = fn.Func.Nname + call.List.Set(n.List.Slice()) + call = typecheck(call, ctxStmt) + call = walkexpr(call, init) + return call } // substArgTypes substitutes the given list of types for @@ -3933,7 +3990,7 @@ func canMergeLoads() bool { // isRuneCount reports whether n is of the form len([]rune(string)). // These are optimized into a call to runtime.countrunes. func isRuneCount(n *Node) bool { - return Debug['N'] == 0 && !instrumenting && n.Op == OLEN && n.Left.Op == OSTR2RUNES + return Debug.N == 0 && !instrumenting && n.Op == OLEN && n.Left.Op == OSTR2RUNES } func walkCheckPtrAlignment(n *Node, init *Nodes, count *Node) *Node { @@ -4002,14 +4059,10 @@ func walkCheckPtrArithmetic(n *Node, init *Nodes) *Node { case OADD: walk(n.Left) walk(n.Right) - case OSUB: + case OSUB, OANDNOT: walk(n.Left) - case OAND: - if n.Implicit() { // was OANDNOT - walk(n.Left) - } case OCONVNOP: - if n.Left.Type.Etype == TUNSAFEPTR { + if n.Left.Type.IsUnsafePtr() { n.Left = cheapexpr(n.Left, init) originals = append(originals, convnop(n.Left, types.Types[TUNSAFEPTR])) } diff --git a/src/cmd/compile/internal/logopt/log_opts.go b/src/cmd/compile/internal/logopt/log_opts.go index 22a94b0f2d3..37a049d6403 100644 --- a/src/cmd/compile/internal/logopt/log_opts.go +++ b/src/cmd/compile/internal/logopt/log_opts.go @@ -19,6 +19,7 @@ import ( "strconv" "strings" "sync" + "unicode" ) // This implements (non)optimization logging for -json option to the Go compiler @@ -223,11 +224,11 @@ type Diagnostic struct { // A LoggedOpt is what the compiler produces and accumulates, // to be converted to JSON for human or IDE consumption. type LoggedOpt struct { - pos src.XPos // Source code position at which the event occurred. If it is inlined, outer and all inlined locations will appear in JSON. - pass string // For human/adhoc consumption; does not appear in JSON (yet) - fname string // For human/adhoc consumption; does not appear in JSON (yet) - what string // The (non) optimization; "nilcheck", "boundsCheck", "inline", "noInline" - target []interface{} // Optional target(s) or parameter(s) of "what" -- what was inlined, why it was not, size of copy, etc. 1st is most important/relevant. + pos src.XPos // Source code position at which the event occurred. If it is inlined, outer and all inlined locations will appear in JSON. + compilerPass string // Compiler pass. For human/adhoc consumption; does not appear in JSON (yet) + functionName string // Function name. For human/adhoc consumption; does not appear in JSON (yet) + what string // The (non) optimization; "nilcheck", "boundsCheck", "inline", "noInline" + target []interface{} // Optional target(s) or parameter(s) of "what" -- what was inlined, why it was not, size of copy, etc. 1st is most important/relevant. } type logFormat uint8 @@ -240,12 +241,13 @@ const ( var Format = None var dest string +// LogJsonOption parses and validates the version,directory value attached to the -json compiler flag. func LogJsonOption(flagValue string) { version, directory := parseLogFlag("json", flagValue) if version != 0 { log.Fatal("-json version must be 0") } - checkLogPath("json", directory) + dest = checkLogPath(directory) Format = Json0 } @@ -268,51 +270,80 @@ func parseLogFlag(flag, value string) (version int, directory string) { return } +// isWindowsDriveURI returns true if the file URI is of the format used by +// Windows URIs. The url.Parse package does not specially handle Windows paths +// (see golang/go#6027), so we check if the URI path has a drive prefix (e.g. "/C:"). +// (copied from tools/internal/span/uri.go) +// this is less comprehensive that the processing in filepath.IsAbs on Windows. +func isWindowsDriveURIPath(uri string) bool { + if len(uri) < 4 { + return false + } + return uri[0] == '/' && unicode.IsLetter(rune(uri[1])) && uri[2] == ':' +} + +func parseLogPath(destination string) (string, string) { + if filepath.IsAbs(destination) { + return filepath.Clean(destination), "" + } + if strings.HasPrefix(destination, "file://") { // IKWIAD, or Windows C:\foo\bar\baz + uri, err := url.Parse(destination) + if err != nil { + return "", fmt.Sprintf("optimizer logging destination looked like file:// URI but failed to parse: err=%v", err) + } + destination = uri.Host + uri.Path + if isWindowsDriveURIPath(destination) { + // strip leading / from /C: + // unlike tools/internal/span/uri.go, do not uppercase the drive letter -- let filepath.Clean do what it does. + destination = destination[1:] + } + return filepath.Clean(destination), "" + } + return "", fmt.Sprintf("optimizer logging destination %s was neither %s-prefixed directory nor file://-prefixed file URI", destination, string(filepath.Separator)) +} + // checkLogPath does superficial early checking of the string specifying // the directory to which optimizer logging is directed, and if // it passes the test, stores the string in LO_dir -func checkLogPath(flag, destination string) { - sep := string(os.PathSeparator) - if strings.HasPrefix(destination, "/") || strings.HasPrefix(destination, sep) { - err := os.MkdirAll(destination, 0755) - if err != nil { - log.Fatalf("optimizer logging destination ',' but could not create : err=%v", err) - } - } else if strings.HasPrefix(destination, "file://") { // IKWIAD, or Windows C:\foo\bar\baz - uri, err := url.Parse(destination) - if err != nil { - log.Fatalf("optimizer logging destination looked like file:// URI but failed to parse: err=%v", err) - } - destination = uri.Host + uri.Path - err = os.MkdirAll(destination, 0755) - if err != nil { - log.Fatalf("optimizer logging destination ',' but could not create %s: err=%v", destination, err) - } - } else { - log.Fatalf("optimizer logging destination %s was neither %s-prefixed directory nor file://-prefixed file URI", destination, sep) +func checkLogPath(destination string) string { + path, complaint := parseLogPath(destination) + if complaint != "" { + log.Fatalf(complaint) } - dest = destination + err := os.MkdirAll(path, 0755) + if err != nil { + log.Fatalf("optimizer logging destination ',' but could not create : err=%v", err) + } + return path } var loggedOpts []*LoggedOpt var mu = sync.Mutex{} // mu protects loggedOpts. -func NewLoggedOpt(pos src.XPos, what, pass, fname string, args ...interface{}) *LoggedOpt { +// NewLoggedOpt allocates a new LoggedOpt, to later be passed to either NewLoggedOpt or LogOpt as "args". +// Pos is the source position (including inlining), what is the message, pass is which pass created the message, +// funcName is the name of the function +// A typical use for this to accumulate an explanation for a missed optimization, for example, why did something escape? +func NewLoggedOpt(pos src.XPos, what, pass, funcName string, args ...interface{}) *LoggedOpt { pass = strings.Replace(pass, " ", "_", -1) - return &LoggedOpt{pos, pass, fname, what, args} + return &LoggedOpt{pos, pass, funcName, what, args} } -func LogOpt(pos src.XPos, what, pass, fname string, args ...interface{}) { +// Logopt logs information about a (usually missed) optimization performed by the compiler. +// Pos is the source position (including inlining), what is the message, pass is which pass created the message, +// funcName is the name of the function +func LogOpt(pos src.XPos, what, pass, funcName string, args ...interface{}) { if Format == None { return } - lo := NewLoggedOpt(pos, what, pass, fname, args...) + lo := NewLoggedOpt(pos, what, pass, funcName, args...) mu.Lock() defer mu.Unlock() // Because of concurrent calls from back end, no telling what the order will be, but is stable-sorted by outer Pos before use. loggedOpts = append(loggedOpts, lo) } +// Enabled returns whether optimization logging is enabled. func Enabled() bool { switch Format { case None: @@ -459,11 +490,13 @@ func FlushLoggedOpts(ctxt *obj.Link, slashPkgPath string) { } } +// newPointRange returns a single-position Range for the compiler source location p. func newPointRange(p src.Pos) Range { return Range{Start: Position{p.Line(), p.Col()}, End: Position{p.Line(), p.Col()}} } +// newLocation returns the Location for the compiler source location p func newLocation(p src.Pos) Location { loc := Location{URI: uriIfy(uprootedPath(p.Filename())), Range: newPointRange(p)} return loc diff --git a/src/cmd/compile/internal/logopt/logopt_test.go b/src/cmd/compile/internal/logopt/logopt_test.go index df3e70a614f..e121c1abd22 100644 --- a/src/cmd/compile/internal/logopt/logopt_test.go +++ b/src/cmd/compile/internal/logopt/logopt_test.go @@ -51,7 +51,35 @@ func want(t *testing.T, out string, desired string) { func wantN(t *testing.T, out string, desired string, n int) { if strings.Count(out, desired) != n { - t.Errorf("expected exactly %d occurences of %s in \n%s", n, desired, out) + t.Errorf("expected exactly %d occurrences of %s in \n%s", n, desired, out) + } +} + +func TestPathStuff(t *testing.T) { + sep := string(filepath.Separator) + if path, whine := parseLogPath("file:///c:foo"); path != "c:foo" || whine != "" { // good path + t.Errorf("path='%s', whine='%s'", path, whine) + } + if path, whine := parseLogPath("file:///foo"); path != sep+"foo" || whine != "" { // good path + t.Errorf("path='%s', whine='%s'", path, whine) + } + if path, whine := parseLogPath("foo"); path != "" || whine == "" { // BAD path + t.Errorf("path='%s', whine='%s'", path, whine) + } + if sep == "\\" { // On WINDOWS ONLY + if path, whine := parseLogPath("C:/foo"); path != "C:\\foo" || whine != "" { // good path + t.Errorf("path='%s', whine='%s'", path, whine) + } + if path, whine := parseLogPath("c:foo"); path != "" || whine == "" { // BAD path + t.Errorf("path='%s', whine='%s'", path, whine) + } + if path, whine := parseLogPath("/foo"); path != "" || whine == "" { // BAD path + t.Errorf("path='%s', whine='%s'", path, whine) + } + } else { // ON UNIX ONLY + if path, whine := parseLogPath("/foo"); path != sep+"foo" || whine != "" { // good path + t.Errorf("path='%s', whine='%s'", path, whine) + } } } @@ -180,21 +208,20 @@ func s15a8(x *[15]int64) [15]int64 { `"relatedInformation":[{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":4,"character":11},"end":{"line":4,"character":11}}},"message":"inlineLoc"}]}`) want(t, slogged, `{"range":{"start":{"line":11,"character":6},"end":{"line":11,"character":6}},"severity":3,"code":"isInBounds","source":"go compiler","message":""}`) want(t, slogged, `{"range":{"start":{"line":7,"character":6},"end":{"line":7,"character":6}},"severity":3,"code":"canInlineFunction","source":"go compiler","message":"cost: 35"}`) - want(t, slogged, `{"range":{"start":{"line":21,"character":21},"end":{"line":21,"character":21}},"severity":3,"code":"cannotInlineCall","source":"go compiler","message":"foo cannot be inlined (escaping closure variable)"}`) // escape analysis explanation want(t, slogged, `{"range":{"start":{"line":7,"character":13},"end":{"line":7,"character":13}},"severity":3,"code":"leak","source":"go compiler","message":"parameter z leaks to ~r2 with derefs=0",`+ `"relatedInformation":[`+ `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: flow: y = z:"},`+ - `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: from y = \u003cN\u003e (assign-pair)"},`+ - `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: flow: ~r1 = y:"},`+ + `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: from y := z (assign-pair)"},`+ + `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: flow: ~R0 = y:"},`+ `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":4,"character":11},"end":{"line":4,"character":11}}},"message":"inlineLoc"},`+ `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: from y.b (dot of pointer)"},`+ `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":4,"character":11},"end":{"line":4,"character":11}}},"message":"inlineLoc"},`+ `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: from \u0026y.b (address-of)"},`+ `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":4,"character":9},"end":{"line":4,"character":9}}},"message":"inlineLoc"},`+ - `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: from ~r1 = \u003cN\u003e (assign-pair)"},`+ - `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":3},"end":{"line":9,"character":3}}},"message":"escflow: flow: ~r2 = ~r1:"},`+ - `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":3},"end":{"line":9,"character":3}}},"message":"escflow: from return (*int)(~r1) (return)"}]}`) + `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":13},"end":{"line":9,"character":13}}},"message":"escflow: from ~R0 = \u003cN\u003e (assign-pair)"},`+ + `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":3},"end":{"line":9,"character":3}}},"message":"escflow: flow: ~r2 = ~R0:"},`+ + `{"location":{"uri":"file://tmpdir/file.go","range":{"start":{"line":9,"character":3},"end":{"line":9,"character":3}}},"message":"escflow: from return (*int)(~R0) (return)"}]}`) }) } diff --git a/src/cmd/compile/internal/ppc64/ssa.go b/src/cmd/compile/internal/ppc64/ssa.go index f8d9ac23790..3e20c44a4c7 100644 --- a/src/cmd/compile/internal/ppc64/ssa.go +++ b/src/cmd/compile/internal/ppc64/ssa.go @@ -166,34 +166,46 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p2.To.Reg = v.Reg1() case ssa.OpPPC64LoweredAtomicAnd8, - ssa.OpPPC64LoweredAtomicOr8: + ssa.OpPPC64LoweredAtomicAnd32, + ssa.OpPPC64LoweredAtomicOr8, + ssa.OpPPC64LoweredAtomicOr32: // LWSYNC - // LBAR (Rarg0), Rtmp + // LBAR/LWAR (Rarg0), Rtmp // AND/OR Rarg1, Rtmp - // STBCCC Rtmp, (Rarg0) + // STBCCC/STWCCC Rtmp, (Rarg0) // BNE -3(PC) + ld := ppc64.ALBAR + st := ppc64.ASTBCCC + if v.Op == ssa.OpPPC64LoweredAtomicAnd32 || v.Op == ssa.OpPPC64LoweredAtomicOr32 { + ld = ppc64.ALWAR + st = ppc64.ASTWCCC + } r0 := v.Args[0].Reg() r1 := v.Args[1].Reg() // LWSYNC - Assuming shared data not write-through-required nor // caching-inhibited. See Appendix B.2.2.2 in the ISA 2.07b. plwsync := s.Prog(ppc64.ALWSYNC) plwsync.To.Type = obj.TYPE_NONE - p := s.Prog(ppc64.ALBAR) + // LBAR or LWAR + p := s.Prog(ld) p.From.Type = obj.TYPE_MEM p.From.Reg = r0 p.To.Type = obj.TYPE_REG p.To.Reg = ppc64.REGTMP + // AND/OR reg1,out p1 := s.Prog(v.Op.Asm()) p1.From.Type = obj.TYPE_REG p1.From.Reg = r1 p1.To.Type = obj.TYPE_REG p1.To.Reg = ppc64.REGTMP - p2 := s.Prog(ppc64.ASTBCCC) + // STBCCC or STWCCC + p2 := s.Prog(st) p2.From.Type = obj.TYPE_REG p2.From.Reg = ppc64.REGTMP p2.To.Type = obj.TYPE_MEM p2.To.Reg = r0 p2.RegTo2 = ppc64.REGTMP + // BNE retry p3 := s.Prog(ppc64.ABNE) p3.To.Type = obj.TYPE_BRANCH gc.Patch(p3, p) @@ -565,6 +577,42 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p = s.Prog(obj.ANOP) gc.Patch(pbover, p) + case ssa.OpPPC64CLRLSLWI: + r := v.Reg() + r1 := v.Args[0].Reg() + shifts := v.AuxInt + p := s.Prog(v.Op.Asm()) + // clrlslwi ra,rs,mb,sh will become rlwinm ra,rs,sh,mb-sh,31-sh as described in ISA + p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: ssa.GetPPC64Shiftmb(shifts)} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: ssa.GetPPC64Shiftsh(shifts)}) + p.Reg = r1 + p.To.Type = obj.TYPE_REG + p.To.Reg = r + + case ssa.OpPPC64CLRLSLDI: + r := v.Reg() + r1 := v.Args[0].Reg() + shifts := v.AuxInt + p := s.Prog(v.Op.Asm()) + // clrlsldi ra,rs,mb,sh will become rldic ra,rs,sh,mb-sh + p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: ssa.GetPPC64Shiftmb(shifts)} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: ssa.GetPPC64Shiftsh(shifts)}) + p.Reg = r1 + p.To.Type = obj.TYPE_REG + p.To.Reg = r + + // Mask has been set as sh + case ssa.OpPPC64RLDICL: + r := v.Reg() + r1 := v.Args[0].Reg() + shifts := v.AuxInt + p := s.Prog(v.Op.Asm()) + p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: ssa.GetPPC64Shiftsh(shifts)} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: ssa.GetPPC64Shiftmb(shifts)}) + p.Reg = r1 + p.To.Type = obj.TYPE_REG + p.To.Reg = r + case ssa.OpPPC64ADD, ssa.OpPPC64FADD, ssa.OpPPC64FADDS, ssa.OpPPC64SUB, ssa.OpPPC64FSUB, ssa.OpPPC64FSUBS, ssa.OpPPC64MULLD, ssa.OpPPC64MULLW, ssa.OpPPC64DIVDU, ssa.OpPPC64DIVWU, ssa.OpPPC64SRAD, ssa.OpPPC64SRAW, ssa.OpPPC64SRD, ssa.OpPPC64SRW, ssa.OpPPC64SLD, ssa.OpPPC64SLW, @@ -601,6 +649,24 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p.To.Type = obj.TYPE_REG p.To.Reg = v.Reg() + // Auxint holds encoded rotate + mask + case ssa.OpPPC64RLWINM, ssa.OpPPC64RLWMI: + rot, _, _, mask := ssa.DecodePPC64RotateMask(v.AuxInt) + p := s.Prog(v.Op.Asm()) + p.To = obj.Addr{Type: obj.TYPE_REG, Reg: v.Reg()} + p.Reg = v.Args[0].Reg() + p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: int64(rot)} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: int64(mask)}) + + // Auxint holds mask + case ssa.OpPPC64RLWNM: + _, _, _, mask := ssa.DecodePPC64RotateMask(v.AuxInt) + p := s.Prog(v.Op.Asm()) + p.To = obj.Addr{Type: obj.TYPE_REG, Reg: v.Reg()} + p.Reg = v.Args[0].Reg() + p.From = obj.Addr{Type: obj.TYPE_REG, Reg: v.Args[1].Reg()} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: int64(mask)}) + case ssa.OpPPC64MADDLD: r := v.Reg() r1 := v.Args[0].Reg() @@ -641,7 +707,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p.From.Reg = v.Args[0].Reg() case ssa.OpPPC64ADDconst, ssa.OpPPC64ANDconst, ssa.OpPPC64ORconst, ssa.OpPPC64XORconst, - ssa.OpPPC64SRADconst, ssa.OpPPC64SRAWconst, ssa.OpPPC64SRDconst, ssa.OpPPC64SRWconst, ssa.OpPPC64SLDconst, ssa.OpPPC64SLWconst: + ssa.OpPPC64SRADconst, ssa.OpPPC64SRAWconst, ssa.OpPPC64SRDconst, ssa.OpPPC64SRWconst, + ssa.OpPPC64SLDconst, ssa.OpPPC64SLWconst, ssa.OpPPC64EXTSWSLconst, ssa.OpPPC64MULLWconst, ssa.OpPPC64MULLDconst: p := s.Prog(v.Op.Asm()) p.Reg = v.Args[0].Reg() p.From.Type = obj.TYPE_CONST @@ -1714,6 +1781,9 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { pp := s.Call(v) pp.To.Reg = ppc64.REG_LR + // Insert a hint this is not a subroutine return. + pp.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: 1}) + if gc.Ctxt.Flag_shared { // When compiling Go into PIC, the function we just // called via pointer might have been implemented in diff --git a/src/cmd/compile/internal/riscv64/ggen.go b/src/cmd/compile/internal/riscv64/ggen.go index be31fad4417..f7c03fe7c21 100644 --- a/src/cmd/compile/internal/riscv64/ggen.go +++ b/src/cmd/compile/internal/riscv64/ggen.go @@ -25,7 +25,15 @@ func zeroRange(pp *gc.Progs, p *obj.Prog, off, cnt int64, _ *uint32) *obj.Prog { return p } - // TODO(jsing): Add a duff zero implementation for medium sized ranges. + if cnt <= int64(128*gc.Widthptr) { + p = pp.Appendpp(p, riscv.AADDI, obj.TYPE_CONST, 0, off, obj.TYPE_REG, riscv.REG_A0, 0) + p.Reg = riscv.REG_SP + p = pp.Appendpp(p, obj.ADUFFZERO, obj.TYPE_NONE, 0, 0, obj.TYPE_MEM, 0, 0) + p.To.Name = obj.NAME_EXTERN + p.To.Sym = gc.Duffzero + p.To.Offset = 8 * (128 - cnt/int64(gc.Widthptr)) + return p + } // Loop, zeroing pointer width bytes at a time. // ADD $(off), SP, T0 diff --git a/src/cmd/compile/internal/riscv64/ssa.go b/src/cmd/compile/internal/riscv64/ssa.go index 73f0dbc1951..0beb5b4bd18 100644 --- a/src/cmd/compile/internal/riscv64/ssa.go +++ b/src/cmd/compile/internal/riscv64/ssa.go @@ -190,7 +190,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { // input args need no code case ssa.OpPhi: gc.CheckLoweredPhi(v) - case ssa.OpCopy, ssa.OpRISCV64MOVconvert: + case ssa.OpCopy, ssa.OpRISCV64MOVconvert, ssa.OpRISCV64MOVDreg: if v.Type.IsMemory() { return } @@ -208,6 +208,11 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p.From.Reg = rs p.To.Type = obj.TYPE_REG p.To.Reg = rd + case ssa.OpRISCV64MOVDnop: + if v.Reg() != v.Args[0].Reg() { + v.Fatalf("input[0] and output not in same register %s", v.LongString()) + } + // nothing to do case ssa.OpLoadReg: if v.Type.IsFlags() { v.Fatalf("load flags not implemented: %v", v.LongString()) @@ -228,6 +233,37 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { gc.AddrAuto(&p.To, v) case ssa.OpSP, ssa.OpSB, ssa.OpGetG: // nothing to do + case ssa.OpRISCV64MOVBreg, ssa.OpRISCV64MOVHreg, ssa.OpRISCV64MOVWreg, + ssa.OpRISCV64MOVBUreg, ssa.OpRISCV64MOVHUreg, ssa.OpRISCV64MOVWUreg: + a := v.Args[0] + for a.Op == ssa.OpCopy || a.Op == ssa.OpRISCV64MOVDreg { + a = a.Args[0] + } + as := v.Op.Asm() + rs := v.Args[0].Reg() + rd := v.Reg() + if a.Op == ssa.OpLoadReg { + t := a.Type + switch { + case v.Op == ssa.OpRISCV64MOVBreg && t.Size() == 1 && t.IsSigned(), + v.Op == ssa.OpRISCV64MOVHreg && t.Size() == 2 && t.IsSigned(), + v.Op == ssa.OpRISCV64MOVWreg && t.Size() == 4 && t.IsSigned(), + v.Op == ssa.OpRISCV64MOVBUreg && t.Size() == 1 && !t.IsSigned(), + v.Op == ssa.OpRISCV64MOVHUreg && t.Size() == 2 && !t.IsSigned(), + v.Op == ssa.OpRISCV64MOVWUreg && t.Size() == 4 && !t.IsSigned(): + // arg is a proper-typed load and already sign/zero-extended + if rs == rd { + return + } + as = riscv.AMOV + default: + } + } + p := s.Prog(as) + p.From.Type = obj.TYPE_REG + p.From.Reg = rs + p.To.Type = obj.TYPE_REG + p.To.Reg = rd case ssa.OpRISCV64ADD, ssa.OpRISCV64SUB, ssa.OpRISCV64SUBW, ssa.OpRISCV64XOR, ssa.OpRISCV64OR, ssa.OpRISCV64AND, ssa.OpRISCV64SLL, ssa.OpRISCV64SRA, ssa.OpRISCV64SRL, ssa.OpRISCV64SLT, ssa.OpRISCV64SLTU, ssa.OpRISCV64MUL, ssa.OpRISCV64MULW, ssa.OpRISCV64MULH, @@ -572,6 +608,20 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p.To.Type = obj.TYPE_REG p.To.Reg = v.Reg() + case ssa.OpRISCV64DUFFZERO: + p := s.Prog(obj.ADUFFZERO) + p.To.Type = obj.TYPE_MEM + p.To.Name = obj.NAME_EXTERN + p.To.Sym = gc.Duffzero + p.To.Offset = v.AuxInt + + case ssa.OpRISCV64DUFFCOPY: + p := s.Prog(obj.ADUFFCOPY) + p.To.Type = obj.TYPE_MEM + p.To.Name = obj.NAME_EXTERN + p.To.Sym = gc.Duffcopy + p.To.Offset = v.AuxInt + default: v.Fatalf("Unhandled op %v", v.Op) } diff --git a/src/cmd/compile/internal/s390x/ssa.go b/src/cmd/compile/internal/s390x/ssa.go index 00d253c95a2..8037357131c 100644 --- a/src/cmd/compile/internal/s390x/ssa.go +++ b/src/cmd/compile/internal/s390x/ssa.go @@ -182,11 +182,23 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { i := v.Aux.(s390x.RotateParams) p := s.Prog(v.Op.Asm()) p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: int64(i.Start)} - p.RestArgs = []obj.Addr{ + p.SetRestArgs([]obj.Addr{ {Type: obj.TYPE_CONST, Offset: int64(i.End)}, {Type: obj.TYPE_CONST, Offset: int64(i.Amount)}, {Type: obj.TYPE_REG, Reg: r2}, - } + }) + p.To = obj.Addr{Type: obj.TYPE_REG, Reg: r1} + case ssa.OpS390XRISBGZ: + r1 := v.Reg() + r2 := v.Args[0].Reg() + i := v.Aux.(s390x.RotateParams) + p := s.Prog(v.Op.Asm()) + p.From = obj.Addr{Type: obj.TYPE_CONST, Offset: int64(i.Start)} + p.SetRestArgs([]obj.Addr{ + {Type: obj.TYPE_CONST, Offset: int64(i.End)}, + {Type: obj.TYPE_CONST, Offset: int64(i.Amount)}, + {Type: obj.TYPE_REG, Reg: r2}, + }) p.To = obj.Addr{Type: obj.TYPE_REG, Reg: r1} case ssa.OpS390XADD, ssa.OpS390XADDW, ssa.OpS390XSUB, ssa.OpS390XSUBW, @@ -360,7 +372,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { case ssa.OpS390XSLDconst, ssa.OpS390XSLWconst, ssa.OpS390XSRDconst, ssa.OpS390XSRWconst, ssa.OpS390XSRADconst, ssa.OpS390XSRAWconst, - ssa.OpS390XRLLGconst, ssa.OpS390XRLLconst: + ssa.OpS390XRLLconst: p := s.Prog(v.Op.Asm()) p.From.Type = obj.TYPE_CONST p.From.Offset = v.AuxInt @@ -761,6 +773,14 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) { p.To.Type = obj.TYPE_MEM p.To.Reg = v.Args[0].Reg() gc.AddAux(&p.To, v) + case ssa.OpS390XLAN, ssa.OpS390XLAO: + // LA(N|O) Ry, TMP, 0(Rx) + op := s.Prog(v.Op.Asm()) + op.From.Type = obj.TYPE_REG + op.From.Reg = v.Args[1].Reg() + op.Reg = s390x.REGTMP + op.To.Type = obj.TYPE_MEM + op.To.Reg = v.Args[0].Reg() case ssa.OpS390XLANfloor, ssa.OpS390XLAOfloor: r := v.Args[0].Reg() // clobbered, assumed R1 in comments @@ -905,7 +925,7 @@ func ssaGenBlock(s *gc.SSAGenState, b, next *ssa.Block) { p.From.Type = obj.TYPE_CONST p.From.Offset = int64(s390x.NotEqual & s390x.NotUnordered) // unordered is not possible p.Reg = s390x.REG_R3 - p.RestArgs = []obj.Addr{{Type: obj.TYPE_CONST, Offset: 0}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: 0}) if b.Succs[0].Block() != next { s.Br(s390x.ABR, b.Succs[0].Block()) } @@ -948,17 +968,17 @@ func ssaGenBlock(s *gc.SSAGenState, b, next *ssa.Block) { p.From.Type = obj.TYPE_CONST p.From.Offset = int64(mask & s390x.NotUnordered) // unordered is not possible p.Reg = b.Controls[0].Reg() - p.RestArgs = []obj.Addr{{Type: obj.TYPE_REG, Reg: b.Controls[1].Reg()}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_REG, Reg: b.Controls[1].Reg()}) case ssa.BlockS390XCGIJ, ssa.BlockS390XCIJ: p.From.Type = obj.TYPE_CONST p.From.Offset = int64(mask & s390x.NotUnordered) // unordered is not possible p.Reg = b.Controls[0].Reg() - p.RestArgs = []obj.Addr{{Type: obj.TYPE_CONST, Offset: int64(int8(b.AuxInt))}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: int64(int8(b.AuxInt))}) case ssa.BlockS390XCLGIJ, ssa.BlockS390XCLIJ: p.From.Type = obj.TYPE_CONST p.From.Offset = int64(mask & s390x.NotUnordered) // unordered is not possible p.Reg = b.Controls[0].Reg() - p.RestArgs = []obj.Addr{{Type: obj.TYPE_CONST, Offset: int64(uint8(b.AuxInt))}} + p.SetFrom3(obj.Addr{Type: obj.TYPE_CONST, Offset: int64(uint8(b.AuxInt))}) default: b.Fatalf("branch not implemented: %s", b.LongString()) } diff --git a/src/cmd/compile/internal/ssa/addressingmodes.go b/src/cmd/compile/internal/ssa/addressingmodes.go index aae0def27fc..1baf143869a 100644 --- a/src/cmd/compile/internal/ssa/addressingmodes.go +++ b/src/cmd/compile/internal/ssa/addressingmodes.go @@ -59,22 +59,22 @@ func addressingModes(f *Func) { v.AuxInt += p.AuxInt case [2]auxType{auxSymValAndOff, auxInt32}: vo := ValAndOff(v.AuxInt) - if !vo.canAdd(p.AuxInt) { + if !vo.canAdd64(p.AuxInt) { continue } - v.AuxInt = vo.add(p.AuxInt) + v.AuxInt = int64(vo.addOffset64(p.AuxInt)) case [2]auxType{auxSymValAndOff, auxSymOff}: vo := ValAndOff(v.AuxInt) if v.Aux != nil && p.Aux != nil { continue } - if !vo.canAdd(p.AuxInt) { + if !vo.canAdd64(p.AuxInt) { continue } if p.Aux != nil { v.Aux = p.Aux } - v.AuxInt = vo.add(p.AuxInt) + v.AuxInt = int64(vo.addOffset64(p.AuxInt)) case [2]auxType{auxSymOff, auxNone}: // nothing to do case [2]auxType{auxSymValAndOff, auxNone}: diff --git a/src/cmd/compile/internal/ssa/branchelim.go b/src/cmd/compile/internal/ssa/branchelim.go index 4f9fd8e22e1..1d34f8160b1 100644 --- a/src/cmd/compile/internal/ssa/branchelim.go +++ b/src/cmd/compile/internal/ssa/branchelim.go @@ -35,7 +35,7 @@ func branchelim(f *Func) { for _, b := range f.Blocks { for _, v := range b.Values { switch v.Op { - case OpLoad, OpAtomicLoad8, OpAtomicLoad32, OpAtomicLoad64, OpAtomicLoadPtr, OpAtomicLoadAcq32: + case OpLoad, OpAtomicLoad8, OpAtomicLoad32, OpAtomicLoad64, OpAtomicLoadPtr, OpAtomicLoadAcq32, OpAtomicLoadAcq64: loadAddr.add(v.Args[0].ID) case OpMove: loadAddr.add(v.Args[1].ID) diff --git a/src/cmd/compile/internal/ssa/check.go b/src/cmd/compile/internal/ssa/check.go index 828f645b399..5f5dfc328a3 100644 --- a/src/cmd/compile/internal/ssa/check.go +++ b/src/cmd/compile/internal/ssa/check.go @@ -165,6 +165,18 @@ func checkFunc(f *Func) { f.Fatalf("value %v has Aux type %T, want string", v, v.Aux) } canHaveAux = true + case auxCallOff: + canHaveAuxInt = true + fallthrough + case auxCall: + if ac, ok := v.Aux.(*AuxCall); ok { + if v.Op == OpStaticCall && ac.Fn == nil { + f.Fatalf("value %v has *AuxCall with nil Fn", v) + } + } else { + f.Fatalf("value %v has Aux type %T, want *AuxCall", v, v.Aux) + } + canHaveAux = true case auxSym, auxTyp: canHaveAux = true case auxSymOff, auxSymValAndOff, auxTypSize: @@ -257,6 +269,38 @@ func checkFunc(f *Func) { f.Fatalf("bad %s type: want uintptr, have %s", v.Op, v.Type.String()) } + case OpStringLen: + if v.Type != c.Types.Int { + f.Fatalf("bad %s type: want int, have %s", + v.Op, v.Type.String()) + } + case OpLoad: + if !v.Args[1].Type.IsMemory() { + f.Fatalf("bad arg 1 type to %s: want mem, have %s", + v.Op, v.Args[1].Type.String()) + } + case OpStore: + if !v.Type.IsMemory() { + f.Fatalf("bad %s type: want mem, have %s", + v.Op, v.Type.String()) + } + if !v.Args[2].Type.IsMemory() { + f.Fatalf("bad arg 2 type to %s: want mem, have %s", + v.Op, v.Args[2].Type.String()) + } + case OpCondSelect: + if !v.Args[2].Type.IsBoolean() { + f.Fatalf("bad arg 2 type to %s: want boolean, have %s", + v.Op, v.Args[2].Type.String()) + } + case OpAddPtr: + if !v.Args[0].Type.IsPtrShaped() && v.Args[0].Type != c.Types.Uintptr { + f.Fatalf("bad arg 0 type to %s: want ptr, have %s", v.Op, v.Args[0].LongString()) + } + if !v.Args[1].Type.IsInteger() { + f.Fatalf("bad arg 1 type to %s: want integer, have %s", v.Op, v.Args[1].LongString()) + } + } // TODO: check for cycles in values diff --git a/src/cmd/compile/internal/ssa/compile.go b/src/cmd/compile/internal/ssa/compile.go index 444475d67ab..63994d1778e 100644 --- a/src/cmd/compile/internal/ssa/compile.go +++ b/src/cmd/compile/internal/ssa/compile.go @@ -47,6 +47,9 @@ func Compile(f *Func) { stack := make([]byte, 16384) n := runtime.Stack(stack, false) stack = stack[:n] + if f.HTMLWriter != nil { + f.HTMLWriter.flushPhases() + } f.Fatalf("panic during %s while compiling %s:\n\n%v\n\n%s\n", phaseName, f.Name, err, stack) } }() @@ -201,6 +204,13 @@ func (p *pass) addDump(s string) { p.dump[s] = true } +func (p *pass) String() string { + if p == nil { + return "nil pass" + } + return p.name +} + // Run consistency checker between each phase var ( checkEnabled = false @@ -294,37 +304,39 @@ commas. For example: ` } - if phase == "check" && flag == "on" { - checkEnabled = val != 0 - debugPoset = checkEnabled // also turn on advanced self-checking in prove's datastructure - return "" - } - if phase == "check" && flag == "off" { - checkEnabled = val == 0 - debugPoset = checkEnabled - return "" - } - if phase == "check" && flag == "seed" { - checkEnabled = true - checkRandSeed = val - debugPoset = checkEnabled - return "" + if phase == "check" { + switch flag { + case "on": + checkEnabled = val != 0 + debugPoset = checkEnabled // also turn on advanced self-checking in prove's datastructure + return "" + case "off": + checkEnabled = val == 0 + debugPoset = checkEnabled + return "" + case "seed": + checkEnabled = true + checkRandSeed = val + debugPoset = checkEnabled + return "" + } } alltime := false allmem := false alldump := false if phase == "all" { - if flag == "time" { + switch flag { + case "time": alltime = val != 0 - } else if flag == "mem" { + case "mem": allmem = val != 0 - } else if flag == "dump" { + case "dump": alldump = val != 0 if alldump { BuildDump = valString } - } else { + default: return fmt.Sprintf("Did not find a flag matching %s in -d=ssa/%s debug option", flag, phase) } } @@ -419,7 +431,7 @@ var passes = [...]pass{ {name: "early copyelim", fn: copyelim}, {name: "early deadcode", fn: deadcode}, // remove generated dead code to avoid doing pointless work during opt {name: "short circuit", fn: shortcircuit}, - {name: "decompose args", fn: decomposeArgs, required: true}, + {name: "decompose args", fn: decomposeArgs, required: !go116lateCallExpansion, disabled: go116lateCallExpansion}, // handled by late call lowering {name: "decompose user", fn: decomposeUser, required: true}, {name: "pre-opt deadcode", fn: deadcode}, {name: "opt", fn: opt, required: true}, // NB: some generic rules know the name of the opt pass. TODO: split required rules and optimizing rules @@ -432,6 +444,7 @@ var passes = [...]pass{ {name: "prove", fn: prove}, {name: "early fuse", fn: fuseEarly}, {name: "decompose builtin", fn: decomposeBuiltIn, required: true}, + {name: "expand calls", fn: expandCalls, required: true}, {name: "softfloat", fn: softfloat, required: true}, {name: "late opt", fn: opt, required: true}, // TODO: split required rules and optimizing rules {name: "dead auto elim", fn: elimDeadAutosGeneric}, diff --git a/src/cmd/compile/internal/ssa/config.go b/src/cmd/compile/internal/ssa/config.go index 4b2f06def11..0fe0337ddfa 100644 --- a/src/cmd/compile/internal/ssa/config.go +++ b/src/cmd/compile/internal/ssa/config.go @@ -38,7 +38,6 @@ type Config struct { useSSE bool // Use SSE for non-float operations useAvg bool // Use optimizations that need Avg* operations useHmul bool // Use optimizations that need Hmul* operations - use387 bool // GO386=387 SoftFloat bool // Race bool // race detector enabled NeedsFpScratch bool // No direct move between GP and FP register sets @@ -150,6 +149,7 @@ type Frontend interface { SplitStruct(LocalSlot, int) LocalSlot SplitArray(LocalSlot) LocalSlot // array must be length 1 SplitInt64(LocalSlot) (LocalSlot, LocalSlot) // returns (hi, lo) + SplitSlot(parent *LocalSlot, suffix string, offset int64, t *types.Type) LocalSlot // DerefItab dereferences an itab function // entry, given the symbol of the itab and @@ -196,6 +196,14 @@ const ( ClassParamOut // return value ) +const go116lateCallExpansion = true + +// LateCallExpansionEnabledWithin returns true if late call expansion should be tested +// within compilation of a function/method. +func LateCallExpansionEnabledWithin(f *Func) bool { + return go116lateCallExpansion +} + // NewConfig returns a new configuration object for the given architecture. func NewConfig(arch string, types Types, ctxt *obj.Link, optimize bool) *Config { c := &Config{arch: arch, Types: types} @@ -248,7 +256,7 @@ func NewConfig(arch string, types Types, ctxt *obj.Link, optimize bool) *Config c.FPReg = framepointerRegARM64 c.LinkReg = linkRegARM64 c.hasGReg = true - c.noDuffDevice = objabi.GOOS == "darwin" // darwin linker cannot handle BR26 reloc with non-zero addend + c.noDuffDevice = objabi.GOOS == "darwin" || objabi.GOOS == "ios" // darwin linker cannot handle BR26 reloc with non-zero addend case "ppc64": c.BigEndian = true fallthrough @@ -379,9 +387,4 @@ func NewConfig(arch string, types Types, ctxt *obj.Link, optimize bool) *Config return c } -func (c *Config) Set387(b bool) { - c.NeedsFpScratch = b - c.use387 = b -} - func (c *Config) Ctxt() *obj.Link { return c.ctxt } diff --git a/src/cmd/compile/internal/ssa/decompose.go b/src/cmd/compile/internal/ssa/decompose.go index ab27ba85ae7..bf7f1e826b1 100644 --- a/src/cmd/compile/internal/ssa/decompose.go +++ b/src/cmd/compile/internal/ssa/decompose.go @@ -6,6 +6,7 @@ package ssa import ( "cmd/compile/internal/types" + "sort" ) // decompose converts phi ops on compound builtin types into phi @@ -31,77 +32,79 @@ func decomposeBuiltIn(f *Func) { } // Split up named values into their components. + // accumulate old names for aggregates (that are decomposed) in toDelete for efficient bulk deletion, + // accumulate new LocalSlots in newNames for addition after the iteration. This decomposition is for + // builtin types with leaf components, and thus there is no need to reprocess the newly create LocalSlots. + var toDelete []namedVal var newNames []LocalSlot - for _, name := range f.Names { + for i, name := range f.Names { t := name.Type switch { case t.IsInteger() && t.Size() > f.Config.RegSize: hiName, loName := f.fe.SplitInt64(name) newNames = append(newNames, hiName, loName) - for _, v := range f.NamedValues[name] { + for j, v := range f.NamedValues[name] { if v.Op != OpInt64Make { continue } f.NamedValues[hiName] = append(f.NamedValues[hiName], v.Args[0]) f.NamedValues[loName] = append(f.NamedValues[loName], v.Args[1]) + toDelete = append(toDelete, namedVal{i, j}) } - delete(f.NamedValues, name) case t.IsComplex(): rName, iName := f.fe.SplitComplex(name) newNames = append(newNames, rName, iName) - for _, v := range f.NamedValues[name] { + for j, v := range f.NamedValues[name] { if v.Op != OpComplexMake { continue } f.NamedValues[rName] = append(f.NamedValues[rName], v.Args[0]) f.NamedValues[iName] = append(f.NamedValues[iName], v.Args[1]) - + toDelete = append(toDelete, namedVal{i, j}) } - delete(f.NamedValues, name) case t.IsString(): ptrName, lenName := f.fe.SplitString(name) newNames = append(newNames, ptrName, lenName) - for _, v := range f.NamedValues[name] { + for j, v := range f.NamedValues[name] { if v.Op != OpStringMake { continue } f.NamedValues[ptrName] = append(f.NamedValues[ptrName], v.Args[0]) f.NamedValues[lenName] = append(f.NamedValues[lenName], v.Args[1]) + toDelete = append(toDelete, namedVal{i, j}) } - delete(f.NamedValues, name) case t.IsSlice(): ptrName, lenName, capName := f.fe.SplitSlice(name) newNames = append(newNames, ptrName, lenName, capName) - for _, v := range f.NamedValues[name] { + for j, v := range f.NamedValues[name] { if v.Op != OpSliceMake { continue } f.NamedValues[ptrName] = append(f.NamedValues[ptrName], v.Args[0]) f.NamedValues[lenName] = append(f.NamedValues[lenName], v.Args[1]) f.NamedValues[capName] = append(f.NamedValues[capName], v.Args[2]) + toDelete = append(toDelete, namedVal{i, j}) } - delete(f.NamedValues, name) case t.IsInterface(): typeName, dataName := f.fe.SplitInterface(name) newNames = append(newNames, typeName, dataName) - for _, v := range f.NamedValues[name] { + for j, v := range f.NamedValues[name] { if v.Op != OpIMake { continue } f.NamedValues[typeName] = append(f.NamedValues[typeName], v.Args[0]) f.NamedValues[dataName] = append(f.NamedValues[dataName], v.Args[1]) + toDelete = append(toDelete, namedVal{i, j}) } - delete(f.NamedValues, name) case t.IsFloat(): // floats are never decomposed, even ones bigger than RegSize - newNames = append(newNames, name) case t.Size() > f.Config.RegSize: f.Fatalf("undecomposed named type %s %v", name, t) - default: - newNames = append(newNames, name) } } - f.Names = newNames + + deleteNamedVals(f, toDelete) + f.Names = append(f.Names, newNames...) } func decomposeBuiltInPhi(v *Value) { @@ -263,14 +266,20 @@ func decomposeUserArrayInto(f *Func, name LocalSlot, slots []LocalSlot) []LocalS f.Fatalf("array not of size 1") } elemName := f.fe.SplitArray(name) + var keep []*Value for _, v := range f.NamedValues[name] { if v.Op != OpArrayMake1 { + keep = append(keep, v) continue } f.NamedValues[elemName] = append(f.NamedValues[elemName], v.Args[0]) } - // delete the name for the array as a whole - delete(f.NamedValues, name) + if len(keep) == 0 { + // delete the name for the array as a whole + delete(f.NamedValues, name) + } else { + f.NamedValues[name] = keep + } if t.Elem().IsArray() { return decomposeUserArrayInto(f, elemName, slots) @@ -300,17 +309,23 @@ func decomposeUserStructInto(f *Func, name LocalSlot, slots []LocalSlot) []Local } makeOp := StructMakeOp(n) + var keep []*Value // create named values for each struct field for _, v := range f.NamedValues[name] { if v.Op != makeOp { + keep = append(keep, v) continue } for i := 0; i < len(fnames); i++ { f.NamedValues[fnames[i]] = append(f.NamedValues[fnames[i]], v.Args[i]) } } - // remove the name of the struct as a whole - delete(f.NamedValues, name) + if len(keep) == 0 { + // delete the name for the struct as a whole + delete(f.NamedValues, name) + } else { + f.NamedValues[name] = keep + } // now that this f.NamedValues contains values for the struct // fields, recurse into nested structs @@ -400,3 +415,35 @@ func StructMakeOp(nf int) Op { } panic("too many fields in an SSAable struct") } + +type namedVal struct { + locIndex, valIndex int // f.NamedValues[f.Names[locIndex]][valIndex] = key +} + +// deleteNamedVals removes particular values with debugger names from f's naming data structures +func deleteNamedVals(f *Func, toDelete []namedVal) { + // Arrange to delete from larger indices to smaller, to ensure swap-with-end deletion does not invalid pending indices. + sort.Slice(toDelete, func(i, j int) bool { + if toDelete[i].locIndex != toDelete[j].locIndex { + return toDelete[i].locIndex > toDelete[j].locIndex + } + return toDelete[i].valIndex > toDelete[j].valIndex + + }) + + // Get rid of obsolete names + for _, d := range toDelete { + loc := f.Names[d.locIndex] + vals := f.NamedValues[loc] + l := len(vals) - 1 + if l > 0 { + vals[d.valIndex] = vals[l] + f.NamedValues[loc] = vals[:l] + } else { + delete(f.NamedValues, loc) + l = len(f.Names) - 1 + f.Names[d.locIndex] = f.Names[l] + f.Names = f.Names[:l] + } + } +} diff --git a/src/cmd/compile/internal/ssa/expand_calls.go b/src/cmd/compile/internal/ssa/expand_calls.go new file mode 100644 index 00000000000..fbde19d94c2 --- /dev/null +++ b/src/cmd/compile/internal/ssa/expand_calls.go @@ -0,0 +1,974 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package ssa + +import ( + "cmd/compile/internal/types" + "cmd/internal/src" + "fmt" + "sort" +) + +type selKey struct { + from *Value + offset int64 + size int64 + typ *types.Type +} + +type offsetKey struct { + from *Value + offset int64 + pt *types.Type +} + +// expandCalls converts LE (Late Expansion) calls that act like they receive value args into a lower-level form +// that is more oriented to a platform's ABI. The SelectN operations that extract results are rewritten into +// more appropriate forms, and any StructMake or ArrayMake inputs are decomposed until non-struct values are +// reached. On the callee side, OpArg nodes are not decomposed until this phase is run. +// TODO results should not be lowered until this phase. +func expandCalls(f *Func) { + // Calls that need lowering have some number of inputs, including a memory input, + // and produce a tuple of (value1, value2, ..., mem) where valueK may or may not be SSA-able. + + // With the current ABI those inputs need to be converted into stores to memory, + // rethreading the call's memory input to the first, and the new call now receiving the last. + + // With the current ABI, the outputs need to be converted to loads, which will all use the call's + // memory output as their input. + if !LateCallExpansionEnabledWithin(f) { + return + } + debug := f.pass.debug > 0 + + if debug { + fmt.Printf("\nexpandsCalls(%s)\n", f.Name) + } + + canSSAType := f.fe.CanSSA + regSize := f.Config.RegSize + sp, _ := f.spSb() + typ := &f.Config.Types + ptrSize := f.Config.PtrSize + + // For 32-bit, need to deal with decomposition of 64-bit integers, which depends on endianness. + var hiOffset, lowOffset int64 + if f.Config.BigEndian { + lowOffset = 4 + } else { + hiOffset = 4 + } + + namedSelects := make(map[*Value][]namedVal) + + sdom := f.Sdom() + + common := make(map[selKey]*Value) + + // intPairTypes returns the pair of 32-bit int types needed to encode a 64-bit integer type on a target + // that has no 64-bit integer registers. + intPairTypes := func(et types.EType) (tHi, tLo *types.Type) { + tHi = typ.UInt32 + if et == types.TINT64 { + tHi = typ.Int32 + } + tLo = typ.UInt32 + return + } + + // isAlreadyExpandedAggregateType returns whether a type is an SSA-able "aggregate" (multiple register) type + // that was expanded in an earlier phase (currently, expand_calls is intended to run after decomposeBuiltin, + // so this is all aggregate types -- small struct and array, complex, interface, string, slice, and 64-bit + // integer on 32-bit). + isAlreadyExpandedAggregateType := func(t *types.Type) bool { + if !canSSAType(t) { + return false + } + return t.IsStruct() || t.IsArray() || t.IsComplex() || t.IsInterface() || t.IsString() || t.IsSlice() || + t.Size() > regSize && t.IsInteger() + } + + offsets := make(map[offsetKey]*Value) + + // offsetFrom creates an offset from a pointer, simplifying chained offsets and offsets from SP + // TODO should also optimize offsets from SB? + offsetFrom := func(from *Value, offset int64, pt *types.Type) *Value { + if offset == 0 && from.Type == pt { // this is not actually likely + return from + } + // Simplify, canonicalize + for from.Op == OpOffPtr { + offset += from.AuxInt + from = from.Args[0] + } + if from == sp { + return f.ConstOffPtrSP(pt, offset, sp) + } + key := offsetKey{from, offset, pt} + v := offsets[key] + if v != nil { + return v + } + v = from.Block.NewValue1I(from.Pos.WithNotStmt(), OpOffPtr, pt, offset, from) + offsets[key] = v + return v + } + + // splitSlots splits one "field" (specified by sfx, offset, and ty) out of the LocalSlots in ls and returns the new LocalSlots this generates. + splitSlots := func(ls []LocalSlot, sfx string, offset int64, ty *types.Type) []LocalSlot { + var locs []LocalSlot + for i := range ls { + locs = append(locs, f.fe.SplitSlot(&ls[i], sfx, offset, ty)) + } + return locs + } + + // removeTrivialWrapperTypes unwraps layers of + // struct { singleField SomeType } and [1]SomeType + // until a non-wrapper type is reached. This is useful + // for working with assignments to/from interface data + // fields (either second operand to OpIMake or OpIData) + // where the wrapping or type conversion can be elided + // because of type conversions/assertions in source code + // that do not appear in SSA. + removeTrivialWrapperTypes := func(t *types.Type) *types.Type { + for { + if t.IsStruct() && t.NumFields() == 1 { + t = t.Field(0).Type + continue + } + if t.IsArray() && t.NumElem() == 1 { + t = t.Elem() + continue + } + break + } + return t + } + + // Calls that need lowering have some number of inputs, including a memory input, + // and produce a tuple of (value1, value2, ..., mem) where valueK may or may not be SSA-able. + + // With the current ABI those inputs need to be converted into stores to memory, + // rethreading the call's memory input to the first, and the new call now receiving the last. + + // With the current ABI, the outputs need to be converted to loads, which will all use the call's + // memory output as their input. + + // rewriteSelect recursively walks from leaf selector to a root (OpSelectN, OpLoad, OpArg) + // through a chain of Struct/Array/builtin Select operations. If the chain of selectors does not + // end in an expected root, it does nothing (this can happen depending on compiler phase ordering). + // The "leaf" provides the type, the root supplies the container, and the leaf-to-root path + // accumulates the offset. + // It emits the code necessary to implement the leaf select operation that leads to the root. + // + // TODO when registers really arrive, must also decompose anything split across two registers or registers and memory. + var rewriteSelect func(leaf *Value, selector *Value, offset int64) []LocalSlot + rewriteSelect = func(leaf *Value, selector *Value, offset int64) []LocalSlot { + if debug { + fmt.Printf("rewriteSelect(%s, %s, %d)\n", leaf.LongString(), selector.LongString(), offset) + } + var locs []LocalSlot + leafType := leaf.Type + if len(selector.Args) > 0 { + w := selector.Args[0] + if w.Op == OpCopy { + for w.Op == OpCopy { + w = w.Args[0] + } + selector.SetArg(0, w) + } + } + switch selector.Op { + case OpArg: + if !isAlreadyExpandedAggregateType(selector.Type) { + if leafType == selector.Type { // OpIData leads us here, sometimes. + leaf.copyOf(selector) + } else { + f.Fatalf("Unexpected OpArg type, selector=%s, leaf=%s\n", selector.LongString(), leaf.LongString()) + } + if debug { + fmt.Printf("\tOpArg, break\n") + } + break + } + if leaf.Op == OpIData { + leafType = removeTrivialWrapperTypes(leaf.Type) + } + aux := selector.Aux + auxInt := selector.AuxInt + offset + if leaf.Block == selector.Block { + leaf.reset(OpArg) + leaf.Aux = aux + leaf.AuxInt = auxInt + leaf.Type = leafType + } else { + w := selector.Block.NewValue0IA(leaf.Pos, OpArg, leafType, auxInt, aux) + leaf.copyOf(w) + if debug { + fmt.Printf("\tnew %s\n", w.LongString()) + } + } + for _, s := range namedSelects[selector] { + locs = append(locs, f.Names[s.locIndex]) + } + + case OpLoad: // We end up here because of IData of immediate structures. + // Failure case: + // (note the failure case is very rare; w/o this case, make.bash and run.bash both pass, as well as + // the hard cases of building {syscall,math,math/cmplx,math/bits,go/constant} on ppc64le and mips-softfloat). + // + // GOSSAFUNC='(*dumper).dump' go build -gcflags=-l -tags=math_big_pure_go cmd/compile/internal/gc + // cmd/compile/internal/gc/dump.go:136:14: internal compiler error: '(*dumper).dump': not lowered: v827, StructSelect PTR PTR + // b2: ← b1 + // v20 (+142) = StaticLECall {AuxCall{reflect.Value.Interface([reflect.Value,0])[interface {},24]}} [40] v8 v1 + // v21 (142) = SelectN [1] v20 + // v22 (142) = SelectN [0] v20 + // b15: ← b8 + // v71 (+143) = IData v22 (v[Nodes]) + // v73 (+146) = StaticLECall <[]*Node,mem> {AuxCall{"".Nodes.Slice([Nodes,0])[[]*Node,8]}} [32] v71 v21 + // + // translates (w/o the "case OpLoad:" above) to: + // + // b2: ← b1 + // v20 (+142) = StaticCall {AuxCall{reflect.Value.Interface([reflect.Value,0])[interface {},24]}} [40] v715 + // v23 (142) = Load <*uintptr> v19 v20 + // v823 (142) = IsNonNil v23 + // v67 (+143) = Load <*[]*Node> v880 v20 + // b15: ← b8 + // v827 (146) = StructSelect <*[]*Node> [0] v67 + // v846 (146) = Store {*[]*Node} v769 v827 v20 + // v73 (+146) = StaticCall {AuxCall{"".Nodes.Slice([Nodes,0])[[]*Node,8]}} [32] v846 + // i.e., the struct select is generated and remains in because it is not applied to an actual structure. + // The OpLoad was created to load the single field of the IData + // This case removes that StructSelect. + if leafType != selector.Type { + f.Fatalf("Unexpected Load as selector, leaf=%s, selector=%s\n", leaf.LongString(), selector.LongString()) + } + leaf.copyOf(selector) + for _, s := range namedSelects[selector] { + locs = append(locs, f.Names[s.locIndex]) + } + + case OpSelectN: + // TODO these may be duplicated. Should memoize. Intermediate selectors will go dead, no worries there. + call := selector.Args[0] + aux := call.Aux.(*AuxCall) + which := selector.AuxInt + if which == aux.NResults() { // mem is after the results. + // rewrite v as a Copy of call -- the replacement call will produce a mem. + leaf.copyOf(call) + } else { + leafType := removeTrivialWrapperTypes(leaf.Type) + if canSSAType(leafType) { + pt := types.NewPtr(leafType) + off := offsetFrom(sp, offset+aux.OffsetOfResult(which), pt) + // Any selection right out of the arg area/registers has to be same Block as call, use call as mem input. + if leaf.Block == call.Block { + leaf.reset(OpLoad) + leaf.SetArgs2(off, call) + leaf.Type = leafType + } else { + w := call.Block.NewValue2(leaf.Pos, OpLoad, leafType, off, call) + leaf.copyOf(w) + if debug { + fmt.Printf("\tnew %s\n", w.LongString()) + } + } + for _, s := range namedSelects[selector] { + locs = append(locs, f.Names[s.locIndex]) + } + } else { + f.Fatalf("Should not have non-SSA-able OpSelectN, selector=%s", selector.LongString()) + } + } + + case OpStructSelect: + w := selector.Args[0] + var ls []LocalSlot + if w.Type.Etype != types.TSTRUCT { // IData artifact + ls = rewriteSelect(leaf, w, offset) + } else { + ls = rewriteSelect(leaf, w, offset+w.Type.FieldOff(int(selector.AuxInt))) + if w.Op != OpIData { + for _, l := range ls { + locs = append(locs, f.fe.SplitStruct(l, int(selector.AuxInt))) + } + } + } + + case OpArraySelect: + w := selector.Args[0] + rewriteSelect(leaf, w, offset+selector.Type.Size()*selector.AuxInt) + + case OpInt64Hi: + w := selector.Args[0] + ls := rewriteSelect(leaf, w, offset+hiOffset) + locs = splitSlots(ls, ".hi", hiOffset, leafType) + + case OpInt64Lo: + w := selector.Args[0] + ls := rewriteSelect(leaf, w, offset+lowOffset) + locs = splitSlots(ls, ".lo", lowOffset, leafType) + + case OpStringPtr: + ls := rewriteSelect(leaf, selector.Args[0], offset) + locs = splitSlots(ls, ".ptr", 0, typ.BytePtr) + + case OpSlicePtr: + w := selector.Args[0] + ls := rewriteSelect(leaf, w, offset) + locs = splitSlots(ls, ".ptr", 0, types.NewPtr(w.Type.Elem())) + + case OpITab: + w := selector.Args[0] + ls := rewriteSelect(leaf, w, offset) + sfx := ".itab" + if w.Type.IsEmptyInterface() { + sfx = ".type" + } + locs = splitSlots(ls, sfx, 0, typ.Uintptr) + + case OpComplexReal: + ls := rewriteSelect(leaf, selector.Args[0], offset) + locs = splitSlots(ls, ".real", 0, leafType) + + case OpComplexImag: + ls := rewriteSelect(leaf, selector.Args[0], offset+leafType.Width) // result is FloatNN, width of result is offset of imaginary part. + locs = splitSlots(ls, ".imag", leafType.Width, leafType) + + case OpStringLen, OpSliceLen: + ls := rewriteSelect(leaf, selector.Args[0], offset+ptrSize) + locs = splitSlots(ls, ".len", ptrSize, leafType) + + case OpIData: + ls := rewriteSelect(leaf, selector.Args[0], offset+ptrSize) + locs = splitSlots(ls, ".data", ptrSize, leafType) + + case OpSliceCap: + ls := rewriteSelect(leaf, selector.Args[0], offset+2*ptrSize) + locs = splitSlots(ls, ".cap", 2*ptrSize, leafType) + + case OpCopy: // If it's an intermediate result, recurse + locs = rewriteSelect(leaf, selector.Args[0], offset) + for _, s := range namedSelects[selector] { + // this copy may have had its own name, preserve that, too. + locs = append(locs, f.Names[s.locIndex]) + } + + default: + // Ignore dead ends. These can occur if this phase is run before decompose builtin (which is not intended, but allowed). + } + + return locs + } + + // storeArgOrLoad converts stores of SSA-able aggregate arguments (passed to a call) into a series of primitive-typed + // stores of non-aggregate types. It recursively walks up a chain of selectors until it reaches a Load or an Arg. + // If it does not reach a Load or an Arg, nothing happens; this allows a little freedom in phase ordering. + var storeArgOrLoad func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offset int64) *Value + + // decomposeArgOrLoad is a helper for storeArgOrLoad. + // It decomposes a Load or an Arg into smaller parts, parameterized by the decomposeOne and decomposeTwo functions + // passed to it, and returns the new mem. If the type does not match one of the expected aggregate types, it returns nil instead. + decomposeArgOrLoad := func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offset int64, + decomposeOne func(pos src.XPos, b *Block, base, source, mem *Value, t1 *types.Type, offArg, offStore int64) *Value, + decomposeTwo func(pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value) *Value { + u := source.Type + switch u.Etype { + case types.TARRAY: + elem := u.Elem() + for i := int64(0); i < u.NumElem(); i++ { + elemOff := i * elem.Size() + mem = decomposeOne(pos, b, base, source, mem, elem, source.AuxInt+elemOff, offset+elemOff) + pos = pos.WithNotStmt() + } + return mem + case types.TSTRUCT: + for i := 0; i < u.NumFields(); i++ { + fld := u.Field(i) + mem = decomposeOne(pos, b, base, source, mem, fld.Type, source.AuxInt+fld.Offset, offset+fld.Offset) + pos = pos.WithNotStmt() + } + return mem + case types.TINT64, types.TUINT64: + if t.Width == regSize { + break + } + tHi, tLo := intPairTypes(t.Etype) + mem = decomposeOne(pos, b, base, source, mem, tHi, source.AuxInt+hiOffset, offset+hiOffset) + pos = pos.WithNotStmt() + return decomposeOne(pos, b, base, source, mem, tLo, source.AuxInt+lowOffset, offset+lowOffset) + case types.TINTER: + return decomposeTwo(pos, b, base, source, mem, typ.Uintptr, typ.BytePtr, source.AuxInt, offset) + case types.TSTRING: + return decomposeTwo(pos, b, base, source, mem, typ.BytePtr, typ.Int, source.AuxInt, offset) + case types.TCOMPLEX64: + return decomposeTwo(pos, b, base, source, mem, typ.Float32, typ.Float32, source.AuxInt, offset) + case types.TCOMPLEX128: + return decomposeTwo(pos, b, base, source, mem, typ.Float64, typ.Float64, source.AuxInt, offset) + case types.TSLICE: + mem = decomposeTwo(pos, b, base, source, mem, typ.BytePtr, typ.Int, source.AuxInt, offset) + return decomposeOne(pos, b, base, source, mem, typ.Int, source.AuxInt+2*ptrSize, offset+2*ptrSize) + } + return nil + } + + // storeOneArg creates a decomposed (one step) arg that is then stored. + // pos and b locate the store instruction, base is the base of the store target, source is the "base" of the value input, + // mem is the input mem, t is the type in question, and offArg and offStore are the offsets from the respective bases. + storeOneArg := func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offArg, offStore int64) *Value { + w := common[selKey{source, offArg, t.Width, t}] + if w == nil { + w = source.Block.NewValue0IA(source.Pos, OpArg, t, offArg, source.Aux) + common[selKey{source, offArg, t.Width, t}] = w + } + return storeArgOrLoad(pos, b, base, w, mem, t, offStore) + } + + // storeOneLoad creates a decomposed (one step) load that is then stored. + storeOneLoad := func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offArg, offStore int64) *Value { + from := offsetFrom(source.Args[0], offArg, types.NewPtr(t)) + w := source.Block.NewValue2(source.Pos, OpLoad, t, from, mem) + return storeArgOrLoad(pos, b, base, w, mem, t, offStore) + } + + storeTwoArg := func(pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value { + mem = storeOneArg(pos, b, base, source, mem, t1, offArg, offStore) + pos = pos.WithNotStmt() + t1Size := t1.Size() + return storeOneArg(pos, b, base, source, mem, t2, offArg+t1Size, offStore+t1Size) + } + + storeTwoLoad := func(pos src.XPos, b *Block, base, source, mem *Value, t1, t2 *types.Type, offArg, offStore int64) *Value { + mem = storeOneLoad(pos, b, base, source, mem, t1, offArg, offStore) + pos = pos.WithNotStmt() + t1Size := t1.Size() + return storeOneLoad(pos, b, base, source, mem, t2, offArg+t1Size, offStore+t1Size) + } + + storeArgOrLoad = func(pos src.XPos, b *Block, base, source, mem *Value, t *types.Type, offset int64) *Value { + if debug { + fmt.Printf("\tstoreArgOrLoad(%s; %s; %s; %s; %d)\n", base.LongString(), source.LongString(), mem.String(), t.String(), offset) + } + + switch source.Op { + case OpCopy: + return storeArgOrLoad(pos, b, base, source.Args[0], mem, t, offset) + + case OpLoad: + ret := decomposeArgOrLoad(pos, b, base, source, mem, t, offset, storeOneLoad, storeTwoLoad) + if ret != nil { + return ret + } + + case OpArg: + ret := decomposeArgOrLoad(pos, b, base, source, mem, t, offset, storeOneArg, storeTwoArg) + if ret != nil { + return ret + } + + case OpArrayMake0, OpStructMake0: + return mem + + case OpStructMake1, OpStructMake2, OpStructMake3, OpStructMake4: + for i := 0; i < t.NumFields(); i++ { + fld := t.Field(i) + mem = storeArgOrLoad(pos, b, base, source.Args[i], mem, fld.Type, offset+fld.Offset) + pos = pos.WithNotStmt() + } + return mem + + case OpArrayMake1: + return storeArgOrLoad(pos, b, base, source.Args[0], mem, t.Elem(), offset) + + case OpInt64Make: + tHi, tLo := intPairTypes(t.Etype) + mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, tHi, offset+hiOffset) + pos = pos.WithNotStmt() + return storeArgOrLoad(pos, b, base, source.Args[1], mem, tLo, offset+lowOffset) + + case OpComplexMake: + tPart := typ.Float32 + wPart := t.Width / 2 + if wPart == 8 { + tPart = typ.Float64 + } + mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, tPart, offset) + pos = pos.WithNotStmt() + return storeArgOrLoad(pos, b, base, source.Args[1], mem, tPart, offset+wPart) + + case OpIMake: + mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, typ.Uintptr, offset) + pos = pos.WithNotStmt() + return storeArgOrLoad(pos, b, base, source.Args[1], mem, typ.BytePtr, offset+ptrSize) + + case OpStringMake: + mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, typ.BytePtr, offset) + pos = pos.WithNotStmt() + return storeArgOrLoad(pos, b, base, source.Args[1], mem, typ.Int, offset+ptrSize) + + case OpSliceMake: + mem = storeArgOrLoad(pos, b, base, source.Args[0], mem, typ.BytePtr, offset) + pos = pos.WithNotStmt() + mem = storeArgOrLoad(pos, b, base, source.Args[1], mem, typ.Int, offset+ptrSize) + return storeArgOrLoad(pos, b, base, source.Args[2], mem, typ.Int, offset+2*ptrSize) + } + + // For nodes that cannot be taken apart -- OpSelectN, other structure selectors. + switch t.Etype { + case types.TARRAY: + elt := t.Elem() + if source.Type != t && t.NumElem() == 1 && elt.Width == t.Width && t.Width == regSize { + t = removeTrivialWrapperTypes(t) + // it could be a leaf type, but the "leaf" could be complex64 (for example) + return storeArgOrLoad(pos, b, base, source, mem, t, offset) + } + for i := int64(0); i < t.NumElem(); i++ { + sel := source.Block.NewValue1I(pos, OpArraySelect, elt, i, source) + mem = storeArgOrLoad(pos, b, base, sel, mem, elt, offset+i*elt.Width) + pos = pos.WithNotStmt() + } + return mem + + case types.TSTRUCT: + if source.Type != t && t.NumFields() == 1 && t.Field(0).Type.Width == t.Width && t.Width == regSize { + // This peculiar test deals with accesses to immediate interface data. + // It works okay because everything is the same size. + // Example code that triggers this can be found in go/constant/value.go, function ToComplex + // v119 (+881) = IData v6 + // v121 (+882) = StaticLECall {AuxCall{"".itof([intVal,0])[floatVal,8]}} [16] v119 v1 + // This corresponds to the generic rewrite rule "(StructSelect [0] (IData x)) => (IData x)" + // Guard against "struct{struct{*foo}}" + // Other rewriting phases create minor glitches when they transform IData, for instance the + // interface-typed Arg "x" of ToFloat in go/constant/value.go + // v6 (858) = Arg {x} (x[Value], x[Value]) + // is rewritten by decomposeArgs into + // v141 (858) = Arg {x} + // v139 (858) = Arg <*uint8> {x} [8] + // because of a type case clause on line 862 of go/constant/value.go + // case intVal: + // return itof(x) + // v139 is later stored as an intVal == struct{val *big.Int} which naively requires the fields of + // of a *uint8, which does not succeed. + t = removeTrivialWrapperTypes(t) + // it could be a leaf type, but the "leaf" could be complex64 (for example) + return storeArgOrLoad(pos, b, base, source, mem, t, offset) + } + + for i := 0; i < t.NumFields(); i++ { + fld := t.Field(i) + sel := source.Block.NewValue1I(pos, OpStructSelect, fld.Type, int64(i), source) + mem = storeArgOrLoad(pos, b, base, sel, mem, fld.Type, offset+fld.Offset) + pos = pos.WithNotStmt() + } + return mem + + case types.TINT64, types.TUINT64: + if t.Width == regSize { + break + } + tHi, tLo := intPairTypes(t.Etype) + sel := source.Block.NewValue1(pos, OpInt64Hi, tHi, source) + mem = storeArgOrLoad(pos, b, base, sel, mem, tHi, offset+hiOffset) + pos = pos.WithNotStmt() + sel = source.Block.NewValue1(pos, OpInt64Lo, tLo, source) + return storeArgOrLoad(pos, b, base, sel, mem, tLo, offset+lowOffset) + + case types.TINTER: + sel := source.Block.NewValue1(pos, OpITab, typ.BytePtr, source) + mem = storeArgOrLoad(pos, b, base, sel, mem, typ.BytePtr, offset) + pos = pos.WithNotStmt() + sel = source.Block.NewValue1(pos, OpIData, typ.BytePtr, source) + return storeArgOrLoad(pos, b, base, sel, mem, typ.BytePtr, offset+ptrSize) + + case types.TSTRING: + sel := source.Block.NewValue1(pos, OpStringPtr, typ.BytePtr, source) + mem = storeArgOrLoad(pos, b, base, sel, mem, typ.BytePtr, offset) + pos = pos.WithNotStmt() + sel = source.Block.NewValue1(pos, OpStringLen, typ.Int, source) + return storeArgOrLoad(pos, b, base, sel, mem, typ.Int, offset+ptrSize) + + case types.TSLICE: + et := types.NewPtr(t.Elem()) + sel := source.Block.NewValue1(pos, OpSlicePtr, et, source) + mem = storeArgOrLoad(pos, b, base, sel, mem, et, offset) + pos = pos.WithNotStmt() + sel = source.Block.NewValue1(pos, OpSliceLen, typ.Int, source) + mem = storeArgOrLoad(pos, b, base, sel, mem, typ.Int, offset+ptrSize) + sel = source.Block.NewValue1(pos, OpSliceCap, typ.Int, source) + return storeArgOrLoad(pos, b, base, sel, mem, typ.Int, offset+2*ptrSize) + + case types.TCOMPLEX64: + sel := source.Block.NewValue1(pos, OpComplexReal, typ.Float32, source) + mem = storeArgOrLoad(pos, b, base, sel, mem, typ.Float32, offset) + pos = pos.WithNotStmt() + sel = source.Block.NewValue1(pos, OpComplexImag, typ.Float32, source) + return storeArgOrLoad(pos, b, base, sel, mem, typ.Float32, offset+4) + + case types.TCOMPLEX128: + sel := source.Block.NewValue1(pos, OpComplexReal, typ.Float64, source) + mem = storeArgOrLoad(pos, b, base, sel, mem, typ.Float64, offset) + pos = pos.WithNotStmt() + sel = source.Block.NewValue1(pos, OpComplexImag, typ.Float64, source) + return storeArgOrLoad(pos, b, base, sel, mem, typ.Float64, offset+8) + } + + dst := offsetFrom(base, offset, types.NewPtr(t)) + x := b.NewValue3A(pos, OpStore, types.TypeMem, t, dst, source, mem) + if debug { + fmt.Printf("\t\tstoreArg returns %s\n", x.LongString()) + } + return x + } + + // rewriteArgs removes all the Args from a call and converts the call args into appropriate + // stores (or later, register movement). Extra args for interface and closure calls are ignored, + // but removed. + rewriteArgs := func(v *Value, firstArg int) *Value { + // Thread the stores on the memory arg + aux := v.Aux.