Stowage/go
2
0
Fork 0
mirror of https://github.com/golang/go.git synced 2025-12-08 06:10:04 +00:00
Code Issues Projects Releases Packages Wiki Activity
go/src/cmd/compile/internal/ssa/gen/S390XOps.go

666 lines
42 KiB
Go
Raw Normal View History

cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// Copyright 2016 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
import "strings"
// Notes:
// - Integer types live in the low portion of registers. Upper portions are junk.
// - Boolean types use the low-order byte of a register. 0=false, 1=true.
// Upper bytes are junk.
// - When doing sub-register operations, we try to write the whole
// destination register to avoid a partial-register write.
// - Unused portions of AuxInt (or the Val portion of ValAndOff) are
cmd/compile/internal/ssa/gen: add comment on SB-addressing on s390x During the review of CL 38801 it was noted that it would be nice to have a bit more clarity on how-and-why SB addressing is handled strangely on s390x. This additional comment should hopefully help. In general SB is handled differently because not all instructions have variants that use relative addressing. Change-Id: I3379012ae3f167478c191c435939c3b876c645ed Reviewed-on: https://go-review.googlesource.com/38952 Reviewed-by: Keith Randall <khr@golang.org>
2017-03-30 15:58:01 -04:00
// filled by sign-extending the used portion. Users of AuxInt which interpret
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// AuxInt as unsigned (e.g. shifts) must be careful.
cmd/compile/internal/ssa/gen: add comment on SB-addressing on s390x During the review of CL 38801 it was noted that it would be nice to have a bit more clarity on how-and-why SB addressing is handled strangely on s390x. This additional comment should hopefully help. In general SB is handled differently because not all instructions have variants that use relative addressing. Change-Id: I3379012ae3f167478c191c435939c3b876c645ed Reviewed-on: https://go-review.googlesource.com/38952 Reviewed-by: Keith Randall <khr@golang.org>
2017-03-30 15:58:01 -04:00
// - The SB 'register' is implemented using instruction-relative addressing. This
// places some limitations on when and how memory operands that are addressed
// relative to SB can be used:
//
// 1. Pseudo-instructions do not always map to a single machine instruction when
// using the SB 'register' to address data. This is because many machine
// instructions do not have relative long (RL suffix) equivalents. For example,
// ADDload, which is assembled as AG.
//
// 2. Loads and stores using relative addressing require the data be aligned
// according to its size (8-bytes for double words, 4-bytes for words
// and so on).
//
// We can always work around these by inserting LARL instructions (load address
// relative long) in the assembler, but typically this results in worse code
// generation because the address can't be re-used. Inserting instructions in the
// assembler also means clobbering the temp register and it is a long-term goal
// to prevent the compiler doing this so that it can be allocated as a normal
// register.
//
// For more information about the z/Architecture, the instruction set and the
// addressing modes it supports take a look at the z/Architecture Principles of
// Operation: http://publibfp.boulder.ibm.com/epubs/pdf/dz9zr010.pdf
//
// Suffixes encode the bit width of pseudo-instructions.
// D (double word) = 64 bit (frequently omitted)
// W (word) = 32 bit
// H (half word) = 16 bit
// B (byte) = 8 bit
// S (single prec.) = 32 bit (double precision is omitted)
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// copied from ../../s390x/reg.go
var regNamesS390X = []string{
"R0",
"R1",
"R2",
"R3",
"R4",
"R5",
"R6",
"R7",
"R8",
"R9",
"R10",
"R11",
"R12",
"g", // R13
"R14",
"SP", // R15
"F0",
"F1",
"F2",
"F3",
"F4",
"F5",
"F6",
"F7",
"F8",
"F9",
"F10",
"F11",
"F12",
"F13",
"F14",
"F15",
//pseudo-registers
"SB",
}
func init() {
// Make map from reg names to reg integers.
if len(regNamesS390X) > 64 {
panic("too many registers")
}
num := map[string]int{}
for i, name := range regNamesS390X {
num[name] = i
}
buildReg := func(s string) regMask {
m := regMask(0)
for _, r := range strings.Split(s, " ") {
if n, ok := num[r]; ok {
m |= regMask(1) << uint(n)
continue
}
panic("register " + r + " not found")
}
return m
}
// Common individual register masks
var (
sp = buildReg("SP")
sb = buildReg("SB")
r0 = buildReg("R0")
// R10 and R11 are reserved by the assembler.
cmd/compile: make link register allocatable in non-leaf functions We save and restore the link register in non-leaf functions because it is clobbered by CALLs. It is therefore available for general purpose use. Only enabled on s390x currently. The RC4 benchmarks in particular benefit from the extra register: name old speed new speed delta RC4_128 243MB/s ± 2% 341MB/s ± 2% +40.46% (p=0.008 n=5+5) RC4_1K 267MB/s ± 0% 359MB/s ± 1% +34.32% (p=0.008 n=5+5) RC4_8K 271MB/s ± 0% 362MB/s ± 0% +33.61% (p=0.008 n=5+5) Change-Id: Id23bff95e771da9425353da2f32668b8e34ba09f Reviewed-on: https://go-review.googlesource.com/30597 Reviewed-by: Cherry Zhang <cherryyz@google.com> Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-10-06 15:06:45 -04:00
gp = buildReg("R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14")
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
gpsp = gp | sp
// R0 is considered to contain the value 0 in address calculations.
ptr = gp &^ r0
ptrsp = ptr | sp
ptrspsb = ptrsp | sb
fp = buildReg("F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15")
callerSave = gp | fp
)
// Common slices of register masks
var (
gponly = []regMask{gp}
fponly = []regMask{fp}
)
// Common regInfo
var (
gp01 = regInfo{inputs: []regMask{}, outputs: gponly}
gp11 = regInfo{inputs: []regMask{gp}, outputs: gponly}
gp11sp = regInfo{inputs: []regMask{gpsp}, outputs: gponly}
gp21 = regInfo{inputs: []regMask{gp, gp}, outputs: gponly}
gp21sp = regInfo{inputs: []regMask{gpsp, gp}, outputs: gponly}
// R0 evaluates to 0 when used as the number of bits to shift
// so we need to exclude it from that operand.
sh21 = regInfo{inputs: []regMask{gp, ptr}, outputs: gponly}
addr = regInfo{inputs: []regMask{sp | sb}, outputs: gponly}
addridx = regInfo{inputs: []regMask{sp | sb, ptrsp}, outputs: gponly}
gp2flags = regInfo{inputs: []regMask{gpsp, gpsp}}
gp1flags = regInfo{inputs: []regMask{gpsp}}
flagsgp = regInfo{outputs: gponly}
gp2flags1 = regInfo{inputs: []regMask{gp, gp}, outputs: gponly}
gpload = regInfo{inputs: []regMask{ptrspsb, 0}, outputs: gponly}
gploadidx = regInfo{inputs: []regMask{ptrspsb, ptrsp, 0}, outputs: gponly}
cmd/compile: merge loads into operations on s390x Adds the new canMergeLoad function which can be used by rules to decide whether a load can be merged into an operation. The function ensures that the merge will not reorder the load relative to memory operations (for example, stores) in such a way that the block can no longer be scheduled. This new function enables transformations such as: MOVD 0(R1), R2 ADD R2, R3 to: ADD 0(R1), R3 The two-operand form of the following instructions can now read a single memory operand: - ADD - ADDC - ADDW - MULLD - MULLW - SUB - SUBC - SUBE - SUBW - AND - ANDW - OR - ORW - XOR - XORW Improves SHA3 performance by 6-8%. Updates #15054. Change-Id: Ibcb9122126cd1a26f2c01c0dfdbb42fe5e7b5b94 Reviewed-on: https://go-review.googlesource.com/29272 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-14 10:42:14 -04:00
gpopload = regInfo{inputs: []regMask{gp, ptrsp, 0}, outputs: gponly}
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
gpstore = regInfo{inputs: []regMask{ptrspsb, gpsp, 0}}
gpstoreconst = regInfo{inputs: []regMask{ptrspsb, 0}}
gpstoreidx = regInfo{inputs: []regMask{ptrsp, ptrsp, gpsp, 0}}
cmd/compile: improve load/store merging on s390x This commit makes the process of load/store merging more incremental for both big and little endian operations. It also adds support for 32-bit shifts (needed to merge 16- and 32-bit loads/stores). In addition, the merging of little endian stores is now supported. Little endian stores are now up to 30 times faster. Change-Id: Iefdd81eda4a65b335f23c3ff222146540083ad9c Reviewed-on: https://go-review.googlesource.com/29956 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-27 20:30:01 -04:00
gpstorebr = regInfo{inputs: []regMask{ptrsp, gpsp, 0}}
cmd/compile: intrinsify atomic operations on s390x Implements the following intrinsics on s390x: - AtomicAdd{32,64} - AtomicCompareAndSwap{32,64} - AtomicExchange{32,64} - AtomicLoad{32,64,Ptr} - AtomicStore{32,64,PtrNoWB} I haven't added rules for And8 or Or8 yet. Change-Id: I647af023a8e513718e90e98a60191e7af6167314 Reviewed-on: https://go-review.googlesource.com/31614 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-10-19 16:41:01 -04:00
gpstorelaa = regInfo{inputs: []regMask{ptrspsb, gpsp, 0}, outputs: gponly}
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
gpmvc = regInfo{inputs: []regMask{ptrsp, ptrsp, 0}}
fp01 = regInfo{inputs: []regMask{}, outputs: fponly}
fp21 = regInfo{inputs: []regMask{fp, fp}, outputs: fponly}
cmd/compile: emit fused multiply-{add,subtract} instructions on s390x Explcitly block fused multiply-add pattern matching when a cast is used after the multiplication, for example: - (a * b) + c // can emit fused multiply-add - float64(a * b) + c // cannot emit fused multiply-add float{32,64} and complex{64,128} casts of matching types are now kept as OCONV operations rather than being replaced with OCONVNOP operations because they now imply a rounding operation (and therefore aren't a no-op anymore). Operations (for example, multiplication) on complex types may utilize fused multiply-add and -subtract instructions internally. There is no way to disable this behavior at the moment. Improves the performance of the floating point implementation of poly1305: name old speed new speed delta 64 246MB/s ± 0% 275MB/s ± 0% +11.48% (p=0.000 n=10+8) 1K 312MB/s ± 0% 357MB/s ± 0% +14.41% (p=0.000 n=10+10) 64Unaligned 246MB/s ± 0% 274MB/s ± 0% +11.43% (p=0.000 n=10+10) 1KUnaligned 312MB/s ± 0% 357MB/s ± 0% +14.39% (p=0.000 n=10+8) Updates #17895. Change-Id: Ia771d275bb9150d1a598f8cc773444663de5ce16 Reviewed-on: https://go-review.googlesource.com/36963 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-02-12 22:12:12 -05:00
fp31 = regInfo{inputs: []regMask{fp, fp, fp}, outputs: fponly}
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
fp21clobber = regInfo{inputs: []regMask{fp, fp}, outputs: fponly}
fpgp = regInfo{inputs: fponly, outputs: gponly}
gpfp = regInfo{inputs: gponly, outputs: fponly}
fp11 = regInfo{inputs: fponly, outputs: fponly}
fp11clobber = regInfo{inputs: fponly, outputs: fponly}
fp2flags = regInfo{inputs: []regMask{fp, fp}}
fpload = regInfo{inputs: []regMask{ptrspsb, 0}, outputs: fponly}
fploadidx = regInfo{inputs: []regMask{ptrsp, ptrsp, 0}, outputs: fponly}
fpstore = regInfo{inputs: []regMask{ptrspsb, fp, 0}}
fpstoreidx = regInfo{inputs: []regMask{ptrsp, ptrsp, fp, 0}}
cmd/compile: intrinsify atomic operations on s390x Implements the following intrinsics on s390x: - AtomicAdd{32,64} - AtomicCompareAndSwap{32,64} - AtomicExchange{32,64} - AtomicLoad{32,64,Ptr} - AtomicStore{32,64,PtrNoWB} I haven't added rules for And8 or Or8 yet. Change-Id: I647af023a8e513718e90e98a60191e7af6167314 Reviewed-on: https://go-review.googlesource.com/31614 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-10-19 16:41:01 -04:00
// LoweredAtomicCas may overwrite arg1, so force it to R0 for now.
