mirror
/
go
mirror of https://github.com/golang/go.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

cmd/compile: remove noDuffDevice 

noDuffDevice was for Plan 9, but Plan 9 doesn't need it anymore.
It was also being set in s390x, mips, mipsle, and wasm, but
on those systems it had no effect since the SSA rules for those
architectures don't refer to it at all.

Change-Id: Ib85c0832674c714f3ad5091f0a022eb7cd3ebcdf
Reviewed-on: https://go-review.googlesource.com/c/go/+/655878
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Auto-Submit: Russ Cox <rsc@golang.org>
This commit is contained in:
Russ Cox 2025-03-07 11:30:07 -05:00 committed by Gopher Robot
parent c9b07e8871
commit 26040b1dd7
15 changed files with 69 additions and 85 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
src/cmd/compile/internal/ssa/_gen/386.rules
View File

@ -246,7 +246,7 @@
// Medium copying uses a duff device.
(Move [s] dst src mem)
&& s > 8 && s <= 4*128 && s%4 == 0
&& !config.noDuffDevice && logLargeCopy(v, s) =>
&& logLargeCopy(v, s) =>
(DUFFCOPY [10*(128-s/4)] dst src mem)
// 10 and 128 are magic constants. 10 is the number of bytes to encode:
// MOVL (SI), CX
@ -256,7 +256,7 @@
// and 128 is the number of such blocks. See src/runtime/duff_386.s:duffcopy.
// Large copying uses REP MOVSL.
(Move [s] dst src mem) && (s > 4*128 || config.noDuffDevice) && s%4 == 0 && logLargeCopy(v, s) =>
(Move [s] dst src mem) && s > 4*128 && s%4 == 0 && logLargeCopy(v, s) =>
(REPMOVSL dst src (MOVLconst [int32(s/4)]) mem)
// Lowering Zero instructions
@ -299,8 +299,7 @@
// Medium zeroing uses a duff device.
(Zero [s] destptr mem)
&& s > 16 && s <= 4*128 && s%4 == 0
&& !config.noDuffDevice =>
&& s > 16 && s <= 4*128 && s%4 == 0 =>
(DUFFZERO [1*(128-s/4)] destptr (MOVLconst [0]) mem)
// 1 and 128 are magic constants. 1 is the number of bytes to encode STOSL.
// 128 is the number of STOSL instructions in duffzero.
@ -308,7 +307,7 @@
// Large zeroing uses REP STOSQ.
(Zero [s] destptr mem)
&& (s > 4*128 || (config.noDuffDevice && s > 16))
&& s > 4*128
&& s%4 == 0 =>
(REPSTOSL destptr (MOVLconst [int32(s/4)]) (MOVLconst [0]) mem)

9
src/cmd/compile/internal/ssa/_gen/AMD64.rules
View File

@ -327,11 +327,11 @@
// Medium copying uses a duff device.
(Move [s] dst src mem)
&& s > 64 && s <= 16*64 && s%16 == 0
&& !config.noDuffDevice && logLargeCopy(v, s) =>
&& logLargeCopy(v, s) =>
(DUFFCOPY [s] dst src mem)
// Large copying uses REP MOVSQ.
(Move [s] dst src mem) && (s > 16*64 || config.noDuffDevice) && s%8 == 0 && logLargeCopy(v, s) =>
(Move [s] dst src mem) && s > 16*64 && s%8 == 0 && logLargeCopy(v, s) =>
(REPMOVSQ dst src (MOVQconst [s/8]) mem)
// Lowering Zero instructions
@ -397,13 +397,12 @@
// Medium zeroing uses a duff device.
(Zero [s] destptr mem)
&& s > 64 && s <= 1024 && s%16 == 0 && !config.noDuffDevice =>
&& s > 64 && s <= 1024 && s%16 == 0 =>
(DUFFZERO [s] destptr mem)
// Large zeroing uses REP STOSQ.
(Zero [s] destptr mem)
&& (s > 1024 || (config.noDuffDevice && s > 64))
&& s%8 == 0 =>
&& s > 1024 && s%8 == 0 =>
(REPSTOSQ destptr (MOVQconst [s/8]) (MOVQconst [0]) mem)
// Lowering constants

