This CL eliminates unnecessary type conversion operations: OpZeroExt16to64 and OpZeroExt8to64.
If the input argrument is a nonzero value, then ORconst operation can also be eliminated.
Benchmarks:
name old time/op new time/op delta
TrailingZeros-8 2.75ns ± 0% 2.75ns ± 0% ~ (all equal)
TrailingZeros8-8 3.49ns ± 1% 2.93ns ± 0% -16.00% (p=0.000 n=10+10)
TrailingZeros16-8 3.49ns ± 1% 2.93ns ± 0% -16.05% (p=0.000 n=9+10)
TrailingZeros32-8 2.67ns ± 1% 2.68ns ± 1% ~ (p=0.468 n=10+10)
TrailingZeros64-8 2.67ns ± 1% 2.65ns ± 0% -0.62% (p=0.022 n=10+9)
code:
func f16(x uint) { z = bits.TrailingZeros16(uint16(x)) }
Before:
"".f16 STEXT size=48 args=0x8 locals=0x0 leaf
0x0000 00000 (test.go:7) TEXT "".f16(SB), LEAF|NOFRAME|ABIInternal, $0-8
0x0000 00000 (test.go:7) FUNCDATA ZR, gclocals·
33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (test.go:7) FUNCDATA $1, gclocals·
33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (test.go:7) FUNCDATA $3, gclocals·
33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (test.go:7) PCDATA $2, ZR
0x0000 00000 (test.go:7) PCDATA ZR, ZR
0x0000 00000 (test.go:7) MOVD "".x(FP), R0
0x0004 00004 (test.go:7) MOVHU R0, R0
0x0008 00008 (test.go:7) ORR $65536, R0, R0
0x000c 00012 (test.go:7) RBIT R0, R0
0x0010 00016 (test.go:7) CLZ R0, R0
0x0014 00020 (test.go:7) MOVD R0, "".z(SB)
0x0020 00032 (test.go:7) RET (R30)
This line of code is unnecessary:
0x0004 00004 (test.go:7) MOVHU R0, R0
After:
"".f16 STEXT size=32 args=0x8 locals=0x0 leaf
0x0000 00000 (test.go:7) TEXT "".f16(SB), LEAF|NOFRAME|ABIInternal, $0-8
0x0000 00000 (test.go:7) FUNCDATA ZR, gclocals·
33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (test.go:7) FUNCDATA $1, gclocals·
33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (test.go:7) FUNCDATA $3, gclocals·
33cdeccccebe80329f1fdbee7f5874cb(SB)
0x0000 00000 (test.go:7) PCDATA $2, ZR
0x0000 00000 (test.go:7) PCDATA ZR, ZR
0x0000 00000 (test.go:7) MOVD "".x(FP), R0
0x0004 00004 (test.go:7) ORR $65536, R0, R0
0x0008 00008 (test.go:7) RBITW R0, R0
0x000c 00012 (test.go:7) CLZW R0, R0
0x0010 00016 (test.go:7) MOVD R0, "".z(SB)
0x001c 00028 (test.go:7) RET (R30)
The situation of TrailingZeros8 is similar to TrailingZeros16.
Change-Id: I473bdca06be8460a0be87abbae6fe640017e4c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/156999
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
return s.newValue1(ssa.OpCtz16, types.Types[TINT], args[0])
},
- sys.AMD64)
+ sys.AMD64, sys.ARM64)
addF("math/bits", "TrailingZeros16",
func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
x := s.newValue1(ssa.OpZeroExt16to64, types.Types[TUINT64], args[0])
y := s.newValue2(ssa.OpOr64, types.Types[TUINT64], x, c)
return s.newValue1(ssa.OpCtz64, types.Types[TINT], y)
},
- sys.ARM64, sys.S390X, sys.PPC64)
+ sys.S390X, sys.PPC64)
addF("math/bits", "TrailingZeros8",
func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
x := s.newValue1(ssa.OpZeroExt8to32, types.Types[TUINT32], args[0])
func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
return s.newValue1(ssa.OpCtz8, types.Types[TINT], args[0])
},
- sys.AMD64)
+ sys.AMD64, sys.ARM64)
addF("math/bits", "TrailingZeros8",
func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
x := s.newValue1(ssa.OpZeroExt8to64, types.Types[TUINT64], args[0])
y := s.newValue2(ssa.OpOr64, types.Types[TUINT64], x, c)
return s.newValue1(ssa.OpCtz64, types.Types[TINT], y)
},
- sys.ARM64, sys.S390X)
+ sys.S390X)
alias("math/bits", "ReverseBytes64", "runtime/internal/sys", "Bswap64", all...)
alias("math/bits", "ReverseBytes32", "runtime/internal/sys", "Bswap32", all...)
// ReverseBytes inlines correctly, no need to intrinsify it.