cas = regInfo{inputs: []regMask{ptrsp, r0, gpsp, 0}, outputs: []regMask{gp, 0}, clobbers: r0}
// LoweredAtomicExchange overwrites the output before executing
// CS{,G}, so the output register must not be the same as the
// input register. For now we just force the output register to
// R0.
exchange = regInfo{inputs: []regMask{ptrsp, gpsp &^ r0, 0}, outputs: []regMask{r0, 0}}
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
)
var S390Xops = []opData{
// fp ops
cmd/compile: emit fused multiply-{add,subtract} instructions on s390x Explcitly block fused multiply-add pattern matching when a cast is used after the multiplication, for example: - (a * b) + c // can emit fused multiply-add - float64(a * b) + c // cannot emit fused multiply-add float{32,64} and complex{64,128} casts of matching types are now kept as OCONV operations rather than being replaced with OCONVNOP operations because they now imply a rounding operation (and therefore aren't a no-op anymore). Operations (for example, multiplication) on complex types may utilize fused multiply-add and -subtract instructions internally. There is no way to disable this behavior at the moment. Improves the performance of the floating point implementation of poly1305: name old speed new speed delta 64 246MB/s ± 0% 275MB/s ± 0% +11.48% (p=0.000 n=10+8) 1K 312MB/s ± 0% 357MB/s ± 0% +14.41% (p=0.000 n=10+10) 64Unaligned 246MB/s ± 0% 274MB/s ± 0% +11.43% (p=0.000 n=10+10) 1KUnaligned 312MB/s ± 0% 357MB/s ± 0% +14.39% (p=0.000 n=10+8) Updates #17895. Change-Id: Ia771d275bb9150d1a598f8cc773444663de5ce16 Reviewed-on: https://go-review.googlesource.com/36963 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-02-12 22:12:12 -05:00
{name: "FADDS", argLength: 2, reg: fp21clobber, asm: "FADDS", commutative: true, resultInArg0: true, clobberFlags: true}, // fp32 arg0 + arg1
{name: "FADD", argLength: 2, reg: fp21clobber, asm: "FADD", commutative: true, resultInArg0: true, clobberFlags: true}, // fp64 arg0 + arg1
{name: "FSUBS", argLength: 2, reg: fp21clobber, asm: "FSUBS", resultInArg0: true, clobberFlags: true}, // fp32 arg0 - arg1
{name: "FSUB", argLength: 2, reg: fp21clobber, asm: "FSUB", resultInArg0: true, clobberFlags: true}, // fp64 arg0 - arg1
{name: "FMULS", argLength: 2, reg: fp21, asm: "FMULS", commutative: true, resultInArg0: true}, // fp32 arg0 * arg1
{name: "FMUL", argLength: 2, reg: fp21, asm: "FMUL", commutative: true, resultInArg0: true}, // fp64 arg0 * arg1
{name: "FDIVS", argLength: 2, reg: fp21, asm: "FDIVS", resultInArg0: true}, // fp32 arg0 / arg1
{name: "FDIV", argLength: 2, reg: fp21, asm: "FDIV", resultInArg0: true}, // fp64 arg0 / arg1
{name: "FNEGS", argLength: 1, reg: fp11clobber, asm: "FNEGS", clobberFlags: true}, // fp32 -arg0
{name: "FNEG", argLength: 1, reg: fp11clobber, asm: "FNEG", clobberFlags: true}, // fp64 -arg0
{name: "FMADDS", argLength: 3, reg: fp31, asm: "FMADDS", resultInArg0: true}, // fp32 arg1 * arg2 + arg0
{name: "FMADD", argLength: 3, reg: fp31, asm: "FMADD", resultInArg0: true}, // fp64 arg1 * arg2 + arg0
{name: "FMSUBS", argLength: 3, reg: fp31, asm: "FMSUBS", resultInArg0: true}, // fp32 arg1 * arg2 - arg0
{name: "FMSUB", argLength: 3, reg: fp31, asm: "FMSUB", resultInArg0: true}, // fp64 arg1 * arg2 - arg0
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "FMOVSload", argLength: 2, reg: fpload, asm: "FMOVS", aux: "SymOff", faultOnNilArg0: true, symEffect: "Read"}, // fp32 load
{name: "FMOVDload", argLength: 2, reg: fpload, asm: "FMOVD", aux: "SymOff", faultOnNilArg0: true, symEffect: "Read"}, // fp64 load
{name: "FMOVSconst", reg: fp01, asm: "FMOVS", aux: "Float32", rematerializeable: true}, // fp32 constant
{name: "FMOVDconst", reg: fp01, asm: "FMOVD", aux: "Float64", rematerializeable: true}, // fp64 constant
{name: "FMOVSloadidx", argLength: 3, reg: fploadidx, asm: "FMOVS", aux: "SymOff", symEffect: "Read"}, // fp32 load indexed by i
{name: "FMOVDloadidx", argLength: 3, reg: fploadidx, asm: "FMOVD", aux: "SymOff", symEffect: "Read"}, // fp64 load indexed by i
{name: "FMOVSstore", argLength: 3, reg: fpstore, asm: "FMOVS", aux: "SymOff", faultOnNilArg0: true, symEffect: "Write"}, // fp32 store
{name: "FMOVDstore", argLength: 3, reg: fpstore, asm: "FMOVD", aux: "SymOff", faultOnNilArg0: true, symEffect: "Write"}, // fp64 store
{name: "FMOVSstoreidx", argLength: 4, reg: fpstoreidx, asm: "FMOVS", aux: "SymOff", symEffect: "Write"}, // fp32 indexed by i store
{name: "FMOVDstoreidx", argLength: 4, reg: fpstoreidx, asm: "FMOVD", aux: "SymOff", symEffect: "Write"}, // fp64 indexed by i store
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// binary ops
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "ADD", argLength: 2, reg: gp21sp, asm: "ADD", commutative: true, clobberFlags: true}, // arg0 + arg1
{name: "ADDW", argLength: 2, reg: gp21sp, asm: "ADDW", commutative: true, clobberFlags: true}, // arg0 + arg1
{name: "ADDconst", argLength: 1, reg: gp11sp, asm: "ADD", aux: "Int64", typ: "UInt64", clobberFlags: true}, // arg0 + auxint
{name: "ADDWconst", argLength: 1, reg: gp11sp, asm: "ADDW", aux: "Int32", clobberFlags: true}, // arg0 + auxint
{name: "ADDload", argLength: 3, reg: gpopload, asm: "ADD", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 + *arg1. arg2=mem
{name: "ADDWload", argLength: 3, reg: gpopload, asm: "ADDW", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 + *arg1. arg2=mem
{name: "SUB", argLength: 2, reg: gp21, asm: "SUB", clobberFlags: true}, // arg0 - arg1
{name: "SUBW", argLength: 2, reg: gp21, asm: "SUBW", clobberFlags: true}, // arg0 - arg1
{name: "SUBconst", argLength: 1, reg: gp11, asm: "SUB", aux: "Int64", resultInArg0: true, clobberFlags: true}, // arg0 - auxint
{name: "SUBWconst", argLength: 1, reg: gp11, asm: "SUBW", aux: "Int32", resultInArg0: true, clobberFlags: true}, // arg0 - auxint
{name: "SUBload", argLength: 3, reg: gpopload, asm: "SUB", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 - *arg1. arg2=mem
{name: "SUBWload", argLength: 3, reg: gpopload, asm: "SUBW", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 - *arg1. arg2=mem
{name: "MULLD", argLength: 2, reg: gp21, asm: "MULLD", typ: "Int64", commutative: true, resultInArg0: true, clobberFlags: true}, // arg0 * arg1
{name: "MULLW", argLength: 2, reg: gp21, asm: "MULLW", typ: "Int32", commutative: true, resultInArg0: true, clobberFlags: true}, // arg0 * arg1
{name: "MULLDconst", argLength: 1, reg: gp11, asm: "MULLD", aux: "Int64", typ: "Int64", resultInArg0: true, clobberFlags: true}, // arg0 * auxint
{name: "MULLWconst", argLength: 1, reg: gp11, asm: "MULLW", aux: "Int32", typ: "Int32", resultInArg0: true, clobberFlags: true}, // arg0 * auxint
{name: "MULLDload", argLength: 3, reg: gpopload, asm: "MULLD", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 * *arg1. arg2=mem
{name: "MULLWload", argLength: 3, reg: gpopload, asm: "MULLW", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 * *arg1. arg2=mem
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile: automatically handle commuting ops in rewrite rules Note that this is a redo of an undo of the original buggy CL 38666. We have lots of rewrite rules that vary only in the fact that we have 2 versions for the 2 different orderings of various commuting ops. For example: (ADDL x (MOVLconst [c])) -> (ADDLconst [c] x) (ADDL (MOVLconst [c]) x) -> (ADDLconst [c] x) It can get unwieldly quickly, especially when there is more than one commuting op in a rule. Our existing "fix" for this problem is to have rules that canonicalize the operations first. For example: (Eq64 x (Const64 <t> [c])) && x.Op != OpConst64 -> (Eq64 (Const64 <t> [c]) x) Subsequent rules can then assume if there is a constant arg to Eq64, it will be the first one. This fix kinda works, but it is fragile and only works when we remember to include the required extra rules. The fundamental problem is that the rule matcher doesn't know anything about commuting ops. This CL fixes that fact. We already have information about which ops commute. (The register allocator takes advantage of commutivity.) The rule generator now automatically generates multiple rules for a single source rule when there are commutative ops in the rule. We can now drop all of our almost-duplicate source-level rules and the canonicalization rules. I have some CLs in progress that will be a lot less verbose when the rule generator handles commutivity for me. I had to reorganize the load-combining rules a bit. The 8-way OR rules generated 128 different reorderings, which was causing the generator to put too much code in the rewrite*.go files (the big ones were going from 25K lines to 132K lines). Instead I reorganized the rules to combine pairs of loads at a time. The generated rule files are now actually a bit (5%) smaller. Make.bash times are ~unchanged. Compiler benchmarks are not observably different. Probably because we don't spend much compiler time in rule matching anyway. I've also done a pass over all of our ops adding commutative markings for ops which hadn't had them previously. Fixes #18292 Change-Id: Ic1c0e43fbf579539f459971625f69690c9ab8805 Reviewed-on: https://go-review.googlesource.com/38801 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2017-03-30 03:30:22 +00:00