8
src/cmd/compile/internal/ssa/_gen/ARM.rules
View File

@ -297,12 +297,12 @@
// 4 and 128 are magic constants, see runtime/mkduff.go
(Zero [s] {t} ptr mem)
&& s%4 == 0 && s > 4 && s <= 512
&& t.Alignment()%4 == 0 && !config.noDuffDevice =>
&& t.Alignment()%4 == 0 =>
(DUFFZERO [4 * (128 - s/4)] ptr (MOVWconst [0]) mem)
// Large zeroing uses a loop
(Zero [s] {t} ptr mem)
&& (s > 512 || config.noDuffDevice) || t.Alignment()%4 != 0 =>
&& s > 512 || t.Alignment()%4 != 0 =>
(LoweredZero [t.Alignment()]
ptr
(ADDconst <ptr.Type> ptr [int32(s-moveSize(t.Alignment(), config))])
@ -337,12 +337,12 @@
// 8 and 128 are magic constants, see runtime/mkduff.go
(Move [s] {t} dst src mem)
&& s%4 == 0 && s > 4 && s <= 512
&& t.Alignment()%4 == 0 && !config.noDuffDevice && logLargeCopy(v, s) =>
&& t.Alignment()%4 == 0 && logLargeCopy(v, s) =>
(DUFFCOPY [8 * (128 - s/4)] dst src mem)
// Large move uses a loop
(Move [s] {t} dst src mem)
&& ((s > 512 || config.noDuffDevice) || t.Alignment()%4 != 0) && logLargeCopy(v, s) =>
&& (s > 512 || t.Alignment()%4 != 0) && logLargeCopy(v, s) =>
(LoweredMove [t.Alignment()]
dst
src

9
src/cmd/compile/internal/ssa/_gen/ARM64.rules
View File

@ -420,13 +420,12 @@
// medium zeroing uses a duff device
// 4, 16, and 64 are magic constants, see runtime/mkduff.go
(Zero [s] ptr mem)
&& s%16 == 0 && s > 64 && s <= 16*64
&& !config.noDuffDevice =>
&& s%16 == 0 && s > 64 && s <= 16*64 =>
(DUFFZERO [4 * (64 - s/16)] ptr mem)
// large zeroing uses a loop
(Zero [s] ptr mem)
&& s%16 == 0 && (s > 16*64 || config.noDuffDevice) =>
&& s%16 == 0 && s > 16*64 =>
(LoweredZero
ptr
(ADDconst <ptr.Type> [s-16] ptr)
@ -504,7 +503,7 @@
// medium move uses a duff device
(Move [s] dst src mem)
&& s > 64 && s <= 16*64 && s%16 == 0
&& !config.noDuffDevice && logLargeCopy(v, s) =>
&& logLargeCopy(v, s) =>
(DUFFCOPY [8 * (64 - s/16)] dst src mem)
// 8 is the number of bytes to encode:
//
@ -515,7 +514,7 @@
// large move uses a loop
(Move [s] dst src mem)
&& s%16 == 0 && (s > 16*64 || config.noDuffDevice)
&& s%16 == 0 && s > 16*64
&& logLargeCopy(v, s) =>
(LoweredMove
dst

5
src/cmd/compile/internal/ssa/_gen/LOONG64.rules
View File

@ -341,8 +341,7 @@
// medium zeroing uses a duff device
(Zero [s] ptr mem)
&& s%8 == 0 && s > 16 && s <= 8*128
&& !config.noDuffDevice =>
&& s%8 == 0 && s > 16 && s <= 8*128 =>
(DUFFZERO [8 * (128 - s/8)] ptr mem)
// large zeroing uses a loop
@ -406,7 +405,7 @@
// medium move uses a duff device
(Move [s] dst src mem)
&& s%8 == 0 && s > 16 && s <= 8*128
&& !config.noDuffDevice && logLargeCopy(v, s) =>
&& logLargeCopy(v, s) =>
(DUFFCOPY [16 * (128 - s/8)] dst src mem)
// 16 and 128 are magic constants. 16 is the number of bytes to encode:
// MOVV (R20), R30