(*AuxCall) + pos := v.Pos.WithNotStmt() + m0 := v.Args[len(v.Args)-1] + mem := m0 + for i, a := range v.Args { + if i < firstArg { + continue + } + if a == m0 { // mem is last. + break + } + auxI := int64(i - firstArg) + if a.Op == OpDereference { + if a.MemoryArg() != m0 { + f.Fatalf("Op...LECall and OpDereference have mismatched mem, %s and %s", v.LongString(), a.LongString()) + } + // "Dereference" of addressed (probably not-SSA-eligible) value becomes Move + // TODO this will be more complicated with registers in the picture. + source := a.Args[0] + dst := f.ConstOffPtrSP(source.Type, aux.OffsetOfArg(auxI), sp) + if a.Uses == 1 && a.Block == v.Block { + a.reset(OpMove) + a.Pos = pos + a.Type = types.TypeMem + a.Aux = aux.TypeOfArg(auxI) + a.AuxInt = aux.SizeOfArg(auxI) + a.SetArgs3(dst, source, mem) + mem = a + } else { + mem = v.Block.NewValue3A(pos, OpMove, types.TypeMem, aux.TypeOfArg(auxI), dst, source, mem) + mem.AuxInt = aux.SizeOfArg(auxI) + } + } else { + if debug { + fmt.Printf("storeArg %s, %v, %d\n", a.LongString(), aux.TypeOfArg(auxI), aux.OffsetOfArg(auxI)) + } + mem = storeArgOrLoad(pos, v.Block, sp, a, mem, aux.TypeOfArg(auxI), aux.OffsetOfArg(auxI)) + } + } + v.resetArgs() + return mem + } + + // TODO if too slow, whole program iteration can be replaced w/ slices of appropriate values, accumulated in first loop here. + + // Step 0: rewrite the calls to convert incoming args to stores. + for _, b := range f.Blocks { + for _, v := range b.Values { + switch v.Op { + case OpStaticLECall: + mem := rewriteArgs(v, 0) + v.SetArgs1(mem) + case OpClosureLECall: + code := v.Args[0] + context := v.Args[1] + mem := rewriteArgs(v, 2) + v.SetArgs3(code, context, mem) + case OpInterLECall: + code := v.Args[0] + mem := rewriteArgs(v, 1) + v.SetArgs2(code, mem) + } + } + } + + for i, name := range f.Names { + t := name.Type + if isAlreadyExpandedAggregateType(t) { + for j, v := range f.NamedValues[name] { + if v.Op == OpSelectN || v.Op == OpArg && isAlreadyExpandedAggregateType(v.Type) { + ns := namedSelects[v] + namedSelects[v] = append(ns, namedVal{locIndex: i, valIndex: j}) + } + } + } + } + + // Step 1: any stores of aggregates remaining are believed to be sourced from call results or args. + // Decompose those stores into a series of smaller stores, adding selection ops as necessary. + for _, b := range f.Blocks { + for _, v := range b.Values { + if v.Op == OpStore { + t := v.Aux.(*types.Type) + source := v.Args[1] + tSrc := source.Type + iAEATt := isAlreadyExpandedAggregateType(t) + + if !iAEATt { + // guarding against store immediate struct into interface data field -- store type is *uint8 + // TODO can this happen recursively? + iAEATt = isAlreadyExpandedAggregateType(tSrc) + if iAEATt { + t = tSrc + } + } + if iAEATt { + if debug { + fmt.Printf("Splitting store %s\n", v.LongString()) + } + dst, mem := v.Args[0], v.Args[2] + mem = storeArgOrLoad(v.Pos, b, dst, source, mem, t, 0) + v.copyOf(mem) + } + } + } + } + + val2Preds := make(map[*Value]int32) // Used to accumulate dependency graph of selection operations for topological ordering. + + // Step 2: transform or accumulate selection operations for rewrite in topological order. + // + // Aggregate types that have already (in earlier phases) been transformed must be lowered comprehensively to finish + // the transformation (user-defined structs and arrays, slices, strings, interfaces, complex, 64-bit on 32-bit architectures), + // + // Any select-for-addressing applied to call results can be transformed directly. + for _, b := range f.Blocks { + for _, v := range b.Values { + // Accumulate chains of selectors for processing in topological order + switch v.Op { + case OpStructSelect, OpArraySelect, + OpIData, OpITab, + OpStringPtr, OpStringLen, + OpSlicePtr, OpSliceLen, OpSliceCap, + OpComplexReal, OpComplexImag, + OpInt64Hi, OpInt64Lo: + w := v.Args[0] + switch w.Op { + case OpStructSelect, OpArraySelect, OpSelectN, OpArg: + val2Preds[w] += 1 + if debug { + fmt.Printf("v2p[%s] = %d\n", w.LongString(), val2Preds[w]) + } + } + fallthrough + + case OpSelectN: + if _, ok := val2Preds[v]; !ok { + val2Preds[v] = 0 + if debug { + fmt.Printf("v2p[%s] = %d\n", v.LongString(), val2Preds[v]) + } + } + + case OpArg: + if !isAlreadyExpandedAggregateType(v.Type) { + continue + } + if _, ok := val2Preds[v]; !ok { + val2Preds[v] = 0 + if debug { + fmt.Printf("v2p[%s] = %d\n", v.LongString(), val2Preds[v]) + } + } + + case OpSelectNAddr: + // Do these directly, there are no chains of selectors. + call := v.Args[0] + which := v.AuxInt + aux := call.Aux.(*AuxCall) + pt := v.Type + off := offsetFrom(sp, aux.OffsetOfResult(which), pt) + v.copyOf(off) + } + } + } + + // Step 3: Compute topological order of selectors, + // then process it in reverse to eliminate duplicates, + // then forwards to rewrite selectors. + // + // All chains of selectors end up in same block as the call. + + // Compilation must be deterministic, so sort after extracting first zeroes from map. + // Sorting allows dominators-last order within each batch, + // so that the backwards scan for duplicates will most often find copies from dominating blocks (it is best-effort). + var toProcess []*Value + less := func(i, j int) bool { + vi, vj := toProcess[i], toProcess[j] + bi, bj := vi.Block, vj.Block + if bi == bj { + return vi.ID < vj.ID + } + return sdom.domorder(bi) > sdom.domorder(bj) // reverse the order to put dominators last. + } + + // Accumulate order in allOrdered + var allOrdered []*Value + for v, n := range val2Preds { + if n == 0 { + allOrdered = append(allOrdered, v) + } + } + last := 0 // allOrdered[0:last] has been top-sorted and processed + for len(val2Preds) > 0 { + toProcess = allOrdered[last:] + last = len(allOrdered) + sort.SliceStable(toProcess, less) + for _, v := range toProcess { + delete(val2Preds, v) + if v.Op == OpArg { + continue // no Args[0], hence done. + } + w := v.Args[0] + n, ok := val2Preds[w] + if !ok { + continue + } + if n == 1 { + allOrdered = append(allOrdered, w) + delete(val2Preds, w) + continue + } + val2Preds[w] = n - 1 + } + } + + common = make(map[selKey]*Value) + // Rewrite duplicate selectors as copies where possible. + for i := len(allOrdered) - 1; i >= 0; i-- { + v := allOrdered[i] + if v.Op == OpArg { + continue + } + w := v.Args[0] + if w.Op == OpCopy { + for w.Op == OpCopy { + w = w.Args[0] + } + v.SetArg(0, w) + } + typ := v.Type + if typ.IsMemory() { + continue // handled elsewhere, not an indexable result + } + size := typ.Width + offset := int64(0) + switch v.Op { + case OpStructSelect: + if w.Type.Etype == types.TSTRUCT { + offset = w.Type.FieldOff(int(v.AuxInt)) + } else { // Immediate interface data artifact, offset is zero. + f.Fatalf("Expand calls interface data problem, func %s, v=%s, w=%s\n", f.Name, v.LongString(), w.LongString()) + } + case OpArraySelect: + offset = size * v.AuxInt + case OpSelectN: + offset = w.Aux.(*AuxCall).OffsetOfResult(v.AuxInt) + case OpInt64Hi: + offset = hiOffset + case OpInt64Lo: + offset = lowOffset + case OpStringLen, OpSliceLen, OpIData: + offset = ptrSize + case OpSliceCap: + offset = 2 * ptrSize + case OpComplexImag: + offset = size + } + sk := selKey{from: w, size: size, offset: offset, typ: typ} + dupe := common[sk] + if dupe == nil { + common[sk] = v + } else if sdom.IsAncestorEq(dupe.Block, v.Block) { + v.copyOf(dupe) + } else { + // Because values are processed in dominator order, the old common[s] will never dominate after a miss is seen. + // Installing the new value might match some future values. + common[sk] = v + } + } + + // Indices of entries in f.Names that need to be deleted. + var toDelete []namedVal + + // Rewrite selectors. + for i, v := range allOrdered { + if debug { + b := v.Block + fmt.Printf("allOrdered[%d] = b%d, %s, uses=%d\n", i, b.ID, v.LongString(), v.Uses) + } + if v.Uses == 0 { + v.reset(OpInvalid) + continue + } + if v.Op == OpCopy { + continue + } + locs := rewriteSelect(v, v, 0) + // Install new names. + if v.Type.IsMemory() { + continue + } + // Leaf types may have debug locations + if !isAlreadyExpandedAggregateType(v.Type) { + for _, l := range locs { + f.NamedValues[l] = append(f.NamedValues[l], v) + } + f.Names = append(f.Names, locs...) + continue + } + // Not-leaf types that had debug locations need to lose them. + if ns, ok := namedSelects[v]; ok { + toDelete = append(toDelete, ns...) + } + } + + deleteNamedVals(f, toDelete) + + // Step 4: rewrite the calls themselves, correcting the type + for _, b := range f.Blocks { + for _, v := range b.Values { + switch v.Op { + case OpStaticLECall: + v.Op = OpStaticCall + v.Type = types.TypeMem + case OpClosureLECall: + v.Op = OpClosureCall + v.Type = types.TypeMem + case OpInterLECall: + v.Op = OpInterCall + v.Type = types.TypeMem + } + } + } + + // Step 5: elide any copies introduced. + for _, b := range f.Blocks { + for _, v := range b.Values { + for i, a := range v.Args { + if a.Op != OpCopy { + continue + } + aa := copySource(a) + v.SetArg(i, aa) + for a.Uses == 0 { + b := a.Args[0] + a.reset(OpInvalid) + a = b + } + } + } + } +} diff --git a/src/cmd/compile/internal/ssa/export_test.go b/src/cmd/compile/internal/ssa/export_test.go index 51665c60e27..b4c3e5cfdfb 100644 --- a/src/cmd/compile/internal/ssa/export_test.go +++ b/src/cmd/compile/internal/ssa/export_test.go @@ -125,6 +125,10 @@ func (d DummyFrontend) SplitStruct(s LocalSlot, i int) LocalSlot { func (d DummyFrontend) SplitArray(s LocalSlot) LocalSlot { return LocalSlot{N: s.N, Type: s.Type.Elem(), Off: s.Off} } + +func (d DummyFrontend) SplitSlot(parent *LocalSlot, suffix string, offset int64, t *types.Type) LocalSlot { + return LocalSlot{N: parent.N, Type: t, Off: offset} +} func (DummyFrontend) Line(_ src.XPos) string { return "unknown.go:0" } diff --git a/src/cmd/compile/internal/ssa/flagalloc.go b/src/cmd/compile/internal/ssa/flagalloc.go index d50b6159121..61c45a6be7b 100644 --- a/src/cmd/compile/internal/ssa/flagalloc.go +++ b/src/cmd/compile/internal/ssa/flagalloc.go @@ -191,11 +191,6 @@ func flagalloc(f *Func) { b.FlagsLiveAtEnd = end[b.ID] != nil } - const go115flagallocdeadcode = true - if !go115flagallocdeadcode { - return - } - // Remove any now-dead values. // The number of values to remove is likely small, // and removing them requires processing all values in a block, diff --git a/src/cmd/compile/internal/ssa/func.go b/src/cmd/compile/internal/ssa/func.go index 32df0c06f3e..e6f899a2c77 100644 --- a/src/cmd/compile/internal/ssa/func.go +++ b/src/cmd/compile/internal/ssa/func.go @@ -672,7 +672,7 @@ func (f *Func) Idom() []*Block { return f.cachedIdom } -// sdom returns a sparse tree representing the dominator relationships +// Sdom returns a sparse tree representing the dominator relationships // among the blocks of f. func (f *Func) Sdom() SparseTree { if f.cachedSdom == nil { @@ -775,3 +775,25 @@ func (f *Func) logDebugHashMatch(evname, name string) { func DebugNameMatch(evname, name string) bool { return os.Getenv(evname) == name } + +func (f *Func) spSb() (sp, sb *Value) { + initpos := f.Entry.Pos + for _, v := range f.Entry.Values { + if v.Op == OpSB { + sb = v + } + if v.Op == OpSP { + sp = v + } + if sb != nil && sp != nil { + break + } + } + if sb == nil { + sb = f.Entry.NewValue0(initpos.WithNotStmt(), OpSB, f.Config.Types.Uintptr) + } + if sp == nil { + sp = f.Entry.NewValue0(initpos.WithNotStmt(), OpSP, f.Config.Types.Uintptr) + } + return +} diff --git a/src/cmd/compile/internal/ssa/func_test.go b/src/cmd/compile/internal/ssa/func_test.go index 5f6f80f72a9..568c6436f5b 100644 --- a/src/cmd/compile/internal/ssa/func_test.go +++ b/src/cmd/compile/internal/ssa/func_test.go @@ -38,6 +38,7 @@ package ssa import ( "cmd/compile/internal/types" + "cmd/internal/obj" "cmd/internal/src" "fmt" "reflect" @@ -140,6 +141,12 @@ var emptyPass pass = pass{ name: "empty pass", } +// AuxCallLSym returns an AuxCall initialized with an LSym that should pass "check" +// as the Aux of a static call. +func AuxCallLSym(name string) *AuxCall { + return &AuxCall{Fn: &obj.LSym{}} +} + // Fun takes the name of an entry bloc and a series of Bloc calls, and // returns a fun containing the composed Func. entry must be a name // supplied to one of the Bloc functions. Each of the bloc names and diff --git a/src/cmd/compile/internal/ssa/fuse_test.go b/src/cmd/compile/internal/ssa/fuse_test.go index 5fe3da93ca0..15190997f22 100644 --- a/src/cmd/compile/internal/ssa/fuse_test.go +++ b/src/cmd/compile/internal/ssa/fuse_test.go @@ -142,10 +142,10 @@ func TestFuseSideEffects(t *testing.T) { Valu("b", OpArg, c.config.Types.Bool, 0, nil), If("b", "then", "else")), Bloc("then", - Valu("call1", OpStaticCall, types.TypeMem, 0, nil, "mem"), + Valu("call1", OpStaticCall, types.TypeMem, 0, AuxCallLSym("_"), "mem"), Goto("empty")), Bloc("else", - Valu("call2", OpStaticCall, types.TypeMem, 0, nil, "mem"), + Valu("call2", OpStaticCall, types.TypeMem, 0, AuxCallLSym("_"), "mem"), Goto("empty")), Bloc("empty", Goto("loop")), diff --git a/src/cmd/compile/internal/ssa/gen/386.rules b/src/cmd/compile/internal/ssa/gen/386.rules index 4a8244eb27f..fbc12fd6721 100644 --- a/src/cmd/compile/internal/ssa/gen/386.rules +++ b/src/cmd/compile/internal/ssa/gen/386.rules @@ -38,10 +38,8 @@ (Xor(32|16|8) ...) => (XORL ...) (Neg(32|16|8) ...) => (NEGL ...) -(Neg32F x) && !config.use387 => (PXOR x (MOVSSconst [float32(math.Copysign(0, -1))])) -(Neg64F x) && !config.use387 => (PXOR x (MOVSDconst [math.Copysign(0, -1)])) -(Neg32F x) && config.use387 => (FCHS x) -(Neg64F x) && config.use387 => (FCHS x) +(Neg32F x) => (PXOR x (MOVSSconst [float32(math.Copysign(0, -1))])) +(Neg64F x) => (PXOR x (MOVSDconst [math.Copysign(0, -1)])) (Com(32|16|8) ...) => (NOTL ...) @@ -312,7 +310,7 @@ (Const32 ...) => (MOVLconst ...) (Const(32|64)F ...) => (MOVS(S|D)const ...) (ConstNil) => (MOVLconst [0]) -(ConstBool [c]) => (MOVLconst [int32(b2i(c))]) +(ConstBool [c]) => (MOVLconst [b2i32(c)]) // Lowering calls (StaticCall ...) => (CALLstatic ...) @@ -533,6 +531,7 @@ // fold ADDL into LEAL (ADDLconst [c] (LEAL [d] {s} x)) && is32Bit(int64(c)+int64(d)) => (LEAL [c+d] {s} x) (LEAL [c] {s} (ADDLconst [d] x)) && is32Bit(int64(c)+int64(d)) => (LEAL [c+d] {s} x) +(ADDLconst [c] x:(SP)) => (LEAL [c] x) // so it is rematerializeable (LEAL [c] {s} (ADDL x y)) && x.Op != OpSB && y.Op != OpSB => (LEAL1 [c] {s} x y) (ADDL x (LEAL [c] {s} y)) && x.Op != OpSB && y.Op != OpSB => (LEAL1 [c] {s} x y) @@ -642,36 +641,36 @@ // it compiles to a thunk call). (MOV(L|W|B|SS|SD|BLSX|WLSX)load [off1] {sym1} (LEAL [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) => - (MOV(L|W|B|SS|SD|BLSX|WLSX)load [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOV(L|W|B|SS|SD|BLSX|WLSX)load [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOV(L|W|B|SS|SD)store [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) => - (MOV(L|W|B|SS|SD)store [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOV(L|W|B|SS|SD)store [off1+off2] {mergeSym(sym1,sym2)} base val mem) (MOV(L|W|B)storeconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && sc.canAdd32(off) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOV(L|W|B)storeconst [sc.addOffset32(off)] {mergeSymTyped(sym1, sym2)} ptr mem) + (MOV(L|W|B)storeconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) ((ADD|SUB|MUL|AND|OR|XOR)Lload [off1] {sym1} val (LEAL [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) => - ((ADD|SUB|MUL|AND|OR|XOR)Lload [off1+off2] {mergeSymTyped(sym1,sym2)} val base mem) + ((ADD|SUB|MUL|AND|OR|XOR)Lload [off1+off2] {mergeSym(sym1,sym2)} val base mem) ((ADD|SUB|MUL|DIV)SSload [off1] {sym1} val (LEAL [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) => - ((ADD|SUB|MUL|DIV)SSload [off1+off2] {mergeSymTyped(sym1,sym2)} val base mem) + ((ADD|SUB|MUL|DIV)SSload [off1+off2] {mergeSym(sym1,sym2)} val base mem) ((ADD|SUB|MUL|DIV)SDload [off1] {sym1} val (LEAL [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) => - ((ADD|SUB|MUL|DIV)SDload [off1+off2] {mergeSymTyped(sym1,sym2)} val base mem) + ((ADD|SUB|MUL|DIV)SDload [off1+off2] {mergeSym(sym1,sym2)} val base mem) ((ADD|SUB|AND|OR|XOR)Lmodify [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) => - ((ADD|SUB|AND|OR|XOR)Lmodify [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + ((ADD|SUB|AND|OR|XOR)Lmodify [off1+off2] {mergeSym(sym1,sym2)} base val mem) ((ADD|AND|OR|XOR)Lconstmodify [valoff1] {sym1} (LEAL [off2] {sym2} base) mem) && valoff1.canAdd32(off2) && canMergeSym(sym1, sym2) && (base.Op != OpSB || !config.ctxt.Flag_shared) => - ((ADD|AND|OR|XOR)Lconstmodify [valoff1.addOffset32(off2)] {mergeSymTyped(sym1,sym2)} base mem) + ((ADD|AND|OR|XOR)Lconstmodify [valoff1.addOffset32(off2)] {mergeSym(sym1,sym2)} base mem) // Merge load/store to op ((ADD|AND|OR|XOR|SUB|MUL)L x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|AND|OR|XOR|SUB|MUL)Lload x [off] {sym} ptr mem) -((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && !config.use387 && clobber(l) => ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem) -((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && !config.use387 && clobber(l) => ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem) +((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem) +((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem) (MOVLstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Lload x [off] {sym} ptr mem) mem) && y.Uses==1 && clobber(y) => ((ADD|AND|OR|XOR)Lmodify [off] {sym} ptr x mem) (MOVLstore {sym} [off] ptr y:((ADD|SUB|AND|OR|XOR)L l:(MOVLload [off] {sym} ptr mem) x) mem) && y.Uses==1 && l.Uses==1 && clobber(y, l) => ((ADD|SUB|AND|OR|XOR)Lmodify [off] {sym} ptr x mem) @@ -681,37 +680,37 @@ // fold LEALs together (LEAL [off1] {sym1} (LEAL [off2] {sym2} x)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (LEAL [off1+off2] {mergeSymTyped(sym1,sym2)} x) + (LEAL [off1+off2] {mergeSym(sym1,sym2)} x) // LEAL into LEAL1 (LEAL1 [off1] {sym1} (LEAL [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => - (LEAL1 [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (LEAL1 [off1+off2] {mergeSym(sym1,sym2)} x y) // LEAL1 into LEAL (LEAL [off1] {sym1} (LEAL1 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (LEAL1 [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (LEAL1 [off1+off2] {mergeSym(sym1,sym2)} x y) // LEAL into LEAL[248] (LEAL2 [off1] {sym1} (LEAL [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => - (LEAL2 [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (LEAL2 [off1+off2] {mergeSym(sym1,sym2)} x y) (LEAL4 [off1] {sym1} (LEAL [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => - (LEAL4 [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (LEAL4 [off1+off2] {mergeSym(sym1,sym2)} x y) (LEAL8 [off1] {sym1} (LEAL [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => - (LEAL8 [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (LEAL8 [off1+off2] {mergeSym(sym1,sym2)} x y) // LEAL[248] into LEAL (LEAL [off1] {sym1} (LEAL2 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (LEAL2 [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (LEAL2 [off1+off2] {mergeSym(sym1,sym2)} x y) (LEAL [off1] {sym1} (LEAL4 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (LEAL4 [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (LEAL4 [off1+off2] {mergeSym(sym1,sym2)} x y) (LEAL [off1] {sym1} (LEAL8 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (LEAL8 [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (LEAL8 [off1+off2] {mergeSym(sym1,sym2)} x y) // LEAL[1248] into LEAL[1248]. Only some such merges are possible. (LEAL1 [off1] {sym1} x (LEAL1 [off2] {sym2} y y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (LEAL2 [off1+off2] {mergeSymTyped(sym1, sym2)} x y) + (LEAL2 [off1+off2] {mergeSym(sym1, sym2)} x y) (LEAL1 [off1] {sym1} x (LEAL1 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (LEAL2 [off1+off2] {mergeSymTyped(sym1, sym2)} y x) + (LEAL2 [off1+off2] {mergeSym(sym1, sym2)} y x) (LEAL2 [off1] {sym} x (LEAL1 [off2] {nil} y y)) && is32Bit(int64(off1)+2*int64(off2)) => (LEAL4 [off1+2*off2] {sym} x y) (LEAL4 [off1] {sym} x (LEAL1 [off2] {nil} y y)) && is32Bit(int64(off1)+4*int64(off2)) => @@ -995,49 +994,49 @@ && x.Uses == 1 && a.Off() + 1 == c.Off() && clobber(x) - => (MOVWstoreconst [makeValAndOff32(int32(a.Val()&0xff | c.Val()<<8), int32(a.Off()))] {s} p mem) + => (MOVWstoreconst [makeValAndOff32(int32(a.Val()&0xff | c.Val()<<8), a.Off32())] {s} p mem) (MOVBstoreconst [a] {s} p x:(MOVBstoreconst [c] {s} p mem)) && x.Uses == 1 && a.Off() + 1 == c.Off() && clobber(x) - => (MOVWstoreconst [makeValAndOff32(int32(a.Val()&0xff | c.Val()<<8), int32(a.Off()))] {s} p mem) + => (MOVWstoreconst [makeValAndOff32(int32(a.Val()&0xff | c.Val()<<8), a.Off32())] {s} p mem) (MOVBstoreconst [c] {s} p1 x:(MOVBstoreconst [a] {s} p0 mem)) && x.Uses == 1 && a.Off() == c.Off() && sequentialAddresses(p0, p1, 1) && clobber(x) - => (MOVWstoreconst [makeValAndOff32(int32(a.Val()&0xff | c.Val()<<8), int32(a.Off()))] {s} p0 mem) + => (MOVWstoreconst [makeValAndOff32(int32(a.Val()&0xff | c.Val()<<8), a.Off32())] {s} p0 mem) (MOVBstoreconst [a] {s} p0 x:(MOVBstoreconst [c] {s} p1 mem)) && x.Uses == 1 && a.Off() == c.Off() && sequentialAddresses(p0, p1, 1) && clobber(x) - => (MOVWstoreconst [makeValAndOff32(int32(a.Val()&0xff | c.Val()<<8), int32(a.Off()))] {s} p0 mem) + => (MOVWstoreconst [makeValAndOff32(int32(a.Val()&0xff | c.Val()<<8), a.Off32())] {s} p0 mem) (MOVWstoreconst [c] {s} p x:(MOVWstoreconst [a] {s} p mem)) && x.Uses == 1 && a.Off() + 2 == c.Off() && clobber(x) - => (MOVLstoreconst [makeValAndOff32(int32(a.Val()&0xffff | c.Val()<<16), int32(a.Off()))] {s} p mem) + => (MOVLstoreconst [makeValAndOff32(int32(a.Val()&0xffff | c.Val()<<16), a.Off32())] {s} p mem) (MOVWstoreconst [a] {s} p x:(MOVWstoreconst [c] {s} p mem)) && x.Uses == 1 && ValAndOff(a).Off() + 2 == ValAndOff(c).Off() && clobber(x) - => (MOVLstoreconst [makeValAndOff32(int32(a.Val()&0xffff | c.Val()<<16), int32(a.Off()))] {s} p mem) + => (MOVLstoreconst [makeValAndOff32(int32(a.Val()&0xffff | c.Val()<<16), a.Off32())] {s} p mem) (MOVWstoreconst [c] {s} p1 x:(MOVWstoreconst [a] {s} p0 mem)) && x.Uses == 1 && a.Off() == c.Off() && sequentialAddresses(p0, p1, 2) && clobber(x) - => (MOVLstoreconst [makeValAndOff32(int32(a.Val()&0xffff | c.Val()<<16), int32(a.Off()))] {s} p0 mem) + => (MOVLstoreconst [makeValAndOff32(int32(a.Val()&0xffff | c.Val()<<16), a.Off32())] {s} p0 mem) (MOVWstoreconst [a] {s} p0 x:(MOVWstoreconst [c] {s} p1 mem)) && x.Uses == 1 && a.Off() == c.Off() && sequentialAddresses(p0, p1, 2) && clobber(x) - => (MOVLstoreconst [makeValAndOff32(int32(a.Val()&0xffff | c.Val()<<16), int32(a.Off()))] {s} p0 mem) + => (MOVLstoreconst [makeValAndOff32(int32(a.Val()&0xffff | c.Val()<<16), a.Off32())] {s} p0 mem) // Combine stores into larger (unaligned) stores. (MOVBstore [i] {s} p (SHR(W|L)const [8] w) x:(MOVBstore [i-1] {s} p w mem)) diff --git a/src/cmd/compile/internal/ssa/gen/386Ops.go b/src/cmd/compile/internal/ssa/gen/386Ops.go index 1061e5579db..737b99c3716 100644 --- a/src/cmd/compile/internal/ssa/gen/386Ops.go +++ b/src/cmd/compile/internal/ssa/gen/386Ops.go @@ -51,17 +51,6 @@ var regNames386 = []string{ "SB", } -// Notes on 387 support. -// - The 387 has a weird stack-register setup for floating-point registers. -// We use these registers when SSE registers are not available (when GO386=387). -// - We use the same register names (X0-X7) but they refer to the 387 -// floating-point registers. That way, most of the SSA backend is unchanged. -// - The instruction generation pass maintains an SSE->387 register mapping. -// This mapping is updated whenever the FP stack is pushed or popped so that -// we can always find a given SSE register even when the TOS pointer has changed. -// - To facilitate the mapping from SSE to 387, we enforce that -// every basic block starts and ends with an empty floating-point stack. - func init() { // Make map from reg names to reg integers. if len(regNames386) > 64 { @@ -463,9 +452,9 @@ func init() { faultOnNilArg0: true, }, - {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", clobberFlags: true, call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("DX"), 0}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("DX"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem // arg0 = destination pointer // arg1 = source pointer @@ -552,9 +541,6 @@ func init() { {name: "FlagGT_UGT"}, // signed > and unsigned < {name: "FlagGT_ULT"}, // signed > and unsigned > - // Special op for -x on 387 - {name: "FCHS", argLength: 1, reg: fp11}, - // Special ops for PIC floating-point constants. // MOVSXconst1 loads the address of the constant-pool entry into a register. // MOVSXconst2 loads the constant from that address. diff --git a/src/cmd/compile/internal/ssa/gen/AMD64.rules b/src/cmd/compile/internal/ssa/gen/AMD64.rules index 8898fe55eb4..a866a967b92 100644 --- a/src/cmd/compile/internal/ssa/gen/AMD64.rules +++ b/src/cmd/compile/internal/ssa/gen/AMD64.rules @@ -401,38 +401,38 @@ (Const32F ...) => (MOVSSconst ...) (Const64F ...) => (MOVSDconst ...) (ConstNil ) => (MOVQconst [0]) -(ConstBool [c]) => (MOVLconst [int32(b2i(c))]) +(ConstBool [c]) => (MOVLconst [b2i32(c)]) // Lowering calls -(StaticCall ...) -> (CALLstatic ...) -(ClosureCall ...) -> (CALLclosure ...) -(InterCall ...) -> (CALLinter ...) +(StaticCall ...) => (CALLstatic ...) +(ClosureCall ...) => (CALLclosure ...) +(InterCall ...) => (CALLinter ...) // Lowering conditional moves // If the condition is a SETxx, we can just run a CMOV from the comparison that was // setting the flags. // Legend: HI=unsigned ABOVE, CS=unsigned BELOW, CC=unsigned ABOVE EQUAL, LS=unsigned BELOW EQUAL (CondSelect x y (SET(EQ|NE|L|G|LE|GE|A|B|AE|BE|EQF|NEF|GF|GEF) cond)) && (is64BitInt(t) || isPtr(t)) - -> (CMOVQ(EQ|NE|LT|GT|LE|GE|HI|CS|CC|LS|EQF|NEF|GTF|GEF) y x cond) + => (CMOVQ(EQ|NE|LT|GT|LE|GE|HI|CS|CC|LS|EQF|NEF|GTF|GEF) y x cond) (CondSelect x y (SET(EQ|NE|L|G|LE|GE|A|B|AE|BE|EQF|NEF|GF|GEF) cond)) && is32BitInt(t) - -> (CMOVL(EQ|NE|LT|GT|LE|GE|HI|CS|CC|LS|EQF|NEF|GTF|GEF) y x cond) + => (CMOVL(EQ|NE|LT|GT|LE|GE|HI|CS|CC|LS|EQF|NEF|GTF|GEF) y x cond) (CondSelect x y (SET(EQ|NE|L|G|LE|GE|A|B|AE|BE|EQF|NEF|GF|GEF) cond)) && is16BitInt(t) - -> (CMOVW(EQ|NE|LT|GT|LE|GE|HI|CS|CC|LS|EQF|NEF|GTF|GEF) y x cond) + => (CMOVW(EQ|NE|LT|GT|LE|GE|HI|CS|CC|LS|EQF|NEF|GTF|GEF) y x cond) // If the condition does not set the flags, we need to generate a comparison. (CondSelect x y check) && !check.Type.IsFlags() && check.Type.Size() == 1 - -> (CondSelect x y (MOVBQZX check)) + => (CondSelect x y (MOVBQZX check)) (CondSelect x y check) && !check.Type.IsFlags() && check.Type.Size() == 2 - -> (CondSelect x y (MOVWQZX check)) + => (CondSelect x y (MOVWQZX check)) (CondSelect x y check) && !check.Type.IsFlags() && check.Type.Size() == 4 - -> (CondSelect x y (MOVLQZX check)) + => (CondSelect x y (MOVLQZX check)) (CondSelect x y check) && !check.Type.IsFlags() && check.Type.Size() == 8 && (is64BitInt(t) || isPtr(t)) - -> (CMOVQNE y x (CMPQconst [0] check)) + => (CMOVQNE y x (CMPQconst [0] check)) (CondSelect x y check) && !check.Type.IsFlags() && check.Type.Size() == 8 && is32BitInt(t) - -> (CMOVLNE y x (CMPQconst [0] check)) + => (CMOVLNE y x (CMPQconst [0] check)) (CondSelect x y check) && !check.Type.IsFlags() && check.Type.Size() == 8 && is16BitInt(t) - -> (CMOVWNE y x (CMPQconst [0] check)) + => (CMOVWNE y x (CMPQconst [0] check)) // Absorb InvertFlags (CMOVQ(EQ|NE|LT|GT|LE|GE|HI|CS|CC|LS) x y (InvertFlags cond)) @@ -465,7 +465,7 @@ (GetCallerSP ...) => (LoweredGetCallerSP ...) (HasCPUFeature {s}) => (SETNE (CMPQconst [0] (LoweredHasCPUFeature {s}))) -(Addr ...) -> (LEAQ ...) +(Addr {sym} base) => (LEAQ {sym} base) (LocalAddr {sym} base _) => (LEAQ {sym} base) (MOVBstore [off] {sym} ptr y:(SETL x) mem) && y.Uses == 1 => (SETLstore [off] {sym} ptr x mem) @@ -501,10 +501,10 @@ (If cond yes no) => (NE (TESTB cond cond) yes no) // Atomic loads. Other than preserving their ordering with respect to other loads, nothing special here. -(AtomicLoad8 ...) -> (MOVBatomicload ...) -(AtomicLoad32 ...) -> (MOVLatomicload ...) -(AtomicLoad64 ...) -> (MOVQatomicload ...) -(AtomicLoadPtr ...) -> (MOVQatomicload ...) +(AtomicLoad8 ptr mem) => (MOVBatomicload ptr mem) +(AtomicLoad32 ptr mem) => (MOVLatomicload ptr mem) +(AtomicLoad64 ptr mem) => (MOVQatomicload ptr mem) +(AtomicLoadPtr ptr mem) => (MOVQatomicload ptr mem) // Atomic stores. We use XCHG to prevent the hardware reordering a subsequent load. // TODO: most runtime uses of atomic stores don't need that property. Use normal stores for those? @@ -526,15 +526,17 @@ (Select1 (AddTupleFirst64 _ tuple)) => (Select1 tuple) // Atomic compare and swap. -(AtomicCompareAndSwap32 ...) -> (CMPXCHGLlock ...) -(AtomicCompareAndSwap64 ...) -> (CMPXCHGQlock ...) +(AtomicCompareAndSwap32 ptr old new_ mem) => (CMPXCHGLlock ptr old new_ mem) +(AtomicCompareAndSwap64 ptr old new_ mem) => (CMPXCHGQlock ptr old new_ mem) // Atomic memory updates. -(AtomicAnd8 ...) -> (ANDBlock ...) -(AtomicOr8 ...) -> (ORBlock ...) +(AtomicAnd8 ptr val mem) => (ANDBlock ptr val mem) +(AtomicAnd32 ptr val mem) => (ANDLlock ptr val mem) +(AtomicOr8 ptr val mem) => (ORBlock ptr val mem) +(AtomicOr32 ptr val mem) => (ORLlock ptr val mem) // Write barrier. -(WB ...) -> (LoweredWB ...) +(WB ...) => (LoweredWB ...) (PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem) (PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem) @@ -581,7 +583,7 @@ ((NE|EQ) (TESTQconst [c] x)) && isUint64PowerOfTwo(int64(c)) => ((ULT|UGE) (BTQconst [int8(log32(c))] x)) ((NE|EQ) (TESTQ (MOVQconst [c]) x)) && isUint64PowerOfTwo(c) - => ((ULT|UGE) (BTQconst [int8(log2(c))] x)) + => ((ULT|UGE) (BTQconst [int8(log64(c))] x)) (SET(NE|EQ) (TESTL (SHLL (MOVLconst [1]) x) y)) => (SET(B|AE) (BTL x y)) (SET(NE|EQ) (TESTQ (SHLQ (MOVQconst [1]) x) y)) => (SET(B|AE) (BTQ x y)) (SET(NE|EQ) (TESTLconst [c] x)) && isUint32PowerOfTwo(int64(c)) @@ -589,7 +591,7 @@ (SET(NE|EQ) (TESTQconst [c] x)) && isUint64PowerOfTwo(int64(c)) => (SET(B|AE) (BTQconst [int8(log32(c))] x)) (SET(NE|EQ) (TESTQ (MOVQconst [c]) x)) && isUint64PowerOfTwo(c) - => (SET(B|AE) (BTQconst [int8(log2(c))] x)) + => (SET(B|AE) (BTQconst [int8(log64(c))] x)) // SET..store variant (SET(NE|EQ)store [off] {sym} ptr (TESTL (SHLL (MOVLconst [1]) x) y) mem) => (SET(B|AE)store [off] {sym} ptr (BTL x y) mem) @@ -600,7 +602,7 @@ (SET(NE|EQ)store [off] {sym} ptr (TESTQconst [c] x) mem) && isUint64PowerOfTwo(int64(c)) => (SET(B|AE)store [off] {sym} ptr (BTQconst [int8(log32(c))] x) mem) (SET(NE|EQ)store [off] {sym} ptr (TESTQ (MOVQconst [c]) x) mem) && isUint64PowerOfTwo(c) - => (SET(B|AE)store [off] {sym} ptr (BTQconst [int8(log2(c))] x) mem) + => (SET(B|AE)store [off] {sym} ptr (BTQconst [int8(log64(c))] x) mem) // Handle bit-testing in the form (a>>b)&1 != 0 by building the above rules // and further combining shifts. @@ -629,7 +631,7 @@ ((ORL|XORL)const [c] x) && isUint32PowerOfTwo(int64(c)) && uint64(c) >= 128 => (BT(S|C)Lconst [int8(log32(c))] x) ((ORQ|XORQ) (MOVQconst [c]) x) && isUint64PowerOfTwo(c) && uint64(c) >= 128 - => (BT(S|C)Qconst [int8(log2(c))] x) + => (BT(S|C)Qconst [int8(log64(c))] x) ((ORL|XORL) (MOVLconst [c]) x) && isUint32PowerOfTwo(int64(c)) && uint64(c) >= 128 => (BT(S|C)Lconst [int8(log32(c))] x) @@ -640,7 +642,7 @@ (ANDLconst [c] x) && isUint32PowerOfTwo(int64(^c)) && uint64(^c) >= 128 => (BTRLconst [int8(log32(^c))] x) (ANDQ (MOVQconst [c]) x) && isUint64PowerOfTwo(^c) && uint64(^c) >= 128 - => (BTRQconst [int8(log2(^c))] x) + => (BTRQconst [int8(log64(^c))] x) (ANDL (MOVLconst [c]) x) && isUint32PowerOfTwo(int64(^c)) && uint64(^c) >= 128 => (BTRLconst [int8(log32(^c))] x) @@ -853,7 +855,7 @@ (ANDL (SHRW x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16]))) (SBBLcarrymask (CMP(Q|L)const (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16])) [16])))) && v.Type.Size() == 2 - -> (ROLW x y) + => (ROLW x y) (ORL (SHRW x (AND(Q|L)const y [15])) (SHLL x (NEG(Q|L) (ADD(Q|L)const (AND(Q|L)const y [15]) [-16])))) && v.Type.Size() == 2 @@ -957,7 +959,7 @@ (MUL(Q|L)const [73] x) => (LEA(Q|L)8 x (LEA(Q|L)8 x x)) (MUL(Q|L)const [81] x) => (LEA(Q|L)8 (LEA(Q|L)8 x x) (LEA(Q|L)8 x x)) -(MUL(Q|L)const [c] x) && isPowerOfTwo(int64(c)+1) && c >= 15 => (SUB(Q|L) (SHL(Q|L)const [int8(log2(int64(c)+1))] x) x) +(MUL(Q|L)const [c] x) && isPowerOfTwo64(int64(c)+1) && c >= 15 => (SUB(Q|L) (SHL(Q|L)const [int8(log64(int64(c)+1))] x) x) (MUL(Q|L)const [c] x) && isPowerOfTwo32(c-1) && c >= 17 => (LEA(Q|L)1 (SHL(Q|L)const [int8(log32(c-1))] x) x) (MUL(Q|L)const [c] x) && isPowerOfTwo32(c-2) && c >= 34 => (LEA(Q|L)2 (SHL(Q|L)const [int8(log32(c-2))] x) x) (MUL(Q|L)const [c] x) && isPowerOfTwo32(c-4) && c >= 68 => (LEA(Q|L)4 (SHL(Q|L)const [int8(log32(c-4))] x) x) @@ -1134,268 +1136,270 @@ // We need to fold LEAQ into the MOVx ops so that the live variable analysis knows // what variables are being read/written by the ops. (MOV(Q|L|W|B|SS|SD|O|BQSX|WQSX|LQSX)load [off1] {sym1} (LEAQ [off2] {sym2} base) mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (MOV(Q|L|W|B|SS|SD|O|BQSX|WQSX|LQSX)load [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOV(Q|L|W|B|SS|SD|O)store [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (MOV(Q|L|W|B|SS|SD|O)store [off1+off2] {mergeSym(sym1,sym2)} base val mem) -(MOV(Q|L|W|B)storeconst [sc] {sym1} (LEAQ [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) -> - (MOV(Q|L|W|B)storeconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem) +(MOV(Q|L|W|B)storeconst [sc] {sym1} (LEAQ [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd32(off) => + (MOV(Q|L|W|B)storeconst [ValAndOff(sc).addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) (SET(L|G|B|A|LE|GE|BE|AE|EQ|NE)store [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (SET(L|G|B|A|LE|GE|BE|AE|EQ|NE)store [off1+off2] {mergeSym(sym1,sym2)} base val mem) ((ADD|SUB|AND|OR|XOR)Qload [off1] {sym1} val (LEAQ [off2] {sym2} base) mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => ((ADD|SUB|AND|OR|XOR)Qload [off1+off2] {mergeSym(sym1,sym2)} val base mem) ((ADD|SUB|AND|OR|XOR)Lload [off1] {sym1} val (LEAQ [off2] {sym2} base) mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => ((ADD|SUB|AND|OR|XOR)Lload [off1+off2] {mergeSym(sym1,sym2)} val base mem) (CMP(Q|L|W|B)load [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (CMP(Q|L|W|B)load [off1+off2] {mergeSym(sym1,sym2)} base val mem) (CMP(Q|L|W|B)constload [valoff1] {sym1} (LEAQ [off2] {sym2} base) mem) - && ValAndOff(valoff1).canAdd(off2) && canMergeSym(sym1, sym2) -> - (CMP(Q|L|W|B)constload [ValAndOff(valoff1).add(off2)] {mergeSym(sym1,sym2)} base mem) + && ValAndOff(valoff1).canAdd32(off2) && canMergeSym(sym1, sym2) => + (CMP(Q|L|W|B)constload [ValAndOff(valoff1).addOffset32(off2)] {mergeSym(sym1,sym2)} base mem) ((ADD|SUB|MUL|DIV)SSload [off1] {sym1} val (LEAQ [off2] {sym2} base) mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => ((ADD|SUB|MUL|DIV)SSload [off1+off2] {mergeSym(sym1,sym2)} val base mem) ((ADD|SUB|MUL|DIV)SDload [off1] {sym1} val (LEAQ [off2] {sym2} base) mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => ((ADD|SUB|MUL|DIV)SDload [off1+off2] {mergeSym(sym1,sym2)} val base mem) ((ADD|AND|OR|XOR|BTC|BTR|BTS)Qconstmodify [valoff1] {sym1} (LEAQ [off2] {sym2} base) mem) - && ValAndOff(valoff1).canAdd(off2) && canMergeSym(sym1, sym2) -> - ((ADD|AND|OR|XOR|BTC|BTR|BTS)Qconstmodify [ValAndOff(valoff1).add(off2)] {mergeSym(sym1,sym2)} base mem) + && ValAndOff(valoff1).canAdd32(off2) && canMergeSym(sym1, sym2) => + ((ADD|AND|OR|XOR|BTC|BTR|BTS)Qconstmodify [ValAndOff(valoff1).addOffset32(off2)] {mergeSym(sym1,sym2)} base mem) ((ADD|AND|OR|XOR|BTC|BTR|BTS)Lconstmodify [valoff1] {sym1} (LEAQ [off2] {sym2} base) mem) - && ValAndOff(valoff1).canAdd(off2) && canMergeSym(sym1, sym2) -> - ((ADD|AND|OR|XOR|BTC|BTR|BTS)Lconstmodify [ValAndOff(valoff1).add(off2)] {mergeSym(sym1,sym2)} base mem) + && ValAndOff(valoff1).canAdd32(off2) && canMergeSym(sym1, sym2) => + ((ADD|AND|OR|XOR|BTC|BTR|BTS)Lconstmodify [ValAndOff(valoff1).addOffset32(off2)] {mergeSym(sym1,sym2)} base mem) ((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Qmodify [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => ((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Qmodify [off1+off2] {mergeSym(sym1,sym2)} base val mem) ((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Lmodify [off1] {sym1} (LEAQ [off2] {sym2} base) val mem) - && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> + && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => ((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Lmodify [off1+off2] {mergeSym(sym1,sym2)} base val mem) // fold LEAQs together -(LEAQ [off1] {sym1} (LEAQ [off2] {sym2} x)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> +(LEAQ [off1] {sym1} (LEAQ [off2] {sym2} x)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (LEAQ [off1+off2] {mergeSym(sym1,sym2)} x) // LEAQ into LEAQ1 -(LEAQ1 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB -> +(LEAQ1 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => (LEAQ1 [off1+off2] {mergeSym(sym1,sym2)} x y) // LEAQ1 into LEAQ -(LEAQ [off1] {sym1} (LEAQ1 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> +(LEAQ [off1] {sym1} (LEAQ1 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (LEAQ1 [off1+off2] {mergeSym(sym1,sym2)} x y) // LEAQ into LEAQ[248] -(LEAQ2 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB -> +(LEAQ2 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => (LEAQ2 [off1+off2] {mergeSym(sym1,sym2)} x y) -(LEAQ4 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB -> +(LEAQ4 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => (LEAQ4 [off1+off2] {mergeSym(sym1,sym2)} x y) -(LEAQ8 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && x.Op != OpSB -> +(LEAQ8 [off1] {sym1} (LEAQ [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => (LEAQ8 [off1+off2] {mergeSym(sym1,sym2)} x y) // LEAQ[248] into LEAQ -(LEAQ [off1] {sym1} (LEAQ2 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> +(LEAQ [off1] {sym1} (LEAQ2 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (LEAQ2 [off1+off2] {mergeSym(sym1,sym2)} x y) -(LEAQ [off1] {sym1} (LEAQ4 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> +(LEAQ [off1] {sym1} (LEAQ4 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (LEAQ4 [off1+off2] {mergeSym(sym1,sym2)} x y) -(LEAQ [off1] {sym1} (LEAQ8 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> +(LEAQ [off1] {sym1} (LEAQ8 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (LEAQ8 [off1+off2] {mergeSym(sym1,sym2)} x y) // LEAQ[1248] into LEAQ[1248]. Only some such merges are possible. -(LEAQ1 [off1] {sym1} x (LEAQ1 [off2] {sym2} y y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> +(LEAQ1 [off1] {sym1} x (LEAQ1 [off2] {sym2} y y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (LEAQ2 [off1+off2] {mergeSym(sym1, sym2)} x y) -(LEAQ1 [off1] {sym1} x (LEAQ1 [off2] {sym2} x y)) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> +(LEAQ1 [off1] {sym1} x (LEAQ1 [off2] {sym2} x y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => (LEAQ2 [off1+off2] {mergeSym(sym1, sym2)} y x) -(LEAQ2 [off1] {sym1} x (LEAQ1 [off2] {sym2} y y)) && is32Bit(off1+2*off2) && sym2 == nil -> +(LEAQ2 [off1] {sym1} x (LEAQ1 [off2] {sym2} y y)) && is32Bit(int64(off1)+2*int64(off2)) && sym2 == nil => (LEAQ4 [off1+2*off2] {sym1} x y) -(LEAQ4 [off1] {sym1} x (LEAQ1 [off2] {sym2} y y)) && is32Bit(off1+4*off2) && sym2 == nil -> +(LEAQ4 [off1] {sym1} x (LEAQ1 [off2] {sym2} y y)) && is32Bit(int64(off1)+4*int64(off2)) && sym2 == nil => (LEAQ8 [off1+4*off2] {sym1} x y) // TODO: more? // Lower LEAQ2/4/8 when the offset is a constant -(LEAQ2 [off] {sym} x (MOV(Q|L)const [scale])) && is32Bit(off+scale*2) -> - (LEAQ [off+scale*2] {sym} x) -(LEAQ4 [off] {sym} x (MOV(Q|L)const [scale])) && is32Bit(off+scale*4) -> - (LEAQ [off+scale*4] {sym} x) -(LEAQ8 [off] {sym} x (MOV(Q|L)const [scale])) && is32Bit(off+scale*8) -> - (LEAQ [off+scale*8] {sym} x) +(LEAQ2 [off] {sym} x (MOV(Q|L)const [scale])) && is32Bit(int64(off)+int64(scale)*2) => + (LEAQ [off+int32(scale)*2] {sym} x) +(LEAQ4 [off] {sym} x (MOV(Q|L)const [scale])) && is32Bit(int64(off)+int64(scale)*4) => + (LEAQ [off+int32(scale)*4] {sym} x) +(LEAQ8 [off] {sym} x (MOV(Q|L)const [scale])) && is32Bit(int64(off)+int64(scale)*8) => + (LEAQ [off+int32(scale)*8] {sym} x) // Absorb InvertFlags into branches. -(LT (InvertFlags cmp) yes no) -> (GT cmp yes no) -(GT (InvertFlags cmp) yes no) -> (LT cmp yes no) -(LE (InvertFlags cmp) yes no) -> (GE cmp yes no) -(GE (InvertFlags cmp) yes no) -> (LE cmp yes no) -(ULT (InvertFlags cmp) yes no) -> (UGT cmp yes no) -(UGT (InvertFlags cmp) yes no) -> (ULT cmp yes no) -(ULE (InvertFlags cmp) yes no) -> (UGE cmp yes no) -(UGE (InvertFlags cmp) yes no) -> (ULE cmp yes no) -(EQ (InvertFlags cmp) yes no) -> (EQ cmp yes no) -(NE (InvertFlags cmp) yes no) -> (NE cmp yes no) +(LT (InvertFlags cmp) yes no) => (GT cmp yes no) +(GT (InvertFlags cmp) yes no) => (LT cmp yes no) +(LE (InvertFlags cmp) yes no) => (GE cmp yes no) +(GE (InvertFlags cmp) yes no) => (LE cmp yes no) +(ULT (InvertFlags cmp) yes no) => (UGT cmp yes no) +(UGT (InvertFlags cmp) yes no) => (ULT cmp yes no) +(ULE (InvertFlags cmp) yes no) => (UGE cmp yes no) +(UGE (InvertFlags cmp) yes no) => (ULE cmp yes no) +(EQ (InvertFlags cmp) yes no) => (EQ cmp yes no) +(NE (InvertFlags cmp) yes no) => (NE cmp yes no) // Constant comparisons. -(CMPQconst (MOVQconst [x]) [y]) && x==y -> (FlagEQ) -(CMPQconst (MOVQconst [x]) [y]) && x (FlagLT_ULT) -(CMPQconst (MOVQconst [x]) [y]) && xuint64(y) -> (FlagLT_UGT) -(CMPQconst (MOVQconst [x]) [y]) && x>y && uint64(x) (FlagGT_ULT) -(CMPQconst (MOVQconst [x]) [y]) && x>y && uint64(x)>uint64(y) -> (FlagGT_UGT) -(CMPLconst (MOVLconst [x]) [y]) && int32(x)==int32(y) -> (FlagEQ) -(CMPLconst (MOVLconst [x]) [y]) && int32(x) (FlagLT_ULT) -(CMPLconst (MOVLconst [x]) [y]) && int32(x)uint32(y) -> (FlagLT_UGT) -(CMPLconst (MOVLconst [x]) [y]) && int32(x)>int32(y) && uint32(x) (FlagGT_ULT) -(CMPLconst (MOVLconst [x]) [y]) && int32(x)>int32(y) && uint32(x)>uint32(y) -> (FlagGT_UGT) -(CMPWconst (MOVLconst [x]) [y]) && int16(x)==int16(y) -> (FlagEQ) -(CMPWconst (MOVLconst [x]) [y]) && int16(x) (FlagLT_ULT) -(CMPWconst (MOVLconst [x]) [y]) && int16(x)uint16(y) -> (FlagLT_UGT) -(CMPWconst (MOVLconst [x]) [y]) && int16(x)>int16(y) && uint16(x) (FlagGT_ULT) -(CMPWconst (MOVLconst [x]) [y]) && int16(x)>int16(y) && uint16(x)>uint16(y) -> (FlagGT_UGT) -(CMPBconst (MOVLconst [x]) [y]) && int8(x)==int8(y) -> (FlagEQ) -(CMPBconst (MOVLconst [x]) [y]) && int8(x) (FlagLT_ULT) -(CMPBconst (MOVLconst [x]) [y]) && int8(x)uint8(y) -> (FlagLT_UGT) -(CMPBconst (MOVLconst [x]) [y]) && int8(x)>int8(y) && uint8(x) (FlagGT_ULT) -(CMPBconst (MOVLconst [x]) [y]) && int8(x)>int8(y) && uint8(x)>uint8(y) -> (FlagGT_UGT) +(CMPQconst (MOVQconst [x]) [y]) && x==int64(y) => (FlagEQ) +(CMPQconst (MOVQconst [x]) [y]) && x (FlagLT_ULT) +(CMPQconst (MOVQconst [x]) [y]) && xuint64(int64(y)) => (FlagLT_UGT) +(CMPQconst (MOVQconst [x]) [y]) && x>int64(y) && uint64(x) (FlagGT_ULT) +(CMPQconst (MOVQconst [x]) [y]) && x>int64(y) && uint64(x)>uint64(int64(y)) => (FlagGT_UGT) +(CMPLconst (MOVLconst [x]) [y]) && x==y => (FlagEQ) +(CMPLconst (MOVLconst [x]) [y]) && x (FlagLT_ULT) +(CMPLconst (MOVLconst [x]) [y]) && xuint32(y) => (FlagLT_UGT) +(CMPLconst (MOVLconst [x]) [y]) && x>y && uint32(x) (FlagGT_ULT) +(CMPLconst (MOVLconst [x]) [y]) && x>y && uint32(x)>uint32(y) => (FlagGT_UGT) +(CMPWconst (MOVLconst [x]) [y]) && int16(x)==y => (FlagEQ) +(CMPWconst (MOVLconst [x]) [y]) && int16(x) (FlagLT_ULT) +(CMPWconst (MOVLconst [x]) [y]) && int16(x)uint16(y) => (FlagLT_UGT) +(CMPWconst (MOVLconst [x]) [y]) && int16(x)>y && uint16(x) (FlagGT_ULT) +(CMPWconst (MOVLconst [x]) [y]) && int16(x)>y && uint16(x)>uint16(y) => (FlagGT_UGT) +(CMPBconst (MOVLconst [x]) [y]) && int8(x)==y => (FlagEQ) +(CMPBconst (MOVLconst [x]) [y]) && int8(x) (FlagLT_ULT) +(CMPBconst (MOVLconst [x]) [y]) && int8(x)uint8(y) => (FlagLT_UGT) +(CMPBconst (MOVLconst [x]) [y]) && int8(x)>y && uint8(x) (FlagGT_ULT) +(CMPBconst (MOVLconst [x]) [y]) && int8(x)>y && uint8(x)>uint8(y) => (FlagGT_UGT) // CMPQconst requires a 32 bit const, but we can still constant-fold 64 bit consts. // In theory this applies to any of the simplifications above, // but CMPQ is the only one I've actually seen occur. -(CMPQ (MOVQconst [x]) (MOVQconst [y])) && x==y -> (FlagEQ) -(CMPQ (MOVQconst [x]) (MOVQconst [y])) && x (FlagLT_ULT) -(CMPQ (MOVQconst [x]) (MOVQconst [y])) && xuint64(y) -> (FlagLT_UGT) -(CMPQ (MOVQconst [x]) (MOVQconst [y])) && x>y && uint64(x) (FlagGT_ULT) -(CMPQ (MOVQconst [x]) (MOVQconst [y])) && x>y && uint64(x)>uint64(y) -> (FlagGT_UGT) +(CMPQ (MOVQconst [x]) (MOVQconst [y])) && x==y => (FlagEQ) +(CMPQ (MOVQconst [x]) (MOVQconst [y])) && x (FlagLT_ULT) +(CMPQ (MOVQconst [x]) (MOVQconst [y])) && xuint64(y) => (FlagLT_UGT) +(CMPQ (MOVQconst [x]) (MOVQconst [y])) && x>y && uint64(x) (FlagGT_ULT) +(CMPQ (MOVQconst [x]) (MOVQconst [y])) && x>y && uint64(x)>uint64(y) => (FlagGT_UGT) // Other known comparisons. -(CMPQconst (MOVBQZX _) [c]) && 0xFF < c -> (FlagLT_ULT) -(CMPQconst (MOVWQZX _) [c]) && 0xFFFF < c -> (FlagLT_ULT) -(CMPQconst (MOVLQZX _) [c]) && 0xFFFFFFFF < c -> (FlagLT_ULT) -(CMPLconst (SHRLconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 32 && (1< (FlagLT_ULT) -(CMPQconst (SHRQconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 64 && (1< (FlagLT_ULT) -(CMPQconst (ANDQconst _ [m]) [n]) && 0 <= m && m < n -> (FlagLT_ULT) -(CMPQconst (ANDLconst _ [m]) [n]) && 0 <= m && m < n -> (FlagLT_ULT) -(CMPLconst (ANDLconst _ [m]) [n]) && 0 <= int32(m) && int32(m) < int32(n) -> (FlagLT_ULT) -(CMPWconst (ANDLconst _ [m]) [n]) && 0 <= int16(m) && int16(m) < int16(n) -> (FlagLT_ULT) -(CMPBconst (ANDLconst _ [m]) [n]) && 0 <= int8(m) && int8(m) < int8(n) -> (FlagLT_ULT) +(CMPQconst (MOVBQZX _) [c]) && 0xFF < c => (FlagLT_ULT) +(CMPQconst (MOVWQZX _) [c]) && 0xFFFF < c => (FlagLT_ULT) +(CMPLconst (SHRLconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 32 && (1< (FlagLT_ULT) +(CMPQconst (SHRQconst _ [c]) [n]) && 0 <= n && 0 < c && c <= 64 && (1< (FlagLT_ULT) +(CMPQconst (ANDQconst _ [m]) [n]) && 0 <= m && m < n => (FlagLT_ULT) +(CMPQconst (ANDLconst _ [m]) [n]) && 0 <= m && m < n => (FlagLT_ULT) +(CMPLconst (ANDLconst _ [m]) [n]) && 0 <= m && m < n => (FlagLT_ULT) +(CMPWconst (ANDLconst _ [m]) [n]) && 0 <= int16(m) && int16(m) < n => (FlagLT_ULT) +(CMPBconst (ANDLconst _ [m]) [n]) && 0 <= int8(m) && int8(m) < n => (FlagLT_ULT) // TESTQ c c sets flags like CMPQ c 0. -(TEST(Q|L)const [c] (MOV(Q|L)const [c])) && c == 0 -> (FlagEQ) -(TEST(Q|L)const [c] (MOV(Q|L)const [c])) && c < 0 -> (FlagLT_UGT) -(TEST(Q|L)const [c] (MOV(Q|L)const [c])) && c > 0 -> (FlagGT_UGT) +(TESTQconst [c] (MOVQconst [d])) && int64(c) == d && c == 0 => (FlagEQ) +(TESTLconst [c] (MOVLconst [c])) && c == 0 => (FlagEQ) +(TESTQconst [c] (MOVQconst [d])) && int64(c) == d && c < 0 => (FlagLT_UGT) +(TESTLconst [c] (MOVLconst [c])) && c < 0 => (FlagLT_UGT) +(TESTQconst [c] (MOVQconst [d])) && int64(c) == d && c > 0 => (FlagGT_UGT) +(TESTLconst [c] (MOVLconst [c])) && c > 0 => (FlagGT_UGT) // TODO: DIVxU also. // Absorb flag constants into SBB ops. -(SBBQcarrymask (FlagEQ)) -> (MOVQconst [0]) -(SBBQcarrymask (FlagLT_ULT)) -> (MOVQconst [-1]) -(SBBQcarrymask (FlagLT_UGT)) -> (MOVQconst [0]) -(SBBQcarrymask (FlagGT_ULT)) -> (MOVQconst [-1]) -(SBBQcarrymask (FlagGT_UGT)) -> (MOVQconst [0]) -(SBBLcarrymask (FlagEQ)) -> (MOVLconst [0]) -(SBBLcarrymask (FlagLT_ULT)) -> (MOVLconst [-1]) -(SBBLcarrymask (FlagLT_UGT)) -> (MOVLconst [0]) -(SBBLcarrymask (FlagGT_ULT)) -> (MOVLconst [-1]) -(SBBLcarrymask (FlagGT_UGT)) -> (MOVLconst [0]) +(SBBQcarrymask (FlagEQ)) => (MOVQconst [0]) +(SBBQcarrymask (FlagLT_ULT)) => (MOVQconst [-1]) +(SBBQcarrymask (FlagLT_UGT)) => (MOVQconst [0]) +(SBBQcarrymask (FlagGT_ULT)) => (MOVQconst [-1]) +(SBBQcarrymask (FlagGT_UGT)) => (MOVQconst [0]) +(SBBLcarrymask (FlagEQ)) => (MOVLconst [0]) +(SBBLcarrymask (FlagLT_ULT)) => (MOVLconst [-1]) +(SBBLcarrymask (FlagLT_UGT)) => (MOVLconst [0]) +(SBBLcarrymask (FlagGT_ULT)) => (MOVLconst [-1]) +(SBBLcarrymask (FlagGT_UGT)) => (MOVLconst [0]) // Absorb flag constants into branches. -((EQ|LE|GE|ULE|UGE) (FlagEQ) yes no) -> (First yes no) -((NE|LT|GT|ULT|UGT) (FlagEQ) yes no) -> (First no yes) -((NE|LT|LE|ULT|ULE) (FlagLT_ULT) yes no) -> (First yes no) -((EQ|GT|GE|UGT|UGE) (FlagLT_ULT) yes no) -> (First no yes) -((NE|LT|LE|UGT|UGE) (FlagLT_UGT) yes no) -> (First yes no) -((EQ|GT|GE|ULT|ULE) (FlagLT_UGT) yes no) -> (First no yes) -((NE|GT|GE|ULT|ULE) (FlagGT_ULT) yes no) -> (First yes no) -((EQ|LT|LE|UGT|UGE) (FlagGT_ULT) yes no) -> (First no yes) -((NE|GT|GE|UGT|UGE) (FlagGT_UGT) yes no) -> (First yes no) -((EQ|LT|LE|ULT|ULE) (FlagGT_UGT) yes no) -> (First no yes) +((EQ|LE|GE|ULE|UGE) (FlagEQ) yes no) => (First yes no) +((NE|LT|GT|ULT|UGT) (FlagEQ) yes no) => (First no yes) +((NE|LT|LE|ULT|ULE) (FlagLT_ULT) yes no) => (First yes no) +((EQ|GT|GE|UGT|UGE) (FlagLT_ULT) yes no) => (First no yes) +((NE|LT|LE|UGT|UGE) (FlagLT_UGT) yes no) => (First yes no) +((EQ|GT|GE|ULT|ULE) (FlagLT_UGT) yes no) => (First no yes) +((NE|GT|GE|ULT|ULE) (FlagGT_ULT) yes no) => (First yes no) +((EQ|LT|LE|UGT|UGE) (FlagGT_ULT) yes no) => (First no yes) +((NE|GT|GE|UGT|UGE) (FlagGT_UGT) yes no) => (First yes no) +((EQ|LT|LE|ULT|ULE) (FlagGT_UGT) yes no) => (First no yes) // Absorb flag constants into SETxx ops. -((SETEQ|SETLE|SETGE|SETBE|SETAE) (FlagEQ)) -> (MOVLconst [1]) -((SETNE|SETL|SETG|SETB|SETA) (FlagEQ)) -> (MOVLconst [0]) -((SETNE|SETL|SETLE|SETB|SETBE) (FlagLT_ULT)) -> (MOVLconst [1]) -((SETEQ|SETG|SETGE|SETA|SETAE) (FlagLT_ULT)) -> (MOVLconst [0]) -((SETNE|SETL|SETLE|SETA|SETAE) (FlagLT_UGT)) -> (MOVLconst [1]) -((SETEQ|SETG|SETGE|SETB|SETBE) (FlagLT_UGT)) -> (MOVLconst [0]) -((SETNE|SETG|SETGE|SETB|SETBE) (FlagGT_ULT)) -> (MOVLconst [1]) -((SETEQ|SETL|SETLE|SETA|SETAE) (FlagGT_ULT)) -> (MOVLconst [0]) -((SETNE|SETG|SETGE|SETA|SETAE) (FlagGT_UGT)) -> (MOVLconst [1]) -((SETEQ|SETL|SETLE|SETB|SETBE) (FlagGT_UGT)) -> (MOVLconst [0]) +((SETEQ|SETLE|SETGE|SETBE|SETAE) (FlagEQ)) => (MOVLconst [1]) +((SETNE|SETL|SETG|SETB|SETA) (FlagEQ)) => (MOVLconst [0]) +((SETNE|SETL|SETLE|SETB|SETBE) (FlagLT_ULT)) => (MOVLconst [1]) +((SETEQ|SETG|SETGE|SETA|SETAE) (FlagLT_ULT)) => (MOVLconst [0]) +((SETNE|SETL|SETLE|SETA|SETAE) (FlagLT_UGT)) => (MOVLconst [1]) +((SETEQ|SETG|SETGE|SETB|SETBE) (FlagLT_UGT)) => (MOVLconst [0]) +((SETNE|SETG|SETGE|SETB|SETBE) (FlagGT_ULT)) => (MOVLconst [1]) +((SETEQ|SETL|SETLE|SETA|SETAE) (FlagGT_ULT)) => (MOVLconst [0]) +((SETNE|SETG|SETGE|SETA|SETAE) (FlagGT_UGT)) => (MOVLconst [1]) +((SETEQ|SETL|SETLE|SETB|SETBE) (FlagGT_UGT)) => (MOVLconst [0]) -(SETEQstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETEQstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETEQstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETEQstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETEQstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETEQstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETEQstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETEQstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETEQstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETEQstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETNEstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETNEstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETNEstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETNEstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETNEstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETNEstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETNEstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETNEstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETNEstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETNEstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETLstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETLstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETLstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETLstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETLstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETLstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETLstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETLstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETLstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETLstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETLEstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETLEstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETLEstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETLEstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETLEstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETLEstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETLEstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETLEstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETLEstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETLEstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETGstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETGstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETGstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETGstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETGstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETGstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETGstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETGstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETGstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETGstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETGEstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETGEstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETGEstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETGEstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETGEstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETGEstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETGEstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETGEstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETGEstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETGEstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETBstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETBstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETBstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETBstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETBstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETBstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETBstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETBstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETBstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETBstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETBEstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETBEstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETBEstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETBEstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETBEstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETBEstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETBEstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETBEstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETBEstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETBEstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETAstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETAstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETAstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETAstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETAstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETAstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETAstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETAstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETAstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETAstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETAEstore [off] {sym} ptr (FlagEQ) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETAEstore [off] {sym} ptr (FlagLT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETAEstore [off] {sym} ptr (FlagLT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) -(SETAEstore [off] {sym} ptr (FlagGT_ULT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) -(SETAEstore [off] {sym} ptr (FlagGT_UGT) mem) -> (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETAEstore [off] {sym} ptr (FlagEQ) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETAEstore [off] {sym} ptr (FlagLT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETAEstore [off] {sym} ptr (FlagLT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) +(SETAEstore [off] {sym} ptr (FlagGT_ULT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [0]) mem) +(SETAEstore [off] {sym} ptr (FlagGT_UGT) mem) => (MOVBstore [off] {sym} ptr (MOVLconst [1]) mem) // Remove redundant *const ops -(ADDQconst [0] x) -> x -(ADDLconst [c] x) && int32(c)==0 -> x -(SUBQconst [0] x) -> x -(SUBLconst [c] x) && int32(c) == 0 -> x -(ANDQconst [0] _) -> (MOVQconst [0]) -(ANDLconst [c] _) && int32(c)==0 -> (MOVLconst [0]) -(ANDQconst [-1] x) -> x -(ANDLconst [c] x) && int32(c)==-1 -> x -(ORQconst [0] x) -> x -(ORLconst [c] x) && int32(c)==0 -> x -(ORQconst [-1] _) -> (MOVQconst [-1]) -(ORLconst [c] _) && int32(c)==-1 -> (MOVLconst [-1]) -(XORQconst [0] x) -> x -(XORLconst [c] x) && int32(c)==0 -> x +(ADDQconst [0] x) => x +(ADDLconst [c] x) && c==0 => x +(SUBQconst [0] x) => x +(SUBLconst [c] x) && c==0 => x +(ANDQconst [0] _) => (MOVQconst [0]) +(ANDLconst [c] _) && c==0 => (MOVLconst [0]) +(ANDQconst [-1] x) => x +(ANDLconst [c] x) && c==-1 => x +(ORQconst [0] x) => x +(ORLconst [c] x) && c==0 => x +(ORQconst [-1] _) => (MOVQconst [-1]) +(ORLconst [c] _) && c==-1 => (MOVLconst [-1]) +(XORQconst [0] x) => x +(XORLconst [c] x) && c==0 => x // TODO: since we got rid of the W/B versions, we might miss // things like (ANDLconst [0x100] x) which were formerly // (ANDBconst [0] x). Probably doesn't happen very often. @@ -1404,99 +1408,99 @@ // Remove redundant ops // Not in generic rules, because they may appear after lowering e. g. Slicemask -(NEG(Q|L) (NEG(Q|L) x)) -> x -(NEG(Q|L) s:(SUB(Q|L) x y)) && s.Uses == 1 -> (SUB(Q|L) y x) +(NEG(Q|L) (NEG(Q|L) x)) => x +(NEG(Q|L) s:(SUB(Q|L) x y)) && s.Uses == 1 => (SUB(Q|L) y x) // Convert constant subtracts to constant adds -(SUBQconst [c] x) && c != -(1<<31) -> (ADDQconst [-c] x) -(SUBLconst [c] x) -> (ADDLconst [int64(int32(-c))] x) +(SUBQconst [c] x) && c != -(1<<31) => (ADDQconst [-c] x) +(SUBLconst [c] x) => (ADDLconst [-c] x) // generic constant folding // TODO: more of this -(ADDQconst [c] (MOVQconst [d])) -> (MOVQconst [c+d]) -(ADDLconst [c] (MOVLconst [d])) -> (MOVLconst [int64(int32(c+d))]) -(ADDQconst [c] (ADDQconst [d] x)) && is32Bit(c+d) -> (ADDQconst [c+d] x) -(ADDLconst [c] (ADDLconst [d] x)) -> (ADDLconst [int64(int32(c+d))] x) -(SUBQconst (MOVQconst [d]) [c]) -> (MOVQconst [d-c]) -(SUBQconst (SUBQconst x [d]) [c]) && is32Bit(-c-d) -> (ADDQconst [-c-d] x) -(SARQconst [c] (MOVQconst [d])) -> (MOVQconst [d>>uint64(c)]) -(SARLconst [c] (MOVQconst [d])) -> (MOVQconst [int64(int32(d))>>uint64(c)]) -(SARWconst [c] (MOVQconst [d])) -> (MOVQconst [int64(int16(d))>>uint64(c)]) -(SARBconst [c] (MOVQconst [d])) -> (MOVQconst [int64(int8(d))>>uint64(c)]) -(NEGQ (MOVQconst [c])) -> (MOVQconst [-c]) -(NEGL (MOVLconst [c])) -> (MOVLconst [int64(int32(-c))]) -(MULQconst [c] (MOVQconst [d])) -> (MOVQconst [c*d]) -(MULLconst [c] (MOVLconst [d])) -> (MOVLconst [int64(int32(c*d))]) -(ANDQconst [c] (MOVQconst [d])) -> (MOVQconst [c&d]) -(ANDLconst [c] (MOVLconst [d])) -> (MOVLconst [c&d]) -(ORQconst [c] (MOVQconst [d])) -> (MOVQconst [c|d]) -(ORLconst [c] (MOVLconst [d])) -> (MOVLconst [c|d]) -(XORQconst [c] (MOVQconst [d])) -> (MOVQconst [c^d]) -(XORLconst [c] (MOVLconst [d])) -> (MOVLconst [c^d]) -(NOTQ (MOVQconst [c])) -> (MOVQconst [^c]) -(NOTL (MOVLconst [c])) -> (MOVLconst [^c]) -(BTSQconst [c] (MOVQconst [d])) -> (MOVQconst [d|(1< (MOVLconst [d|(1< (MOVQconst [d&^(1< (MOVLconst [d&^(1< (MOVQconst [d^(1< (MOVLconst [d^(1< (MOVQconst [int64(c)+d]) +(ADDLconst [c] (MOVLconst [d])) => (MOVLconst [c+d]) +(ADDQconst [c] (ADDQconst [d] x)) && is32Bit(int64(c)+int64(d)) => (ADDQconst [c+d] x) +(ADDLconst [c] (ADDLconst [d] x)) => (ADDLconst [c+d] x) +(SUBQconst (MOVQconst [d]) [c]) => (MOVQconst [d-int64(c)]) +(SUBQconst (SUBQconst x [d]) [c]) && is32Bit(int64(-c)-int64(d)) => (ADDQconst [-c-d] x) +(SARQconst [c] (MOVQconst [d])) => (MOVQconst [d>>uint64(c)]) +(SARLconst [c] (MOVQconst [d])) => (MOVQconst [int64(int32(d))>>uint64(c)]) +(SARWconst [c] (MOVQconst [d])) => (MOVQconst [int64(int16(d))>>uint64(c)]) +(SARBconst [c] (MOVQconst [d])) => (MOVQconst [int64(int8(d))>>uint64(c)]) +(NEGQ (MOVQconst [c])) => (MOVQconst [-c]) +(NEGL (MOVLconst [c])) => (MOVLconst [-c]) +(MULQconst [c] (MOVQconst [d])) => (MOVQconst [int64(c)*d]) +(MULLconst [c] (MOVLconst [d])) => (MOVLconst [c*d]) +(ANDQconst [c] (MOVQconst [d])) => (MOVQconst [int64(c)&d]) +(ANDLconst [c] (MOVLconst [d])) => (MOVLconst [c&d]) +(ORQconst [c] (MOVQconst [d])) => (MOVQconst [int64(c)|d]) +(ORLconst [c] (MOVLconst [d])) => (MOVLconst [c|d]) +(XORQconst [c] (MOVQconst [d])) => (MOVQconst [int64(c)^d]) +(XORLconst [c] (MOVLconst [d])) => (MOVLconst [c^d]) +(NOTQ (MOVQconst [c])) => (MOVQconst [^c]) +(NOTL (MOVLconst [c])) => (MOVLconst [^c]) +(BTSQconst [c] (MOVQconst [d])) => (MOVQconst [d|(1< (MOVLconst [d|(1< (MOVQconst [d&^(1< (MOVLconst [d&^(1< (MOVQconst [d^(1< (MOVLconst [d^(1< (MOVQconst [c|d]) +(ORQ (MOVQconst [c]) (MOVQconst [d])) => (MOVQconst [c|d]) // generic simplifications // TODO: more of this -(ADDQ x (NEGQ y)) -> (SUBQ x y) -(ADDL x (NEGL y)) -> (SUBL x y) -(SUBQ x x) -> (MOVQconst [0]) -(SUBL x x) -> (MOVLconst [0]) -(ANDQ x x) -> x -(ANDL x x) -> x -(ORQ x x) -> x -(ORL x x) -> x -(XORQ x x) -> (MOVQconst [0]) -(XORL x x) -> (MOVLconst [0]) +(ADDQ x (NEGQ y)) => (SUBQ x y) +(ADDL x (NEGL y)) => (SUBL x y) +(SUBQ x x) => (MOVQconst [0]) +(SUBL x x) => (MOVLconst [0]) +(ANDQ x x) => x +(ANDL x x) => x +(ORQ x x) => x +(ORL x x) => x +(XORQ x x) => (MOVQconst [0]) +(XORL x x) => (MOVLconst [0]) -(SHLLconst [d] (MOVLconst [c])) -> (MOVLconst [int64(int32(c)) << uint64(d)]) -(SHLQconst [d] (MOVQconst [c])) -> (MOVQconst [c << uint64(d)]) -(SHLQconst [d] (MOVLconst [c])) -> (MOVQconst [int64(int32(c)) << uint64(d)]) +(SHLLconst [d] (MOVLconst [c])) => (MOVLconst [c << uint64(d)]) +(SHLQconst [d] (MOVQconst [c])) => (MOVQconst [c << uint64(d)]) +(SHLQconst [d] (MOVLconst [c])) => (MOVQconst [int64(c) << uint64(d)]) // Fold NEG into ADDconst/MULconst. Take care to keep c in 32 bit range. -(NEGQ (ADDQconst [c] (NEGQ x))) && c != -(1<<31) -> (ADDQconst [-c] x) -(MULQconst [c] (NEGQ x)) && c != -(1<<31) -> (MULQconst [-c] x) +(NEGQ (ADDQconst [c] (NEGQ x))) && c != -(1<<31) => (ADDQconst [-c] x) +(MULQconst [c] (NEGQ x)) && c != -(1<<31) => (MULQconst [-c] x) // checking AND against 0. -(CMPQconst a:(ANDQ x y) [0]) && a.Uses == 1 -> (TESTQ x y) -(CMPLconst a:(ANDL x y) [0]) && a.Uses == 1 -> (TESTL x y) -(CMPWconst a:(ANDL x y) [0]) && a.Uses == 1 -> (TESTW x y) -(CMPBconst a:(ANDL x y) [0]) && a.Uses == 1 -> (TESTB x y) -(CMPQconst a:(ANDQconst [c] x) [0]) && a.Uses == 1 -> (TESTQconst [c] x) -(CMPLconst a:(ANDLconst [c] x) [0]) && a.Uses == 1 -> (TESTLconst [c] x) -(CMPWconst a:(ANDLconst [c] x) [0]) && a.Uses == 1 -> (TESTWconst [int64(int16(c))] x) -(CMPBconst a:(ANDLconst [c] x) [0]) && a.Uses == 1 -> (TESTBconst [int64(int8(c))] x) +(CMPQconst a:(ANDQ x y) [0]) && a.Uses == 1 => (TESTQ x y) +(CMPLconst a:(ANDL x y) [0]) && a.Uses == 1 => (TESTL x y) +(CMPWconst a:(ANDL x y) [0]) && a.Uses == 1 => (TESTW x y) +(CMPBconst a:(ANDL x y) [0]) && a.Uses == 1 => (TESTB x y) +(CMPQconst a:(ANDQconst [c] x) [0]) && a.Uses == 1 => (TESTQconst [c] x) +(CMPLconst a:(ANDLconst [c] x) [0]) && a.Uses == 1 => (TESTLconst [c] x) +(CMPWconst a:(ANDLconst [c] x) [0]) && a.Uses == 1 => (TESTWconst [int16(c)] x) +(CMPBconst a:(ANDLconst [c] x) [0]) && a.Uses == 1 => (TESTBconst [int8(c)] x) // Convert TESTx to TESTxconst if possible. -(TESTQ (MOVQconst [c]) x) && is32Bit(c) -> (TESTQconst [c] x) -(TESTL (MOVLconst [c]) x) -> (TESTLconst [c] x) -(TESTW (MOVLconst [c]) x) -> (TESTWconst [c] x) -(TESTB (MOVLconst [c]) x) -> (TESTBconst [c] x) +(TESTQ (MOVQconst [c]) x) && is32Bit(c) => (TESTQconst [int32(c)] x) +(TESTL (MOVLconst [c]) x) => (TESTLconst [c] x) +(TESTW (MOVLconst [c]) x) => (TESTWconst [int16(c)] x) +(TESTB (MOVLconst [c]) x) => (TESTBconst [int8(c)] x) // TEST %reg,%reg is shorter than CMP -(CMPQconst x [0]) -> (TESTQ x x) -(CMPLconst x [0]) -> (TESTL x x) -(CMPWconst x [0]) -> (TESTW x x) -(CMPBconst x [0]) -> (TESTB x x) -(TESTQconst [-1] x) && x.Op != OpAMD64MOVQconst -> (TESTQ x x) -(TESTLconst [-1] x) && x.Op != OpAMD64MOVLconst -> (TESTL x x) -(TESTWconst [-1] x) && x.Op != OpAMD64MOVLconst -> (TESTW x x) -(TESTBconst [-1] x) && x.Op != OpAMD64MOVLconst -> (TESTB x x) +(CMPQconst x [0]) => (TESTQ x x) +(CMPLconst x [0]) => (TESTL x x) +(CMPWconst x [0]) => (TESTW x x) +(CMPBconst x [0]) => (TESTB x x) +(TESTQconst [-1] x) && x.Op != OpAMD64MOVQconst => (TESTQ x x) +(TESTLconst [-1] x) && x.Op != OpAMD64MOVLconst => (TESTL x x) +(TESTWconst [-1] x) && x.Op != OpAMD64MOVLconst => (TESTW x x) +(TESTBconst [-1] x) && x.Op != OpAMD64MOVLconst => (TESTB x x) // Convert LEAQ1 back to ADDQ if we can -(LEAQ1 [0] x y) && v.Aux == nil -> (ADDQ x y) +(LEAQ1 [0] x y) && v.Aux == nil => (ADDQ x y) // Combining byte loads into larger (unaligned) loads. // There are many ways these combinations could occur. This is @@ -1512,7 +1516,7 @@ && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0, x1, sh) - -> @mergePoint(b,x0,x1) (MOVWload [i0] {s} p mem) + => @mergePoint(b,x0,x1) (MOVWload [i0] {s} p mem) (OR(L|Q) x0:(MOVBload [i] {s} p0 mem) sh:(SHL(L|Q)const [8] x1:(MOVBload [i] {s} p1 mem))) @@ -1522,7 +1526,7 @@ && sequentialAddresses(p0, p1, 1) && mergePoint(b,x0,x1) != nil && clobber(x0, x1, sh) - -> @mergePoint(b,x0,x1) (MOVWload [i] {s} p0 mem) + => @mergePoint(b,x0,x1) (MOVWload [i] {s} p0 mem) (OR(L|Q) x0:(MOVWload [i0] {s} p mem) sh:(SHL(L|Q)const [16] x1:(MOVWload [i1] {s} p mem))) @@ -1532,7 +1536,7 @@ && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0, x1, sh) - -> @mergePoint(b,x0,x1) (MOVLload [i0] {s} p mem) + => @mergePoint(b,x0,x1) (MOVLload [i0] {s} p mem) (OR(L|Q) x0:(MOVWload [i] {s} p0 mem) sh:(SHL(L|Q)const [16] x1:(MOVWload [i] {s} p1 mem))) @@ -1542,7 +1546,7 @@ && sequentialAddresses(p0, p1, 2) && mergePoint(b,x0,x1) != nil && clobber(x0, x1, sh) - -> @mergePoint(b,x0,x1) (MOVLload [i] {s} p0 mem) + => @mergePoint(b,x0,x1) (MOVLload [i] {s} p0 mem) (ORQ x0:(MOVLload [i0] {s} p mem) sh:(SHLQconst [32] x1:(MOVLload [i1] {s} p mem))) @@ -1552,7 +1556,7 @@ && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0, x1, sh) - -> @mergePoint(b,x0,x1) (MOVQload [i0] {s} p mem) + => @mergePoint(b,x0,x1) (MOVQload [i0] {s} p mem) (ORQ x0:(MOVLload [i] {s} p0 mem) sh:(SHLQconst [32] x1:(MOVLload [i] {s} p1 mem))) @@ -1562,7 +1566,7 @@ && sequentialAddresses(p0, p1, 4) && mergePoint(b,x0,x1) != nil && clobber(x0, x1, sh) - -> @mergePoint(b,x0,x1) (MOVQload [i] {s} p0 mem) + => @mergePoint(b,x0,x1) (MOVQload [i] {s} p0 mem) (OR(L|Q) s1:(SHL(L|Q)const [j1] x1:(MOVBload [i1] {s} p mem)) @@ -1579,7 +1583,7 @@ && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0, x1, s0, s1, or) - -> @mergePoint(b,x0,x1,y) (OR(L|Q) (SHL(L|Q)const [j0] (MOVWload [i0] {s} p mem)) y) + => @mergePoint(b,x0,x1,y) (OR(L|Q) (SHL(L|Q)const [j0] (MOVWload [i0] {s} p mem)) y) (OR(L|Q) s1:(SHL(L|Q)const [j1] x1:(MOVBload [i] {s} p1 mem)) @@ -1596,7 +1600,7 @@ && sequentialAddresses(p0, p1, 1) && mergePoint(b,x0,x1,y) != nil && clobber(x0, x1, s0, s1, or) - -> @mergePoint(b,x0,x1,y) (OR(L|Q) (SHL(L|Q)const [j0] (MOVWload [i] {s} p0 mem)) y) + => @mergePoint(b,x0,x1,y) (OR(L|Q) (SHL(L|Q)const [j0] (MOVWload [i] {s} p0 mem)) y) (ORQ s1:(SHLQconst [j1] x1:(MOVWload [i1] {s} p mem)) @@ -1613,7 +1617,7 @@ && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0, x1, s0, s1, or) - -> @mergePoint(b,x0,x1,y) (ORQ (SHLQconst [j0] (MOVLload [i0] {s} p mem)) y) + => @mergePoint(b,x0,x1,y) (ORQ (SHLQconst [j0] (MOVLload [i0] {s} p mem)) y) (ORQ s1:(SHLQconst [j1] x1:(MOVWload [i] {s} p1 mem)) @@ -1630,7 +1634,7 @@ && sequentialAddresses(p0, p1, 2) && mergePoint(b,x0,x1,y) != nil && clobber(x0, x1, s0, s1, or) - -> @mergePoint(b,x0,x1,y) (ORQ (SHLQconst [j0] (MOVLload [i] {s} p0 mem)) y) + => @mergePoint(b,x0,x1,y) (ORQ (SHLQconst [j0] (MOVLload [i] {s} p0 mem)) y) // Big-endian loads @@ -1643,7 +1647,7 @@ && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0, x1, sh) - -> @mergePoint(b,x0,x1) (ROLWconst [8] (MOVWload [i0] {s} p mem)) + => @mergePoint(b,x0,x1) (ROLWconst [8] (MOVWload [i0] {s} p mem)) (OR(L|Q) x1:(MOVBload [i] {s} p1 mem) @@ -1654,7 +1658,7 @@ && sequentialAddresses(p0, p1, 1) && mergePoint(b,x0,x1) != nil && clobber(x0, x1, sh) - -> @mergePoint(b,x0,x1) (ROLWconst [8] (MOVWload [i] {s} p0 mem)) + => @mergePoint(b,x0,x1) (ROLWconst [8] (MOVWload [i] {s} p0 mem)) (OR(L|Q) r1:(ROLWconst [8] x1:(MOVWload [i1] {s} p mem)) @@ -1667,7 +1671,7 @@ && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0, x1, r0, r1, sh) - -> @mergePoint(b,x0,x1) (BSWAPL (MOVLload [i0] {s} p mem)) + => @mergePoint(b,x0,x1) (BSWAPL (MOVLload [i0] {s} p mem)) (OR(L|Q) r1:(ROLWconst [8] x1:(MOVWload [i] {s} p1 mem)) @@ -1680,7 +1684,7 @@ && sequentialAddresses(p0, p1, 2) && mergePoint(b,x0,x1) != nil && clobber(x0, x1, r0, r1, sh) - -> @mergePoint(b,x0,x1) (BSWAPL (MOVLload [i] {s} p0 mem)) + => @mergePoint(b,x0,x1) (BSWAPL (MOVLload [i] {s} p0 mem)) (ORQ r1:(BSWAPL x1:(MOVLload [i1] {s} p mem)) @@ -1693,7 +1697,7 @@ && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0, x1, r0, r1, sh) - -> @mergePoint(b,x0,x1) (BSWAPQ (MOVQload [i0] {s} p mem)) + => @mergePoint(b,x0,x1) (BSWAPQ (MOVQload [i0] {s} p mem)) (ORQ r1:(BSWAPL x1:(MOVLload [i] {s} p1 mem)) @@ -1706,7 +1710,7 @@ && sequentialAddresses(p0, p1, 4) && mergePoint(b,x0,x1) != nil && clobber(x0, x1, r0, r1, sh) - -> @mergePoint(b,x0,x1) (BSWAPQ (MOVQload [i] {s} p0 mem)) + => @mergePoint(b,x0,x1) (BSWAPQ (MOVQload [i] {s} p0 mem)) (OR(L|Q) s0:(SHL(L|Q)const [j0] x0:(MOVBload [i0] {s} p mem)) @@ -1723,7 +1727,7 @@ && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0, x1, s0, s1, or) - -> @mergePoint(b,x0,x1,y) (OR(L|Q) (SHL(L|Q)const [j1] (ROLWconst [8] (MOVWload [i0] {s} p mem))) y) + => @mergePoint(b,x0,x1,y) (OR(L|Q) (SHL(L|Q)const [j1] (ROLWconst [8] (MOVWload [i0] {s} p mem))) y) (OR(L|Q) s0:(SHL(L|Q)const [j0] x0:(MOVBload [i] {s} p0 mem)) @@ -1740,7 +1744,7 @@ && sequentialAddresses(p0, p1, 1) && mergePoint(b,x0,x1,y) != nil && clobber(x0, x1, s0, s1, or) - -> @mergePoint(b,x0,x1,y) (OR(L|Q) (SHL(L|Q)const [j1] (ROLWconst [8] (MOVWload [i] {s} p0 mem))) y) + => @mergePoint(b,x0,x1,y) (OR(L|Q) (SHL(L|Q)const [j1] (ROLWconst [8] (MOVWload [i] {s} p0 mem))) y) (ORQ s0:(SHLQconst [j0] r0:(ROLWconst [8] x0:(MOVWload [i0] {s} p mem))) @@ -1759,7 +1763,7 @@ && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0, x1, r0, r1, s0, s1, or) - -> @mergePoint(b,x0,x1,y) (ORQ (SHLQconst [j1] (BSWAPL (MOVLload [i0] {s} p mem))) y) + => @mergePoint(b,x0,x1,y) (ORQ (SHLQconst [j1] (BSWAPL (MOVLload [i0] {s} p mem))) y) (ORQ s0:(SHLQconst [j0] r0:(ROLWconst [8] x0:(MOVWload [i] {s} p0 mem))) @@ -1778,20 +1782,20 @@ && sequentialAddresses(p0, p1, 2) && mergePoint(b,x0,x1,y) != nil && clobber(x0, x1, r0, r1, s0, s1, or) - -> @mergePoint(b,x0,x1,y) (ORQ (SHLQconst [j1] (BSWAPL (MOVLload [i] {s} p0 mem))) y) + => @mergePoint(b,x0,x1,y) (ORQ (SHLQconst [j1] (BSWAPL (MOVLload [i] {s} p0 mem))) y) // Combine 2 byte stores + shift into rolw 8 + word store (MOVBstore [i] {s} p w x0:(MOVBstore [i-1] {s} p (SHRWconst [8] w) mem)) && x0.Uses == 1 && clobber(x0) - -> (MOVWstore [i-1] {s} p (ROLWconst [8] w) mem) + => (MOVWstore [i-1] {s} p (ROLWconst [8] w) mem) (MOVBstore [i] {s} p1 w x0:(MOVBstore [i] {s} p0 (SHRWconst [8] w) mem)) && x0.Uses == 1 && sequentialAddresses(p0, p1, 1) && clobber(x0) - -> (MOVWstore [i] {s} p0 (ROLWconst [8] w) mem) + => (MOVWstore [i] {s} p0 (ROLWconst [8] w) mem) // Combine stores + shifts into bswap and larger (unaligned) stores (MOVBstore [i] {s} p w @@ -1802,7 +1806,7 @@ && x1.Uses == 1 && x2.Uses == 1 && clobber(x0, x1, x2) - -> (MOVLstore [i-3] {s} p (BSWAPL w) mem) + => (MOVLstore [i-3] {s} p (BSWAPL w) mem) (MOVBstore [i] {s} p3 w x2:(MOVBstore [i] {s} p2 (SHRLconst [8] w) x1:(MOVBstore [i] {s} p1 (SHRLconst [16] w) @@ -1814,7 +1818,7 @@ && sequentialAddresses(p1, p2, 1) && sequentialAddresses(p2, p3, 1) && clobber(x0, x1, x2) - -> (MOVLstore [i] {s} p0 (BSWAPL w) mem) + => (MOVLstore [i] {s} p0 (BSWAPL w) mem) (MOVBstore [i] {s} p w x6:(MOVBstore [i-1] {s} p (SHRQconst [8] w) @@ -1832,7 +1836,7 @@ && x5.Uses == 1 && x6.Uses == 1 && clobber(x0, x1, x2, x3, x4, x5, x6) - -> (MOVQstore [i-7] {s} p (BSWAPQ w) mem) + => (MOVQstore [i-7] {s} p (BSWAPQ w) mem) (MOVBstore [i] {s} p7 w x6:(MOVBstore [i] {s} p6 (SHRQconst [8] w) x5:(MOVBstore [i] {s} p5 (SHRQconst [16] w) @@ -1856,114 +1860,114 @@ && sequentialAddresses(p5, p6, 1) && sequentialAddresses(p6, p7, 1) && clobber(x0, x1, x2, x3, x4, x5, x6) - -> (MOVQstore [i] {s} p0 (BSWAPQ w) mem) + => (MOVQstore [i] {s} p0 (BSWAPQ w) mem) // Combine constant stores into larger (unaligned) stores. (MOVBstoreconst [c] {s} p x:(MOVBstoreconst [a] {s} p mem)) && x.Uses == 1 - && ValAndOff(a).Off() + 1 == ValAndOff(c).Off() + && a.Off() + 1 == c.Off() && clobber(x) - -> (MOVWstoreconst [makeValAndOff(ValAndOff(a).Val()&0xff | ValAndOff(c).Val()<<8, ValAndOff(a).Off())] {s} p mem) + => (MOVWstoreconst [makeValAndOff64(a.Val()&0xff | c.Val()<<8, a.Off())] {s} p mem) (MOVBstoreconst [a] {s} p x:(MOVBstoreconst [c] {s} p mem)) && x.Uses == 1 - && ValAndOff(a).Off() + 1 == ValAndOff(c).Off() + && a.Off() + 1 == c.Off() && clobber(x) - -> (MOVWstoreconst [makeValAndOff(ValAndOff(a).Val()&0xff | ValAndOff(c).Val()<<8, ValAndOff(a).Off())] {s} p mem) + => (MOVWstoreconst [makeValAndOff64(a.Val()&0xff | c.Val()<<8, a.Off())] {s} p mem) (MOVWstoreconst [c] {s} p x:(MOVWstoreconst [a] {s} p mem)) && x.Uses == 1 - && ValAndOff(a).Off() + 2 == ValAndOff(c).Off() + && a.Off() + 2 == c.Off() && clobber(x) - -> (MOVLstoreconst [makeValAndOff(ValAndOff(a).Val()&0xffff | ValAndOff(c).Val()<<16, ValAndOff(a).Off())] {s} p mem) + => (MOVLstoreconst [makeValAndOff64(a.Val()&0xffff | c.Val()<<16, a.Off())] {s} p mem) (MOVWstoreconst [a] {s} p x:(MOVWstoreconst [c] {s} p mem)) && x.Uses == 1 - && ValAndOff(a).Off() + 2 == ValAndOff(c).Off() + && a.Off() + 2 == c.Off() && clobber(x) - -> (MOVLstoreconst [makeValAndOff(ValAndOff(a).Val()&0xffff | ValAndOff(c).Val()<<16, ValAndOff(a).Off())] {s} p mem) + => (MOVLstoreconst [makeValAndOff64(a.Val()&0xffff | c.Val()<<16, a.Off())] {s} p mem) (MOVLstoreconst [c] {s} p x:(MOVLstoreconst [a] {s} p mem)) && x.Uses == 1 - && ValAndOff(a).Off() + 4 == ValAndOff(c).Off() + && a.Off() + 4 == c.Off() && clobber(x) - -> (MOVQstore [ValAndOff(a).Off()] {s} p (MOVQconst [ValAndOff(a).Val()&0xffffffff | ValAndOff(c).Val()<<32]) mem) + => (MOVQstore [a.Off32()] {s} p (MOVQconst [a.Val()&0xffffffff | c.Val()<<32]) mem) (MOVLstoreconst [a] {s} p x:(MOVLstoreconst [c] {s} p mem)) && x.Uses == 1 - && ValAndOff(a).Off() + 4 == ValAndOff(c).Off() + && a.Off() + 4 == c.Off() && clobber(x) - -> (MOVQstore [ValAndOff(a).Off()] {s} p (MOVQconst [ValAndOff(a).Val()&0xffffffff | ValAndOff(c).Val()<<32]) mem) + => (MOVQstore [a.Off32()] {s} p (MOVQconst [a.Val()&0xffffffff | c.Val()<<32]) mem) (MOVQstoreconst [c] {s} p x:(MOVQstoreconst [c2] {s} p mem)) && config.useSSE && x.Uses == 1 - && ValAndOff(c2).Off() + 8 == ValAndOff(c).Off() - && ValAndOff(c).Val() == 0 - && ValAndOff(c2).Val() == 0 + && c2.Off() + 8 == c.Off() + && c.Val() == 0 + && c2.Val() == 0 && clobber(x) - -> (MOVOstore [ValAndOff(c2).Off()] {s} p (MOVOconst [0]) mem) + => (MOVOstore [c2.Off32()] {s} p (MOVOconst [0]) mem) // Combine stores into larger (unaligned) stores. Little endian. (MOVBstore [i] {s} p (SHR(W|L|Q)const [8] w) x:(MOVBstore [i-1] {s} p w mem)) && x.Uses == 1 && clobber(x) - -> (MOVWstore [i-1] {s} p w mem) + => (MOVWstore [i-1] {s} p w mem) (MOVBstore [i] {s} p w x:(MOVBstore [i+1] {s} p (SHR(W|L|Q)const [8] w) mem)) && x.Uses == 1 && clobber(x) - -> (MOVWstore [i] {s} p w mem) + => (MOVWstore [i] {s} p w mem) (MOVBstore [i] {s} p (SHR(L|Q)const [j] w) x:(MOVBstore [i-1] {s} p w0:(SHR(L|Q)const [j-8] w) mem)) && x.Uses == 1 && clobber(x) - -> (MOVWstore [i-1] {s} p w0 mem) + => (MOVWstore [i-1] {s} p w0 mem) (MOVBstore [i] {s} p1 (SHR(W|L|Q)const [8] w) x:(MOVBstore [i] {s} p0 w mem)) && x.Uses == 1 && sequentialAddresses(p0, p1, 1) && clobber(x) - -> (MOVWstore [i] {s} p0 w mem) + => (MOVWstore [i] {s} p0 w mem) (MOVBstore [i] {s} p0 w x:(MOVBstore [i] {s} p1 (SHR(W|L|Q)const [8] w) mem)) && x.Uses == 1 && sequentialAddresses(p0, p1, 1) && clobber(x) - -> (MOVWstore [i] {s} p0 w mem) + => (MOVWstore [i] {s} p0 w mem) (MOVBstore [i] {s} p1 (SHR(L|Q)const [j] w) x:(MOVBstore [i] {s} p0 w0:(SHR(L|Q)const [j-8] w) mem)) && x.Uses == 1 && sequentialAddresses(p0, p1, 1) && clobber(x) - -> (MOVWstore [i] {s} p0 w0 mem) + => (MOVWstore [i] {s} p0 w0 mem) (MOVWstore [i] {s} p (SHR(L|Q)const [16] w) x:(MOVWstore [i-2] {s} p w mem)) && x.Uses == 1 && clobber(x) - -> (MOVLstore [i-2] {s} p w mem) + => (MOVLstore [i-2] {s} p w mem) (MOVWstore [i] {s} p (SHR(L|Q)const [j] w) x:(MOVWstore [i-2] {s} p w0:(SHR(L|Q)const [j-16] w) mem)) && x.Uses == 1 && clobber(x) - -> (MOVLstore [i-2] {s} p w0 mem) + => (MOVLstore [i-2] {s} p w0 mem) (MOVWstore [i] {s} p1 (SHR(L|Q)const [16] w) x:(MOVWstore [i] {s} p0 w mem)) && x.Uses == 1 && sequentialAddresses(p0, p1, 2) && clobber(x) - -> (MOVLstore [i] {s} p0 w mem) + => (MOVLstore [i] {s} p0 w mem) (MOVWstore [i] {s} p1 (SHR(L|Q)const [j] w) x:(MOVWstore [i] {s} p0 w0:(SHR(L|Q)const [j-16] w) mem)) && x.Uses == 1 && sequentialAddresses(p0, p1, 2) && clobber(x) - -> (MOVLstore [i] {s} p0 w0 mem) + => (MOVLstore [i] {s} p0 w0 mem) (MOVLstore [i] {s} p (SHRQconst [32] w) x:(MOVLstore [i-4] {s} p w mem)) && x.Uses == 1 && clobber(x) - -> (MOVQstore [i-4] {s} p w mem) + => (MOVQstore [i-4] {s} p w mem) (MOVLstore [i] {s} p (SHRQconst [j] w) x:(MOVLstore [i-4] {s} p w0:(SHRQconst [j-32] w) mem)) && x.Uses == 1 && clobber(x) - -> (MOVQstore [i-4] {s} p w0 mem) + => (MOVQstore [i-4] {s} p w0 mem) (MOVLstore [i] {s} p1 (SHRQconst [32] w) x:(MOVLstore [i] {s} p0 w mem)) && x.Uses == 1 && sequentialAddresses(p0, p1, 4) && clobber(x) - -> (MOVQstore [i] {s} p0 w mem) + => (MOVQstore [i] {s} p0 w mem) (MOVLstore [i] {s} p1 (SHRQconst [j] w) x:(MOVLstore [i] {s} p0 w0:(SHRQconst [j-32] w) mem)) && x.Uses == 1 && sequentialAddresses(p0, p1, 4) && clobber(x) - -> (MOVQstore [i] {s} p0 w0 mem) + => (MOVQstore [i] {s} p0 w0 mem) (MOVBstore [i] {s} p x1:(MOVBload [j] {s2} p2 mem) @@ -1973,7 +1977,7 @@ && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1, x2, mem2) - -> (MOVWstore [i-1] {s} p (MOVWload [j-1] {s2} p2 mem) mem) + => (MOVWstore [i-1] {s} p (MOVWload [j-1] {s2} p2 mem) mem) (MOVWstore [i] {s} p x1:(MOVWload [j] {s2} p2 mem) @@ -1983,7 +1987,7 @@ && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1, x2, mem2) - -> (MOVLstore [i-2] {s} p (MOVLload [j-2] {s2} p2 mem) mem) + => (MOVLstore [i-2] {s} p (MOVLload [j-2] {s2} p2 mem) mem) (MOVLstore [i] {s} p x1:(MOVLload [j] {s2} p2 mem) @@ -1993,178 +1997,178 @@ && x2.Uses == 1 && mem2.Uses == 1 && clobber(x1, x2, mem2) - -> (MOVQstore [i-4] {s} p (MOVQload [j-4] {s2} p2 mem) mem) + => (MOVQstore [i-4] {s} p (MOVQload [j-4] {s2} p2 mem) mem) -(MOVQload [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) -> +(MOVQload [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(int64(off1)+int64(off2)) => (MOVQload [off1+off2] {mergeSym(sym1,sym2)} base mem) -(MOVLload [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) -> +(MOVLload [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(int64(off1)+int64(off2)) => (MOVLload [off1+off2] {mergeSym(sym1,sym2)} base mem) -(MOVWload [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) -> +(MOVWload [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(int64(off1)+int64(off2)) => (MOVWload [off1+off2] {mergeSym(sym1,sym2)} base mem) -(MOVBload [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) -> +(MOVBload [off1] {sym1} (LEAL [off2] {sym2} base) mem) && canMergeSym(sym1, sym2) && is32Bit(int64(off1)+int64(off2)) => (MOVBload [off1+off2] {mergeSym(sym1,sym2)} base mem) -(MOVQstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) -> +(MOVQstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(int64(off1)+int64(off2)) => (MOVQstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) -(MOVLstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) -> +(MOVLstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(int64(off1)+int64(off2)) => (MOVLstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) -(MOVWstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) -> +(MOVWstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(int64(off1)+int64(off2)) => (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) -(MOVBstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(off1+off2) -> +(MOVBstore [off1] {sym1} (LEAL [off2] {sym2} base) val mem) && canMergeSym(sym1, sym2) && is32Bit(int64(off1)+int64(off2)) => (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) -(MOVQstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) -> - (MOVQstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem) -(MOVLstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) -> - (MOVLstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem) -(MOVWstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) -> - (MOVWstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem) -(MOVBstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && ValAndOff(sc).canAdd(off) -> - (MOVBstoreconst [ValAndOff(sc).add(off)] {mergeSym(sym1, sym2)} ptr mem) +(MOVQstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && sc.canAdd32(off) => + (MOVQstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) +(MOVLstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && sc.canAdd32(off) => + (MOVLstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) +(MOVWstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && sc.canAdd32(off) => + (MOVWstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) +(MOVBstoreconst [sc] {sym1} (LEAL [off] {sym2} ptr) mem) && canMergeSym(sym1, sym2) && sc.canAdd32(off) => + (MOVBstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) -(MOVQload [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVQload [off1+off2] {sym} ptr mem) -(MOVLload [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVLload [off1+off2] {sym} ptr mem) -(MOVWload [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVWload [off1+off2] {sym} ptr mem) -(MOVBload [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(off1+off2) -> (MOVBload [off1+off2] {sym} ptr mem) -(MOVQstore [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVQstore [off1+off2] {sym} ptr val mem) -(MOVLstore [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVLstore [off1+off2] {sym} ptr val mem) -(MOVWstore [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVWstore [off1+off2] {sym} ptr val mem) -(MOVBstore [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(off1+off2) -> (MOVBstore [off1+off2] {sym} ptr val mem) -(MOVQstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) -> - (MOVQstoreconst [ValAndOff(sc).add(off)] {s} ptr mem) -(MOVLstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) -> - (MOVLstoreconst [ValAndOff(sc).add(off)] {s} ptr mem) -(MOVWstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) -> - (MOVWstoreconst [ValAndOff(sc).add(off)] {s} ptr mem) -(MOVBstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && ValAndOff(sc).canAdd(off) -> - (MOVBstoreconst [ValAndOff(sc).add(off)] {s} ptr mem) +(MOVQload [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (MOVQload [off1+off2] {sym} ptr mem) +(MOVLload [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (MOVLload [off1+off2] {sym} ptr mem) +(MOVWload [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (MOVWload [off1+off2] {sym} ptr mem) +(MOVBload [off1] {sym} (ADDLconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (MOVBload [off1+off2] {sym} ptr mem) +(MOVQstore [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(int64(off1)+int64(off2)) => (MOVQstore [off1+off2] {sym} ptr val mem) +(MOVLstore [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(int64(off1)+int64(off2)) => (MOVLstore [off1+off2] {sym} ptr val mem) +(MOVWstore [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(int64(off1)+int64(off2)) => (MOVWstore [off1+off2] {sym} ptr val mem) +(MOVBstore [off1] {sym} (ADDLconst [off2] ptr) val mem) && is32Bit(int64(off1)+int64(off2)) => (MOVBstore [off1+off2] {sym} ptr val mem) +(MOVQstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && sc.canAdd32(off) => + (MOVQstoreconst [sc.addOffset32(off)] {s} ptr mem) +(MOVLstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && sc.canAdd32(off) => + (MOVLstoreconst [sc.addOffset32(off)] {s} ptr mem) +(MOVWstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && sc.canAdd32(off) => + (MOVWstoreconst [sc.addOffset32(off)] {s} ptr mem) +(MOVBstoreconst [sc] {s} (ADDLconst [off] ptr) mem) && sc.canAdd32(off) => + (MOVBstoreconst [sc.addOffset32(off)] {s} ptr mem) // Merge load and op // TODO: add indexed variants? -((ADD|SUB|AND|OR|XOR)Q x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|SUB|AND|OR|XOR)Qload x [off] {sym} ptr mem) -((ADD|SUB|AND|OR|XOR)L x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|SUB|AND|OR|XOR)Lload x [off] {sym} ptr mem) -((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem) -((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem) -(MOVLstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Lload x [off] {sym} ptr mem) mem) && y.Uses==1 && clobber(y) -> ((ADD|AND|OR|XOR)Lmodify [off] {sym} ptr x mem) -(MOVLstore {sym} [off] ptr y:((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)L l:(MOVLload [off] {sym} ptr mem) x) mem) && y.Uses==1 && l.Uses==1 && clobber(y, l) -> +((ADD|SUB|AND|OR|XOR)Q x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|SUB|AND|OR|XOR)Qload x [off] {sym} ptr mem) +((ADD|SUB|AND|OR|XOR)L x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|SUB|AND|OR|XOR)Lload x [off] {sym} ptr mem) +((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem) +((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) => ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem) +(MOVLstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Lload x [off] {sym} ptr mem) mem) && y.Uses==1 && clobber(y) => ((ADD|AND|OR|XOR)Lmodify [off] {sym} ptr x mem) +(MOVLstore {sym} [off] ptr y:((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)L l:(MOVLload [off] {sym} ptr mem) x) mem) && y.Uses==1 && l.Uses==1 && clobber(y, l) => ((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Lmodify [off] {sym} ptr x mem) -(MOVQstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Qload x [off] {sym} ptr mem) mem) && y.Uses==1 && clobber(y) -> ((ADD|AND|OR|XOR)Qmodify [off] {sym} ptr x mem) -(MOVQstore {sym} [off] ptr y:((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Q l:(MOVQload [off] {sym} ptr mem) x) mem) && y.Uses==1 && l.Uses==1 && clobber(y, l) -> +(MOVQstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Qload x [off] {sym} ptr mem) mem) && y.Uses==1 && clobber(y) => ((ADD|AND|OR|XOR)Qmodify [off] {sym} ptr x mem) +(MOVQstore {sym} [off] ptr y:((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Q l:(MOVQload [off] {sym} ptr mem) x) mem) && y.Uses==1 && l.Uses==1 && clobber(y, l) => ((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Qmodify [off] {sym} ptr x mem) // Merge ADDQconst and LEAQ into atomic loads. -(MOV(Q|L|B)atomicload [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> +(MOV(Q|L|B)atomicload [off1] {sym} (ADDQconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (MOV(Q|L|B)atomicload [off1+off2] {sym} ptr mem) -(MOV(Q|L|B)atomicload [off1] {sym1} (LEAQ [off2] {sym2} ptr) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) -> - (MOV(Q|L|B)atomicload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) +(MOV(Q|L|B)atomicload [off1] {sym1} (LEAQ [off2] {sym2} ptr) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => + (MOV(Q|L|B)atomicload [off1+off2] {mergeSym(sym1, sym2)} ptr mem) // Merge ADDQconst and LEAQ into atomic stores. -(XCHGQ [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> +(XCHGQ [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (XCHGQ [off1+off2] {sym} val ptr mem) -(XCHGQ [off1] {sym1} val (LEAQ [off2] {sym2} ptr) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && ptr.Op != OpSB -> +(XCHGQ [off1] {sym1} val (LEAQ [off2] {sym2} ptr) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && ptr.Op != OpSB => (XCHGQ [off1+off2] {mergeSym(sym1,sym2)} val ptr mem) -(XCHGL [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> +(XCHGL [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (XCHGL [off1+off2] {sym} val ptr mem) -(XCHGL [off1] {sym1} val (LEAQ [off2] {sym2} ptr) mem) && is32Bit(off1+off2) && canMergeSym(sym1, sym2) && ptr.Op != OpSB -> +(XCHGL [off1] {sym1} val (LEAQ [off2] {sym2} ptr) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && ptr.Op != OpSB => (XCHGL [off1+off2] {mergeSym(sym1,sym2)} val ptr mem) // Merge ADDQconst into atomic adds. // TODO: merging LEAQ doesn't work, assembler doesn't like the resulting instructions. -(XADDQlock [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> +(XADDQlock [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (XADDQlock [off1+off2] {sym} val ptr mem) -(XADDLlock [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(off1+off2) -> +(XADDLlock [off1] {sym} val (ADDQconst [off2] ptr) mem) && is32Bit(int64(off1)+int64(off2)) => (XADDLlock [off1+off2] {sym} val ptr mem) // Merge ADDQconst into atomic compare and swaps. // TODO: merging LEAQ doesn't work, assembler doesn't like the resulting instructions. -(CMPXCHGQlock [off1] {sym} (ADDQconst [off2] ptr) old new_ mem) && is32Bit(off1+off2) -> +(CMPXCHGQlock [off1] {sym} (ADDQconst [off2] ptr) old new_ mem) && is32Bit(int64(off1)+int64(off2)) => (CMPXCHGQlock [off1+off2] {sym} ptr old new_ mem) -(CMPXCHGLlock [off1] {sym} (ADDQconst [off2] ptr) old new_ mem) && is32Bit(off1+off2) -> +(CMPXCHGLlock [off1] {sym} (ADDQconst [off2] ptr) old new_ mem) && is32Bit(int64(off1)+int64(off2)) => (CMPXCHGLlock [off1+off2] {sym} ptr old new_ mem) // We don't need the conditional move if we know the arg of BSF is not zero. -(CMOVQEQ x _ (Select1 (BSFQ (ORQconst [c] _)))) && c != 0 -> x +(CMOVQEQ x _ (Select1 (BSFQ (ORQconst [c] _)))) && c != 0 => x // Extension is unnecessary for trailing zeros. -(BSFQ (ORQconst [1<<8] (MOVBQZX x))) -> (BSFQ (ORQconst [1<<8] x)) -(BSFQ (ORQconst [1<<16] (MOVWQZX x))) -> (BSFQ (ORQconst [1<<16] x)) +(BSFQ (ORQconst [1<<8] (MOVBQZX x))) => (BSFQ (ORQconst [1<<8] x)) +(BSFQ (ORQconst [1<<16] (MOVWQZX x))) => (BSFQ (ORQconst [1<<16] x)) // Redundant sign/zero extensions // Note: see issue 21963. We have to make sure we use the right type on // the resulting extension (the outer type, not the inner type). -(MOVLQSX (MOVLQSX x)) -> (MOVLQSX x) -(MOVLQSX (MOVWQSX x)) -> (MOVWQSX x) -(MOVLQSX (MOVBQSX x)) -> (MOVBQSX x) -(MOVWQSX (MOVWQSX x)) -> (MOVWQSX x) -(MOVWQSX (MOVBQSX x)) -> (MOVBQSX x) -(MOVBQSX (MOVBQSX x)) -> (MOVBQSX x) -(MOVLQZX (MOVLQZX x)) -> (MOVLQZX x) -(MOVLQZX (MOVWQZX x)) -> (MOVWQZX x) -(MOVLQZX (MOVBQZX x)) -> (MOVBQZX x) -(MOVWQZX (MOVWQZX x)) -> (MOVWQZX x) -(MOVWQZX (MOVBQZX x)) -> (MOVBQZX x) -(MOVBQZX (MOVBQZX x)) -> (MOVBQZX x) +(MOVLQSX (MOVLQSX x)) => (MOVLQSX x) +(MOVLQSX (MOVWQSX x)) => (MOVWQSX x) +(MOVLQSX (MOVBQSX x)) => (MOVBQSX x) +(MOVWQSX (MOVWQSX x)) => (MOVWQSX x) +(MOVWQSX (MOVBQSX x)) => (MOVBQSX x) +(MOVBQSX (MOVBQSX x)) => (MOVBQSX x) +(MOVLQZX (MOVLQZX x)) => (MOVLQZX x) +(MOVLQZX (MOVWQZX x)) => (MOVWQZX x) +(MOVLQZX (MOVBQZX x)) => (MOVBQZX x) +(MOVWQZX (MOVWQZX x)) => (MOVWQZX x) +(MOVWQZX (MOVBQZX x)) => (MOVBQZX x) +(MOVBQZX (MOVBQZX x)) => (MOVBQZX x) (MOVQstore [off] {sym} ptr a:((ADD|AND|OR|XOR|BTC|BTR|BTS)Qconst [c] l:(MOVQload [off] {sym} ptr2 mem)) mem) - && isSamePtr(ptr, ptr2) && a.Uses == 1 && l.Uses == 1 && validValAndOff(c,off) && clobber(l, a) -> - ((ADD|AND|OR|XOR|BTC|BTR|BTS)Qconstmodify {sym} [makeValAndOff(c,off)] ptr mem) + && isSamePtr(ptr, ptr2) && a.Uses == 1 && l.Uses == 1 && validValAndOff(int64(c),int64(off)) && clobber(l, a) => + ((ADD|AND|OR|XOR|BTC|BTR|BTS)Qconstmodify {sym} [makeValAndOff32(int32(c),off)] ptr mem) (MOVLstore [off] {sym} ptr a:((ADD|AND|OR|XOR|BTC|BTR|BTS)Lconst [c] l:(MOVLload [off] {sym} ptr2 mem)) mem) - && isSamePtr(ptr, ptr2) && a.Uses == 1 && l.Uses == 1 && validValAndOff(c,off) && clobber(l, a) -> - ((ADD|AND|OR|XOR|BTC|BTR|BTS)Lconstmodify {sym} [makeValAndOff(c,off)] ptr mem) + && isSamePtr(ptr, ptr2) && a.Uses == 1 && l.Uses == 1 && validValAndOff(int64(c),int64(off)) && clobber(l, a) => + ((ADD|AND|OR|XOR|BTC|BTR|BTS)Lconstmodify {sym} [makeValAndOff32(int32(c),off)] ptr mem) // float <-> int register moves, with no conversion. // These come up when compiling math.{Float{32,64}bits,Float{32,64}frombits}. -(MOVQload [off] {sym} ptr (MOVSDstore [off] {sym} ptr val _)) -> (MOVQf2i val) -(MOVLload [off] {sym} ptr (MOVSSstore [off] {sym} ptr val _)) -> (MOVLf2i val) -(MOVSDload [off] {sym} ptr (MOVQstore [off] {sym} ptr val _)) -> (MOVQi2f val) -(MOVSSload [off] {sym} ptr (MOVLstore [off] {sym} ptr val _)) -> (MOVLi2f val) +(MOVQload [off] {sym} ptr (MOVSDstore [off] {sym} ptr val _)) => (MOVQf2i val) +(MOVLload [off] {sym} ptr (MOVSSstore [off] {sym} ptr val _)) => (MOVLf2i val) +(MOVSDload [off] {sym} ptr (MOVQstore [off] {sym} ptr val _)) => (MOVQi2f val) +(MOVSSload [off] {sym} ptr (MOVLstore [off] {sym} ptr val _)) => (MOVLi2f val) // Other load-like ops. -(ADDQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> (ADDQ x (MOVQf2i y)) -(ADDLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> (ADDL x (MOVLf2i y)) -(SUBQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> (SUBQ x (MOVQf2i y)) -(SUBLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> (SUBL x (MOVLf2i y)) -(ANDQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> (ANDQ x (MOVQf2i y)) -(ANDLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> (ANDL x (MOVLf2i y)) -( ORQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> ( ORQ x (MOVQf2i y)) -( ORLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> ( ORL x (MOVLf2i y)) -(XORQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) -> (XORQ x (MOVQf2i y)) -(XORLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) -> (XORL x (MOVLf2i y)) +(ADDQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) => (ADDQ x (MOVQf2i y)) +(ADDLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) => (ADDL x (MOVLf2i y)) +(SUBQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) => (SUBQ x (MOVQf2i y)) +(SUBLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) => (SUBL x (MOVLf2i y)) +(ANDQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) => (ANDQ x (MOVQf2i y)) +(ANDLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) => (ANDL x (MOVLf2i y)) +( ORQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) => ( ORQ x (MOVQf2i y)) +( ORLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) => ( ORL x (MOVLf2i y)) +(XORQload x [off] {sym} ptr (MOVSDstore [off] {sym} ptr y _)) => (XORQ x (MOVQf2i y)) +(XORLload x [off] {sym} ptr (MOVSSstore [off] {sym} ptr y _)) => (XORL x (MOVLf2i y)) -(ADDSDload x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) -> (ADDSD x (MOVQi2f y)) -(ADDSSload x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) -> (ADDSS x (MOVLi2f y)) -(SUBSDload x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) -> (SUBSD x (MOVQi2f y)) -(SUBSSload x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) -> (SUBSS x (MOVLi2f y)) -(MULSDload x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) -> (MULSD x (MOVQi2f y)) -(MULSSload x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) -> (MULSS x (MOVLi2f y)) +(ADDSDload x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) => (ADDSD x (MOVQi2f y)) +(ADDSSload x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) => (ADDSS x (MOVLi2f y)) +(SUBSDload x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) => (SUBSD x (MOVQi2f y)) +(SUBSSload x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) => (SUBSS x (MOVLi2f y)) +(MULSDload x [off] {sym} ptr (MOVQstore [off] {sym} ptr y _)) => (MULSD x (MOVQi2f y)) +(MULSSload x [off] {sym} ptr (MOVLstore [off] {sym} ptr y _)) => (MULSS x (MOVLi2f y)) // Redirect stores to use the other register set. -(MOVQstore [off] {sym} ptr (MOVQf2i val) mem) -> (MOVSDstore [off] {sym} ptr val mem) -(MOVLstore [off] {sym} ptr (MOVLf2i val) mem) -> (MOVSSstore [off] {sym} ptr val mem) -(MOVSDstore [off] {sym} ptr (MOVQi2f val) mem) -> (MOVQstore [off] {sym} ptr val mem) -(MOVSSstore [off] {sym} ptr (MOVLi2f val) mem) -> (MOVLstore [off] {sym} ptr val mem) +(MOVQstore [off] {sym} ptr (MOVQf2i val) mem) => (MOVSDstore [off] {sym} ptr val mem) +(MOVLstore [off] {sym} ptr (MOVLf2i val) mem) => (MOVSSstore [off] {sym} ptr val mem) +(MOVSDstore [off] {sym} ptr (MOVQi2f val) mem) => (MOVQstore [off] {sym} ptr val mem) +(MOVSSstore [off] {sym} ptr (MOVLi2f val) mem) => (MOVLstore [off] {sym} ptr val mem) // Load args directly into the register class where it will be used. // We do this by just modifying the type of the Arg. -(MOVQf2i (Arg [off] {sym})) && t.Size() == u.Size() -> @b.Func.Entry (Arg [off] {sym}) -(MOVLf2i (Arg [off] {sym})) && t.Size() == u.Size() -> @b.Func.Entry (Arg [off] {sym}) -(MOVQi2f (Arg [off] {sym})) && t.Size() == u.Size() -> @b.Func.Entry (Arg [off] {sym}) -(MOVLi2f (Arg [off] {sym})) && t.Size() == u.Size() -> @b.Func.Entry (Arg [off] {sym}) +(MOVQf2i (Arg [off] {sym})) && t.Size() == u.Size() => @b.Func.Entry (Arg [off] {sym}) +(MOVLf2i (Arg [off] {sym})) && t.Size() == u.Size() => @b.Func.Entry (Arg [off] {sym}) +(MOVQi2f (Arg [off] {sym})) && t.Size() == u.Size() => @b.Func.Entry (Arg [off] {sym}) +(MOVLi2f (Arg [off] {sym})) && t.Size() == u.Size() => @b.Func.Entry (Arg [off] {sym}) // LEAQ is rematerializeable, so this helps to avoid register spill. // See issue 22947 for details -(ADD(Q|L)const [off] x:(SP)) -> (LEA(Q|L) [off] x) +(ADD(Q|L)const [off] x:(SP)) => (LEA(Q|L) [off] x) // HMULx is commutative, but its first argument must go in AX. // If possible, put a rematerializeable value in the first argument slot, // to reduce the odds that another value will be have to spilled // specifically to free up AX. -(HMUL(Q|L) x y) && !x.rematerializeable() && y.rematerializeable() -> (HMUL(Q|L) y x) -(HMUL(Q|L)U x y) && !x.rematerializeable() && y.rematerializeable() -> (HMUL(Q|L)U y x) +(HMUL(Q|L) x y) && !x.rematerializeable() && y.rematerializeable() => (HMUL(Q|L) y x) +(HMUL(Q|L)U x y) && !x.rematerializeable() && y.rematerializeable() => (HMUL(Q|L)U y x) // Fold loads into compares // Note: these may be undone by the flagalloc pass. -(CMP(Q|L|W|B) l:(MOV(Q|L|W|B)load {sym} [off] ptr mem) x) && canMergeLoad(v, l) && clobber(l) -> (CMP(Q|L|W|B)load {sym} [off] ptr x mem) -(CMP(Q|L|W|B) x l:(MOV(Q|L|W|B)load {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) -> (InvertFlags (CMP(Q|L|W|B)load {sym} [off] ptr x mem)) +(CMP(Q|L|W|B) l:(MOV(Q|L|W|B)load {sym} [off] ptr mem) x) && canMergeLoad(v, l) && clobber(l) => (CMP(Q|L|W|B)load {sym} [off] ptr x mem) +(CMP(Q|L|W|B) x l:(MOV(Q|L|W|B)load {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (InvertFlags (CMP(Q|L|W|B)load {sym} [off] ptr x mem)) (CMP(Q|L)const l:(MOV(Q|L)load {sym} [off] ptr mem) [c]) && l.Uses == 1 @@ -2175,22 +2179,22 @@ && clobber(l) => @l.Block (CMP(W|B)constload {sym} [makeValAndOff32(int32(c),off)] ptr mem) -(CMPQload {sym} [off] ptr (MOVQconst [c]) mem) && validValAndOff(c,off) -> (CMPQconstload {sym} [makeValAndOff(c,off)] ptr mem) -(CMPLload {sym} [off] ptr (MOVLconst [c]) mem) && validValAndOff(c,off) -> (CMPLconstload {sym} [makeValAndOff(c,off)] ptr mem) -(CMPWload {sym} [off] ptr (MOVLconst [c]) mem) && validValAndOff(int64(int16(c)),off) -> (CMPWconstload {sym} [makeValAndOff(int64(int16(c)),off)] ptr mem) -(CMPBload {sym} [off] ptr (MOVLconst [c]) mem) && validValAndOff(int64(int8(c)),off) -> (CMPBconstload {sym} [makeValAndOff(int64(int8(c)),off)] ptr mem) +(CMPQload {sym} [off] ptr (MOVQconst [c]) mem) && validValAndOff(c,int64(off)) => (CMPQconstload {sym} [makeValAndOff64(c,int64(off))] ptr mem) +(CMPLload {sym} [off] ptr (MOVLconst [c]) mem) && validValAndOff(int64(c),int64(off)) => (CMPLconstload {sym} [makeValAndOff32(c,off)] ptr mem) +(CMPWload {sym} [off] ptr (MOVLconst [c]) mem) && validValAndOff(int64(int16(c)),int64(off)) => (CMPWconstload {sym} [makeValAndOff32(int32(int16(c)),off)] ptr mem) +(CMPBload {sym} [off] ptr (MOVLconst [c]) mem) && validValAndOff(int64(int8(c)),int64(off)) => (CMPBconstload {sym} [makeValAndOff32(int32(int8(c)),off)] ptr mem) (TEST(Q|L|W|B) l:(MOV(Q|L|W|B)load {sym} [off] ptr mem) l2) && l == l2 && l.Uses == 2 - && validValAndOff(0,off) - && clobber(l) -> - @l.Block (CMP(Q|L|W|B)constload {sym} [makeValAndOff(0,off)] ptr mem) + && validValAndOff(0, int64(off)) + && clobber(l) => + @l.Block (CMP(Q|L|W|B)constload {sym} [makeValAndOff64(0, int64(off))] ptr mem) -(MOVBload [off] {sym} (SB) _) && symIsRO(sym) -> (MOVLconst [int64(read8(sym, off))]) -(MOVWload [off] {sym} (SB) _) && symIsRO(sym) -> (MOVLconst [int64(read16(sym, off, config.ctxt.Arch.ByteOrder))]) -(MOVLload [off] {sym} (SB) _) && symIsRO(sym) -> (MOVQconst [int64(read32(sym, off, config.ctxt.Arch.ByteOrder))]) -(MOVQload [off] {sym} (SB) _) && symIsRO(sym) -> (MOVQconst [int64(read64(sym, off, config.ctxt.Arch.ByteOrder))]) -(MOVOstore [dstOff] {dstSym} ptr (MOVOload [srcOff] {srcSym} (SB) _) mem) && symIsRO(srcSym) -> - (MOVQstore [dstOff+8] {dstSym} ptr (MOVQconst [int64(read64(srcSym, srcOff+8, config.ctxt.Arch.ByteOrder))]) - (MOVQstore [dstOff] {dstSym} ptr (MOVQconst [int64(read64(srcSym, srcOff, config.ctxt.Arch.ByteOrder))]) mem)) +(MOVBload [off] {sym} (SB) _) && symIsRO(sym) => (MOVLconst [int32(read8(sym, int64(off)))]) +(MOVWload [off] {sym} (SB) _) && symIsRO(sym) => (MOVLconst [int32(read16(sym, int64(off), config.ctxt.Arch.ByteOrder))]) +(MOVLload [off] {sym} (SB) _) && symIsRO(sym) => (MOVQconst [int64(read32(sym, int64(off), config.ctxt.Arch.ByteOrder))]) +(MOVQload [off] {sym} (SB) _) && symIsRO(sym) => (MOVQconst [int64(read64(sym, int64(off), config.ctxt.Arch.ByteOrder))]) +(MOVOstore [dstOff] {dstSym} ptr (MOVOload [srcOff] {srcSym} (SB) _) mem) && symIsRO(srcSym) => + (MOVQstore [dstOff+8] {dstSym} ptr (MOVQconst [int64(read64(srcSym, int64(srcOff)+8, config.ctxt.Arch.ByteOrder))]) + (MOVQstore [dstOff] {dstSym} ptr (MOVQconst [int64(read64(srcSym, int64(srcOff), config.ctxt.Arch.ByteOrder))]) mem)) diff --git a/src/cmd/compile/internal/ssa/gen/AMD64Ops.go b/src/cmd/compile/internal/ssa/gen/AMD64Ops.go index e6d66957ddd..de5372670b3 100644 --- a/src/cmd/compile/internal/ssa/gen/AMD64Ops.go +++ b/src/cmd/compile/internal/ssa/gen/AMD64Ops.go @@ -767,9 +767,9 @@ func init() { faultOnNilArg0: true, }, - {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", clobberFlags: true, call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("DX"), 0}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("DX"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem // arg0 = destination pointer // arg1 = source pointer @@ -902,7 +902,9 @@ func init() { // Atomic memory updates. {name: "ANDBlock", argLength: 3, reg: gpstore, asm: "ANDB", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"}, // *(arg0+auxint+aux) &= arg1 + {name: "ANDLlock", argLength: 3, reg: gpstore, asm: "ANDL", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"}, // *(arg0+auxint+aux) &= arg1 {name: "ORBlock", argLength: 3, reg: gpstore, asm: "ORB", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"}, // *(arg0+auxint+aux) |= arg1 + {name: "ORLlock", argLength: 3, reg: gpstore, asm: "ORL", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"}, // *(arg0+auxint+aux) |= arg1 } var AMD64blocks = []blockData{ diff --git a/src/cmd/compile/internal/ssa/gen/ARM.rules b/src/cmd/compile/internal/ssa/gen/ARM.rules index 983f8848497..11c36b5da35 100644 --- a/src/cmd/compile/internal/ssa/gen/ARM.rules +++ b/src/cmd/compile/internal/ssa/gen/ARM.rules @@ -2,83 +2,83 @@ // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. -(Add(Ptr|32|16|8) ...) -> (ADD ...) -(Add(32|64)F ...) -> (ADD(F|D) ...) -(Add32carry ...) -> (ADDS ...) -(Add32withcarry ...) -> (ADC ...) +(Add(Ptr|32|16|8) ...) => (ADD ...) +(Add(32|64)F ...) => (ADD(F|D) ...) +(Add32carry ...) => (ADDS ...) +(Add32withcarry ...) => (ADC ...) -(Sub(Ptr|32|16|8) ...) -> (SUB ...) -(Sub(32|64)F ...) -> (SUB(F|D) ...) -(Sub32carry ...) -> (SUBS ...) -(Sub32withcarry ...) -> (SBC ...) +(Sub(Ptr|32|16|8) ...) => (SUB ...) +(Sub(32|64)F ...) => (SUB(F|D) ...) +(Sub32carry ...) => (SUBS ...) +(Sub32withcarry ...) => (SBC ...) -(Mul(32|16|8) ...) -> (MUL ...) -(Mul(32|64)F ...) -> (MUL(F|D) ...) -(Hmul(32|32u) ...) -> (HMU(L|LU) ...) -(Mul32uhilo ...) -> (MULLU ...) +(Mul(32|16|8) ...) => (MUL ...) +(Mul(32|64)F ...) => (MUL(F|D) ...) +(Hmul(32|32u) ...) => (HMU(L|LU) ...) +(Mul32uhilo ...) => (MULLU ...) -(Div32 x y) -> +(Div32 x y) => (SUB (XOR // negate the result if one operand is negative (Select0 (CALLudiv (SUB (XOR x (Signmask x)) (Signmask x)) // negate x if negative (SUB (XOR y (Signmask y)) (Signmask y)))) // negate y if negative (Signmask (XOR x y))) (Signmask (XOR x y))) -(Div32u x y) -> (Select0 (CALLudiv x y)) -(Div16 x y) -> (Div32 (SignExt16to32 x) (SignExt16to32 y)) -(Div16u x y) -> (Div32u (ZeroExt16to32 x) (ZeroExt16to32 y)) -(Div8 x y) -> (Div32 (SignExt8to32 x) (SignExt8to32 y)) -(Div8u x y) -> (Div32u (ZeroExt8to32 x) (ZeroExt8to32 y)) -(Div(32|64)F ...) -> (DIV(F|D) ...) +(Div32u x y) => (Select0 (CALLudiv x y)) +(Div16 x y) => (Div32 (SignExt16to32 x) (SignExt16to32 y)) +(Div16u x y) => (Div32u (ZeroExt16to32 x) (ZeroExt16to32 y)) +(Div8 x y) => (Div32 (SignExt8to32 x) (SignExt8to32 y)) +(Div8u x y) => (Div32u (ZeroExt8to32 x) (ZeroExt8to32 y)) +(Div(32|64)F ...) => (DIV(F|D) ...) -(Mod32 x y) -> +(Mod32 x y) => (SUB (XOR // negate the result if x is negative (Select1 (CALLudiv (SUB (XOR x (Signmask x)) (Signmask x)) // negate x if negative (SUB (XOR y (Signmask y)) (Signmask y)))) // negate y if negative (Signmask x)) (Signmask x)) -(Mod32u x y) -> (Select1 (CALLudiv x y)) -(Mod16 x y) -> (Mod32 (SignExt16to32 x) (SignExt16to32 y)) -(Mod16u x y) -> (Mod32u (ZeroExt16to32 x) (ZeroExt16to32 y)) -(Mod8 x y) -> (Mod32 (SignExt8to32 x) (SignExt8to32 y)) -(Mod8u x y) -> (Mod32u (ZeroExt8to32 x) (ZeroExt8to32 y)) +(Mod32u x y) => (Select1 (CALLudiv x y)) +(Mod16 x y) => (Mod32 (SignExt16to32 x) (SignExt16to32 y)) +(Mod16u x y) => (Mod32u (ZeroExt16to32 x) (ZeroExt16to32 y)) +(Mod8 x y) => (Mod32 (SignExt8to32 x) (SignExt8to32 y)) +(Mod8u x y) => (Mod32u (ZeroExt8to32 x) (ZeroExt8to32 y)) // (x + y) / 2 with x>=y -> (x - y) / 2 + y -(Avg32u x y) -> (ADD (SRLconst (SUB x y) [1]) y) +(Avg32u x y) => (ADD (SRLconst (SUB x y) [1]) y) -(And(32|16|8) ...) -> (AND ...) -(Or(32|16|8) ...) -> (OR ...) -(Xor(32|16|8) ...) -> (XOR ...) +(And(32|16|8) ...) => (AND ...) +(Or(32|16|8) ...) => (OR ...) +(Xor(32|16|8) ...) => (XOR ...) // unary ops -(Neg(32|16|8) x) -> (RSBconst [0] x) -(Neg(32|64)F ...) -> (NEG(F|D) ...) +(Neg(32|16|8) x) => (RSBconst [0] x) +(Neg(32|64)F ...) => (NEG(F|D) ...) -(Com(32|16|8) ...) -> (MVN ...) +(Com(32|16|8) ...) => (MVN ...) -(Sqrt ...) -> (SQRTD ...) -(Abs ...) -> (ABSD ...) +(Sqrt ...) => (SQRTD ...) +(Abs ...) => (ABSD ...) // TODO: optimize this for ARMv5 and ARMv6 -(Ctz32NonZero ...) -> (Ctz32 ...) -(Ctz16NonZero ...) -> (Ctz32 ...) -(Ctz8NonZero ...) -> (Ctz32 ...) +(Ctz32NonZero ...) => (Ctz32 ...) +(Ctz16NonZero ...) => (Ctz32 ...) +(Ctz8NonZero ...) => (Ctz32 ...) // count trailing zero for ARMv5 and ARMv6 // 32 - CLZ(x&-x - 1) -(Ctz32 x) && objabi.GOARM<=6 -> +(Ctz32 x) && objabi.GOARM<=6 => (RSBconst [32] (CLZ (SUBconst (AND x (RSBconst [0] x)) [1]))) -(Ctz16 x) && objabi.GOARM<=6 -> +(Ctz16 x) && objabi.GOARM<=6 => (RSBconst [32] (CLZ (SUBconst (AND (ORconst [0x10000] x) (RSBconst [0] (ORconst [0x10000] x))) [1]))) -(Ctz8 x) && objabi.GOARM<=6 -> +(Ctz8 x) && objabi.GOARM<=6 => (RSBconst [32] (CLZ (SUBconst (AND (ORconst [0x100] x) (RSBconst [0] (ORconst [0x100] x))) [1]))) // count trailing zero for ARMv7 -(Ctz32 x) && objabi.GOARM==7 -> (CLZ (RBIT x)) -(Ctz16 x) && objabi.GOARM==7 -> (CLZ (RBIT (ORconst [0x10000] x))) -(Ctz8 x) && objabi.GOARM==7 -> (CLZ (RBIT (ORconst [0x100] x))) +(Ctz32 x) && objabi.GOARM==7 => (CLZ (RBIT x)) +(Ctz16 x) && objabi.GOARM==7 => (CLZ (RBIT (ORconst [0x10000] x))) +(Ctz8 x) && objabi.GOARM==7 => (CLZ (RBIT (ORconst [0x100] x))) // bit length -(BitLen32 x) -> (RSBconst [32] (CLZ x)) +(BitLen32 x) => (RSBconst [32] (CLZ x)) // byte swap for ARMv5 // let (a, b, c, d) be the bytes of x from high to low @@ -89,203 +89,203 @@ // t5 = x right rotate 8 bits -- (d, a, b, c ) // result = t4 ^ t5 -- (d, c, b, a ) // using shifted ops this can be done in 4 instructions. -(Bswap32 x) && objabi.GOARM==5 -> +(Bswap32 x) && objabi.GOARM==5 => (XOR (SRLconst (BICconst (XOR x (SRRconst [16] x)) [0xff0000]) [8]) (SRRconst x [8])) // byte swap for ARMv6 and above -(Bswap32 x) && objabi.GOARM>=6 -> (REV x) +(Bswap32 x) && objabi.GOARM>=6 => (REV x) // boolean ops -- booleans are represented with 0=false, 1=true -(AndB ...) -> (AND ...) -(OrB ...) -> (OR ...) -(EqB x y) -> (XORconst [1] (XOR x y)) -(NeqB ...) -> (XOR ...) -(Not x) -> (XORconst [1] x) +(AndB ...) => (AND ...) +(OrB ...) => (OR ...) +(EqB x y) => (XORconst [1] (XOR x y)) +(NeqB ...) => (XOR ...) +(Not x) => (XORconst [1] x) // shifts // hardware instruction uses only the low byte of the shift // we compare to 256 to ensure Go semantics for large shifts -(Lsh32x32 x y) -> (CMOVWHSconst (SLL x y) (CMPconst [256] y) [0]) -(Lsh32x16 x y) -> (CMOVWHSconst (SLL x (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) -(Lsh32x8 x y) -> (SLL x (ZeroExt8to32 y)) +(Lsh32x32 x y) => (CMOVWHSconst (SLL x y) (CMPconst [256] y) [0]) +(Lsh32x16 x y) => (CMOVWHSconst (SLL x (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) +(Lsh32x8 x y) => (SLL x (ZeroExt8to32 y)) -(Lsh16x32 x y) -> (CMOVWHSconst (SLL x y) (CMPconst [256] y) [0]) -(Lsh16x16 x y) -> (CMOVWHSconst (SLL x (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) -(Lsh16x8 x y) -> (SLL x (ZeroExt8to32 y)) +(Lsh16x32 x y) => (CMOVWHSconst (SLL x y) (CMPconst [256] y) [0]) +(Lsh16x16 x y) => (CMOVWHSconst (SLL x (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) +(Lsh16x8 x y) => (SLL x (ZeroExt8to32 y)) -(Lsh8x32 x y) -> (CMOVWHSconst (SLL x y) (CMPconst [256] y) [0]) -(Lsh8x16 x y) -> (CMOVWHSconst (SLL x (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) -(Lsh8x8 x y) -> (SLL x (ZeroExt8to32 y)) +(Lsh8x32 x y) => (CMOVWHSconst (SLL x y) (CMPconst [256] y) [0]) +(Lsh8x16 x y) => (CMOVWHSconst (SLL x (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) +(Lsh8x8 x y) => (SLL x (ZeroExt8to32 y)) -(Rsh32Ux32 x y) -> (CMOVWHSconst (SRL x y) (CMPconst [256] y) [0]) -(Rsh32Ux16 x y) -> (CMOVWHSconst (SRL x (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) -(Rsh32Ux8 x y) -> (SRL x (ZeroExt8to32 y)) +(Rsh32Ux32 x y) => (CMOVWHSconst (SRL x y) (CMPconst [256] y) [0]) +(Rsh32Ux16 x y) => (CMOVWHSconst (SRL x (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) +(Rsh32Ux8 x y) => (SRL x (ZeroExt8to32 y)) -(Rsh16Ux32 x y) -> (CMOVWHSconst (SRL (ZeroExt16to32 x) y) (CMPconst [256] y) [0]) -(Rsh16Ux16 x y) -> (CMOVWHSconst (SRL (ZeroExt16to32 x) (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) -(Rsh16Ux8 x y) -> (SRL (ZeroExt16to32 x) (ZeroExt8to32 y)) +(Rsh16Ux32 x y) => (CMOVWHSconst (SRL (ZeroExt16to32 x) y) (CMPconst [256] y) [0]) +(Rsh16Ux16 x y) => (CMOVWHSconst (SRL (ZeroExt16to32 x) (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) +(Rsh16Ux8 x y) => (SRL (ZeroExt16to32 x) (ZeroExt8to32 y)) -(Rsh8Ux32 x y) -> (CMOVWHSconst (SRL (ZeroExt8to32 x) y) (CMPconst [256] y) [0]) -(Rsh8Ux16 x y) -> (CMOVWHSconst (SRL (ZeroExt8to32 x) (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) -(Rsh8Ux8 x y) -> (SRL (ZeroExt8to32 x) (ZeroExt8to32 y)) +(Rsh8Ux32 x y) => (CMOVWHSconst (SRL (ZeroExt8to32 x) y) (CMPconst [256] y) [0]) +(Rsh8Ux16 x y) => (CMOVWHSconst (SRL (ZeroExt8to32 x) (ZeroExt16to32 y)) (CMPconst [256] (ZeroExt16to32 y)) [0]) +(Rsh8Ux8 x y) => (SRL (ZeroExt8to32 x) (ZeroExt8to32 y)) -(Rsh32x32 x y) -> (SRAcond x y (CMPconst [256] y)) -(Rsh32x16 x y) -> (SRAcond x (ZeroExt16to32 y) (CMPconst [256] (ZeroExt16to32 y))) -(Rsh32x8 x y) -> (SRA x (ZeroExt8to32 y)) +(Rsh32x32 x y) => (SRAcond x y (CMPconst [256] y)) +(Rsh32x16 x y) => (SRAcond x (ZeroExt16to32 y) (CMPconst [256] (ZeroExt16to32 y))) +(Rsh32x8 x y) => (SRA x (ZeroExt8to32 y)) -(Rsh16x32 x y) -> (SRAcond (SignExt16to32 x) y (CMPconst [256] y)) -(Rsh16x16 x y) -> (SRAcond (SignExt16to32 x) (ZeroExt16to32 y) (CMPconst [256] (ZeroExt16to32 y))) -(Rsh16x8 x y) -> (SRA (SignExt16to32 x) (ZeroExt8to32 y)) +(Rsh16x32 x y) => (SRAcond (SignExt16to32 x) y (CMPconst [256] y)) +(Rsh16x16 x y) => (SRAcond (SignExt16to32 x) (ZeroExt16to32 y) (CMPconst [256] (ZeroExt16to32 y))) +(Rsh16x8 x y) => (SRA (SignExt16to32 x) (ZeroExt8to32 y)) -(Rsh8x32 x y) -> (SRAcond (SignExt8to32 x) y (CMPconst [256] y)) -(Rsh8x16 x y) -> (SRAcond (SignExt8to32 x) (ZeroExt16to32 y) (CMPconst [256] (ZeroExt16to32 y))) -(Rsh8x8 x y) -> (SRA (SignExt8to32 x) (ZeroExt8to32 y)) +(Rsh8x32 x y) => (SRAcond (SignExt8to32 x) y (CMPconst [256] y)) +(Rsh8x16 x y) => (SRAcond (SignExt8to32 x) (ZeroExt16to32 y) (CMPconst [256] (ZeroExt16to32 y))) +(Rsh8x8 x y) => (SRA (SignExt8to32 x) (ZeroExt8to32 y)) // constant shifts // generic opt rewrites all constant shifts to shift by Const64 -(Lsh32x64 x (Const64 [c])) && uint64(c) < 32 -> (SLLconst x [c]) -(Rsh32x64 x (Const64 [c])) && uint64(c) < 32 -> (SRAconst x [c]) -(Rsh32Ux64 x (Const64 [c])) && uint64(c) < 32 -> (SRLconst x [c]) -(Lsh16x64 x (Const64 [c])) && uint64(c) < 16 -> (SLLconst x [c]) -(Rsh16x64 x (Const64 [c])) && uint64(c) < 16 -> (SRAconst (SLLconst x [16]) [c+16]) -(Rsh16Ux64 x (Const64 [c])) && uint64(c) < 16 -> (SRLconst (SLLconst x [16]) [c+16]) -(Lsh8x64 x (Const64 [c])) && uint64(c) < 8 -> (SLLconst x [c]) -(Rsh8x64 x (Const64 [c])) && uint64(c) < 8 -> (SRAconst (SLLconst x [24]) [c+24]) -(Rsh8Ux64 x (Const64 [c])) && uint64(c) < 8 -> (SRLconst (SLLconst x [24]) [c+24]) +(Lsh32x64 x (Const64 [c])) && uint64(c) < 32 => (SLLconst x [int32(c)]) +(Rsh32x64 x (Const64 [c])) && uint64(c) < 32 => (SRAconst x [int32(c)]) +(Rsh32Ux64 x (Const64 [c])) && uint64(c) < 32 => (SRLconst x [int32(c)]) +(Lsh16x64 x (Const64 [c])) && uint64(c) < 16 => (SLLconst x [int32(c)]) +(Rsh16x64 x (Const64 [c])) && uint64(c) < 16 => (SRAconst (SLLconst x [16]) [int32(c+16)]) +(Rsh16Ux64 x (Const64 [c])) && uint64(c) < 16 => (SRLconst (SLLconst x [16]) [int32(c+16)]) +(Lsh8x64 x (Const64 [c])) && uint64(c) < 8 => (SLLconst x [int32(c)]) +(Rsh8x64 x (Const64 [c])) && uint64(c) < 8 => (SRAconst (SLLconst x [24]) [int32(c+24)]) +(Rsh8Ux64 x (Const64 [c])) && uint64(c) < 8 => (SRLconst (SLLconst x [24]) [int32(c+24)]) // large constant shifts -(Lsh32x64 _ (Const64 [c])) && uint64(c) >= 32 -> (Const32 [0]) -(Rsh32Ux64 _ (Const64 [c])) && uint64(c) >= 32 -> (Const32 [0]) -(Lsh16x64 _ (Const64 [c])) && uint64(c) >= 16 -> (Const16 [0]) -(Rsh16Ux64 _ (Const64 [c])) && uint64(c) >= 16 -> (Const16 [0]) -(Lsh8x64 _ (Const64 [c])) && uint64(c) >= 8 -> (Const8 [0]) -(Rsh8Ux64 _ (Const64 [c])) && uint64(c) >= 8 -> (Const8 [0]) +(Lsh32x64 _ (Const64 [c])) && uint64(c) >= 32 => (Const32 [0]) +(Rsh32Ux64 _ (Const64 [c])) && uint64(c) >= 32 => (Const32 [0]) +(Lsh16x64 _ (Const64 [c])) && uint64(c) >= 16 => (Const16 [0]) +(Rsh16Ux64 _ (Const64 [c])) && uint64(c) >= 16 => (Const16 [0]) +(Lsh8x64 _ (Const64 [c])) && uint64(c) >= 8 => (Const8 [0]) +(Rsh8Ux64 _ (Const64 [c])) && uint64(c) >= 8 => (Const8 [0]) // large constant signed right shift, we leave the sign bit -(Rsh32x64 x (Const64 [c])) && uint64(c) >= 32 -> (SRAconst x [31]) -(Rsh16x64 x (Const64 [c])) && uint64(c) >= 16 -> (SRAconst (SLLconst x [16]) [31]) -(Rsh8x64 x (Const64 [c])) && uint64(c) >= 8 -> (SRAconst (SLLconst x [24]) [31]) +(Rsh32x64 x (Const64 [c])) && uint64(c) >= 32 => (SRAconst x [31]) +(Rsh16x64 x (Const64 [c])) && uint64(c) >= 16 => (SRAconst (SLLconst x [16]) [31]) +(Rsh8x64 x (Const64 [c])) && uint64(c) >= 8 => (SRAconst (SLLconst x [24]) [31]) // constants -(Const(8|16|32) ...) -> (MOVWconst ...) -(Const(32F|64F) ...) -> (MOV(F|D)const ...) -(ConstNil) -> (MOVWconst [0]) -(ConstBool ...) -> (MOVWconst ...) +(Const(8|16|32) [val]) => (MOVWconst [int32(val)]) +(Const(32|64)F [val]) => (MOV(F|D)const [float64(val)]) +(ConstNil) => (MOVWconst [0]) +(ConstBool [b]) => (MOVWconst [b2i32(b)]) // truncations // Because we ignore high parts of registers, truncates are just copies. -(Trunc16to8 ...) -> (Copy ...) -(Trunc32to8 ...) -> (Copy ...) -(Trunc32to16 ...) -> (Copy ...) +(Trunc16to8 ...) => (Copy ...) +(Trunc32to8 ...) => (Copy ...) +(Trunc32to16 ...) => (Copy ...) // Zero-/Sign-extensions -(ZeroExt8to16 ...) -> (MOVBUreg ...) -(ZeroExt8to32 ...) -> (MOVBUreg ...) -(ZeroExt16to32 ...) -> (MOVHUreg ...) +(ZeroExt8to16 ...) => (MOVBUreg ...) +(ZeroExt8to32 ...) => (MOVBUreg ...) +(ZeroExt16to32 ...) => (MOVHUreg ...) -(SignExt8to16 ...) -> (MOVBreg ...) -(SignExt8to32 ...) -> (MOVBreg ...) -(SignExt16to32 ...) -> (MOVHreg ...) +(SignExt8to16 ...) => (MOVBreg ...) +(SignExt8to32 ...) => (MOVBreg ...) +(SignExt16to32 ...) => (MOVHreg ...) -(Signmask x) -> (SRAconst x [31]) -(Zeromask x) -> (SRAconst (RSBshiftRL x x [1]) [31]) // sign bit of uint32(x)>>1 - x -(Slicemask x) -> (SRAconst (RSBconst [0] x) [31]) +(Signmask x) => (SRAconst x [31]) +(Zeromask x) => (SRAconst (RSBshiftRL x x [1]) [31]) // sign bit of uint32(x)>>1 - x +(Slicemask x) => (SRAconst (RSBconst [0] x) [31]) // float <-> int conversion -(Cvt32to32F ...) -> (MOVWF ...) -(Cvt32to64F ...) -> (MOVWD ...) -(Cvt32Uto32F ...) -> (MOVWUF ...) -(Cvt32Uto64F ...) -> (MOVWUD ...) -(Cvt32Fto32 ...) -> (MOVFW ...) -(Cvt64Fto32 ...) -> (MOVDW ...) -(Cvt32Fto32U ...) -> (MOVFWU ...) -(Cvt64Fto32U ...) -> (MOVDWU ...) -(Cvt32Fto64F ...) -> (MOVFD ...) -(Cvt64Fto32F ...) -> (MOVDF ...) +(Cvt32to32F ...) => (MOVWF ...) +(Cvt32to64F ...) => (MOVWD ...) +(Cvt32Uto32F ...) => (MOVWUF ...) +(Cvt32Uto64F ...) => (MOVWUD ...) +(Cvt32Fto32 ...) => (MOVFW ...) +(Cvt64Fto32 ...) => (MOVDW ...) +(Cvt32Fto32U ...) => (MOVFWU ...) +(Cvt64Fto32U ...) => (MOVDWU ...) +(Cvt32Fto64F ...) => (MOVFD ...) +(Cvt64Fto32F ...) => (MOVDF ...) -(Round(32|64)F ...) -> (Copy ...) +(Round(32|64)F ...) => (Copy ...) -(CvtBoolToUint8 ...) -> (Copy ...) +(CvtBoolToUint8 ...) => (Copy ...) // fused-multiply-add -(FMA x y z) -> (FMULAD z x y) +(FMA x y z) => (FMULAD z x y) // comparisons -(Eq8 x y) -> (Equal (CMP (ZeroExt8to32 x) (ZeroExt8to32 y))) -(Eq16 x y) -> (Equal (CMP (ZeroExt16to32 x) (ZeroExt16to32 y))) -(Eq32 x y) -> (Equal (CMP x y)) -(EqPtr x y) -> (Equal (CMP x y)) -(Eq(32|64)F x y) -> (Equal (CMP(F|D) x y)) +(Eq8 x y) => (Equal (CMP (ZeroExt8to32 x) (ZeroExt8to32 y))) +(Eq16 x y) => (Equal (CMP (ZeroExt16to32 x) (ZeroExt16to32 y))) +(Eq32 x y) => (Equal (CMP x y)) +(EqPtr x y) => (Equal (CMP x y)) +(Eq(32|64)F x y) => (Equal (CMP(F|D) x y)) -(Neq8 x y) -> (NotEqual (CMP (ZeroExt8to32 x) (ZeroExt8to32 y))) -(Neq16 x y) -> (NotEqual (CMP (ZeroExt16to32 x) (ZeroExt16to32 y))) -(Neq32 x y) -> (NotEqual (CMP x y)) -(NeqPtr x y) -> (NotEqual (CMP x y)) -(Neq(32|64)F x y) -> (NotEqual (CMP(F|D) x y)) +(Neq8 x y) => (NotEqual (CMP (ZeroExt8to32 x) (ZeroExt8to32 y))) +(Neq16 x y) => (NotEqual (CMP (ZeroExt16to32 x) (ZeroExt16to32 y))) +(Neq32 x y) => (NotEqual (CMP x y)) +(NeqPtr x y) => (NotEqual (CMP x y)) +(Neq(32|64)F x y) => (NotEqual (CMP(F|D) x y)) -(Less8 x y) -> (LessThan (CMP (SignExt8to32 x) (SignExt8to32 y))) -(Less16 x y) -> (LessThan (CMP (SignExt16to32 x) (SignExt16to32 y))) -(Less32 x y) -> (LessThan (CMP x y)) -(Less(32|64)F x y) -> (GreaterThan (CMP(F|D) y x)) // reverse operands to work around NaN +(Less8 x y) => (LessThan (CMP (SignExt8to32 x) (SignExt8to32 y))) +(Less16 x y) => (LessThan (CMP (SignExt16to32 x) (SignExt16to32 y))) +(Less32 x y) => (LessThan (CMP x y)) +(Less(32|64)F x y) => (GreaterThan (CMP(F|D) y x)) // reverse operands to work around NaN -(Less8U x y) -> (LessThanU (CMP (ZeroExt8to32 x) (ZeroExt8to32 y))) -(Less16U x y) -> (LessThanU (CMP (ZeroExt16to32 x) (ZeroExt16to32 y))) -(Less32U x y) -> (LessThanU (CMP x y)) +(Less8U x y) => (LessThanU (CMP (ZeroExt8to32 x) (ZeroExt8to32 y))) +(Less16U x y) => (LessThanU (CMP (ZeroExt16to32 x) (ZeroExt16to32 y))) +(Less32U x y) => (LessThanU (CMP x y)) -(Leq8 x y) -> (LessEqual (CMP (SignExt8to32 x) (SignExt8to32 y))) -(Leq16 x y) -> (LessEqual (CMP (SignExt16to32 x) (SignExt16to32 y))) -(Leq32 x y) -> (LessEqual (CMP x y)) -(Leq(32|64)F x y) -> (GreaterEqual (CMP(F|D) y x)) // reverse operands to work around NaN +(Leq8 x y) => (LessEqual (CMP (SignExt8to32 x) (SignExt8to32 y))) +(Leq16 x y) => (LessEqual (CMP (SignExt16to32 x) (SignExt16to32 y))) +(Leq32 x y) => (LessEqual (CMP x y)) +(Leq(32|64)F x y) => (GreaterEqual (CMP(F|D) y x)) // reverse operands to work around NaN -(Leq8U x y) -> (LessEqualU (CMP (ZeroExt8to32 x) (ZeroExt8to32 y))) -(Leq16U x y) -> (LessEqualU (CMP (ZeroExt16to32 x) (ZeroExt16to32 y))) -(Leq32U x y) -> (LessEqualU (CMP x y)) +(Leq8U x y) => (LessEqualU (CMP (ZeroExt8to32 x) (ZeroExt8to32 y))) +(Leq16U x y) => (LessEqualU (CMP (ZeroExt16to32 x) (ZeroExt16to32 y))) +(Leq32U x y) => (LessEqualU (CMP x y)) -(OffPtr [off] ptr:(SP)) -> (MOVWaddr [off] ptr) -(OffPtr [off] ptr) -> (ADDconst [off] ptr) +(OffPtr [off] ptr:(SP)) => (MOVWaddr [int32(off)] ptr) +(OffPtr [off] ptr) => (ADDconst [int32(off)] ptr) -(Addr ...) -> (MOVWaddr ...) -(LocalAddr {sym} base _) -> (MOVWaddr {sym} base) +(Addr {sym} base) => (MOVWaddr {sym} base) +(LocalAddr {sym} base _) => (MOVWaddr {sym} base) // loads -(Load ptr mem) && t.IsBoolean() -> (MOVBUload ptr mem) -(Load ptr mem) && (is8BitInt(t) && isSigned(t)) -> (MOVBload ptr mem) -(Load ptr mem) && (is8BitInt(t) && !isSigned(t)) -> (MOVBUload ptr mem) -(Load ptr mem) && (is16BitInt(t) && isSigned(t)) -> (MOVHload ptr mem) -(Load ptr mem) && (is16BitInt(t) && !isSigned(t)) -> (MOVHUload ptr mem) -(Load ptr mem) && (is32BitInt(t) || isPtr(t)) -> (MOVWload ptr mem) -(Load ptr mem) && is32BitFloat(t) -> (MOVFload ptr mem) -(Load ptr mem) && is64BitFloat(t) -> (MOVDload ptr mem) +(Load ptr mem) && t.IsBoolean() => (MOVBUload ptr mem) +(Load ptr mem) && (is8BitInt(t) && isSigned(t)) => (MOVBload ptr mem) +(Load ptr mem) && (is8BitInt(t) && !isSigned(t)) => (MOVBUload ptr mem) +(Load ptr mem) && (is16BitInt(t) && isSigned(t)) => (MOVHload ptr mem) +(Load ptr mem) && (is16BitInt(t) && !isSigned(t)) => (MOVHUload ptr mem) +(Load ptr mem) && (is32BitInt(t) || isPtr(t)) => (MOVWload ptr mem) +(Load ptr mem) && is32BitFloat(t) => (MOVFload ptr mem) +(Load ptr mem) && is64BitFloat(t) => (MOVDload ptr mem) // stores -(Store {t} ptr val mem) && t.