{name: "MULHD", argLength: 2, reg: gp21, asm: "MULHD", typ: "Int64", commutative: true, resultInArg0: true, clobberFlags: true}, // (arg0 * arg1) >> width
{name: "MULHDU", argLength: 2, reg: gp21, asm: "MULHDU", typ: "Int64", commutative: true, resultInArg0: true, clobberFlags: true}, // (arg0 * arg1) >> width
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
{name: "DIVD", argLength: 2, reg: gp21, asm: "DIVD", resultInArg0: true, clobberFlags: true}, // arg0 / arg1
{name: "DIVW", argLength: 2, reg: gp21, asm: "DIVW", resultInArg0: true, clobberFlags: true}, // arg0 / arg1
{name: "DIVDU", argLength: 2, reg: gp21, asm: "DIVDU", resultInArg0: true, clobberFlags: true}, // arg0 / arg1
{name: "DIVWU", argLength: 2, reg: gp21, asm: "DIVWU", resultInArg0: true, clobberFlags: true}, // arg0 / arg1
{name: "MODD", argLength: 2, reg: gp21, asm: "MODD", resultInArg0: true, clobberFlags: true}, // arg0 % arg1
{name: "MODW", argLength: 2, reg: gp21, asm: "MODW", resultInArg0: true, clobberFlags: true}, // arg0 % arg1
{name: "MODDU", argLength: 2, reg: gp21, asm: "MODDU", resultInArg0: true, clobberFlags: true}, // arg0 % arg1
{name: "MODWU", argLength: 2, reg: gp21, asm: "MODWU", resultInArg0: true, clobberFlags: true}, // arg0 % arg1
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "AND", argLength: 2, reg: gp21, asm: "AND", commutative: true, clobberFlags: true}, // arg0 & arg1
{name: "ANDW", argLength: 2, reg: gp21, asm: "ANDW", commutative: true, clobberFlags: true}, // arg0 & arg1
{name: "ANDconst", argLength: 1, reg: gp11, asm: "AND", aux: "Int64", resultInArg0: true, clobberFlags: true}, // arg0 & auxint
{name: "ANDWconst", argLength: 1, reg: gp11, asm: "ANDW", aux: "Int32", resultInArg0: true, clobberFlags: true}, // arg0 & auxint
{name: "ANDload", argLength: 3, reg: gpopload, asm: "AND", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 & *arg1. arg2=mem
{name: "ANDWload", argLength: 3, reg: gpopload, asm: "ANDW", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 & *arg1. arg2=mem
{name: "OR", argLength: 2, reg: gp21, asm: "OR", commutative: true, clobberFlags: true}, // arg0 | arg1
{name: "ORW", argLength: 2, reg: gp21, asm: "ORW", commutative: true, clobberFlags: true}, // arg0 | arg1
{name: "ORconst", argLength: 1, reg: gp11, asm: "OR", aux: "Int64", resultInArg0: true, clobberFlags: true}, // arg0 | auxint
{name: "ORWconst", argLength: 1, reg: gp11, asm: "ORW", aux: "Int32", resultInArg0: true, clobberFlags: true}, // arg0 | auxint
{name: "ORload", argLength: 3, reg: gpopload, asm: "OR", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 | *arg1. arg2=mem
{name: "ORWload", argLength: 3, reg: gpopload, asm: "ORW", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 | *arg1. arg2=mem
{name: "XOR", argLength: 2, reg: gp21, asm: "XOR", commutative: true, clobberFlags: true}, // arg0 ^ arg1
{name: "XORW", argLength: 2, reg: gp21, asm: "XORW", commutative: true, clobberFlags: true}, // arg0 ^ arg1
{name: "XORconst", argLength: 1, reg: gp11, asm: "XOR", aux: "Int64", resultInArg0: true, clobberFlags: true}, // arg0 ^ auxint
{name: "XORWconst", argLength: 1, reg: gp11, asm: "XORW", aux: "Int32", resultInArg0: true, clobberFlags: true}, // arg0 ^ auxint
{name: "XORload", argLength: 3, reg: gpopload, asm: "XOR", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 ^ *arg1. arg2=mem
{name: "XORWload", argLength: 3, reg: gpopload, asm: "XORW", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, symEffect: "Read"}, // arg0 ^ *arg1. arg2=mem
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
{name: "CMP", argLength: 2, reg: gp2flags, asm: "CMP", typ: "Flags"}, // arg0 compare to arg1
{name: "CMPW", argLength: 2, reg: gp2flags, asm: "CMPW", typ: "Flags"}, // arg0 compare to arg1
{name: "CMPU", argLength: 2, reg: gp2flags, asm: "CMPU", typ: "Flags"}, // arg0 compare to arg1
{name: "CMPWU", argLength: 2, reg: gp2flags, asm: "CMPWU", typ: "Flags"}, // arg0 compare to arg1
{name: "CMPconst", argLength: 1, reg: gp1flags, asm: "CMP", typ: "Flags", aux: "Int64"}, // arg0 compare to auxint
{name: "CMPWconst", argLength: 1, reg: gp1flags, asm: "CMPW", typ: "Flags", aux: "Int32"}, // arg0 compare to auxint
{name: "CMPUconst", argLength: 1, reg: gp1flags, asm: "CMPU", typ: "Flags", aux: "Int64"}, // arg0 compare to auxint
{name: "CMPWUconst", argLength: 1, reg: gp1flags, asm: "CMPWU", typ: "Flags", aux: "Int32"}, // arg0 compare to auxint
{name: "FCMPS", argLength: 2, reg: fp2flags, asm: "CEBR", typ: "Flags"}, // arg0 compare to arg1, f32
{name: "FCMP", argLength: 2, reg: fp2flags, asm: "FCMPU", typ: "Flags"}, // arg0 compare to arg1, f64
cmd/compile: automatically handle commuting ops in rewrite rules Note that this is a redo of an undo of the original buggy CL 38666. We have lots of rewrite rules that vary only in the fact that we have 2 versions for the 2 different orderings of various commuting ops. For example: (ADDL x (MOVLconst [c])) -> (ADDLconst [c] x) (ADDL (MOVLconst [c]) x) -> (ADDLconst [c] x) It can get unwieldly quickly, especially when there is more than one commuting op in a rule. Our existing "fix" for this problem is to have rules that canonicalize the operations first. For example: (Eq64 x (Const64 <t> [c])) && x.Op != OpConst64 -> (Eq64 (Const64 <t> [c]) x) Subsequent rules can then assume if there is a constant arg to Eq64, it will be the first one. This fix kinda works, but it is fragile and only works when we remember to include the required extra rules. The fundamental problem is that the rule matcher doesn't know anything about commuting ops. This CL fixes that fact. We already have information about which ops commute. (The register allocator takes advantage of commutivity.) The rule generator now automatically generates multiple rules for a single source rule when there are commutative ops in the rule. We can now drop all of our almost-duplicate source-level rules and the canonicalization rules. I have some CLs in progress that will be a lot less verbose when the rule generator handles commutivity for me. I had to reorganize the load-combining rules a bit. The 8-way OR rules generated 128 different reorderings, which was causing the generator to put too much code in the rewrite*.go files (the big ones were going from 25K lines to 132K lines). Instead I reorganized the rules to combine pairs of loads at a time. The generated rule files are now actually a bit (5%) smaller. Make.bash times are ~unchanged. Compiler benchmarks are not observably different. Probably because we don't spend much compiler time in rule matching anyway. I've also done a pass over all of our ops adding commutative markings for ops which hadn't had them previously. Fixes #18292 Change-Id: Ic1c0e43fbf579539f459971625f69690c9ab8805 Reviewed-on: https://go-review.googlesource.com/38801 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2017-03-30 03:30:22 +00:00
{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: "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