6
src/cmd/compile/internal/ssa/_gen/MIPS64.rules
View File

@ -306,12 +306,12 @@
// 8, and 128 are magic constants, see runtime/mkduff.go
(Zero [s] {t} ptr mem)
&& s%8 == 0 && s > 24 && s <= 8*128
&& t.Alignment()%8 == 0 && !config.noDuffDevice =>
&& t.Alignment()%8 == 0 =>
(DUFFZERO [8 * (128 - s/8)] ptr mem)
// large or unaligned zeroing uses a loop
(Zero [s] {t} ptr mem)
&& (s > 8*128 || config.noDuffDevice) || t.Alignment()%8 != 0 =>
&& s > 8*128 || t.Alignment()%8 != 0 =>
(LoweredZero [t.Alignment()]
ptr
(ADDVconst <ptr.Type> ptr [s-moveSize(t.Alignment(), config)])
@ -382,7 +382,7 @@
// medium move uses a duff device
(Move [s] {t} dst src mem)
&& s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0
&& !config.noDuffDevice && logLargeCopy(v, s) =>
&& logLargeCopy(v, s) =>
(DUFFCOPY [16 * (128 - s/8)] dst src mem)
// 16 and 128 are magic constants. 16 is the number of bytes to encode:
// MOVV (R1), R23

4
src/cmd/compile/internal/ssa/_gen/RISCV64.rules
View File

@ -356,7 +356,7 @@
// 8 and 128 are magic constants, see runtime/mkduff.go
(Zero [s] {t} ptr mem)
&& s%8 == 0 && s <= 8*128
&& t.Alignment()%8 == 0 && !config.noDuffDevice =>
&& t.Alignment()%8 == 0 =>
(DUFFZERO [8 * (128 - s/8)] ptr mem)
// Generic zeroing uses a loop
@ -445,7 +445,7 @@
// 16 and 128 are magic constants, see runtime/mkduff.go
(Move [s] {t} dst src mem)
&& s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0
&& !config.noDuffDevice && logLargeCopy(v, s) =>
&& logLargeCopy(v, s) =>
(DUFFCOPY [16 * (128 - s/8)] dst src mem)
// Generic move uses a loop

4
src/cmd/compile/internal/ssa/config.go
View File

@ -41,7 +41,6 @@ type Config struct {
hasGReg bool // has hardware g register
ctxt *obj.Link // Generic arch information
optimize bool // Do optimization
noDuffDevice bool // Don't use Duff's device
useAvg bool // Use optimizations that need Avg* operations
useHmul bool // Use optimizations that need Hmul* operations
SoftFloat bool //
@ -295,7 +294,6 @@ func NewConfig(arch string, types Types, ctxt *obj.Link, optimize, softfloat boo
c.FPReg = framepointerRegS390X
c.LinkReg = linkRegS390X
c.hasGReg = true
c.noDuffDevice = true
c.BigEndian = true
c.unalignedOK = true
c.haveBswap64 = true
@ -316,7 +314,6 @@ func NewConfig(arch string, types Types, ctxt *obj.Link, optimize, softfloat boo
c.FPReg = framepointerRegMIPS
c.LinkReg = linkRegMIPS
c.hasGReg = true
c.noDuffDevice = true
case "riscv64":
c.PtrSize = 8
c.RegSize = 8
@ -344,7 +341,6 @@ func NewConfig(arch string, types Types, ctxt *obj.Link, optimize, softfloat boo
c.FPReg = framepointerRegWasm
c.LinkReg = linkRegWasm
c.hasGReg = true
c.noDuffDevice = true
c.useAvg = false
c.useHmul = false
default:

