cmd/compile: optimize RotateLeft8/16 on loong64

goos: linux
goarch: loong64
pkg: math/bits
cpu: Loongson-3A6000 @ 2500.00MHz
             │  bench.old   │              bench.new               │
             │    sec/op    │    sec/op     vs base                │
RotateLeft8     1.401n ± 0%    1.201n ± 0%  -14.28% (p=0.000 n=20)
RotateLeft16   1.4010n ± 0%   0.8032n ± 0%  -42.67% (p=0.000 n=20)
geomean         1.401n        0.9822n       -29.90%

goos: linux
goarch: loong64
pkg: math/bits
cpu: Loongson-3A5000 @ 2500.00MHz
             │  bench.old  │              bench.new              │
             │   sec/op    │   sec/op     vs base                │
RotateLeft8    1.576n ± 0%   1.310n ± 0%  -16.88% (p=0.000 n=20)
RotateLeft16   1.576n ± 0%   1.166n ± 0%  -26.02% (p=0.000 n=20)
geomean        1.576n        1.236n       -21.58%

Change-Id: I39c18306be0b8fd31b57bd0911714abd1783b50e
Reviewed-on: https://go-review.googlesource.com/c/go/+/604738
Auto-Submit: abner chenc <chenguoqi@loongson.cn>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Reviewed-by: abner chenc <chenguoqi@loongson.cn>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Tim King <taking@google.com>

diff --git a/src/cmd/compile/internal/ssa/_gen/LOONG64.rules b/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
index 0f58cbcc11715ac67af5ba4ac5adda26fc33e680..ab280f8f79d66c051066cd3570601498c8504bea 100644 (file)
--- a/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
+++ b/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
@@ -119,7 +119,9 @@
 
 // rotates
 (RotateLeft8 <t> x (MOVVconst [c])) => (Or8 (Lsh8x64 <t> x (MOVVconst [c&7])) (Rsh8Ux64 <t> x (MOVVconst [-c&7])))
+(RotateLeft8 <t> x y) => (OR <t> (SLLV <t> x (ANDconst <typ.Int64> [7] y)) (SRLV <t> (ZeroExt8to64 x) (ANDconst <typ.Int64> [7] (NEGV <typ.Int64> y))))
 (RotateLeft16 <t> x (MOVVconst [c])) => (Or16 (Lsh16x64 <t> x (MOVVconst [c&15])) (Rsh16Ux64 <t> x (MOVVconst [-c&15])))
+(RotateLeft16 <t> x y) => (ROTR <t> (OR <typ.UInt32> (ZeroExt16to32 x) (SLLVconst <t> (ZeroExt16to32 x) [16])) (NEGV <typ.Int64> y))
 (RotateLeft32 x y) => (ROTR  x (NEGV <y.Type> y))
 (RotateLeft64 x y) => (ROTRV x (NEGV <y.Type> y))
 

diff --git a/src/cmd/compile/internal/ssa/rewrite.go b/src/cmd/compile/internal/ssa/rewrite.go
index b95afd9f2dba2e970f9b40ddb247f2594be616d3..c7d5dc71060b46dc651d918a998940f840b8652e 100644 (file)
--- a/src/cmd/compile/internal/ssa/rewrite.go
+++ b/src/cmd/compile/internal/ssa/rewrite.go
@@ -2255,9 +2255,9 @@ func canRotate(c *Config, bits int64) bool {
                return false
        }
        switch c.arch {
-       case "386", "amd64", "arm64", "riscv64":
+       case "386", "amd64", "arm64", "loong64", "riscv64":
                return true
-       case "arm", "s390x", "ppc64", "ppc64le", "wasm", "loong64":
+       case "arm", "s390x", "ppc64", "ppc64le", "wasm":
                return bits >= 32
        default:
                return false