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile: automatically handle commuting ops in rewrite rules Note that this is a redo of an undo of the original buggy CL 38666. We have lots of rewrite rules that vary only in the fact that we have 2 versions for the 2 different orderings of various commuting ops. For example: (ADDL x (MOVLconst [c])) -> (ADDLconst [c] x) (ADDL (MOVLconst [c]) x) -> (ADDLconst [c] x) It can get unwieldly quickly, especially when there is more than one commuting op in a rule. Our existing "fix" for this problem is to have rules that canonicalize the operations first. For example: (Eq64 x (Const64 <t> [c])) && x.Op != OpConst64 -> (Eq64 (Const64 <t> [c]) x) Subsequent rules can then assume if there is a constant arg to Eq64, it will be the first one. This fix kinda works, but it is fragile and only works when we remember to include the required extra rules. The fundamental problem is that the rule matcher doesn't know anything about commuting ops. This CL fixes that fact. We already have information about which ops commute. (The register allocator takes advantage of commutivity.) The rule generator now automatically generates multiple rules for a single source rule when there are commutative ops in the rule. We can now drop all of our almost-duplicate source-level rules and the canonicalization rules. I have some CLs in progress that will be a lot less verbose when the rule generator handles commutivity for me. I had to reorganize the load-combining rules a bit. The 8-way OR rules generated 128 different reorderings, which was causing the generator to put too much code in the rewrite*.go files (the big ones were going from 25K lines to 132K lines). Instead I reorganized the rules to combine pairs of loads at a time. The generated rule files are now actually a bit (5%) smaller. Make.bash times are ~unchanged. Compiler benchmarks are not observably different. Probably because we don't spend much compiler time in rule matching anyway. I've also done a pass over all of our ops adding commutative markings for ops which hadn't had them previously. Fixes #18292 Change-Id: Ic1c0e43fbf579539f459971625f69690c9ab8805 Reviewed-on: https://go-review.googlesource.com/38801 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2017-03-30 03:30:22 +00:00
{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 arg0 >> arg1, shift amount is mod 32
{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 arg0 >> auxint, shift amount 0-31
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// Arithmetic shifts clobber flags.
cmd/compile: automatically handle commuting ops in rewrite rules Note that this is a redo of an undo of the original buggy CL 38666. We have lots of rewrite rules that vary only in the fact that we have 2 versions for the 2 different orderings of various commuting ops. For example: (ADDL x (MOVLconst [c])) -> (ADDLconst [c] x) (ADDL (MOVLconst [c]) x) -> (ADDLconst [c] x) It can get unwieldly quickly, especially when there is more than one commuting op in a rule. Our existing "fix" for this problem is to have rules that canonicalize the operations first. For example: (Eq64 x (Const64 <t> [c])) && x.Op != OpConst64 -> (Eq64 (Const64 <t> [c]) x) Subsequent rules can then assume if there is a constant arg to Eq64, it will be the first one. This fix kinda works, but it is fragile and only works when we remember to include the required extra rules. The fundamental problem is that the rule matcher doesn't know anything about commuting ops. This CL fixes that fact. We already have information about which ops commute. (The register allocator takes advantage of commutivity.) The rule generator now automatically generates multiple rules for a single source rule when there are commutative ops in the rule. We can now drop all of our almost-duplicate source-level rules and the canonicalization rules. I have some CLs in progress that will be a lot less verbose when the rule generator handles commutivity for me. I had to reorganize the load-combining rules a bit. The 8-way OR rules generated 128 different reorderings, which was causing the generator to put too much code in the rewrite*.go files (the big ones were going from 25K lines to 132K lines). Instead I reorganized the rules to combine pairs of loads at a time. The generated rule files are now actually a bit (5%) smaller. Make.bash times are ~unchanged. Compiler benchmarks are not observably different. Probably because we don't spend much compiler time in rule matching anyway. I've also done a pass over all of our ops adding commutative markings for ops which hadn't had them previously. Fixes #18292 Change-Id: Ic1c0e43fbf579539f459971625f69690c9ab8805 Reviewed-on: https://go-review.googlesource.com/38801 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2017-03-30 03:30:22 +00:00
{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 arg0 >> arg1, shift amount is mod 32
{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 arg0 >> auxint, shift amount 0-31
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile: automatically handle commuting ops in rewrite rules Note that this is a redo of an undo of the original buggy CL 38666. We have lots of rewrite rules that vary only in the fact that we have 2 versions for the 2 different orderings of various commuting ops. For example: (ADDL x (MOVLconst [c])) -> (ADDLconst [c] x) (ADDL (MOVLconst [c]) x) -> (ADDLconst [c] x) It can get unwieldly quickly, especially when there is more than one commuting op in a rule. Our existing "fix" for this problem is to have rules that canonicalize the operations first. For example: (Eq64 x (Const64 <t> [c])) && x.Op != OpConst64 -> (Eq64 (Const64 <t> [c]) x) Subsequent rules can then assume if there is a constant arg to Eq64, it will be the first one. This fix kinda works, but it is fragile and only works when we remember to include the required extra rules. The fundamental problem is that the rule matcher doesn't know anything about commuting ops. This CL fixes that fact. We already have information about which ops commute. (The register allocator takes advantage of commutivity.) The rule generator now automatically generates multiple rules for a single source rule when there are commutative ops in the rule. We can now drop all of our almost-duplicate source-level rules and the canonicalization rules. I have some CLs in progress that will be a lot less verbose when the rule generator handles commutivity for me. I had to reorganize the load-combining rules a bit. The 8-way OR rules generated 128 different reorderings, which was causing the generator to put too much code in the rewrite*.go files (the big ones were going from 25K lines to 132K lines). Instead I reorganized the rules to combine pairs of loads at a time. The generated rule files are now actually a bit (5%) smaller. Make.bash times are ~unchanged. Compiler benchmarks are not observably different. Probably because we don't spend much compiler time in rule matching anyway. I've also done a pass over all of our ops adding commutative markings for ops which hadn't had them previously. Fixes #18292 Change-Id: Ic1c0e43fbf579539f459971625f69690c9ab8805 Reviewed-on: https://go-review.googlesource.com/38801 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2017-03-30 03:30:22 +00:00
{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
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// unary ops
{name: "NEG", argLength: 1, reg: gp11, asm: "NEG", clobberFlags: true}, // -arg0
{name: "NEGW", argLength: 1, reg: gp11, asm: "NEGW", clobberFlags: true}, // -arg0
{name: "NOT", argLength: 1, reg: gp11, resultInArg0: true, clobberFlags: true}, // ^arg0
{name: "NOTW", argLength: 1, reg: gp11, resultInArg0: true, clobberFlags: true}, // ^arg0
{name: "FSQRT", argLength: 1, reg: fp11, asm: "FSQRT"}, // sqrt(arg0)
{name: "SUBEcarrymask", argLength: 1, reg: flagsgp, asm: "SUBE"}, // (int64)(-1) if carry is set, 0 if carry is clear.
{name: "SUBEWcarrymask", argLength: 1, reg: flagsgp, asm: "SUBE"}, // (int32)(-1) if carry is set, 0 if carry is clear.
// Note: 32-bits subtraction is not implemented in S390X. Temporarily use SUBE (64-bits).