18
src/cmd/compile/internal/ssa/rewrite386.go
View File

@ -8924,7 +8924,6 @@ func rewriteValue386_OpMove(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: (Move [0] _ _ mem)
// result: mem
@ -9113,14 +9112,14 @@ func rewriteValue386_OpMove(v *Value) bool {
return true
}
// match: (Move [s] dst src mem)
// cond: s > 8 && s <= 4*128 && s%4 == 0 && !config.noDuffDevice && logLargeCopy(v, s)
// cond: s > 8 && s <= 4*128 && s%4 == 0 && logLargeCopy(v, s)
// result: (DUFFCOPY [10*(128-s/4)] dst src mem)
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !(s > 8 && s <= 4*128 && s%4 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
if !(s > 8 && s <= 4*128 && s%4 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(Op386DUFFCOPY)
@ -9129,14 +9128,14 @@ func rewriteValue386_OpMove(v *Value) bool {
return true
}
// match: (Move [s] dst src mem)
// cond: (s > 4*128 || config.noDuffDevice) && s%4 == 0 && logLargeCopy(v, s)
// cond: s > 4*128 && s%4 == 0 && logLargeCopy(v, s)
// result: (REPMOVSL dst src (MOVLconst [int32(s/4)]) mem)
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !((s > 4*128 || config.noDuffDevice) && s%4 == 0 && logLargeCopy(v, s)) {
if !(s > 4*128 && s%4 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(Op386REPMOVSL)
@ -10575,7 +10574,6 @@ func rewriteValue386_OpZero(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: (Zero [0] _ mem)
// result: mem
@ -10769,13 +10767,13 @@ func rewriteValue386_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] destptr mem)
// cond: s > 16 && s <= 4*128 && s%4 == 0 && !config.noDuffDevice
// cond: s > 16 && s <= 4*128 && s%4 == 0
// result: (DUFFZERO [1*(128-s/4)] destptr (MOVLconst [0]) mem)
for {
s := auxIntToInt64(v.AuxInt)
destptr := v_0
mem := v_1
if !(s > 16 && s <= 4*128 && s%4 == 0 && !config.noDuffDevice) {
if !(s > 16 && s <= 4*128 && s%4 == 0) {
break
}
v.reset(Op386DUFFZERO)
@ -10786,13 +10784,13 @@ func rewriteValue386_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] destptr mem)
// cond: (s > 4*128 || (config.noDuffDevice && s > 16)) && s%4 == 0
// cond: s > 4*128 && s%4 == 0
// result: (REPSTOSL destptr (MOVLconst [int32(s/4)]) (MOVLconst [0]) mem)
for {
s := auxIntToInt64(v.AuxInt)
destptr := v_0
mem := v_1
if !((s > 4*128 || (config.noDuffDevice && s > 16)) && s%4 == 0) {
if !(s > 4*128 && s%4 == 0) {
break
}
v.reset(Op386REPSTOSL)

18
src/cmd/compile/internal/ssa/rewriteAMD64.go
View File

@ -27830,7 +27830,6 @@ func rewriteValueAMD64_OpMove(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: (Move [0] _ _ mem)
// result: mem
@ -28230,14 +28229,14 @@ func rewriteValueAMD64_OpMove(v *Value) bool {
return true
}
// match: (Move [s] dst src mem)
// cond: s > 64 && s <= 16*64 && s%16 == 0 && !config.noDuffDevice && logLargeCopy(v, s)
// cond: s > 64 && s <= 16*64 && s%16 == 0 && logLargeCopy(v, s)
// result: (DUFFCOPY [s] dst src mem)
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !(s > 64 && s <= 16*64 && s%16 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
if !(s > 64 && s <= 16*64 && s%16 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(OpAMD64DUFFCOPY)
@ -28246,14 +28245,14 @@ func rewriteValueAMD64_OpMove(v *Value) bool {
return true
}
// match: (Move [s] dst src mem)
// cond: (s > 16*64 || config.noDuffDevice) && s%8 == 0 && logLargeCopy(v, s)
// cond: s > 16*64 && s%8 == 0 && logLargeCopy(v, s)
// result: (REPMOVSQ dst src (MOVQconst [s/8]) mem)
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !((s > 16*64 || config.noDuffDevice) && s%8 == 0 && logLargeCopy(v, s)) {
if !(s > 16*64 && s%8 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(OpAMD64REPMOVSQ)
@ -30443,7 +30442,6 @@ func rewriteValueAMD64_OpZero(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: (Zero [0] _ mem)
// result: mem
@ -30745,13 +30743,13 @@ func rewriteValueAMD64_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] destptr mem)
// cond: s > 64 && s <= 1024 && s%16 == 0 && !config.noDuffDevice
// cond: s > 64 && s <= 1024 && s%16 == 0
// result: (DUFFZERO [s] destptr mem)
for {
s := auxIntToInt64(v.AuxInt)
destptr := v_0
mem := v_1
if !(s > 64 && s <= 1024 && s%16 == 0 && !config.noDuffDevice) {
if !(s > 64 && s <= 1024 && s%16 == 0) {
break
}
v.reset(OpAMD64DUFFZERO)
@ -30760,13 +30758,13 @@ func rewriteValueAMD64_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] destptr mem)
// cond: (s > 1024 || (config.noDuffDevice && s > 64)) && s%8 == 0
// cond: s > 1024 && s%8 == 0
// result: (REPSTOSQ destptr (MOVQconst [s/8]) (MOVQconst [0]) mem)
for {
s := auxIntToInt64(v.AuxInt)
destptr := v_0
mem := v_1
if !((s > 1024 || (config.noDuffDevice && s > 64)) && s%8 == 0) {
if !(s > 1024 && s%8 == 0) {
break
}
v.reset(OpAMD64REPSTOSQ)