(Ctz64NonZero x) -> (Ctz64 x)
(Ctz32NonZero x) -> (Ctz32 x)
+(Ctz16NonZero x) -> (Ctz32 x)
+(Ctz8NonZero x) -> (Ctz32 x)
(Ctz64 <t> x) -> (CLZ (RBIT <t> x))
(Ctz32 <t> x) -> (CLZW (RBITW <t> x))
+(Ctz16 <t> x) -> (CLZW <t> (RBITW <typ.UInt32> (ORconst <typ.UInt32> [0x10000] x)))
+(Ctz8 <t> x) -> (CLZW <t> (RBITW <typ.UInt32> (ORconst <typ.UInt32> [0x100] x)))
(PopCount64 <t> x) -> (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> x))))
(PopCount32 <t> x) -> (FMOVDfpgp <t> (VUADDLV <typ.Float64> (VCNT <typ.Float64> (FMOVDgpfp <typ.Float64> (ZeroExt32to64 x)))))
return rewriteValueARM64_OpConstBool_0(v)
case OpConstNil:
return rewriteValueARM64_OpConstNil_0(v)
+ case OpCtz16:
+ return rewriteValueARM64_OpCtz16_0(v)
+ case OpCtz16NonZero:
+ return rewriteValueARM64_OpCtz16NonZero_0(v)
case OpCtz32:
return rewriteValueARM64_OpCtz32_0(v)
case OpCtz32NonZero:
return rewriteValueARM64_OpCtz64_0(v)
case OpCtz64NonZero:
return rewriteValueARM64_OpCtz64NonZero_0(v)
+ case OpCtz8:
+ return rewriteValueARM64_OpCtz8_0(v)
+ case OpCtz8NonZero:
+ return rewriteValueARM64_OpCtz8NonZero_0(v)
case OpCvt32Fto32:
return rewriteValueARM64_OpCvt32Fto32_0(v)
case OpCvt32Fto32U:
return true
}
}
+func rewriteValueARM64_OpCtz16_0(v *Value) bool {
+ b := v.Block
+ _ = b
+ typ := &b.Func.Config.Types
+ _ = typ
+ // match: (Ctz16 <t> x)
+ // cond:
+ // result: (CLZW <t> (RBITW <typ.UInt32> (ORconst <typ.UInt32> [0x10000] x)))
+ for {
+ t := v.Type
+ x := v.Args[0]
+ v.reset(OpARM64CLZW)
+ v.Type = t
+ v0 := b.NewValue0(v.Pos, OpARM64RBITW, typ.UInt32)
+ v1 := b.NewValue0(v.Pos, OpARM64ORconst, typ.UInt32)
+ v1.AuxInt = 0x10000
+ v1.AddArg(x)
+ v0.AddArg(v1)
+ v.AddArg(v0)
+ return true
+ }
+}
+func rewriteValueARM64_OpCtz16NonZero_0(v *Value) bool {
+ // match: (Ctz16NonZero x)
+ // cond:
+ // result: (Ctz32 x)
+ for {
+ x := v.Args[0]
+ v.reset(OpCtz32)
+ v.AddArg(x)
+ return true
+ }
+}
func rewriteValueARM64_OpCtz32_0(v *Value) bool {
b := v.Block
_ = b
return true
}
}
+func rewriteValueARM64_OpCtz8_0(v *Value) bool {
+ b := v.Block
+ _ = b
+ typ := &b.Func.Config.Types
+ _ = typ
+ // match: (Ctz8 <t> x)
+ // cond:
+ // result: (CLZW <t> (RBITW <typ.UInt32> (ORconst <typ.UInt32> [0x100] x)))
+ for {
+ t := v.Type
+ x := v.Args[0]
+ v.reset(OpARM64CLZW)
+ v.Type = t
+ v0 := b.NewValue0(v.Pos, OpARM64RBITW, typ.UInt32)
+ v1 := b.NewValue0(v.Pos, OpARM64ORconst, typ.UInt32)
+ v1.AuxInt = 0x100
+ v1.AddArg(x)
+ v0.AddArg(v1)
+ v.AddArg(v0)
+ return true
+ }
+}
+func rewriteValueARM64_OpCtz8NonZero_0(v *Value) bool {
+ // match: (Ctz8NonZero x)
+ // cond:
+ // result: (Ctz32 x)
+ for {
+ x := v.Args[0]
+ v.reset(OpCtz32)
+ v.AddArg(x)
+ return true
+ }
+}
func rewriteValueARM64_OpCvt32Fto32_0(v *Value) bool {
// match: (Cvt32Fto32 x)
// cond:
func TrailingZeros(n uint) int {
// amd64:"BSFQ","MOVL\t\\$64","CMOVQEQ"
+ // arm64:"RBIT","CLZ"
// s390x:"FLOGR"
// ppc64:"ANDN","POPCNTD"
// ppc64le:"ANDN","POPCNTD"
func TrailingZeros64(n uint64) int {
// amd64:"BSFQ","MOVL\t\\$64","CMOVQEQ"
+ // arm64:"RBIT","CLZ"
// s390x:"FLOGR"
// ppc64:"ANDN","POPCNTD"
// ppc64le:"ANDN","POPCNTD"
func TrailingZeros32(n uint32) int {
// amd64:"BTSQ\\t\\$32","BSFQ"
+ // arm64:"RBITW","CLZW"
// s390x:"FLOGR","MOVWZ"
// ppc64:"ANDN","POPCNTW"
// ppc64le:"ANDN","POPCNTW"
func TrailingZeros16(n uint16) int {
// amd64:"BSFL","BTSL\\t\\$16"
+ // arm64:"ORR\t\\$65536","RBITW","CLZW",-"MOVHU\tR",-"RBIT\t",-"CLZ\t"
// s390x:"FLOGR","OR\t\\$65536"
// ppc64:"POPCNTD","OR\\t\\$65536"
// ppc64le:"POPCNTD","OR\\t\\$65536"
func TrailingZeros8(n uint8) int {
// amd64:"BSFL","BTSL\\t\\$8"
+ // arm64:"ORR\t\\$256","RBITW","CLZW",-"MOVBU\tR",-"RBIT\t",-"CLZ\t"
// s390x:"FLOGR","OR\t\\$256"
return bits.TrailingZeros8(n)
}
i := 0
for n != 0 {
// amd64:"BSFL",-"BTSL"
+ // arm64:"RBITW","CLZW",-"ORR"
i += bits.TrailingZeros16(n)
n &= n - 1
}
i := 0
for n != 0 {
// amd64:"BSFL",-"BTSL"
+ // arm64:"RBITW","CLZW",-"ORR"
i += bits.TrailingZeros8(n)
n &= n - 1
}