{name: "MOVDEQ", argLength: 3, reg: gp2flags1, resultInArg0: true, asm: "MOVDEQ"}, // extract == condition from arg0
{name: "MOVDNE", argLength: 3, reg: gp2flags1, resultInArg0: true, asm: "MOVDNE"}, // extract != condition from arg0
{name: "MOVDLT", argLength: 3, reg: gp2flags1, resultInArg0: true, asm: "MOVDLT"}, // extract signed < condition from arg0
{name: "MOVDLE", argLength: 3, reg: gp2flags1, resultInArg0: true, asm: "MOVDLE"}, // extract signed <= condition from arg0
{name: "MOVDGT", argLength: 3, reg: gp2flags1, resultInArg0: true, asm: "MOVDGT"}, // extract signed > condition from arg0
{name: "MOVDGE", argLength: 3, reg: gp2flags1, resultInArg0: true, asm: "MOVDGE"}, // extract signed >= condition from arg0
// Different rules for floating point conditions because
// any comparison involving a NaN is always false and thus
// the patterns for inverting conditions cannot be used.
{name: "MOVDGTnoinv", argLength: 3, reg: gp2flags1, resultInArg0: true, asm: "MOVDGT"}, // extract floating > condition from arg0
{name: "MOVDGEnoinv", argLength: 3, reg: gp2flags1, resultInArg0: true, asm: "MOVDGE"}, // extract floating >= condition from arg0
{name: "MOVBreg", argLength: 1, reg: gp11sp, asm: "MOVB", typ: "Int64"}, // sign extend arg0 from int8 to int64
{name: "MOVBZreg", argLength: 1, reg: gp11sp, asm: "MOVBZ", typ: "UInt64"}, // zero extend arg0 from int8 to int64
{name: "MOVHreg", argLength: 1, reg: gp11sp, asm: "MOVH", typ: "Int64"}, // sign extend arg0 from int16 to int64
{name: "MOVHZreg", argLength: 1, reg: gp11sp, asm: "MOVHZ", typ: "UInt64"}, // zero extend arg0 from int16 to int64
{name: "MOVWreg", argLength: 1, reg: gp11sp, asm: "MOVW", typ: "Int64"}, // sign extend arg0 from int32 to int64
{name: "MOVWZreg", argLength: 1, reg: gp11sp, asm: "MOVWZ", typ: "UInt64"}, // zero extend arg0 from int32 to int64
cmd/compile: fix type propagation through s390x SSA rules This CL fixes two issues: 1. Load ops were initially always lowered to unsigned loads, even for signed types. This was fine by itself however LoadReg ops (used to re-load spilled values) were lowered to signed loads for signed types. This meant that spills could invalidate optimizations that assumed the original unsigned load. 2. Types were not always being maintained correctly through rules designed to eliminate unnecessary zero and sign extensions. Fixes #18906. Change-Id: I95785dcadba03f7e3e94524677e7d8d3d3b9b737 Reviewed-on: https://go-review.googlesource.com/36256 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2017-02-03 04:55:34 -05:00
{name: "MOVDreg", argLength: 1, reg: gp11sp, asm: "MOVD"}, // move from arg0
{name: "MOVDnop", argLength: 1, reg: gp11, resultInArg0: true}, // nop, return arg0 in same register
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
{name: "MOVDconst", reg: gp01, asm: "MOVD", typ: "UInt64", aux: "Int64", rematerializeable: true}, // auxint
{name: "CFDBRA", argLength: 1, reg: fpgp, asm: "CFDBRA"}, // convert float64 to int32
{name: "CGDBRA", argLength: 1, reg: fpgp, asm: "CGDBRA"}, // convert float64 to int64
{name: "CFEBRA", argLength: 1, reg: fpgp, asm: "CFEBRA"}, // convert float32 to int32
{name: "CGEBRA", argLength: 1, reg: fpgp, asm: "CGEBRA"}, // convert float32 to int64
{name: "CEFBRA", argLength: 1, reg: gpfp, asm: "CEFBRA"}, // convert int32 to float32
{name: "CDFBRA", argLength: 1, reg: gpfp, asm: "CDFBRA"}, // convert int32 to float64
{name: "CEGBRA", argLength: 1, reg: gpfp, asm: "CEGBRA"}, // convert int64 to float32
{name: "CDGBRA", argLength: 1, reg: gpfp, asm: "CDGBRA"}, // convert int64 to float64
{name: "LEDBR", argLength: 1, reg: fp11, asm: "LEDBR"}, // convert float64 to float32
{name: "LDEBR", argLength: 1, reg: fp11, asm: "LDEBR"}, // convert float32 to float64
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "MOVDaddr", argLength: 1, reg: addr, aux: "SymOff", rematerializeable: true, clobberFlags: true, symEffect: "Read"}, // arg0 + auxint + offset encoded in aux
{name: "MOVDaddridx", argLength: 2, reg: addridx, aux: "SymOff", clobberFlags: true, symEffect: "Read"}, // arg0 + arg1 + auxint + aux
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// auxint+aux == add auxint and the offset of the symbol in aux (if any) to the effective address
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "MOVBZload", argLength: 2, reg: gpload, asm: "MOVBZ", aux: "SymOff", typ: "UInt8", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // load byte from arg0+auxint+aux. arg1=mem. Zero extend.
{name: "MOVBload", argLength: 2, reg: gpload, asm: "MOVB", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // ditto, sign extend to int64
{name: "MOVHZload", argLength: 2, reg: gpload, asm: "MOVHZ", aux: "SymOff", typ: "UInt16", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // load 2 bytes from arg0+auxint+aux. arg1=mem. Zero extend.
{name: "MOVHload", argLength: 2, reg: gpload, asm: "MOVH", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // ditto, sign extend to int64
{name: "MOVWZload", argLength: 2, reg: gpload, asm: "MOVWZ", aux: "SymOff", typ: "UInt32", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // load 4 bytes from arg0+auxint+aux. arg1=mem. Zero extend.
{name: "MOVWload", argLength: 2, reg: gpload, asm: "MOVW", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // ditto, sign extend to int64
{name: "MOVDload", argLength: 2, reg: gpload, asm: "MOVD", aux: "SymOff", typ: "UInt64", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // load 8 bytes from arg0+auxint+aux. arg1=mem
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile: intrinsify Ctz{32,64} and Bswap{32,64} on s390x Also adds the 'find leftmost one' instruction (FLOGR) and replaces the WORD-encoded use of FLOGR in math/big with it. Change-Id: I18e7cd19e75b8501a6ae8bd925471f7e37ded206 Reviewed-on: https://go-review.googlesource.com/29372 Reviewed-by: Cherry Zhang <cherryyz@google.com> Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-09-16 21:42:18 -04:00
{name: "MOVWBR", argLength: 1, reg: gp11, asm: "MOVWBR"}, // arg0 swap bytes
{name: "MOVDBR", argLength: 1, reg: gp11, asm: "MOVDBR"}, // arg0 swap bytes
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "MOVHBRload", argLength: 2, reg: gpload, asm: "MOVHBR", aux: "SymOff", typ: "UInt16", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // load 2 bytes from arg0+auxint+aux. arg1=mem. Reverse bytes.
{name: "MOVWBRload", argLength: 2, reg: gpload, asm: "MOVWBR", aux: "SymOff", typ: "UInt32", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // load 4 bytes from arg0+auxint+aux. arg1=mem. Reverse bytes.
{name: "MOVDBRload", argLength: 2, reg: gpload, asm: "MOVDBR", aux: "SymOff", typ: "UInt64", clobberFlags: true, faultOnNilArg0: true, symEffect: "Read"}, // load 8 bytes from arg0+auxint+aux. arg1=mem. Reverse bytes.
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "MOVBstore", argLength: 3, reg: gpstore, asm: "MOVB", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store byte in arg1 to arg0+auxint+aux. arg2=mem
{name: "MOVHstore", argLength: 3, reg: gpstore, asm: "MOVH", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store 2 bytes in arg1 to arg0+auxint+aux. arg2=mem
{name: "MOVWstore", argLength: 3, reg: gpstore, asm: "MOVW", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store 4 bytes in arg1 to arg0+auxint+aux. arg2=mem
{name: "MOVDstore", argLength: 3, reg: gpstore, asm: "MOVD", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store 8 bytes in arg1 to arg0+auxint+aux. arg2=mem
{name: "MOVHBRstore", argLength: 3, reg: gpstorebr, asm: "MOVHBR", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store 2 bytes in arg1 to arg0+auxint+aux. arg2=mem. Reverse bytes.
{name: "MOVWBRstore", argLength: 3, reg: gpstorebr, asm: "MOVWBR", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store 4 bytes in arg1 to arg0+auxint+aux. arg2=mem. Reverse bytes.
{name: "MOVDBRstore", argLength: 3, reg: gpstorebr, asm: "MOVDBR", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store 8 bytes in arg1 to arg0+auxint+aux. arg2=mem. Reverse bytes.
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "MVC", argLength: 3, reg: gpmvc, asm: "MVC", aux: "SymValAndOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, faultOnNilArg1: true, symEffect: "None"}, // arg0=destptr, arg1=srcptr, arg2=mem, auxint=size,off
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// indexed loads/stores
// TODO(mundaym): add sign-extended indexed loads
cmd/compile: automatically handle commuting ops in rewrite rules Note that this is a redo of an undo of the original buggy CL 38666. We have lots of rewrite rules that vary only in the fact that we have 2 versions for the 2 different orderings of various commuting ops. For example: (ADDL x (MOVLconst [c])) -> (ADDLconst [c] x) (ADDL (MOVLconst [c]) x) -> (ADDLconst [c] x) It can get unwieldly quickly, especially when there is more than one commuting op in a rule. Our existing "fix" for this problem is to have rules that canonicalize the operations first. For example: (Eq64 x (Const64 <t> [c])) && x.Op != OpConst64 -> (Eq64 (Const64 <t> [c]) x) Subsequent rules can then assume if there is a constant arg to Eq64, it will be the first one. This fix kinda works, but it is fragile and only works when we remember to include the required extra rules. The fundamental problem is that the rule matcher doesn't know anything about commuting ops. This CL fixes that fact. We already have information about which ops commute. (The register allocator takes advantage of commutivity.) The rule generator now automatically generates multiple rules for a single source rule when there are commutative ops in the rule. We can now drop all of our almost-duplicate source-level rules and the canonicalization rules. I have some CLs in progress that will be a lot less verbose when the rule generator handles commutivity for me. I had to reorganize the load-combining rules a bit. The 8-way OR rules generated 128 different reorderings, which was causing the generator to put too much code in the rewrite*.go files (the big ones were going from 25K lines to 132K lines). Instead I reorganized the rules to combine pairs of loads at a time. The generated rule files are now actually a bit (5%) smaller. Make.bash times are ~unchanged. Compiler benchmarks are not observably different. Probably because we don't spend much compiler time in rule matching anyway. I've also done a pass over all of our ops adding commutative markings for ops which hadn't had them previously. Fixes #18292 Change-Id: Ic1c0e43fbf579539f459971625f69690c9ab8805 Reviewed-on: https://go-review.googlesource.com/38801 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2017-03-30 03:30:22 +00:00
{name: "MOVBZloadidx", argLength: 3, reg: gploadidx, commutative: true, asm: "MOVBZ", aux: "SymOff", typ: "UInt8", clobberFlags: true, symEffect: "Read"}, // load a byte from arg0+arg1+auxint+aux. arg2=mem
{name: "MOVHZloadidx", argLength: 3, reg: gploadidx, commutative: true, asm: "MOVHZ", aux: "SymOff", typ: "UInt16", clobberFlags: true, symEffect: "Read"}, // load 2 bytes from arg0+arg1+auxint+aux. arg2=mem
{name: "MOVWZloadidx", argLength: 3, reg: gploadidx, commutative: true, asm: "MOVWZ", aux: "SymOff", typ: "UInt32", clobberFlags: true, symEffect: "Read"}, // load 4 bytes from arg0+arg1+auxint+aux. arg2=mem
{name: "MOVDloadidx", argLength: 3, reg: gploadidx, commutative: true, asm: "MOVD", aux: "SymOff", typ: "UInt64", clobberFlags: true, symEffect: "Read"}, // load 8 bytes from arg0+arg1+auxint+aux. arg2=mem
{name: "MOVHBRloadidx", argLength: 3, reg: gploadidx, commutative: true, asm: "MOVHBR", aux: "SymOff", typ: "Int16", clobberFlags: true, symEffect: "Read"}, // load 2 bytes from arg0+arg1+auxint+aux. arg2=mem. Reverse bytes.
{name: "MOVWBRloadidx", argLength: 3, reg: gploadidx, commutative: true, asm: "MOVWBR", aux: "SymOff", typ: "Int32", clobberFlags: true, symEffect: "Read"}, // load 4 bytes from arg0+arg1+auxint+aux. arg2=mem. Reverse bytes.
{name: "MOVDBRloadidx", argLength: 3, reg: gploadidx, commutative: true, asm: "MOVDBR", aux: "SymOff", typ: "Int64", clobberFlags: true, symEffect: "Read"}, // load 8 bytes from arg0+arg1+auxint+aux. arg2=mem. Reverse bytes.
{name: "MOVBstoreidx", argLength: 4, reg: gpstoreidx, commutative: true, asm: "MOVB", aux: "SymOff", clobberFlags: true, symEffect: "Write"}, // store byte in arg2 to arg0+arg1+auxint+aux. arg3=mem
{name: "MOVHstoreidx", argLength: 4, reg: gpstoreidx, commutative: true, asm: "MOVH", aux: "SymOff", clobberFlags: true, symEffect: "Write"}, // store 2 bytes in arg2 to arg0+arg1+auxint+aux. arg3=mem
{name: "MOVWstoreidx", argLength: 4, reg: gpstoreidx, commutative: true, asm: "MOVW", aux: "SymOff", clobberFlags: true, symEffect: "Write"}, // store 4 bytes in arg2 to arg0+arg1+auxint+aux. arg3=mem
{name: "MOVDstoreidx", argLength: 4, reg: gpstoreidx, commutative: true, asm: "MOVD", aux: "SymOff", clobberFlags: true, symEffect: "Write"}, // store 8 bytes in arg2 to arg0+arg1+auxint+aux. arg3=mem
{name: "MOVHBRstoreidx", argLength: 4, reg: gpstoreidx, commutative: true, asm: "MOVHBR", aux: "SymOff", clobberFlags: true, symEffect: "Write"}, // store 2 bytes in arg2 to arg0+arg1+auxint+aux. arg3=mem. Reverse bytes.
{name: "MOVWBRstoreidx", argLength: 4, reg: gpstoreidx, commutative: true, asm: "MOVWBR", aux: "SymOff", clobberFlags: true, symEffect: "Write"}, // store 4 bytes in arg2 to arg0+arg1+auxint+aux. arg3=mem. Reverse bytes.
{name: "MOVDBRstoreidx", argLength: 4, reg: gpstoreidx, commutative: true, asm: "MOVDBR", aux: "SymOff", clobberFlags: true, symEffect: "Write"}, // store 8 bytes in arg2 to arg0+arg1+auxint+aux. arg3=mem. Reverse bytes.
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// For storeconst ops, the AuxInt field encodes both
// the value to store and an address offset of the store.
// Cast AuxInt to a ValAndOff to extract Val and Off fields.
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "MOVBstoreconst", argLength: 2, reg: gpstoreconst, asm: "MOVB", aux: "SymValAndOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store low byte of ValAndOff(AuxInt).Val() to arg0+ValAndOff(AuxInt).Off()+aux. arg1=mem
{name: "MOVHstoreconst", argLength: 2, reg: gpstoreconst, asm: "MOVH", aux: "SymValAndOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store low 2 bytes of ...
{name: "MOVWstoreconst", argLength: 2, reg: gpstoreconst, asm: "MOVW", aux: "SymValAndOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store low 4 bytes of ...
{name: "MOVDstoreconst", argLength: 2, reg: gpstoreconst, asm: "MOVD", aux: "SymValAndOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"}, // store 8 bytes of ...
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "CLEAR", argLength: 2, reg: regInfo{inputs: []regMask{ptr, 0}}, asm: "CLEAR", aux: "SymValAndOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, symEffect: "Write"},
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
cmd/compile: experiment which clobbers all dead pointer fields The experiment "clobberdead" clobbers all pointer fields that the compiler thinks are dead, just before and after every safepoint. Useful for debugging the generation of live pointer bitmaps. Helped find the following issues: Update #15936 Update #16026 Update #16095 Update #18860 Change-Id: Id1d12f86845e3d93bae903d968b1eac61fc461f9 Reviewed-on: https://go-review.googlesource.com/23924 Run-TryBot: Keith Randall <khr@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2016-06-08 22:02:08 -07:00
{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
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
{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
// (InvertFlags (CMP a b)) == (CMP b a)
// InvertFlags is a pseudo-op which can't appear in assembly output.
{name: "InvertFlags", argLength: 1}, // reverse direction of arg0
// Pseudo-ops
{name: "LoweredGetG", argLength: 1, reg: gp01}, // arg0=mem
// Scheduler ensures LoweredGetClosurePtr occurs only in entry block,
// and sorts it to the very beginning of the block to prevent other
// use of R12 (the closure pointer)
{name: "LoweredGetClosurePtr", reg: regInfo{outputs: []regMask{buildReg("R12")}}},
// arg0=ptr,arg1=mem, returns void. Faults if ptr is nil.
cmd/compile: remove duplicate nilchecks Mark nil check operations as faulting if their arg is zero. This lets the late nilcheck pass remove duplicates. Fixes #17242. Change-Id: I4c9938d8a5a1e43edd85b4a66f0b34004860bcd9 Reviewed-on: https://go-review.googlesource.com/29952 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2016-09-27 14:39:27 -07:00
{name: "LoweredNilCheck", argLength: 2, reg: regInfo{inputs: []regMask{ptrsp}}, clobberFlags: true, nilCheck: true, faultOnNilArg0: true},
cmd/compile: emit fused multiply-{add,subtract} instructions on s390x Explcitly block fused multiply-add pattern matching when a cast is used after the multiplication, for example: - (a * b) + c // can emit fused multiply-add - float64(a * b) + c // cannot emit fused multiply-add float{32,64} and complex{64,128} casts of matching types are now kept as OCONV operations rather than being replaced with OCONVNOP operations because they now imply a rounding operation (and therefore aren't a no-op anymore). Operations (for example, multiplication) on complex types may utilize fused multiply-add and -subtract instructions internally. There is no way to disable this behavior at the moment. Improves the performance of the floating point implementation of poly1305: name old speed new speed delta 64 246MB/s ± 0% 275MB/s ± 0% +11.48% (p=0.000 n=10+8) 1K 312MB/s ± 0% 357MB/s ± 0% +14.41% (p=0.000 n=10+10) 64Unaligned 246MB/s ± 0% 274MB/s ± 0% +11.43% (p=0.000 n=10+10) 1KUnaligned 312MB/s ± 0% 357MB/s ± 0% +14.39% (p=0.000 n=10+8) Updates #17895. Change-Id: Ia771d275bb9150d1a598f8cc773444663de5ce16 Reviewed-on: https://go-review.googlesource.com/36963 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-02-12 22:12:12 -05:00
// Round ops to block fused-multiply-add extraction.