(*types.Type).Size() == 1 -> (MOVBstore ptr val mem) -(Store {t} ptr val mem) && t.(*types.Type).Size() == 2 -> (MOVHstore ptr val mem) -(Store {t} ptr val mem) && t.(*types.Type).Size() == 4 && !is32BitFloat(val.Type) -> (MOVWstore ptr val mem) -(Store {t} ptr val mem) && t.(*types.Type).Size() == 4 && is32BitFloat(val.Type) -> (MOVFstore ptr val mem) -(Store {t} ptr val mem) && t.(*types.Type).Size() == 8 && is64BitFloat(val.Type) -> (MOVDstore ptr val mem) +(Store {t} ptr val mem) && t.Size() == 1 => (MOVBstore ptr val mem) +(Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem) +(Store {t} ptr val mem) && t.Size() == 4 && !is32BitFloat(val.Type) => (MOVWstore ptr val mem) +(Store {t} ptr val mem) && t.Size() == 4 && is32BitFloat(val.Type) => (MOVFstore ptr val mem) +(Store {t} ptr val mem) && t.Size() == 8 && is64BitFloat(val.Type) => (MOVDstore ptr val mem) // zero instructions -(Zero [0] _ mem) -> mem -(Zero [1] ptr mem) -> (MOVBstore ptr (MOVWconst [0]) mem) -(Zero [2] {t} ptr mem) && t.(*types.Type).Alignment()%2 == 0 -> +(Zero [0] _ mem) => mem +(Zero [1] ptr mem) => (MOVBstore ptr (MOVWconst [0]) mem) +(Zero [2] {t} ptr mem) && t.Alignment()%2 == 0 => (MOVHstore ptr (MOVWconst [0]) mem) -(Zero [2] ptr mem) -> +(Zero [2] ptr mem) => (MOVBstore [1] ptr (MOVWconst [0]) (MOVBstore [0] ptr (MOVWconst [0]) mem)) -(Zero [4] {t} ptr mem) && t.(*types.Type).Alignment()%4 == 0 -> +(Zero [4] {t} ptr mem) && t.Alignment()%4 == 0 => (MOVWstore ptr (MOVWconst [0]) mem) -(Zero [4] {t} ptr mem) && t.(*types.Type).Alignment()%2 == 0 -> +(Zero [4] {t} ptr mem) && t.Alignment()%2 == 0 => (MOVHstore [2] ptr (MOVWconst [0]) (MOVHstore [0] ptr (MOVWconst [0]) mem)) -(Zero [4] ptr mem) -> +(Zero [4] ptr mem) => (MOVBstore [3] ptr (MOVWconst [0]) (MOVBstore [2] ptr (MOVWconst [0]) (MOVBstore [1] ptr (MOVWconst [0]) (MOVBstore [0] ptr (MOVWconst [0]) mem)))) -(Zero [3] ptr mem) -> +(Zero [3] ptr mem) => (MOVBstore [2] ptr (MOVWconst [0]) (MOVBstore [1] ptr (MOVWconst [0]) (MOVBstore [0] ptr (MOVWconst [0]) mem))) @@ -294,38 +294,38 @@ // 4 and 128 are magic constants, see runtime/mkduff.go (Zero [s] {t} ptr mem) && s%4 == 0 && s > 4 && s <= 512 - && t.(*types.Type).Alignment()%4 == 0 && !config.noDuffDevice -> + && t.Alignment()%4 == 0 && !config.noDuffDevice => (DUFFZERO [4 * (128 - s/4)] ptr (MOVWconst [0]) mem) // Large zeroing uses a loop (Zero [s] {t} ptr mem) - && (s > 512 || config.noDuffDevice) || t.(*types.Type).Alignment()%4 != 0 -> - (LoweredZero [t.(*types.Type).Alignment()] + && (s > 512 || config.noDuffDevice) || t.Alignment()%4 != 0 => + (LoweredZero [t.Alignment()] ptr - (ADDconst ptr [s-moveSize(t.(*types.Type).Alignment(), config)]) + (ADDconst ptr [int32(s-moveSize(t.Alignment(), config))]) (MOVWconst [0]) mem) // moves -(Move [0] _ _ mem) -> mem -(Move [1] dst src mem) -> (MOVBstore dst (MOVBUload src mem) mem) -(Move [2] {t} dst src mem) && t.(*types.Type).Alignment()%2 == 0 -> +(Move [0] _ _ mem) => mem +(Move [1] dst src mem) => (MOVBstore dst (MOVBUload src mem) mem) +(Move [2] {t} dst src mem) && t.Alignment()%2 == 0 => (MOVHstore dst (MOVHUload src mem) mem) -(Move [2] dst src mem) -> +(Move [2] dst src mem) => (MOVBstore [1] dst (MOVBUload [1] src mem) (MOVBstore dst (MOVBUload src mem) mem)) -(Move [4] {t} dst src mem) && t.(*types.Type).Alignment()%4 == 0 -> +(Move [4] {t} dst src mem) && t.Alignment()%4 == 0 => (MOVWstore dst (MOVWload src mem) mem) -(Move [4] {t} dst src mem) && t.(*types.Type).Alignment()%2 == 0 -> +(Move [4] {t} dst src mem) && t.Alignment()%2 == 0 => (MOVHstore [2] dst (MOVHUload [2] src mem) (MOVHstore dst (MOVHUload src mem) mem)) -(Move [4] dst src mem) -> +(Move [4] dst src mem) => (MOVBstore [3] dst (MOVBUload [3] src mem) (MOVBstore [2] dst (MOVBUload [2] src mem) (MOVBstore [1] dst (MOVBUload [1] src mem) (MOVBstore dst (MOVBUload src mem) mem)))) -(Move [3] dst src mem) -> +(Move [3] dst src mem) => (MOVBstore [2] dst (MOVBUload [2] src mem) (MOVBstore [1] dst (MOVBUload [1] src mem) (MOVBstore dst (MOVBUload src mem) mem))) @@ -334,279 +334,279 @@ // 8 and 128 are magic constants, see runtime/mkduff.go (Move [s] {t} dst src mem) && s%4 == 0 && s > 4 && s <= 512 - && t.(*types.Type).Alignment()%4 == 0 && !config.noDuffDevice && logLargeCopy(v, s) -> + && t.Alignment()%4 == 0 && !config.noDuffDevice && logLargeCopy(v, s) => (DUFFCOPY [8 * (128 - s/4)] dst src mem) // Large move uses a loop (Move [s] {t} dst src mem) - && ((s > 512 || config.noDuffDevice) || t.(*types.Type).Alignment()%4 != 0) && logLargeCopy(v, s) -> - (LoweredMove [t.(*types.Type).Alignment()] + && ((s > 512 || config.noDuffDevice) || t.Alignment()%4 != 0) && logLargeCopy(v, s) => + (LoweredMove [t.Alignment()] dst src - (ADDconst src [s-moveSize(t.(*types.Type).Alignment(), config)]) + (ADDconst src [int32(s-moveSize(t.Alignment(), config))]) mem) // calls -(StaticCall ...) -> (CALLstatic ...) -(ClosureCall ...) -> (CALLclosure ...) -(InterCall ...) -> (CALLinter ...) +(StaticCall ...) => (CALLstatic ...) +(ClosureCall ...) => (CALLclosure ...) +(InterCall ...) => (CALLinter ...) // checks -(NilCheck ...) -> (LoweredNilCheck ...) -(IsNonNil ptr) -> (NotEqual (CMPconst [0] ptr)) -(IsInBounds idx len) -> (LessThanU (CMP idx len)) -(IsSliceInBounds idx len) -> (LessEqualU (CMP idx len)) +(NilCheck ...) => (LoweredNilCheck ...) +(IsNonNil ptr) => (NotEqual (CMPconst [0] ptr)) +(IsInBounds idx len) => (LessThanU (CMP idx len)) +(IsSliceInBounds idx len) => (LessEqualU (CMP idx len)) // pseudo-ops -(GetClosurePtr ...) -> (LoweredGetClosurePtr ...) -(GetCallerSP ...) -> (LoweredGetCallerSP ...) -(GetCallerPC ...) -> (LoweredGetCallerPC ...) +(GetClosurePtr ...) => (LoweredGetClosurePtr ...) +(GetCallerSP ...) => (LoweredGetCallerSP ...) +(GetCallerPC ...) => (LoweredGetCallerPC ...) // Absorb pseudo-ops into blocks. -(If (Equal cc) yes no) -> (EQ cc yes no) -(If (NotEqual cc) yes no) -> (NE cc yes no) -(If (LessThan cc) yes no) -> (LT cc yes no) -(If (LessThanU cc) yes no) -> (ULT cc yes no) -(If (LessEqual cc) yes no) -> (LE cc yes no) -(If (LessEqualU cc) yes no) -> (ULE cc yes no) -(If (GreaterThan cc) yes no) -> (GT cc yes no) -(If (GreaterThanU cc) yes no) -> (UGT cc yes no) -(If (GreaterEqual cc) yes no) -> (GE cc yes no) -(If (GreaterEqualU cc) yes no) -> (UGE cc yes no) +(If (Equal cc) yes no) => (EQ cc yes no) +(If (NotEqual cc) yes no) => (NE cc yes no) +(If (LessThan cc) yes no) => (LT cc yes no) +(If (LessThanU cc) yes no) => (ULT cc yes no) +(If (LessEqual cc) yes no) => (LE cc yes no) +(If (LessEqualU cc) yes no) => (ULE cc yes no) +(If (GreaterThan cc) yes no) => (GT cc yes no) +(If (GreaterThanU cc) yes no) => (UGT cc yes no) +(If (GreaterEqual cc) yes no) => (GE cc yes no) +(If (GreaterEqualU cc) yes no) => (UGE cc yes no) -(If cond yes no) -> (NE (CMPconst [0] cond) yes no) +(If cond yes no) => (NE (CMPconst [0] cond) yes no) // Absorb boolean tests into block -(NE (CMPconst [0] (Equal cc)) yes no) -> (EQ cc yes no) -(NE (CMPconst [0] (NotEqual cc)) yes no) -> (NE cc yes no) -(NE (CMPconst [0] (LessThan cc)) yes no) -> (LT cc yes no) -(NE (CMPconst [0] (LessThanU cc)) yes no) -> (ULT cc yes no) -(NE (CMPconst [0] (LessEqual cc)) yes no) -> (LE cc yes no) -(NE (CMPconst [0] (LessEqualU cc)) yes no) -> (ULE cc yes no) -(NE (CMPconst [0] (GreaterThan cc)) yes no) -> (GT cc yes no) -(NE (CMPconst [0] (GreaterThanU cc)) yes no) -> (UGT cc yes no) -(NE (CMPconst [0] (GreaterEqual cc)) yes no) -> (GE cc yes no) -(NE (CMPconst [0] (GreaterEqualU cc)) yes no) -> (UGE cc yes no) +(NE (CMPconst [0] (Equal cc)) yes no) => (EQ cc yes no) +(NE (CMPconst [0] (NotEqual cc)) yes no) => (NE cc yes no) +(NE (CMPconst [0] (LessThan cc)) yes no) => (LT cc yes no) +(NE (CMPconst [0] (LessThanU cc)) yes no) => (ULT cc yes no) +(NE (CMPconst [0] (LessEqual cc)) yes no) => (LE cc yes no) +(NE (CMPconst [0] (LessEqualU cc)) yes no) => (ULE cc yes no) +(NE (CMPconst [0] (GreaterThan cc)) yes no) => (GT cc yes no) +(NE (CMPconst [0] (GreaterThanU cc)) yes no) => (UGT cc yes no) +(NE (CMPconst [0] (GreaterEqual cc)) yes no) => (GE cc yes no) +(NE (CMPconst [0] (GreaterEqualU cc)) yes no) => (UGE cc yes no) // Write barrier. -(WB ...) -> (LoweredWB ...) +(WB ...) => (LoweredWB ...) -(PanicBounds [kind] x y mem) && boundsABI(kind) == 0 -> (LoweredPanicBoundsA [kind] x y mem) -(PanicBounds [kind] x y mem) && boundsABI(kind) == 1 -> (LoweredPanicBoundsB [kind] x y mem) -(PanicBounds [kind] x y mem) && boundsABI(kind) == 2 -> (LoweredPanicBoundsC [kind] x y mem) +(PanicBounds [kind] x y mem) && boundsABI(kind) == 0 => (LoweredPanicBoundsA [kind] x y mem) +(PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem) +(PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem) -(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 0 -> (LoweredPanicExtendA [kind] hi lo y mem) -(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 1 -> (LoweredPanicExtendB [kind] hi lo y mem) -(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 2 -> (LoweredPanicExtendC [kind] hi lo y mem) +(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 0 => (LoweredPanicExtendA [kind] hi lo y mem) +(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 1 => (LoweredPanicExtendB [kind] hi lo y mem) +(PanicExtend [kind] hi lo y mem) && boundsABI(kind) == 2 => (LoweredPanicExtendC [kind] hi lo y mem) // Optimizations // fold offset into address -(ADDconst [off1] (MOVWaddr [off2] {sym} ptr)) -> (MOVWaddr [off1+off2] {sym} ptr) -(SUBconst [off1] (MOVWaddr [off2] {sym} ptr)) -> (MOVWaddr [off2-off1] {sym} ptr) +(ADDconst [off1] (MOVWaddr [off2] {sym} ptr)) => (MOVWaddr [off1+off2] {sym} ptr) +(SUBconst [off1] (MOVWaddr [off2] {sym} ptr)) => (MOVWaddr [off2-off1] {sym} ptr) // fold address into load/store -(MOVBload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVBload [off1+off2] {sym} ptr mem) -(MOVBload [off1] {sym} (SUBconst [off2] ptr) mem) -> (MOVBload [off1-off2] {sym} ptr mem) -(MOVBUload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVBUload [off1+off2] {sym} ptr mem) -(MOVBUload [off1] {sym} (SUBconst [off2] ptr) mem) -> (MOVBUload [off1-off2] {sym} ptr mem) -(MOVHload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVHload [off1+off2] {sym} ptr mem) -(MOVHload [off1] {sym} (SUBconst [off2] ptr) mem) -> (MOVHload [off1-off2] {sym} ptr mem) -(MOVHUload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVHUload [off1+off2] {sym} ptr mem) -(MOVHUload [off1] {sym} (SUBconst [off2] ptr) mem) -> (MOVHUload [off1-off2] {sym} ptr mem) -(MOVWload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVWload [off1+off2] {sym} ptr mem) -(MOVWload [off1] {sym} (SUBconst [off2] ptr) mem) -> (MOVWload [off1-off2] {sym} ptr mem) -(MOVFload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVFload [off1+off2] {sym} ptr mem) -(MOVFload [off1] {sym} (SUBconst [off2] ptr) mem) -> (MOVFload [off1-off2] {sym} ptr mem) -(MOVDload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVDload [off1+off2] {sym} ptr mem) -(MOVDload [off1] {sym} (SUBconst [off2] ptr) mem) -> (MOVDload [off1-off2] {sym} ptr mem) +(MOVBload [off1] {sym} (ADDconst [off2] ptr) mem) => (MOVBload [off1+off2] {sym} ptr mem) +(MOVBload [off1] {sym} (SUBconst [off2] ptr) mem) => (MOVBload [off1-off2] {sym} ptr mem) +(MOVBUload [off1] {sym} (ADDconst [off2] ptr) mem) => (MOVBUload [off1+off2] {sym} ptr mem) +(MOVBUload [off1] {sym} (SUBconst [off2] ptr) mem) => (MOVBUload [off1-off2] {sym} ptr mem) +(MOVHload [off1] {sym} (ADDconst [off2] ptr) mem) => (MOVHload [off1+off2] {sym} ptr mem) +(MOVHload [off1] {sym} (SUBconst [off2] ptr) mem) => (MOVHload [off1-off2] {sym} ptr mem) +(MOVHUload [off1] {sym} (ADDconst [off2] ptr) mem) => (MOVHUload [off1+off2] {sym} ptr mem) +(MOVHUload [off1] {sym} (SUBconst [off2] ptr) mem) => (MOVHUload [off1-off2] {sym} ptr mem) +(MOVWload [off1] {sym} (ADDconst [off2] ptr) mem) => (MOVWload [off1+off2] {sym} ptr mem) +(MOVWload [off1] {sym} (SUBconst [off2] ptr) mem) => (MOVWload [off1-off2] {sym} ptr mem) +(MOVFload [off1] {sym} (ADDconst [off2] ptr) mem) => (MOVFload [off1+off2] {sym} ptr mem) +(MOVFload [off1] {sym} (SUBconst [off2] ptr) mem) => (MOVFload [off1-off2] {sym} ptr mem) +(MOVDload [off1] {sym} (ADDconst [off2] ptr) mem) => (MOVDload [off1+off2] {sym} ptr mem) +(MOVDload [off1] {sym} (SUBconst [off2] ptr) mem) => (MOVDload [off1-off2] {sym} ptr mem) -(MOVBstore [off1] {sym} (ADDconst [off2] ptr) val mem) -> (MOVBstore [off1+off2] {sym} ptr val mem) -(MOVBstore [off1] {sym} (SUBconst [off2] ptr) val mem) -> (MOVBstore [off1-off2] {sym} ptr val mem) -(MOVHstore [off1] {sym} (ADDconst [off2] ptr) val mem) -> (MOVHstore [off1+off2] {sym} ptr val mem) -(MOVHstore [off1] {sym} (SUBconst [off2] ptr) val mem) -> (MOVHstore [off1-off2] {sym} ptr val mem) -(MOVWstore [off1] {sym} (ADDconst [off2] ptr) val mem) -> (MOVWstore [off1+off2] {sym} ptr val mem) -(MOVWstore [off1] {sym} (SUBconst [off2] ptr) val mem) -> (MOVWstore [off1-off2] {sym} ptr val mem) -(MOVFstore [off1] {sym} (ADDconst [off2] ptr) val mem) -> (MOVFstore [off1+off2] {sym} ptr val mem) -(MOVFstore [off1] {sym} (SUBconst [off2] ptr) val mem) -> (MOVFstore [off1-off2] {sym} ptr val mem) -(MOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) -> (MOVDstore [off1+off2] {sym} ptr val mem) -(MOVDstore [off1] {sym} (SUBconst [off2] ptr) val mem) -> (MOVDstore [off1-off2] {sym} ptr val mem) +(MOVBstore [off1] {sym} (ADDconst [off2] ptr) val mem) => (MOVBstore [off1+off2] {sym} ptr val mem) +(MOVBstore [off1] {sym} (SUBconst [off2] ptr) val mem) => (MOVBstore [off1-off2] {sym} ptr val mem) +(MOVHstore [off1] {sym} (ADDconst [off2] ptr) val mem) => (MOVHstore [off1+off2] {sym} ptr val mem) +(MOVHstore [off1] {sym} (SUBconst [off2] ptr) val mem) => (MOVHstore [off1-off2] {sym} ptr val mem) +(MOVWstore [off1] {sym} (ADDconst [off2] ptr) val mem) => (MOVWstore [off1+off2] {sym} ptr val mem) +(MOVWstore [off1] {sym} (SUBconst [off2] ptr) val mem) => (MOVWstore [off1-off2] {sym} ptr val mem) +(MOVFstore [off1] {sym} (ADDconst [off2] ptr) val mem) => (MOVFstore [off1+off2] {sym} ptr val mem) +(MOVFstore [off1] {sym} (SUBconst [off2] ptr) val mem) => (MOVFstore [off1-off2] {sym} ptr val mem) +(MOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem) => (MOVDstore [off1+off2] {sym} ptr val mem) +(MOVDstore [off1] {sym} (SUBconst [off2] ptr) val mem) => (MOVDstore [off1-off2] {sym} ptr val mem) -(MOVBload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) -> +(MOVBload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => (MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) -(MOVBUload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) -> +(MOVBUload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => (MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) -(MOVHload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) -> +(MOVHload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => (MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) -(MOVHUload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) -> +(MOVHUload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => (MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) -(MOVWload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) -> +(MOVWload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => (MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) -(MOVFload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) -> +(MOVFload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => (MOVFload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) -(MOVDload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) -> +(MOVDload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => (MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) -(MOVBstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) -> +(MOVBstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) -(MOVHstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) -> +(MOVHstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) -(MOVWstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) -> +(MOVWstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) -(MOVFstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) -> +(MOVFstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => (MOVFstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) -(MOVDstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) -> +(MOVDstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) // replace load from same location as preceding store with zero/sign extension (or copy in case of full width) -(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBreg x) -(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVBUreg x) -(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHreg x) -(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVHUreg x) -(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x +(MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVBreg x) +(MOVBUload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVBUreg x) +(MOVHload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVHreg x) +(MOVHUload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVHUreg x) +(MOVWload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => x -(MOVFload [off] {sym} ptr (MOVFstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x -(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x +(MOVFload [off] {sym} ptr (MOVFstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => x +(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => x -(MOVWloadidx ptr idx (MOVWstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) -> x -(MOVWloadshiftLL ptr idx [c] (MOVWstoreshiftLL ptr2 idx [d] x _)) && c==d && isSamePtr(ptr, ptr2) -> x -(MOVWloadshiftRL ptr idx [c] (MOVWstoreshiftRL ptr2 idx [d] x _)) && c==d && isSamePtr(ptr, ptr2) -> x -(MOVWloadshiftRA ptr idx [c] (MOVWstoreshiftRA ptr2 idx [d] x _)) && c==d && isSamePtr(ptr, ptr2) -> x -(MOVBUloadidx ptr idx (MOVBstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) -> (MOVBUreg x) -(MOVBloadidx ptr idx (MOVBstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) -> (MOVBreg x) -(MOVHUloadidx ptr idx (MOVHstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) -> (MOVHUreg x) -(MOVHloadidx ptr idx (MOVHstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) -> (MOVHreg x) +(MOVWloadidx ptr idx (MOVWstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) => x +(MOVWloadshiftLL ptr idx [c] (MOVWstoreshiftLL ptr2 idx [d] x _)) && c==d && isSamePtr(ptr, ptr2) => x +(MOVWloadshiftRL ptr idx [c] (MOVWstoreshiftRL ptr2 idx [d] x _)) && c==d && isSamePtr(ptr, ptr2) => x +(MOVWloadshiftRA ptr idx [c] (MOVWstoreshiftRA ptr2 idx [d] x _)) && c==d && isSamePtr(ptr, ptr2) => x +(MOVBUloadidx ptr idx (MOVBstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) => (MOVBUreg x) +(MOVBloadidx ptr idx (MOVBstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) => (MOVBreg x) +(MOVHUloadidx ptr idx (MOVHstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) => (MOVHUreg x) +(MOVHloadidx ptr idx (MOVHstoreidx ptr2 idx x _)) && isSamePtr(ptr, ptr2) => (MOVHreg x) // fold constant into arithmatic ops -(ADD x (MOVWconst [c])) -> (ADDconst [c] x) -(SUB (MOVWconst [c]) x) -> (RSBconst [c] x) -(SUB x (MOVWconst [c])) -> (SUBconst [c] x) -(RSB (MOVWconst [c]) x) -> (SUBconst [c] x) -(RSB x (MOVWconst [c])) -> (RSBconst [c] x) +(ADD x (MOVWconst [c])) => (ADDconst [c] x) +(SUB (MOVWconst [c]) x) => (RSBconst [c] x) +(SUB x (MOVWconst [c])) => (SUBconst [c] x) +(RSB (MOVWconst [c]) x) => (SUBconst [c] x) +(RSB x (MOVWconst [c])) => (RSBconst [c] x) -(ADDS x (MOVWconst [c])) -> (ADDSconst [c] x) -(SUBS x (MOVWconst [c])) -> (SUBSconst [c] x) +(ADDS x (MOVWconst [c])) => (ADDSconst [c] x) +(SUBS x (MOVWconst [c])) => (SUBSconst [c] x) -(ADC (MOVWconst [c]) x flags) -> (ADCconst [c] x flags) -(SBC (MOVWconst [c]) x flags) -> (RSCconst [c] x flags) -(SBC x (MOVWconst [c]) flags) -> (SBCconst [c] x flags) +(ADC (MOVWconst [c]) x flags) => (ADCconst [c] x flags) +(SBC (MOVWconst [c]) x flags) => (RSCconst [c] x flags) +(SBC x (MOVWconst [c]) flags) => (SBCconst [c] x flags) -(AND x (MOVWconst [c])) -> (ANDconst [c] x) -(OR x (MOVWconst [c])) -> (ORconst [c] x) -(XOR x (MOVWconst [c])) -> (XORconst [c] x) -(BIC x (MOVWconst [c])) -> (BICconst [c] x) +(AND x (MOVWconst [c])) => (ANDconst [c] x) +(OR x (MOVWconst [c])) => (ORconst [c] x) +(XOR x (MOVWconst [c])) => (XORconst [c] x) +(BIC x (MOVWconst [c])) => (BICconst [c] x) -(SLL x (MOVWconst [c])) -> (SLLconst x [c&31]) // Note: I don't think we ever generate bad constant shifts (i.e. c>=32) -(SRL x (MOVWconst [c])) -> (SRLconst x [c&31]) -(SRA x (MOVWconst [c])) -> (SRAconst x [c&31]) +(SLL x (MOVWconst [c])) => (SLLconst x [c&31]) // Note: I don't think we ever generate bad constant shifts (i.e. c>=32) +(SRL x (MOVWconst [c])) => (SRLconst x [c&31]) +(SRA x (MOVWconst [c])) => (SRAconst x [c&31]) -(CMP x (MOVWconst [c])) -> (CMPconst [c] x) -(CMP (MOVWconst [c]) x) -> (InvertFlags (CMPconst [c] x)) -(CMN x (MOVWconst [c])) -> (CMNconst [c] x) -(TST x (MOVWconst [c])) -> (TSTconst [c] x) -(TEQ x (MOVWconst [c])) -> (TEQconst [c] x) +(CMP x (MOVWconst [c])) => (CMPconst [c] x) +(CMP (MOVWconst [c]) x) => (InvertFlags (CMPconst [c] x)) +(CMN x (MOVWconst [c])) => (CMNconst [c] x) +(TST x (MOVWconst [c])) => (TSTconst [c] x) +(TEQ x (MOVWconst [c])) => (TEQconst [c] x) // Canonicalize the order of arguments to comparisons - helps with CSE. -(CMP x y) && x.ID > y.ID -> (InvertFlags (CMP y x)) +(CMP x y) && x.ID > y.ID => (InvertFlags (CMP y x)) // don't extend after proper load // MOVWreg instruction is not emitted if src and dst registers are same, but it ensures the type. -(MOVBreg x:(MOVBload _ _)) -> (MOVWreg x) -(MOVBUreg x:(MOVBUload _ _)) -> (MOVWreg x) -(MOVHreg x:(MOVBload _ _)) -> (MOVWreg x) -(MOVHreg x:(MOVBUload _ _)) -> (MOVWreg x) -(MOVHreg x:(MOVHload _ _)) -> (MOVWreg x) -(MOVHUreg x:(MOVBUload _ _)) -> (MOVWreg x) -(MOVHUreg x:(MOVHUload _ _)) -> (MOVWreg x) +(MOVBreg x:(MOVBload _ _)) => (MOVWreg x) +(MOVBUreg x:(MOVBUload _ _)) => (MOVWreg x) +(MOVHreg x:(MOVBload _ _)) => (MOVWreg x) +(MOVHreg x:(MOVBUload _ _)) => (MOVWreg x) +(MOVHreg x:(MOVHload _ _)) => (MOVWreg x) +(MOVHUreg x:(MOVBUload _ _)) => (MOVWreg x) +(MOVHUreg x:(MOVHUload _ _)) => (MOVWreg x) // fold extensions and ANDs together -(MOVBUreg (ANDconst [c] x)) -> (ANDconst [c&0xff] x) -(MOVHUreg (ANDconst [c] x)) -> (ANDconst [c&0xffff] x) -(MOVBreg (ANDconst [c] x)) && c & 0x80 == 0 -> (ANDconst [c&0x7f] x) -(MOVHreg (ANDconst [c] x)) && c & 0x8000 == 0 -> (ANDconst [c&0x7fff] x) +(MOVBUreg (ANDconst [c] x)) => (ANDconst [c&0xff] x) +(MOVHUreg (ANDconst [c] x)) => (ANDconst [c&0xffff] x) +(MOVBreg (ANDconst [c] x)) && c & 0x80 == 0 => (ANDconst [c&0x7f] x) +(MOVHreg (ANDconst [c] x)) && c & 0x8000 == 0 => (ANDconst [c&0x7fff] x) // fold double extensions -(MOVBreg x:(MOVBreg _)) -> (MOVWreg x) -(MOVBUreg x:(MOVBUreg _)) -> (MOVWreg x) -(MOVHreg x:(MOVBreg _)) -> (MOVWreg x) -(MOVHreg x:(MOVBUreg _)) -> (MOVWreg x) -(MOVHreg x:(MOVHreg _)) -> (MOVWreg x) -(MOVHUreg x:(MOVBUreg _)) -> (MOVWreg x) -(MOVHUreg x:(MOVHUreg _)) -> (MOVWreg x) +(MOVBreg x:(MOVBreg _)) => (MOVWreg x) +(MOVBUreg x:(MOVBUreg _)) => (MOVWreg x) +(MOVHreg x:(MOVBreg _)) => (MOVWreg x) +(MOVHreg x:(MOVBUreg _)) => (MOVWreg x) +(MOVHreg x:(MOVHreg _)) => (MOVWreg x) +(MOVHUreg x:(MOVBUreg _)) => (MOVWreg x) +(MOVHUreg x:(MOVHUreg _)) => (MOVWreg x) // don't extend before store -(MOVBstore [off] {sym} ptr (MOVBreg x) mem) -> (MOVBstore [off] {sym} ptr x mem) -(MOVBstore [off] {sym} ptr (MOVBUreg x) mem) -> (MOVBstore [off] {sym} ptr x mem) -(MOVBstore [off] {sym} ptr (MOVHreg x) mem) -> (MOVBstore [off] {sym} ptr x mem) -(MOVBstore [off] {sym} ptr (MOVHUreg x) mem) -> (MOVBstore [off] {sym} ptr x mem) -(MOVHstore [off] {sym} ptr (MOVHreg x) mem) -> (MOVHstore [off] {sym} ptr x mem) -(MOVHstore [off] {sym} ptr (MOVHUreg x) mem) -> (MOVHstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVBUreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVHreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem) +(MOVHstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVHstore [off] {sym} ptr x mem) // if a register move has only 1 use, just use the same register without emitting instruction // MOVWnop doesn't emit instruction, only for ensuring the type. -(MOVWreg x) && x.Uses == 1 -> (MOVWnop x) +(MOVWreg x) && x.Uses == 1 => (MOVWnop x) // mul by constant -(MUL x (MOVWconst [c])) && int32(c) == -1 -> (RSBconst [0] x) -(MUL _ (MOVWconst [0])) -> (MOVWconst [0]) -(MUL x (MOVWconst [1])) -> x -(MUL x (MOVWconst [c])) && isPowerOfTwo(c) -> (SLLconst [log2(c)] x) -(MUL x (MOVWconst [c])) && isPowerOfTwo(c-1) && int32(c) >= 3 -> (ADDshiftLL x x [log2(c-1)]) -(MUL x (MOVWconst [c])) && isPowerOfTwo(c+1) && int32(c) >= 7 -> (RSBshiftLL x x [log2(c+1)]) -(MUL x (MOVWconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (SLLconst [log2(c/3)] (ADDshiftLL x x [1])) -(MUL x (MOVWconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (SLLconst [log2(c/5)] (ADDshiftLL x x [2])) -(MUL x (MOVWconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (SLLconst [log2(c/7)] (RSBshiftLL x x [3])) -(MUL x (MOVWconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (SLLconst [log2(c/9)] (ADDshiftLL x x [3])) +(MUL x (MOVWconst [c])) && int32(c) == -1 => (RSBconst [0] x) +(MUL _ (MOVWconst [0])) => (MOVWconst [0]) +(MUL x (MOVWconst [1])) => x +(MUL x (MOVWconst [c])) && isPowerOfTwo32(c) => (SLLconst [int32(log32(c))] x) +(MUL x (MOVWconst [c])) && isPowerOfTwo32(c-1) && c >= 3 => (ADDshiftLL x x [int32(log32(c-1))]) +(MUL x (MOVWconst [c])) && isPowerOfTwo32(c+1) && c >= 7 => (RSBshiftLL x x [int32(log32(c+1))]) +(MUL x (MOVWconst [c])) && c%3 == 0 && isPowerOfTwo32(c/3) => (SLLconst [int32(log32(c/3))] (ADDshiftLL x x [1])) +(MUL x (MOVWconst [c])) && c%5 == 0 && isPowerOfTwo32(c/5) => (SLLconst [int32(log32(c/5))] (ADDshiftLL x x [2])) +(MUL x (MOVWconst [c])) && c%7 == 0 && isPowerOfTwo32(c/7) => (SLLconst [int32(log32(c/7))] (RSBshiftLL x x [3])) +(MUL x (MOVWconst [c])) && c%9 == 0 && isPowerOfTwo32(c/9) => (SLLconst [int32(log32(c/9))] (ADDshiftLL x x [3])) -(MULA x (MOVWconst [c]) a) && int32(c) == -1 -> (SUB a x) -(MULA _ (MOVWconst [0]) a) -> a -(MULA x (MOVWconst [1]) a) -> (ADD x a) -(MULA x (MOVWconst [c]) a) && isPowerOfTwo(c) -> (ADD (SLLconst [log2(c)] x) a) -(MULA x (MOVWconst [c]) a) && isPowerOfTwo(c-1) && int32(c) >= 3 -> (ADD (ADDshiftLL x x [log2(c-1)]) a) -(MULA x (MOVWconst [c]) a) && isPowerOfTwo(c+1) && int32(c) >= 7 -> (ADD (RSBshiftLL x x [log2(c+1)]) a) -(MULA x (MOVWconst [c]) a) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (ADD (SLLconst [log2(c/3)] (ADDshiftLL x x [1])) a) -(MULA x (MOVWconst [c]) a) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (ADD (SLLconst [log2(c/5)] (ADDshiftLL x x [2])) a) -(MULA x (MOVWconst [c]) a) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (ADD (SLLconst [log2(c/7)] (RSBshiftLL x x [3])) a) -(MULA x (MOVWconst [c]) a) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (ADD (SLLconst [log2(c/9)] (ADDshiftLL x x [3])) a) +(MULA x (MOVWconst [c]) a) && c == -1 => (SUB a x) +(MULA _ (MOVWconst [0]) a) => a +(MULA x (MOVWconst [1]) a) => (ADD x a) +(MULA x (MOVWconst [c]) a) && isPowerOfTwo32(c) => (ADD (SLLconst [int32(log32(c))] x) a) +(MULA x (MOVWconst [c]) a) && isPowerOfTwo32(c-1) && c >= 3 => (ADD (ADDshiftLL x x [int32(log32(c-1))]) a) +(MULA x (MOVWconst [c]) a) && isPowerOfTwo32(c+1) && c >= 7 => (ADD (RSBshiftLL x x [int32(log32(c+1))]) a) +(MULA x (MOVWconst [c]) a) && c%3 == 0 && isPowerOfTwo32(c/3) => (ADD (SLLconst [int32(log32(c/3))] (ADDshiftLL x x [1])) a) +(MULA x (MOVWconst [c]) a) && c%5 == 0 && isPowerOfTwo32(c/5) => (ADD (SLLconst [int32(log32(c/5))] (ADDshiftLL x x [2])) a) +(MULA x (MOVWconst [c]) a) && c%7 == 0 && isPowerOfTwo32(c/7) => (ADD (SLLconst [int32(log32(c/7))] (RSBshiftLL x x [3])) a) +(MULA x (MOVWconst [c]) a) && c%9 == 0 && isPowerOfTwo32(c/9) => (ADD (SLLconst [int32(log32(c/9))] (ADDshiftLL x x [3])) a) -(MULA (MOVWconst [c]) x a) && int32(c) == -1 -> (SUB a x) -(MULA (MOVWconst [0]) _ a) -> a -(MULA (MOVWconst [1]) x a) -> (ADD x a) -(MULA (MOVWconst [c]) x a) && isPowerOfTwo(c) -> (ADD (SLLconst [log2(c)] x) a) -(MULA (MOVWconst [c]) x a) && isPowerOfTwo(c-1) && int32(c) >= 3 -> (ADD (ADDshiftLL x x [log2(c-1)]) a) -(MULA (MOVWconst [c]) x a) && isPowerOfTwo(c+1) && int32(c) >= 7 -> (ADD (RSBshiftLL x x [log2(c+1)]) a) -(MULA (MOVWconst [c]) x a) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (ADD (SLLconst [log2(c/3)] (ADDshiftLL x x [1])) a) -(MULA (MOVWconst [c]) x a) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (ADD (SLLconst [log2(c/5)] (ADDshiftLL x x [2])) a) -(MULA (MOVWconst [c]) x a) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (ADD (SLLconst [log2(c/7)] (RSBshiftLL x x [3])) a) -(MULA (MOVWconst [c]) x a) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (ADD (SLLconst [log2(c/9)] (ADDshiftLL x x [3])) a) +(MULA (MOVWconst [c]) x a) && c == -1 => (SUB a x) +(MULA (MOVWconst [0]) _ a) => a +(MULA (MOVWconst [1]) x a) => (ADD x a) +(MULA (MOVWconst [c]) x a) && isPowerOfTwo32(c) => (ADD (SLLconst [int32(log32(c))] x) a) +(MULA (MOVWconst [c]) x a) && isPowerOfTwo32(c-1) && c >= 3 => (ADD (ADDshiftLL x x [int32(log32(c-1))]) a) +(MULA (MOVWconst [c]) x a) && isPowerOfTwo32(c+1) && c >= 7 => (ADD (RSBshiftLL x x [int32(log32(c+1))]) a) +(MULA (MOVWconst [c]) x a) && c%3 == 0 && isPowerOfTwo32(c/3) => (ADD (SLLconst [int32(log32(c/3))] (ADDshiftLL x x [1])) a) +(MULA (MOVWconst [c]) x a) && c%5 == 0 && isPowerOfTwo32(c/5) => (ADD (SLLconst [int32(log32(c/5))] (ADDshiftLL x x [2])) a) +(MULA (MOVWconst [c]) x a) && c%7 == 0 && isPowerOfTwo32(c/7) => (ADD (SLLconst [int32(log32(c/7))] (RSBshiftLL x x [3])) a) +(MULA (MOVWconst [c]) x a) && c%9 == 0 && isPowerOfTwo32(c/9) => (ADD (SLLconst [int32(log32(c/9))] (ADDshiftLL x x [3])) a) -(MULS x (MOVWconst [c]) a) && int32(c) == -1 -> (ADD a x) -(MULS _ (MOVWconst [0]) a) -> a -(MULS x (MOVWconst [1]) a) -> (RSB x a) -(MULS x (MOVWconst [c]) a) && isPowerOfTwo(c) -> (RSB (SLLconst [log2(c)] x) a) -(MULS x (MOVWconst [c]) a) && isPowerOfTwo(c-1) && int32(c) >= 3 -> (RSB (ADDshiftLL x x [log2(c-1)]) a) -(MULS x (MOVWconst [c]) a) && isPowerOfTwo(c+1) && int32(c) >= 7 -> (RSB (RSBshiftLL x x [log2(c+1)]) a) -(MULS x (MOVWconst [c]) a) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (RSB (SLLconst [log2(c/3)] (ADDshiftLL x x [1])) a) -(MULS x (MOVWconst [c]) a) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (RSB (SLLconst [log2(c/5)] (ADDshiftLL x x [2])) a) -(MULS x (MOVWconst [c]) a) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (RSB (SLLconst [log2(c/7)] (RSBshiftLL x x [3])) a) -(MULS x (MOVWconst [c]) a) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (RSB (SLLconst [log2(c/9)] (ADDshiftLL x x [3])) a) +(MULS x (MOVWconst [c]) a) && c == -1 => (ADD a x) +(MULS _ (MOVWconst [0]) a) => a +(MULS x (MOVWconst [1]) a) => (RSB x a) +(MULS x (MOVWconst [c]) a) && isPowerOfTwo32(c) => (RSB (SLLconst [int32(log32(c))] x) a) +(MULS x (MOVWconst [c]) a) && isPowerOfTwo32(c-1) && c >= 3 => (RSB (ADDshiftLL x x [int32(log32(c-1))]) a) +(MULS x (MOVWconst [c]) a) && isPowerOfTwo32(c+1) && c >= 7 => (RSB (RSBshiftLL x x [int32(log32(c+1))]) a) +(MULS x (MOVWconst [c]) a) && c%3 == 0 && isPowerOfTwo32(c/3) => (RSB (SLLconst [int32(log32(c/3))] (ADDshiftLL x x [1])) a) +(MULS x (MOVWconst [c]) a) && c%5 == 0 && isPowerOfTwo32(c/5) => (RSB (SLLconst [int32(log32(c/5))] (ADDshiftLL x x [2])) a) +(MULS x (MOVWconst [c]) a) && c%7 == 0 && isPowerOfTwo32(c/7) => (RSB (SLLconst [int32(log32(c/7))] (RSBshiftLL x x [3])) a) +(MULS x (MOVWconst [c]) a) && c%9 == 0 && isPowerOfTwo32(c/9) => (RSB (SLLconst [int32(log32(c/9))] (ADDshiftLL x x [3])) a) -(MULS (MOVWconst [c]) x a) && int32(c) == -1 -> (ADD a x) -(MULS (MOVWconst [0]) _ a) -> a -(MULS (MOVWconst [1]) x a) -> (RSB x a) -(MULS (MOVWconst [c]) x a) && isPowerOfTwo(c) -> (RSB (SLLconst [log2(c)] x) a) -(MULS (MOVWconst [c]) x a) && isPowerOfTwo(c-1) && int32(c) >= 3 -> (RSB (ADDshiftLL x x [log2(c-1)]) a) -(MULS (MOVWconst [c]) x a) && isPowerOfTwo(c+1) && int32(c) >= 7 -> (RSB (RSBshiftLL x x [log2(c+1)]) a) -(MULS (MOVWconst [c]) x a) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (RSB (SLLconst [log2(c/3)] (ADDshiftLL x x [1])) a) -(MULS (MOVWconst [c]) x a) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (RSB (SLLconst [log2(c/5)] (ADDshiftLL x x [2])) a) -(MULS (MOVWconst [c]) x a) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (RSB (SLLconst [log2(c/7)] (RSBshiftLL x x [3])) a) -(MULS (MOVWconst [c]) x a) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (RSB (SLLconst [log2(c/9)] (ADDshiftLL x x [3])) a) +(MULS (MOVWconst [c]) x a) && c == -1 => (ADD a x) +(MULS (MOVWconst [0]) _ a) => a +(MULS (MOVWconst [1]) x a) => (RSB x a) +(MULS (MOVWconst [c]) x a) && isPowerOfTwo32(c) => (RSB (SLLconst [int32(log32(c))] x) a) +(MULS (MOVWconst [c]) x a) && isPowerOfTwo32(c-1) && c >= 3 => (RSB (ADDshiftLL x x [int32(log32(c-1))]) a) +(MULS (MOVWconst [c]) x a) && isPowerOfTwo32(c+1) && c >= 7 => (RSB (RSBshiftLL x x [int32(log32(c+1))]) a) +(MULS (MOVWconst [c]) x a) && c%3 == 0 && isPowerOfTwo32(c/3) => (RSB (SLLconst [int32(log32(c/3))] (ADDshiftLL x x [1])) a) +(MULS (MOVWconst [c]) x a) && c%5 == 0 && isPowerOfTwo32(c/5) => (RSB (SLLconst [int32(log32(c/5))] (ADDshiftLL x x [2])) a) +(MULS (MOVWconst [c]) x a) && c%7 == 0 && isPowerOfTwo32(c/7) => (RSB (SLLconst [int32(log32(c/7))] (RSBshiftLL x x [3])) a) +(MULS (MOVWconst [c]) x a) && c%9 == 0 && isPowerOfTwo32(c/9) => (RSB (SLLconst [int32(log32(c/9))] (ADDshiftLL x x [3])) a) // div by constant -(Select0 (CALLudiv x (MOVWconst [1]))) -> x -(Select1 (CALLudiv _ (MOVWconst [1]))) -> (MOVWconst [0]) -(Select0 (CALLudiv x (MOVWconst [c]))) && isPowerOfTwo(c) -> (SRLconst [log2(c)] x) -(Select1 (CALLudiv x (MOVWconst [c]))) && isPowerOfTwo(c) -> (ANDconst [c-1] x) +(Select0 (CALLudiv x (MOVWconst [1]))) => x +(Select1 (CALLudiv _ (MOVWconst [1]))) => (MOVWconst [0]) +(Select0 (CALLudiv x (MOVWconst [c]))) && isPowerOfTwo32(c) => (SRLconst [int32(log32(c))] x) +(Select1 (CALLudiv x (MOVWconst [c]))) && isPowerOfTwo32(c) => (ANDconst [c-1] x) // constant comparisons (CMPconst (MOVWconst [x]) [y]) => (FlagConstant [subFlags32(x,y)]) @@ -664,16 +664,16 @@ (GEnoov (FlagConstant [fc]) yes no) && !fc.geNoov() => (First no yes) // absorb InvertFlags into branches -(LT (InvertFlags cmp) yes no) -> (GT cmp yes no) -(GT (InvertFlags cmp) yes no) -> (LT cmp yes no) -(LE (InvertFlags cmp) yes no) -> (GE cmp yes no) -(GE (InvertFlags cmp) yes no) -> (LE cmp yes no) -(ULT (InvertFlags cmp) yes no) -> (UGT cmp yes no) -(UGT (InvertFlags cmp) yes no) -> (ULT cmp yes no) -(ULE (InvertFlags cmp) yes no) -> (UGE cmp yes no) -(UGE (InvertFlags cmp) yes no) -> (ULE cmp yes no) -(EQ (InvertFlags cmp) yes no) -> (EQ cmp yes no) -(NE (InvertFlags cmp) yes no) -> (NE cmp yes no) +(LT (InvertFlags cmp) yes no) => (GT cmp yes no) +(GT (InvertFlags cmp) yes no) => (LT cmp yes no) +(LE (InvertFlags cmp) yes no) => (GE cmp yes no) +(GE (InvertFlags cmp) yes no) => (LE cmp yes no) +(ULT (InvertFlags cmp) yes no) => (UGT cmp yes no) +(UGT (InvertFlags cmp) yes no) => (ULT cmp yes no) +(ULE (InvertFlags cmp) yes no) => (UGE cmp yes no) +(UGE (InvertFlags cmp) yes no) => (ULE cmp yes no) +(EQ (InvertFlags cmp) yes no) => (EQ cmp yes no) +(NE (InvertFlags cmp) yes no) => (NE cmp yes no) (LTnoov (InvertFlags cmp) yes no) => (GTnoov cmp yes no) (GEnoov (InvertFlags cmp) yes no) => (LEnoov cmp yes no) (LEnoov (InvertFlags cmp) yes no) => (GEnoov cmp yes no) @@ -692,16 +692,16 @@ (GreaterEqualU (FlagConstant [fc])) => (MOVWconst [b2i32(fc.uge())]) // absorb InvertFlags into boolean values -(Equal (InvertFlags x)) -> (Equal x) -(NotEqual (InvertFlags x)) -> (NotEqual x) -(LessThan (InvertFlags x)) -> (GreaterThan x) -(LessThanU (InvertFlags x)) -> (GreaterThanU x) -(GreaterThan (InvertFlags x)) -> (LessThan x) -(GreaterThanU (InvertFlags x)) -> (LessThanU x) -(LessEqual (InvertFlags x)) -> (GreaterEqual x) -(LessEqualU (InvertFlags x)) -> (GreaterEqualU x) -(GreaterEqual (InvertFlags x)) -> (LessEqual x) -(GreaterEqualU (InvertFlags x)) -> (LessEqualU x) +(Equal (InvertFlags x)) => (Equal x) +(NotEqual (InvertFlags x)) => (NotEqual x) +(LessThan (InvertFlags x)) => (GreaterThan x) +(LessThanU (InvertFlags x)) => (GreaterThanU x) +(GreaterThan (InvertFlags x)) => (LessThan x) +(GreaterThanU (InvertFlags x)) => (LessThanU x) +(LessEqual (InvertFlags x)) => (GreaterEqual x) +(LessEqualU (InvertFlags x)) => (GreaterEqualU x) +(GreaterEqual (InvertFlags x)) => (LessEqual x) +(GreaterEqualU (InvertFlags x)) => (LessEqualU x) // absorb flag constants into conditional instructions (CMOVWLSconst _ (FlagConstant [fc]) [c]) && fc.ule() => (MOVWconst [c]) @@ -710,766 +710,766 @@ (CMOVWHSconst _ (FlagConstant [fc]) [c]) && fc.uge() => (MOVWconst [c]) (CMOVWHSconst x (FlagConstant [fc]) [c]) && fc.ult() => x -(CMOVWLSconst x (InvertFlags flags) [c]) -> (CMOVWHSconst x flags [c]) -(CMOVWHSconst x (InvertFlags flags) [c]) -> (CMOVWLSconst x flags [c]) +(CMOVWLSconst x (InvertFlags flags) [c]) => (CMOVWHSconst x flags [c]) +(CMOVWHSconst x (InvertFlags flags) [c]) => (CMOVWLSconst x flags [c]) (SRAcond x _ (FlagConstant [fc])) && fc.uge() => (SRAconst x [31]) (SRAcond x y (FlagConstant [fc])) && fc.ult() => (SRA x y) // remove redundant *const ops -(ADDconst [0] x) -> x -(SUBconst [0] x) -> x -(ANDconst [0] _) -> (MOVWconst [0]) -(ANDconst [c] x) && int32(c)==-1 -> x -(ORconst [0] x) -> x -(ORconst [c] _) && int32(c)==-1 -> (MOVWconst [-1]) -(XORconst [0] x) -> x -(BICconst [0] x) -> x -(BICconst [c] _) && int32(c)==-1 -> (MOVWconst [0]) +(ADDconst [0] x) => x +(SUBconst [0] x) => x +(ANDconst [0] _) => (MOVWconst [0]) +(ANDconst [c] x) && int32(c)==-1 => x +(ORconst [0] x) => x +(ORconst [c] _) && int32(c)==-1 => (MOVWconst [-1]) +(XORconst [0] x) => x +(BICconst [0] x) => x +(BICconst [c] _) && int32(c)==-1 => (MOVWconst [0]) // generic constant folding -(ADDconst [c] x) && !isARMImmRot(uint32(c)) && isARMImmRot(uint32(-c)) -> (SUBconst [int64(int32(-c))] x) -(SUBconst [c] x) && !isARMImmRot(uint32(c)) && isARMImmRot(uint32(-c)) -> (ADDconst [int64(int32(-c))] x) -(ANDconst [c] x) && !isARMImmRot(uint32(c)) && isARMImmRot(^uint32(c)) -> (BICconst [int64(int32(^uint32(c)))] x) -(BICconst [c] x) && !isARMImmRot(uint32(c)) && isARMImmRot(^uint32(c)) -> (ANDconst [int64(int32(^uint32(c)))] x) -(ADDconst [c] x) && objabi.GOARM==7 && !isARMImmRot(uint32(c)) && uint32(c)>0xffff && uint32(-c)<=0xffff -> (SUBconst [int64(int32(-c))] x) -(SUBconst [c] x) && objabi.GOARM==7 && !isARMImmRot(uint32(c)) && uint32(c)>0xffff && uint32(-c)<=0xffff -> (ANDconst [int64(int32(-c))] x) -(ANDconst [c] x) && objabi.GOARM==7 && !isARMImmRot(uint32(c)) && uint32(c)>0xffff && ^uint32(c)<=0xffff -> (BICconst [int64(int32(^uint32(c)))] x) -(BICconst [c] x) && objabi.GOARM==7 && !isARMImmRot(uint32(c)) && uint32(c)>0xffff && ^uint32(c)<=0xffff -> (ANDconst [int64(int32(^uint32(c)))] x) -(ADDconst [c] (MOVWconst [d])) -> (MOVWconst [int64(int32(c+d))]) -(ADDconst [c] (ADDconst [d] x)) -> (ADDconst [int64(int32(c+d))] x) -(ADDconst [c] (SUBconst [d] x)) -> (ADDconst [int64(int32(c-d))] x) -(ADDconst [c] (RSBconst [d] x)) -> (RSBconst [int64(int32(c+d))] x) -(ADCconst [c] (ADDconst [d] x) flags) -> (ADCconst [int64(int32(c+d))] x flags) -(ADCconst [c] (SUBconst [d] x) flags) -> (ADCconst [int64(int32(c-d))] x flags) -(SUBconst [c] (MOVWconst [d])) -> (MOVWconst [int64(int32(d-c))]) -(SUBconst [c] (SUBconst [d] x)) -> (ADDconst [int64(int32(-c-d))] x) -(SUBconst [c] (ADDconst [d] x)) -> (ADDconst [int64(int32(-c+d))] x) -(SUBconst [c] (RSBconst [d] x)) -> (RSBconst [int64(int32(-c+d))] x) -(SBCconst [c] (ADDconst [d] x) flags) -> (SBCconst [int64(int32(c-d))] x flags) -(SBCconst [c] (SUBconst [d] x) flags) -> (SBCconst [int64(int32(c+d))] x flags) -(RSBconst [c] (MOVWconst [d])) -> (MOVWconst [int64(int32(c-d))]) -(RSBconst [c] (RSBconst [d] x)) -> (ADDconst [int64(int32(c-d))] x) -(RSBconst [c] (ADDconst [d] x)) -> (RSBconst [int64(int32(c-d))] x) -(RSBconst [c] (SUBconst [d] x)) -> (RSBconst [int64(int32(c+d))] x) -(RSCconst [c] (ADDconst [d] x) flags) -> (RSCconst [int64(int32(c-d))] x flags) -(RSCconst [c] (SUBconst [d] x) flags) -> (RSCconst [int64(int32(c+d))] x flags) -(SLLconst [c] (MOVWconst [d])) -> (MOVWconst [int64(int32(uint32(d)< (MOVWconst [int64(int32(uint32(d)>>uint64(c)))]) -(SRAconst [c] (MOVWconst [d])) -> (MOVWconst [int64(int32(d)>>uint64(c))]) -(MUL (MOVWconst [c]) (MOVWconst [d])) -> (MOVWconst [int64(int32(c*d))]) -(MULA (MOVWconst [c]) (MOVWconst [d]) a) -> (ADDconst [int64(int32(c*d))] a) -(MULS (MOVWconst [c]) (MOVWconst [d]) a) -> (SUBconst [int64(int32(c*d))] a) -(Select0 (CALLudiv (MOVWconst [c]) (MOVWconst [d]))) -> (MOVWconst [int64(int32(uint32(c)/uint32(d)))]) -(Select1 (CALLudiv (MOVWconst [c]) (MOVWconst [d]))) -> (MOVWconst [int64(int32(uint32(c)%uint32(d)))]) -(ANDconst [c] (MOVWconst [d])) -> (MOVWconst [c&d]) -(ANDconst [c] (ANDconst [d] x)) -> (ANDconst [c&d] x) -(ORconst [c] (MOVWconst [d])) -> (MOVWconst [c|d]) -(ORconst [c] (ORconst [d] x)) -> (ORconst [c|d] x) -(XORconst [c] (MOVWconst [d])) -> (MOVWconst [c^d]) -(XORconst [c] (XORconst [d] x)) -> (XORconst [c^d] x) -(BICconst [c] (MOVWconst [d])) -> (MOVWconst [d&^c]) -(BICconst [c] (BICconst [d] x)) -> (BICconst [int64(int32(c|d))] x) -(MVN (MOVWconst [c])) -> (MOVWconst [^c]) -(MOVBreg (MOVWconst [c])) -> (MOVWconst [int64(int8(c))]) -(MOVBUreg (MOVWconst [c])) -> (MOVWconst [int64(uint8(c))]) -(MOVHreg (MOVWconst [c])) -> (MOVWconst [int64(int16(c))]) -(MOVHUreg (MOVWconst [c])) -> (MOVWconst [int64(uint16(c))]) -(MOVWreg (MOVWconst [c])) -> (MOVWconst [c]) +(ADDconst [c] x) && !isARMImmRot(uint32(c)) && isARMImmRot(uint32(-c)) => (SUBconst [-c] x) +(SUBconst [c] x) && !isARMImmRot(uint32(c)) && isARMImmRot(uint32(-c)) => (ADDconst [-c] x) +(ANDconst [c] x) && !isARMImmRot(uint32(c)) && isARMImmRot(^uint32(c)) => (BICconst [int32(^uint32(c))] x) +(BICconst [c] x) && !isARMImmRot(uint32(c)) && isARMImmRot(^uint32(c)) => (ANDconst [int32(^uint32(c))] x) +(ADDconst [c] x) && objabi.GOARM==7 && !isARMImmRot(uint32(c)) && uint32(c)>0xffff && uint32(-c)<=0xffff => (SUBconst [-c] x) +(SUBconst [c] x) && objabi.GOARM==7 && !isARMImmRot(uint32(c)) && uint32(c)>0xffff && uint32(-c)<=0xffff => (ADDconst [-c] x) +(ANDconst [c] x) && objabi.GOARM==7 && !isARMImmRot(uint32(c)) && uint32(c)>0xffff && ^uint32(c)<=0xffff => (BICconst [int32(^uint32(c))] x) +(BICconst [c] x) && objabi.GOARM==7 && !isARMImmRot(uint32(c)) && uint32(c)>0xffff && ^uint32(c)<=0xffff => (ANDconst [int32(^uint32(c))] x) +(ADDconst [c] (MOVWconst [d])) => (MOVWconst [c+d]) +(ADDconst [c] (ADDconst [d] x)) => (ADDconst [c+d] x) +(ADDconst [c] (SUBconst [d] x)) => (ADDconst [c-d] x) +(ADDconst [c] (RSBconst [d] x)) => (RSBconst [c+d] x) +(ADCconst [c] (ADDconst [d] x) flags) => (ADCconst [c+d] x flags) +(ADCconst [c] (SUBconst [d] x) flags) => (ADCconst [c-d] x flags) +(SUBconst [c] (MOVWconst [d])) => (MOVWconst [d-c]) +(SUBconst [c] (SUBconst [d] x)) => (ADDconst [-c-d] x) +(SUBconst [c] (ADDconst [d] x)) => (ADDconst [-c+d] x) +(SUBconst [c] (RSBconst [d] x)) => (RSBconst [-c+d] x) +(SBCconst [c] (ADDconst [d] x) flags) => (SBCconst [c-d] x flags) +(SBCconst [c] (SUBconst [d] x) flags) => (SBCconst [c+d] x flags) +(RSBconst [c] (MOVWconst [d])) => (MOVWconst [c-d]) +(RSBconst [c] (RSBconst [d] x)) => (ADDconst [c-d] x) +(RSBconst [c] (ADDconst [d] x)) => (RSBconst [c-d] x) +(RSBconst [c] (SUBconst [d] x)) => (RSBconst [c+d] x) +(RSCconst [c] (ADDconst [d] x) flags) => (RSCconst [c-d] x flags) +(RSCconst [c] (SUBconst [d] x) flags) => (RSCconst [c+d] x flags) +(SLLconst [c] (MOVWconst [d])) => (MOVWconst [d< (MOVWconst [int32(uint32(d)>>uint64(c))]) +(SRAconst [c] (MOVWconst [d])) => (MOVWconst [d>>uint64(c)]) +(MUL (MOVWconst [c]) (MOVWconst [d])) => (MOVWconst [c*d]) +(MULA (MOVWconst [c]) (MOVWconst [d]) a) => (ADDconst [c*d] a) +(MULS (MOVWconst [c]) (MOVWconst [d]) a) => (SUBconst [c*d] a) +(Select0 (CALLudiv (MOVWconst [c]) (MOVWconst [d]))) && d != 0 => (MOVWconst [int32(uint32(c)/uint32(d))]) +(Select1 (CALLudiv (MOVWconst [c]) (MOVWconst [d]))) && d != 0 => (MOVWconst [int32(uint32(c)%uint32(d))]) +(ANDconst [c] (MOVWconst [d])) => (MOVWconst [c&d]) +(ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x) +(ORconst [c] (MOVWconst [d])) => (MOVWconst [c|d]) +(ORconst [c] (ORconst [d] x)) => (ORconst [c|d] x) +(XORconst [c] (MOVWconst [d])) => (MOVWconst [c^d]) +(XORconst [c] (XORconst [d] x)) => (XORconst [c^d] x) +(BICconst [c] (MOVWconst [d])) => (MOVWconst [d&^c]) +(BICconst [c] (BICconst [d] x)) => (BICconst [c|d] x) +(MVN (MOVWconst [c])) => (MOVWconst [^c]) +(MOVBreg (MOVWconst [c])) => (MOVWconst [int32(int8(c))]) +(MOVBUreg (MOVWconst [c])) => (MOVWconst [int32(uint8(c))]) +(MOVHreg (MOVWconst [c])) => (MOVWconst [int32(int16(c))]) +(MOVHUreg (MOVWconst [c])) => (MOVWconst [int32(uint16(c))]) +(MOVWreg (MOVWconst [c])) => (MOVWconst [c]) // BFX: Width = c >> 8, LSB = c & 0xff, result = d << (32 - Width - LSB) >> (32 - Width) -(BFX [c] (MOVWconst [d])) -> (MOVWconst [int64(int32(d)<<(32-uint32(c&0xff)-uint32(c>>8))>>(32-uint32(c>>8)))]) -(BFXU [c] (MOVWconst [d])) -> (MOVWconst [int64(int32(uint32(d)<<(32-uint32(c&0xff)-uint32(c>>8))>>(32-uint32(c>>8))))]) +(BFX [c] (MOVWconst [d])) => (MOVWconst [d<<(32-uint32(c&0xff)-uint32(c>>8))>>(32-uint32(c>>8))]) +(BFXU [c] (MOVWconst [d])) => (MOVWconst [int32(uint32(d)<<(32-uint32(c&0xff)-uint32(c>>8))>>(32-uint32(c>>8)))]) // absorb shifts into ops -(ADD x (SLLconst [c] y)) -> (ADDshiftLL x y [c]) -(ADD x (SRLconst [c] y)) -> (ADDshiftRL x y [c]) -(ADD x (SRAconst [c] y)) -> (ADDshiftRA x y [c]) -(ADD x (SLL y z)) -> (ADDshiftLLreg x y z) -(ADD x (SRL y z)) -> (ADDshiftRLreg x y z) -(ADD x (SRA y z)) -> (ADDshiftRAreg x y z) -(ADC x (SLLconst [c] y) flags) -> (ADCshiftLL x y [c] flags) -(ADC x (SRLconst [c] y) flags) -> (ADCshiftRL x y [c] flags) -(ADC x (SRAconst [c] y) flags) -> (ADCshiftRA x y [c] flags) -(ADC x (SLL y z) flags) -> (ADCshiftLLreg x y z flags) -(ADC x (SRL y z) flags) -> (ADCshiftRLreg x y z flags) -(ADC x (SRA y z) flags) -> (ADCshiftRAreg x y z flags) -(ADDS x (SLLconst [c] y)) -> (ADDSshiftLL x y [c]) -(ADDS x (SRLconst [c] y)) -> (ADDSshiftRL x y [c]) -(ADDS x (SRAconst [c] y)) -> (ADDSshiftRA x y [c]) -(ADDS x (SLL y z)) -> (ADDSshiftLLreg x y z) -(ADDS x (SRL y z)) -> (ADDSshiftRLreg x y z) -(ADDS x (SRA y z)) -> (ADDSshiftRAreg x y z) -(SUB x (SLLconst [c] y)) -> (SUBshiftLL x y [c]) -(SUB (SLLconst [c] y) x) -> (RSBshiftLL x y [c]) -(SUB x (SRLconst [c] y)) -> (SUBshiftRL x y [c]) -(SUB (SRLconst [c] y) x) -> (RSBshiftRL x y [c]) -(SUB x (SRAconst [c] y)) -> (SUBshiftRA x y [c]) -(SUB (SRAconst [c] y) x) -> (RSBshiftRA x y [c]) -(SUB x (SLL y z)) -> (SUBshiftLLreg x y z) -(SUB (SLL y z) x) -> (RSBshiftLLreg x y z) -(SUB x (SRL y z)) -> (SUBshiftRLreg x y z) -(SUB (SRL y z) x) -> (RSBshiftRLreg x y z) -(SUB x (SRA y z)) -> (SUBshiftRAreg x y z) -(SUB (SRA y z) x) -> (RSBshiftRAreg x y z) -(SBC x (SLLconst [c] y) flags) -> (SBCshiftLL x y [c] flags) -(SBC (SLLconst [c] y) x flags) -> (RSCshiftLL x y [c] flags) -(SBC x (SRLconst [c] y) flags) -> (SBCshiftRL x y [c] flags) -(SBC (SRLconst [c] y) x flags) -> (RSCshiftRL x y [c] flags) -(SBC x (SRAconst [c] y) flags) -> (SBCshiftRA x y [c] flags) -(SBC (SRAconst [c] y) x flags) -> (RSCshiftRA x y [c] flags) -(SBC x (SLL y z) flags) -> (SBCshiftLLreg x y z flags) -(SBC (SLL y z) x flags) -> (RSCshiftLLreg x y z flags) -(SBC x (SRL y z) flags) -> (SBCshiftRLreg x y z flags) -(SBC (SRL y z) x flags) -> (RSCshiftRLreg x y z flags) -(SBC x (SRA y z) flags) -> (SBCshiftRAreg x y z flags) -(SBC (SRA y z) x flags) -> (RSCshiftRAreg x y z flags) -(SUBS x (SLLconst [c] y)) -> (SUBSshiftLL x y [c]) -(SUBS (SLLconst [c] y) x) -> (RSBSshiftLL x y [c]) -(SUBS x (SRLconst [c] y)) -> (SUBSshiftRL x y [c]) -(SUBS (SRLconst [c] y) x) -> (RSBSshiftRL x y [c]) -(SUBS x (SRAconst [c] y)) -> (SUBSshiftRA x y [c]) -(SUBS (SRAconst [c] y) x) -> (RSBSshiftRA x y [c]) -(SUBS x (SLL y z)) -> (SUBSshiftLLreg x y z) -(SUBS (SLL y z) x) -> (RSBSshiftLLreg x y z) -(SUBS x (SRL y z)) -> (SUBSshiftRLreg x y z) -(SUBS (SRL y z) x) -> (RSBSshiftRLreg x y z) -(SUBS x (SRA y z)) -> (SUBSshiftRAreg x y z) -(SUBS (SRA y z) x) -> (RSBSshiftRAreg x y z) -(RSB x (SLLconst [c] y)) -> (RSBshiftLL x y [c]) -(RSB (SLLconst [c] y) x) -> (SUBshiftLL x y [c]) -(RSB x (SRLconst [c] y)) -> (RSBshiftRL x y [c]) -(RSB (SRLconst [c] y) x) -> (SUBshiftRL x y [c]) -(RSB x (SRAconst [c] y)) -> (RSBshiftRA x y [c]) -(RSB (SRAconst [c] y) x) -> (SUBshiftRA x y [c]) -(RSB x (SLL y z)) -> (RSBshiftLLreg x y z) -(RSB (SLL y z) x) -> (SUBshiftLLreg x y z) -(RSB x (SRL y z)) -> (RSBshiftRLreg x y z) -(RSB (SRL y z) x) -> (SUBshiftRLreg x y z) -(RSB x (SRA y z)) -> (RSBshiftRAreg x y z) -(RSB (SRA y z) x) -> (SUBshiftRAreg x y z) -(AND x (SLLconst [c] y)) -> (ANDshiftLL x y [c]) -(AND x (SRLconst [c] y)) -> (ANDshiftRL x y [c]) -(AND x (SRAconst [c] y)) -> (ANDshiftRA x y [c]) -(AND x (SLL y z)) -> (ANDshiftLLreg x y z) -(AND x (SRL y z)) -> (ANDshiftRLreg x y z) -(AND x (SRA y z)) -> (ANDshiftRAreg x y z) -(OR x (SLLconst [c] y)) -> (ORshiftLL x y [c]) -(OR x (SRLconst [c] y)) -> (ORshiftRL x y [c]) -(OR x (SRAconst [c] y)) -> (ORshiftRA x y [c]) -(OR x (SLL y z)) -> (ORshiftLLreg x y z) -(OR x (SRL y z)) -> (ORshiftRLreg x y z) -(OR x (SRA y z)) -> (ORshiftRAreg x y z) -(XOR x (SLLconst [c] y)) -> (XORshiftLL x y [c]) -(XOR x (SRLconst [c] y)) -> (XORshiftRL x y [c]) -(XOR x (SRAconst [c] y)) -> (XORshiftRA x y [c]) -(XOR x (SRRconst [c] y)) -> (XORshiftRR x y [c]) -(XOR x (SLL y z)) -> (XORshiftLLreg x y z) -(XOR x (SRL y z)) -> (XORshiftRLreg x y z) -(XOR x (SRA y z)) -> (XORshiftRAreg x y z) -(BIC x (SLLconst [c] y)) -> (BICshiftLL x y [c]) -(BIC x (SRLconst [c] y)) -> (BICshiftRL x y [c]) -(BIC x (SRAconst [c] y)) -> (BICshiftRA x y [c]) -(BIC x (SLL y z)) -> (BICshiftLLreg x y z) -(BIC x (SRL y z)) -> (BICshiftRLreg x y z) -(BIC x (SRA y z)) -> (BICshiftRAreg x y z) -(MVN (SLLconst [c] x)) -> (MVNshiftLL x [c]) -(MVN (SRLconst [c] x)) -> (MVNshiftRL x [c]) -(MVN (SRAconst [c] x)) -> (MVNshiftRA x [c]) -(MVN (SLL x y)) -> (MVNshiftLLreg x y) -(MVN (SRL x y)) -> (MVNshiftRLreg x y) -(MVN (SRA x y)) -> (MVNshiftRAreg x y) +(ADD x (SLLconst [c] y)) => (ADDshiftLL x y [c]) +(ADD x (SRLconst [c] y)) => (ADDshiftRL x y [c]) +(ADD x (SRAconst [c] y)) => (ADDshiftRA x y [c]) +(ADD x (SLL y z)) => (ADDshiftLLreg x y z) +(ADD x (SRL y z)) => (ADDshiftRLreg x y z) +(ADD x (SRA y z)) => (ADDshiftRAreg x y z) +(ADC x (SLLconst [c] y) flags) => (ADCshiftLL x y [c] flags) +(ADC x (SRLconst [c] y) flags) => (ADCshiftRL x y [c] flags) +(ADC x (SRAconst [c] y) flags) => (ADCshiftRA x y [c] flags) +(ADC x (SLL y z) flags) => (ADCshiftLLreg x y z flags) +(ADC x (SRL y z) flags) => (ADCshiftRLreg x y z flags) +(ADC x (SRA y z) flags) => (ADCshiftRAreg x y z flags) +(ADDS x (SLLconst [c] y)) => (ADDSshiftLL x y [c]) +(ADDS x (SRLconst [c] y)) => (ADDSshiftRL x y [c]) +(ADDS x (SRAconst [c] y)) => (ADDSshiftRA x y [c]) +(ADDS x (SLL y z)) => (ADDSshiftLLreg x y z) +(ADDS x (SRL y z)) => (ADDSshiftRLreg x y z) +(ADDS x (SRA y z)) => (ADDSshiftRAreg x y z) +(SUB x (SLLconst [c] y)) => (SUBshiftLL x y [c]) +(SUB (SLLconst [c] y) x) => (RSBshiftLL x y [c]) +(SUB x (SRLconst [c] y)) => (SUBshiftRL x y [c]) +(SUB (SRLconst [c] y) x) => (RSBshiftRL x y [c]) +(SUB x (SRAconst [c] y)) => (SUBshiftRA x y [c]) +(SUB (SRAconst [c] y) x) => (RSBshiftRA x y [c]) +(SUB x (SLL y z)) => (SUBshiftLLreg x y z) +(SUB (SLL y z) x) => (RSBshiftLLreg x y z) +(SUB x (SRL y z)) => (SUBshiftRLreg x y z) +(SUB (SRL y z) x) => (RSBshiftRLreg x y z) +(SUB x (SRA y z)) => (SUBshiftRAreg x y z) +(SUB (SRA y z) x) => (RSBshiftRAreg x y z) +(SBC x (SLLconst [c] y) flags) => (SBCshiftLL x y [c] flags) +(SBC (SLLconst [c] y) x flags) => (RSCshiftLL x y [c] flags) +(SBC x (SRLconst [c] y) flags) => (SBCshiftRL x y [c] flags) +(SBC (SRLconst [c] y) x flags) => (RSCshiftRL x y [c] flags) +(SBC x (SRAconst [c] y) flags) => (SBCshiftRA x y [c] flags) +(SBC (SRAconst [c] y) x flags) => (RSCshiftRA x y [c] flags) +(SBC x (SLL y z) flags) => (SBCshiftLLreg x y z flags) +(SBC (SLL y z) x flags) => (RSCshiftLLreg x y z flags) +(SBC x (SRL y z) flags) => (SBCshiftRLreg x y z flags) +(SBC (SRL y z) x flags) => (RSCshiftRLreg x y z flags) +(SBC x (SRA y z) flags) => (SBCshiftRAreg x y z flags) +(SBC (SRA y z) x flags) => (RSCshiftRAreg x y z flags) +(SUBS x (SLLconst [c] y)) => (SUBSshiftLL x y [c]) +(SUBS (SLLconst [c] y) x) => (RSBSshiftLL x y [c]) +(SUBS x (SRLconst [c] y)) => (SUBSshiftRL x y [c]) +(SUBS (SRLconst [c] y) x) => (RSBSshiftRL x y [c]) +(SUBS x (SRAconst [c] y)) => (SUBSshiftRA x y [c]) +(SUBS (SRAconst [c] y) x) => (RSBSshiftRA x y [c]) +(SUBS x (SLL y z)) => (SUBSshiftLLreg x y z) +(SUBS (SLL y z) x) => (RSBSshiftLLreg x y z) +(SUBS x (SRL y z)) => (SUBSshiftRLreg x y z) +(SUBS (SRL y z) x) => (RSBSshiftRLreg x y z) +(SUBS x (SRA y z)) => (SUBSshiftRAreg x y z) +(SUBS (SRA y z) x) => (RSBSshiftRAreg x y z) +(RSB x (SLLconst [c] y)) => (RSBshiftLL x y [c]) +(RSB (SLLconst [c] y) x) => (SUBshiftLL x y [c]) +(RSB x (SRLconst [c] y)) => (RSBshiftRL x y [c]) +(RSB (SRLconst [c] y) x) => (SUBshiftRL x y [c]) +(RSB x (SRAconst [c] y)) => (RSBshiftRA x y [c]) +(RSB (SRAconst [c] y) x) => (SUBshiftRA x y [c]) +(RSB x (SLL y z)) => (RSBshiftLLreg x y z) +(RSB (SLL y z) x) => (SUBshiftLLreg x y z) +(RSB x (SRL y z)) => (RSBshiftRLreg x y z) +(RSB (SRL y z) x) => (SUBshiftRLreg x y z) +(RSB x (SRA y z)) => (RSBshiftRAreg x y z) +(RSB (SRA y z) x) => (SUBshiftRAreg x y z) +(AND x (SLLconst [c] y)) => (ANDshiftLL x y [c]) +(AND x (SRLconst [c] y)) => (ANDshiftRL x y [c]) +(AND x (SRAconst [c] y)) => (ANDshiftRA x y [c]) +(AND x (SLL y z)) => (ANDshiftLLreg x y z) +(AND x (SRL y z)) => (ANDshiftRLreg x y z) +(AND x (SRA y z)) => (ANDshiftRAreg x y z) +(OR x (SLLconst [c] y)) => (ORshiftLL x y [c]) +(OR x (SRLconst [c] y)) => (ORshiftRL x y [c]) +(OR x (SRAconst [c] y)) => (ORshiftRA x y [c]) +(OR x (SLL y z)) => (ORshiftLLreg x y z) +(OR x (SRL y z)) => (ORshiftRLreg x y z) +(OR x (SRA y z)) => (ORshiftRAreg x y z) +(XOR x (SLLconst [c] y)) => (XORshiftLL x y [c]) +(XOR x (SRLconst [c] y)) => (XORshiftRL x y [c]) +(XOR x (SRAconst [c] y)) => (XORshiftRA x y [c]) +(XOR x (SRRconst [c] y)) => (XORshiftRR x y [c]) +(XOR x (SLL y z)) => (XORshiftLLreg x y z) +(XOR x (SRL y z)) => (XORshiftRLreg x y z) +(XOR x (SRA y z)) => (XORshiftRAreg x y z) +(BIC x (SLLconst [c] y)) => (BICshiftLL x y [c]) +(BIC x (SRLconst [c] y)) => (BICshiftRL x y [c]) +(BIC x (SRAconst [c] y)) => (BICshiftRA x y [c]) +(BIC x (SLL y z)) => (BICshiftLLreg x y z) +(BIC x (SRL y z)) => (BICshiftRLreg x y z) +(BIC x (SRA y z)) => (BICshiftRAreg x y z) +(MVN (SLLconst [c] x)) => (MVNshiftLL x [c]) +(MVN (SRLconst [c] x)) => (MVNshiftRL x [c]) +(MVN (SRAconst [c] x)) => (MVNshiftRA x [c]) +(MVN (SLL x y)) => (MVNshiftLLreg x y) +(MVN (SRL x y)) => (MVNshiftRLreg x y) +(MVN (SRA x y)) => (MVNshiftRAreg x y) -(CMP x (SLLconst [c] y)) -> (CMPshiftLL x y [c]) -(CMP (SLLconst [c] y) x) -> (InvertFlags (CMPshiftLL x y [c])) -(CMP x (SRLconst [c] y)) -> (CMPshiftRL x y [c]) -(CMP (SRLconst [c] y) x) -> (InvertFlags (CMPshiftRL x y [c])) -(CMP x (SRAconst [c] y)) -> (CMPshiftRA x y [c]) -(CMP (SRAconst [c] y) x) -> (InvertFlags (CMPshiftRA x y [c])) -(CMP x (SLL y z)) -> (CMPshiftLLreg x y z) -(CMP (SLL y z) x) -> (InvertFlags (CMPshiftLLreg x y z)) -(CMP x (SRL y z)) -> (CMPshiftRLreg x y z) -(CMP (SRL y z) x) -> (InvertFlags (CMPshiftRLreg x y z)) -(CMP x (SRA y z)) -> (CMPshiftRAreg x y z) -(CMP (SRA y z) x) -> (InvertFlags (CMPshiftRAreg x y z)) -(TST x (SLLconst [c] y)) -> (TSTshiftLL x y [c]) -(TST x (SRLconst [c] y)) -> (TSTshiftRL x y [c]) -(TST x (SRAconst [c] y)) -> (TSTshiftRA x y [c]) -(TST x (SLL y z)) -> (TSTshiftLLreg x y z) -(TST x (SRL y z)) -> (TSTshiftRLreg x y z) -(TST x (SRA y z)) -> (TSTshiftRAreg x y z) -(TEQ x (SLLconst [c] y)) -> (TEQshiftLL x y [c]) -(TEQ x (SRLconst [c] y)) -> (TEQshiftRL x y [c]) -(TEQ x (SRAconst [c] y)) -> (TEQshiftRA x y [c]) -(TEQ x (SLL y z)) -> (TEQshiftLLreg x y z) -(TEQ x (SRL y z)) -> (TEQshiftRLreg x y z) -(TEQ x (SRA y z)) -> (TEQshiftRAreg x y z) -(CMN x (SLLconst [c] y)) -> (CMNshiftLL x y [c]) -(CMN x (SRLconst [c] y)) -> (CMNshiftRL x y [c]) -(CMN x (SRAconst [c] y)) -> (CMNshiftRA x y [c]) -(CMN x (SLL y z)) -> (CMNshiftLLreg x y z) -(CMN x (SRL y z)) -> (CMNshiftRLreg x y z) -(CMN x (SRA y z)) -> (CMNshiftRAreg x y z) +(CMP x (SLLconst [c] y)) => (CMPshiftLL x y [c]) +(CMP (SLLconst [c] y) x) => (InvertFlags (CMPshiftLL x y [c])) +(CMP x (SRLconst [c] y)) => (CMPshiftRL x y [c]) +(CMP (SRLconst [c] y) x) => (InvertFlags (CMPshiftRL x y [c])) +(CMP x (SRAconst [c] y)) => (CMPshiftRA x y [c]) +(CMP (SRAconst [c] y) x) => (InvertFlags (CMPshiftRA x y [c])) +(CMP x (SLL y z)) => (CMPshiftLLreg x y z) +(CMP (SLL y z) x) => (InvertFlags (CMPshiftLLreg x y z)) +(CMP x (SRL y z)) => (CMPshiftRLreg x y z) +(CMP (SRL y z) x) => (InvertFlags (CMPshiftRLreg x y z)) +(CMP x (SRA y z)) => (CMPshiftRAreg x y z) +(CMP (SRA y z) x) => (InvertFlags (CMPshiftRAreg x y z)) +(TST x (SLLconst [c] y)) => (TSTshiftLL x y [c]) +(TST x (SRLconst [c] y)) => (TSTshiftRL x y [c]) +(TST x (SRAconst [c] y)) => (TSTshiftRA x y [c]) +(TST x (SLL y z)) => (TSTshiftLLreg x y z) +(TST x (SRL y z)) => (TSTshiftRLreg x y z) +(TST x (SRA y z)) => (TSTshiftRAreg x y z) +(TEQ x (SLLconst [c] y)) => (TEQshiftLL x y [c]) +(TEQ x (SRLconst [c] y)) => (TEQshiftRL x y [c]) +(TEQ x (SRAconst [c] y)) => (TEQshiftRA x y [c]) +(TEQ x (SLL y z)) => (TEQshiftLLreg x y z) +(TEQ x (SRL y z)) => (TEQshiftRLreg x y z) +(TEQ x (SRA y z)) => (TEQshiftRAreg x y z) +(CMN x (SLLconst [c] y)) => (CMNshiftLL x y [c]) +(CMN x (SRLconst [c] y)) => (CMNshiftRL x y [c]) +(CMN x (SRAconst [c] y)) => (CMNshiftRA x y [c]) +(CMN x (SLL y z)) => (CMNshiftLLreg x y z) +(CMN x (SRL y z)) => (CMNshiftRLreg x y z) +(CMN x (SRA y z)) => (CMNshiftRAreg x y z) // prefer *const ops to *shift ops -(ADDshiftLL (MOVWconst [c]) x [d]) -> (ADDconst [c] (SLLconst x [d])) -(ADDshiftRL (MOVWconst [c]) x [d]) -> (ADDconst [c] (SRLconst x [d])) -(ADDshiftRA (MOVWconst [c]) x [d]) -> (ADDconst [c] (SRAconst x [d])) -(ADCshiftLL (MOVWconst [c]) x [d] flags) -> (ADCconst [c] (SLLconst x [d]) flags) -(ADCshiftRL (MOVWconst [c]) x [d] flags) -> (ADCconst [c] (SRLconst x [d]) flags) -(ADCshiftRA (MOVWconst [c]) x [d] flags) -> (ADCconst [c] (SRAconst x [d]) flags) -(ADDSshiftLL (MOVWconst [c]) x [d]) -> (ADDSconst [c] (SLLconst x [d])) -(ADDSshiftRL (MOVWconst [c]) x [d]) -> (ADDSconst [c] (SRLconst x [d])) -(ADDSshiftRA (MOVWconst [c]) x [d]) -> (ADDSconst [c] (SRAconst x [d])) -(SUBshiftLL (MOVWconst [c]) x [d]) -> (RSBconst [c] (SLLconst x [d])) -(SUBshiftRL (MOVWconst [c]) x [d]) -> (RSBconst [c] (SRLconst x [d])) -(SUBshiftRA (MOVWconst [c]) x [d]) -> (RSBconst [c] (SRAconst x [d])) -(SBCshiftLL (MOVWconst [c]) x [d] flags) -> (RSCconst [c] (SLLconst x [d]) flags) -(SBCshiftRL (MOVWconst [c]) x [d] flags) -> (RSCconst [c] (SRLconst x [d]) flags) -(SBCshiftRA (MOVWconst [c]) x [d] flags) -> (RSCconst [c] (SRAconst x [d]) flags) -(SUBSshiftLL (MOVWconst [c]) x [d]) -> (RSBSconst [c] (SLLconst x [d])) -(SUBSshiftRL (MOVWconst [c]) x [d]) -> (RSBSconst [c] (SRLconst x [d])) -(SUBSshiftRA (MOVWconst [c]) x [d]) -> (RSBSconst [c] (SRAconst x [d])) -(RSBshiftLL (MOVWconst [c]) x [d]) -> (SUBconst [c] (SLLconst x [d])) -(RSBshiftRL (MOVWconst [c]) x [d]) -> (SUBconst [c] (SRLconst x [d])) -(RSBshiftRA (MOVWconst [c]) x [d]) -> (SUBconst [c] (SRAconst