go/src/cmd/internal/obj/pcln.go

359 lines
10 KiB
Go
Raw Normal View History

// Copyright 2013 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 obj
import (
"cmd/internal/src"
"log"
)
func addvarint(d *Pcdata, v uint32) {
for ; v >= 0x80; v >>= 7 {
d.P = append(d.P, uint8(v|0x80))
}
d.P = append(d.P, uint8(v))
}
// funcpctab writes to dst a pc-value table mapping the code in func to the values
// returned by valfunc parameterized by arg. The invocation of valfunc to update the
// current value is, for each p,
//
// val = valfunc(func, val, p, 0, arg);
// record val as value at p->pc;
// val = valfunc(func, val, p, 1, arg);
//
// where func is the function, val is the current value, p is the instruction being
// considered, and arg can be used to further parameterize valfunc.
func funcpctab(ctxt *Link, dst *Pcdata, func_ *LSym, desc string, valfunc func(*Link, *LSym, int32, *Prog, int32, interface{}) int32, arg interface{}) {
dbg := desc == ctxt.Debugpcln
dst.P = dst.P[:0]
if dbg {
ctxt.Logf("funcpctab %s [valfunc=%s]\n", func_.Name, desc)
}
val := int32(-1)
oldval := val
if func_.Func.Text == nil {
return
}
pc := func_.Func.Text.Pc
if dbg {
ctxt.Logf("%6x %6d %v\n", uint64(pc), val, func_.Func.Text)
}
started := false
var delta uint32
for p := func_.Func.Text; p != nil; p = p.Link {
// Update val. If it's not changing, keep going.
val = valfunc(ctxt, func_, val, p, 0, arg)
if val == oldval && started {
val = valfunc(ctxt, func_, val, p, 1, arg)
if dbg {
ctxt.Logf("%6x %6s %v\n", uint64(p.Pc), "", p)
}
continue
}
// If the pc of the next instruction is the same as the
// pc of this instruction, this instruction is not a real
// instruction. Keep going, so that we only emit a delta
// for a true instruction boundary in the program.
if p.Link != nil && p.Link.Pc == p.Pc {
val = valfunc(ctxt, func_, val, p, 1, arg)
if dbg {
ctxt.Logf("%6x %6s %v\n", uint64(p.Pc), "", p)
}
continue
}
// The table is a sequence of (value, pc) pairs, where each
// pair states that the given value is in effect from the current position
// up to the given pc, which becomes the new current position.
// To generate the table as we scan over the program instructions,
// we emit a "(value" when pc == func->value, and then
// each time we observe a change in value we emit ", pc) (value".
// When the scan is over, we emit the closing ", pc)".
//
// The table is delta-encoded. The value deltas are signed and
// transmitted in zig-zag form, where a complement bit is placed in bit 0,
// and the pc deltas are unsigned. Both kinds of deltas are sent
// as variable-length little-endian base-128 integers,
// where the 0x80 bit indicates that the integer continues.
if dbg {
ctxt.Logf("%6x %6d %v\n", uint64(p.Pc), val, p)
}
if started {
addvarint(dst, uint32((p.Pc-pc)/int64(ctxt.Arch.MinLC)))
pc = p.Pc
}
delta = uint32(val) - uint32(oldval)
if delta>>31 != 0 {
delta = 1 | ^(delta << 1)
} else {
delta <<= 1
}
addvarint(dst, delta)
oldval = val
started = true
val = valfunc(ctxt, func_, val, p, 1, arg)
}
if started {
if dbg {
ctxt.Logf("%6x done\n", uint64(func_.Func.Text.Pc+func_.Size))
}
addvarint(dst, uint32((func_.Size-pc)/int64(ctxt.Arch.MinLC)))
addvarint(dst, 0) // terminator
}
if dbg {
ctxt.Logf("wrote %d bytes to %p\n", len(dst.P), dst)
for i := 0; i < len(dst.P); i++ {
ctxt.Logf(" %02x", dst.P[i])
}
ctxt.Logf("\n")
}
}
// pctofileline computes either the file number (arg == 0)
// or the line number (arg == 1) to use at p.