{name: "LoweredRound32F", argLength: 1, reg: fp11, resultInArg0: true},
{name: "LoweredRound64F", argLength: 1, reg: fp11, resultInArg0: true},
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// MOVDconvert converts between pointers and integers.
// We have a special op for this so as to not confuse GC
// (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
{name: "MOVDconvert", argLength: 2, reg: gp11sp, asm: "MOVD"},
// Constant flag values. For any comparison, there are 5 possible
// outcomes: the three from the signed total order (<,==,>) and the
// three from the unsigned total order. The == cases overlap.
// Note: there's a sixth "unordered" outcome for floating-point
// comparisons, but we don't use such a beast yet.
// These ops are for temporary use by rewrite rules. They
// cannot appear in the generated assembly.
{name: "FlagEQ"}, // equal
{name: "FlagLT"}, // <
{name: "FlagGT"}, // >
cmd/compile: intrinsify atomic operations on s390x Implements the following intrinsics on s390x: - AtomicAdd{32,64} - AtomicCompareAndSwap{32,64} - AtomicExchange{32,64} - AtomicLoad{32,64,Ptr} - AtomicStore{32,64,PtrNoWB} I haven't added rules for And8 or Or8 yet. Change-Id: I647af023a8e513718e90e98a60191e7af6167314 Reviewed-on: https://go-review.googlesource.com/31614 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-10-19 16:41:01 -04:00
// Atomic loads. These are just normal loads but return <value,memory> tuples
// so they can be properly ordered with other loads.
// load from arg0+auxint+aux. arg1=mem.
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "MOVWZatomicload", argLength: 2, reg: gpload, asm: "MOVWZ", aux: "SymOff", faultOnNilArg0: true, symEffect: "Read"},
{name: "MOVDatomicload", argLength: 2, reg: gpload, asm: "MOVD", aux: "SymOff", faultOnNilArg0: true, symEffect: "Read"},
cmd/compile: intrinsify atomic operations on s390x Implements the following intrinsics on s390x: - AtomicAdd{32,64} - AtomicCompareAndSwap{32,64} - AtomicExchange{32,64} - AtomicLoad{32,64,Ptr} - AtomicStore{32,64,PtrNoWB} I haven't added rules for And8 or Or8 yet. Change-Id: I647af023a8e513718e90e98a60191e7af6167314 Reviewed-on: https://go-review.googlesource.com/31614 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-10-19 16:41:01 -04:00
// Atomic stores. These are just normal stores.
// store arg1 to arg0+auxint+aux. arg2=mem.
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "MOVWatomicstore", argLength: 3, reg: gpstore, asm: "MOVW", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "Write"},
{name: "MOVDatomicstore", argLength: 3, reg: gpstore, asm: "MOVD", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "Write"},
cmd/compile: intrinsify atomic operations on s390x Implements the following intrinsics on s390x: - AtomicAdd{32,64} - AtomicCompareAndSwap{32,64} - AtomicExchange{32,64} - AtomicLoad{32,64,Ptr} - AtomicStore{32,64,PtrNoWB} I haven't added rules for And8 or Or8 yet. Change-Id: I647af023a8e513718e90e98a60191e7af6167314 Reviewed-on: https://go-review.googlesource.com/31614 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-10-19 16:41:01 -04:00
// Atomic adds.
// *(arg0+auxint+aux) += arg1. arg2=mem.
// Returns a tuple of <old contents of *(arg0+auxint+aux), memory>.
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "LAA", argLength: 3, reg: gpstorelaa, asm: "LAA", typ: "(UInt32,Mem)", aux: "SymOff", faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"},
{name: "LAAG", argLength: 3, reg: gpstorelaa, asm: "LAAG", typ: "(UInt64,Mem)", aux: "SymOff", faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"},
cmd/compile: intrinsify atomic operations on s390x Implements the following intrinsics on s390x: - AtomicAdd{32,64} - AtomicCompareAndSwap{32,64} - AtomicExchange{32,64} - AtomicLoad{32,64,Ptr} - AtomicStore{32,64,PtrNoWB} I haven't added rules for And8 or Or8 yet. Change-Id: I647af023a8e513718e90e98a60191e7af6167314 Reviewed-on: https://go-review.googlesource.com/31614 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-10-19 16:41:01 -04:00
{name: "AddTupleFirst32", argLength: 2}, // arg0=tuple <x,y>. Returns <x+arg1,y>.
{name: "AddTupleFirst64", argLength: 2}, // arg0=tuple <x,y>. Returns <x+arg1,y>.
// Compare and swap.
// arg0 = pointer, arg1 = old value, arg2 = new value, arg3 = memory.
// if *(arg0+auxint+aux) == arg1 {
// *(arg0+auxint+aux) = arg2
// return (true, memory)
// } else {
// return (false, memory)
// }
// Note that these instructions also return the old value in arg1, but we ignore it.
// TODO: have these return flags instead of bool. The current system generates:
// CS ...
// MOVD $0, ret
// BNE 2(PC)
// MOVD $1, ret
// CMPW ret, $0
// BNE ...
// instead of just
// CS ...
// BEQ ...
// but we can't do that because memory-using ops can't generate flags yet
// (flagalloc wants to move flag-generating instructions around).
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "LoweredAtomicCas32", argLength: 4, reg: cas, asm: "CS", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"},
{name: "LoweredAtomicCas64", argLength: 4, reg: cas, asm: "CSG", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"},
cmd/compile: intrinsify atomic operations on s390x Implements the following intrinsics on s390x: - AtomicAdd{32,64} - AtomicCompareAndSwap{32,64} - AtomicExchange{32,64} - AtomicLoad{32,64,Ptr} - AtomicStore{32,64,PtrNoWB} I haven't added rules for And8 or Or8 yet. Change-Id: I647af023a8e513718e90e98a60191e7af6167314 Reviewed-on: https://go-review.googlesource.com/31614 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-10-19 16:41:01 -04:00
// Lowered atomic swaps, emulated using compare-and-swap.
// store arg1 to arg0+auxint+aux, arg2=mem.
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
{name: "LoweredAtomicExchange32", argLength: 3, reg: exchange, asm: "CS", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"},
{name: "LoweredAtomicExchange64", argLength: 3, reg: exchange, asm: "CSG", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true, symEffect: "RdWr"},
cmd/compile: intrinsify atomic operations on s390x Implements the following intrinsics on s390x: - AtomicAdd{32,64} - AtomicCompareAndSwap{32,64} - AtomicExchange{32,64} - AtomicLoad{32,64,Ptr} - AtomicStore{32,64,PtrNoWB} I haven't added rules for And8 or Or8 yet. Change-Id: I647af023a8e513718e90e98a60191e7af6167314 Reviewed-on: https://go-review.googlesource.com/31614 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2016-10-19 16:41:01 -04:00
cmd/compile: intrinsify Ctz{32,64} and Bswap{32,64} on s390x Also adds the 'find leftmost one' instruction (FLOGR) and replaces the WORD-encoded use of FLOGR in math/big with it. Change-Id: I18e7cd19e75b8501a6ae8bd925471f7e37ded206 Reviewed-on: https://go-review.googlesource.com/29372 Reviewed-by: Cherry Zhang <cherryyz@google.com> Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-09-16 21:42:18 -04:00
// find leftmost one
{
name: "FLOGR",
argLength: 1,
reg: regInfo{inputs: gponly, outputs: []regMask{buildReg("R0")}, clobbers: buildReg("R1")},
asm: "FLOGR",
typ: "UInt64",
clobberFlags: true,
},
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
// store multiple
{
cmd/compile: redo nil checks Get rid of BlockCheck. Josh goaded me into it, and I went down a rabbithole making it happen. NilCheck now panics if the pointer is nil and returns void, as before. BlockCheck is gone, and NilCheck is no longer a Control value for any block. It just exists (and deadcode knows not to throw it away). I rewrote the nilcheckelim pass to handle this case. In particular, there can now be multiple NilCheck ops per block. I moved all of the arch-dependent nil check elimination done as part of ssaGenValue into its own proper pass, so we don't have to duplicate that code for every architecture. Making the arch-dependent nil check its own pass means I needed to add a bunch of flags to the opcode table so I could write the code without arch-dependent ops everywhere. Change-Id: I419f891ac9b0de313033ff09115c374163416a9f Reviewed-on: https://go-review.googlesource.com/29120 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-09-13 17:01:01 -07:00
name: "STMG2",
argLength: 4,
reg: regInfo{inputs: []regMask{ptrsp, buildReg("R1"), buildReg("R2"), 0}},
aux: "SymOff",
typ: "Mem",
asm: "STMG",
faultOnNilArg0: true,
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
symEffect: "Write",
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
},
{
cmd/compile: redo nil checks Get rid of BlockCheck. Josh goaded me into it, and I went down a rabbithole making it happen. NilCheck now panics if the pointer is nil and returns void, as before. BlockCheck is gone, and NilCheck is no longer a Control value for any block. It just exists (and deadcode knows not to throw it away). I rewrote the nilcheckelim pass to handle this case. In particular, there can now be multiple NilCheck ops per block. I moved all of the arch-dependent nil check elimination done as part of ssaGenValue into its own proper pass, so we don't have to duplicate that code for every architecture. Making the arch-dependent nil check its own pass means I needed to add a bunch of flags to the opcode table so I could write the code without arch-dependent ops everywhere. Change-Id: I419f891ac9b0de313033ff09115c374163416a9f Reviewed-on: https://go-review.googlesource.com/29120 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-09-13 17:01:01 -07:00