diff --git a/src/cmd/compile/internal/ssa/rewriteLOONG64.go b/src/cmd/compile/internal/ssa/rewriteLOONG64.go
index f4b1997e18901d8e250252b7a24bdad1ac386b5a..e17c305f4feb5eb7908cf7a3bc31d4eeda3c5cba 100644 (file)
--- a/src/cmd/compile/internal/ssa/rewriteLOONG64.go
+++ b/src/cmd/compile/internal/ssa/rewriteLOONG64.go
@@ -6102,7 +6102,26 @@ func rewriteValueLOONG64_OpRotateLeft16(v *Value) bool {
                v.AddArg2(v0, v2)
                return true
        }
-       return false
+       // match: (RotateLeft16 <t> x y)
+       // result: (ROTR <t> (OR <typ.UInt32> (ZeroExt16to32 x) (SLLVconst <t> (ZeroExt16to32 x) [16])) (NEGV <typ.Int64> y))
+       for {
+               t := v.Type
+               x := v_0
+               y := v_1
+               v.reset(OpLOONG64ROTR)
+               v.Type = t
+               v0 := b.NewValue0(v.Pos, OpLOONG64OR, typ.UInt32)
+               v1 := b.NewValue0(v.Pos, OpZeroExt16to32, typ.UInt32)
+               v1.AddArg(x)
+               v2 := b.NewValue0(v.Pos, OpLOONG64SLLVconst, t)
+               v2.AuxInt = int64ToAuxInt(16)
+               v2.AddArg(v1)
+               v0.AddArg2(v1, v2)
+               v3 := b.NewValue0(v.Pos, OpLOONG64NEGV, typ.Int64)
+               v3.AddArg(y)
+               v.AddArg2(v0, v3)
+               return true
+       }
 }
 func rewriteValueLOONG64_OpRotateLeft32(v *Value) bool {
        v_1 := v.Args[1]
@@ -6162,7 +6181,31 @@ func rewriteValueLOONG64_OpRotateLeft8(v *Value) bool {
                v.AddArg2(v0, v2)
                return true
        }
-       return false
+       // match: (RotateLeft8 <t> x y)
+       // result: (OR <t> (SLLV <t> x (ANDconst <typ.Int64> [7] y)) (SRLV <t> (ZeroExt8to64 x) (ANDconst <typ.Int64> [7] (NEGV <typ.Int64> y))))
+       for {
+               t := v.Type
+               x := v_0
+               y := v_1
+               v.reset(OpLOONG64OR)
+               v.Type = t
+               v0 := b.NewValue0(v.Pos, OpLOONG64SLLV, t)
+               v1 := b.NewValue0(v.Pos, OpLOONG64ANDconst, typ.Int64)
+               v1.AuxInt = int64ToAuxInt(7)
+               v1.AddArg(y)
+               v0.AddArg2(x, v1)
+               v2 := b.NewValue0(v.Pos, OpLOONG64SRLV, t)
+               v3 := b.NewValue0(v.Pos, OpZeroExt8to64, typ.UInt64)
+               v3.AddArg(x)
+               v4 := b.NewValue0(v.Pos, OpLOONG64ANDconst, typ.Int64)
+               v4.AuxInt = int64ToAuxInt(7)
+               v5 := b.NewValue0(v.Pos, OpLOONG64NEGV, typ.Int64)
+               v5.AddArg(y)
+               v4.AddArg(v5)
+               v2.AddArg2(v3, v4)
+               v.AddArg2(v0, v2)
+               return true
+       }
 }
 func rewriteValueLOONG64_OpRsh16Ux16(v *Value) bool {
        v_1 := v.Args[1]

diff --git a/test/codegen/mathbits.go b/test/codegen/mathbits.go
index af2c9e790cea8f38482e29d7be9a8d141bad05cc..4754f29525a55efe5e2b94e6d2f3116dad00d111 100644 (file)
--- a/test/codegen/mathbits.go
+++ b/test/codegen/mathbits.go
@@ -230,6 +230,7 @@ func ReverseBytes16(n uint16) uint16 {
 func RotateLeft64(n uint64) uint64 {
        // amd64:"ROLQ"
        // arm64:"ROR"
+       // loong64:"ROTRV"
        // ppc64x:"ROTL"
        // riscv64:"RORI"
        // s390x:"RISBGZ\t[$]0, [$]63, [$]37, "
@@ -241,6 +242,7 @@ func RotateLeft32(n uint32) uint32 {
        // amd64:"ROLL" 386:"ROLL"
        // arm:`MOVW\tR[0-9]+@>23`
        // arm64:"RORW"
+       // loong64:"ROTR\t"
        // ppc64x:"ROTLW"
        // riscv64:"RORIW"
        // s390x:"RLL"
@@ -251,18 +253,21 @@ func RotateLeft32(n uint32) uint32 {
 func RotateLeft16(n uint16, s int) uint16 {
        // amd64:"ROLW" 386:"ROLW"
        // arm64:"RORW",-"CSEL"
+       // loong64:"ROTR\t","SLLV"
        return bits.RotateLeft16(n, s)
 }
 
 func RotateLeft8(n uint8, s int) uint8 {
        // amd64:"ROLB" 386:"ROLB"
        // arm64:"LSL","LSR",-"CSEL"
+       // loong64:"OR","SLLV","SRLV"
        return bits.RotateLeft8(n, s)
 }
 
 func RotateLeftVariable(n uint, m int) uint {
        // amd64:"ROLQ"
        // arm64:"ROR"
+       // loong64:"ROTRV"
        // ppc64x:"ROTL"
        // riscv64:"ROL"
        // s390x:"RLLG"
@@ -273,6 +278,7 @@ func RotateLeftVariable(n uint, m int) uint {
 func RotateLeftVariable64(n uint64, m int) uint64 {
        // amd64:"ROLQ"
        // arm64:"ROR"
+       // loong64:"ROTRV"
        // ppc64x:"ROTL"
        // riscv64:"ROL"
        // s390x:"RLLG"
@@ -284,6 +290,7 @@ func RotateLeftVariable32(n uint32, m int) uint32 {
        // arm:`MOVW\tR[0-9]+@>R[0-9]+`
        // amd64:"ROLL"
        // arm64:"RORW"
+       // loong64:"ROTR\t"
        // ppc64x:"ROTLW"
        // riscv64:"ROLW"
        // s390x:"RLL"