// Because p.Pos applies to p, phase == 0 (before p)
// takes care of the update.
func pctofileline(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg interface{}) int32 {
if p.As == ATEXT || p.As == ANOP || p.Pos.Line() == 0 || phase == 1 {
return oldval
}
[dev.inline] cmd/internal/src: replace src.Pos with syntax.Pos This replaces the src.Pos LineHist-based position tracking with the syntax.Pos implementation and updates all uses. The LineHist table is not used anymore - the respective code is still there but should be removed eventually. CL forthcoming. Passes toolstash -cmp when comparing to the master repo (with the exception of a couple of swapped assembly instructions, likely due to different instruction scheduling because the line-based sorting has changed; though this is won't affect correctness). The sizes of various important compiler data structures have increased significantly (see the various sizes_test.go files); this is probably the reason for an increase of compilation times (to be addressed). Here are the results of compilebench -count 5, run on a "quiet" machine (no apps running besides a terminal): name old time/op new time/op delta Template 256ms ± 1% 280ms ±15% +9.54% (p=0.008 n=5+5) Unicode 132ms ± 1% 132ms ± 1% ~ (p=0.690 n=5+5) GoTypes 891ms ± 1% 917ms ± 2% +2.88% (p=0.008 n=5+5) Compiler 3.84s ± 2% 3.99s ± 2% +3.95% (p=0.016 n=5+5) MakeBash 47.1s ± 1% 47.2s ± 2% ~ (p=0.841 n=5+5) name old user-ns/op new user-ns/op delta Template 309M ± 1% 326M ± 2% +5.18% (p=0.008 n=5+5) Unicode 165M ± 1% 168M ± 4% ~ (p=0.421 n=5+5) GoTypes 1.14G ± 2% 1.18G ± 1% +3.47% (p=0.008 n=5+5) Compiler 5.00G ± 1% 5.16G ± 1% +3.12% (p=0.008 n=5+5) Change-Id: I241c4246cdff627d7ecb95cac23060b38f9775ec Reviewed-on: https://go-review.googlesource.com/34273 Run-TryBot: Robert Griesemer <gri@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2016-12-09 17:15:05 -08:00
f, l := linkgetlineFromPos(ctxt, p.Pos)
if arg == nil {
return l
}
pcln := arg.(*Pcln)
if f == pcln.Lastfile {
return int32(pcln.Lastindex)
}
for i, file := range pcln.File {
if file == f {
pcln.Lastfile = f
pcln.Lastindex = i
return int32(i)
}
}
i := len(pcln.File)
pcln.File = append(pcln.File, f)
pcln.Lastfile = f
pcln.Lastindex = i
return int32(i)
}
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
// pcinlineState holds the state used to create a function's inlining
// tree and the PC-value table that maps PCs to nodes in that tree.
type pcinlineState struct {
globalToLocal map[int]int
localTree InlTree
}
// addBranch adds a branch from the global inlining tree in ctxt to
// the function's local inlining tree, returning the index in the local tree.
func (s *pcinlineState) addBranch(ctxt *Link, globalIndex int) int {
if globalIndex < 0 {
return -1
}
localIndex, ok := s.globalToLocal[globalIndex]
if ok {
return localIndex
}
// Since tracebacks don't include column information, we could
// use one node for multiple calls of the same function on the
// same line (e.g., f(x) + f(y)). For now, we use one node for
// each inlined call.
call := ctxt.InlTree.nodes[globalIndex]
call.Parent = s.addBranch(ctxt, call.Parent)
localIndex = len(s.localTree.nodes)
s.localTree.nodes = append(s.localTree.nodes, call)
s.globalToLocal[globalIndex] = localIndex
return localIndex
}
// pctoinline computes the index into the local inlining tree to use at p.
// If p is not the result of inlining, pctoinline returns -1. Because p.Pos
// applies to p, phase == 0 (before p) takes care of the update.
func (s *pcinlineState) pctoinline(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg interface{}) int32 {
if phase == 1 {
return oldval
}
posBase := ctxt.PosTable.Pos(p.Pos).Base()
if posBase == nil {
return -1
}
globalIndex := posBase.InliningIndex()
if globalIndex < 0 {
return -1
}
if s.globalToLocal == nil {
s.globalToLocal = make(map[int]int)
}
return int32(s.addBranch(ctxt, globalIndex))
}
// pctospadj computes the sp adjustment in effect.
// It is oldval plus any adjustment made by p itself.
// The adjustment by p takes effect only after p, so we
// apply the change during phase == 1.
func pctospadj(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg interface{}) int32 {
if oldval == -1 { // starting
oldval = 0
}
if phase == 0 {
return oldval
}
if oldval+p.Spadj < -10000 || oldval+p.Spadj > 1100000000 {
ctxt.Diag("overflow in spadj: %d + %d = %d", oldval, p.Spadj, oldval+p.Spadj)
ctxt.DiagFlush()
log.Fatalf("bad code")
}
return oldval + p.Spadj
}
// pctostmt returns either,
// if phase==0, then whether the current instruction is a step-target (Dwarf is_stmt)
// else (phase == 1), zero.
//
func pctostmt(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg interface{}) int32 {
if phase == 1 {
return 0 // Ignored; also different from initial value of -1, if that ever matters.