16
src/cmd/compile/internal/ssa/rewriteARM.go
View File

@ -14751,7 +14751,7 @@ func rewriteValueARM_OpMove(v *Value) bool {
return true
}
// match: (Move [s] {t} dst src mem)
// cond: s%4 == 0 && s > 4 && s <= 512 && t.Alignment()%4 == 0 && !config.noDuffDevice && logLargeCopy(v, s)
// cond: s%4 == 0 && s > 4 && s <= 512 && t.Alignment()%4 == 0 && logLargeCopy(v, s)
// result: (DUFFCOPY [8 * (128 - s/4)] dst src mem)
for {
s := auxIntToInt64(v.AuxInt)
@ -14759,7 +14759,7 @@ func rewriteValueARM_OpMove(v *Value) bool {
dst := v_0
src := v_1
mem := v_2
if !(s%4 == 0 && s > 4 && s <= 512 && t.Alignment()%4 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
if !(s%4 == 0 && s > 4 && s <= 512 && t.Alignment()%4 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(OpARMDUFFCOPY)
@ -14768,7 +14768,7 @@ func rewriteValueARM_OpMove(v *Value) bool {
return true
}
// match: (Move [s] {t} dst src mem)
// cond: ((s > 512 || config.noDuffDevice) || t.Alignment()%4 != 0) && logLargeCopy(v, s)
// cond: (s > 512 || t.Alignment()%4 != 0) && logLargeCopy(v, s)
// result: (LoweredMove [t.Alignment()] dst src (ADDconst <src.Type> src [int32(s-moveSize(t.Alignment(), config))]) mem)
for {
s := auxIntToInt64(v.AuxInt)
@ -14776,7 +14776,7 @@ func rewriteValueARM_OpMove(v *Value) bool {
dst := v_0
src := v_1
mem := v_2
if !(((s > 512 || config.noDuffDevice) || t.Alignment()%4 != 0) && logLargeCopy(v, s)) {
if !((s > 512 || t.Alignment()%4 != 0) && logLargeCopy(v, s)) {
break
}
v.reset(OpARMLoweredMove)
@ -16175,14 +16175,14 @@ func rewriteValueARM_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] {t} ptr mem)
// cond: s%4 == 0 && s > 4 && s <= 512 && t.Alignment()%4 == 0 && !config.noDuffDevice
// cond: s%4 == 0 && s > 4 && s <= 512 && t.Alignment()%4 == 0
// result: (DUFFZERO [4 * (128 - s/4)] ptr (MOVWconst [0]) mem)
for {
s := auxIntToInt64(v.AuxInt)
t := auxToType(v.Aux)
ptr := v_0
mem := v_1
if !(s%4 == 0 && s > 4 && s <= 512 && t.Alignment()%4 == 0 && !config.noDuffDevice) {
if !(s%4 == 0 && s > 4 && s <= 512 && t.Alignment()%4 == 0) {
break
}
v.reset(OpARMDUFFZERO)
@ -16193,14 +16193,14 @@ func rewriteValueARM_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] {t} ptr mem)
// cond: (s > 512 || config.noDuffDevice) || t.Alignment()%4 != 0
// cond: s > 512 || t.Alignment()%4 != 0
// result: (LoweredZero [t.Alignment()] ptr (ADDconst <ptr.Type> ptr [int32(s-moveSize(t.Alignment(), config))]) (MOVWconst [0]) mem)
for {
s := auxIntToInt64(v.AuxInt)
t := auxToType(v.Aux)
ptr := v_0
mem := v_1
if !((s > 512 || config.noDuffDevice) || t.Alignment()%4 != 0) {
if !(s > 512 || t.Alignment()%4 != 0) {
break
}
v.reset(OpARMLoweredZero)