x [d])) -(RSCshiftLL (MOVWconst [c]) x [d] flags) -> (SBCconst [c] (SLLconst x [d]) flags) -(RSCshiftRL (MOVWconst [c]) x [d] flags) -> (SBCconst [c] (SRLconst x [d]) flags) -(RSCshiftRA (MOVWconst [c]) x [d] flags) -> (SBCconst [c] (SRAconst x [d]) flags) -(RSBSshiftLL (MOVWconst [c]) x [d]) -> (SUBSconst [c] (SLLconst x [d])) -(RSBSshiftRL (MOVWconst [c]) x [d]) -> (SUBSconst [c] (SRLconst x [d])) -(RSBSshiftRA (MOVWconst [c]) x [d]) -> (SUBSconst [c] (SRAconst x [d])) -(ANDshiftLL (MOVWconst [c]) x [d]) -> (ANDconst [c] (SLLconst x [d])) -(ANDshiftRL (MOVWconst [c]) x [d]) -> (ANDconst [c] (SRLconst x [d])) -(ANDshiftRA (MOVWconst [c]) x [d]) -> (ANDconst [c] (SRAconst x [d])) -(ORshiftLL (MOVWconst [c]) x [d]) -> (ORconst [c] (SLLconst x [d])) -(ORshiftRL (MOVWconst [c]) x [d]) -> (ORconst [c] (SRLconst x [d])) -(ORshiftRA (MOVWconst [c]) x [d]) -> (ORconst [c] (SRAconst x [d])) -(XORshiftLL (MOVWconst [c]) x [d]) -> (XORconst [c] (SLLconst x [d])) -(XORshiftRL (MOVWconst [c]) x [d]) -> (XORconst [c] (SRLconst x [d])) -(XORshiftRA (MOVWconst [c]) x [d]) -> (XORconst [c] (SRAconst x [d])) -(XORshiftRR (MOVWconst [c]) x [d]) -> (XORconst [c] (SRRconst x [d])) -(CMPshiftLL (MOVWconst [c]) x [d]) -> (InvertFlags (CMPconst [c] (SLLconst x [d]))) -(CMPshiftRL (MOVWconst [c]) x [d]) -> (InvertFlags (CMPconst [c] (SRLconst x [d]))) -(CMPshiftRA (MOVWconst [c]) x [d]) -> (InvertFlags (CMPconst [c] (SRAconst x [d]))) -(TSTshiftLL (MOVWconst [c]) x [d]) -> (TSTconst [c] (SLLconst x [d])) -(TSTshiftRL (MOVWconst [c]) x [d]) -> (TSTconst [c] (SRLconst x [d])) -(TSTshiftRA (MOVWconst [c]) x [d]) -> (TSTconst [c] (SRAconst x [d])) -(TEQshiftLL (MOVWconst [c]) x [d]) -> (TEQconst [c] (SLLconst x [d])) -(TEQshiftRL (MOVWconst [c]) x [d]) -> (TEQconst [c] (SRLconst x [d])) -(TEQshiftRA (MOVWconst [c]) x [d]) -> (TEQconst [c] (SRAconst x [d])) -(CMNshiftLL (MOVWconst [c]) x [d]) -> (CMNconst [c] (SLLconst x [d])) -(CMNshiftRL (MOVWconst [c]) x [d]) -> (CMNconst [c] (SRLconst x [d])) -(CMNshiftRA (MOVWconst [c]) x [d]) -> (CMNconst [c] (SRAconst x [d])) +(ADDshiftLL (MOVWconst [c]) x [d]) => (ADDconst [c] (SLLconst x [d])) +(ADDshiftRL (MOVWconst [c]) x [d]) => (ADDconst [c] (SRLconst x [d])) +(ADDshiftRA (MOVWconst [c]) x [d]) => (ADDconst [c] (SRAconst x [d])) +(ADCshiftLL (MOVWconst [c]) x [d] flags) => (ADCconst [c] (SLLconst x [d]) flags) +(ADCshiftRL (MOVWconst [c]) x [d] flags) => (ADCconst [c] (SRLconst x [d]) flags) +(ADCshiftRA (MOVWconst [c]) x [d] flags) => (ADCconst [c] (SRAconst x [d]) flags) +(ADDSshiftLL (MOVWconst [c]) x [d]) => (ADDSconst [c] (SLLconst x [d])) +(ADDSshiftRL (MOVWconst [c]) x [d]) => (ADDSconst [c] (SRLconst x [d])) +(ADDSshiftRA (MOVWconst [c]) x [d]) => (ADDSconst [c] (SRAconst x [d])) +(SUBshiftLL (MOVWconst [c]) x [d]) => (RSBconst [c] (SLLconst x [d])) +(SUBshiftRL (MOVWconst [c]) x [d]) => (RSBconst [c] (SRLconst x [d])) +(SUBshiftRA (MOVWconst [c]) x [d]) => (RSBconst [c] (SRAconst x [d])) +(SBCshiftLL (MOVWconst [c]) x [d] flags) => (RSCconst [c] (SLLconst x [d]) flags) +(SBCshiftRL (MOVWconst [c]) x [d] flags) => (RSCconst [c] (SRLconst x [d]) flags) +(SBCshiftRA (MOVWconst [c]) x [d] flags) => (RSCconst [c] (SRAconst x [d]) flags) +(SUBSshiftLL (MOVWconst [c]) x [d]) => (RSBSconst [c] (SLLconst x [d])) +(SUBSshiftRL (MOVWconst [c]) x [d]) => (RSBSconst [c] (SRLconst x [d])) +(SUBSshiftRA (MOVWconst [c]) x [d]) => (RSBSconst [c] (SRAconst x [d])) +(RSBshiftLL (MOVWconst [c]) x [d]) => (SUBconst [c] (SLLconst x [d])) +(RSBshiftRL (MOVWconst [c]) x [d]) => (SUBconst [c] (SRLconst x [d])) +(RSBshiftRA (MOVWconst [c]) x [d]) => (SUBconst [c] (SRAconst x [d])) +(RSCshiftLL (MOVWconst [c]) x [d] flags) => (SBCconst [c] (SLLconst x [d]) flags) +(RSCshiftRL (MOVWconst [c]) x [d] flags) => (SBCconst [c] (SRLconst x [d]) flags) +(RSCshiftRA (MOVWconst [c]) x [d] flags) => (SBCconst [c] (SRAconst x [d]) flags) +(RSBSshiftLL (MOVWconst [c]) x [d]) => (SUBSconst [c] (SLLconst x [d])) +(RSBSshiftRL (MOVWconst [c]) x [d]) => (SUBSconst [c] (SRLconst x [d])) +(RSBSshiftRA (MOVWconst [c]) x [d]) => (SUBSconst [c] (SRAconst x [d])) +(ANDshiftLL (MOVWconst [c]) x [d]) => (ANDconst [c] (SLLconst x [d])) +(ANDshiftRL (MOVWconst [c]) x [d]) => (ANDconst [c] (SRLconst x [d])) +(ANDshiftRA (MOVWconst [c]) x [d]) => (ANDconst [c] (SRAconst x [d])) +(ORshiftLL (MOVWconst [c]) x [d]) => (ORconst [c] (SLLconst x [d])) +(ORshiftRL (MOVWconst [c]) x [d]) => (ORconst [c] (SRLconst x [d])) +(ORshiftRA (MOVWconst [c]) x [d]) => (ORconst [c] (SRAconst x [d])) +(XORshiftLL (MOVWconst [c]) x [d]) => (XORconst [c] (SLLconst x [d])) +(XORshiftRL (MOVWconst [c]) x [d]) => (XORconst [c] (SRLconst x [d])) +(XORshiftRA (MOVWconst [c]) x [d]) => (XORconst [c] (SRAconst x [d])) +(XORshiftRR (MOVWconst [c]) x [d]) => (XORconst [c] (SRRconst x [d])) +(CMPshiftLL (MOVWconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SLLconst x [d]))) +(CMPshiftRL (MOVWconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SRLconst x [d]))) +(CMPshiftRA (MOVWconst [c]) x [d]) => (InvertFlags (CMPconst [c] (SRAconst x [d]))) +(TSTshiftLL (MOVWconst [c]) x [d]) => (TSTconst [c] (SLLconst x [d])) +(TSTshiftRL (MOVWconst [c]) x [d]) => (TSTconst [c] (SRLconst x [d])) +(TSTshiftRA (MOVWconst [c]) x [d]) => (TSTconst [c] (SRAconst x [d])) +(TEQshiftLL (MOVWconst [c]) x [d]) => (TEQconst [c] (SLLconst x [d])) +(TEQshiftRL (MOVWconst [c]) x [d]) => (TEQconst [c] (SRLconst x [d])) +(TEQshiftRA (MOVWconst [c]) x [d]) => (TEQconst [c] (SRAconst x [d])) +(CMNshiftLL (MOVWconst [c]) x [d]) => (CMNconst [c] (SLLconst x [d])) +(CMNshiftRL (MOVWconst [c]) x [d]) => (CMNconst [c] (SRLconst x [d])) +(CMNshiftRA (MOVWconst [c]) x [d]) => (CMNconst [c] (SRAconst x [d])) -(ADDshiftLLreg (MOVWconst [c]) x y) -> (ADDconst [c] (SLL x y)) -(ADDshiftRLreg (MOVWconst [c]) x y) -> (ADDconst [c] (SRL x y)) -(ADDshiftRAreg (MOVWconst [c]) x y) -> (ADDconst [c] (SRA x y)) -(ADCshiftLLreg (MOVWconst [c]) x y flags) -> (ADCconst [c] (SLL x y) flags) -(ADCshiftRLreg (MOVWconst [c]) x y flags) -> (ADCconst [c] (SRL x y) flags) -(ADCshiftRAreg (MOVWconst [c]) x y flags) -> (ADCconst [c] (SRA x y) flags) -(ADDSshiftLLreg (MOVWconst [c]) x y) -> (ADDSconst [c] (SLL x y)) -(ADDSshiftRLreg (MOVWconst [c]) x y) -> (ADDSconst [c] (SRL x y)) -(ADDSshiftRAreg (MOVWconst [c]) x y) -> (ADDSconst [c] (SRA x y)) -(SUBshiftLLreg (MOVWconst [c]) x y) -> (RSBconst [c] (SLL x y)) -(SUBshiftRLreg (MOVWconst [c]) x y) -> (RSBconst [c] (SRL x y)) -(SUBshiftRAreg (MOVWconst [c]) x y) -> (RSBconst [c] (SRA x y)) -(SBCshiftLLreg (MOVWconst [c]) x y flags) -> (RSCconst [c] (SLL x y) flags) -(SBCshiftRLreg (MOVWconst [c]) x y flags) -> (RSCconst [c] (SRL x y) flags) -(SBCshiftRAreg (MOVWconst [c]) x y flags) -> (RSCconst [c] (SRA x y) flags) -(SUBSshiftLLreg (MOVWconst [c]) x y) -> (RSBSconst [c] (SLL x y)) -(SUBSshiftRLreg (MOVWconst [c]) x y) -> (RSBSconst [c] (SRL x y)) -(SUBSshiftRAreg (MOVWconst [c]) x y) -> (RSBSconst [c] (SRA x y)) -(RSBshiftLLreg (MOVWconst [c]) x y) -> (SUBconst [c] (SLL x y)) -(RSBshiftRLreg (MOVWconst [c]) x y) -> (SUBconst [c] (SRL x y)) -(RSBshiftRAreg (MOVWconst [c]) x y) -> (SUBconst [c] (SRA x y)) -(RSCshiftLLreg (MOVWconst [c]) x y flags) -> (SBCconst [c] (SLL x y) flags) -(RSCshiftRLreg (MOVWconst [c]) x y flags) -> (SBCconst [c] (SRL x y) flags) -(RSCshiftRAreg (MOVWconst [c]) x y flags) -> (SBCconst [c] (SRA x y) flags) -(RSBSshiftLLreg (MOVWconst [c]) x y) -> (SUBSconst [c] (SLL x y)) -(RSBSshiftRLreg (MOVWconst [c]) x y) -> (SUBSconst [c] (SRL x y)) -(RSBSshiftRAreg (MOVWconst [c]) x y) -> (SUBSconst [c] (SRA x y)) -(ANDshiftLLreg (MOVWconst [c]) x y) -> (ANDconst [c] (SLL x y)) -(ANDshiftRLreg (MOVWconst [c]) x y) -> (ANDconst [c] (SRL x y)) -(ANDshiftRAreg (MOVWconst [c]) x y) -> (ANDconst [c] (SRA x y)) -(ORshiftLLreg (MOVWconst [c]) x y) -> (ORconst [c] (SLL x y)) -(ORshiftRLreg (MOVWconst [c]) x y) -> (ORconst [c] (SRL x y)) -(ORshiftRAreg (MOVWconst [c]) x y) -> (ORconst [c] (SRA x y)) -(XORshiftLLreg (MOVWconst [c]) x y) -> (XORconst [c] (SLL x y)) -(XORshiftRLreg (MOVWconst [c]) x y) -> (XORconst [c] (SRL x y)) -(XORshiftRAreg (MOVWconst [c]) x y) -> (XORconst [c] (SRA x y)) -(CMPshiftLLreg (MOVWconst [c]) x y) -> (InvertFlags (CMPconst [c] (SLL x y))) -(CMPshiftRLreg (MOVWconst [c]) x y) -> (InvertFlags (CMPconst [c] (SRL x y))) -(CMPshiftRAreg (MOVWconst [c]) x y) -> (InvertFlags (CMPconst [c] (SRA x y))) -(TSTshiftLLreg (MOVWconst [c]) x y) -> (TSTconst [c] (SLL x y)) -(TSTshiftRLreg (MOVWconst [c]) x y) -> (TSTconst [c] (SRL x y)) -(TSTshiftRAreg (MOVWconst [c]) x y) -> (TSTconst [c] (SRA x y)) -(TEQshiftLLreg (MOVWconst [c]) x y) -> (TEQconst [c] (SLL x y)) -(TEQshiftRLreg (MOVWconst [c]) x y) -> (TEQconst [c] (SRL x y)) -(TEQshiftRAreg (MOVWconst [c]) x y) -> (TEQconst [c] (SRA x y)) -(CMNshiftLLreg (MOVWconst [c]) x y) -> (CMNconst [c] (SLL x y)) -(CMNshiftRLreg (MOVWconst [c]) x y) -> (CMNconst [c] (SRL x y)) -(CMNshiftRAreg (MOVWconst [c]) x y) -> (CMNconst [c] (SRA x y)) +(ADDshiftLLreg (MOVWconst [c]) x y) => (ADDconst [c] (SLL x y)) +(ADDshiftRLreg (MOVWconst [c]) x y) => (ADDconst [c] (SRL x y)) +(ADDshiftRAreg (MOVWconst [c]) x y) => (ADDconst [c] (SRA x y)) +(ADCshiftLLreg (MOVWconst [c]) x y flags) => (ADCconst [c] (SLL x y) flags) +(ADCshiftRLreg (MOVWconst [c]) x y flags) => (ADCconst [c] (SRL x y) flags) +(ADCshiftRAreg (MOVWconst [c]) x y flags) => (ADCconst [c] (SRA x y) flags) +(ADDSshiftLLreg (MOVWconst [c]) x y) => (ADDSconst [c] (SLL x y)) +(ADDSshiftRLreg (MOVWconst [c]) x y) => (ADDSconst [c] (SRL x y)) +(ADDSshiftRAreg (MOVWconst [c]) x y) => (ADDSconst [c] (SRA x y)) +(SUBshiftLLreg (MOVWconst [c]) x y) => (RSBconst [c] (SLL x y)) +(SUBshiftRLreg (MOVWconst [c]) x y) => (RSBconst [c] (SRL x y)) +(SUBshiftRAreg (MOVWconst [c]) x y) => (RSBconst [c] (SRA x y)) +(SBCshiftLLreg (MOVWconst [c]) x y flags) => (RSCconst [c] (SLL x y) flags) +(SBCshiftRLreg (MOVWconst [c]) x y flags) => (RSCconst [c] (SRL x y) flags) +(SBCshiftRAreg (MOVWconst [c]) x y flags) => (RSCconst [c] (SRA x y) flags) +(SUBSshiftLLreg (MOVWconst [c]) x y) => (RSBSconst [c] (SLL x y)) +(SUBSshiftRLreg (MOVWconst [c]) x y) => (RSBSconst [c] (SRL x y)) +(SUBSshiftRAreg (MOVWconst [c]) x y) => (RSBSconst [c] (SRA x y)) +(RSBshiftLLreg (MOVWconst [c]) x y) => (SUBconst [c] (SLL x y)) +(RSBshiftRLreg (MOVWconst [c]) x y) => (SUBconst [c] (SRL x y)) +(RSBshiftRAreg (MOVWconst [c]) x y) => (SUBconst [c] (SRA x y)) +(RSCshiftLLreg (MOVWconst [c]) x y flags) => (SBCconst [c] (SLL x y) flags) +(RSCshiftRLreg (MOVWconst [c]) x y flags) => (SBCconst [c] (SRL x y) flags) +(RSCshiftRAreg (MOVWconst [c]) x y flags) => (SBCconst [c] (SRA x y) flags) +(RSBSshiftLLreg (MOVWconst [c]) x y) => (SUBSconst [c] (SLL x y)) +(RSBSshiftRLreg (MOVWconst [c]) x y) => (SUBSconst [c] (SRL x y)) +(RSBSshiftRAreg (MOVWconst [c]) x y) => (SUBSconst [c] (SRA x y)) +(ANDshiftLLreg (MOVWconst [c]) x y) => (ANDconst [c] (SLL x y)) +(ANDshiftRLreg (MOVWconst [c]) x y) => (ANDconst [c] (SRL x y)) +(ANDshiftRAreg (MOVWconst [c]) x y) => (ANDconst [c] (SRA x y)) +(ORshiftLLreg (MOVWconst [c]) x y) => (ORconst [c] (SLL x y)) +(ORshiftRLreg (MOVWconst [c]) x y) => (ORconst [c] (SRL x y)) +(ORshiftRAreg (MOVWconst [c]) x y) => (ORconst [c] (SRA x y)) +(XORshiftLLreg (MOVWconst [c]) x y) => (XORconst [c] (SLL x y)) +(XORshiftRLreg (MOVWconst [c]) x y) => (XORconst [c] (SRL x y)) +(XORshiftRAreg (MOVWconst [c]) x y) => (XORconst [c] (SRA x y)) +(CMPshiftLLreg (MOVWconst [c]) x y) => (InvertFlags (CMPconst [c] (SLL x y))) +(CMPshiftRLreg (MOVWconst [c]) x y) => (InvertFlags (CMPconst [c] (SRL x y))) +(CMPshiftRAreg (MOVWconst [c]) x y) => (InvertFlags (CMPconst [c] (SRA x y))) +(TSTshiftLLreg (MOVWconst [c]) x y) => (TSTconst [c] (SLL x y)) +(TSTshiftRLreg (MOVWconst [c]) x y) => (TSTconst [c] (SRL x y)) +(TSTshiftRAreg (MOVWconst [c]) x y) => (TSTconst [c] (SRA x y)) +(TEQshiftLLreg (MOVWconst [c]) x y) => (TEQconst [c] (SLL x y)) +(TEQshiftRLreg (MOVWconst [c]) x y) => (TEQconst [c] (SRL x y)) +(TEQshiftRAreg (MOVWconst [c]) x y) => (TEQconst [c] (SRA x y)) +(CMNshiftLLreg (MOVWconst [c]) x y) => (CMNconst [c] (SLL x y)) +(CMNshiftRLreg (MOVWconst [c]) x y) => (CMNconst [c] (SRL x y)) +(CMNshiftRAreg (MOVWconst [c]) x y) => (CMNconst [c] (SRA x y)) // constant folding in *shift ops -(ADDshiftLL x (MOVWconst [c]) [d]) -> (ADDconst x [int64(int32(uint32(c)< (ADDconst x [int64(int32(uint32(c)>>uint64(d)))]) -(ADDshiftRA x (MOVWconst [c]) [d]) -> (ADDconst x [int64(int32(c)>>uint64(d))]) -(ADCshiftLL x (MOVWconst [c]) [d] flags) -> (ADCconst x [int64(int32(uint32(c)< (ADCconst x [int64(int32(uint32(c)>>uint64(d)))] flags) -(ADCshiftRA x (MOVWconst [c]) [d] flags) -> (ADCconst x [int64(int32(c)>>uint64(d))] flags) -(ADDSshiftLL x (MOVWconst [c]) [d]) -> (ADDSconst x [int64(int32(uint32(c)< (ADDSconst x [int64(int32(uint32(c)>>uint64(d)))]) -(ADDSshiftRA x (MOVWconst [c]) [d]) -> (ADDSconst x [int64(int32(c)>>uint64(d))]) -(SUBshiftLL x (MOVWconst [c]) [d]) -> (SUBconst x [int64(int32(uint32(c)< (SUBconst x [int64(int32(uint32(c)>>uint64(d)))]) -(SUBshiftRA x (MOVWconst [c]) [d]) -> (SUBconst x [int64(int32(c)>>uint64(d))]) -(SBCshiftLL x (MOVWconst [c]) [d] flags) -> (SBCconst x [int64(int32(uint32(c)< (SBCconst x [int64(int32(uint32(c)>>uint64(d)))] flags) -(SBCshiftRA x (MOVWconst [c]) [d] flags) -> (SBCconst x [int64(int32(c)>>uint64(d))] flags) -(SUBSshiftLL x (MOVWconst [c]) [d]) -> (SUBSconst x [int64(int32(uint32(c)< (SUBSconst x [int64(int32(uint32(c)>>uint64(d)))]) -(SUBSshiftRA x (MOVWconst [c]) [d]) -> (SUBSconst x [int64(int32(c)>>uint64(d))]) -(RSBshiftLL x (MOVWconst [c]) [d]) -> (RSBconst x [int64(int32(uint32(c)< (RSBconst x [int64(int32(uint32(c)>>uint64(d)))]) -(RSBshiftRA x (MOVWconst [c]) [d]) -> (RSBconst x [int64(int32(c)>>uint64(d))]) -(RSCshiftLL x (MOVWconst [c]) [d] flags) -> (RSCconst x [int64(int32(uint32(c)< (RSCconst x [int64(int32(uint32(c)>>uint64(d)))] flags) -(RSCshiftRA x (MOVWconst [c]) [d] flags) -> (RSCconst x [int64(int32(c)>>uint64(d))] flags) -(RSBSshiftLL x (MOVWconst [c]) [d]) -> (RSBSconst x [int64(int32(uint32(c)< (RSBSconst x [int64(int32(uint32(c)>>uint64(d)))]) -(RSBSshiftRA x (MOVWconst [c]) [d]) -> (RSBSconst x [int64(int32(c)>>uint64(d))]) -(ANDshiftLL x (MOVWconst [c]) [d]) -> (ANDconst x [int64(int32(uint32(c)< (ANDconst x [int64(int32(uint32(c)>>uint64(d)))]) -(ANDshiftRA x (MOVWconst [c]) [d]) -> (ANDconst x [int64(int32(c)>>uint64(d))]) -(ORshiftLL x (MOVWconst [c]) [d]) -> (ORconst x [int64(int32(uint32(c)< (ORconst x [int64(int32(uint32(c)>>uint64(d)))]) -(ORshiftRA x (MOVWconst [c]) [d]) -> (ORconst x [int64(int32(c)>>uint64(d))]) -(XORshiftLL x (MOVWconst [c]) [d]) -> (XORconst x [int64(int32(uint32(c)< (XORconst x [int64(int32(uint32(c)>>uint64(d)))]) -(XORshiftRA x (MOVWconst [c]) [d]) -> (XORconst x [int64(int32(c)>>uint64(d))]) -(XORshiftRR x (MOVWconst [c]) [d]) -> (XORconst x [int64(int32(uint32(c)>>uint64(d)|uint32(c)< (BICconst x [int64(int32(uint32(c)< (BICconst x [int64(int32(uint32(c)>>uint64(d)))]) -(BICshiftRA x (MOVWconst [c]) [d]) -> (BICconst x [int64(int32(c)>>uint64(d))]) -(MVNshiftLL (MOVWconst [c]) [d]) -> (MOVWconst [^int64(uint32(c)< (MOVWconst [^int64(uint32(c)>>uint64(d))]) -(MVNshiftRA (MOVWconst [c]) [d]) -> (MOVWconst [^int64(int32(c)>>uint64(d))]) -(CMPshiftLL x (MOVWconst [c]) [d]) -> (CMPconst x [int64(int32(uint32(c)< (CMPconst x [int64(int32(uint32(c)>>uint64(d)))]) -(CMPshiftRA x (MOVWconst [c]) [d]) -> (CMPconst x [int64(int32(c)>>uint64(d))]) -(TSTshiftLL x (MOVWconst [c]) [d]) -> (TSTconst x [int64(int32(uint32(c)< (TSTconst x [int64(int32(uint32(c)>>uint64(d)))]) -(TSTshiftRA x (MOVWconst [c]) [d]) -> (TSTconst x [int64(int32(c)>>uint64(d))]) -(TEQshiftLL x (MOVWconst [c]) [d]) -> (TEQconst x [int64(int32(uint32(c)< (TEQconst x [int64(int32(uint32(c)>>uint64(d)))]) -(TEQshiftRA x (MOVWconst [c]) [d]) -> (TEQconst x [int64(int32(c)>>uint64(d))]) -(CMNshiftLL x (MOVWconst [c]) [d]) -> (CMNconst x [int64(int32(uint32(c)< (CMNconst x [int64(int32(uint32(c)>>uint64(d)))]) -(CMNshiftRA x (MOVWconst [c]) [d]) -> (CMNconst x [int64(int32(c)>>uint64(d))]) +(ADDshiftLL x (MOVWconst [c]) [d]) => (ADDconst x [c< (ADDconst x [int32(uint32(c)>>uint64(d))]) +(ADDshiftRA x (MOVWconst [c]) [d]) => (ADDconst x [c>>uint64(d)]) +(ADCshiftLL x (MOVWconst [c]) [d] flags) => (ADCconst x [c< (ADCconst x [int32(uint32(c)>>uint64(d))] flags) +(ADCshiftRA x (MOVWconst [c]) [d] flags) => (ADCconst x [c>>uint64(d)] flags) +(ADDSshiftLL x (MOVWconst [c]) [d]) => (ADDSconst x [c< (ADDSconst x [int32(uint32(c)>>uint64(d))]) +(ADDSshiftRA x (MOVWconst [c]) [d]) => (ADDSconst x [c>>uint64(d)]) +(SUBshiftLL x (MOVWconst [c]) [d]) => (SUBconst x [c< (SUBconst x [int32(uint32(c)>>uint64(d))]) +(SUBshiftRA x (MOVWconst [c]) [d]) => (SUBconst x [c>>uint64(d)]) +(SBCshiftLL x (MOVWconst [c]) [d] flags) => (SBCconst x [c< (SBCconst x [int32(uint32(c)>>uint64(d))] flags) +(SBCshiftRA x (MOVWconst [c]) [d] flags) => (SBCconst x [c>>uint64(d)] flags) +(SUBSshiftLL x (MOVWconst [c]) [d]) => (SUBSconst x [c< (SUBSconst x [int32(uint32(c)>>uint64(d))]) +(SUBSshiftRA x (MOVWconst [c]) [d]) => (SUBSconst x [c>>uint64(d)]) +(RSBshiftLL x (MOVWconst [c]) [d]) => (RSBconst x [c< (RSBconst x [int32(uint32(c)>>uint64(d))]) +(RSBshiftRA x (MOVWconst [c]) [d]) => (RSBconst x [c>>uint64(d)]) +(RSCshiftLL x (MOVWconst [c]) [d] flags) => (RSCconst x [c< (RSCconst x [int32(uint32(c)>>uint64(d))] flags) +(RSCshiftRA x (MOVWconst [c]) [d] flags) => (RSCconst x [c>>uint64(d)] flags) +(RSBSshiftLL x (MOVWconst [c]) [d]) => (RSBSconst x [c< (RSBSconst x [int32(uint32(c)>>uint64(d))]) +(RSBSshiftRA x (MOVWconst [c]) [d]) => (RSBSconst x [c>>uint64(d)]) +(ANDshiftLL x (MOVWconst [c]) [d]) => (ANDconst x [c< (ANDconst x [int32(uint32(c)>>uint64(d))]) +(ANDshiftRA x (MOVWconst [c]) [d]) => (ANDconst x [c>>uint64(d)]) +(ORshiftLL x (MOVWconst [c]) [d]) => (ORconst x [c< (ORconst x [int32(uint32(c)>>uint64(d))]) +(ORshiftRA x (MOVWconst [c]) [d]) => (ORconst x [c>>uint64(d)]) +(XORshiftLL x (MOVWconst [c]) [d]) => (XORconst x [c< (XORconst x [int32(uint32(c)>>uint64(d))]) +(XORshiftRA x (MOVWconst [c]) [d]) => (XORconst x [c>>uint64(d)]) +(XORshiftRR x (MOVWconst [c]) [d]) => (XORconst x [int32(uint32(c)>>uint64(d)|uint32(c)< (BICconst x [c< (BICconst x [int32(uint32(c)>>uint64(d))]) +(BICshiftRA x (MOVWconst [c]) [d]) => (BICconst x [c>>uint64(d)]) +(MVNshiftLL (MOVWconst [c]) [d]) => (MOVWconst [^(c< (MOVWconst [^int32(uint32(c)>>uint64(d))]) +(MVNshiftRA (MOVWconst [c]) [d]) => (MOVWconst [int32(c)>>uint64(d)]) +(CMPshiftLL x (MOVWconst [c]) [d]) => (CMPconst x [c< (CMPconst x [int32(uint32(c)>>uint64(d))]) +(CMPshiftRA x (MOVWconst [c]) [d]) => (CMPconst x [c>>uint64(d)]) +(TSTshiftLL x (MOVWconst [c]) [d]) => (TSTconst x [c< (TSTconst x [int32(uint32(c)>>uint64(d))]) +(TSTshiftRA x (MOVWconst [c]) [d]) => (TSTconst x [c>>uint64(d)]) +(TEQshiftLL x (MOVWconst [c]) [d]) => (TEQconst x [c< (TEQconst x [int32(uint32(c)>>uint64(d))]) +(TEQshiftRA x (MOVWconst [c]) [d]) => (TEQconst x [c>>uint64(d)]) +(CMNshiftLL x (MOVWconst [c]) [d]) => (CMNconst x [c< (CMNconst x [int32(uint32(c)>>uint64(d))]) +(CMNshiftRA x (MOVWconst [c]) [d]) => (CMNconst x [c>>uint64(d)]) -(ADDshiftLLreg x y (MOVWconst [c])) -> (ADDshiftLL x y [c]) -(ADDshiftRLreg x y (MOVWconst [c])) -> (ADDshiftRL x y [c]) -(ADDshiftRAreg x y (MOVWconst [c])) -> (ADDshiftRA x y [c]) -(ADCshiftLLreg x y (MOVWconst [c]) flags) -> (ADCshiftLL x y [c] flags) -(ADCshiftRLreg x y (MOVWconst [c]) flags) -> (ADCshiftRL x y [c] flags) -(ADCshiftRAreg x y (MOVWconst [c]) flags) -> (ADCshiftRA x y [c] flags) -(ADDSshiftLLreg x y (MOVWconst [c])) -> (ADDSshiftLL x y [c]) -(ADDSshiftRLreg x y (MOVWconst [c])) -> (ADDSshiftRL x y [c]) -(ADDSshiftRAreg x y (MOVWconst [c])) -> (ADDSshiftRA x y [c]) -(SUBshiftLLreg x y (MOVWconst [c])) -> (SUBshiftLL x y [c]) -(SUBshiftRLreg x y (MOVWconst [c])) -> (SUBshiftRL x y [c]) -(SUBshiftRAreg x y (MOVWconst [c])) -> (SUBshiftRA x y [c]) -(SBCshiftLLreg x y (MOVWconst [c]) flags) -> (SBCshiftLL x y [c] flags) -(SBCshiftRLreg x y (MOVWconst [c]) flags) -> (SBCshiftRL x y [c] flags) -(SBCshiftRAreg x y (MOVWconst [c]) flags) -> (SBCshiftRA x y [c] flags) -(SUBSshiftLLreg x y (MOVWconst [c])) -> (SUBSshiftLL x y [c]) -(SUBSshiftRLreg x y (MOVWconst [c])) -> (SUBSshiftRL x y [c]) -(SUBSshiftRAreg x y (MOVWconst [c])) -> (SUBSshiftRA x y [c]) -(RSBshiftLLreg x y (MOVWconst [c])) -> (RSBshiftLL x y [c]) -(RSBshiftRLreg x y (MOVWconst [c])) -> (RSBshiftRL x y [c]) -(RSBshiftRAreg x y (MOVWconst [c])) -> (RSBshiftRA x y [c]) -(RSCshiftLLreg x y (MOVWconst [c]) flags) -> (RSCshiftLL x y [c] flags) -(RSCshiftRLreg x y (MOVWconst [c]) flags) -> (RSCshiftRL x y [c] flags) -(RSCshiftRAreg x y (MOVWconst [c]) flags) -> (RSCshiftRA x y [c] flags) -(RSBSshiftLLreg x y (MOVWconst [c])) -> (RSBSshiftLL x y [c]) -(RSBSshiftRLreg x y (MOVWconst [c])) -> (RSBSshiftRL x y [c]) -(RSBSshiftRAreg x y (MOVWconst [c])) -> (RSBSshiftRA x y [c]) -(ANDshiftLLreg x y (MOVWconst [c])) -> (ANDshiftLL x y [c]) -(ANDshiftRLreg x y (MOVWconst [c])) -> (ANDshiftRL x y [c]) -(ANDshiftRAreg x y (MOVWconst [c])) -> (ANDshiftRA x y [c]) -(ORshiftLLreg x y (MOVWconst [c])) -> (ORshiftLL x y [c]) -(ORshiftRLreg x y (MOVWconst [c])) -> (ORshiftRL x y [c]) -(ORshiftRAreg x y (MOVWconst [c])) -> (ORshiftRA x y [c]) -(XORshiftLLreg x y (MOVWconst [c])) -> (XORshiftLL x y [c]) -(XORshiftRLreg x y (MOVWconst [c])) -> (XORshiftRL x y [c]) -(XORshiftRAreg x y (MOVWconst [c])) -> (XORshiftRA x y [c]) -(BICshiftLLreg x y (MOVWconst [c])) -> (BICshiftLL x y [c]) -(BICshiftRLreg x y (MOVWconst [c])) -> (BICshiftRL x y [c]) -(BICshiftRAreg x y (MOVWconst [c])) -> (BICshiftRA x y [c]) -(MVNshiftLLreg x (MOVWconst [c])) -> (MVNshiftLL x [c]) -(MVNshiftRLreg x (MOVWconst [c])) -> (MVNshiftRL x [c]) -(MVNshiftRAreg x (MOVWconst [c])) -> (MVNshiftRA x [c]) -(CMPshiftLLreg x y (MOVWconst [c])) -> (CMPshiftLL x y [c]) -(CMPshiftRLreg x y (MOVWconst [c])) -> (CMPshiftRL x y [c]) -(CMPshiftRAreg x y (MOVWconst [c])) -> (CMPshiftRA x y [c]) -(TSTshiftLLreg x y (MOVWconst [c])) -> (TSTshiftLL x y [c]) -(TSTshiftRLreg x y (MOVWconst [c])) -> (TSTshiftRL x y [c]) -(TSTshiftRAreg x y (MOVWconst [c])) -> (TSTshiftRA x y [c]) -(TEQshiftLLreg x y (MOVWconst [c])) -> (TEQshiftLL x y [c]) -(TEQshiftRLreg x y (MOVWconst [c])) -> (TEQshiftRL x y [c]) -(TEQshiftRAreg x y (MOVWconst [c])) -> (TEQshiftRA x y [c]) -(CMNshiftLLreg x y (MOVWconst [c])) -> (CMNshiftLL x y [c]) -(CMNshiftRLreg x y (MOVWconst [c])) -> (CMNshiftRL x y [c]) -(CMNshiftRAreg x y (MOVWconst [c])) -> (CMNshiftRA x y [c]) +(ADDshiftLLreg x y (MOVWconst [c])) => (ADDshiftLL x y [c]) +(ADDshiftRLreg x y (MOVWconst [c])) => (ADDshiftRL x y [c]) +(ADDshiftRAreg x y (MOVWconst [c])) => (ADDshiftRA x y [c]) +(ADCshiftLLreg x y (MOVWconst [c]) flags) => (ADCshiftLL x y [c] flags) +(ADCshiftRLreg x y (MOVWconst [c]) flags) => (ADCshiftRL x y [c] flags) +(ADCshiftRAreg x y (MOVWconst [c]) flags) => (ADCshiftRA x y [c] flags) +(ADDSshiftLLreg x y (MOVWconst [c])) => (ADDSshiftLL x y [c]) +(ADDSshiftRLreg x y (MOVWconst [c])) => (ADDSshiftRL x y [c]) +(ADDSshiftRAreg x y (MOVWconst [c])) => (ADDSshiftRA x y [c]) +(SUBshiftLLreg x y (MOVWconst [c])) => (SUBshiftLL x y [c]) +(SUBshiftRLreg x y (MOVWconst [c])) => (SUBshiftRL x y [c]) +(SUBshiftRAreg x y (MOVWconst [c])) => (SUBshiftRA x y [c]) +(SBCshiftLLreg x y (MOVWconst [c]) flags) => (SBCshiftLL x y [c] flags) +(SBCshiftRLreg x y (MOVWconst [c]) flags) => (SBCshiftRL x y [c] flags) +(SBCshiftRAreg x y (MOVWconst [c]) flags) => (SBCshiftRA x y [c] flags) +(SUBSshiftLLreg x y (MOVWconst [c])) => (SUBSshiftLL x y [c]) +(SUBSshiftRLreg x y (MOVWconst [c])) => (SUBSshiftRL x y [c]) +(SUBSshiftRAreg x y (MOVWconst [c])) => (SUBSshiftRA x y [c]) +(RSBshiftLLreg x y (MOVWconst [c])) => (RSBshiftLL x y [c]) +(RSBshiftRLreg x y (MOVWconst [c])) => (RSBshiftRL x y [c]) +(RSBshiftRAreg x y (MOVWconst [c])) => (RSBshiftRA x y [c]) +(RSCshiftLLreg x y (MOVWconst [c]) flags) => (RSCshiftLL x y [c] flags) +(RSCshiftRLreg x y (MOVWconst [c]) flags) => (RSCshiftRL x y [c] flags) +(RSCshiftRAreg x y (MOVWconst [c]) flags) => (RSCshiftRA x y [c] flags) +(RSBSshiftLLreg x y (MOVWconst [c])) => (RSBSshiftLL x y [c]) +(RSBSshiftRLreg x y (MOVWconst [c])) => (RSBSshiftRL x y [c]) +(RSBSshiftRAreg x y (MOVWconst [c])) => (RSBSshiftRA x y [c]) +(ANDshiftLLreg x y (MOVWconst [c])) => (ANDshiftLL x y [c]) +(ANDshiftRLreg x y (MOVWconst [c])) => (ANDshiftRL x y [c]) +(ANDshiftRAreg x y (MOVWconst [c])) => (ANDshiftRA x y [c]) +(ORshiftLLreg x y (MOVWconst [c])) => (ORshiftLL x y [c]) +(ORshiftRLreg x y (MOVWconst [c])) => (ORshiftRL x y [c]) +(ORshiftRAreg x y (MOVWconst [c])) => (ORshiftRA x y [c]) +(XORshiftLLreg x y (MOVWconst [c])) => (XORshiftLL x y [c]) +(XORshiftRLreg x y (MOVWconst [c])) => (XORshiftRL x y [c]) +(XORshiftRAreg x y (MOVWconst [c])) => (XORshiftRA x y [c]) +(BICshiftLLreg x y (MOVWconst [c])) => (BICshiftLL x y [c]) +(BICshiftRLreg x y (MOVWconst [c])) => (BICshiftRL x y [c]) +(BICshiftRAreg x y (MOVWconst [c])) => (BICshiftRA x y [c]) +(MVNshiftLLreg x (MOVWconst [c])) => (MVNshiftLL x [c]) +(MVNshiftRLreg x (MOVWconst [c])) => (MVNshiftRL x [c]) +(MVNshiftRAreg x (MOVWconst [c])) => (MVNshiftRA x [c]) +(CMPshiftLLreg x y (MOVWconst [c])) => (CMPshiftLL x y [c]) +(CMPshiftRLreg x y (MOVWconst [c])) => (CMPshiftRL x y [c]) +(CMPshiftRAreg x y (MOVWconst [c])) => (CMPshiftRA x y [c]) +(TSTshiftLLreg x y (MOVWconst [c])) => (TSTshiftLL x y [c]) +(TSTshiftRLreg x y (MOVWconst [c])) => (TSTshiftRL x y [c]) +(TSTshiftRAreg x y (MOVWconst [c])) => (TSTshiftRA x y [c]) +(TEQshiftLLreg x y (MOVWconst [c])) => (TEQshiftLL x y [c]) +(TEQshiftRLreg x y (MOVWconst [c])) => (TEQshiftRL x y [c]) +(TEQshiftRAreg x y (MOVWconst [c])) => (TEQshiftRA x y [c]) +(CMNshiftLLreg x y (MOVWconst [c])) => (CMNshiftLL x y [c]) +(CMNshiftRLreg x y (MOVWconst [c])) => (CMNshiftRL x y [c]) +(CMNshiftRAreg x y (MOVWconst [c])) => (CMNshiftRA x y [c]) // Generate rotates -(ADDshiftLL [c] (SRLconst x [32-c]) x) -> (SRRconst [32-c] x) -( ORshiftLL [c] (SRLconst x [32-c]) x) -> (SRRconst [32-c] x) -(XORshiftLL [c] (SRLconst x [32-c]) x) -> (SRRconst [32-c] x) -(ADDshiftRL [c] (SLLconst x [32-c]) x) -> (SRRconst [ c] x) -( ORshiftRL [c] (SLLconst x [32-c]) x) -> (SRRconst [ c] x) -(XORshiftRL [c] (SLLconst x [32-c]) x) -> (SRRconst [ c] x) +(ADDshiftLL [c] (SRLconst x [32-c]) x) => (SRRconst [32-c] x) +( ORshiftLL [c] (SRLconst x [32-c]) x) => (SRRconst [32-c] x) +(XORshiftLL [c] (SRLconst x [32-c]) x) => (SRRconst [32-c] x) +(ADDshiftRL [c] (SLLconst x [32-c]) x) => (SRRconst [ c] x) +( ORshiftRL [c] (SLLconst x [32-c]) x) => (SRRconst [ c] x) +(XORshiftRL [c] (SLLconst x [32-c]) x) => (SRRconst [ c] x) -(RotateLeft32 x (MOVWconst [c])) -> (SRRconst [-c&31] x) -(RotateLeft16 x (MOVWconst [c])) -> (Or16 (Lsh16x32 x (MOVWconst [c&15])) (Rsh16Ux32 x (MOVWconst [-c&15]))) -(RotateLeft8 x (MOVWconst [c])) -> (Or8 (Lsh8x32 x (MOVWconst [c&7])) (Rsh8Ux32 x (MOVWconst [-c&7]))) -(RotateLeft32 x y) -> (SRR x (RSBconst [0] y)) +(RotateLeft32 x (MOVWconst [c])) => (SRRconst [-c&31] x) +(RotateLeft16 x (MOVWconst [c])) => (Or16 (Lsh16x32 x (MOVWconst [c&15])) (Rsh16Ux32 x (MOVWconst [-c&15]))) +(RotateLeft8 x (MOVWconst [c])) => (Or8 (Lsh8x32 x (MOVWconst [c&7])) (Rsh8Ux32 x (MOVWconst [-c&7]))) +(RotateLeft32 x y) => (SRR x (RSBconst [0] y)) // ((x>>8) | (x<<8)) -> (REV16 x), the type of x is uint16, "|" can also be "^" or "+". // UBFX instruction is supported by ARMv6T2, ARMv7 and above versions, REV16 is supported by // ARMv6 and above versions. So for ARMv6, we need to match SLLconst, SRLconst and ORshiftLL. ((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (BFXU [int32(armBFAuxInt(8, 8))] x) x) => (REV16 x) -((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (SRLconst [24] (SLLconst [16] x)) x) && objabi.GOARM>=6 -> (REV16 x) +((ADDshiftLL|ORshiftLL|XORshiftLL) [8] (SRLconst [24] (SLLconst [16] x)) x) && objabi.GOARM>=6 => (REV16 x) // use indexed loads and stores -(MOVWload [0] {sym} (ADD ptr idx) mem) && sym == nil -> (MOVWloadidx ptr idx mem) -(MOVWstore [0] {sym} (ADD ptr idx) val mem) && sym == nil -> (MOVWstoreidx ptr idx val mem) -(MOVWload [0] {sym} (ADDshiftLL ptr idx [c]) mem) && sym == nil -> (MOVWloadshiftLL ptr idx [c] mem) -(MOVWload [0] {sym} (ADDshiftRL ptr idx [c]) mem) && sym == nil -> (MOVWloadshiftRL ptr idx [c] mem) -(MOVWload [0] {sym} (ADDshiftRA ptr idx [c]) mem) && sym == nil -> (MOVWloadshiftRA ptr idx [c] mem) -(MOVWstore [0] {sym} (ADDshiftLL ptr idx [c]) val mem) && sym == nil -> (MOVWstoreshiftLL ptr idx [c] val mem) -(MOVWstore [0] {sym} (ADDshiftRL ptr idx [c]) val mem) && sym == nil -> (MOVWstoreshiftRL ptr idx [c] val mem) -(MOVWstore [0] {sym} (ADDshiftRA ptr idx [c]) val mem) && sym == nil -> (MOVWstoreshiftRA ptr idx [c] val mem) -(MOVBUload [0] {sym} (ADD ptr idx) mem) && sym == nil -> (MOVBUloadidx ptr idx mem) -(MOVBload [0] {sym} (ADD ptr idx) mem) && sym == nil -> (MOVBloadidx ptr idx mem) -(MOVBstore [0] {sym} (ADD ptr idx) val mem) && sym == nil -> (MOVBstoreidx ptr idx val mem) -(MOVHUload [0] {sym} (ADD ptr idx) mem) && sym == nil -> (MOVHUloadidx ptr idx mem) -(MOVHload [0] {sym} (ADD ptr idx) mem) && sym == nil -> (MOVHloadidx ptr idx mem) -(MOVHstore [0] {sym} (ADD ptr idx) val mem) && sym == nil -> (MOVHstoreidx ptr idx val mem) +(MOVWload [0] {sym} (ADD ptr idx) mem) && sym == nil => (MOVWloadidx ptr idx mem) +(MOVWstore [0] {sym} (ADD ptr idx) val mem) && sym == nil => (MOVWstoreidx ptr idx val mem) +(MOVWload [0] {sym} (ADDshiftLL ptr idx [c]) mem) && sym == nil => (MOVWloadshiftLL ptr idx [c] mem) +(MOVWload [0] {sym} (ADDshiftRL ptr idx [c]) mem) && sym == nil => (MOVWloadshiftRL ptr idx [c] mem) +(MOVWload [0] {sym} (ADDshiftRA ptr idx [c]) mem) && sym == nil => (MOVWloadshiftRA ptr idx [c] mem) +(MOVWstore [0] {sym} (ADDshiftLL ptr idx [c]) val mem) && sym == nil => (MOVWstoreshiftLL ptr idx [c] val mem) +(MOVWstore [0] {sym} (ADDshiftRL ptr idx [c]) val mem) && sym == nil => (MOVWstoreshiftRL ptr idx [c] val mem) +(MOVWstore [0] {sym} (ADDshiftRA ptr idx [c]) val mem) && sym == nil => (MOVWstoreshiftRA ptr idx [c] val mem) +(MOVBUload [0] {sym} (ADD ptr idx) mem) && sym == nil => (MOVBUloadidx ptr idx mem) +(MOVBload [0] {sym} (ADD ptr idx) mem) && sym == nil => (MOVBloadidx ptr idx mem) +(MOVBstore [0] {sym} (ADD ptr idx) val mem) && sym == nil => (MOVBstoreidx ptr idx val mem) +(MOVHUload [0] {sym} (ADD ptr idx) mem) && sym == nil => (MOVHUloadidx ptr idx mem) +(MOVHload [0] {sym} (ADD ptr idx) mem) && sym == nil => (MOVHloadidx ptr idx mem) +(MOVHstore [0] {sym} (ADD ptr idx) val mem) && sym == nil => (MOVHstoreidx ptr idx val mem) // constant folding in indexed loads and stores -(MOVWloadidx ptr (MOVWconst [c]) mem) -> (MOVWload [c] ptr mem) -(MOVWloadidx (MOVWconst [c]) ptr mem) -> (MOVWload [c] ptr mem) -(MOVBloadidx ptr (MOVWconst [c]) mem) -> (MOVBload [c] ptr mem) -(MOVBloadidx (MOVWconst [c]) ptr mem) -> (MOVBload [c] ptr mem) -(MOVBUloadidx ptr (MOVWconst [c]) mem) -> (MOVBUload [c] ptr mem) -(MOVBUloadidx (MOVWconst [c]) ptr mem) -> (MOVBUload [c] ptr mem) -(MOVHUloadidx ptr (MOVWconst [c]) mem) -> (MOVHUload [c] ptr mem) -(MOVHUloadidx (MOVWconst [c]) ptr mem) -> (MOVHUload [c] ptr mem) -(MOVHloadidx ptr (MOVWconst [c]) mem) -> (MOVHload [c] ptr mem) -(MOVHloadidx (MOVWconst [c]) ptr mem) -> (MOVHload [c] ptr mem) +(MOVWloadidx ptr (MOVWconst [c]) mem) => (MOVWload [c] ptr mem) +(MOVWloadidx (MOVWconst [c]) ptr mem) => (MOVWload [c] ptr mem) +(MOVBloadidx ptr (MOVWconst [c]) mem) => (MOVBload [c] ptr mem) +(MOVBloadidx (MOVWconst [c]) ptr mem) => (MOVBload [c] ptr mem) +(MOVBUloadidx ptr (MOVWconst [c]) mem) => (MOVBUload [c] ptr mem) +(MOVBUloadidx (MOVWconst [c]) ptr mem) => (MOVBUload [c] ptr mem) +(MOVHUloadidx ptr (MOVWconst [c]) mem) => (MOVHUload [c] ptr mem) +(MOVHUloadidx (MOVWconst [c]) ptr mem) => (MOVHUload [c] ptr mem) +(MOVHloadidx ptr (MOVWconst [c]) mem) => (MOVHload [c] ptr mem) +(MOVHloadidx (MOVWconst [c]) ptr mem) => (MOVHload [c] ptr mem) -(MOVWstoreidx ptr (MOVWconst [c]) val mem) -> (MOVWstore [c] ptr val mem) -(MOVWstoreidx (MOVWconst [c]) ptr val mem) -> (MOVWstore [c] ptr val mem) -(MOVBstoreidx ptr (MOVWconst [c]) val mem) -> (MOVBstore [c] ptr val mem) -(MOVBstoreidx (MOVWconst [c]) ptr val mem) -> (MOVBstore [c] ptr val mem) -(MOVHstoreidx ptr (MOVWconst [c]) val mem) -> (MOVHstore [c] ptr val mem) -(MOVHstoreidx (MOVWconst [c]) ptr val mem) -> (MOVHstore [c] ptr val mem) +(MOVWstoreidx ptr (MOVWconst [c]) val mem) => (MOVWstore [c] ptr val mem) +(MOVWstoreidx (MOVWconst [c]) ptr val mem) => (MOVWstore [c] ptr val mem) +(MOVBstoreidx ptr (MOVWconst [c]) val mem) => (MOVBstore [c] ptr val mem) +(MOVBstoreidx (MOVWconst [c]) ptr val mem) => (MOVBstore [c] ptr val mem) +(MOVHstoreidx ptr (MOVWconst [c]) val mem) => (MOVHstore [c] ptr val mem) +(MOVHstoreidx (MOVWconst [c]) ptr val mem) => (MOVHstore [c] ptr val mem) -(MOVWloadidx ptr (SLLconst idx [c]) mem) -> (MOVWloadshiftLL ptr idx [c] mem) -(MOVWloadidx (SLLconst idx [c]) ptr mem) -> (MOVWloadshiftLL ptr idx [c] mem) -(MOVWloadidx ptr (SRLconst idx [c]) mem) -> (MOVWloadshiftRL ptr idx [c] mem) -(MOVWloadidx (SRLconst idx [c]) ptr mem) -> (MOVWloadshiftRL ptr idx [c] mem) -(MOVWloadidx ptr (SRAconst idx [c]) mem) -> (MOVWloadshiftRA ptr idx [c] mem) -(MOVWloadidx (SRAconst idx [c]) ptr mem) -> (MOVWloadshiftRA ptr idx [c] mem) +(MOVWloadidx ptr (SLLconst idx [c]) mem) => (MOVWloadshiftLL ptr idx [c] mem) +(MOVWloadidx (SLLconst idx [c]) ptr mem) => (MOVWloadshiftLL ptr idx [c] mem) +(MOVWloadidx ptr (SRLconst idx [c]) mem) => (MOVWloadshiftRL ptr idx [c] mem) +(MOVWloadidx (SRLconst idx [c]) ptr mem) => (MOVWloadshiftRL ptr idx [c] mem) +(MOVWloadidx ptr (SRAconst idx [c]) mem) => (MOVWloadshiftRA ptr idx [c] mem) +(MOVWloadidx (SRAconst idx [c]) ptr mem) => (MOVWloadshiftRA ptr idx [c] mem) -(MOVWstoreidx ptr (SLLconst idx [c]) val mem) -> (MOVWstoreshiftLL ptr idx [c] val mem) -(MOVWstoreidx (SLLconst idx [c]) ptr val mem) -> (MOVWstoreshiftLL ptr idx [c] val mem) -(MOVWstoreidx ptr (SRLconst idx [c]) val mem) -> (MOVWstoreshiftRL ptr idx [c] val mem) -(MOVWstoreidx (SRLconst idx [c]) ptr val mem) -> (MOVWstoreshiftRL ptr idx [c] val mem) -(MOVWstoreidx ptr (SRAconst idx [c]) val mem) -> (MOVWstoreshiftRA ptr idx [c] val mem) -(MOVWstoreidx (SRAconst idx [c]) ptr val mem) -> (MOVWstoreshiftRA ptr idx [c] val mem) +(MOVWstoreidx ptr (SLLconst idx [c]) val mem) => (MOVWstoreshiftLL ptr idx [c] val mem) +(MOVWstoreidx (SLLconst idx [c]) ptr val mem) => (MOVWstoreshiftLL ptr idx [c] val mem) +(MOVWstoreidx ptr (SRLconst idx [c]) val mem) => (MOVWstoreshiftRL ptr idx [c] val mem) +(MOVWstoreidx (SRLconst idx [c]) ptr val mem) => (MOVWstoreshiftRL ptr idx [c] val mem) +(MOVWstoreidx ptr (SRAconst idx [c]) val mem) => (MOVWstoreshiftRA ptr idx [c] val mem) +(MOVWstoreidx (SRAconst idx [c]) ptr val mem) => (MOVWstoreshiftRA ptr idx [c] val mem) -(MOVWloadshiftLL ptr (MOVWconst [c]) [d] mem) -> (MOVWload [int64(uint32(c)< (MOVWload [int64(uint32(c)>>uint64(d))] ptr mem) -(MOVWloadshiftRA ptr (MOVWconst [c]) [d] mem) -> (MOVWload [int64(int32(c)>>uint64(d))] ptr mem) +(MOVWloadshiftLL ptr (MOVWconst [c]) [d] mem) => (MOVWload [int32(uint32(c)< (MOVWload [int32(uint32(c)>>uint64(d))] ptr mem) +(MOVWloadshiftRA ptr (MOVWconst [c]) [d] mem) => (MOVWload [c>>uint64(d)] ptr mem) -(MOVWstoreshiftLL ptr (MOVWconst [c]) [d] val mem) -> (MOVWstore [int64(uint32(c)< (MOVWstore [int64(uint32(c)>>uint64(d))] ptr val mem) -(MOVWstoreshiftRA ptr (MOVWconst [c]) [d] val mem) -> (MOVWstore [int64(int32(c)>>uint64(d))] ptr val mem) +(MOVWstoreshiftLL ptr (MOVWconst [c]) [d] val mem) => (MOVWstore [int32(uint32(c)< (MOVWstore [int32(uint32(c)>>uint64(d))] ptr val mem) +(MOVWstoreshiftRA ptr (MOVWconst [c]) [d] val mem) => (MOVWstore [c>>uint64(d)] ptr val mem) // generic simplifications -(ADD x (RSBconst [0] y)) -> (SUB x y) -(ADD (RSBconst [c] x) (RSBconst [d] y)) -> (RSBconst [c+d] (ADD x y)) -(SUB x x) -> (MOVWconst [0]) -(RSB x x) -> (MOVWconst [0]) -(AND x x) -> x -(OR x x) -> x -(XOR x x) -> (MOVWconst [0]) -(BIC x x) -> (MOVWconst [0]) +(ADD x (RSBconst [0] y)) => (SUB x y) +(ADD (RSBconst [c] x) (RSBconst [d] y)) => (RSBconst [c+d] (ADD x y)) +(SUB x x) => (MOVWconst [0]) +(RSB x x) => (MOVWconst [0]) +(AND x x) => x +(OR x x) => x +(XOR x x) => (MOVWconst [0]) +(BIC x x) => (MOVWconst [0]) -(ADD (MUL x y) a) -> (MULA x y a) -(SUB a (MUL x y)) && objabi.GOARM == 7 -> (MULS x y a) -(RSB (MUL x y) a) && objabi.GOARM == 7 -> (MULS x y a) +(ADD (MUL x y) a) => (MULA x y a) +(SUB a (MUL x y)) && objabi.GOARM == 7 => (MULS x y a) +(RSB (MUL x y) a) && objabi.GOARM == 7 => (MULS x y a) -(NEGF (MULF x y)) && objabi.GOARM >= 6 -> (NMULF x y) -(NEGD (MULD x y)) && objabi.GOARM >= 6 -> (NMULD x y) -(MULF (NEGF x) y) && objabi.GOARM >= 6 -> (NMULF x y) -(MULD (NEGD x) y) && objabi.GOARM >= 6 -> (NMULD x y) -(NMULF (NEGF x) y) -> (MULF x y) -(NMULD (NEGD x) y) -> (MULD x y) +(NEGF (MULF x y)) && objabi.GOARM >= 6 => (NMULF x y) +(NEGD (MULD x y)) && objabi.GOARM >= 6 => (NMULD x y) +(MULF (NEGF x) y) && objabi.GOARM >= 6 => (NMULF x y) +(MULD (NEGD x) y) && objabi.GOARM >= 6 => (NMULD x y) +(NMULF (NEGF x) y) => (MULF x y) +(NMULD (NEGD x) y) => (MULD x y) // the result will overwrite the addend, since they are in the same register -(ADDF a (MULF x y)) && a.Uses == 1 && objabi.GOARM >= 6 -> (MULAF a x y) -(ADDF a (NMULF x y)) && a.Uses == 1 && objabi.GOARM >= 6 -> (MULSF a x y) -(ADDD a (MULD x y)) && a.Uses == 1 && objabi.GOARM >= 6 -> (MULAD a x y) -(ADDD a (NMULD x y)) && a.Uses == 1 && objabi.GOARM >= 6 -> (MULSD a x y) -(SUBF a (MULF x y)) && a.Uses == 1 && objabi.GOARM >= 6 -> (MULSF a x y) -(SUBF a (NMULF x y)) && a.Uses == 1 && objabi.GOARM >= 6 -> (MULAF a x y) -(SUBD a (MULD x y)) && a.Uses == 1 && objabi.GOARM >= 6 -> (MULSD a x y) -(SUBD a (NMULD x y)) && a.Uses == 1 && objabi.GOARM >= 6 -> (MULAD a x y) +(ADDF a (MULF x y)) && a.Uses == 1 && objabi.GOARM >= 6 => (MULAF a x y) +(ADDF a (NMULF x y)) && a.Uses == 1 && objabi.GOARM >= 6 => (MULSF a x y) +(ADDD a (MULD x y)) && a.Uses == 1 && objabi.GOARM >= 6 => (MULAD a x y) +(ADDD a (NMULD x y)) && a.Uses == 1 && objabi.GOARM >= 6 => (MULSD a x y) +(SUBF a (MULF x y)) && a.Uses == 1 && objabi.GOARM >= 6 => (MULSF a x y) +(SUBF a (NMULF x y)) && a.Uses == 1 && objabi.GOARM >= 6 => (MULAF a x y) +(SUBD a (MULD x y)) && a.Uses == 1 && objabi.GOARM >= 6 => (MULSD a x y) +(SUBD a (NMULD x y)) && a.Uses == 1 && objabi.GOARM >= 6 => (MULAD a x y) -(AND x (MVN y)) -> (BIC x y) +(AND x (MVN y)) => (BIC x y) // simplification with *shift ops -(SUBshiftLL x (SLLconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(SUBshiftRL x (SRLconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(SUBshiftRA x (SRAconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(RSBshiftLL x (SLLconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(RSBshiftRL x (SRLconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(RSBshiftRA x (SRAconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(ANDshiftLL x y:(SLLconst x [c]) [d]) && c==d -> y -(ANDshiftRL x y:(SRLconst x [c]) [d]) && c==d -> y -(ANDshiftRA x y:(SRAconst x [c]) [d]) && c==d -> y -(ORshiftLL x y:(SLLconst x [c]) [d]) && c==d -> y -(ORshiftRL x y:(SRLconst x [c]) [d]) && c==d -> y -(ORshiftRA x y:(SRAconst x [c]) [d]) && c==d -> y -(XORshiftLL x (SLLconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(XORshiftRL x (SRLconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(XORshiftRA x (SRAconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(BICshiftLL x (SLLconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(BICshiftRL x (SRLconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(BICshiftRA x (SRAconst x [c]) [d]) && c==d -> (MOVWconst [0]) -(AND x (MVNshiftLL y [c])) -> (BICshiftLL x y [c]) -(AND x (MVNshiftRL y [c])) -> (BICshiftRL x y [c]) -(AND x (MVNshiftRA y [c])) -> (BICshiftRA x y [c]) +(SUBshiftLL x (SLLconst x [c]) [d]) && c==d => (MOVWconst [0]) +(SUBshiftRL x (SRLconst x [c]) [d]) && c==d => (MOVWconst [0]) +(SUBshiftRA x (SRAconst x [c]) [d]) && c==d => (MOVWconst [0]) +(RSBshiftLL x (SLLconst x [c]) [d]) && c==d => (MOVWconst [0]) +(RSBshiftRL x (SRLconst x [c]) [d]) && c==d => (MOVWconst [0]) +(RSBshiftRA x (SRAconst x [c]) [d]) && c==d => (MOVWconst [0]) +(ANDshiftLL x y:(SLLconst x [c]) [d]) && c==d => y +(ANDshiftRL x y:(SRLconst x [c]) [d]) && c==d => y +(ANDshiftRA x y:(SRAconst x [c]) [d]) && c==d => y +(ORshiftLL x y:(SLLconst x [c]) [d]) && c==d => y +(ORshiftRL x y:(SRLconst x [c]) [d]) && c==d => y +(ORshiftRA x y:(SRAconst x [c]) [d]) && c==d => y +(XORshiftLL x (SLLconst x [c]) [d]) && c==d => (MOVWconst [0]) +(XORshiftRL x (SRLconst x [c]) [d]) && c==d => (MOVWconst [0]) +(XORshiftRA x (SRAconst x [c]) [d]) && c==d => (MOVWconst [0]) +(BICshiftLL x (SLLconst x [c]) [d]) && c==d => (MOVWconst [0]) +(BICshiftRL x (SRLconst x [c]) [d]) && c==d => (MOVWconst [0]) +(BICshiftRA x (SRAconst x [c]) [d]) && c==d => (MOVWconst [0]) +(AND x (MVNshiftLL y [c])) => (BICshiftLL x y [c]) +(AND x (MVNshiftRL y [c])) => (BICshiftRL x y [c]) +(AND x (MVNshiftRA y [c])) => (BICshiftRA x y [c]) // floating point optimizations -(CMPF x (MOVFconst [0])) -> (CMPF0 x) -(CMPD x (MOVDconst [0])) -> (CMPD0 x) +(CMPF x (MOVFconst [0])) => (CMPF0 x) +(CMPD x (MOVDconst [0])) => (CMPD0 x) // bit extraction -(SRAconst (SLLconst x [c]) [d]) && objabi.GOARM==7 && uint64(d)>=uint64(c) && uint64(d)<=31 -> (BFX [(d-c)|(32-d)<<8] x) -(SRLconst (SLLconst x [c]) [d]) && objabi.GOARM==7 && uint64(d)>=uint64(c) && uint64(d)<=31 -> (BFXU [(d-c)|(32-d)<<8] x) +(SRAconst (SLLconst x [c]) [d]) && objabi.GOARM==7 && uint64(d)>=uint64(c) && uint64(d)<=31 => (BFX [(d-c)|(32-d)<<8] x) +(SRLconst (SLLconst x [c]) [d]) && objabi.GOARM==7 && uint64(d)>=uint64(c) && uint64(d)<=31 => (BFXU [(d-c)|(32-d)<<8] x) // comparison simplification -(CMP x (RSBconst [0] y)) -> (CMN x y) -(CMN x (RSBconst [0] y)) -> (CMP x y) -(EQ (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 -> (EQ (CMP x y) yes no) -(EQ (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 -> (EQ (CMP a (MUL x y)) yes no) -(EQ (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 -> (EQ (CMPconst [c] x) yes no) -(EQ (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 -> (EQ (CMPshiftLL x y [c]) yes no) -(EQ (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 -> (EQ (CMPshiftRL x y [c]) yes no) -(EQ (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 -> (EQ (CMPshiftRA x y [c]) yes no) -(EQ (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 -> (EQ (CMPshiftLLreg x y z) yes no) -(EQ (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 -> (EQ (CMPshiftRLreg x y z) yes no) -(EQ (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 -> (EQ (CMPshiftRAreg x y z) yes no) -(NE (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 -> (NE (CMP x y) yes no) -(NE (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 -> (NE (CMP a (MUL x y)) yes no) -(NE (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 -> (NE (CMPconst [c] x) yes no) -(NE (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 -> (NE (CMPshiftLL x y [c]) yes no) -(NE (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 -> (NE (CMPshiftRL x y [c]) yes no) -(NE (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 -> (NE (CMPshiftRA x y [c]) yes no) -(NE (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 -> (NE (CMPshiftLLreg x y z) yes no) -(NE (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 -> (NE (CMPshiftRLreg x y z) yes no) -(NE (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 -> (NE (CMPshiftRAreg x y z) yes no) -(EQ (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 -> (EQ (CMN x y) yes no) -(EQ (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 -> (EQ (CMN a (MUL x y)) yes no) -(EQ (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 -> (EQ (CMNconst [c] x) yes no) -(EQ (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 -> (EQ (CMNshiftLL x y [c]) yes no) -(EQ (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 -> (EQ (CMNshiftRL x y [c]) yes no) -(EQ (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 -> (EQ (CMNshiftRA x y [c]) yes no) -(EQ (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 -> (EQ (CMNshiftLLreg x y z) yes no) -(EQ (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 -> (EQ (CMNshiftRLreg x y z) yes no) -(EQ (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 -> (EQ (CMNshiftRAreg x y z) yes no) -(NE (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 -> (NE (CMN x y) yes no) -(NE (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 -> (NE (CMN a (MUL x y)) yes no) -(NE (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 -> (NE (CMNconst [c] x) yes no) -(NE (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 -> (NE (CMNshiftLL x y [c]) yes no) -(NE (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 -> (NE (CMNshiftRL x y [c]) yes no) -(NE (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 -> (NE (CMNshiftRA x y [c]) yes no) -(NE (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 -> (NE (CMNshiftLLreg x y z) yes no) -(NE (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 -> (NE (CMNshiftRLreg x y z) yes no) -(NE (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 -> (NE (CMNshiftRAreg x y z) yes no) -(EQ (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 -> (EQ (TST x y) yes no) -(EQ (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 -> (EQ (TSTconst [c] x) yes no) -(EQ (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 -> (EQ (TSTshiftLL x y [c]) yes no) -(EQ (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 -> (EQ (TSTshiftRL x y [c]) yes no) -(EQ (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 -> (EQ (TSTshiftRA x y [c]) yes no) -(EQ (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 -> (EQ (TSTshiftLLreg x y z) yes no) -(EQ (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 -> (EQ (TSTshiftRLreg x y z) yes no) -(EQ (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 -> (EQ (TSTshiftRAreg x y z) yes no) -(NE (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 -> (NE (TST x y) yes no) -(NE (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 -> (NE (TSTconst [c] x) yes no) -(NE (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 -> (NE (TSTshiftLL x y [c]) yes no) -(NE (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 -> (NE (TSTshiftRL x y [c]) yes no) -(NE (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 -> (NE (TSTshiftRA x y [c]) yes no) -(NE (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 -> (NE (TSTshiftLLreg x y z) yes no) -(NE (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 -> (NE (TSTshiftRLreg x y z) yes no) -(NE (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 -> (NE (TSTshiftRAreg x y z) yes no) -(EQ (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 -> (EQ (TEQ x y) yes no) -(EQ (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 -> (EQ (TEQconst [c] x) yes no) -(EQ (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 -> (EQ (TEQshiftLL x y [c]) yes no) -(EQ (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 -> (EQ (TEQshiftRL x y [c]) yes no) -(EQ (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 -> (EQ (TEQshiftRA x y [c]) yes no) -(EQ (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 -> (EQ (TEQshiftLLreg x y z) yes no) -(EQ (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 -> (EQ (TEQshiftRLreg x y z) yes no) -(EQ (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 -> (EQ (TEQshiftRAreg x y z) yes no) -(NE (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 -> (NE (TEQ x y) yes no) -(NE (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 -> (NE (TEQconst [c] x) yes no) -(NE (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 -> (NE (TEQshiftLL x y [c]) yes no) -(NE (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 -> (NE (TEQshiftRL x y [c]) yes no) -(NE (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 -> (NE (TEQshiftRA x y [c]) yes no) -(NE (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 -> (NE (TEQshiftLLreg x y z) yes no) -(NE (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 -> (NE (TEQshiftRLreg x y z) yes no) -(NE (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 -> (NE (TEQshiftRAreg x y z) yes no) -(LT (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 -> (LTnoov (CMP x y) yes no) -(LT (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 -> (LTnoov (CMP a (MUL x y)) yes no) -(LT (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 -> (LTnoov (CMPconst [c] x) yes no) -(LT (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 -> (LTnoov (CMPshiftLL x y [c]) yes no) -(LT (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 -> (LTnoov (CMPshiftRL x y [c]) yes no) -(LT (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 -> (LTnoov (CMPshiftRA x y [c]) yes no) -(LT (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 -> (LTnoov (CMPshiftLLreg x y z) yes no) -(LT (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 -> (LTnoov (CMPshiftRLreg x y z) yes no) -(LT (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 -> (LTnoov (CMPshiftRAreg x y z) yes no) -(LE (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 -> (LEnoov (CMP x y) yes no) -(LE (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 -> (LEnoov (CMP a (MUL x y)) yes no) -(LE (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 -> (LEnoov (CMPconst [c] x) yes no) -(LE (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 -> (LEnoov (CMPshiftLL x y [c]) yes no) -(LE (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 -> (LEnoov (CMPshiftRL x y [c]) yes no) -(LE (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 -> (LEnoov (CMPshiftRA x y [c]) yes no) -(LE (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 -> (LEnoov (CMPshiftLLreg x y z) yes no) -(LE (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 -> (LEnoov (CMPshiftRLreg x y z) yes no) -(LE (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 -> (LEnoov (CMPshiftRAreg x y z) yes no) -(LT (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 -> (LTnoov (CMN x y) yes no) -(LT (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 -> (LTnoov (CMN a (MUL x y)) yes no) -(LT (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 -> (LTnoov (CMNconst [c] x) yes no) -(LT (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 -> (LTnoov (CMNshiftLL x y [c]) yes no) -(LT (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 -> (LTnoov (CMNshiftRL x y [c]) yes no) -(LT (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 -> (LTnoov (CMNshiftRA x y [c]) yes no) -(LT (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 -> (LTnoov (CMNshiftLLreg x y z) yes no) -(LT (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 -> (LTnoov (CMNshiftRLreg x y z) yes no) -(LT (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 -> (LTnoov (CMNshiftRAreg x y z) yes no) -(LE (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 -> (LEnoov (CMN x y) yes no) -(LE (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 -> (LEnoov (CMN a (MUL x y)) yes no) -(LE (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 -> (LEnoov (CMNconst [c] x) yes no) -(LE (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 -> (LEnoov (CMNshiftLL x y [c]) yes no) -(LE (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 -> (LEnoov (CMNshiftRL x y [c]) yes no) -(LE (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 -> (LEnoov (CMNshiftRA x y [c]) yes no) -(LE (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 -> (LEnoov (CMNshiftLLreg x y z) yes no) -(LE (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 -> (LEnoov (CMNshiftRLreg x y z) yes no) -(LE (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 -> (LEnoov (CMNshiftRAreg x y z) yes no) -(LT (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 -> (LT (TST x y) yes no) -(LT (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 -> (LT (TSTconst [c] x) yes no) -(LT (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 -> (LT (TSTshiftLL x y [c]) yes no) -(LT (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 -> (LT (TSTshiftRL x y [c]) yes no) -(LT (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 -> (LT (TSTshiftRA x y [c]) yes no) -(LT (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 -> (LT (TSTshiftLLreg x y z) yes no) -(LT (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 -> (LT (TSTshiftRLreg x y z) yes no) -(LT (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 -> (LT (TSTshiftRAreg x y z) yes no) -(LE (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 -> (LE (TST x y) yes no) -(LE (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 -> (LE (TSTconst [c] x) yes no) -(LE (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 -> (LE (TSTshiftLL x y [c]) yes no) -(LE (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 -> (LE (TSTshiftRL x y [c]) yes no) -(LE (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 -> (LE (TSTshiftRA x y [c]) yes no) -(LE (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 -> (LE (TSTshiftLLreg x y z) yes no) -(LE (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 -> (LE (TSTshiftRLreg x y z) yes no) -(LE (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 -> (LE (TSTshiftRAreg x y z) yes no) -(LT (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 -> (LT (TEQ x y) yes no) -(LT (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 -> (LT (TEQconst [c] x) yes no) -(LT (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 -> (LT (TEQshiftLL x y [c]) yes no) -(LT (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 -> (LT (TEQshiftRL x y [c]) yes no) -(LT (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 -> (LT (TEQshiftRA x y [c]) yes no) -(LT (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 -> (LT (TEQshiftLLreg x y z) yes no) -(LT (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 -> (LT (TEQshiftRLreg x y z) yes no) -(LT (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 -> (LT (TEQshiftRAreg x y z) yes no) -(LE (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 -> (LE (TEQ x y) yes no) -(LE (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 -> (LE (TEQconst [c] x) yes no) -(LE (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 -> (LE (TEQshiftLL x y [c]) yes no) -(LE (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 -> (LE (TEQshiftRL x y [c]) yes no) -(LE (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 -> (LE (TEQshiftRA x y [c]) yes no) -(LE (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 -> (LE (TEQshiftLLreg x y z) yes no) -(LE (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 -> (LE (TEQshiftRLreg x y z) yes no) -(LE (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 -> (LE (TEQshiftRAreg x y z) yes no) -(GT (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 -> (GTnoov (CMP x y) yes no) -(GT (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 -> (GTnoov (CMP a (MUL x y)) yes no) -(GT (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 -> (GTnoov (CMPconst [c] x) yes no) -(GT (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 -> (GTnoov (CMPshiftLL x y [c]) yes no) -(GT (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 -> (GTnoov (CMPshiftRL x y [c]) yes no) -(GT (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 -> (GTnoov (CMPshiftRA x y [c]) yes no) -(GT (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 -> (GTnoov (CMPshiftLLreg x y z) yes no) -(GT (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 -> (GTnoov (CMPshiftRLreg x y z) yes no) -(GT (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 -> (GTnoov (CMPshiftRAreg x y z) yes no) -(GE (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 -> (GEnoov (CMP x y) yes no) -(GE (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 -> (GEnoov (CMP a (MUL x y)) yes no) -(GE (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 -> (GEnoov (CMPconst [c] x) yes no) -(GE (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 -> (GEnoov (CMPshiftLL x y [c]) yes no) -(GE (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 -> (GEnoov (CMPshiftRL x y [c]) yes no) -(GE (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 -> (GEnoov (CMPshiftRA x y [c]) yes no) -(GE (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 -> (GEnoov (CMPshiftLLreg x y z) yes no) -(GE (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 -> (GEnoov (CMPshiftRLreg x y z) yes no) -(GE (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 -> (GEnoov (CMPshiftRAreg x y z) yes no) -(GT (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 -> (GTnoov (CMN x y) yes no) -(GT (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 -> (GTnoov (CMNconst [c] x) yes no) -(GT (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 -> (GTnoov (CMNshiftLL x y [c]) yes no) -(GT (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 -> (GTnoov (CMNshiftRL x y [c]) yes no) -(GT (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 -> (GTnoov (CMNshiftRA x y [c]) yes no) -(GT (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 -> (GTnoov (CMNshiftLLreg x y z) yes no) -(GT (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 -> (GTnoov (CMNshiftRLreg x y z) yes no) -(GT (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 -> (GTnoov (CMNshiftRAreg x y z) yes no) -(GE (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 -> (GEnoov (CMN x y) yes no) -(GE (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 -> (GEnoov (CMN a (MUL x y)) yes no) -(GE (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 -> (GEnoov (CMNconst [c] x) yes no) -(GE (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 -> (GEnoov (CMNshiftLL x y [c]) yes no) -(GE (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 -> (GEnoov (CMNshiftRL x y [c]) yes no) -(GE (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 -> (GEnoov (CMNshiftRA x y [c]) yes no) -(GE (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 -> (GEnoov (CMNshiftLLreg x y z) yes no) -(GE (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 -> (GEnoov (CMNshiftRLreg x y z) yes no) -(GE (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 -> (GEnoov (CMNshiftRAreg x y z) yes no) -(GT (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 -> (GT (TST x y) yes no) -(GT (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 -> (GTnoov (CMN a (MUL x y)) yes no) -(GT (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 -> (GT (TSTconst [c] x) yes no) -(GT (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 -> (GT (TSTshiftLL x y [c]) yes no) -(GT (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 -> (GT (TSTshiftRL x y [c]) yes no) -(GT (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 -> (GT (TSTshiftRA x y [c]) yes no) -(GT (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 -> (GT (TSTshiftLLreg x y z) yes no) -(GT (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 -> (GT (TSTshiftRLreg x y z) yes no) -(GT (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 -> (GT (TSTshiftRAreg x y z) yes no) -(GE (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 -> (GE (TST x y) yes no) -(GE (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 -> (GE (TSTconst [c] x) yes no) -(GE (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 -> (GE (TSTshiftLL x y [c]) yes no) -(GE (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 -> (GE (TSTshiftRL x y [c]) yes no) -(GE (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 -> (GE (TSTshiftRA x y [c]) yes no) -(GE (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 -> (GE (TSTshiftLLreg x y z) yes no) -(GE (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 -> (GE (TSTshiftRLreg x y z) yes no) -(GE (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 -> (GE (TSTshiftRAreg x y z) yes no) -(GT (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 -> (GT (TEQ x y) yes no) -(GT (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 -> (GT (TEQconst [c] x) yes no) -(GT (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 -> (GT (TEQshiftLL x y [c]) yes no) -(GT (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 -> (GT (TEQshiftRL x y [c]) yes no) -(GT (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 -> (GT (TEQshiftRA x y [c]) yes no) -(GT (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 -> (GT (TEQshiftLLreg x y z) yes no) -(GT (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 -> (GT (TEQshiftRLreg x y z) yes no) -(GT (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 -> (GT (TEQshiftRAreg x y z) yes no) -(GE (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 -> (GE (TEQ x y) yes no) -(GE (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 -> (GE (TEQconst [c] x) yes no) -(GE (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 -> (GE (TEQshiftLL x y [c]) yes no) -(GE (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 -> (GE (TEQshiftRL x y [c]) yes no) -(GE (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 -> (GE (TEQshiftRA x y [c]) yes no) -(GE (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 -> (GE (TEQshiftLLreg x y z) yes no) -(GE (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 -> (GE (TEQshiftRLreg x y z) yes no) -(GE (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 -> (GE (TEQshiftRAreg x y z) yes no) +((LT|LE|EQ|NE|GE|GT) (CMP x (RSBconst [0] y))) => ((LT|LE|EQ|NE|GE|GT) (CMN x y)) // sense of carry bit not preserved +((LT|LE|EQ|NE|GE|GT) (CMN x (RSBconst [0] y))) => ((LT|LE|EQ|NE|GE|GT) (CMP x y)) // sense of carry bit not preserved +(EQ (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 => (EQ (CMP x y) yes no) +(EQ (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 => (EQ (CMP a (MUL x y)) yes no) +(EQ (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 => (EQ (CMPconst [c] x) yes no) +(EQ (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 => (EQ (CMPshiftLL x y [c]) yes no) +(EQ (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 => (EQ (CMPshiftRL x y [c]) yes no) +(EQ (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 => (EQ (CMPshiftRA x y [c]) yes no) +(EQ (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 => (EQ (CMPshiftLLreg x y z) yes no) +(EQ (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 => (EQ (CMPshiftRLreg x y z) yes no) +(EQ (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 => (EQ (CMPshiftRAreg x y z) yes no) +(NE (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 => (NE (CMP x y) yes no) +(NE (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 => (NE (CMP a (MUL x y)) yes no) +(NE (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 => (NE (CMPconst [c] x) yes no) +(NE (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 => (NE (CMPshiftLL x y [c]) yes no) +(NE (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 => (NE (CMPshiftRL x y [c]) yes no) +(NE (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 => (NE (CMPshiftRA x y [c]) yes no) +(NE (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 => (NE (CMPshiftLLreg x y z) yes no) +(NE (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 => (NE (CMPshiftRLreg x y z) yes no) +(NE (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 => (NE (CMPshiftRAreg x y z) yes no) +(EQ (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 => (EQ (CMN x y) yes no) +(EQ (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 => (EQ (CMN a (MUL x y)) yes no) +(EQ (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 => (EQ (CMNconst [c] x) yes no) +(EQ (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 => (EQ (CMNshiftLL x y [c]) yes no) +(EQ (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 => (EQ (CMNshiftRL x y [c]) yes no) +(EQ (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 => (EQ (CMNshiftRA x y [c]) yes no) +(EQ (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 => (EQ (CMNshiftLLreg x y z) yes no) +(EQ (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 => (EQ (CMNshiftRLreg x y z) yes no) +(EQ (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 => (EQ (CMNshiftRAreg x y z) yes no) +(NE (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 => (NE (CMN x y) yes no) +(NE (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 => (NE (CMN a (MUL x y)) yes no) +(NE (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 => (NE (CMNconst [c] x) yes no) +(NE (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 => (NE (CMNshiftLL x y [c]) yes no) +(NE (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 => (NE (CMNshiftRL x y [c]) yes no) +(NE (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 => (NE (CMNshiftRA x y [c]) yes no) +(NE (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 => (NE (CMNshiftLLreg x y z) yes no) +(NE (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 => (NE (CMNshiftRLreg x y z) yes no) +(NE (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 => (NE (CMNshiftRAreg x y z) yes no) +(EQ (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 => (EQ (TST x y) yes no) +(EQ (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 => (EQ (TSTconst [c] x) yes no) +(EQ (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 => (EQ (TSTshiftLL x y [c]) yes no) +(EQ (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 => (EQ (TSTshiftRL x y [c]) yes no) +(EQ (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 => (EQ (TSTshiftRA x y [c]) yes no) +(EQ (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 => (EQ (TSTshiftLLreg x y z) yes no) +(EQ (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 => (EQ (TSTshiftRLreg x y z) yes no) +(EQ (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 => (EQ (TSTshiftRAreg x y z) yes no) +(NE (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 => (NE (TST x y) yes no) +(NE (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 => (NE (TSTconst [c] x) yes no) +(NE (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 => (NE (TSTshiftLL x y [c]) yes no) +(NE (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 => (NE (TSTshiftRL x y [c]) yes no) +(NE (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 => (NE (TSTshiftRA x y [c]) yes no) +(NE (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 => (NE (TSTshiftLLreg x y z) yes no) +(NE (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 => (NE (TSTshiftRLreg x y z) yes no) +(NE (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 => (NE (TSTshiftRAreg x y z) yes no) +(EQ (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 => (EQ (TEQ x y) yes no) +(EQ (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 => (EQ (TEQconst [c] x) yes no) +(EQ (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 => (EQ (TEQshiftLL x y [c]) yes no) +(EQ (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 => (EQ (TEQshiftRL x y [c]) yes no) +(EQ (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 => (EQ (TEQshiftRA x y [c]) yes no) +(EQ (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 => (EQ (TEQshiftLLreg x y z) yes no) +(EQ (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 => (EQ (TEQshiftRLreg x y z) yes no) +(EQ (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 => (EQ (TEQshiftRAreg x y z) yes no) +(NE (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 => (NE (TEQ x y) yes no) +(NE (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 => (NE (TEQconst [c] x) yes no) +(NE (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 => (NE (TEQshiftLL x y [c]) yes no) +(NE (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 => (NE (TEQshiftRL x y [c]) yes no) +(NE (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 => (NE (TEQshiftRA x y [c]) yes no) +(NE (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 => (NE (TEQshiftLLreg x y z) yes no) +(NE (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 => (NE (TEQshiftRLreg x y z) yes no) +(NE (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 => (NE (TEQshiftRAreg x y z) yes no) +(LT (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 => (LTnoov (CMP x y) yes no) +(LT (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 => (LTnoov (CMP a (MUL x y)) yes no) +(LT (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 => (LTnoov (CMPconst [c] x) yes no) +(LT (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 => (LTnoov (CMPshiftLL x y [c]) yes no) +(LT (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 => (LTnoov (CMPshiftRL x y [c]) yes no) +(LT (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 => (LTnoov (CMPshiftRA x y [c]) yes no) +(LT (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 => (LTnoov (CMPshiftLLreg x y z) yes no) +(LT (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 => (LTnoov (CMPshiftRLreg x y z) yes no) +(LT (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 => (LTnoov (CMPshiftRAreg x y z) yes no) +(LE (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 => (LEnoov (CMP x y) yes no) +(LE (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 => (LEnoov (CMP a (MUL x y)) yes no) +(LE (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 => (LEnoov (CMPconst [c] x) yes no) +(LE (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 => (LEnoov (CMPshiftLL x y [c]) yes no) +(LE (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 => (LEnoov (CMPshiftRL x y [c]) yes no) +(LE (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 => (LEnoov (CMPshiftRA x y [c]) yes no) +(LE (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 => (LEnoov (CMPshiftLLreg x y z) yes no) +(LE (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 => (LEnoov (CMPshiftRLreg x y z) yes no) +(LE (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 => (LEnoov (CMPshiftRAreg x y z) yes no) +(LT (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 => (LTnoov (CMN x y) yes no) +(LT (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 => (LTnoov (CMN a (MUL x y)) yes no) +(LT (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 => (LTnoov (CMNconst [c] x) yes no) +(LT (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 => (LTnoov (CMNshiftLL x y [c]) yes no) +(LT (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 => (LTnoov (CMNshiftRL x y [c]) yes no) +(LT (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 => (LTnoov (CMNshiftRA x y [c]) yes no) +(LT (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 => (LTnoov (CMNshiftLLreg x y z) yes no) +(LT (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 => (LTnoov (CMNshiftRLreg x y z) yes no) +(LT (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 => (LTnoov (CMNshiftRAreg x y z) yes no) +(LE (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 => (LEnoov (CMN x y) yes no) +(LE (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 => (LEnoov (CMN a (MUL x y)) yes no) +(LE (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 => (LEnoov (CMNconst [c] x) yes no) +(LE (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 => (LEnoov (CMNshiftLL x y [c]) yes no) +(LE (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 => (LEnoov (CMNshiftRL x y [c]) yes no) +(LE (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 => (LEnoov (CMNshiftRA x y [c]) yes no) +(LE (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 => (LEnoov (CMNshiftLLreg x y z) yes no) +(LE (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 => (LEnoov (CMNshiftRLreg x y z) yes no) +(LE (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 => (LEnoov (CMNshiftRAreg x y z) yes no) +(LT (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 => (LTnoov (TST x y) yes no) +(LT (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 => (LTnoov (TSTconst [c] x) yes no) +(LT (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 => (LTnoov (TSTshiftLL x y [c]) yes no) +(LT (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 => (LTnoov (TSTshiftRL x y [c]) yes no) +(LT (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 => (LTnoov (TSTshiftRA x y [c]) yes no) +(LT (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 => (LTnoov (TSTshiftLLreg x y z) yes no) +(LT (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 => (LTnoov (TSTshiftRLreg x y z) yes no) +(LT (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 => (LTnoov (TSTshiftRAreg x y z) yes no) +(LE (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 => (LEnoov (TST x y) yes no) +(LE (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 => (LEnoov (TSTconst [c] x) yes no) +(LE (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 => (LEnoov (TSTshiftLL x y [c]) yes no) +(LE (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 => (LEnoov (TSTshiftRL x y [c]) yes no) +(LE (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 => (LEnoov (TSTshiftRA x y [c]) yes no) +(LE (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 => (LEnoov (TSTshiftLLreg x y z) yes no) +(LE (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 => (LEnoov (TSTshiftRLreg x y z) yes no) +(LE (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 => (LEnoov (TSTshiftRAreg x y z) yes no) +(LT (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 => (LTnoov (TEQ x y) yes no) +(LT (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 => (LTnoov (TEQconst [c] x) yes no) +(LT (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 => (LTnoov (TEQshiftLL x y [c]) yes no) +(LT (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 => (LTnoov (TEQshiftRL x y [c]) yes no) +(LT (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 => (LTnoov (TEQshiftRA x y [c]) yes no) +(LT (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 => (LTnoov (TEQshiftLLreg x y z) yes no) +(LT (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 => (LTnoov (TEQshiftRLreg x y z) yes no) +(LT (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 => (LTnoov (TEQshiftRAreg x y z) yes no) +(LE (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 => (LEnoov (TEQ x y) yes no) +(LE (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 => (LEnoov (TEQconst [c] x) yes no) +(LE (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 => (LEnoov (TEQshiftLL x y [c]) yes no) +(LE (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 => (LEnoov (TEQshiftRL x y [c]) yes no) +(LE (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 => (LEnoov (TEQshiftRA x y [c]) yes no) +(LE (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 => (LEnoov (TEQshiftLLreg x y z) yes no) +(LE (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 => (LEnoov (TEQshiftRLreg x y z) yes no) +(LE (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 => (LEnoov (TEQshiftRAreg x y z) yes no) +(GT (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 => (GTnoov (CMP x y) yes no) +(GT (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 => (GTnoov (CMP a (MUL x y)) yes no) +(GT (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 => (GTnoov (CMPconst [c] x) yes no) +(GT (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 => (GTnoov (CMPshiftLL x y [c]) yes no) +(GT (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 => (GTnoov (CMPshiftRL x y [c]) yes no) +(GT (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 => (GTnoov (CMPshiftRA x y [c]) yes no) +(GT (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 => (GTnoov (CMPshiftLLreg x y z) yes no) +(GT (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 => (GTnoov (CMPshiftRLreg x y z) yes no) +(GT (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 => (GTnoov (CMPshiftRAreg x y z) yes no) +(GE (CMPconst [0] l:(SUB x y)) yes no) && l.Uses==1 => (GEnoov (CMP x y) yes no) +(GE (CMPconst [0] l:(MULS x y a)) yes no) && l.Uses==1 => (GEnoov (CMP a (MUL x y)) yes no) +(GE (CMPconst [0] l:(SUBconst [c] x)) yes no) && l.Uses==1 => (GEnoov (CMPconst [c] x) yes no) +(GE (CMPconst [0] l:(SUBshiftLL x y [c])) yes no) && l.Uses==1 => (GEnoov (CMPshiftLL x y [c]) yes no) +(GE (CMPconst [0] l:(SUBshiftRL x y [c])) yes no) && l.Uses==1 => (GEnoov (CMPshiftRL x y [c]) yes no) +(GE (CMPconst [0] l:(SUBshiftRA x y [c])) yes no) && l.Uses==1 => (GEnoov (CMPshiftRA x y [c]) yes no) +(GE (CMPconst [0] l:(SUBshiftLLreg x y z)) yes no) && l.Uses==1 => (GEnoov (CMPshiftLLreg x y z) yes no) +(GE (CMPconst [0] l:(SUBshiftRLreg x y z)) yes no) && l.Uses==1 => (GEnoov (CMPshiftRLreg x y z) yes no) +(GE (CMPconst [0] l:(SUBshiftRAreg x y z)) yes no) && l.Uses==1 => (GEnoov (CMPshiftRAreg x y z) yes no) +(GT (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 => (GTnoov (CMN x y) yes no) +(GT (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 => (GTnoov (CMNconst [c] x) yes no) +(GT (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 => (GTnoov (CMNshiftLL x y [c]) yes no) +(GT (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 => (GTnoov (CMNshiftRL x y [c]) yes no) +(GT (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 => (GTnoov (CMNshiftRA x y [c]) yes no) +(GT (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 => (GTnoov (CMNshiftLLreg x y z) yes no) +(GT (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 => (GTnoov (CMNshiftRLreg x y z) yes no) +(GT (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 => (GTnoov (CMNshiftRAreg x y z) yes no) +(GE (CMPconst [0] l:(ADD x y)) yes no) && l.Uses==1 => (GEnoov (CMN x y) yes no) +(GE (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 => (GEnoov (CMN a (MUL x y)) yes no) +(GE (CMPconst [0] l:(ADDconst [c] x)) yes no) && l.Uses==1 => (GEnoov (CMNconst [c] x) yes no) +(GE (CMPconst [0] l:(ADDshiftLL x y [c])) yes no) && l.Uses==1 => (GEnoov (CMNshiftLL x y [c]) yes no) +(GE (CMPconst [0] l:(ADDshiftRL x y [c])) yes no) && l.Uses==1 => (GEnoov (CMNshiftRL x y [c]) yes no) +(GE (CMPconst [0] l:(ADDshiftRA x y [c])) yes no) && l.Uses==1 => (GEnoov (CMNshiftRA x y [c]) yes no) +(GE (CMPconst [0] l:(ADDshiftLLreg x y z)) yes no) && l.Uses==1 => (GEnoov (CMNshiftLLreg x y z) yes no) +(GE (CMPconst [0] l:(ADDshiftRLreg x y z)) yes no) && l.Uses==1 => (GEnoov (CMNshiftRLreg x y z) yes no) +(GE (CMPconst [0] l:(ADDshiftRAreg x y z)) yes no) && l.Uses==1 => (GEnoov (CMNshiftRAreg x y z) yes no) +(GT (CMPconst [0] l:(MULA x y a)) yes no) && l.Uses==1 => (GTnoov (CMN a (MUL x y)) yes no) +(GT (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 => (GTnoov (TST x y) yes no) +(GT (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 => (GTnoov (TSTconst [c] x) yes no) +(GT (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 => (GTnoov (TSTshiftLL x y [c]) yes no) +(GT (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 => (GTnoov (TSTshiftRL x y [c]) yes no) +(GT (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 => (GTnoov (TSTshiftRA x y [c]) yes no) +(GT (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 => (GTnoov (TSTshiftLLreg x y z) yes no) +(GT (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 => (GTnoov (TSTshiftRLreg x y z) yes no) +(GT (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 => (GTnoov (TSTshiftRAreg x y z) yes no) +(GE (CMPconst [0] l:(AND x y)) yes no) && l.Uses==1 => (GEnoov (TST x y) yes no) +(GE (CMPconst [0] l:(ANDconst [c] x)) yes no) && l.Uses==1 => (GEnoov (TSTconst [c] x) yes no) +(GE (CMPconst [0] l:(ANDshiftLL x y [c])) yes no) && l.Uses==1 => (GEnoov (TSTshiftLL x y [c]) yes no) +(GE (CMPconst [0] l:(ANDshiftRL x y [c])) yes no) && l.Uses==1 => (GEnoov (TSTshiftRL x y [c]) yes no) +(GE (CMPconst [0] l:(ANDshiftRA x y [c])) yes no) && l.Uses==1 => (GEnoov (TSTshiftRA x y [c]) yes no) +(GE (CMPconst [0] l:(ANDshiftLLreg x y z)) yes no) && l.Uses==1 => (GEnoov (TSTshiftLLreg x y z) yes no) +(GE (CMPconst [0] l:(ANDshiftRLreg x y z)) yes no) && l.Uses==1 => (GEnoov (TSTshiftRLreg x y z) yes no) +(GE (CMPconst [0] l:(ANDshiftRAreg x y z)) yes no) && l.Uses==1 => (GEnoov (TSTshiftRAreg x y z) yes no) +(GT (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 => (GTnoov (TEQ x y) yes no) +(GT (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 => (GTnoov (TEQconst [c] x) yes no) +(GT (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 => (GTnoov (TEQshiftLL x y [c]) yes no) +(GT (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 => (GTnoov (TEQshiftRL x y [c]) yes no) +(GT (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 => (GTnoov (TEQshiftRA x y [c]) yes no) +(GT (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 => (GTnoov (TEQshiftLLreg x y z) yes no) +(GT (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 => (GTnoov (TEQshiftRLreg x y z) yes no) +(GT (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 => (GTnoov (TEQshiftRAreg x y z) yes no) +(GE (CMPconst [0] l:(XOR x y)) yes no) && l.Uses==1 => (GEnoov (TEQ x y) yes no) +(GE (CMPconst [0] l:(XORconst [c] x)) yes no) && l.Uses==1 => (GEnoov (TEQconst [c] x) yes no) +(GE (CMPconst [0] l:(XORshiftLL x y [c])) yes no) && l.Uses==1 => (GEnoov (TEQshiftLL x y [c]) yes no) +(GE (CMPconst [0] l:(XORshiftRL x y [c])) yes no) && l.Uses==1 => (GEnoov (TEQshiftRL x y [c]) yes no) +(GE (CMPconst [0] l:(XORshiftRA x y [c])) yes no) && l.Uses==1 => (GEnoov (TEQshiftRA x y [c]) yes no) +(GE (CMPconst [0] l:(XORshiftLLreg x y z)) yes no) && l.Uses==1 => (GEnoov (TEQshiftLLreg x y z) yes no) +(GE (CMPconst [0] l:(XORshiftRLreg x y z)) yes no) && l.Uses==1 => (GEnoov (TEQshiftRLreg x y z) yes no) +(GE (CMPconst [0] l:(XORshiftRAreg x y z)) yes no) && l.Uses==1 => (GEnoov (TEQshiftRAreg x y z) yes no) -(MOVBUload [off] {sym} (SB) _) && symIsRO(sym) -> (MOVWconst [int64(read8(sym, off))]) -(MOVHUload [off] {sym} (SB) _) && symIsRO(sym) -> (MOVWconst [int64(read16(sym, off, config.ctxt.Arch.ByteOrder))]) -(MOVWload [off] {sym} (SB) _) && symIsRO(sym) -> (MOVWconst [int64(int32(read32(sym, off, config.ctxt.Arch.ByteOrder)))]) +(MOVBUload [off] {sym} (SB) _) && symIsRO(sym) => (MOVWconst [int32(read8(sym, int64(off)))]) +(MOVHUload [off] {sym} (SB) _) && symIsRO(sym) => (MOVWconst [int32(read16(sym, int64(off), config.ctxt.Arch.ByteOrder))]) +(MOVWload [off] {sym} (SB) _) && symIsRO(sym) => (MOVWconst [int32(read32(sym, int64(off), config.ctxt.Arch.ByteOrder))]) diff --git a/src/cmd/compile/internal/ssa/gen/ARM64.rules b/src/cmd/compile/internal/ssa/gen/ARM64.rules index 311067e87a3..3f4d0c1c527 100644 --- a/src/cmd/compile/internal/ssa/gen/ARM64.rules +++ b/src/cmd/compile/internal/ssa/gen/ARM64.rules @@ -543,15 +543,24 @@ (AtomicStore64 ...) => (STLR ...) (AtomicStorePtrNoWB ...) => (STLR ...) -(AtomicExchange(32|64) ...) => (LoweredAtomicExchange(32|64) ...) -(AtomicAdd(32|64) ...) => (LoweredAtomicAdd(32|64) ...) +(AtomicExchange(32|64) ...) => (LoweredAtomicExchange(32|64) ...) +(AtomicAdd(32|64) ...) => (LoweredAtomicAdd(32|64) ...) (AtomicCompareAndSwap(32|64) ...) => (LoweredAtomicCas(32|64) ...) -// Currently the updated value is not used, but we need a register to temporarily hold it. -(AtomicAnd8 ptr val mem) => (Select1 (LoweredAtomicAnd8 ptr val mem)) -(AtomicOr8 ptr val mem) => (Select1 (LoweredAtomicOr8 ptr val mem)) +(AtomicAdd(32|64)Variant ...) => (LoweredAtomicAdd(32|64)Variant ...) +(AtomicExchange(32|64)Variant ...) => (LoweredAtomicExchange(32|64)Variant ...) +(AtomicCompareAndSwap(32|64)Variant ...) => (LoweredAtomicCas(32|64)Variant ...) -(AtomicAdd(32|64)Variant ...) => (LoweredAtomicAdd(32|64)Variant ...) +// Currently the updated value is not used, but we need a register to temporarily hold it. +(AtomicAnd8 ptr val mem) => (Select1 (LoweredAtomicAnd8 ptr val mem)) +(AtomicAnd32 ptr val mem) => (Select1 (LoweredAtomicAnd32 ptr val mem)) +(AtomicOr8 ptr val mem) => (Select1 (LoweredAtomicOr8 ptr val mem)) +(AtomicOr32 ptr val mem) => (Select1 (LoweredAtomicOr32 ptr val mem)) + +(AtomicAnd8Variant ptr val mem) => (Select1 (LoweredAtomicAnd8Variant ptr val mem)) +(AtomicAnd32Variant ptr val mem) => (Select1 (LoweredAtomicAnd32Variant ptr val mem)) +(AtomicOr8Variant ptr val mem) => (Select1 (LoweredAtomicOr8Variant ptr val mem)) +(AtomicOr32Variant ptr val mem) => (Select1 (LoweredAtomicOr32Variant ptr val mem)) // Write barrier. (WB ...) => (LoweredWB ...) @@ -859,88 +868,88 @@ (MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVBload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVBUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVBUload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVHload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVHUload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVWload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVWload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVWUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVWUload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVDload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (FMOVSload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (FMOVSload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (FMOVSload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (FMOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (FMOVDload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (FMOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVBstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVBstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVHstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVHstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVWstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVWstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVDstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (STP [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val1 val2 mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (STP [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val1 val2 mem) + (STP [off1+off2] {mergeSym(sym1,sym2)} ptr val1 val2 mem) (FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (FMOVSstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (FMOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (FMOVDstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVBstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVBstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVHstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVWstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVWstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVDstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVDstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVDstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVQstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) && (ptr.Op != OpSB || !config.ctxt.Flag_shared) => - (MOVQstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVQstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) // store zero (MOVBstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVBstorezero [off] {sym} ptr mem) @@ -1116,280 +1125,282 @@ // if a register move has only 1 use, just use the same register without emitting instruction // MOVDnop doesn't emit instruction, only for ensuring the type. -(MOVDreg x) && x.Uses == 1 -> (MOVDnop x) +(MOVDreg x) && x.Uses == 1 => (MOVDnop x) // fold constant into arithmatic ops -(ADD x (MOVDconst [c])) -> (ADDconst [c] x) -(SUB x (MOVDconst [c])) -> (SUBconst [c] x) -(AND x (MOVDconst [c])) -> (ANDconst [c] x) -(OR x (MOVDconst [c])) -> (ORconst [c] x) -(XOR x (MOVDconst [c])) -> (XORconst [c] x) -(TST x (MOVDconst [c])) -> (TSTconst [c] x) -(TSTW x (MOVDconst [c])) -> (TSTWconst [c] x) -(CMN x (MOVDconst [c])) -> (CMNconst [c] x) -(CMNW x (MOVDconst [c])) -> (CMNWconst [c] x) -(BIC x (MOVDconst [c])) -> (ANDconst [^c] x) -(EON x (MOVDconst [c])) -> (XORconst [^c] x) -(ORN x (MOVDconst [c])) -> (ORconst [^c] x) +(ADD x (MOVDconst [c])) => (ADDconst [c] x) +(SUB x (MOVDconst [c])) => (SUBconst [c] x) +(AND x (MOVDconst [c])) => (ANDconst [c] x) +(OR x (MOVDconst [c])) => (ORconst [c] x) +(XOR x (MOVDconst [c])) => (XORconst [c] x) +(TST x (MOVDconst [c])) => (TSTconst [c] x) +(TSTW x (MOVDconst [c])) => (TSTWconst [int32(c)] x) +(CMN x (MOVDconst [c])) => (CMNconst [c] x) +(CMNW x (MOVDconst [c])) => (CMNWconst [int32(c)] x) +(BIC x (MOVDconst [c])) => (ANDconst [^c] x) +(EON x (MOVDconst [c])) => (XORconst [^c] x) +(ORN x (MOVDconst [c])) => (ORconst [^c] x) -(SLL x (MOVDconst [c])) -> (SLLconst x [c&63]) // Note: I don't think we ever generate bad constant shifts (i.e. c>=64) -(SRL x (MOVDconst [c])) -> (SRLconst x [c&63]) -(SRA x (MOVDconst [c])) -> (SRAconst x [c&63]) +(SLL x (MOVDconst [c])) => (SLLconst x [c&63]) // Note: I don't think we ever generate bad constant shifts (i.e. c>=64) +(SRL x (MOVDconst [c])) => (SRLconst x [c&63]) +(SRA x (MOVDconst [c])) => (SRAconst x [c&63]) -(CMP x (MOVDconst [c])) -> (CMPconst [c] x) -(CMP (MOVDconst [c]) x) -> (InvertFlags (CMPconst [c] x)) +(CMP x (MOVDconst [c])) => (CMPconst [c] x) +(CMP (MOVDconst [c]) x) => (InvertFlags (CMPconst [c] x)) (CMPW x (MOVDconst [c])) => (CMPWconst [int32(c)] x) (CMPW (MOVDconst [c]) x) => (InvertFlags (CMPWconst [int32(c)] x)) // Canonicalize the order of arguments to comparisons - helps with CSE. -((CMP|CMPW) x y) && x.ID > y.ID -> (InvertFlags ((CMP|CMPW) y x)) +((CMP|CMPW) x y) && x.ID > y.ID => (InvertFlags ((CMP|CMPW) y x)) -// mul-neg -> mneg -(NEG (MUL x y)) -> (MNEG x y) -(NEG (MULW x y)) -> (MNEGW x y) -(MUL (NEG x) y) -> (MNEG x y) -(MULW (NEG x) y) -> (MNEGW x y) +// mul-neg => mneg +(NEG (MUL x y)) => (MNEG x y) +(NEG (MULW x y)) => (MNEGW x y) +(MUL (NEG x) y) => (MNEG x y) +(MULW (NEG x) y) => (MNEGW x y) // madd/msub -(ADD a l:(MUL x y)) && l.Uses==1 && clobber(l) -> (MADD a x y) -(SUB a l:(MUL x y)) && l.Uses==1 && clobber(l) -> (MSUB a x y) -(ADD a l:(MNEG x y)) && l.Uses==1 && clobber(l) -> (MSUB a x y) -(SUB a l:(MNEG x y)) && l.Uses==1 && clobber(l) -> (MADD a x y) +(ADD a l:(MUL x y)) && l.Uses==1 && clobber(l) => (MADD a x y) +(SUB a l:(MUL x y)) && l.Uses==1 && clobber(l) => (MSUB a x y) +(ADD a l:(MNEG x y)) && l.Uses==1 && clobber(l) => (MSUB a x y) +(SUB a l:(MNEG x y)) && l.Uses==1 && clobber(l) => (MADD a x y) -(ADD a l:(MULW x y)) && a.Type.Size() != 8 && l.Uses==1 && clobber(l) -> (MADDW a x y) -(SUB a l:(MULW x y)) && a.Type.Size() != 8 && l.Uses==1 && clobber(l) -> (MSUBW a x y) -(ADD a l:(MNEGW x y)) && a.Type.Size() != 8 && l.Uses==1 && clobber(l) -> (MSUBW a x y) -(SUB a l:(MNEGW x y)) && a.Type.Size() != 8 && l.Uses==1 && clobber(l) -> (MADDW a x y) +(ADD a l:(MULW x y)) && a.Type.Size() != 8 && l.Uses==1 && clobber(l) => (MADDW a x y) +(SUB a l:(MULW x y)) && a.Type.Size() != 8 && l.Uses==1 && clobber(l) => (MSUBW a x y) +(ADD a l:(MNEGW x y)) && a.Type.Size() != 8 && l.Uses==1 && clobber(l) => (MSUBW a x y) +(SUB a l:(MNEGW x y)) && a.Type.Size() != 8 && l.Uses==1 && clobber(l) => (MADDW a x y) // optimize ADCSflags, SBCSflags and friends -(ADCSflags x y (Select1 (ADDSconstflags [-1] (ADCzerocarry c)))) -> (ADCSflags x y c) -(ADCSflags x y (Select1 (ADDSconstflags [-1] (MOVDconst [0])))) -> (ADDSflags x y) -(SBCSflags x y (Select1 (NEGSflags (NEG (NGCzerocarry bo))))) -> (SBCSflags x y bo) -(SBCSflags x y (Select1 (NEGSflags (MOVDconst [0])))) -> (SUBSflags x y) +(ADCSflags x y (Select1 (ADDSconstflags [-1] (ADCzerocarry c)))) => (ADCSflags x y c) +(ADCSflags x y (Select1 (ADDSconstflags [-1] (MOVDconst [0])))) => (ADDSflags x y) +(SBCSflags x y (Select1 (NEGSflags (NEG (NGCzerocarry bo))))) => (SBCSflags x y bo) +(SBCSflags x y (Select1 (NEGSflags (MOVDconst [0])))) => (SUBSflags x y) // mul by constant -(MUL x (MOVDconst [-1])) -> (NEG x) -(MUL _ (MOVDconst [0])) -> (MOVDconst [0]) -(MUL x (MOVDconst [1])) -> x -(MUL x (MOVDconst [c])) && isPowerOfTwo(c) -> (SLLconst [log2(c)] x) -(MUL x (MOVDconst [c])) && isPowerOfTwo(c-1) && c >= 3 -> (ADDshiftLL x x [log2(c-1)]) -(MUL x (MOVDconst [c])) && isPowerOfTwo(c+1) && c >= 7 -> (ADDshiftLL (NEG x) x [log2(c+1)]) -(MUL x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) -> (SLLconst [log2(c/3)] (ADDshiftLL x x [1])) -(MUL x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) -> (SLLconst [log2(c/5)] (ADDshiftLL x x [2])) -(MUL x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) -> (SLLconst [log2(c/7)] (ADDshiftLL (NEG x) x [3])) -(MUL x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) -> (SLLconst [log2(c/9)] (ADDshiftLL x x [3])) +(MUL x (MOVDconst [-1])) => (NEG x) +(MUL _ (MOVDconst [0])) => (MOVDconst [0]) +(MUL x (MOVDconst [1])) => x +(MUL x (MOVDconst [c])) && isPowerOfTwo64(c) => (SLLconst [log64(c)] x) +(MUL x (MOVDconst [c])) && isPowerOfTwo64(c-1) && c >= 3 => (ADDshiftLL x x [log64(c-1)]) +(MUL x (MOVDconst [c])) && isPowerOfTwo64(c+1) && c >= 7 => (ADDshiftLL (NEG x) x [log64(c+1)]) +(MUL x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo64(c/3) => (SLLconst [log64(c/3)] (ADDshiftLL x x [1])) +(MUL x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo64(c/5) => (SLLconst [log64(c/5)] (ADDshiftLL x x [2])) +(MUL x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo64(c/7) => (SLLconst [log64(c/7)] (ADDshiftLL (NEG x) x [3])) +(MUL x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo64(c/9) => (SLLconst [log64(c/9)] (ADDshiftLL x x [3])) -(MULW x (MOVDconst [c])) && int32(c)==-1 -> (NEG x) -(MULW _ (MOVDconst [c])) && int32(c)==0 -> (MOVDconst [0]) -(MULW x (MOVDconst [c])) && int32(c)==1 -> x -(MULW x (MOVDconst [c])) && isPowerOfTwo(c) -> (SLLconst [log2(c)] x) -(MULW x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c) >= 3 -> (ADDshiftLL x x [log2(c-1)]) -(MULW x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c) >= 7 -> (ADDshiftLL (NEG x) x [log2(c+1)]) -(MULW x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (SLLconst [log2(c/3)] (ADDshiftLL x x [1])) -(MULW x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (SLLconst [log2(c/5)] (ADDshiftLL x x [2])) -(MULW x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (SLLconst [log2(c/7)] (ADDshiftLL (NEG x) x [3])) -(MULW x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (SLLconst [log2(c/9)] (ADDshiftLL x x [3])) +(MULW x (MOVDconst [c])) && int32(c)==-1 => (NEG x) +(MULW _ (MOVDconst [c])) && int32(c)==0 => (MOVDconst [0]) +(MULW x (MOVDconst [c])) && int32(c)==1 => x +(MULW x (MOVDconst [c])) && isPowerOfTwo64(c) => (SLLconst [log64(c)] x) +(MULW x (MOVDconst [c])) && isPowerOfTwo64(c-1) && int32(c) >= 3 => (ADDshiftLL x x [log64(c-1)]) +(MULW x (MOVDconst [c])) && isPowerOfTwo64(c+1) && int32(c) >= 7 => (ADDshiftLL (NEG x) x [log64(c+1)]) +(MULW x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo64(c/3) && is32Bit(c) => (SLLconst [log64(c/3)] (ADDshiftLL x x [1])) +(MULW x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo64(c/5) && is32Bit(c) => (SLLconst [log64(c/5)] (ADDshiftLL x x [2])) +(MULW x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo64(c/7) && is32Bit(c) => (SLLconst [log64(c/7)] (ADDshiftLL (NEG x) x [3])) +(MULW x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo64(c/9) && is32Bit(c) => (SLLconst [log64(c/9)] (ADDshiftLL x x [3])) // mneg by constant -(MNEG x (MOVDconst [-1])) -> x -(MNEG _ (MOVDconst [0])) -> (MOVDconst [0]) -(MNEG x (MOVDconst [1])) -> (NEG x) -(MNEG x (MOVDconst [c])) && isPowerOfTwo(c) -> (NEG (SLLconst [log2(c)] x)) -(MNEG x (MOVDconst [c])) && isPowerOfTwo(c-1) && c >= 3 -> (NEG (ADDshiftLL x x [log2(c-1)])) -(MNEG x (MOVDconst [c])) && isPowerOfTwo(c+1) && c >= 7 -> (NEG (ADDshiftLL (NEG x) x [log2(c+1)])) -(MNEG x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) -> (SLLconst [log2(c/3)] (SUBshiftLL x x [2])) -(MNEG x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) -> (NEG (SLLconst [log2(c/5)] (ADDshiftLL x x [2]))) -(MNEG x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) -> (SLLconst [log2(c/7)] (SUBshiftLL x x [3])) -(MNEG x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) -> (NEG (SLLconst [log2(c/9)] (ADDshiftLL x x [3]))) +(MNEG x (MOVDconst [-1])) => x +(MNEG _ (MOVDconst [0])) => (MOVDconst [0]) +(MNEG x (MOVDconst [1])) => (NEG x) +(MNEG x (MOVDconst [c])) && isPowerOfTwo64(c) => (NEG (SLLconst [log64(c)] x)) +(MNEG x (MOVDconst [c])) && isPowerOfTwo64(c-1) && c >= 3 => (NEG (ADDshiftLL x x [log64(c-1)])) +(MNEG x (MOVDconst [c])) && isPowerOfTwo64(c+1) && c >= 7 => (NEG (ADDshiftLL (NEG x) x [log64(c+1)])) +(MNEG x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo64(c/3) => (SLLconst [log64(c/3)] (SUBshiftLL x x [2])) +(MNEG x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo64(c/5) => (NEG (SLLconst [log64(c/5)] (ADDshiftLL x x [2]))) +(MNEG x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo64(c/7) => (SLLconst [log64(c/7)] (SUBshiftLL x x [3])) +(MNEG x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo64(c/9) => (NEG (SLLconst [log64(c/9)] (ADDshiftLL x x [3]))) -(MNEGW x (MOVDconst [c])) && int32(c)==-1 -> x -(MNEGW _ (MOVDconst [c])) && int32(c)==0 -> (MOVDconst [0]) -(MNEGW x (MOVDconst [c])) && int32(c)==1 -> (NEG x) -(MNEGW x (MOVDconst [c])) && isPowerOfTwo(c) -> (NEG (SLLconst [log2(c)] x)) -(MNEGW x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c) >= 3 -> (NEG (ADDshiftLL x x [log2(c-1)])) -(MNEGW x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c) >= 7 -> (NEG (ADDshiftLL (NEG x) x [log2(c+1)])) -(MNEGW x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (SLLconst [log2(c/3)] (SUBshiftLL x x [2])) -(MNEGW x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (NEG (SLLconst [log2(c/5)] (ADDshiftLL x x [2]))) -(MNEGW x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (SLLconst [log2(c/7)] (SUBshiftLL x x [3])) -(MNEGW x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (NEG (SLLconst [log2(c/9)] (ADDshiftLL x x [3]))) -(MADD a x (MOVDconst [-1])) -> (SUB a x) -(MADD a _ (MOVDconst [0])) -> a -(MADD a x (MOVDconst [1])) -> (ADD a x) -(MADD a x (MOVDconst [c])) && isPowerOfTwo(c) -> (ADDshiftLL a x [log2(c)]) -(MADD a x (MOVDconst [c])) && isPowerOfTwo(c-1) && c>=3 -> (ADD a (ADDshiftLL x x [log2(c-1)])) -(MADD a x (MOVDconst [c])) && isPowerOfTwo(c+1) && c>=7 -> (SUB a (SUBshiftLL x x [log2(c+1)])) -(MADD a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) -> (SUBshiftLL a (SUBshiftLL x x [2]) [log2(c/3)]) -(MADD a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) -> (ADDshiftLL a (ADDshiftLL x x [2]) [log2(c/5)]) -(MADD a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) -> (SUBshiftLL a (SUBshiftLL x x [3]) [log2(c/7)]) -(MADD a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) -> (ADDshiftLL a (ADDshiftLL x x [3]) [log2(c/9)]) +(MNEGW x (MOVDconst [c])) && int32(c)==-1 => x +(MNEGW _ (MOVDconst [c])) && int32(c)==0 => (MOVDconst [0]) +(MNEGW x (MOVDconst [c])) && int32(c)==1 => (NEG x) +(MNEGW x (MOVDconst [c])) && isPowerOfTwo64(c) => (NEG (SLLconst [log64(c)] x)) +(MNEGW x (MOVDconst [c])) && isPowerOfTwo64(c-1) && int32(c) >= 3 => (NEG (ADDshiftLL x x [log64(c-1)])) +(MNEGW x (MOVDconst [c])) && isPowerOfTwo64(c+1) && int32(c) >= 7 => (NEG (ADDshiftLL (NEG x) x [log64(c+1)])) +(MNEGW x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo64(c/3) && is32Bit(c) => (SLLconst [log64(c/3)] (SUBshiftLL x x [2])) +(MNEGW x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo64(c/5) && is32Bit(c) => (NEG (SLLconst [log64(c/5)] (ADDshiftLL x x [2]))) +(MNEGW x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo64(c/7) && is32Bit(c) => (SLLconst [log64(c/7)] (SUBshiftLL x x [3])) +(MNEGW x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo64(c/9) && is32Bit(c) => (NEG (SLLconst [log64(c/9)] (ADDshiftLL x x [3]))) -(MADD a (MOVDconst [-1]) x) -> (SUB a x) -(MADD a (MOVDconst [0]) _) -> a -(MADD a (MOVDconst [1]) x) -> (ADD a x) -(MADD a (MOVDconst [c]) x) && isPowerOfTwo(c) -> (ADDshiftLL a x [log2(c)]) -(MADD a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && c>=3 -> (ADD a (ADDshiftLL x x [log2(c-1)])) -(MADD a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && c>=7 -> (SUB a (SUBshiftLL x x [log2(c+1)])) -(MADD a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) -> (SUBshiftLL a (SUBshiftLL x x [2]) [log2(c/3)]) -(MADD a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) -> (ADDshiftLL a (ADDshiftLL x x [2]) [log2(c/5)]) -(MADD a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) -> (SUBshiftLL a (SUBshiftLL x x [3]) [log2(c/7)]) -(MADD a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) -> (ADDshiftLL a (ADDshiftLL x x [3]) [log2(c/9)]) -(MADDW a x (MOVDconst [c])) && int32(c)==-1 -> (SUB a x) -(MADDW a _ (MOVDconst [c])) && int32(c)==0 -> a -(MADDW a x (MOVDconst [c])) && int32(c)==1 -> (ADD a x) -(MADDW a x (MOVDconst [c])) && isPowerOfTwo(c) -> (ADDshiftLL a x [log2(c)]) -(MADDW a x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c)>=3 -> (ADD a (ADDshiftLL x x [log2(c-1)])) -(MADDW a x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c)>=7 -> (SUB a (SUBshiftLL x x [log2(c+1)])) -(MADDW a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (SUBshiftLL a (SUBshiftLL x x [2]) [log2(c/3)]) -(MADDW a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (ADDshiftLL a (ADDshiftLL x x [2]) [log2(c/5)]) -(MADDW a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (SUBshiftLL a (SUBshiftLL x x [3]) [log2(c/7)]) -(MADDW a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (ADDshiftLL a (ADDshiftLL x x [3]) [log2(c/9)]) +(MADD a x (MOVDconst [-1])) => (SUB a x) +(MADD a _ (MOVDconst [0])) => a +(MADD a x (MOVDconst [1])) => (ADD a x) +(MADD a x (MOVDconst [c])) && isPowerOfTwo64(c) => (ADDshiftLL a x [log64(c)]) +(MADD a x (MOVDconst [c])) && isPowerOfTwo64(c-1) && c>=3 => (ADD a (ADDshiftLL x x [log64(c-1)])) +(MADD a x (MOVDconst [c])) && isPowerOfTwo64(c+1) && c>=7 => (SUB a (SUBshiftLL x x [log64(c+1)])) +(MADD a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo64(c/3) => (SUBshiftLL a (SUBshiftLL x x [2]) [log64(c/3)]) +(MADD a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo64(c/5) => (ADDshiftLL a (ADDshiftLL x x [2]) [log64(c/5)]) +(MADD a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo64(c/7) => (SUBshiftLL a (SUBshiftLL x x [3]) [log64(c/7)]) +(MADD a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo64(c/9) => (ADDshiftLL a (ADDshiftLL x x [3]) [log64(c/9)]) -(MADDW a (MOVDconst [c]) x) && int32(c)==-1 -> (SUB a x) -(MADDW a (MOVDconst [c]) _) && int32(c)==0 -> a -(MADDW a (MOVDconst [c]) x) && int32(c)==1 -> (ADD a x) -(MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c) -> (ADDshiftLL a x [log2(c)]) -(MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && int32(c)>=3 -> (ADD a (ADDshiftLL x x [log2(c-1)])) -(MADDW a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && int32(c)>=7 -> (SUB a (SUBshiftLL x x [log2(c+1)])) -(MADDW a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (SUBshiftLL a (SUBshiftLL x x [2]) [log2(c/3)]) -(MADDW a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (ADDshiftLL a (ADDshiftLL x x [2]) [log2(c/5)]) -(MADDW a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (SUBshiftLL a (SUBshiftLL x x [3]) [log2(c/7)]) -(MADDW a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (ADDshiftLL a (ADDshiftLL x x [3]) [log2(c/9)]) +(MADD a (MOVDconst [-1]) x) => (SUB a x) +(MADD a (MOVDconst [0]) _) => a +(MADD a (MOVDconst [1]) x) => (ADD a x) +(MADD a (MOVDconst [c]) x) && isPowerOfTwo64(c) => (ADDshiftLL a x [log64(c)]) +(MADD a (MOVDconst [c]) x) && isPowerOfTwo64(c-1) && c>=3 => (ADD a (ADDshiftLL x x [log64(c-1)])) +(MADD a (MOVDconst [c]) x) && isPowerOfTwo64(c+1) && c>=7 => (SUB a (SUBshiftLL x x [log64(c+1)])) +(MADD a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo64(c/3) => (SUBshiftLL a (SUBshiftLL x x [2]) [log64(c/3)]) +(MADD a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo64(c/5) => (ADDshiftLL a (ADDshiftLL x x [2]) [log64(c/5)]) +(MADD a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo64(c/7) => (SUBshiftLL a (SUBshiftLL x x [3]) [log64(c/7)]) +(MADD a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo64(c/9) => (ADDshiftLL a (ADDshiftLL x x [3]) [log64(c/9)]) -(MSUB a x (MOVDconst [-1])) -> (ADD a x) -(MSUB a _ (MOVDconst [0])) -> a -(MSUB a x (MOVDconst [1])) -> (SUB a x) -(MSUB a x (MOVDconst [c])) && isPowerOfTwo(c) -> (SUBshiftLL a x [log2(c)]) -(MSUB a x (MOVDconst [c])) && isPowerOfTwo(c-1) && c>=3 -> (SUB a (ADDshiftLL x x [log2(c-1)])) -(MSUB a x (MOVDconst [c])) && isPowerOfTwo(c+1) && c>=7 -> (ADD a (SUBshiftLL x x [log2(c+1)])) -(MSUB a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) -> (ADDshiftLL a (SUBshiftLL x x [2]) [log2(c/3)]) -(MSUB a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) -> (SUBshiftLL a (ADDshiftLL x x [2]) [log2(c/5)]) -(MSUB a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) -> (ADDshiftLL a (SUBshiftLL x x [3]) [log2(c/7)]) -(MSUB a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) -> (SUBshiftLL a (ADDshiftLL x x [3]) [log2(c/9)]) +(MADDW a x (MOVDconst [c])) && int32(c)==-1 => (SUB a x) +(MADDW a _ (MOVDconst [c])) && int32(c)==0 => a +(MADDW a x (MOVDconst [c])) && int32(c)==1 => (ADD a x) +(MADDW a x (MOVDconst [c])) && isPowerOfTwo64(c) => (ADDshiftLL a x [log64(c)]) +(MADDW a x (MOVDconst [c])) && isPowerOfTwo64(c-1) && int32(c)>=3 => (ADD a (ADDshiftLL x x [log64(c-1)])) +(MADDW a x (MOVDconst [c])) && isPowerOfTwo64(c+1) && int32(c)>=7 => (SUB a (SUBshiftLL x x [log64(c+1)])) +(MADDW a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo64(c/3) && is32Bit(c) => (SUBshiftLL a (SUBshiftLL x x [2]) [log64(c/3)]) +(MADDW a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo64(c/5) && is32Bit(c) => (ADDshiftLL a (ADDshiftLL x x [2]) [log64(c/5)]) +(MADDW a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo64(c/7) && is32Bit(c) => (SUBshiftLL a (SUBshiftLL x x [3]) [log64(c/7)]) +(MADDW a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo64(c/9) && is32Bit(c) => (ADDshiftLL a (ADDshiftLL x x [3]) [log64(c/9)]) -(MSUB a (MOVDconst [-1]) x) -> (ADD a x) -(MSUB a (MOVDconst [0]) _) -> a -(MSUB a (MOVDconst [1]) x) -> (SUB a x) -(MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c) -> (SUBshiftLL a x [log2(c)]) -(MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && c>=3 -> (SUB a (ADDshiftLL x x [log2(c-1)])) -(MSUB a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && c>=7 -> (ADD a (SUBshiftLL x x [log2(c+1)])) -(MSUB a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) -> (ADDshiftLL a (SUBshiftLL x x [2]) [log2(c/3)]) -(MSUB a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) -> (SUBshiftLL a (ADDshiftLL x x [2]) [log2(c/5)]) -(MSUB a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) -> (ADDshiftLL a (SUBshiftLL x x [3]) [log2(c/7)]) -(MSUB a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) -> (SUBshiftLL a (ADDshiftLL x x [3]) [log2(c/9)]) +(MADDW a (MOVDconst [c]) x) && int32(c)==-1 => (SUB a x) +(MADDW a (MOVDconst [c]) _) && int32(c)==0 => a +(MADDW a (MOVDconst [c]) x) && int32(c)==1 => (ADD a x) +(MADDW a (MOVDconst [c]) x) && isPowerOfTwo64(c) => (ADDshiftLL a x [log64(c)]) +(MADDW a (MOVDconst [c]) x) && isPowerOfTwo64(c-1) && int32(c)>=3 => (ADD a (ADDshiftLL x x [log64(c-1)])) +(MADDW a (MOVDconst [c]) x) && isPowerOfTwo64(c+1) && int32(c)>=7 => (SUB a (SUBshiftLL x x [log64(c+1)])) +(MADDW a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo64(c/3) && is32Bit(c) => (SUBshiftLL a (SUBshiftLL x x [2]) [log64(c/3)]) +(MADDW a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo64(c/5) && is32Bit(c) => (ADDshiftLL a (ADDshiftLL x x [2]) [log64(c/5)]) +(MADDW a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo64(c/7) && is32Bit(c) => (SUBshiftLL a (SUBshiftLL x x [3]) [log64(c/7)]) +(MADDW a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo64(c/9) && is32Bit(c) => (ADDshiftLL a (ADDshiftLL x x [3]) [log64(c/9)]) -(MSUBW a x (MOVDconst [c])) && int32(c)==-1 -> (ADD a x) -(MSUBW a _ (MOVDconst [c])) && int32(c)==0 -> a -(MSUBW a x (MOVDconst [c])) && int32(c)==1 -> (SUB a x) -(MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c) -> (SUBshiftLL a x [log2(c)]) -(MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c-1) && int32(c)>=3 -> (SUB a (ADDshiftLL x x [log2(c-1)])) -(MSUBW a x (MOVDconst [c])) && isPowerOfTwo(c+1) && int32(c)>=7 -> (ADD a (SUBshiftLL x x [log2(c+1)])) -(MSUBW a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (ADDshiftLL a (SUBshiftLL x x [2]) [log2(c/3)]) -(MSUBW a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (SUBshiftLL a (ADDshiftLL x x [2]) [log2(c/5)]) -(MSUBW a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (ADDshiftLL a (SUBshiftLL x x [3]) [log2(c/7)]) -(MSUBW a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (SUBshiftLL a (ADDshiftLL x x [3]) [log2(c/9)]) +(MSUB a x (MOVDconst [-1])) => (ADD a x) +(MSUB a _ (MOVDconst [0])) => a +(MSUB a x (MOVDconst [1])) => (SUB a x) +(MSUB a x (MOVDconst [c])) && isPowerOfTwo64(c) => (SUBshiftLL a x [log64(c)]) +(MSUB a x (MOVDconst [c])) && isPowerOfTwo64(c-1) && c>=3 => (SUB a (ADDshiftLL x x [log64(c-1)])) +(MSUB a x (MOVDconst [c])) && isPowerOfTwo64(c+1) && c>=7 => (ADD a (SUBshiftLL x x [log64(c+1)])) +(MSUB a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo64(c/3) => (ADDshiftLL a (SUBshiftLL x x [2]) [log64(c/3)]) +(MSUB a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo64(c/5) => (SUBshiftLL a (ADDshiftLL x x [2]) [log64(c/5)]) +(MSUB a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo64(c/7) => (ADDshiftLL a (SUBshiftLL x x [3]) [log64(c/7)]) +(MSUB a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo64(c/9) => (SUBshiftLL a (ADDshiftLL x x [3]) [log64(c/9)]) -(MSUBW a (MOVDconst [c]) x) && int32(c)==-1 -> (ADD a x) -(MSUBW a (MOVDconst [c]) _) && int32(c)==0 -> a -(MSUBW a (MOVDconst [c]) x) && int32(c)==1 -> (SUB a x) -(MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c) -> (SUBshiftLL a x [log2(c)]) -(MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c-1) && int32(c)>=3 -> (SUB a (ADDshiftLL x x [log2(c-1)])) -(MSUBW a (MOVDconst [c]) x) && isPowerOfTwo(c+1) && int32(c)>=7 -> (ADD a (SUBshiftLL x x [log2(c+1)])) -(MSUBW a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo(c/3) && is32Bit(c) -> (ADDshiftLL a (SUBshiftLL x x [2]) [log2(c/3)]) -(MSUBW a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo(c/5) && is32Bit(c) -> (SUBshiftLL a (ADDshiftLL x x [2]) [log2(c/5)]) -(MSUBW a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo(c/7) && is32Bit(c) -> (ADDshiftLL a (SUBshiftLL x x [3]) [log2(c/7)]) -(MSUBW a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo(c/9) && is32Bit(c) -> (SUBshiftLL a (ADDshiftLL x x [3]) [log2(c/9)]) +(MSUB a (MOVDconst [-1]) x) => (ADD a x) +(MSUB a (MOVDconst [0]) _) => a +(MSUB a (MOVDconst [1]) x) => (SUB a x) +(MSUB a (MOVDconst [c]) x) && isPowerOfTwo64(c) => (SUBshiftLL a x [log64(c)]) +(MSUB a (MOVDconst [c]) x) && isPowerOfTwo64(c-1) && c>=3 => (SUB a (ADDshiftLL x x [log64(c-1)])) +(MSUB a (MOVDconst [c]) x) && isPowerOfTwo64(c+1) && c>=7 => (ADD a (SUBshiftLL x x [log64(c+1)])) +(MSUB a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo64(c/3) => (ADDshiftLL a (SUBshiftLL x x [2]) [log64(c/3)]) +(MSUB a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo64(c/5) => (SUBshiftLL a (ADDshiftLL x x [2]) [log64(c/5)]) +(MSUB a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo64(c/7) => (ADDshiftLL a (SUBshiftLL x x [3]) [log64(c/7)]) +(MSUB a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo64(c/9) => (SUBshiftLL a (ADDshiftLL x x [3]) [log64(c/9)]) + +(MSUBW a x (MOVDconst [c])) && int32(c)==-1 => (ADD a x) +(MSUBW a _ (MOVDconst [c])) && int32(c)==0 => a +(MSUBW a x (MOVDconst [c])) && int32(c)==1 => (SUB a x) +(MSUBW a x (MOVDconst [c])) && isPowerOfTwo64(c) => (SUBshiftLL a x [log64(c)]) +(MSUBW a x (MOVDconst [c])) && isPowerOfTwo64(c-1) && int32(c)>=3 => (SUB a (ADDshiftLL x x [log64(c-1)])) +(MSUBW a x (MOVDconst [c])) && isPowerOfTwo64(c+1) && int32(c)>=7 => (ADD a (SUBshiftLL x x [log64(c+1)])) +(MSUBW a x (MOVDconst [c])) && c%3 == 0 && isPowerOfTwo64(c/3) && is32Bit(c) => (ADDshiftLL a (SUBshiftLL x x [2]) [log64(c/3)]) +(MSUBW a x (MOVDconst [c])) && c%5 == 0 && isPowerOfTwo64(c/5) && is32Bit(c) => (SUBshiftLL a (ADDshiftLL x x [2]) [log64(c/5)]) +(MSUBW a x (MOVDconst [c])) && c%7 == 0 && isPowerOfTwo64(c/7) && is32Bit(c) => (ADDshiftLL a (SUBshiftLL x x [3]) [log64(c/7)]) +(MSUBW a x (MOVDconst [c])) && c%9 == 0 && isPowerOfTwo64(c/9) && is32Bit(c) => (SUBshiftLL a (ADDshiftLL x x [3]) [log64(c/9)]) + +(MSUBW a (MOVDconst [c]) x) && int32(c)==-1 => (ADD a x) +(MSUBW a (MOVDconst [c]) _) && int32(c)==0 => a +(MSUBW a (MOVDconst [c]) x) && int32(c)==1 => (SUB a x) +(MSUBW a (MOVDconst [c]) x) && isPowerOfTwo64(c) => (SUBshiftLL a x [log64(c)]) +(MSUBW a (MOVDconst [c]) x) && isPowerOfTwo64(c-1) && int32(c)>=3 => (SUB a (ADDshiftLL x x [log64(c-1)])) +(MSUBW a (MOVDconst [c]) x) && isPowerOfTwo64(c+1) && int32(c)>=7 => (ADD a (SUBshiftLL x x [log64(c+1)])) +(MSUBW a (MOVDconst [c]) x) && c%3 == 0 && isPowerOfTwo64(c/3) && is32Bit(c) => (ADDshiftLL a (SUBshiftLL x x [2]) [log64(c/3)]) +(MSUBW a (MOVDconst [c]) x) && c%5 == 0 && isPowerOfTwo64(c/5) && is32Bit(c) => (SUBshiftLL a (ADDshiftLL x x [2]) [log64(c/5)]) +(MSUBW a (MOVDconst [c]) x) && c%7 == 0 && isPowerOfTwo64(c/7) && is32Bit(c) => (ADDshiftLL a (SUBshiftLL x x [3]) [log64(c/7)]) +(MSUBW a (MOVDconst [c]) x) && c%9 == 0 && isPowerOfTwo64(c/9) && is32Bit(c) => (SUBshiftLL a (ADDshiftLL x x [3]) [log64(c/9)]) // div by constant -(UDIV x (MOVDconst [1])) -> x -(UDIV x (MOVDconst [c])) && isPowerOfTwo(c) -> (SRLconst [log2(c)] x) -(UDIVW x (MOVDconst [c])) && uint32(c)==1 -> x -(UDIVW x (MOVDconst [c])) && isPowerOfTwo(c) && is32Bit(c) -> (SRLconst [log2(c)] x) -(UMOD _ (MOVDconst [1])) -> (MOVDconst [0]) -(UMOD x (MOVDconst [c])) && isPowerOfTwo(c) -> (ANDconst [c-1] x) -(UMODW _ (MOVDconst [c])) && uint32(c)==1 -> (MOVDconst [0]) -(UMODW x (MOVDconst [c])) && isPowerOfTwo(c) && is32Bit(c) -> (ANDconst [c-1] x) +(UDIV x (MOVDconst [1])) => x +(UDIV x (MOVDconst [c])) && isPowerOfTwo64(c) => (SRLconst [log64(c)] x) +(UDIVW x (MOVDconst [c])) && uint32(c)==1 => x +(UDIVW x (MOVDconst [c])) && isPowerOfTwo64(c) && is32Bit(c) => (SRLconst [log64(c)] x) +(UMOD _ (MOVDconst [1])) => (MOVDconst [0]) +(UMOD x (MOVDconst [c])) && isPowerOfTwo64(c) => (ANDconst [c-1] x) +(UMODW _ (MOVDconst [c])) && uint32(c)==1 => (MOVDconst [0]) +(UMODW x (MOVDconst [c])) && isPowerOfTwo64(c) && is32Bit(c) => (ANDconst [c-1] x) // generic simplifications -(ADD x (NEG y)) -> (SUB x y) -(SUB x x) -> (MOVDconst [0]) -(AND x x) -> x -(OR x x) -> x -(XOR x x) -> (MOVDconst [0]) -(BIC x x) -> (MOVDconst [0]) -(EON x x) -> (MOVDconst [-1]) -(ORN x x) -> (MOVDconst [-1]) -(AND x (MVN y)) -> (BIC x y) -(XOR x (MVN y)) -> (EON x y) -(OR x (MVN y)) -> (ORN x y) -(MVN (XOR x y)) -> (EON x y) +(ADD x (NEG y)) => (SUB x y) +(SUB x x) => (MOVDconst [0]) +(AND x x) => x +(OR x x) => x +(XOR x x) => (MOVDconst [0]) +(BIC x x) => (MOVDconst [0]) +(EON x x) => (MOVDconst [-1]) +(ORN x x) => (MOVDconst [-1]) +(AND x (MVN y)) => (BIC x y) +(XOR x (MVN y)) => (EON x y) +(OR x (MVN y)) => (ORN x y) +(MVN (XOR x y)) => (EON x y) (CSEL [cc] x (MOVDconst [0]) flag) => (CSEL0 [cc] x flag) (CSEL [cc] (MOVDconst [0]) y flag) => (CSEL0 [arm64Negate(cc)] y flag) -(SUB x (SUB y z)) -> (SUB (ADD x z) y) -(SUB (SUB x y) z) -> (SUB x (ADD y z)) +(SUB x (SUB y z)) => (SUB (ADD x z) y) +(SUB (SUB x y) z) => (SUB x (ADD y z)) // remove redundant *const ops -(ADDconst [0] x) -> x -(SUBconst [0] x) -> x -(ANDconst [0] _) -> (MOVDconst [0]) -(ANDconst [-1] x) -> x -(ORconst [0] x) -> x -(ORconst [-1] _) -> (MOVDconst [-1]) -(XORconst [0] x) -> x -(XORconst [-1] x) -> (MVN x) +(ADDconst [0] x) => x +(SUBconst [0] x) => x +(ANDconst [0] _) => (MOVDconst [0]) +(ANDconst [-1] x) => x +(ORconst [0] x) => x +(ORconst [-1] _) => (MOVDconst [-1]) +(XORconst [0] x) => x +(XORconst [-1] x) => (MVN x) // generic constant folding -(ADDconst [c] (MOVDconst [d])) -> (MOVDconst [c+d]) -(ADDconst [c] (ADDconst [d] x)) -> (ADDconst [c+d] x) -(ADDconst [c] (SUBconst [d] x)) -> (ADDconst [c-d] x) -(SUBconst [c] (MOVDconst [d])) -> (MOVDconst [d-c]) -(SUBconst [c] (SUBconst [d] x)) -> (ADDconst [-c-d] x) -(SUBconst [c] (ADDconst [d] x)) -> (ADDconst [-c+d] x) -(SLLconst [c] (MOVDconst [d])) -> (MOVDconst [d< (MOVDconst [int64(uint64(d)>>uint64(c))]) -(SRAconst [c] (MOVDconst [d])) -> (MOVDconst [d>>uint64(c)]) -(MUL (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [c*d]) -(MULW (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [int64(int32(c)*int32(d))]) -(MNEG (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [-c*d]) -(MNEGW (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [-int64(int32(c)*int32(d))]) -(MADD (MOVDconst [c]) x y) -> (ADDconst [c] (MUL x y)) -(MADDW (MOVDconst [c]) x y) -> (ADDconst [c] (MULW x y)) -(MSUB (MOVDconst [c]) x y) -> (ADDconst [c] (MNEG x y)) -(MSUBW (MOVDconst [c]) x y) -> (ADDconst [c] (MNEGW x y)) -(MADD a (MOVDconst [c]) (MOVDconst [d])) -> (ADDconst [c*d] a) -(MADDW a (MOVDconst [c]) (MOVDconst [d])) -> (ADDconst [int64(int32(c)*int32(d))] a) -(MSUB a (MOVDconst [c]) (MOVDconst [d])) -> (SUBconst [c*d] a) -(MSUBW a (MOVDconst [c]) (MOVDconst [d])) -> (SUBconst [int64(int32(c)*int32(d))] a) -(DIV (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [c/d]) -(UDIV (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [int64(uint64(c)/uint64(d))]) -(DIVW (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [int64(int32(c)/int32(d))]) -(UDIVW (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [int64(uint32(c)/uint32(d))]) -(MOD (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [c%d]) -(UMOD (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [int64(uint64(c)%uint64(d))]) -(MODW (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [int64(int32(c)%int32(d))]) -(UMODW (MOVDconst [c]) (MOVDconst [d])) -> (MOVDconst [int64(uint32(c)%uint32(d))]) -(ANDconst [c] (MOVDconst [d])) -> (MOVDconst [c&d]) -(ANDconst [c] (ANDconst [d] x)) -> (ANDconst [c&d] x) -(ANDconst [c] (MOVWUreg x)) -> (ANDconst [c&(1<<32-1)] x) -(ANDconst [c] (MOVHUreg x)) -> (ANDconst [c&(1<<16-1)] x) -(ANDconst [c] (MOVBUreg x)) -> (ANDconst [c&(1<<8-1)] x) -(MOVWUreg (ANDconst [c] x)) -> (ANDconst [c&(1<<32-1)] x) -(MOVHUreg (ANDconst [c] x)) -> (ANDconst [c&(1<<16-1)] x) -(MOVBUreg (ANDconst [c] x)) -> (ANDconst [c&(1<<8-1)] x) -(ORconst [c] (MOVDconst [d])) -> (MOVDconst [c|d]) -(ORconst [c] (ORconst [d] x)) -> (ORconst [c|d] x) -(XORconst [c] (MOVDconst [d])) -> (MOVDconst [c^d]) -(XORconst [c] (XORconst [d] x)) -> (XORconst [c^d] x) -(MVN (MOVDconst [c])) -> (MOVDconst [^c]) -(NEG (MOVDconst [c])) -> (MOVDconst [-c]) -(MOVBreg (MOVDconst [c])) -> (MOVDconst [int64(int8(c))]) -(MOVBUreg (MOVDconst [c])) -> (MOVDconst [int64(uint8(c))]) -(MOVHreg (MOVDconst [c])) -> (MOVDconst [int64(int16(c))]) -(MOVHUreg (MOVDconst [c])) -> (MOVDconst [int64(uint16(c))]) -(MOVWreg (MOVDconst [c])) -> (MOVDconst [int64(int32(c))]) -(MOVWUreg (MOVDconst [c])) -> (MOVDconst [int64(uint32(c))]) -(MOVDreg (MOVDconst [c])) -> (MOVDconst [c]) +(ADDconst [c] (MOVDconst [d])) => (MOVDconst [c+d]) +(ADDconst [c] (ADDconst [d] x)) => (ADDconst [c+d] x) +(ADDconst [c] (SUBconst [d] x)) => (ADDconst [c-d] x) +(SUBconst [c] (MOVDconst [d])) => (MOVDconst [d-c]) +(SUBconst [c] (SUBconst [d] x)) => (ADDconst [-c-d] x) +(SUBconst [c] (ADDconst [d] x)) => (ADDconst [-c+d] x) +(SLLconst [c] (MOVDconst [d])) => (MOVDconst [d< (MOVDconst [int64(uint64(d)>>uint64(c))]) +(SRAconst [c] (MOVDconst [d])) => (MOVDconst [d>>uint64(c)]) +(MUL (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [c*d]) +(MULW (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [int64(int32(c)*int32(d))]) +(MNEG (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [-c*d]) +(MNEGW (MOVDconst [c]) (MOVDconst [d])) => (MOVDconst [-int64(int32(c)*int32(d))]) +(MADD (MOVDconst [c]) x y) => (ADDconst [c] (MUL x y)) +(MADDW (MOVDconst [c]) x y) => (ADDconst [c] (MULW x y)) +(MSUB (MOVDconst [c]) x y) => (ADDconst [c] (MNEG x y)) +(MSUBW (MOVDconst [c]) x y) => (ADDconst [c] (MNEGW x y)) +(MADD a (MOVDconst [c]) (MOVDconst [d])) => (ADDconst [c*d] a) +(MADDW a (MOVDconst [c]) (MOVDconst [d])) => (ADDconst [int64(int32(c)*int32(d))] a) +(MSUB a (MOVDconst [c]) (MOVDconst [d])) => (SUBconst [c*d] a) +(MSUBW a (MOVDconst [c]) (MOVDconst [d])) => (SUBconst [int64(int32(c)*int32(d))] a) +(DIV (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [c/d]) +(UDIV (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint64(c)/uint64(d))]) +(DIVW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(int32(c)/int32(d))]) +(UDIVW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(c)/uint32(d))]) +(MOD (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [c%d]) +(UMOD (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint64(c)%uint64(d))]) +(MODW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(int32(c)%int32(d))]) +(UMODW (MOVDconst [c]) (MOVDconst [d])) && d != 0 => (MOVDconst [int64(uint32(c)%uint32(d))]) +(ANDconst [c] (MOVDconst [d])) => (MOVDconst [c&d]) +(ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x) +(ANDconst [c] (MOVWUreg x)) => (ANDconst [c&(1<<32-1)] x) +(ANDconst [c] (MOVHUreg x)) => (ANDconst [c&(1<<16-1)] x) +(ANDconst [c] (MOVBUreg x)) => (ANDconst [c&(1<<8-1)] x) +(MOVWUreg (ANDconst [c] x)) => (ANDconst [c&(1<<32-1)] x) +(MOVHUreg (ANDconst [c] x)) => (ANDconst [c&(1<<16-1)] x) +(MOVBUreg (ANDconst [c] x)) => (ANDconst [c&(1<<8-1)] x) +(ORconst [c] (MOVDconst [d])) => (MOVDconst [c|d]) +(ORconst [c] (ORconst [d] x)) => (ORconst [c|d] x) +(XORconst [c] (MOVDconst [d])) => (MOVDconst [c^d]) +(XORconst [c] (XORconst [d] x)) => (XORconst [c^d] x) +(MVN (MOVDconst [c])) => (MOVDconst [^c]) +(NEG (MOVDconst [c])) => (MOVDconst [-c]) +(MOVBreg (MOVDconst [c])) => (MOVDconst [int64(int8(c))]) +(MOVBUreg (MOVDconst [c])) => (MOVDconst [int64(uint8(c))]) +(MOVHreg (MOVDconst [c])) => (MOVDconst [int64(int16(c))]) +(MOVHUreg (MOVDconst [c])) => (MOVDconst [int64(uint16(c))]) +(MOVWreg (MOVDconst [c])) => (MOVDconst [int64(int32(c))]) +(MOVWUreg (MOVDconst [c])) => (MOVDconst [int64(uint32(c))]) +(MOVDreg (MOVDconst [c])) => (MOVDconst [c]) // constant comparisons (CMPconst (MOVDconst [x]) [y]) => (FlagConstant [subFlags64(x,y)]) diff --git a/src/cmd/compile/internal/ssa/gen/ARM64Ops.go b/src/cmd/compile/internal/ssa/gen/ARM64Ops.go index e9af261a6ae..87db2b7c9d9 100644 --- a/src/cmd/compile/internal/ssa/gen/ARM64Ops.go +++ b/src/cmd/compile/internal/ssa/gen/ARM64Ops.go @@ -471,9 +471,9 @@ func init() { {name: "CSEL0", argLength: 2, reg: gp1flags1, asm: "CSEL", aux: "CCop"}, // auxint(flags) ? arg0 : 0 // function calls - {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", clobberFlags: true, call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("R26"), 0}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("R26"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem // pseudo-ops {name: "LoweredNilCheck", argLength: 2, reg: regInfo{inputs: []regMask{gpg}}, nilCheck: true, faultOnNilArg0: true}, // panic if arg0 is nil. arg1=mem. @@ -621,6 +621,12 @@ func init() { {name: "LoweredAtomicExchange64", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, {name: "LoweredAtomicExchange32", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + // atomic exchange variant. + // store arg1 to arg0. arg2=mem. returns . auxint must be zero. + // SWPALD Rarg1, (Rarg0), Rout + {name: "LoweredAtomicExchange64Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true}, + {name: "LoweredAtomicExchange32Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true}, + // atomic add. // *arg0 += arg1. arg2=mem. returns . auxint must be zero. // LDAXR (Rarg0), Rout @@ -654,14 +660,45 @@ func init() { {name: "LoweredAtomicCas64", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, {name: "LoweredAtomicCas32", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + // atomic compare and swap variant. + // arg0 = pointer, arg1 = old value, arg2 = new value, arg3 = memory. auxint must be zero. + // if *arg0 == arg1 { + // *arg0 = arg2 + // return (true, memory) + // } else { + // return (false, memory) + // } + // MOV Rarg1, Rtmp + // CASAL Rtmp, (Rarg0), Rarg2 + // CMP Rarg1, Rtmp + // CSET EQ, Rout + {name: "LoweredAtomicCas64Variant", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + {name: "LoweredAtomicCas32Variant", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + // atomic and/or. // *arg0 &= (|=) arg1. arg2=mem. returns . auxint must be zero. - // LDAXRB (Rarg0), Rout + // LDAXR (Rarg0), Rout // AND/OR Rarg1, Rout - // STLXRB Rout, (Rarg0), Rtmp + // STLXR Rout, (Rarg0), Rtmp // CBNZ Rtmp, -3(PC) {name: "LoweredAtomicAnd8", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "AND", typ: "(UInt8,Mem)", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + {name: "LoweredAtomicAnd32", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "AND", typ: "(UInt32,Mem)", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, {name: "LoweredAtomicOr8", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "ORR", typ: "(UInt8,Mem)", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + {name: "LoweredAtomicOr32", argLength: 3, reg: gpxchg, resultNotInArgs: true, asm: "ORR", typ: "(UInt32,Mem)", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + + // atomic and/or variant. + // *arg0 &= (|=) arg1. arg2=mem. returns . auxint must be zero. + // AND: + // MNV Rarg1, Rtemp + // LDANDALB Rtemp, (Rarg0), Rout + // AND Rarg1, Rout + // OR: + // LDORALB Rarg1, (Rarg0), Rout + // ORR Rarg1, Rout + {name: "LoweredAtomicAnd8Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, typ: "(UInt8,Mem)", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + {name: "LoweredAtomicAnd32Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, typ: "(UInt32,Mem)", faultOnNilArg0: true, hasSideEffects: true, unsafePoint: true}, + {name: "LoweredAtomicOr8Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, typ: "(UInt8,Mem)", faultOnNilArg0: true, hasSideEffects: true}, + {name: "LoweredAtomicOr32Variant", argLength: 3, reg: gpxchg, resultNotInArgs: true, typ: "(UInt32,Mem)", faultOnNilArg0: true, hasSideEffects: true}, // LoweredWB invokes runtime.gcWriteBarrier. arg0=destptr, arg1=srcptr, arg2=mem, aux=runtime.gcWriteBarrier // It saves all GP registers if necessary, diff --git a/src/cmd/compile/internal/ssa/gen/ARMOps.go b/src/cmd/compile/internal/ssa/gen/ARMOps.go index 068fecf74c0..70c789937aa 100644 --- a/src/cmd/compile/internal/ssa/gen/ARMOps.go +++ b/src/cmd/compile/internal/ssa/gen/ARMOps.go @@ -428,9 +428,9 @@ func init() { {name: "SRAcond", argLength: 3, reg: gp2flags1, asm: "SRA"}, // arg0 >> 31 if flags indicates HS, arg0 >> arg1 otherwise, signed shift, arg2=flags // function calls - {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", clobberFlags: true, call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("R7"), 0}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("R7"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem // pseudo-ops {name: "LoweredNilCheck", argLength: 2, reg: regInfo{inputs: []regMask{gpg}}, nilCheck: true, faultOnNilArg0: true}, // panic if arg0 is nil. arg1=mem. diff --git a/src/cmd/compile/internal/ssa/gen/MIPS.rules b/src/cmd/compile/internal/ssa/gen/MIPS.rules index 96feaf9234b..8ad2c90ac33 100644 --- a/src/cmd/compile/internal/ssa/gen/MIPS.rules +++ b/src/cmd/compile/internal/ssa/gen/MIPS.rules @@ -96,17 +96,17 @@ (Rsh8Ux16 x y) => (CMOVZ (SRL (ZeroExt8to32 x) (ZeroExt16to32 y) ) (MOVWconst [0]) (SGTUconst [32] (ZeroExt16to32 y))) (Rsh8Ux8 x y) => (CMOVZ (SRL (ZeroExt8to32 x) (ZeroExt8to32 y) ) (MOVWconst [0]) (SGTUconst [32] (ZeroExt8to32 y))) -(Rsh32x32 x y) => (SRA x ( CMOVZ y (MOVWconst [-1]) (SGTUconst [32] y))) -(Rsh32x16 x y) => (SRA x ( CMOVZ (ZeroExt16to32 y) (MOVWconst [-1]) (SGTUconst [32] (ZeroExt16to32 y)))) -(Rsh32x8 x y) => (SRA x ( CMOVZ (ZeroExt8to32 y) (MOVWconst [-1]) (SGTUconst [32] (ZeroExt8to32 y)))) +(Rsh32x32 x y) => (SRA x ( CMOVZ y (MOVWconst [31]) (SGTUconst [32] y))) +(Rsh32x16 x y) => (SRA x ( CMOVZ (ZeroExt16to32 y) (MOVWconst [31]) (SGTUconst [32] (ZeroExt16to32 y)))) +(Rsh32x8 x y) => (SRA x ( CMOVZ (ZeroExt8to32 y) (MOVWconst [31]) (SGTUconst [32] (ZeroExt8to32 y)))) -(Rsh16x32 x y) => (SRA (SignExt16to32 x) ( CMOVZ y (MOVWconst [-1]) (SGTUconst [32] y))) -(Rsh16x16 x y) => (SRA (SignExt16to32 x) ( CMOVZ (ZeroExt16to32 y) (MOVWconst [-1]) (SGTUconst [32] (ZeroExt16to32 y)))) -(Rsh16x8 x y) => (SRA (SignExt16to32 x) ( CMOVZ (ZeroExt8to32 y) (MOVWconst [-1]) (SGTUconst [32] (ZeroExt8to32 y)))) +(Rsh16x32 x y) => (SRA (SignExt16to32 x) ( CMOVZ y (MOVWconst [31]) (SGTUconst [32] y))) +(Rsh16x16 x y) => (SRA (SignExt16to32 x) ( CMOVZ (ZeroExt16to32 y) (MOVWconst [31]) (SGTUconst [32] (ZeroExt16to32 y)))) +(Rsh16x8 x y) => (SRA (SignExt16to32 x) ( CMOVZ (ZeroExt8to32 y) (MOVWconst [31]) (SGTUconst [32] (ZeroExt8to32 y)))) -(Rsh8x32 x y) => (SRA (SignExt16to32 x) ( CMOVZ y (MOVWconst [-1]) (SGTUconst [32] y))) -(Rsh8x16 x y) => (SRA (SignExt16to32 x) ( CMOVZ (ZeroExt16to32 y) (MOVWconst [-1]) (SGTUconst [32] (ZeroExt16to32 y)))) -(Rsh8x8 x y) => (SRA (SignExt16to32 x) ( CMOVZ (ZeroExt8to32 y) (MOVWconst [-1]) (SGTUconst [32] (ZeroExt8to32 y)))) +(Rsh8x32 x y) => (SRA (SignExt16to32 x) ( CMOVZ y (MOVWconst [31]) (SGTUconst [32] y))) +(Rsh8x16 x y) => (SRA (SignExt16to32 x) ( CMOVZ (ZeroExt16to32 y) (MOVWconst [31]) (SGTUconst [32] (ZeroExt16to32 y)))) +(Rsh8x8 x y) => (SRA (SignExt16to32 x) ( CMOVZ (ZeroExt8to32 y) (MOVWconst [31]) (SGTUconst [32] (ZeroExt8to32 y)))) // rotates (RotateLeft8 x (MOVWconst [c])) => (Or8 (Lsh8x32 x (MOVWconst [c&7])) (Rsh8Ux32 x (MOVWconst [-c&7]))) @@ -143,7 +143,7 @@ (Const(32|16|8) [val]) => (MOVWconst [int32(val)]) (Const(32|64)F ...) => (MOV(F|D)const ...) (ConstNil) => (MOVWconst [0]) -(ConstBool [b]) => (MOVWconst [int32(b2i(b))]) +(ConstBool [b]) => (MOVWconst [b2i32(b)]) // truncations // Because we ignore high parts of registers, truncates are just copies. @@ -383,6 +383,9 @@ (ANDconst [3] (XORconst [3] ptr)))))) mem) +(AtomicAnd32 ...) => (LoweredAtomicAnd ...) +(AtomicOr32 ...) => (LoweredAtomicOr ...) + // checks (NilCheck ...) => (LoweredNilCheck ...) @@ -459,36 +462,36 @@ (MOVWstorezero [off1] {sym} x:(ADDconst [off2] ptr) mem) && (is16Bit(int64(off1+off2)) || x.Uses == 1) => (MOVWstorezero [off1+off2] {sym} ptr mem) (MOVBload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVBload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVBUload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVBUload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVHload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHUload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVHUload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVWload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVWload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVFload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVFload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVFload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVDload [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVDload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVBstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => - (MOVBstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVHstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => - (MOVHstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVWstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => - (MOVWstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVFstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => - (MOVFstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVFstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVDstore [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) => - (MOVDstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVBstorezero [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVBstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHstorezero [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVHstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVWstorezero [off1] {sym1} (MOVWaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) => - (MOVWstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) // replace load from same location as preceding store with zero/sign extension (or copy in case of full width) (MOVBload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) => (MOVBreg x) @@ -564,10 +567,9 @@ (XOR x (MOVWconst [c])) => (XORconst [c] x) (NOR x (MOVWconst [c])) => (NORconst [c] x) -(SRA x (MOVWconst [c])) && c >= 32 => (SRAconst x [31]) -(SLL x (MOVWconst [c])) => (SLLconst x [c]) -(SRL x (MOVWconst [c])) => (SRLconst x [c]) -(SRA x (MOVWconst [c])) => (SRAconst x [c]) +(SLL x (MOVWconst [c])) => (SLLconst x [c&31]) +(SRL x (MOVWconst [c])) => (SRLconst x [c&31]) +(SRA x (MOVWconst [c])) => (SRAconst x [c&31]) (SGT (MOVWconst [c]) x) => (SGTconst [c] x) (SGTU (MOVWconst [c]) x) => (SGTUconst [c] x) @@ -581,13 +583,13 @@ (Select0 (MULTU (MOVWconst [1]) _ )) => (MOVWconst [0]) (Select1 (MULTU (MOVWconst [-1]) x )) => (NEG x) (Select0 (MULTU (MOVWconst [-1]) x )) => (CMOVZ (ADDconst [-1] x) (MOVWconst [0]) x) -(Select1 (MULTU (MOVWconst [c]) x )) && isPowerOfTwo(int64(uint32(c))) => (SLLconst [int32(log2uint32(int64(c)))] x) -(Select0 (MULTU (MOVWconst [c]) x )) && isPowerOfTwo(int64(uint32(c))) => (SRLconst [int32(32-log2uint32(int64(c)))] x) +(Select1 (MULTU (MOVWconst [c]) x )) && isPowerOfTwo64(int64(uint32(c))) => (SLLconst [int32(log2uint32(int64(c)))] x) +(Select0 (MULTU (MOVWconst [c]) x )) && isPowerOfTwo64(int64(uint32(c))) => (SRLconst [int32(32-log2uint32(int64(c)))] x) (MUL (MOVWconst [0]) _ ) => (MOVWconst [0]) (MUL (MOVWconst [1]) x ) => x (MUL (MOVWconst [-1]) x ) => (NEG x) -(MUL (MOVWconst [c]) x ) && isPowerOfTwo(int64(uint32(c))) => (SLLconst [int32(log2uint32(int64(c)))] x) +(MUL (MOVWconst [c]) x ) && isPowerOfTwo64(int64(uint32(c))) => (SLLconst [int32(log2uint32(int64(c)))] x) // generic simplifications (ADD x (NEG y)) => (SUB x y) @@ -624,10 +626,10 @@ (MUL (MOVWconst [c]) (MOVWconst [d])) => (MOVWconst [c*d]) (Select1 (MULTU (MOVWconst [c]) (MOVWconst [d]))) => (MOVWconst [int32(uint32(c)*uint32(d))]) (Select0 (MULTU (MOVWconst [c]) (MOVWconst [d]))) => (MOVWconst [int32((int64(uint32(c))*int64(uint32(d)))>>32)]) -(Select1 (DIV (MOVWconst [c]) (MOVWconst [d]))) => (MOVWconst [c/d]) -(Select1 (DIVU (MOVWconst [c]) (MOVWconst [d]))) => (MOVWconst [int32(uint32(c)/uint32(d))]) -(Select0 (DIV (MOVWconst [c]) (MOVWconst [d]))) => (MOVWconst [c%d]) -(Select0 (DIVU (MOVWconst [c]) (MOVWconst [d]))) => (MOVWconst [int32(uint32(c)%uint32(d))]) +(Select1 (DIV (MOVWconst [c]) (MOVWconst [d]))) && d != 0 => (MOVWconst [c/d]) +(Select1 (DIVU (MOVWconst [c]) (MOVWconst [d]))) && d != 0 => (MOVWconst [int32(uint32(c)/uint32(d))]) +(Select0 (DIV (MOVWconst [c]) (MOVWconst [d]))) && d != 0 => (MOVWconst [c%d]) +(Select0 (DIVU (MOVWconst [c]) (MOVWconst [d]))) && d != 0 => (MOVWconst [int32(uint32(c)%uint32(d))]) (ANDconst [c] (MOVWconst [d])) => (MOVWconst [c&d]) (ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x) (ORconst [c] (MOVWconst [d])) => (MOVWconst [c|d]) diff --git a/src/cmd/compile/internal/ssa/gen/MIPS64.rules b/src/cmd/compile/internal/ssa/gen/MIPS64.rules index e008ec87036..088c9b1ac44 100644 --- a/src/cmd/compile/internal/ssa/gen/MIPS64.rules +++ b/src/cmd/compile/internal/ssa/gen/MIPS64.rules @@ -462,44 +462,44 @@ (MOVVstorezero [off1] {sym} (ADDVconst [off2] ptr) mem) && is32Bit(int64(off1)+off2) => (MOVVstorezero [off1+int32(off2)] {sym} ptr mem) (MOVBload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVBload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVBUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVBUload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBUload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVHload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVHload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVHUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVHUload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHUload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVWload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVWload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVWUload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVWUload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWUload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVVload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVVload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVVload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVFload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVFload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVFload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVDload [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVDload [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVDload [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVBstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVBstore [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVBstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem) (MOVHstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVHstore [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVHstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem) (MOVWstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVWstore [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVWstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem) (MOVVstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVVstore [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVVstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem) (MOVFstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVFstore [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVFstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem) (MOVDstore [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVDstore [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVDstore [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr val mem) (MOVBstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVBstorezero [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBstorezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVHstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVHstorezero [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHstorezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVWstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVWstorezero [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWstorezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) (MOVVstorezero [off1] {sym1} (MOVVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVVstorezero [off1+int32(off2)] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVVstorezero [off1+int32(off2)] {mergeSym(sym1,sym2)} ptr mem) // store zero (MOVBstore [off] {sym} ptr (MOVVconst [0]) mem) => (MOVBstorezero [off] {sym} ptr mem) @@ -580,13 +580,13 @@ (Select1 (MULVU x (MOVVconst [-1]))) => (NEGV x) (Select1 (MULVU _ (MOVVconst [0]))) => (MOVVconst [0]) (Select1 (MULVU x (MOVVconst [1]))) => x -(Select1 (MULVU x (MOVVconst [c]))) && isPowerOfTwo(c) => (SLLVconst [log2(c)] x) +(Select1 (MULVU x (MOVVconst [c]))) && isPowerOfTwo64(c) => (SLLVconst [log64(c)] x) // div by constant (Select1 (DIVVU x (MOVVconst [1]))) => x -(Select1 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo(c) => (SRLVconst [log2(c)] x) +(Select1 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo64(c) => (SRLVconst [log64(c)] x) (Select0 (DIVVU _ (MOVVconst [1]))) => (MOVVconst [0]) // mod -(Select0 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo(c) => (ANDconst [c-1] x) // mod +(Select0 (DIVVU x (MOVVconst [c]))) && isPowerOfTwo64(c) => (ANDconst [c-1] x) // mod // generic simplifications (ADDV x (NEGV y)) => (SUBV x y) @@ -617,10 +617,10 @@ (SRLVconst [c] (MOVVconst [d])) => (MOVVconst [int64(uint64(d)>>uint64(c))]) (SRAVconst [c] (MOVVconst [d])) => (MOVVconst [d>>uint64(c)]) (Select1 (MULVU (MOVVconst [c]) (MOVVconst [d]))) => (MOVVconst [c*d]) -(Select1 (DIVV (MOVVconst [c]) (MOVVconst [d]))) => (MOVVconst [c/d]) -(Select1 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) => (MOVVconst [int64(uint64(c)/uint64(d))]) -(Select0 (DIVV (MOVVconst [c]) (MOVVconst [d]))) => (MOVVconst [c%d]) // mod -(Select0 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) => (MOVVconst [int64(uint64(c)%uint64(d))]) // mod +(Select1 (DIVV (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [c/d]) +(Select1 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [int64(uint64(c)/uint64(d))]) +(Select0 (DIVV (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [c%d]) // mod +(Select0 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [int64(uint64(c)%uint64(d))]) // mod (ANDconst [c] (MOVVconst [d])) => (MOVVconst [c&d]) (ANDconst [c] (ANDconst [d] x)) => (ANDconst [c&d] x) (ORconst [c] (MOVVconst [d])) => (MOVVconst [c|d]) diff --git a/src/cmd/compile/internal/ssa/gen/MIPS64Ops.go b/src/cmd/compile/internal/ssa/gen/MIPS64Ops.go index 5f00c080afb..e1e39335026 100644 --- a/src/cmd/compile/internal/ssa/gen/MIPS64Ops.go +++ b/src/cmd/compile/internal/ssa/gen/MIPS64Ops.go @@ -273,9 +273,9 @@ func init() { {name: "MOVDF", argLength: 1, reg: fp11, asm: "MOVDF"}, // float64 -> float32 // function calls - {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", clobberFlags: true, call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("R22"), 0}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("R22"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem // duffzero // arg0 = address of memory to zero diff --git a/src/cmd/compile/internal/ssa/gen/MIPSOps.go b/src/cmd/compile/internal/ssa/gen/MIPSOps.go index a5f6c8df544..75ab99ea26c 100644 --- a/src/cmd/compile/internal/ssa/gen/MIPSOps.go +++ b/src/cmd/compile/internal/ssa/gen/MIPSOps.go @@ -185,11 +185,11 @@ func init() { // shifts {name: "SLL", argLength: 2, reg: gp21, asm: "SLL"}, // arg0 << arg1, shift amount is mod 32 - {name: "SLLconst", argLength: 1, reg: gp11, asm: "SLL", aux: "Int32"}, // arg0 << auxInt + {name: "SLLconst", argLength: 1, reg: gp11, asm: "SLL", aux: "Int32"}, // arg0 << auxInt, shift amount must be 0 through 31 inclusive {name: "SRL", argLength: 2, reg: gp21, asm: "SRL"}, // arg0 >> arg1, unsigned, shift amount is mod 32 - {name: "SRLconst", argLength: 1, reg: gp11, asm: "SRL", aux: "Int32"}, // arg0 >> auxInt, unsigned + {name: "SRLconst", argLength: 1, reg: gp11, asm: "SRL", aux: "Int32"}, // arg0 >> auxInt, shift amount must be 0 through 31 inclusive {name: "SRA", argLength: 2, reg: gp21, asm: "SRA"}, // arg0 >> arg1, signed, shift amount is mod 32 - {name: "SRAconst", argLength: 1, reg: gp11, asm: "SRA", aux: "Int32"}, // arg0 >> auxInt, signed + {name: "SRAconst", argLength: 1, reg: gp11, asm: "SRA", aux: "Int32"}, // arg0 >> auxInt, signed, shift amount must be 0 through 31 inclusive {name: "CLZ", argLength: 1, reg: gp11, asm: "CLZ"}, @@ -255,9 +255,9 @@ func init() { {name: "MOVDF", argLength: 1, reg: fp11, asm: "MOVDF"}, // float64 -> float32 // function calls - {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", clobberFlags: true, call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("R22"), 0}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{gpsp, buildReg("R22"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem // atomic ops diff --git a/src/cmd/compile/internal/ssa/gen/PPC64.rules b/src/cmd/compile/internal/ssa/gen/PPC64.rules index e5fb1e98c29..c0640461729 100644 --- a/src/cmd/compile/internal/ssa/gen/PPC64.rules +++ b/src/cmd/compile/internal/ssa/gen/PPC64.rules @@ -79,6 +79,23 @@ (Abs ...) => (FABS ...) (FMA ...) => (FMADD ...) +// Lowering extension +// Note: we always extend to 64 bits even though some ops don't need that many result bits. +(SignExt8to(16|32|64) ...) => (MOVBreg ...) +(SignExt16to(32|64) ...) => (MOVHreg ...) +(SignExt32to64 ...) => (MOVWreg ...) + +(ZeroExt8to(16|32|64) ...) => (MOVBZreg ...) +(ZeroExt16to(32|64) ...) => (MOVHZreg ...) +(ZeroExt32to64 ...) => (MOVWZreg ...) + +(Trunc(16|32|64)to8 x) && isSigned(t) => (MOVBreg x) +(Trunc(16|32|64)to8 x) => (MOVBZreg x) +(Trunc(32|64)to16 x) && isSigned(t) => (MOVHreg x) +(Trunc(32|64)to16 x) => (MOVHZreg x) +(Trunc64to32 x) && isSigned(t) => (MOVWreg x) +(Trunc64to32 x) => (MOVWZreg x) + // Lowering constants (Const(64|32|16|8) [val]) => (MOVDconst [int64(val)]) (Const(32|64)F ...) => (FMOV(S|D)const ...) @@ -133,6 +150,31 @@ (ROTLW x (MOVDconst [c])) => (ROTLWconst x [c&31]) (ROTL x (MOVDconst [c])) => (ROTLconst x [c&63]) +// Combine rotate and mask operations +(ANDconst [m] (ROTLWconst [r] x)) && isPPC64WordRotateMask(m) => (RLWINM [encodePPC64RotateMask(r,m,32)] x) +(AND (MOVDconst [m]) (ROTLWconst [r] x)) && isPPC64WordRotateMask(m) => (RLWINM [encodePPC64RotateMask(r,m,32)] x) +(ANDconst [m] (ROTLW x r)) && isPPC64WordRotateMask(m) => (RLWNM [encodePPC64RotateMask(0,m,32)] x r) +(AND (MOVDconst [m]) (ROTLW x r)) && isPPC64WordRotateMask(m) => (RLWNM [encodePPC64RotateMask(0,m,32)] x r) + +// Note, any rotated word bitmask is still a valid word bitmask. +(ROTLWconst [r] (AND (MOVDconst [m]) x)) && isPPC64WordRotateMask(m) => (RLWINM [encodePPC64RotateMask(r,rotateLeft32(m,r),32)] x) +(ROTLWconst [r] (ANDconst [m] x)) && isPPC64WordRotateMask(m) => (RLWINM [encodePPC64RotateMask(r,rotateLeft32(m,r),32)] x) + +(ANDconst [m] (SRWconst x [s])) && mergePPC64RShiftMask(m,s,32) == 0 => (MOVDconst [0]) +(ANDconst [m] (SRWconst x [s])) && mergePPC64AndSrwi(m,s) != 0 => (RLWINM [mergePPC64AndSrwi(m,s)] x) +(AND (MOVDconst [m]) (SRWconst x [s])) && mergePPC64RShiftMask(m,s,32) == 0 => (MOVDconst [0]) +(AND (MOVDconst [m]) (SRWconst x [s])) && mergePPC64AndSrwi(m,s) != 0 => (RLWINM [mergePPC64AndSrwi(m,s)] x) + +(SRWconst (ANDconst [m] x) [s]) && mergePPC64RShiftMask(m>>uint(s),s,32) == 0 => (MOVDconst [0]) +(SRWconst (ANDconst [m] x) [s]) && mergePPC64AndSrwi(m>>uint(s),s) != 0 => (RLWINM [mergePPC64AndSrwi(m>>uint(s),s)] x) +(SRWconst (AND (MOVDconst [m]) x) [s]) && mergePPC64RShiftMask(m>>uint(s),s,32) == 0 => (MOVDconst [0]) +(SRWconst (AND (MOVDconst [m]) x) [s]) && mergePPC64AndSrwi(m>>uint(s),s) != 0 => (RLWINM [mergePPC64AndSrwi(m>>uint(s),s)] x) + +// Merge shift right + shift left and clear left (e.g for a table lookup) +(CLRLSLDI [c] (SRWconst [s] x)) && mergePPC64ClrlsldiSrw(int64(c),s) != 0 => (RLWINM [mergePPC64ClrlsldiSrw(int64(c),s)] x) +(SLDconst [l] (SRWconst [r] x)) && mergePPC64SldiSrw(l,r) != 0 => (RLWINM [mergePPC64SldiSrw(l,r)] x) +// The following reduction shows up frequently too. e.g b[(x>>14)&0xFF] +(CLRLSLDI [c] i:(RLWINM [s] x)) && mergePPC64ClrlsldiRlwinm(c,s) != 0 => (RLWINM [mergePPC64ClrlsldiRlwinm(c,s)] x) // large constant shifts (Lsh64x64 _ (MOVDconst [c])) && uint64(c) >= 64 => (MOVDconst [0]) @@ -780,6 +822,21 @@ (MOVWreg y:(MOVWZreg x)) => (MOVWreg x) (MOVWZreg y:(MOVWreg x)) => (MOVWZreg x) +// Truncate then logical then truncate: omit first, lesser or equal truncate +(MOVWZreg ((OR|XOR|AND) x (MOVWZreg y))) => (MOVWZreg ((OR|XOR|AND) x y)) +(MOVHZreg ((OR|XOR|AND) x (MOVWZreg y))) => (MOVHZreg ((OR|XOR|AND) x y)) +(MOVHZreg ((OR|XOR|AND) x (MOVHZreg y))) => (MOVHZreg ((OR|XOR|AND) x y)) +(MOVBZreg ((OR|XOR|AND) x (MOVWZreg y))) => (MOVBZreg ((OR|XOR|AND) x y)) +(MOVBZreg ((OR|XOR|AND) x (MOVHZreg y))) => (MOVBZreg ((OR|XOR|AND) x y)) +(MOVBZreg ((OR|XOR|AND) x (MOVBZreg y))) => (MOVBZreg ((OR|XOR|AND) x y)) + +(MOV(B|H|W)Zreg z:(ANDconst [c] (MOVBZload ptr x))) => z +(MOVBZreg z:(AND y (MOVBZload ptr x))) => z +(MOV(H|W)Zreg z:(ANDconst [c] (MOVHZload ptr x))) => z +(MOVHZreg z:(AND y (MOVHZload ptr x))) => z +(MOVWZreg z:(ANDconst [c] (MOVWZload ptr x))) => z +(MOVWZreg z:(AND y (MOVWZload ptr x))) => z + // Arithmetic constant ops (ADD x (MOVDconst [c])) && is32Bit(c) => (ADDconst [c] x) @@ -788,6 +845,9 @@ (SUB x (MOVDconst [c])) && is32Bit(-c) => (ADDconst [-c] x) (ADDconst [c] (MOVDaddr [d] {sym} x)) && is32Bit(c+int64(d)) => (MOVDaddr [int32(c+int64(d))] {sym} x) +(ADDconst [c] x:(SP)) && is32Bit(c) => (MOVDaddr [int32(c)] x) // so it is rematerializeable + +(MULL(W|D) x (MOVDconst [c])) && is16Bit(c) => (MULL(W|D)const [int32(c)] x) // Subtract from (with carry, but ignored) constant. // Note, these clobber the carry bit. @@ -829,48 +889,48 @@ // is only one use. (MOVBstore [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (MOVBstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVHstore [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (MOVHstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVWstore [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (MOVWstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVDstore [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) && (off1+off2)%4 == 0 => - (MOVDstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (FMOVSstore [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (FMOVSstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (FMOVDstore [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (FMOVDstore [off1+off2] {mergeSymTyped(sym1,sym2)} ptr val mem) + (FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem) (MOVBZload [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (MOVBZload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBZload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHload [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (MOVHload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHZload [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (MOVHZload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHZload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVWload [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) && (off1+off2)%4 == 0 => - (MOVWload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVWZload [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (MOVWZload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWZload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVDload [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) && (off1+off2)%4 == 0 => - (MOVDload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (FMOVSload [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (FMOVSload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (FMOVSload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (FMOVDload [off1] {sym1} p:(MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is16Bit(int64(off1+off2)) && (ptr.Op != OpSB || p.Uses == 1) => - (FMOVDload [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (FMOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem) // Fold offsets for loads. (FMOVSload [off1] {sym} (ADDconst [off2] ptr) mem) && is16Bit(int64(off1)+off2) => (FMOVSload [off1+int32(off2)] {sym} ptr mem) @@ -920,23 +980,23 @@ // Fold symbols into storezero (MOVDstorezero [off1] {sym1} p:(MOVDaddr [off2] {sym2} x) mem) && canMergeSym(sym1,sym2) && (x.Op != OpSB || p.Uses == 1) && (off1+off2)%4 == 0 => - (MOVDstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} x mem) + (MOVDstorezero [off1+off2] {mergeSym(sym1,sym2)} x mem) (MOVWstorezero [off1] {sym1} p:(MOVDaddr [off2] {sym2} x) mem) && canMergeSym(sym1,sym2) && (x.Op != OpSB || p.Uses == 1) => - (MOVWstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} x mem) + (MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} x mem) (MOVHstorezero [off1] {sym1} p:(MOVDaddr [off2] {sym2} x) mem) && canMergeSym(sym1,sym2) && (x.Op != OpSB || p.Uses == 1) => - (MOVHstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} x mem) + (MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} x mem) (MOVBstorezero [off1] {sym1} p:(MOVDaddr [off2] {sym2} x) mem) && canMergeSym(sym1,sym2) && (x.Op != OpSB || p.Uses == 1) => - (MOVBstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} x mem) + (MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} x mem) // atomic intrinsics (AtomicLoad(8|32|64|Ptr) ptr mem) => (LoweredAtomicLoad(8|32|64|Ptr) [1] ptr mem) -(AtomicLoadAcq32 ptr mem) => (LoweredAtomicLoad32 [0] ptr mem) +(AtomicLoadAcq(32|64) ptr mem) => (LoweredAtomicLoad(32|64) [0] ptr mem) (AtomicStore(8|32|64) ptr val mem) => (LoweredAtomicStore(8|32|64) [1] ptr val mem) -(AtomicStoreRel32 ptr val mem) => (LoweredAtomicStore32 [0] ptr val mem) +(AtomicStoreRel(32|64) ptr val mem) => (LoweredAtomicStore(32|64) [0] ptr val mem) //(AtomicStorePtrNoWB ptr val mem) => (STLR ptr val mem) (AtomicExchange(32|64) ...) => (LoweredAtomicExchange(32|64) ...) @@ -946,25 +1006,10 @@ (AtomicCompareAndSwap(32|64) ptr old new_ mem) => (LoweredAtomicCas(32|64) [1] ptr old new_ mem) (AtomicCompareAndSwapRel32 ptr old new_ mem) => (LoweredAtomicCas32 [0] ptr old new_ mem) -(AtomicAnd8 ...) => (LoweredAtomicAnd8 ...) -(AtomicOr8 ...) => (LoweredAtomicOr8 ...) - -// Lowering extension -// Note: we always extend to 64 bits even though some ops don't need that many result bits. -(SignExt8to(16|32|64) ...) => (MOVBreg ...) -(SignExt16to(32|64) ...) => (MOVHreg ...) -(SignExt32to64 ...) => (MOVWreg ...) - -(ZeroExt8to(16|32|64) ...) => (MOVBZreg ...) -(ZeroExt16to(32|64) ...) => (MOVHZreg ...) -(ZeroExt32to64 ...) => (MOVWZreg ...) - -(Trunc(16|32|64)to8 x) && isSigned(t) => (MOVBreg x) -(Trunc(16|32|64)to8 x) => (MOVBZreg x) -(Trunc(32|64)to16 x) && isSigned(t) => (MOVHreg x) -(Trunc(32|64)to16 x) => (MOVHZreg x) -(Trunc64to32 x) && isSigned(t) => (MOVWreg x) -(Trunc64to32 x) => (MOVWZreg x) +(AtomicAnd8 ...) => (LoweredAtomicAnd8 ...) +(AtomicAnd32 ...) => (LoweredAtomicAnd32 ...) +(AtomicOr8 ...) => (LoweredAtomicOr8 ...) +(AtomicOr32 ...) => (LoweredAtomicOr32 ...) (Slicemask x) => (SRADconst (NEG x) [63]) @@ -996,6 +1041,21 @@ (MOVWreg (MOVDconst [c])) => (MOVDconst [int64(int32(c))]) (MOVWZreg (MOVDconst [c])) => (MOVDconst [int64(uint32(c))]) +// Implement clrsldi and clrslwi extended mnemonics as described in +// ISA 3.0 section C.8. AuxInt field contains values needed for +// the instructions, packed together since there is only one available. +(SLDconst [c] z:(MOVBZreg x)) && c < 8 && z.Uses == 1 => (CLRLSLDI [newPPC64ShiftAuxInt(c,56,63,64)] x) +(SLDconst [c] z:(MOVHZreg x)) && c < 16 && z.Uses == 1 => (CLRLSLDI [newPPC64ShiftAuxInt(c,48,63,64)] x) +(SLDconst [c] z:(MOVWZreg x)) && c < 32 && z.Uses == 1 => (CLRLSLDI [newPPC64ShiftAuxInt(c,32,63,64)] x) + +(SLDconst [c] z:(ANDconst [d] x)) && z.Uses == 1 && isPPC64ValidShiftMask(d) && c <= (64-getPPC64ShiftMaskLength(d)) => (CLRLSLDI [newPPC64ShiftAuxInt(c,64-getPPC64ShiftMaskLength(d),63,64)] x) +(SLDconst [c] z:(AND (MOVDconst [d]) x)) && z.Uses == 1 && isPPC64ValidShiftMask(d) && c<=(64-getPPC64ShiftMaskLength(d)) => (CLRLSLDI [newPPC64ShiftAuxInt(c,64-getPPC64ShiftMaskLength(d),63,64)] x) +(SLWconst [c] z:(MOVBZreg x)) && z.Uses == 1 && c < 8 => (CLRLSLWI [newPPC64ShiftAuxInt(c,24,31,32)] x) +(SLWconst [c] z:(MOVHZreg x)) && z.Uses == 1 && c < 16 => (CLRLSLWI [newPPC64ShiftAuxInt(c,16,31,32)] x) +(SLWconst [c] z:(ANDconst [d] x)) && z.Uses == 1 && isPPC64ValidShiftMask(d) && c<=(32-getPPC64ShiftMaskLength(d)) => (CLRLSLWI [newPPC64ShiftAuxInt(c,32-getPPC64ShiftMaskLength(d),31,32)] x) +(SLWconst [c] z:(AND (MOVDconst [d]) x)) && z.Uses == 1 && isPPC64ValidShiftMask(d) && c<=(32-getPPC64ShiftMaskLength(d)) => (CLRLSLWI [newPPC64ShiftAuxInt(c,32-getPPC64ShiftMaskLength(d),31,32)] x) +// special case for power9 +(SL(W|D)const [c] z:(MOVWreg x)) && c < 32 && objabi.GOPPC64 >= 9 => (EXTSWSLconst [c] x) // Lose widening ops fed to stores (MOVBstore [off] {sym} ptr (MOV(B|BZ|H|HZ|W|WZ)reg x) mem) => (MOVBstore [off] {sym} ptr x mem) diff --git a/src/cmd/compile/internal/ssa/gen/PPC64Ops.go b/src/cmd/compile/internal/ssa/gen/PPC64Ops.go index 44f6a74c63e..f7198b90c32 100644 --- a/src/cmd/compile/internal/ssa/gen/PPC64Ops.go +++ b/src/cmd/compile/internal/ssa/gen/PPC64Ops.go @@ -137,6 +137,7 @@ func init() { gp01 = regInfo{inputs: nil, outputs: []regMask{gp}} gp11 = regInfo{inputs: []regMask{gp | sp | sb}, outputs: []regMask{gp}} gp21 = regInfo{inputs: []regMask{gp | sp | sb, gp | sp | sb}, outputs: []regMask{gp}} + gp21a0 = regInfo{inputs: []regMask{gp, gp | sp | sb}, outputs: []regMask{gp}} gp31 = regInfo{inputs: []regMask{gp | sp | sb, gp | sp | sb, gp | sp | sb}, outputs: []regMask{gp}} gp22 = regInfo{inputs: []regMask{gp | sp | sb, gp | sp | sb}, outputs: []regMask{gp, gp}} gp32 = regInfo{inputs: []regMask{gp | sp | sb, gp | sp | sb, gp | sp | sb}, outputs: []regMask{gp, gp}} @@ -181,6 +182,8 @@ func init() { {name: "MULLD", argLength: 2, reg: gp21, asm: "MULLD", typ: "Int64", commutative: true}, // arg0*arg1 (signed 64-bit) {name: "MULLW", argLength: 2, reg: gp21, asm: "MULLW", typ: "Int32", commutative: true}, // arg0*arg1 (signed 32-bit) + {name: "MULLDconst", argLength: 1, reg: gp11, asm: "MULLD", aux: "Int32", typ: "Int64"}, // arg0*auxInt (signed 64-bit) + {name: "MULLWconst", argLength: 1, reg: gp11, asm: "MULLW", aux: "Int32", typ: "Int64"}, // arg0*auxInt (signed 64-bit) {name: "MADDLD", argLength: 3, reg: gp31, asm: "MADDLD", typ: "Int64"}, // (arg0*arg1)+arg2 (signed 64-bit) {name: "MULHD", argLength: 2, reg: gp21, asm: "MULHD", commutative: true}, // (arg0 * arg1) >> 64, signed @@ -206,6 +209,11 @@ func init() { {name: "ROTL", argLength: 2, reg: gp21, asm: "ROTL"}, // arg0 rotate left by arg1 mod 64 {name: "ROTLW", argLength: 2, reg: gp21, asm: "ROTLW"}, // uint32(arg0) rotate left by arg1 mod 32 + // The following are ops to implement the extended mnemonics for shifts as described in section C.8 of the ISA. + // The constant shift values are packed into the aux int32. + {name: "RLDICL", argLength: 1, reg: gp11, asm: "RLDICL", aux: "Int32"}, // arg0 extract bits identified by shift params" + {name: "CLRLSLWI", argLength: 1, reg: gp11, asm: "CLRLSLWI", aux: "Int32"}, // + {name: "CLRLSLDI", argLength: 1, reg: gp11, asm: "CLRLSLDI", aux: "Int32"}, // {name: "LoweredAdd64Carry", argLength: 3, reg: gp32, resultNotInArgs: true}, // arg0 + arg1 + carry, returns (sum, carry) @@ -218,6 +226,11 @@ func init() { {name: "ROTLconst", argLength: 1, reg: gp11, asm: "ROTL", aux: "Int64"}, // arg0 rotate left by auxInt bits {name: "ROTLWconst", argLength: 1, reg: gp11, asm: "ROTLW", aux: "Int64"}, // uint32(arg0) rotate left by auxInt bits + {name: "EXTSWSLconst", argLength: 1, reg: gp11, asm: "EXTSWSLI", aux: "Int64"}, + + {name: "RLWINM", argLength: 1, reg: gp11, asm: "RLWNM", aux: "Int64"}, // Rotate and mask by immediate "rlwinm". encodePPC64RotateMask describes aux + {name: "RLWNM", argLength: 2, reg: gp21, asm: "RLWNM", aux: "Int64"}, // Rotate and mask by "rlwnm". encodePPC64RotateMask describes aux + {name: "RLWMI", argLength: 2, reg: gp21a0, asm: "RLWMI", aux: "Int64", resultInArg0: true}, // "rlwimi" similar aux encoding as above {name: "CNTLZD", argLength: 1, reg: gp11, asm: "CNTLZD", clobberFlags: true}, // count leading zeros {name: "CNTLZW", argLength: 1, reg: gp11, asm: "CNTLZW", clobberFlags: true}, // count leading zeros (32 bit) @@ -414,9 +427,9 @@ func init() { {name: "LoweredRound32F", argLength: 1, reg: fp11, resultInArg0: true, zeroWidth: true}, {name: "LoweredRound64F", argLength: 1, reg: fp11, resultInArg0: true, zeroWidth: true}, - {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", clobberFlags: true, call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{callptr, ctxt, 0}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{callptr}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{callptr, ctxt, 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{callptr}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem // large or unaligned zeroing // arg0 = address of memory to zero (in R3, changed as side effect) @@ -594,25 +607,22 @@ func init() { {name: "LoweredAtomicLoadPtr", argLength: 2, reg: gpload, typ: "Int64", aux: "Int64", clobberFlags: true, faultOnNilArg0: true}, // atomic add32, 64 - // SYNC + // LWSYNC // LDAR (Rarg0), Rout // ADD Rarg1, Rout // STDCCC Rout, (Rarg0) // BNE -3(PC) - // ISYNC // return new sum - {name: "LoweredAtomicAdd32", argLength: 3, reg: gpxchg, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true}, {name: "LoweredAtomicAdd64", argLength: 3, reg: gpxchg, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true}, // atomic exchange32, 64 - // SYNC + // LWSYNC // LDAR (Rarg0), Rout // STDCCC Rarg1, (Rarg0) // BNE -2(PC) // ISYNC // return old val - {name: "LoweredAtomicExchange32", argLength: 3, reg: gpxchg, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true}, {name: "LoweredAtomicExchange64", argLength: 3, reg: gpxchg, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true}, @@ -635,15 +645,16 @@ func init() { {name: "LoweredAtomicCas64", argLength: 4, reg: gpcas, resultNotInArgs: true, aux: "Int64", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true}, {name: "LoweredAtomicCas32", argLength: 4, reg: gpcas, resultNotInArgs: true, aux: "Int64", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true}, - // atomic 8 and/or. + // atomic 8/32 and/or. // *arg0 &= (|=) arg1. arg2=mem. returns memory. auxint must be zero. - // LBAR (Rarg0), Rtmp + // LBAR/LWAT (Rarg0), Rtmp // AND/OR Rarg1, Rtmp - // STBCCC Rtmp, (Rarg0), Rtmp + // STBCCC/STWCCC Rtmp, (Rarg0), Rtmp // BNE Rtmp, -3(PC) - {name: "LoweredAtomicAnd8", argLength: 3, reg: gpstore, asm: "AND", faultOnNilArg0: true, hasSideEffects: true}, + {name: "LoweredAtomicAnd32", argLength: 3, reg: gpstore, asm: "AND", faultOnNilArg0: true, hasSideEffects: true}, {name: "LoweredAtomicOr8", argLength: 3, reg: gpstore, asm: "OR", faultOnNilArg0: true, hasSideEffects: true}, + {name: "LoweredAtomicOr32", argLength: 3, reg: gpstore, asm: "OR", faultOnNilArg0: true, hasSideEffects: true}, // LoweredWB invokes runtime.gcWriteBarrier. arg0=destptr, arg1=srcptr, arg2=mem, aux=runtime.gcWriteBarrier // It preserves R0 through R17 (except special registers R1, R2, R11, R12, R13), g, and its arguments R20 and R21, diff --git a/src/cmd/compile/internal/ssa/gen/RISCV64.rules b/src/cmd/compile/internal/ssa/gen/RISCV64.rules index 9437c8e9d4b..4380a5efef7 100644 --- a/src/cmd/compile/internal/ssa/gen/RISCV64.rules +++ b/src/cmd/compile/internal/ssa/gen/RISCV64.rules @@ -3,17 +3,12 @@ // license that can be found in the LICENSE file. // Optimizations TODO: -// * Somehow track when values are already zero/signed-extended, avoid re-extending. // * Use SLTI and SLTIU for comparisons to constants, instead of SLT/SLTU with constants in registers -// * Find a more efficient way to do zero/sign extension than left+right shift. -// There are many other options (store then load-extend, LUI+ANDI for zero extend, special case 32->64, ...), -// but left+right shift is simple and uniform, and we don't have real hardware to do perf testing on anyway. // * Use the zero register instead of moving 0 into a register. // * Add rules to avoid generating a temp bool value for (If (SLT[U] ...) ...). // * Optimize left and right shift by simplifying SLTIU, Neg, and ADD for constants. // * Arrange for non-trivial Zero and Move lowerings to use aligned loads and stores. // * Eliminate zero immediate shifts, adds, etc. -// * Use a Duff's device for some moves and zeros. // * Avoid using Neq32 for writeBarrier.enabled checks. // Lowering arithmetic @@ -66,8 +61,8 @@ (Mod32u ...) => (REMUW ...) (Mod16 x y [false]) => (REMW (SignExt16to32 x) (SignExt16to32 y)) (Mod16u x y) => (REMUW (ZeroExt16to32 x) (ZeroExt16to32 y)) -(Mod8 x y) => (REMW (SignExt8to32 x) (SignExt8to32 y)) -(Mod8u x y) => (REMUW (ZeroExt8to32 x) (ZeroExt8to32 y)) +(Mod8 x y) => (REMW (SignExt8to32 x) (SignExt8to32 y)) +(Mod8u x y) => (REMUW (ZeroExt8to32 x) (ZeroExt8to32 y)) (And64 ...) => (AND ...) (And32 ...) => (AND ...) @@ -98,25 +93,21 @@ (Sqrt ...) => (FSQRTD ...) -// Zero and sign extension -// Shift left until the bits we want are at the top of the register. -// Then logical/arithmetic shift right for zero/sign extend. -// We always extend to 64 bits; there's no reason not to, -// and optimization rules can then collapse some extensions. +// Sign and zero extension. -(SignExt8to16 x) => (SRAI [56] (SLLI [56] x)) -(SignExt8to32 x) => (SRAI [56] (SLLI [56] x)) -(SignExt8to64 x) => (SRAI [56] (SLLI [56] x)) -(SignExt16to32 x) => (SRAI [48] (SLLI [48] x)) -(SignExt16to64 x) => (SRAI [48] (SLLI [48] x)) -(SignExt32to64 x) => (ADDIW [0] x) +(SignExt8to16 ...) => (MOVBreg ...) +(SignExt8to32 ...) => (MOVBreg ...) +(SignExt8to64 ...) => (MOVBreg ...) +(SignExt16to32 ...) => (MOVHreg ...) +(SignExt16to64 ...) => (MOVHreg ...) +(SignExt32to64 ...) => (MOVWreg ...) -(ZeroExt8to16 x) => (SRLI [56] (SLLI [56] x)) -(ZeroExt8to32 x) => (SRLI [56] (SLLI [56] x)) -(ZeroExt8to64 x) => (SRLI [56] (SLLI [56] x)) -(ZeroExt16to32 x) => (SRLI [48] (SLLI [48] x)) -(ZeroExt16to64 x) => (SRLI [48] (SLLI [48] x)) -(ZeroExt32to64 x) => (SRLI [32] (SLLI [32] x)) +(ZeroExt8to16 ...) => (MOVBUreg ...) +(ZeroExt8to32 ...) => (MOVBUreg ...) +(ZeroExt8to64 ...) => (MOVBUreg ...) +(ZeroExt16to32 ...) => (MOVHUreg ...) +(ZeroExt16to64 ...) => (MOVHUreg ...) +(ZeroExt32to64 ...) => (MOVWUreg ...) (Cvt32to32F ...) => (FCVTSW ...) (Cvt32to64F ...) => (FCVTDW ...) @@ -261,16 +252,16 @@ (EqPtr x y) => (SEQZ (SUB x y)) (Eq64 x y) => (SEQZ (SUB x y)) (Eq32 x y) => (SEQZ (SUBW x y)) -(Eq16 x y) => (SEQZ (ZeroExt16to64 (SUB x y))) -(Eq8 x y) => (SEQZ (ZeroExt8to64 (SUB x y))) +(Eq16 x y) => (SEQZ (SUB (ZeroExt16to64 x) (ZeroExt16to64 y))) +(Eq8 x y) => (SEQZ (SUB (ZeroExt8to64 x) (ZeroExt8to64 y))) (Eq64F ...) => (FEQD ...) (Eq32F ...) => (FEQS ...) (NeqPtr x y) => (SNEZ (SUB x y)) (Neq64 x y) => (SNEZ (SUB x y)) (Neq32 x y) => (SNEZ (SUBW x y)) -(Neq16 x y) => (SNEZ (ZeroExt16to64 (SUB x y))) -(Neq8 x y) => (SNEZ (ZeroExt8to64 (SUB x y))) +(Neq16 x y) => (SNEZ (SUB (ZeroExt16to64 x) (ZeroExt16to64 y))) +(Neq8 x y) => (SNEZ (SUB (ZeroExt8to64 x) (ZeroExt8to64 y))) (Neq64F ...) => (FNED ...) (Neq32F ...) => (FNES ...) @@ -291,42 +282,42 @@ (Store {t} ptr val mem) && t.Size() == 2 => (MOVHstore ptr val mem) (Store {t} ptr val mem) && t.Size() == 4 && !is32BitFloat(val.Type) => (MOVWstore ptr val mem) (Store {t} ptr val mem) && t.Size() == 8 && !is64BitFloat(val.Type) => (MOVDstore ptr val mem) -(Store {t} ptr val mem) && t.Size() == 4 && is32BitFloat(val.Type) => (FMOVWstore ptr val mem) -(Store {t} ptr val mem) && t.Size() == 8 && is64BitFloat(val.Type) => (FMOVDstore ptr val mem) +(Store {t} ptr val mem) && t.Size() == 4 && is32BitFloat(val.Type) => (FMOVWstore ptr val mem) +(Store {t} ptr val mem) && t.Size() == 8 && is64BitFloat(val.Type) => (FMOVDstore ptr val mem) // We need to fold MOVaddr into the LD/MOVDstore ops so that the live variable analysis // knows what variables are being read/written by the ops. (MOVBUload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVBUload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVBUload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVBload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVBload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVBload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVHUload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVHUload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVHUload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVHload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVHload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVHload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVWUload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVWUload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVWUload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVWload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVWload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVWload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVDload [off1] {sym1} (MOVaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVDload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVDload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVBstore [off1] {sym1} (MOVaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVBstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (MOVHstore [off1] {sym1} (MOVaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVHstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (MOVWstore [off1] {sym1} (MOVaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVWstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (MOVDstore [off1] {sym1} (MOVaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVDstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (MOVBstorezero [off1] {sym1} (MOVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVBstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVHstorezero [off1] {sym1} (MOVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVHstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVWstorezero [off1] {sym1} (MOVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVWstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVDstorezero [off1] {sym1} (MOVaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) && is32Bit(int64(off1)+int64(off2)) => - (MOVDstorezero [off1+off2] {mergeSymTyped(sym1,sym2)} ptr mem) + (MOVDstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem) (MOVBUload [off1] {sym} (ADDI [off2] base) mem) && is32Bit(int64(off1)+off2) => (MOVBUload [off1+int32(off2)] {sym} base mem) @@ -360,13 +351,66 @@ // with OffPtr -> ADDI. (ADDI [c] (MOVaddr [d] {s} x)) && is32Bit(c+int64(d)) => (MOVaddr [int32(c)+d] {s} x) -// Zeroing -// TODO: more optimized zeroing, including attempting to use aligned accesses. -(Zero [0] _ mem) => mem -(Zero [1] ptr mem) => (MOVBstore ptr (MOVBconst) mem) -(Zero [2] ptr mem) => (MOVHstore ptr (MOVHconst) mem) -(Zero [4] ptr mem) => (MOVWstore ptr (MOVWconst) mem) -(Zero [8] ptr mem) => (MOVDstore ptr (MOVDconst) mem) +// Small zeroing +(Zero [0] _ mem) => mem +(Zero [1] ptr mem) => (MOVBstore ptr (MOVBconst [0]) mem) +(Zero [2] {t} ptr mem) && t.Alignment()%2 == 0 => + (MOVHstore ptr (MOVHconst [0]) mem) +(Zero [2] ptr mem) => + (MOVBstore [1] ptr (MOVBconst [0]) + (MOVBstore ptr (MOVBconst [0]) mem)) +(Zero [4] {t} ptr mem) && t.Alignment()%4 == 0 => + (MOVWstore ptr (MOVWconst [0]) mem) +(Zero [4] {t} ptr mem) && t.Alignment()%2 == 0 => + (MOVHstore [2] ptr (MOVHconst [0]) + (MOVHstore ptr (MOVHconst [0]) mem)) +(Zero [4] ptr mem) => + (MOVBstore [3] ptr (MOVBconst [0]) + (MOVBstore [2] ptr (MOVBconst [0]) + (MOVBstore [1] ptr (MOVBconst [0]) + (MOVBstore ptr (MOVBconst [0]) mem)))) +(Zero [8] {t} ptr mem) && t.Alignment()%8 == 0 => + (MOVDstore ptr (MOVDconst [0]) mem) +(Zero [8] {t} ptr mem) && t.Alignment()%4 == 0 => + (MOVWstore [4] ptr (MOVWconst [0]) + (MOVWstore ptr (MOVWconst [0]) mem)) +(Zero [8] {t} ptr mem) && t.Alignment()%2 == 0 => + (MOVHstore [6] ptr (MOVHconst [0]) + (MOVHstore [4] ptr (MOVHconst [0]) + (MOVHstore [2] ptr (MOVHconst [0]) + (MOVHstore ptr (MOVHconst [0]) mem)))) + +(Zero [3] ptr mem) => + (MOVBstore [2] ptr (MOVBconst [0]) + (MOVBstore [1] ptr (MOVBconst [0]) + (MOVBstore ptr (MOVBconst [0]) mem))) +(Zero [6] {t} ptr mem) && t.Alignment()%2 == 0 => + (MOVHstore [4] ptr (MOVHconst [0]) + (MOVHstore [2] ptr (MOVHconst [0]) + (MOVHstore ptr (MOVHconst [0]) mem))) +(Zero [12] {t} ptr mem) && t.Alignment()%4 == 0 => + (MOVWstore [8] ptr (MOVWconst [0]) + (MOVWstore [4] ptr (MOVWconst [0]) + (MOVWstore ptr (MOVWconst [0]) mem))) +(Zero [16] {t} ptr mem) && t.Alignment()%8 == 0 => + (MOVDstore [8] ptr (MOVDconst [0]) + (MOVDstore ptr (MOVDconst [0]) mem)) +(Zero [24] {t} ptr mem) && t.Alignment()%8 == 0 => + (MOVDstore [16] ptr (MOVDconst [0]) + (MOVDstore [8] ptr (MOVDconst [0]) + (MOVDstore ptr (MOVDconst [0]) mem))) +(Zero [32] {t} ptr mem) && t.Alignment()%8 == 0 => + (MOVDstore [24] ptr (MOVDconst [0]) + (MOVDstore [16] ptr (MOVDconst [0]) + (MOVDstore [8] ptr (MOVDconst [0]) + (MOVDstore ptr (MOVDconst [0]) mem)))) + +// Medium 8-aligned zeroing uses a Duff's device +// 8 and 128 are magic constants, see runtime/mkduff.go +(Zero [s] {t} ptr mem) + && s%8 == 0 && s <= 8*128 + && t.Alignment()%8 == 0 && !config.noDuffDevice => + (DUFFZERO [8 * (128 - s/8)] ptr mem) // Generic zeroing uses a loop (Zero [s] {t} ptr mem) => @@ -378,7 +422,7 @@ (Convert ...) => (MOVconvert ...) // Checks -(IsNonNil p) => (NeqPtr (MOVDconst) p) +(IsNonNil p) => (NeqPtr (MOVDconst [0]) p) (IsInBounds ...) => (Less64U ...) (IsSliceInBounds ...) => (Leq64U ...) @@ -395,13 +439,66 @@ (PanicBounds [kind] x y mem) && boundsABI(kind) == 1 => (LoweredPanicBoundsB [kind] x y mem) (PanicBounds [kind] x y mem) && boundsABI(kind) == 2 => (LoweredPanicBoundsC [kind] x y mem) -// Moves -// TODO: more optimized moves, including attempting to use aligned accesses. -(Move [0] _ _ mem) => mem +// Small moves +(Move [0] _ _ mem) => mem (Move [1] dst src mem) => (MOVBstore dst (MOVBload src mem) mem) -(Move [2] dst src mem) => (MOVHstore dst (MOVHload src mem) mem) -(Move [4] dst src mem) => (MOVWstore dst (MOVWload src mem) mem) -(Move [8] dst src mem) => (MOVDstore dst (MOVDload src mem) mem) +(Move [2] {t} dst src mem) && t.Alignment()%2 == 0 => + (MOVHstore dst (MOVHload src mem) mem) +(Move [2] dst src mem) => + (MOVBstore [1] dst (MOVBload [1] src mem) + (MOVBstore dst (MOVBload src mem) mem)) +(Move [4] {t} dst src mem) && t.Alignment()%4 == 0 => + (MOVWstore dst (MOVWload src mem) mem) +(Move [4] {t} dst src mem) && t.Alignment()%2 == 0 => + (MOVHstore [2] dst (MOVHload [2] src mem) + (MOVHstore dst (MOVHload src mem) mem)) +(Move [4] dst src mem) => + (MOVBstore [3] dst (MOVBload [3] src mem) + (MOVBstore [2] dst (MOVBload [2] src mem) + (MOVBstore [1] dst (MOVBload [1] src mem) + (MOVBstore dst (MOVBload src mem) mem)))) +(Move [8] {t} dst src mem) && t.Alignment()%8 == 0 => + (MOVDstore dst (MOVDload src mem) mem) +(Move [8] {t} dst src mem) && t.Alignment()%4 == 0 => + (MOVWstore [4] dst (MOVWload [4] src mem) + (MOVWstore dst (MOVWload src mem) mem)) +(Move [8] {t} dst src mem) && t.Alignment()%2 == 0 => + (MOVHstore [6] dst (MOVHload [6] src mem) + (MOVHstore [4] dst (MOVHload [4] src mem) + (MOVHstore [2] dst (MOVHload [2] src mem) + (MOVHstore dst (MOVHload src mem) mem)))) + +(Move [3] dst src mem) => + (MOVBstore [2] dst (MOVBload [2] src mem) + (MOVBstore [1] dst (MOVBload [1] src mem) + (MOVBstore dst (MOVBload src mem) mem))) +(Move [6] {t} dst src mem) && t.Alignment()%2 == 0 => + (MOVHstore [4] dst (MOVHload [4] src mem) + (MOVHstore [2] dst (MOVHload [2] src mem) + (MOVHstore dst (MOVHload src mem) mem))) +(Move [12] {t} dst src mem) && t.Alignment()%4 == 0 => + (MOVWstore [8] dst (MOVWload [8] src mem) + (MOVWstore [4] dst (MOVWload [4] src mem) + (MOVWstore dst (MOVWload src mem) mem))) +(Move [16] {t} dst src mem) && t.Alignment()%8 == 0 => + (MOVDstore [8] dst (MOVDload [8] src mem) + (MOVDstore dst (MOVDload src mem) mem)) +(Move [24] {t} dst src mem) && t.Alignment()%8 == 0 => + (MOVDstore [16] dst (MOVDload [16] src mem) + (MOVDstore [8] dst (MOVDload [8] src mem) + (MOVDstore dst (MOVDload src mem) mem))) +(Move [32] {t} dst src mem) && t.Alignment()%8 == 0 => + (MOVDstore [24] dst (MOVDload [24] src mem) + (MOVDstore [16] dst (MOVDload [16] src mem) + (MOVDstore [8] dst (MOVDload [8] src mem) + (MOVDstore dst (MOVDload src mem) mem)))) + +// Medium 8-aligned move uses a Duff's device +// 16 and 128 are magic constants, see runtime/mkduff.go +(Move [s] {t} dst src mem) + && s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 + && !config.noDuffDevice && logLargeCopy(v, s) => + (DUFFCOPY [16 * (128 - s/8)] dst src mem) // Generic move uses a loop (Move [s] {t} dst src mem) && (s <= 16 || logLargeCopy(v, s)) => @@ -424,6 +521,8 @@ (OffPtr [off] ptr) && is32Bit(off) => (ADDI [off] ptr) (OffPtr [off] ptr) => (ADD (MOVDconst [off]) ptr) +// TODO(jsing): Check if we actually need MOV{B,H,W}const as most platforms +// use a single MOVDconst op. (Const8 ...) => (MOVBconst ...) (Const16 ...) => (MOVHconst ...) (Const32 ...) => (MOVWconst ...) @@ -501,6 +600,79 @@ (MOVWstore [off] {sym} ptr (MOVWconst [0]) mem) => (MOVWstorezero [off] {sym} ptr mem) (MOVDstore [off] {sym} ptr (MOVDconst [0]) mem) => (MOVDstorezero [off] {sym} ptr mem) +// Avoid sign/zero extension for consts. +(MOVBreg (MOVBconst [c])) => (MOVDconst [int64(c)]) +(MOVHreg (MOVBconst [c])) => (MOVDconst [int64(c)]) +(MOVHreg (MOVHconst [c])) => (MOVDconst [int64(c)]) +(MOVWreg (MOVBconst [c])) => (MOVDconst [int64(c)]) +(MOVWreg (MOVHconst [c])) => (MOVDconst [int64(c)]) +(MOVWreg (MOVWconst [c])) => (MOVDconst [int64(c)]) +(MOVBUreg (MOVBconst [c])) => (MOVDconst [int64(uint8(c))]) +(MOVHUreg (MOVBconst [c])) => (MOVDconst [int64(uint16(c))]) +(MOVHUreg (MOVHconst [c])) => (MOVDconst [int64(uint16(c))]) +(MOVWUreg (MOVBconst [c])) => (MOVDconst [int64(uint32(c))]) +(MOVWUreg (MOVHconst [c])) => (MOVDconst [int64(uint32(c))]) +(MOVWUreg (MOVWconst [c])) => (MOVDconst [int64(uint32(c))]) + +// Avoid sign/zero extension after properly typed load. +(MOVBreg x:(MOVBload _ _)) => (MOVDreg x) +(MOVHreg x:(MOVBload _ _)) => (MOVDreg x) +(MOVHreg x:(MOVBUload _ _)) => (MOVDreg x) +(MOVHreg x:(MOVHload _ _)) => (MOVDreg x) +(MOVWreg x:(MOVBload _ _)) => (MOVDreg x) +(MOVWreg x:(MOVBUload _ _)) => (MOVDreg x) +(MOVWreg x:(MOVHload _ _)) => (MOVDreg x) +(MOVWreg x:(MOVHUload _ _)) => (MOVDreg x) +(MOVWreg x:(MOVWload _ _)) => (MOVDreg x) +(MOVBUreg x:(MOVBUload _ _)) => (MOVDreg x) +(MOVHUreg x:(MOVBUload _ _)) => (MOVDreg x) +(MOVHUreg x:(MOVHUload _ _)) => (MOVDreg x) +(MOVWUreg x:(MOVBUload _ _)) => (MOVDreg x) +(MOVWUreg x:(MOVHUload _ _)) => (MOVDreg x) +(MOVWUreg x:(MOVWUload _ _)) => (MOVDreg x) + +// Fold double extensions. +(MOVBreg x:(MOVBreg _)) => (MOVDreg x) +(MOVHreg x:(MOVBreg _)) => (MOVDreg x) +(MOVHreg x:(MOVBUreg _)) => (MOVDreg x) +(MOVHreg x:(MOVHreg _)) => (MOVDreg x) +(MOVWreg x:(MOVBreg _)) => (MOVDreg x) +(MOVWreg x:(MOVBUreg _)) => (MOVDreg x) +(MOVWreg x:(MOVHreg _)) => (MOVDreg x) +(MOVWreg x:(MOVWreg _)) => (MOVDreg x) +(MOVBUreg x:(MOVBUreg _)) => (MOVDreg x) +(MOVHUreg x:(MOVBUreg _)) => (MOVDreg x) +(MOVHUreg x:(MOVHUreg _)) => (MOVDreg x) +(MOVWUreg x:(MOVBUreg _)) => (MOVDreg x) +(MOVWUreg x:(MOVHUreg _)) => (MOVDreg x) +(MOVWUreg x:(MOVWUreg _)) => (MOVDreg x) + +// Do not extend before store. +(MOVBstore [off] {sym} ptr (MOVBreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVHreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVWreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVBUreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVBstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVBstore [off] {sym} ptr x mem) +(MOVHstore [off] {sym} ptr (MOVHreg x) mem) => (MOVHstore [off] {sym} ptr x mem) +(MOVHstore [off] {sym} ptr (MOVWreg x) mem) => (MOVHstore [off] {sym} ptr x mem) +(MOVHstore [off] {sym} ptr (MOVHUreg x) mem) => (MOVHstore [off] {sym} ptr x mem) +(MOVHstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVHstore [off] {sym} ptr x mem) +(MOVWstore [off] {sym} ptr (MOVWreg x) mem) => (MOVWstore [off] {sym} ptr x mem) +(MOVWstore [off] {sym} ptr (MOVWUreg x) mem) => (MOVWstore [off] {sym} ptr x mem) + +// Replace extend after load with alternate load where possible. +(MOVBreg x:(MOVBUload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVBload [off] {sym} ptr mem) +(MOVHreg x:(MOVHUload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVHload [off] {sym} ptr mem) +(MOVWreg x:(MOVWUload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVWload [off] {sym} ptr mem) +(MOVBUreg x:(MOVBload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVBUload [off] {sym} ptr mem) +(MOVHUreg x:(MOVHload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVHUload [off] {sym} ptr mem) +(MOVWUreg x:(MOVWload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (MOVWUload [off] {sym} ptr mem) + +// If a register move has only 1 use, just use the same register without emitting instruction +// MOVnop does not emit an instruction, only for ensuring the type. +(MOVDreg x) && x.Uses == 1 => (MOVDnop x) + // Fold constant into immediate instructions where possible. (ADD (MOVBconst [val]) x) => (ADDI [int64(val)] x) (ADD (MOVHconst [val]) x) => (ADDI [int64(val)] x) diff --git a/src/cmd/compile/internal/ssa/gen/RISCV64Ops.go b/src/cmd/compile/internal/ssa/gen/RISCV64Ops.go index 8ab4abe04a0..f64319230b0 100644 --- a/src/cmd/compile/internal/ssa/gen/RISCV64Ops.go +++ b/src/cmd/compile/internal/ssa/gen/RISCV64Ops.go @@ -24,10 +24,11 @@ import ( // L = 64 bit int, used when the opcode starts with F const ( - riscv64REG_G = 4 + riscv64REG_G = 27 riscv64REG_CTXT = 20 riscv64REG_LR = 1 riscv64REG_SP = 2 + riscv64REG_TP = 4 riscv64REG_TMP = 31 riscv64REG_ZERO = 0 ) @@ -78,8 +79,8 @@ func init() { // Add general purpose registers to gpMask. switch r { - // ZERO, and TMP are not in any gp mask. - case riscv64REG_ZERO, riscv64REG_TMP: + // ZERO, TP and TMP are not in any gp mask. + case riscv64REG_ZERO, riscv64REG_TP, riscv64REG_TMP: case riscv64REG_G: gpgMask |= mask gpspsbgMask |= mask @@ -192,6 +193,17 @@ func init() { {name: "MOVWstorezero", argLength: 2, reg: gpstore0, aux: "SymOff", asm: "MOVW", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // 32 bits {name: "MOVDstorezero", argLength: 2, reg: gpstore0, aux: "SymOff", asm: "MOV", typ: "Mem", faultOnNilArg0: true, symEffect: "Write"}, // 64 bits + // Conversions + {name: "MOVBreg", argLength: 1, reg: gp11, asm: "MOVB"}, // move from arg0, sign-extended from byte + {name: "MOVHreg", argLength: 1, reg: gp11, asm: "MOVH"}, // move from arg0, sign-extended from half + {name: "MOVWreg", argLength: 1, reg: gp11, asm: "MOVW"}, // move from arg0, sign-extended from word + {name: "MOVDreg", argLength: 1, reg: gp11, asm: "MOV"}, // move from arg0 + {name: "MOVBUreg", argLength: 1, reg: gp11, asm: "MOVBU"}, // move from arg0, unsign-extended from byte + {name: "MOVHUreg", argLength: 1, reg: gp11, asm: "MOVHU"}, // move from arg0, unsign-extended from half + {name: "MOVWUreg", argLength: 1, reg: gp11, asm: "MOVWU"}, // move from arg0, unsign-extended from word + + {name: "MOVDnop", argLength: 1, reg: regInfo{inputs: []regMask{gpMask}, outputs: []regMask{gpMask}}, resultInArg0: true}, // nop, return arg0 in same register + // Shift ops {name: "SLL", argLength: 2, reg: gp21, asm: "SLL"}, // arg0 << (aux1 & 63) {name: "SRA", argLength: 2, reg: gp21, asm: "SRA"}, // arg0 >> (aux1 & 63), signed @@ -224,9 +236,47 @@ func init() { {name: "MOVconvert", argLength: 2, reg: gp11, asm: "MOV"}, // arg0, but converted to int/ptr as appropriate; arg1=mem // Calls - {name: "CALLstatic", argLength: 1, reg: call, aux: "SymOff", call: true, symEffect: "None"}, // call static function aux.(*gc.Sym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: callClosure, aux: "Int64", call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: callInter, aux: "Int64", call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: call, aux: "CallOff", call: true}, // call static function aux.