}
s := p.Pos.IsStmt()
if s == src.PosIsStmt {
return 1
}
if s == src.PosNotStmt { // includes NoSrcPos case
return 0
}
// Line numbers in .s files will have no special setting, therefore default to is_stmt=1.
return 1
}
// pctopcdata computes the pcdata value in effect at p.
// A PCDATA instruction sets the value in effect at future
// non-PCDATA instructions.
// Since PCDATA instructions have no width in the final code,
// it does not matter which phase we use for the update.
func pctopcdata(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg interface{}) int32 {
if phase == 0 || p.As != APCDATA || p.From.Offset != int64(arg.(uint32)) {
return oldval
}
if int64(int32(p.To.Offset)) != p.To.Offset {
ctxt.Diag("overflow in PCDATA instruction: %v", p)
ctxt.DiagFlush()
log.Fatalf("bad code")
}
return int32(p.To.Offset)
}
// stmtData writes out pc-linked is_stmt data for eventual use in the DWARF line numbering table.
func stmtData(ctxt *Link, cursym *LSym) {
var pctostmtData Pcdata
funcpctab(ctxt, &pctostmtData, cursym, "pctostmt", pctostmt, nil)
cursym.Func.dwarfIsStmtSym.P = pctostmtData.P
}
func linkpcln(ctxt *Link, cursym *LSym) {
pcln := &cursym.Func.Pcln
npcdata := 0
nfuncdata := 0
for p := cursym.Func.Text; p != nil; p = p.Link {
// Find the highest ID of any used PCDATA table. This ignores PCDATA table
// that consist entirely of "-1", since that's the assumed default value.
// From.Offset is table ID
// To.Offset is data
if p.As == APCDATA && p.From.Offset >= int64(npcdata) && p.To.Offset != -1 { // ignore -1 as we start at -1, if we only see -1, nothing changed
npcdata = int(p.From.Offset + 1)
}
// Find the highest ID of any FUNCDATA table.
// From.Offset is table ID
if p.As == AFUNCDATA && p.From.Offset >= int64(nfuncdata) {
nfuncdata = int(p.From.Offset + 1)
}
}
pcln.Pcdata = make([]Pcdata, npcdata)
pcln.Pcdata = pcln.Pcdata[:npcdata]
pcln.Funcdata = make([]*LSym, nfuncdata)
pcln.Funcdataoff = make([]int64, nfuncdata)
pcln.Funcdataoff = pcln.Funcdataoff[:nfuncdata]
funcpctab(ctxt, &pcln.Pcsp, cursym, "pctospadj", pctospadj, nil)
funcpctab(ctxt, &pcln.Pcfile, cursym, "pctofile", pctofileline, pcln)
funcpctab(ctxt, &pcln.Pcline, cursym, "pctoline", pctofileline, nil)
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
pcinlineState := new(pcinlineState)
funcpctab(ctxt, &pcln.Pcinline, cursym, "pctoinline", pcinlineState.pctoinline, nil)
pcln.InlTree = pcinlineState.localTree
if ctxt.Debugpcln == "pctoinline" && len(pcln.InlTree.nodes) > 0 {
ctxt.Logf("-- inlining tree for %s:\n", cursym)
dumpInlTree(ctxt, pcln.InlTree)
ctxt.Logf("--\n")
}
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
// tabulate which pc and func data we have.
havepc := make([]uint32, (npcdata+31)/32)
havefunc := make([]uint32, (nfuncdata+31)/32)
for p := cursym.Func.Text; p != nil; p = p.Link {
if p.As == AFUNCDATA {
if (havefunc[p.From.Offset/32]>>uint64(p.From.Offset%32))&1 != 0 {
ctxt.Diag("multiple definitions for FUNCDATA $%d", p.From.Offset)
}
havefunc[p.From.Offset/32] |= 1 << uint64(p.From.Offset%32)
}
if p.As == APCDATA && p.To.Offset != -1 {
havepc[p.From.Offset/32] |= 1 << uint64(p.From.Offset%32)
}
}
// pcdata.
for i := 0; i < npcdata; i++ {
if (havepc[i/32]>>uint(i%32))&1 == 0 {
continue
}
funcpctab(ctxt, &pcln.Pcdata[i], cursym, "pctopcdata", pctopcdata, interface{}(uint32(i)))
}
// funcdata
if nfuncdata > 0 {
var i int
for p := cursym.Func.Text; p != nil; p = p.Link {
if p.As == AFUNCDATA {
i = int(p.From.Offset)
pcln.Funcdataoff[i] = p.To.Offset
if p.To.Type != TYPE_CONST {
// TODO: Dedup.
//funcdata_bytes += p->to.sym->size;
pcln.Funcdata[i] = p.To.Sym
}
}
}
}
}