18
src/cmd/compile/internal/ssa/rewriteARM64.go
View File

@ -20588,7 +20588,6 @@ func rewriteValueARM64_OpMove(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: (Move [0] _ _ mem)
// result: mem
@ -21078,14 +21077,14 @@ func rewriteValueARM64_OpMove(v *Value) bool {
return true
}
// match: (Move [s] dst src mem)
// cond: s > 64 && s <= 16*64 && s%16 == 0 && !config.noDuffDevice && logLargeCopy(v, s)
// cond: s > 64 && s <= 16*64 && s%16 == 0 && logLargeCopy(v, s)
// result: (DUFFCOPY [8 * (64 - s/16)] dst src mem)
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !(s > 64 && s <= 16*64 && s%16 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
if !(s > 64 && s <= 16*64 && s%16 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(OpARM64DUFFCOPY)
@ -21094,14 +21093,14 @@ func rewriteValueARM64_OpMove(v *Value) bool {
return true
}
// match: (Move [s] dst src mem)
// cond: s%16 == 0 && (s > 16*64 || config.noDuffDevice) && logLargeCopy(v, s)
// cond: s%16 == 0 && s > 16*64 && logLargeCopy(v, s)
// result: (LoweredMove dst src (ADDconst <src.Type> src [s-16]) mem)
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !(s%16 == 0 && (s > 16*64 || config.noDuffDevice) && logLargeCopy(v, s)) {
if !(s%16 == 0 && s > 16*64 && logLargeCopy(v, s)) {
break
}
v.reset(OpARM64LoweredMove)
@ -23446,7 +23445,6 @@ func rewriteValueARM64_OpZero(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: (Zero [0] _ mem)
// result: mem
@ -23822,13 +23820,13 @@ func rewriteValueARM64_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] ptr mem)
// cond: s%16 == 0 && s > 64 && s <= 16*64 && !config.noDuffDevice
// cond: s%16 == 0 && s > 64 && s <= 16*64
// result: (DUFFZERO [4 * (64 - s/16)] ptr mem)
for {
s := auxIntToInt64(v.AuxInt)
ptr := v_0
mem := v_1
if !(s%16 == 0 && s > 64 && s <= 16*64 && !config.noDuffDevice) {
if !(s%16 == 0 && s > 64 && s <= 16*64) {
break
}
v.reset(OpARM64DUFFZERO)
@ -23837,13 +23835,13 @@ func rewriteValueARM64_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] ptr mem)
// cond: s%16 == 0 && (s > 16*64 || config.noDuffDevice)
// cond: s%16 == 0 && s > 16*64
// result: (LoweredZero ptr (ADDconst <ptr.Type> [s-16] ptr) mem)
for {
s := auxIntToInt64(v.AuxInt)
ptr := v_0
mem := v_1
if !(s%16 == 0 && (s > 16*64 || config.noDuffDevice)) {
if !(s%16 == 0 && s > 16*64) {
break
}
v.reset(OpARM64LoweredZero)

10
src/cmd/compile/internal/ssa/rewriteLOONG64.go
View File

@ -7961,7 +7961,6 @@ func rewriteValueLOONG64_OpMove(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: (Move [0] _ _ mem)
// result: mem
@ -8311,14 +8310,14 @@ func rewriteValueLOONG64_OpMove(v *Value) bool {
return true
}
// match: (Move [s] dst src mem)
// cond: s%8 == 0 && s > 16 && s <= 8*128 && !config.noDuffDevice && logLargeCopy(v, s)
// cond: s%8 == 0 && s > 16 && s <= 8*128 && logLargeCopy(v, s)
// result: (DUFFCOPY [16 * (128 - s/8)] dst src mem)
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !(s%8 == 0 && s > 16 && s <= 8*128 && !config.noDuffDevice && logLargeCopy(v, s)) {
if !(s%8 == 0 && s > 16 && s <= 8*128 && logLargeCopy(v, s)) {
break
}
v.reset(OpLOONG64DUFFCOPY)
@ -9873,7 +9872,6 @@ func rewriteValueLOONG64_OpZero(v *Value) bool {
v_1 := v.Args[1]
v_0 := v.Args[0]
b := v.Block
config := b.Func.Config
typ := &b.Func.Config.Types
// match: (Zero [0] _ mem)
// result: mem
@ -10167,13 +10165,13 @@ func rewriteValueLOONG64_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] ptr mem)
// cond: s%8 == 0 && s > 16 && s <= 8*128 && !config.noDuffDevice
// cond: s%8 == 0 && s > 16 && s <= 8*128
// result: (DUFFZERO [8 * (128 - s/8)] ptr mem)
for {
s := auxIntToInt64(v.AuxInt)
ptr := v_0
mem := v_1
if !(s%8 == 0 && s > 16 && s <= 8*128 && !config.noDuffDevice) {
if !(s%8 == 0 && s > 16 && s <= 8*128) {
break
}
v.reset(OpLOONG64DUFFZERO)