name: "STMG3",
argLength: 5,
reg: regInfo{inputs: []regMask{ptrsp, buildReg("R1"), buildReg("R2"), buildReg("R3"), 0}},
aux: "SymOff",
typ: "Mem",
asm: "STMG",
faultOnNilArg0: true,
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
symEffect: "Write",
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
},
{
name: "STMG4",
argLength: 6,
reg: regInfo{inputs: []regMask{
ptrsp,
buildReg("R1"),
buildReg("R2"),
buildReg("R3"),
buildReg("R4"),
0,
}},
cmd/compile: redo nil checks Get rid of BlockCheck. Josh goaded me into it, and I went down a rabbithole making it happen. NilCheck now panics if the pointer is nil and returns void, as before. BlockCheck is gone, and NilCheck is no longer a Control value for any block. It just exists (and deadcode knows not to throw it away). I rewrote the nilcheckelim pass to handle this case. In particular, there can now be multiple NilCheck ops per block. I moved all of the arch-dependent nil check elimination done as part of ssaGenValue into its own proper pass, so we don't have to duplicate that code for every architecture. Making the arch-dependent nil check its own pass means I needed to add a bunch of flags to the opcode table so I could write the code without arch-dependent ops everywhere. Change-Id: I419f891ac9b0de313033ff09115c374163416a9f Reviewed-on: https://go-review.googlesource.com/29120 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-09-13 17:01:01 -07:00
aux: "SymOff",
typ: "Mem",
asm: "STMG",
faultOnNilArg0: true,
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
symEffect: "Write",
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
},
{
cmd/compile: redo nil checks Get rid of BlockCheck. Josh goaded me into it, and I went down a rabbithole making it happen. NilCheck now panics if the pointer is nil and returns void, as before. BlockCheck is gone, and NilCheck is no longer a Control value for any block. It just exists (and deadcode knows not to throw it away). I rewrote the nilcheckelim pass to handle this case. In particular, there can now be multiple NilCheck ops per block. I moved all of the arch-dependent nil check elimination done as part of ssaGenValue into its own proper pass, so we don't have to duplicate that code for every architecture. Making the arch-dependent nil check its own pass means I needed to add a bunch of flags to the opcode table so I could write the code without arch-dependent ops everywhere. Change-Id: I419f891ac9b0de313033ff09115c374163416a9f Reviewed-on: https://go-review.googlesource.com/29120 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-09-13 17:01:01 -07:00
name: "STM2",
argLength: 4,
reg: regInfo{inputs: []regMask{ptrsp, buildReg("R1"), buildReg("R2"), 0}},
aux: "SymOff",
typ: "Mem",
asm: "STMY",
faultOnNilArg0: true,
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
symEffect: "Write",
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
},
{
cmd/compile: redo nil checks Get rid of BlockCheck. Josh goaded me into it, and I went down a rabbithole making it happen. NilCheck now panics if the pointer is nil and returns void, as before. BlockCheck is gone, and NilCheck is no longer a Control value for any block. It just exists (and deadcode knows not to throw it away). I rewrote the nilcheckelim pass to handle this case. In particular, there can now be multiple NilCheck ops per block. I moved all of the arch-dependent nil check elimination done as part of ssaGenValue into its own proper pass, so we don't have to duplicate that code for every architecture. Making the arch-dependent nil check its own pass means I needed to add a bunch of flags to the opcode table so I could write the code without arch-dependent ops everywhere. Change-Id: I419f891ac9b0de313033ff09115c374163416a9f Reviewed-on: https://go-review.googlesource.com/29120 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-09-13 17:01:01 -07:00
name: "STM3",
argLength: 5,
reg: regInfo{inputs: []regMask{ptrsp, buildReg("R1"), buildReg("R2"), buildReg("R3"), 0}},
aux: "SymOff",
typ: "Mem",
asm: "STMY",
faultOnNilArg0: true,
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
symEffect: "Write",
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
},
{
name: "STM4",
argLength: 6,
reg: regInfo{inputs: []regMask{
ptrsp,
buildReg("R1"),
buildReg("R2"),
buildReg("R3"),
buildReg("R4"),
0,
}},
cmd/compile: redo nil checks Get rid of BlockCheck. Josh goaded me into it, and I went down a rabbithole making it happen. NilCheck now panics if the pointer is nil and returns void, as before. BlockCheck is gone, and NilCheck is no longer a Control value for any block. It just exists (and deadcode knows not to throw it away). I rewrote the nilcheckelim pass to handle this case. In particular, there can now be multiple NilCheck ops per block. I moved all of the arch-dependent nil check elimination done as part of ssaGenValue into its own proper pass, so we don't have to duplicate that code for every architecture. Making the arch-dependent nil check its own pass means I needed to add a bunch of flags to the opcode table so I could write the code without arch-dependent ops everywhere. Change-Id: I419f891ac9b0de313033ff09115c374163416a9f Reviewed-on: https://go-review.googlesource.com/29120 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-09-13 17:01:01 -07:00
aux: "SymOff",
typ: "Mem",
asm: "STMY",
faultOnNilArg0: true,
cmd/compile/internal/ssa: populate SymEffects for SSA Ops Changes to ${GOARCH}Ops.go files were mechanically produced using github.com/mdempsky/ssa-symops, a one-off tool that inserts "SymEffect: X" elements by pattern matching against the Op names. Change-Id: Ibf3e481ffd588647f2a31662d72114b740ccbfcf Reviewed-on: https://go-review.googlesource.com/38084 Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2017-03-09 14:46:43 -08:00
symEffect: "Write",
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
},
// large move
// auxint = remaining bytes after loop (rem)
// arg0 = address of dst memory (in R1, changed as a side effect)
// arg1 = address of src memory (in R2, changed as a side effect)
// arg2 = pointer to last address to move in loop + 256
// arg3 = mem
// returns mem
//
// mvc: MVC $256, 0(R2), 0(R1)
// MOVD $256(R1), R1
// MOVD $256(R2), R2
// CMP R2, Rarg2
// BNE mvc
// MVC $rem, 0(R2), 0(R1) // if rem > 0
{
name: "LoweredMove",
aux: "Int64",
argLength: 4,
reg: regInfo{
inputs: []regMask{buildReg("R1"), buildReg("R2"), gpsp},
clobbers: buildReg("R1 R2"),
},
cmd/compile: remove nil check for Zero/Move on 386, AMD64, S390X Fixes #18003. Change-Id: Iadcc5c424c64badecfb5fdbd4dbd9197df56182c Reviewed-on: https://go-review.googlesource.com/33421 Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-11-21 11:31:39 -05:00
clobberFlags: true,
typ: "Mem",
faultOnNilArg0: true,
faultOnNilArg1: true,
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
},
// large clear
// auxint = remaining bytes after loop (rem)
// arg0 = address of dst memory (in R1, changed as a side effect)
// arg1 = pointer to last address to zero in loop + 256
// arg2 = mem
// returns mem
//
// clear: CLEAR $256, 0(R1)
// MOVD $256(R1), R1
// CMP R1, Rarg2
// BNE clear
// CLEAR $rem, 0(R1) // if rem > 0
{
name: "LoweredZero",
aux: "Int64",
argLength: 3,
reg: regInfo{
inputs: []regMask{buildReg("R1"), gpsp},
clobbers: buildReg("R1"),
},
cmd/compile: remove nil check for Zero/Move on 386, AMD64, S390X Fixes #18003. Change-Id: Iadcc5c424c64badecfb5fdbd4dbd9197df56182c Reviewed-on: https://go-review.googlesource.com/33421 Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-11-21 11:31:39 -05:00
clobberFlags: true,
typ: "Mem",
faultOnNilArg0: true,
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
},
}
var S390Xblocks = []blockData{
{name: "EQ"},
{name: "NE"},
{name: "LT"},
{name: "LE"},
{name: "GT"},
{name: "GE"},
{name: "GTF"}, // FP comparison
{name: "GEF"}, // FP comparison
}
archs = append(archs, arch{
name: "S390X",
pkg: "cmd/internal/obj/s390x",
genfile: "../../s390x/ssa.go",
ops: S390Xops,
blocks: S390Xblocks,
regnames: regNamesS390X,
gpregmask: gp,
fpregmask: fp,
framepointerreg: -1, // not used
cmd/compile: make link register allocatable in non-leaf functions We save and restore the link register in non-leaf functions because it is clobbered by CALLs. It is therefore available for general purpose use. Only enabled on s390x currently. The RC4 benchmarks in particular benefit from the extra register: name old speed new speed delta RC4_128 243MB/s ± 2% 341MB/s ± 2% +40.46% (p=0.008 n=5+5) RC4_1K 267MB/s ± 0% 359MB/s ± 1% +34.32% (p=0.008 n=5+5) RC4_8K 271MB/s ± 0% 362MB/s ± 0% +33.61% (p=0.008 n=5+5) Change-Id: Id23bff95e771da9425353da2f32668b8e34ba09f Reviewed-on: https://go-review.googlesource.com/30597 Reviewed-by: Cherry Zhang <cherryyz@google.com> Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-10-06 15:06:45 -04:00
linkreg: int8(num["R14"]),
cmd/compile: add SSA backend for s390x and enable by default The new SSA backend modifies the ABI slightly: R0 is now a usable general purpose register. Fixes #16677. Change-Id: I367435ce921e0c7e79e021c80cf8ef5d1d1466cf Reviewed-on: https://go-review.googlesource.com/28978 Run-TryBot: Michael Munday <munday@ca.ibm.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2016-09-12 14:50:10 -04:00
})
}
Reference in a new issue Copy permalink