(*gc.Sym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: callClosure, aux: "CallOff", call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: callInter, aux: "CallOff", call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + + // duffzero + // arg0 = address of memory to zero (in X10, changed as side effect) + // arg1 = mem + // auxint = offset into duffzero code to start executing + // X1 (link register) changed because of function call + // returns mem + { + name: "DUFFZERO", + aux: "Int64", + argLength: 2, + reg: regInfo{ + inputs: []regMask{regNamed["X10"]}, + clobbers: regNamed["X1"] | regNamed["X10"], + }, + typ: "Mem", + faultOnNilArg0: true, + }, + + // duffcopy + // arg0 = address of dst memory (in X11, changed as side effect) + // arg1 = address of src memory (in X10, changed as side effect) + // arg2 = mem + // auxint = offset into duffcopy code to start executing + // X1 (link register) changed because of function call + // returns mem + { + name: "DUFFCOPY", + aux: "Int64", + argLength: 3, + reg: regInfo{ + inputs: []regMask{regNamed["X11"], regNamed["X10"]}, + clobbers: regNamed["X1"] | regNamed["X10"] | regNamed["X11"], + }, + typ: "Mem", + faultOnNilArg0: true, + faultOnNilArg1: true, + }, // Generic moves and zeros diff --git a/src/cmd/compile/internal/ssa/gen/S390X.rules b/src/cmd/compile/internal/ssa/gen/S390X.rules index e564f638d38..39949edbc2b 100644 --- a/src/cmd/compile/internal/ssa/gen/S390X.rules +++ b/src/cmd/compile/internal/ssa/gen/S390X.rules @@ -198,6 +198,9 @@ (RXSBG {s390x.NewRotateParams(59, 60, 3)} (MOVDconst [3<<3]) ptr)) mem) +(AtomicAnd32 ...) => (LAN ...) +(AtomicOr32 ...) => (LAO ...) + // Lowering extension // Note: we always extend to 64 bits even though some ops don't need that many result bits. (SignExt8to(16|32|64) ...) => (MOVBreg ...) @@ -640,8 +643,18 @@ // equivalent to the leftmost 32 bits being set. // TODO(mundaym): modify the assembler to accept 64-bit values // and use isU32Bit(^c). -(AND x (MOVDconst [c])) && is32Bit(c) && c < 0 => (ANDconst [c] x) -(AND x (MOVDconst [c])) && is32Bit(c) && c >= 0 => (MOVWZreg (ANDWconst [int32(c)] x)) +(AND x (MOVDconst [c])) + && s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)) != nil + => (RISBGZ x {*s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c))}) +(AND x (MOVDconst [c])) + && is32Bit(c) + && c < 0 + => (ANDconst [c] x) +(AND x (MOVDconst [c])) + && is32Bit(c) + && c >= 0 + => (MOVWZreg (ANDWconst [int32(c)] x)) + (ANDW x (MOVDconst [c])) => (ANDWconst [int32(c)] x) ((AND|ANDW)const [c] ((AND|ANDW)const [d] x)) => ((AND|ANDW)const [c&d] x) @@ -650,14 +663,20 @@ ((OR|XOR)W x (MOVDconst [c])) => ((OR|XOR)Wconst [int32(c)] x) // Constant shifts. -(S(LD|RD|RAD|LW|RW|RAW) x (MOVDconst [c])) - => (S(LD|RD|RAD|LW|RW|RAW)const x [int8(c&63)]) +(S(LD|RD|RAD) x (MOVDconst [c])) => (S(LD|RD|RAD)const x [int8(c&63)]) +(S(LW|RW|RAW) x (MOVDconst [c])) && c&32 == 0 => (S(LW|RW|RAW)const x [int8(c&31)]) +(S(LW|RW) _ (MOVDconst [c])) && c&32 != 0 => (MOVDconst [0]) +(SRAW x (MOVDconst [c])) && c&32 != 0 => (SRAWconst x [31]) // Shifts only use the rightmost 6 bits of the shift value. +(S(LD|RD|RAD|LW|RW|RAW) x (RISBGZ y {r})) + && r.Amount == 0 + && r.OutMask()&63 == 63 + => (S(LD|RD|RAD|LW|RW|RAW) x y) (S(LD|RD|RAD|LW|RW|RAW) x (AND (MOVDconst [c]) y)) - => (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst [int32(c&63)] y)) + => (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst [int32(c&63)] y)) (S(LD|RD|RAD|LW|RW|RAW) x (ANDWconst [c] y)) && c&63 == 63 - => (S(LD|RD|RAD|LW|RW|RAW) x y) + => (S(LD|RD|RAD|LW|RW|RAW) x y) (SLD x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SLD x y) (SRD x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRD x y) (SRAD x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRAD x y) @@ -665,17 +684,13 @@ (SRW x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRW x y) (SRAW x (MOV(W|H|B|WZ|HZ|BZ)reg y)) => (SRAW x y) -// Constant rotate generation -(RLL x (MOVDconst [c])) => (RLLconst x [int8(c&31)]) -(RLLG x (MOVDconst [c])) => (RLLGconst x [int8(c&63)]) +// Match rotate by constant. +(RLLG x (MOVDconst [c])) => (RISBGZ x {s390x.NewRotateParams(0, 63, int8(c&63))}) +(RLL x (MOVDconst [c])) => (RLLconst x [int8(c&31)]) -(ADD (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (RLLGconst [c] x) -( OR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (RLLGconst [c] x) -(XOR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c => (RLLGconst [c] x) - -(ADDW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (RLLconst [c] x) -( ORW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (RLLconst [c] x) -(XORW (SLWconst x [c]) (SRWconst x [d])) && d == 32-c => (RLLconst [c] x) +// Match rotate by constant pattern. +((ADD|OR|XOR) (SLDconst x [c]) (SRDconst x [64-c])) => (RISBGZ x {s390x.NewRotateParams(0, 63, c)}) +((ADD|OR|XOR)W (SLWconst x [c]) (SRWconst x [32-c])) => (RLLconst x [c]) // Signed 64-bit comparison with immediate. (CMP x (MOVDconst [c])) && is32Bit(c) => (CMPconst x [int32(c)]) @@ -689,15 +704,97 @@ (CMP(W|WU) x (MOVDconst [c])) => (CMP(W|WU)const x [int32(c)]) (CMP(W|WU) (MOVDconst [c]) x) => (InvertFlags (CMP(W|WU)const x [int32(c)])) +// Match (x >> c) << d to 'rotate then insert selected bits [into zero]'. +(SLDconst (SRDconst x [c]) [d]) => (RISBGZ x {s390x.NewRotateParams(max8(0, c-d), 63-d, (d-c)&63)}) + +// Match (x << c) >> d to 'rotate then insert selected bits [into zero]'. +(SRDconst (SLDconst x [c]) [d]) => (RISBGZ x {s390x.NewRotateParams(d, min8(63, 63-c+d), (c-d)&63)}) + +// Absorb input zero extension into 'rotate then insert selected bits [into zero]'. +(RISBGZ (MOVWZreg x) {r}) && r.InMerge(0xffffffff) != nil => (RISBGZ x {*r.InMerge(0xffffffff)}) +(RISBGZ (MOVHZreg x) {r}) && r.InMerge(0x0000ffff) != nil => (RISBGZ x {*r.InMerge(0x0000ffff)}) +(RISBGZ (MOVBZreg x) {r}) && r.InMerge(0x000000ff) != nil => (RISBGZ x {*r.InMerge(0x000000ff)}) + +// Absorb 'rotate then insert selected bits [into zero]' into zero extension. +(MOVWZreg (RISBGZ x {r})) && r.OutMerge(0xffffffff) != nil => (RISBGZ x {*r.OutMerge(0xffffffff)}) +(MOVHZreg (RISBGZ x {r})) && r.OutMerge(0x0000ffff) != nil => (RISBGZ x {*r.OutMerge(0x0000ffff)}) +(MOVBZreg (RISBGZ x {r})) && r.OutMerge(0x000000ff) != nil => (RISBGZ x {*r.OutMerge(0x000000ff)}) + +// Absorb shift into 'rotate then insert selected bits [into zero]'. +// +// Any unsigned shift can be represented as a rotate and mask operation: +// +// x << c => RotateLeft64(x, c) & (^uint64(0) << c) +// x >> c => RotateLeft64(x, -c) & (^uint64(0) >> c) +// +// Therefore when a shift is used as the input to a rotate then insert +// selected bits instruction we can merge the two together. We just have +// to be careful that the resultant mask is representable (non-zero and +// contiguous). For example, assuming that x is variable and c, y and m +// are constants, a shift followed by a rotate then insert selected bits +// could be represented as: +// +// RotateLeft64(RotateLeft64(x, c) & (^uint64(0) << c), y) & m +// +// We can split the rotation by y into two, one rotate for x and one for +// the mask: +// +// RotateLeft64(RotateLeft64(x, c), y) & (RotateLeft64(^uint64(0) << c, y)) & m +// +// The rotations of x by c followed by y can then be combined: +// +// RotateLeft64(x, c+y) & (RotateLeft64(^uint64(0) << c, y)) & m +// ^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +// rotate mask +// +// To perform this optimization we therefore just need to check that it +// is valid to merge the shift mask (^(uint64(0)< (RISBGZ x {(*r.InMerge(^uint64(0)<>c) != nil => (RISBGZ x {(*r.InMerge(^uint64(0)>>c)).RotateLeft(-c)}) + +// Absorb 'rotate then insert selected bits [into zero]' into left shift. +(SLDconst (RISBGZ x {r}) [c]) + && s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask()) != nil + => (RISBGZ x {(*s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask())).RotateLeft(r.Amount)}) + +// Absorb 'rotate then insert selected bits [into zero]' into right shift. +(SRDconst (RISBGZ x {r}) [c]) + && s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask()) != nil + => (RISBGZ x {(*s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask())).RotateLeft(r.Amount)}) + +// Merge 'rotate then insert selected bits [into zero]' instructions together. +(RISBGZ (RISBGZ x {y}) {z}) + && z.InMerge(y.OutMask()) != nil + => (RISBGZ x {(*z.InMerge(y.OutMask())).RotateLeft(y.Amount)}) + +// Convert RISBGZ into 64-bit shift (helps CSE). +(RISBGZ x {r}) && r.End == 63 && r.Start == -r.Amount&63 => (SRDconst x [-r.Amount&63]) +(RISBGZ x {r}) && r.Start == 0 && r.End == 63-r.Amount => (SLDconst x [r.Amount]) + +// Optimize single bit isolation when it is known to be equivalent to +// the most significant bit due to mask produced by arithmetic shift. +// Simply isolate the most significant bit itself and place it in the +// correct position. +// +// Example: (int64(x) >> 63) & 0x8 -> RISBGZ $60, $60, $4, Rsrc, Rdst +(RISBGZ (SRADconst x [c]) {r}) + && r.Start == r.End // single bit selected + && (r.Start+r.Amount)&63 <= c // equivalent to most significant bit of x + => (RISBGZ x {s390x.NewRotateParams(r.Start, r.Start, -r.Start&63)}) + // Canonicalize the order of arguments to comparisons - helps with CSE. ((CMP|CMPW|CMPU|CMPWU) x y) && x.ID > y.ID => (InvertFlags ((CMP|CMPW|CMPU|CMPWU) y x)) -// Using MOV{W,H,B}Zreg instead of AND is cheaper. -(AND x (MOVDconst [0xFF])) => (MOVBZreg x) -(AND x (MOVDconst [0xFFFF])) => (MOVHZreg x) -(AND x (MOVDconst [0xFFFFFFFF])) => (MOVWZreg x) -(ANDWconst [0xFF] x) => (MOVBZreg x) -(ANDWconst [0xFFFF] x) => (MOVHZreg x) +// Use sign/zero extend instead of RISBGZ. +(RISBGZ x {r}) && r == s390x.NewRotateParams(56, 63, 0) => (MOVBZreg x) +(RISBGZ x {r}) && r == s390x.NewRotateParams(48, 63, 0) => (MOVHZreg x) +(RISBGZ x {r}) && r == s390x.NewRotateParams(32, 63, 0) => (MOVWZreg x) + +// Use sign/zero extend instead of ANDW. +(ANDWconst [0x00ff] x) => (MOVBZreg x) +(ANDWconst [0xffff] x) => (MOVHZreg x) // Strength reduce multiplication to the sum (or difference) of two powers of two. // @@ -770,21 +867,22 @@ // detect attempts to set/clear the sign bit // may need to be reworked when NIHH/OIHH are added -(SRDconst [1] (SLDconst [1] (LGDR x))) => (LGDR (LPDFR x)) -(LDGR (SRDconst [1] (SLDconst [1] x))) => (LPDFR (LDGR x)) -(AND (MOVDconst [^(-1<<63)]) (LGDR x)) => (LGDR (LPDFR x)) -(LDGR (AND (MOVDconst [^(-1<<63)]) x)) => (LPDFR (LDGR x)) -(OR (MOVDconst [-1<<63]) (LGDR x)) => (LGDR (LNDFR x)) -(LDGR (OR (MOVDconst [-1<<63]) x)) => (LNDFR (LDGR x)) +(RISBGZ (LGDR x) {r}) && r == s390x.NewRotateParams(1, 63, 0) => (LGDR (LPDFR x)) +(LDGR (RISBGZ x {r})) && r == s390x.NewRotateParams(1, 63, 0) => (LPDFR (LDGR x)) +(OR (MOVDconst [-1<<63]) (LGDR x)) => (LGDR (LNDFR x)) +(LDGR (OR (MOVDconst [-1<<63]) x)) => (LNDFR (LDGR x)) // detect attempts to set the sign bit with load (LDGR x:(ORload [off] {sym} (MOVDconst [-1<<63]) ptr mem)) && x.Uses == 1 && clobber(x) => @x.Block (LNDFR (LDGR (MOVDload [off] {sym} ptr mem))) // detect copysign -(OR (SLDconst [63] (SRDconst [63] (LGDR x))) (LGDR (LPDFR y))) => (LGDR (CPSDR y x)) -(OR (SLDconst [63] (SRDconst [63] (LGDR x))) (MOVDconst [c])) && c & -1<<63 == 0 => (LGDR (CPSDR (FMOVDconst [math.Float64frombits(uint64(c))]) x)) -(OR (AND (MOVDconst [-1<<63]) (LGDR x)) (LGDR (LPDFR y))) => (LGDR (CPSDR y x)) -(OR (AND (MOVDconst [-1<<63]) (LGDR x)) (MOVDconst [c])) && c & -1<<63 == 0 => (LGDR (CPSDR (FMOVDconst [math.Float64frombits(uint64(c))]) x)) +(OR (RISBGZ (LGDR x) {r}) (LGDR (LPDFR y))) + && r == s390x.NewRotateParams(0, 0, 0) + => (LGDR (CPSDR y x)) +(OR (RISBGZ (LGDR x) {r}) (MOVDconst [c])) + && c >= 0 + && r == s390x.NewRotateParams(0, 0, 0) + => (LGDR (CPSDR (FMOVDconst [math.Float64frombits(uint64(c))]) x)) (CPSDR y (FMOVDconst [c])) && !math.Signbit(c) => (LPDFR y) (CPSDR y (FMOVDconst [c])) && math.Signbit(c) => (LNDFR y) @@ -871,67 +969,67 @@ // loads/stores using PC-relative addressing directly must be aligned to the // size of the target. (MOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%8 == 0 && (off1+off2)%8 == 0)) => - (MOVDload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVDload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVWZload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) => - (MOVWZload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVWZload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVHZload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) => - (MOVHZload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVHZload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVBZload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVBZload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVBZload [off1+off2] {mergeSym(sym1,sym2)} base mem) (FMOVSload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (FMOVSload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (FMOVSload [off1+off2] {mergeSym(sym1,sym2)} base mem) (FMOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (FMOVDload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (FMOVDload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVWload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) => - (MOVWload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVWload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVHload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) => - (MOVHload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVHload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} base) mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVBload [off1+off2] {mergeSymTyped(sym1,sym2)} base mem) + (MOVBload [off1+off2] {mergeSym(sym1,sym2)} base mem) (MOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%8 == 0 && (off1+off2)%8 == 0)) => - (MOVDstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (MOVWstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%4 == 0 && (off1+off2)%4 == 0)) => - (MOVWstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (MOVHstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && (base.Op != OpSB || (t.IsPtr() && t.Elem().Alignment()%2 == 0 && (off1+off2)%2 == 0)) => - (MOVHstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (MOVBstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (MOVBstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (FMOVSstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) (FMOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} base) val mem) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) => - (FMOVDstore [off1+off2] {mergeSymTyped(sym1,sym2)} base val mem) + (FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} base val mem) -(ADDload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ADDload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(ADDWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ADDWload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(MULLDload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (MULLDload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(MULLWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (MULLWload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(SUBload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (SUBload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(SUBWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (SUBWload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) +(ADDload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ADDload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(ADDWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ADDWload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(MULLDload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (MULLDload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(MULLWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (MULLWload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(SUBload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (SUBload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(SUBWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (SUBWload [o1+o2] {mergeSym(s1, s2)} x ptr mem) -(ANDload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ANDload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(ANDWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ANDWload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(ORload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ORload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(ORWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ORWload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(XORload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (XORload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) -(XORWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (XORWload [o1+o2] {mergeSymTyped(s1, s2)} x ptr mem) +(ANDload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ANDload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(ANDWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ANDWload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(ORload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ORload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(ORWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (ORWload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(XORload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (XORload [o1+o2] {mergeSym(s1, s2)} x ptr mem) +(XORWload [o1] {s1} x (MOVDaddr [o2] {s2} ptr) mem) && ptr.Op != OpSB && is20Bit(int64(o1)+int64(o2)) && canMergeSym(s1, s2) => (XORWload [o1+o2] {mergeSym(s1, s2)} x ptr mem) // Cannot store constant to SB directly (no 'move relative long immediate' instructions). (MOVDstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) => - (MOVDstoreconst [sc.addOffset32(off)] {mergeSymTyped(sym1, sym2)} ptr mem) + (MOVDstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) (MOVWstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) => - (MOVWstoreconst [sc.addOffset32(off)] {mergeSymTyped(sym1, sym2)} ptr mem) + (MOVWstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) (MOVHstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) => - (MOVHstoreconst [sc.addOffset32(off)] {mergeSymTyped(sym1, sym2)} ptr mem) + (MOVHstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) (MOVBstoreconst [sc] {sym1} (MOVDaddr [off] {sym2} ptr) mem) && ptr.Op != OpSB && canMergeSym(sym1, sym2) && sc.canAdd32(off) => - (MOVBstoreconst [sc.addOffset32(off)] {mergeSymTyped(sym1, sym2)} ptr mem) + (MOVBstoreconst [sc.addOffset32(off)] {mergeSym(sym1, sym2)} ptr mem) // MOVDaddr into MOVDaddridx (MOVDaddridx [off1] {sym1} (MOVDaddr [off2] {sym2} x) y) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && x.Op != OpSB => - (MOVDaddridx [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (MOVDaddridx [off1+off2] {mergeSym(sym1,sym2)} x y) (MOVDaddridx [off1] {sym1} x (MOVDaddr [off2] {sym2} y)) && is32Bit(int64(off1)+int64(off2)) && canMergeSym(sym1, sym2) && y.Op != OpSB => - (MOVDaddridx [off1+off2] {mergeSymTyped(sym1,sym2)} x y) + (MOVDaddridx [off1+off2] {mergeSym(sym1,sym2)} x y) // Absorb InvertFlags into branches. (BRC {c} (InvertFlags cmp) yes no) => (BRC {c.ReverseComparison()} cmp yes no) @@ -963,6 +1061,9 @@ (CMPWconst (ANDWconst _ [m]) [n]) && int32(m) >= 0 && int32(m) < int32(n) => (FlagLT) (CMPWUconst (ANDWconst _ [m]) [n]) && uint32(m) < uint32(n) => (FlagLT) +(CMPconst (RISBGZ x {r}) [c]) && c > 0 && r.OutMask() < uint64(c) => (FlagLT) +(CMPUconst (RISBGZ x {r}) [c]) && r.OutMask() < uint64(uint32(c)) => (FlagLT) + // Constant compare-and-branch with immediate. (CGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Equal != 0 && int64(x) == int64(y) => (First yes no) (CGIJ {c} (MOVDconst [x]) [y] yes no) && c&s390x.Less != 0 && int64(x) < int64(y) => (First yes no) diff --git a/src/cmd/compile/internal/ssa/gen/S390XOps.go b/src/cmd/compile/internal/ssa/gen/S390XOps.go index 710beaddbbb..f0cf2f2f6e6 100644 --- a/src/cmd/compile/internal/ssa/gen/S390XOps.go +++ b/src/cmd/compile/internal/ssa/gen/S390XOps.go @@ -331,25 +331,26 @@ func init() { {name: "LTEBR", argLength: 1, reg: fp1flags, asm: "LTEBR", typ: "Flags"}, // arg0 compare to 0, f32 {name: "SLD", argLength: 2, reg: sh21, asm: "SLD"}, // arg0 << arg1, shift amount is mod 64 - {name: "SLW", argLength: 2, reg: sh21, asm: "SLW"}, // arg0 << arg1, shift amount is mod 32 + {name: "SLW", argLength: 2, reg: sh21, asm: "SLW"}, // arg0 << arg1, shift amount is mod 64 {name: "SLDconst", argLength: 1, reg: gp11, asm: "SLD", aux: "Int8"}, // arg0 << auxint, shift amount 0-63 {name: "SLWconst", argLength: 1, reg: gp11, asm: "SLW", aux: "Int8"}, // arg0 << auxint, shift amount 0-31 {name: "SRD", argLength: 2, reg: sh21, asm: "SRD"}, // unsigned arg0 >> arg1, shift amount is mod 64 - {name: "SRW", argLength: 2, reg: sh21, asm: "SRW"}, // unsigned uint32(arg0) >> arg1, shift amount is mod 32 + {name: "SRW", argLength: 2, reg: sh21, asm: "SRW"}, // unsigned uint32(arg0) >> arg1, shift amount is mod 64 {name: "SRDconst", argLength: 1, reg: gp11, asm: "SRD", aux: "Int8"}, // unsigned arg0 >> auxint, shift amount 0-63 {name: "SRWconst", argLength: 1, reg: gp11, asm: "SRW", aux: "Int8"}, // unsigned uint32(arg0) >> auxint, shift amount 0-31 // Arithmetic shifts clobber flags. {name: "SRAD", argLength: 2, reg: sh21, asm: "SRAD", clobberFlags: true}, // signed arg0 >> arg1, shift amount is mod 64 - {name: "SRAW", argLength: 2, reg: sh21, asm: "SRAW", clobberFlags: true}, // signed int32(arg0) >> arg1, shift amount is mod 32 + {name: "SRAW", argLength: 2, reg: sh21, asm: "SRAW", clobberFlags: true}, // signed int32(arg0) >> arg1, shift amount is mod 64 {name: "SRADconst", argLength: 1, reg: gp11, asm: "SRAD", aux: "Int8", clobberFlags: true}, // signed arg0 >> auxint, shift amount 0-63 {name: "SRAWconst", argLength: 1, reg: gp11, asm: "SRAW", aux: "Int8", clobberFlags: true}, // signed int32(arg0) >> auxint, shift amount 0-31 - {name: "RLLG", argLength: 2, reg: sh21, asm: "RLLG"}, // arg0 rotate left arg1, rotate amount 0-63 - {name: "RLL", argLength: 2, reg: sh21, asm: "RLL"}, // arg0 rotate left arg1, rotate amount 0-31 - {name: "RLLGconst", argLength: 1, reg: gp11, asm: "RLLG", aux: "Int8"}, // arg0 rotate left auxint, rotate amount 0-63 - {name: "RLLconst", argLength: 1, reg: gp11, asm: "RLL", aux: "Int8"}, // arg0 rotate left auxint, rotate amount 0-31 + // Rotate instructions. + // Note: no RLLGconst - use RISBGZ instead. + {name: "RLLG", argLength: 2, reg: sh21, asm: "RLLG"}, // arg0 rotate left arg1, rotate amount 0-63 + {name: "RLL", argLength: 2, reg: sh21, asm: "RLL"}, // arg0 rotate left arg1, rotate amount 0-31 + {name: "RLLconst", argLength: 1, reg: gp11, asm: "RLL", aux: "Int8"}, // arg0 rotate left auxint, rotate amount 0-31 // Rotate then (and|or|xor|insert) selected bits instructions. // @@ -371,6 +372,7 @@ func init() { // +-------------+-------+-----+--------+-----------------------+-----------------------+-----------------------+ // {name: "RXSBG", argLength: 2, reg: gp21, asm: "RXSBG", resultInArg0: true, aux: "S390XRotateParams", clobberFlags: true}, // rotate then xor selected bits + {name: "RISBGZ", argLength: 1, reg: gp11, asm: "RISBGZ", aux: "S390XRotateParams", clobberFlags: true}, // rotate then insert selected bits [into zero] // unary ops {name: "NEG", argLength: 1, reg: gp11, asm: "NEG", clobberFlags: true}, // -arg0 @@ -475,9 +477,9 @@ func init() { {name: "CLEAR", argLength: 2, reg: regInfo{inputs: []regMask{ptr, 0}}, asm: "CLEAR", aux: "SymValAndOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, - {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", clobberFlags: true, call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{ptrsp, buildReg("R12"), 0}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{ptr}, clobbers: callerSave}, aux: "Int64", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "CALLstatic", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "CALLclosure", argLength: 3, reg: regInfo{inputs: []regMask{ptrsp, buildReg("R12"), 0}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "CALLinter", argLength: 2, reg: regInfo{inputs: []regMask{ptr}, clobbers: callerSave}, aux: "CallOff", clobberFlags: true, call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem // (InvertFlags (CMP a b)) == (CMP b a) // InvertFlags is a pseudo-op which can't appear in assembly output. @@ -547,8 +549,10 @@ func init() { // Atomic bitwise operations. // Note: 'floor' operations round the pointer down to the nearest word boundary // which reflects how they are used in the runtime. - {name: "LAOfloor", argLength: 3, reg: gpstorelab, asm: "LAO", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *(floor(arg0, 4)) |= arg1. arg2 = mem. + {name: "LAN", argLength: 3, reg: gpstore, asm: "LAN", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *arg0 &= arg1. arg2 = mem. {name: "LANfloor", argLength: 3, reg: gpstorelab, asm: "LAN", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *(floor(arg0, 4)) &= arg1. arg2 = mem. + {name: "LAO", argLength: 3, reg: gpstore, asm: "LAO", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *arg0 |= arg1. arg2 = mem. + {name: "LAOfloor", argLength: 3, reg: gpstorelab, asm: "LAO", typ: "Mem", clobberFlags: true, hasSideEffects: true}, // *(floor(arg0, 4)) |= arg1. arg2 = mem. // Compare and swap. // arg0 = pointer, arg1 = old value, arg2 = new value, arg3 = memory. diff --git a/src/cmd/compile/internal/ssa/gen/Wasm.rules b/src/cmd/compile/internal/ssa/gen/Wasm.rules index ea12c5d617a..fc45cd3ed5f 100644 --- a/src/cmd/compile/internal/ssa/gen/Wasm.rules +++ b/src/cmd/compile/internal/ssa/gen/Wasm.rules @@ -399,6 +399,7 @@ // folding offset into address (I64AddConst [off] (LoweredAddr {sym} [off2] base)) && isU32Bit(off+int64(off2)) => (LoweredAddr {sym} [int32(off)+off2] base) +(I64AddConst [off] x:(SP)) && isU32Bit(off) => (LoweredAddr [int32(off)] x) // so it is rematerializeable // transforming readonly globals into constants (I64Load [off] (LoweredAddr {sym} [off2] (SB)) _) && symIsRO(sym) && isU32Bit(off+int64(off2)) => (I64Const [int64(read64(sym, off+int64(off2), config.ctxt.Arch.ByteOrder))]) diff --git a/src/cmd/compile/internal/ssa/gen/WasmOps.go b/src/cmd/compile/internal/ssa/gen/WasmOps.go index e43eae17e9d..36c53bc78c2 100644 --- a/src/cmd/compile/internal/ssa/gen/WasmOps.go +++ b/src/cmd/compile/internal/ssa/gen/WasmOps.go @@ -122,9 +122,9 @@ func init() { ) var WasmOps = []opData{ - {name: "LoweredStaticCall", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "SymOff", call: true, symEffect: "None"}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem - {name: "LoweredClosureCall", argLength: 3, reg: regInfo{inputs: []regMask{gp, gp, 0}, clobbers: callerSave}, aux: "Int64", call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem - {name: "LoweredInterCall", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "Int64", call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem + {name: "LoweredStaticCall", argLength: 1, reg: regInfo{clobbers: callerSave}, aux: "CallOff", call: true}, // call static function aux.(*obj.LSym). arg0=mem, auxint=argsize, returns mem + {name: "LoweredClosureCall", argLength: 3, reg: regInfo{inputs: []regMask{gp, gp, 0}, clobbers: callerSave}, aux: "CallOff", call: true}, // call function via closure. arg0=codeptr, arg1=closure, arg2=mem, auxint=argsize, returns mem + {name: "LoweredInterCall", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", call: true}, // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem {name: "LoweredAddr", argLength: 1, reg: gp11, aux: "SymOff", rematerializeable: true, symEffect: "Addr"}, // returns base+aux+auxint, arg0=base {name: "LoweredMove", argLength: 3, reg: regInfo{inputs: []regMask{gp, gp}}, aux: "Int64"}, // large move. arg0=dst, arg1=src, arg2=mem, auxint=len/8, returns mem @@ -137,7 +137,7 @@ func init() { {name: "LoweredWB", argLength: 3, reg: regInfo{inputs: []regMask{gp, gp}}, aux: "Sym", symEffect: "None"}, // invokes runtime.gcWriteBarrier. arg0=destptr, arg1=srcptr, arg2=mem, aux=runtime.gcWriteBarrier // LoweredConvert converts between pointers and integers. - // We have a special op for this so as to not confuse GC + // We have a special op for this so as to not confuse GCCallOff // (particularly stack maps). It takes a memory arg so it // gets correctly ordered with respect to GC safepoints. // arg0=ptr/int arg1=mem, output=int/ptr diff --git a/src/cmd/compile/internal/ssa/gen/dec64.rules b/src/cmd/compile/internal/ssa/gen/dec64.rules index 4f9e863f907..9297ed8d2e0 100644 --- a/src/cmd/compile/internal/ssa/gen/dec64.rules +++ b/src/cmd/compile/internal/ssa/gen/dec64.rules @@ -9,7 +9,6 @@ (Int64Hi (Int64Make hi _)) => hi (Int64Lo (Int64Make _ lo)) => lo - (Load ptr mem) && is64BitInt(t) && !config.BigEndian && t.IsSigned() => (Int64Make (Load (OffPtr [4] ptr) mem) @@ -42,20 +41,21 @@ lo (Store {hi.Type} dst hi mem)) -(Arg {n} [off]) && is64BitInt(v.Type) && !config.BigEndian && v.Type.IsSigned() => +// These are not enabled during decomposeBuiltin if late call expansion, but they are always enabled for softFloat +(Arg {n} [off]) && is64BitInt(v.Type) && !config.BigEndian && v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin") => (Int64Make (Arg {n} [off+4]) (Arg {n} [off])) -(Arg {n} [off]) && is64BitInt(v.Type) && !config.BigEndian && !v.Type.IsSigned() => +(Arg {n} [off]) && is64BitInt(v.Type) && !config.BigEndian && !v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin") => (Int64Make (Arg {n} [off+4]) (Arg {n} [off])) -(Arg {n} [off]) && is64BitInt(v.Type) && config.BigEndian && v.Type.IsSigned() => +(Arg {n} [off]) && is64BitInt(v.Type) && config.BigEndian && v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin") => (Int64Make (Arg {n} [off]) (Arg {n} [off+4])) -(Arg {n} [off]) && is64BitInt(v.Type) && config.BigEndian && !v.Type.IsSigned() => +(Arg {n} [off]) && is64BitInt(v.Type) && config.BigEndian && !v.Type.IsSigned() && !(go116lateCallExpansion && b.Func.pass.name == "decompose builtin") => (Int64Make (Arg {n} [off]) (Arg {n} [off+4])) @@ -143,6 +143,10 @@ (Trunc64to32 (Int64Make _ lo)) => lo (Trunc64to16 (Int64Make _ lo)) => (Trunc32to16 lo) (Trunc64to8 (Int64Make _ lo)) => (Trunc32to8 lo) +// Most general +(Trunc64to32 x) => (Int64Lo x) +(Trunc64to16 x) => (Trunc32to16 (Int64Lo x)) +(Trunc64to8 x) => (Trunc32to8 (Int64Lo x)) (Lsh32x64 _ (Int64Make (Const32 [c]) _)) && c != 0 => (Const32 [0]) (Rsh32x64 x (Int64Make (Const32 [c]) _)) && c != 0 => (Signmask x) @@ -175,156 +179,174 @@ // turn x64 non-constant shifts to x32 shifts // if high 32-bit of the shift is nonzero, make a huge shift (Lsh64x64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Lsh64x32 x (Or32 (Zeromask hi) lo)) + (Lsh64x32 x (Or32 (Zeromask hi) lo)) (Rsh64x64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Rsh64x32 x (Or32 (Zeromask hi) lo)) + (Rsh64x32 x (Or32 (Zeromask hi) lo)) (Rsh64Ux64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Rsh64Ux32 x (Or32 (Zeromask hi) lo)) + (Rsh64Ux32 x (Or32 (Zeromask hi) lo)) (Lsh32x64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Lsh32x32 x (Or32 (Zeromask hi) lo)) + (Lsh32x32 x (Or32 (Zeromask hi) lo)) (Rsh32x64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Rsh32x32 x (Or32 (Zeromask hi) lo)) + (Rsh32x32 x (Or32 (Zeromask hi) lo)) (Rsh32Ux64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Rsh32Ux32 x (Or32 (Zeromask hi) lo)) + (Rsh32Ux32 x (Or32 (Zeromask hi) lo)) (Lsh16x64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Lsh16x32 x (Or32 (Zeromask hi) lo)) + (Lsh16x32 x (Or32 (Zeromask hi) lo)) (Rsh16x64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Rsh16x32 x (Or32 (Zeromask hi) lo)) + (Rsh16x32 x (Or32 (Zeromask hi) lo)) (Rsh16Ux64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Rsh16Ux32 x (Or32 (Zeromask hi) lo)) + (Rsh16Ux32 x (Or32 (Zeromask hi) lo)) (Lsh8x64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Lsh8x32 x (Or32 (Zeromask hi) lo)) + (Lsh8x32 x (Or32 (Zeromask hi) lo)) (Rsh8x64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Rsh8x32 x (Or32 (Zeromask hi) lo)) + (Rsh8x32 x (Or32 (Zeromask hi) lo)) (Rsh8Ux64 x (Int64Make hi lo)) && hi.Op != OpConst32 => - (Rsh8Ux32 x (Or32 (Zeromask hi) lo)) + (Rsh8Ux32 x (Or32 (Zeromask hi) lo)) + +// Most general +(Lsh64x64 x y) => (Lsh64x32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Rsh64x64 x y) => (Rsh64x32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Rsh64Ux64 x y) => (Rsh64Ux32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Lsh32x64 x y) => (Lsh32x32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Rsh32x64 x y) => (Rsh32x32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Rsh32Ux64 x y) => (Rsh32Ux32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Lsh16x64 x y) => (Lsh16x32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Rsh16x64 x y) => (Rsh16x32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Rsh16Ux64 x y) => (Rsh16Ux32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Lsh8x64 x y) => (Lsh8x32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Rsh8x64 x y) => (Rsh8x32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) +(Rsh8Ux64 x y) => (Rsh8Ux32 x (Or32 (Zeromask (Int64Hi y)) (Int64Lo y))) + +// Clean up constants a little +(Or32 (Zeromask (Const32 [c])) y) && c == 0 => y +(Or32 (Zeromask (Const32 [c])) y) && c != 0 => (Const32 [-1]) // 64x left shift // result.hi = hi<>(32-s) | lo<<(s-32) // >> is unsigned, large shifts result 0 // result.lo = lo< +(Lsh64x32 x s) => (Int64Make (Or32 (Or32 - (Lsh32x32 hi s) + (Lsh32x32 (Int64Hi x) s) (Rsh32Ux32 - lo + (Int64Lo x) (Sub32 (Const32 [32]) s))) (Lsh32x32 - lo + (Int64Lo x) (Sub32 s (Const32 [32])))) - (Lsh32x32 lo s)) -(Lsh64x16 (Int64Make hi lo) s) => + (Lsh32x32 (Int64Lo x) s)) +(Lsh64x16 x s) => (Int64Make (Or32 (Or32 - (Lsh32x16 hi s) + (Lsh32x16 (Int64Hi x) s) (Rsh32Ux16 - lo + (Int64Lo x) (Sub16 (Const16 [32]) s))) (Lsh32x16 - lo + (Int64Lo x) (Sub16 s (Const16 [32])))) - (Lsh32x16 lo s)) -(Lsh64x8 (Int64Make hi lo) s) => + (Lsh32x16 (Int64Lo x) s)) +(Lsh64x8 x s) => (Int64Make (Or32 (Or32 - (Lsh32x8 hi s) + (Lsh32x8 (Int64Hi x) s) (Rsh32Ux8 - lo + (Int64Lo x) (Sub8 (Const8 [32]) s))) (Lsh32x8 - lo + (Int64Lo x) (Sub8 s (Const8 [32])))) - (Lsh32x8 lo s)) + (Lsh32x8 (Int64Lo x) s)) // 64x unsigned right shift // result.hi = hi>>s // result.lo = lo>>s | hi<<(32-s) | hi>>(s-32) // >> is unsigned, large shifts result 0 -(Rsh64Ux32 (Int64Make hi lo) s) => +(Rsh64Ux32 x s) => (Int64Make - (Rsh32Ux32 hi s) + (Rsh32Ux32 (Int64Hi x) s) (Or32 (Or32 - (Rsh32Ux32 lo s) + (Rsh32Ux32 (Int64Lo x) s) (Lsh32x32 - hi + (Int64Hi x) (Sub32 (Const32 [32]) s))) (Rsh32Ux32 - hi + (Int64Hi x) (Sub32 s (Const32 [32]))))) -(Rsh64Ux16 (Int64Make hi lo) s) => +(Rsh64Ux16 x s) => (Int64Make - (Rsh32Ux16 hi s) + (Rsh32Ux16 (Int64Hi x) s) (Or32 (Or32 - (Rsh32Ux16 lo s) + (Rsh32Ux16 (Int64Lo x) s) (Lsh32x16 - hi + (Int64Hi x) (Sub16 (Const16 [32]) s))) (Rsh32Ux16 - hi + (Int64Hi x) (Sub16 s (Const16 [32]))))) -(Rsh64Ux8 (Int64Make hi lo) s) => +(Rsh64Ux8 x s) => (Int64Make - (Rsh32Ux8 hi s) + (Rsh32Ux8 (Int64Hi x) s) (Or32 (Or32 - (Rsh32Ux8 lo s) + (Rsh32Ux8 (Int64Lo x) s) (Lsh32x8 - hi + (Int64Hi x) (Sub8 (Const8 [32]) s))) (Rsh32Ux8 - hi + (Int64Hi x) (Sub8 s (Const8 [32]))))) // 64x signed right shift // result.hi = hi>>s // result.lo = lo>>s | hi<<(32-s) | (hi>>(s-32))&zeromask(s>>5) // hi>>(s-32) is signed, large shifts result 0/-1 -(Rsh64x32 (Int64Make hi lo) s) => +(Rsh64x32 x s) => (Int64Make - (Rsh32x32 hi s) + (Rsh32x32 (Int64Hi x) s) (Or32 (Or32 - (Rsh32Ux32 lo s) + (Rsh32Ux32 (Int64Lo x) s) (Lsh32x32 - hi + (Int64Hi x) (Sub32 (Const32 [32]) s))) (And32 (Rsh32x32 - hi + (Int64Hi x) (Sub32 s (Const32 [32]))) (Zeromask (Rsh32Ux32 s (Const32 [5])))))) -(Rsh64x16 (Int64Make hi lo) s) => +(Rsh64x16 x s) => (Int64Make - (Rsh32x16 hi s) + (Rsh32x16 (Int64Hi x) s) (Or32 (Or32 - (Rsh32Ux16 lo s) + (Rsh32Ux16 (Int64Lo x) s) (Lsh32x16 - hi + (Int64Hi x) (Sub16 (Const16 [32]) s))) (And32 (Rsh32x16 - hi + (Int64Hi x) (Sub16 s (Const16 [32]))) (Zeromask (ZeroExt16to32 (Rsh16Ux32 s (Const32 [5]))))))) -(Rsh64x8 (Int64Make hi lo) s) => +(Rsh64x8 x s) => (Int64Make - (Rsh32x8 hi s) + (Rsh32x8 (Int64Hi x) s) (Or32 (Or32 - (Rsh32Ux8 lo s) + (Rsh32Ux8 (Int64Lo x) s) (Lsh32x8 - hi + (Int64Hi x) (Sub8 (Const8 [32]) s))) (And32 (Rsh32x8 - hi + (Int64Hi x) (Sub8 s (Const8 [32]))) (Zeromask (ZeroExt8to32 diff --git a/src/cmd/compile/internal/ssa/gen/generic.rules b/src/cmd/compile/internal/ssa/gen/generic.rules index f7e6bbebac7..81568b7b7af 100644 --- a/src/cmd/compile/internal/ssa/gen/generic.rules +++ b/src/cmd/compile/internal/ssa/gen/generic.rules @@ -1040,6 +1040,46 @@ (ZeroExt32to64 x))) (Const64 [32+umagic32(c).s-1]))) +// For unsigned 64-bit divides on 32-bit machines, +// if the constant fits in 16 bits (so that the last term +// fits in 32 bits), convert to three 32-bit divides by a constant. +// +// If 1<<32 = Q * c + R +// and x = hi << 32 + lo +// +// Then x = (hi/c*c + hi%c) << 32 + lo +// = hi/c*c<<32 + hi%c<<32 + lo +// = hi/c*c<<32 + (hi%c)*(Q*c+R) + lo/c*c + lo%c +// = hi/c*c<<32 + (hi%c)*Q*c + lo/c*c + (hi%c*R+lo%c) +// and x / c = (hi/c)<<32 + (hi%c)*Q + lo/c + (hi%c*R+lo%c)/c +(Div64u x (Const64 [c])) && c > 0 && c <= 0xFFFF && umagicOK32(int32(c)) && config.RegSize == 4 && config.useHmul => + (Add64 + (Add64 + (Add64 + (Lsh64x64 + (ZeroExt32to64 + (Div32u + (Trunc64to32 (Rsh64Ux64 x (Const64 [32]))) + (Const32 [int32(c)]))) + (Const64 [32])) + (ZeroExt32to64 (Div32u (Trunc64to32 x) (Const32 [int32(c)])))) + (Mul64 + (ZeroExt32to64 + (Mod32u + (Trunc64to32 (Rsh64Ux64 x (Const64 [32]))) + (Const32 [int32(c)]))) + (Const64 [int64((1<<32)/c)]))) + (ZeroExt32to64 + (Div32u + (Add32 + (Mod32u (Trunc64to32 x) (Const32 [int32(c)])) + (Mul32 + (Mod32u + (Trunc64to32 (Rsh64Ux64 x (Const64 [32]))) + (Const32 [int32(c)])) + (Const32 [int32((1<<32)%c)]))) + (Const32 [int32(c)])))) + // For 64-bit divides on 64-bit machines // (64-bit divides on 32-bit machines are lowered to a runtime call by the walk pass.) (Div64u x (Const64 [c])) && umagicOK64(c) && config.RegSize == 8 && umagic64(c).m&1 == 0 && config.useHmul => @@ -1933,34 +1973,71 @@ // recognize runtime.newobject and don't Zero/Nilcheck it (Zero (Load (OffPtr [c] (SP)) mem) mem) && mem.Op == OpStaticCall - && isSameSym(mem.Aux, "runtime.newobject") + && isSameCall(mem.Aux, "runtime.newobject") && c == config.ctxt.FixedFrameSize() + config.RegSize // offset of return value => mem (Store (Load (OffPtr [c] (SP)) mem) x mem) && isConstZero(x) && mem.Op == OpStaticCall - && isSameSym(mem.Aux, "runtime.newobject") + && isSameCall(mem.Aux, "runtime.newobject") && c == config.ctxt.FixedFrameSize() + config.RegSize // offset of return value => mem (Store (OffPtr (Load (OffPtr [c] (SP)) mem)) x mem) && isConstZero(x) && mem.Op == OpStaticCall - && isSameSym(mem.Aux, "runtime.newobject") + && isSameCall(mem.Aux, "runtime.newobject") && c == config.ctxt.FixedFrameSize() + config.RegSize // offset of return value => mem // nil checks just need to rewrite to something useless. // they will be deadcode eliminated soon afterwards. (NilCheck (Load (OffPtr [c] (SP)) (StaticCall {sym} _)) _) - && symNamed(sym, "runtime.newobject") + && isSameCall(sym, "runtime.newobject") && c == config.ctxt.FixedFrameSize() + config.RegSize // offset of return value && warnRule(fe.Debug_checknil(), v, "removed nil check") => (Invalid) (NilCheck (OffPtr (Load (OffPtr [c] (SP)) (StaticCall {sym} _))) _) - && symNamed(sym, "runtime.newobject") + && isSameCall(sym, "runtime.newobject") && c == config.ctxt.FixedFrameSize() + config.RegSize // offset of return value && warnRule(fe.Debug_checknil(), v, "removed nil check") => (Invalid) +// for rewriting results of some late-expanded rewrites (below) +(SelectN [0] (MakeResult a ___)) => a +(SelectN [1] (MakeResult a b ___)) => b +(SelectN [2] (MakeResult a b c ___)) => c + +// for late-expanded calls, recognize newobject and remove zeroing and nilchecks +(Zero (SelectN [0] call:(StaticLECall _ _)) mem:(SelectN [1] call)) + && isSameCall(call.Aux, "runtime.newobject") + => mem + +(Store (SelectN [0] call:(StaticLECall _ _)) x mem:(SelectN [1] call)) + && isConstZero(x) + && isSameCall(call.Aux, "runtime.newobject") + => mem + +(Store (OffPtr (SelectN [0] call:(StaticLECall _ _))) x mem:(SelectN [1] call)) + && isConstZero(x) + && isSameCall(call.Aux, "runtime.newobject") + => mem + +(NilCheck (SelectN [0] call:(StaticLECall _ _)) (SelectN [1] call)) + && isSameCall(call.Aux, "runtime.newobject") + && warnRule(fe.Debug_checknil(), v, "removed nil check") + => (Invalid) + +(NilCheck (OffPtr (SelectN [0] call:(StaticLECall _ _))) (SelectN [1] call)) + && isSameCall(call.Aux, "runtime.newobject") + && warnRule(fe.Debug_checknil(), v, "removed nil check") + => (Invalid) + +// for late-expanded calls, recognize memequal applied to a single constant byte +// TODO figure out breakeven number of bytes for this optimization. +(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [1]) mem) + && isSameCall(callAux, "runtime.memequal") + && symIsRO(scon) + => (MakeResult (Eq8 (Load sptr mem) (Const8 [int8(read8(scon,0))])) mem) + // Evaluate constant address comparisons. (EqPtr x x) => (ConstBool [true]) (NeqPtr x x) => (ConstBool [false]) @@ -2010,19 +2087,37 @@ // See the comment in op Move in genericOps.go for discussion of the type. (StaticCall {sym} s1:(Store _ (Const(64|32) [sz]) s2:(Store _ src s3:(Store {t} _ dst mem)))) && sz >= 0 - && symNamed(sym, "runtime.memmove") + && isSameCall(sym, "runtime.memmove") && t.IsPtr() // avoids TUINTPTR, see issue 30061 && s1.Uses == 1 && s2.Uses == 1 && s3.Uses == 1 && isInlinableMemmove(dst, src, int64(sz), config) && clobber(s1, s2, s3) => (Move {t.Elem()} [int64(sz)] dst src mem) +// Inline small or disjoint runtime.memmove calls with constant length. +// See the comment in op Move in genericOps.go for discussion of the type. +(SelectN [0] call:(StaticLECall {sym} dst src (Const(64|32) [sz]) mem)) + && sz >= 0 + && call.Uses == 1 // this will exclude all calls with results + && isSameCall(sym, "runtime.memmove") + && dst.Type.IsPtr() // avoids TUINTPTR, see issue 30061 + && isInlinableMemmove(dst, src, int64(sz), config) + && clobber(call) + => (Move {dst.Type.Elem()} [int64(sz)] dst src mem) + // De-virtualize interface calls into static calls. // Note that (ITab (IMake)) doesn't get // rewritten until after the first opt pass, // so this rule should trigger reliably. -(InterCall [argsize] (Load (OffPtr [off] (ITab (IMake (Addr {itab} (SB)) _))) _) mem) && devirt(v, itab, off) != nil => - (StaticCall [int32(argsize)] {devirt(v, itab, off)} mem) +(InterCall [argsize] {auxCall} (Load (OffPtr [off] (ITab (IMake (Addr {itab} (SB)) _))) _) mem) && devirt(v, auxCall, itab, off) != nil => + (StaticCall [int32(argsize)] {devirt(v, auxCall, itab, off)} mem) + +// De-virtualize late-expanded interface calls into late-expanded static calls. +// Note that (ITab (IMake)) doesn't get rewritten until after the first opt pass, +// so this rule should trigger reliably. +// devirtLECall removes the first argument, adds the devirtualized symbol to the AuxCall, and changes the opcode +(InterLECall [argsize] {auxCall} (Load (OffPtr [off] (ITab (IMake (Addr {itab} (SB)) _))) _) ___) && devirtLESym(v, auxCall, itab, off) != + nil => devirtLECall(v, devirtLESym(v, auxCall, itab, off)) // Move and Zero optimizations. // Move source and destination may overlap. @@ -2404,6 +2499,7 @@ (Store {t5} (OffPtr [o5] dst) d4 (Zero {t1} [n] dst mem))))) +// TODO this does not fire before call expansion; is that acceptable? (StaticCall {sym} x) && needRaceCleanup(sym, v) => x // Collapse moving A -> B -> C into just A -> C. diff --git a/src/cmd/compile/internal/ssa/gen/genericOps.go b/src/cmd/compile/internal/ssa/gen/genericOps.go index 5df0a164bff..8cfda35c225 100644 --- a/src/cmd/compile/internal/ssa/gen/genericOps.go +++ b/src/cmd/compile/internal/ssa/gen/genericOps.go @@ -346,6 +346,7 @@ var genericOps = []opData{ // Memory operations {name: "Load", argLength: 2}, // Load from arg0. arg1=memory + {name: "Dereference", argLength: 2}, // Load from arg0. arg1=memory. Helper op for arg/result passing, result is an otherwise not-SSA-able "value". {name: "Store", argLength: 3, typ: "Mem", aux: "Typ"}, // Store arg1 to arg0. arg2=memory, aux=type. Returns memory. // The source and destination of Move may overlap in some cases. See e.g. // memmove inlining in generic.rules. When inlineablememmovesize (in ../rewrite.go) @@ -387,9 +388,13 @@ var genericOps = []opData{ // as a phantom first argument. // TODO(josharian): ClosureCall and InterCall should have Int32 aux // to match StaticCall's 32 bit arg size limit. - {name: "ClosureCall", argLength: 3, aux: "Int64", call: true}, // arg0=code pointer, arg1=context ptr, arg2=memory. auxint=arg size. Returns memory. - {name: "StaticCall", argLength: 1, aux: "SymOff", call: true, symEffect: "None"}, // call function aux.(*obj.LSym), arg0=memory. auxint=arg size. Returns memory. - {name: "InterCall", argLength: 2, aux: "Int64", call: true}, // interface call. arg0=code pointer, arg1=memory, auxint=arg size. Returns memory. + // TODO(drchase,josharian): could the arg size limit be bundled into the rules for CallOff? + {name: "ClosureCall", argLength: 3, aux: "CallOff", call: true}, // arg0=code pointer, arg1=context ptr, arg2=memory. auxint=arg size. Returns memory. + {name: "StaticCall", argLength: 1, aux: "CallOff", call: true}, // call function aux.(*obj.LSym), arg0=memory. auxint=arg size. Returns memory. + {name: "InterCall", argLength: 2, aux: "CallOff", call: true}, // interface call. arg0=code pointer, arg1=memory, auxint=arg size. Returns memory. + {name: "ClosureLECall", argLength: -1, aux: "CallOff", call: true}, // late-expanded closure call. arg0=code pointer, arg1=context ptr, arg2..argN-1 are inputs, argN is mem. auxint = arg size. Result is tuple of result(s), plus mem. + {name: "StaticLECall", argLength: -1, aux: "CallOff", call: true}, // late-expanded static call function aux.(*ssa.AuxCall.Fn). arg0..argN-1 are inputs, argN is mem. auxint = arg size. Result is tuple of result(s), plus mem. + {name: "InterLECall", argLength: -1, aux: "CallOff", call: true}, // late-expanded interface call. arg0=code pointer, arg1..argN-1 are inputs, argN is mem. auxint = arg size. Result is tuple of result(s), plus mem. // Conversions: signed extensions, zero (unsigned) extensions, truncations {name: "SignExt8to16", argLength: 1, typ: "Int16"}, @@ -531,23 +536,28 @@ var genericOps = []opData{ {name: "Cvt64Fto64U", argLength: 1}, // float64 -> uint64, only used on archs that has the instruction // pseudo-ops for breaking Tuple - {name: "Select0", argLength: 1, zeroWidth: true}, // the first component of a tuple - {name: "Select1", argLength: 1, zeroWidth: true}, // the second component of a tuple + {name: "Select0", argLength: 1, zeroWidth: true}, // the first component of a tuple + {name: "Select1", argLength: 1, zeroWidth: true}, // the second component of a tuple + {name: "SelectN", argLength: 1, aux: "Int64"}, // arg0=result, auxint=field index. Returns the auxint'th member. + {name: "SelectNAddr", argLength: 1, aux: "Int64"}, // arg0=result, auxint=field index. Returns the address of auxint'th member. Used for un-SSA-able result types. + {name: "MakeResult", argLength: -1}, // arg0 .. are components of a "Result" (like the result from a Call). The last arg should be memory (like the result from a call). - // Atomic operations used for semantically inlining runtime/internal/atomic. - // Atomic loads return a new memory so that the loads are properly ordered - // with respect to other loads and stores. - // TODO: use for sync/atomic at some point. + // Atomic operations used for semantically inlining sync/atomic and + // runtime/internal/atomic. Atomic loads return a new memory so that + // the loads are properly ordered with respect to other loads and + // stores. {name: "AtomicLoad8", argLength: 2, typ: "(UInt8,Mem)"}, // Load from arg0. arg1=memory. Returns loaded value and new memory. {name: "AtomicLoad32", argLength: 2, typ: "(UInt32,Mem)"}, // Load from arg0. arg1=memory. Returns loaded value and new memory. {name: "AtomicLoad64", argLength: 2, typ: "(UInt64,Mem)"}, // Load from arg0. arg1=memory. Returns loaded value and new memory. {name: "AtomicLoadPtr", argLength: 2, typ: "(BytePtr,Mem)"}, // Load from arg0. arg1=memory. Returns loaded value and new memory. {name: "AtomicLoadAcq32", argLength: 2, typ: "(UInt32,Mem)"}, // Load from arg0. arg1=memory. Lock acquisition, returns loaded value and new memory. + {name: "AtomicLoadAcq64", argLength: 2, typ: "(UInt64,Mem)"}, // Load from arg0. arg1=memory. Lock acquisition, returns loaded value and new memory. {name: "AtomicStore8", argLength: 3, typ: "Mem", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Returns memory. {name: "AtomicStore32", argLength: 3, typ: "Mem", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Returns memory. {name: "AtomicStore64", argLength: 3, typ: "Mem", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Returns memory. {name: "AtomicStorePtrNoWB", argLength: 3, typ: "Mem", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Returns memory. {name: "AtomicStoreRel32", argLength: 3, typ: "Mem", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Lock release, returns memory. + {name: "AtomicStoreRel64", argLength: 3, typ: "Mem", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Lock release, returns memory. {name: "AtomicExchange32", argLength: 3, typ: "(UInt32,Mem)", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Returns old contents of *arg0 and new memory. {name: "AtomicExchange64", argLength: 3, typ: "(UInt64,Mem)", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Returns old contents of *arg0 and new memory. {name: "AtomicAdd32", argLength: 3, typ: "(UInt32,Mem)", hasSideEffects: true}, // Do *arg0 += arg1. arg2=memory. Returns sum and new memory. @@ -556,14 +566,24 @@ var genericOps = []opData{ {name: "AtomicCompareAndSwap64", argLength: 4, typ: "(Bool,Mem)", hasSideEffects: true}, // if *arg0==arg1, then set *arg0=arg2. Returns true if store happens and new memory. {name: "AtomicCompareAndSwapRel32", argLength: 4, typ: "(Bool,Mem)", hasSideEffects: true}, // if *arg0==arg1, then set *arg0=arg2. Lock release, reports whether store happens and new memory. {name: "AtomicAnd8", argLength: 3, typ: "Mem", hasSideEffects: true}, // *arg0 &= arg1. arg2=memory. Returns memory. + {name: "AtomicAnd32", argLength: 3, typ: "Mem", hasSideEffects: true}, // *arg0 &= arg1. arg2=memory. Returns memory. {name: "AtomicOr8", argLength: 3, typ: "Mem", hasSideEffects: true}, // *arg0 |= arg1. arg2=memory. Returns memory. + {name: "AtomicOr32", argLength: 3, typ: "Mem", hasSideEffects: true}, // *arg0 |= arg1. arg2=memory. Returns memory. // Atomic operation variants // These variants have the same semantics as above atomic operations. // But they are used for generating more efficient code on certain modern machines, with run-time CPU feature detection. // Currently, they are used on ARM64 only. - {name: "AtomicAdd32Variant", argLength: 3, typ: "(UInt32,Mem)", hasSideEffects: true}, // Do *arg0 += arg1. arg2=memory. Returns sum and new memory. - {name: "AtomicAdd64Variant", argLength: 3, typ: "(UInt64,Mem)", hasSideEffects: true}, // Do *arg0 += arg1. arg2=memory. Returns sum and new memory. + {name: "AtomicAdd32Variant", argLength: 3, typ: "(UInt32,Mem)", hasSideEffects: true}, // Do *arg0 += arg1. arg2=memory. Returns sum and new memory. + {name: "AtomicAdd64Variant", argLength: 3, typ: "(UInt64,Mem)", hasSideEffects: true}, // Do *arg0 += arg1. arg2=memory. Returns sum and new memory. + {name: "AtomicExchange32Variant", argLength: 3, typ: "(UInt32,Mem)", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Returns old contents of *arg0 and new memory. + {name: "AtomicExchange64Variant", argLength: 3, typ: "(UInt64,Mem)", hasSideEffects: true}, // Store arg1 to *arg0. arg2=memory. Returns old contents of *arg0 and new memory. + {name: "AtomicCompareAndSwap32Variant", argLength: 4, typ: "(Bool,Mem)", hasSideEffects: true}, // if *arg0==arg1, then set *arg0=arg2. Returns true if store happens and new memory. + {name: "AtomicCompareAndSwap64Variant", argLength: 4, typ: "(Bool,Mem)", hasSideEffects: true}, // if *arg0==arg1, then set *arg0=arg2. Returns true if store happens and new memory. + {name: "AtomicAnd8Variant", argLength: 3, typ: "Mem", hasSideEffects: true}, // *arg0 &= arg1. arg2=memory. Returns memory. + {name: "AtomicAnd32Variant", argLength: 3, typ: "Mem", hasSideEffects: true}, // *arg0 &= arg1. arg2=memory. Returns memory. + {name: "AtomicOr8Variant", argLength: 3, typ: "Mem", hasSideEffects: true}, // *arg0 |= arg1. arg2=memory. Returns memory. + {name: "AtomicOr32Variant", argLength: 3, typ: "Mem", hasSideEffects: true}, // *arg0 |= arg1. arg2=memory. Returns memory. // Clobber experiment op {name: "Clobber", argLength: 0, typ: "Void", aux: "SymOff", symEffect: "None"}, // write an invalid pointer value to the given pointer slot of a stack variable diff --git a/src/cmd/compile/internal/ssa/gen/rulegen.go b/src/cmd/compile/internal/ssa/gen/rulegen.go index 9e2e112cd79..120ccbbdb35 100644 --- a/src/cmd/compile/internal/ssa/gen/rulegen.go +++ b/src/cmd/compile/internal/ssa/gen/rulegen.go @@ -35,8 +35,7 @@ import ( ) // rule syntax: -// sexpr [&& extra conditions] -> [@block] sexpr (untyped) -// sexpr [&& extra conditions] => [@block] sexpr (typed) +// sexpr [&& extra conditions] => [@block] sexpr // // sexpr are s-expressions (lisp-like parenthesized groupings) // sexpr ::= [variable:](opcode sexpr*) @@ -50,8 +49,12 @@ import ( // variable ::= some token // opcode ::= one of the opcodes from the *Ops.go files +// special rules: trailing ellipsis "..." (in the outermost sexpr?) must match on both sides of a rule. +// trailing three underscore "___" in the outermost match sexpr indicate the presence of +// extra ignored args that need not appear in the replacement + // extra conditions is just a chunk of Go that evaluates to a boolean. It may use -// variables declared in the matching sexpr. The variable "v" is predefined to be +// variables declared in the matching tsexpr. The variable "v" is predefined to be // the value matched by the entire rule. // If multiple rules match, the first one in file order is selected. @@ -75,14 +78,8 @@ func normalizeSpaces(s string) string { } // parse returns the matching part of the rule, additional conditions, and the result. -// parse also reports whether the generated code should use strongly typed aux and auxint fields. -func (r Rule) parse() (match, cond, result string, typed bool) { - arrow := "->" - if strings.Contains(r.Rule, "=>") { - arrow = "=>" - typed = true - } - s := strings.Split(r.Rule, arrow) +func (r Rule) parse() (match, cond, result string) { + s := strings.Split(r.Rule, "=>") match = normalizeSpaces(s[0]) result = normalizeSpaces(s[1]) cond = "" @@ -90,7 +87,7 @@ func (r Rule) parse() (match, cond, result string, typed bool) { cond = normalizeSpaces(match[i+2:]) match = normalizeSpaces(match[:i]) } - return match, cond, result, typed + return match, cond, result } func genRules(arch arch) { genRulesSuffix(arch, "") } @@ -116,7 +113,7 @@ func genRulesSuffix(arch arch, suff string) { scanner := bufio.NewScanner(text) rule := "" var lineno int - var ruleLineno int // line number of "->" or "=>" + var ruleLineno int // line number of "=>" for scanner.Scan() { lineno++ line := scanner.Text() @@ -130,13 +127,13 @@ func genRulesSuffix(arch arch, suff string) { if rule == "" { continue } - if !strings.Contains(rule, "->") && !strings.Contains(rule, "=>") { + if !strings.Contains(rule, "=>") { continue } if ruleLineno == 0 { ruleLineno = lineno } - if strings.HasSuffix(rule, "->") || strings.HasSuffix(rule, "=>") { + if strings.HasSuffix(rule, "=>") { continue // continue on the next line } if n := balance(rule); n > 0 { @@ -153,7 +150,7 @@ func genRulesSuffix(arch arch, suff string) { continue } // Do fancier value op matching. - match, _, _, _ := r.parse() + match, _, _ := r.parse() op, oparch, _, _, _, _ := parseValue(match, arch, loc) opname := fmt.Sprintf("Op%s%s", oparch, op.name) oprules[opname] = append(oprules[opname], r) @@ -227,7 +224,7 @@ func genRulesSuffix(arch arch, suff string) { log.Fatalf("unconditional rule %s is followed by other rules", rr.Match) } rr = &RuleRewrite{Loc: rule.Loc} - rr.Match, rr.Cond, rr.Result, rr.Typed = rule.parse() + rr.Match, rr.Cond, rr.Result = rule.parse() pos, _ := genMatch(rr, arch, rr.Match, fn.ArgLen >= 0) if pos == "" { pos = "v.Pos" @@ -786,7 +783,6 @@ type ( Alloc int // for unique var names Loc string // file name & line number of the original rule CommuteDepth int // used to track depth of commute loops - Typed bool // aux and auxint fields should be strongly typed } Declare struct { Name string @@ -840,7 +836,7 @@ func breakf(format string, a ...interface{}) *CondBreak { func genBlockRewrite(rule Rule, arch arch, data blockData) *RuleRewrite { rr := &RuleRewrite{Loc: rule.Loc} - rr.Match, rr.Cond, rr.Result, rr.Typed = rule.parse() + rr.Match, rr.Cond, rr.Result = rule.parse() _, _, auxint, aux, s := extract(rr.Match) // remove parens, then split // check match of control values @@ -884,15 +880,6 @@ func genBlockRewrite(rule Rule, arch arch, data blockData) *RuleRewrite { if e.name == "" { continue } - if !rr.Typed { - if !token.IsIdentifier(e.name) || rr.declared(e.name) { - // code or variable - rr.add(breakf("b.%s != %s", e.field, e.name)) - } else { - rr.add(declf(e.name, "b.%s", e.field)) - } - continue - } if e.dclType == "" { log.Fatalf("op %s has no declared type for %s", data.name, e.field) @@ -961,20 +948,12 @@ func genBlockRewrite(rule Rule, arch arch, data blockData) *RuleRewrite { } if auxint != "" { - if rr.Typed { - // Make sure auxint value has the right type. - rr.add(stmtf("b.AuxInt = %sToAuxInt(%s)", unTitle(outdata.auxIntType()), auxint)) - } else { - rr.add(stmtf("b.AuxInt = %s", auxint)) - } + // Make sure auxint value has the right type. + rr.add(stmtf("b.AuxInt = %sToAuxInt(%s)", unTitle(outdata.auxIntType()), auxint)) } if aux != "" { - if rr.Typed { - // Make sure aux value has the right type. - rr.add(stmtf("b.Aux = %sToAux(%s)", unTitle(outdata.auxType()), aux)) - } else { - rr.add(stmtf("b.Aux = %s", aux)) - } + // Make sure aux value has the right type. + rr.add(stmtf("b.Aux = %sToAux(%s)", unTitle(outdata.auxType()), aux)) } succChanged := false @@ -1019,6 +998,19 @@ func genMatch0(rr *RuleRewrite, arch arch, match, v string, cnt map[string]int, pos = v + ".Pos" } + // If the last argument is ___, it means "don't care about trailing arguments, really" + // The likely/intended use is for rewrites that are too tricky to express in the existing pattern language + // Do a length check early because long patterns fed short (ultimately not-matching) inputs will + // do an indexing error in pattern-matching. + if op.argLength == -1 { + l := len(args) + if l == 0 || args[l-1] != "___" { + rr.add(breakf("len(%s.Args) != %d", v, l)) + } else if l > 1 && args[l-1] == "___" { + rr.add(breakf("len(%s.Args) < %d", v, l-1)) + } + } + for _, e := range []struct { name, field, dclType string }{ @@ -1029,15 +1021,6 @@ func genMatch0(rr *RuleRewrite, arch arch, match, v string, cnt map[string]int, if e.name == "" { continue } - if !rr.Typed { - if !token.IsIdentifier(e.name) || rr.declared(e.name) { - // code or variable - rr.add(breakf("%s.%s != %s", v, e.field, e.name)) - } else { - rr.add(declf(e.name, "%s.%s", v, e.field)) - } - continue - } if e.dclType == "" { log.Fatalf("op %s has no declared type for %s", op.name, e.field) @@ -1159,9 +1142,6 @@ func genMatch0(rr *RuleRewrite, arch arch, match, v string, cnt map[string]int, } } - if op.argLength == -1 { - rr.add(breakf("len(%s.Args) != %d", v, len(args))) - } return pos, checkOp } @@ -1230,20 +1210,12 @@ func genResult0(rr *RuleRewrite, arch arch, result string, top, move bool, pos s } if auxint != "" { - if rr.Typed { - // Make sure auxint value has the right type. - rr.add(stmtf("%s.AuxInt = %sToAuxInt(%s)", v, unTitle(op.auxIntType()), auxint)) - } else { - rr.add(stmtf("%s.AuxInt = %s", v, auxint)) - } + // Make sure auxint value has the right type. + rr.add(stmtf("%s.AuxInt = %sToAuxInt(%s)", v, unTitle(op.auxIntType()), auxint)) } if aux != "" { - if rr.Typed { - // Make sure aux value has the right type. - rr.add(stmtf("%s.Aux = %sToAux(%s)", v, unTitle(op.auxType()), aux)) - } else { - rr.add(stmtf("%s.Aux = %s", v, aux)) - } + // Make sure aux value has the right type. + rr.add(stmtf("%s.Aux = %sToAux(%s)", v, unTitle(op.auxType()), aux)) } all := new(strings.Builder) for i, arg := range args { @@ -1424,7 +1396,7 @@ func parseValue(val string, arch arch, loc string) (op opData, oparch, typ, auxi func opHasAuxInt(op opData) bool { switch op.aux { case "Bool", "Int8", "Int16", "Int32", "Int64", "Int128", "Float32", "Float64", - "SymOff", "SymValAndOff", "TypSize", "ARM64BitField", "FlagConstant", "CCop": + "SymOff", "CallOff", "SymValAndOff", "TypSize", "ARM64BitField", "FlagConstant", "CCop": return true } return false @@ -1432,7 +1404,7 @@ func opHasAuxInt(op opData) bool { func opHasAux(op opData) bool { switch op.aux { - case "String", "Sym", "SymOff", "SymValAndOff", "Typ", "TypSize", + case "String", "Sym", "SymOff", "Call", "CallOff", "SymValAndOff", "Typ", "TypSize", "S390XCCMask", "S390XRotateParams": return true } @@ -1524,7 +1496,7 @@ func excludeFromExpansion(s string, idx []int) bool { return true } right := s[idx[1]:] - if strings.Contains(left, "&&") && (strings.Contains(right, "->") || strings.Contains(right, "=>")) { + if strings.Contains(left, "&&") && strings.Contains(right, "=>") { // Inside && conditions. return true } @@ -1626,7 +1598,6 @@ func normalizeWhitespace(x string) string { x = strings.Replace(x, " )", ")", -1) x = strings.Replace(x, "[ ", "[", -1) x = strings.Replace(x, " ]", "]", -1) - x = strings.Replace(x, ")->", ") ->", -1) x = strings.Replace(x, ")=>", ") =>", -1) return x } @@ -1683,7 +1654,7 @@ func parseEllipsisRules(rules []Rule, arch arch) (newop string, ok bool) { return "", false } rule := rules[0] - match, cond, result, _ := rule.parse() + match, cond, result := rule.parse() if cond != "" || !isEllipsisValue(match) || !isEllipsisValue(result) { if strings.Contains(rule.Rule, "...") { log.Fatalf("%s: found ellipsis in non-ellipsis rule", rule.Loc) @@ -1708,7 +1679,7 @@ func isEllipsisValue(s string) bool { } func checkEllipsisRuleCandidate(rule Rule, arch arch) { - match, cond, result, _ := rule.parse() + match, cond, result := rule.parse() if cond != "" { return } @@ -1718,7 +1689,7 @@ func checkEllipsisRuleCandidate(rule Rule, arch arch) { var usingCopy string var eop opData if result[0] != '(' { - // Check for (Foo x) -> x, which can be converted to (Foo ...) -> (Copy ...). + // Check for (Foo x) => x, which can be converted to (Foo ...) => (Copy ...). args2 = []string{result} usingCopy = " using Copy" } else { @@ -1775,6 +1746,10 @@ func (op opData) auxType() string { return "Sym" case "SymOff": return "Sym" + case "Call": + return "Call" + case "CallOff": + return "Call" case "SymValAndOff": return "Sym" case "Typ": @@ -1809,6 +1784,8 @@ func (op opData) auxIntType() string { return "float32" case "Float64": return "float64" + case "CallOff": + return "int32" case "SymOff": return "int32" case "SymValAndOff": diff --git a/src/cmd/compile/internal/ssa/html.go b/src/cmd/compile/internal/ssa/html.go index ba37a804123..a9d52fa4ee0 100644 --- a/src/cmd/compile/internal/ssa/html.go +++ b/src/cmd/compile/internal/ssa/html.go @@ -28,18 +28,23 @@ type HTMLWriter struct { } func NewHTMLWriter(path string, f *Func, cfgMask string) *HTMLWriter { + path = strings.Replace(path, "/", string(filepath.Separator), -1) out, err := os.OpenFile(path, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644) if err != nil { f.Fatalf("%v", err) } - pwd, err := os.Getwd() - if err != nil { - f.Fatalf("%v", err) + reportPath := path + if !filepath.IsAbs(reportPath) { + pwd, err := os.Getwd() + if err != nil { + f.Fatalf("%v", err) + } + reportPath = filepath.Join(pwd, path) } html := HTMLWriter{ w: out, Func: f, - path: filepath.Join(pwd, path), + path: reportPath, dot: newDotWriter(cfgMask), } html.start() @@ -119,7 +124,8 @@ td.collapsed { } td.collapsed div { - /* TODO: Flip the direction of the phase's title 90 degrees on a collapsed column. */ + text-align: right; + transform: rotate(180deg); writing-mode: vertical-lr; white-space: pre; } @@ -357,6 +363,21 @@ body.darkmode ellipse.outline-black { outline: gray solid 2px; }