12
src/cmd/compile/internal/ssa/rewriteMIPS64.go
View File

@ -6043,7 +6043,7 @@ func rewriteValueMIPS64_OpMove(v *Value) bool {
return true
}
// match: (Move [s] {t} dst src mem)
// cond: s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice && logLargeCopy(v, s)
// cond: s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0 && logLargeCopy(v, s)
// result: (DUFFCOPY [16 * (128 - s/8)] dst src mem)
for {
s := auxIntToInt64(v.AuxInt)
@ -6051,7 +6051,7 @@ func rewriteValueMIPS64_OpMove(v *Value) bool {
dst := v_0
src := v_1
mem := v_2
if !(s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
if !(s%8 == 0 && s >= 24 && s <= 8*128 && t.Alignment()%8 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(OpMIPS64DUFFCOPY)
@ -8200,14 +8200,14 @@ func rewriteValueMIPS64_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] {t} ptr mem)
// cond: s%8 == 0 && s > 24 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice
// cond: s%8 == 0 && s > 24 && s <= 8*128 && t.Alignment()%8 == 0
// result: (DUFFZERO [8 * (128 - s/8)] ptr mem)
for {
s := auxIntToInt64(v.AuxInt)
t := auxToType(v.Aux)
ptr := v_0
mem := v_1
if !(s%8 == 0 && s > 24 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice) {
if !(s%8 == 0 && s > 24 && s <= 8*128 && t.Alignment()%8 == 0) {
break
}
v.reset(OpMIPS64DUFFZERO)
@ -8216,14 +8216,14 @@ func rewriteValueMIPS64_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] {t} ptr mem)
// cond: (s > 8*128 || config.noDuffDevice) || t.Alignment()%8 != 0
// cond: s > 8*128 || t.Alignment()%8 != 0
// result: (LoweredZero [t.Alignment()] ptr (ADDVconst <ptr.Type> ptr [s-moveSize(t.Alignment(), config)]) mem)
for {
s := auxIntToInt64(v.AuxInt)
t := auxToType(v.Aux)
ptr := v_0
mem := v_1
if !((s > 8*128 || config.noDuffDevice) || t.Alignment()%8 != 0) {
if !(s > 8*128 || t.Alignment()%8 != 0) {
break
}
v.reset(OpMIPS64LoweredZero)

8
src/cmd/compile/internal/ssa/rewriteRISCV64.go
View File

@ -3015,7 +3015,7 @@ func rewriteValueRISCV64_OpMove(v *Value) bool {
return true
}
// match: (Move [s] {t} dst src mem)
// cond: s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice && logLargeCopy(v, s)
// cond: s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 && logLargeCopy(v, s)
// result: (DUFFCOPY [16 * (128 - s/8)] dst src mem)
for {
s := auxIntToInt64(v.AuxInt)
@ -3023,7 +3023,7 @@ func rewriteValueRISCV64_OpMove(v *Value) bool {
dst := v_0
src := v_1
mem := v_2
if !(s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice && logLargeCopy(v, s)) {
if !(s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(OpRISCV64DUFFCOPY)
@ -9053,14 +9053,14 @@ func rewriteValueRISCV64_OpZero(v *Value) bool {
return true
}
// match: (Zero [s] {t} ptr mem)
// cond: s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice
// cond: s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0
// result: (DUFFZERO [8 * (128 - s/8)] ptr mem)
for {
s := auxIntToInt64(v.AuxInt)
t := auxToType(v.Aux)
ptr := v_0
mem := v_1
if !(s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0 && !config.noDuffDevice) {
if !(s%8 == 0 && s <= 8*128 && t.Alignment()%8 == 0) {
break
}
v.reset(OpRISCV64DUFFZERO)