cmd/compile: wire up bits.Reverse intrinsics for loong64

Micro-benchmark results on Loongson 3A5000 and 3A6000:

goos: linux
goarch: loong64
pkg: math/bits
cpu: Loongson-3A6000 @ 2500.00MHz
          |  CL 624576   |               this CL                |
          |    sec/op    |    sec/op     vs base                |
Reverse     2.8130n ± 0%   0.8008n ± 0%  -71.53% (p=0.000 n=20)
Reverse8    0.7014n ± 0%   0.4040n ± 0%  -42.40% (p=0.000 n=20)
Reverse16   1.2975n ± 0%   0.6632n ± 1%  -48.89% (p=0.000 n=20)
Reverse32   2.7520n ± 0%   0.4042n ± 0%  -85.31% (p=0.000 n=20)
Reverse64   2.8970n ± 0%   0.4041n ± 0%  -86.05% (p=0.000 n=20)
geomean      1.828n        0.5116n       -72.01%

goos: linux
goarch: loong64
pkg: math/bits
cpu: Loongson-3A5000 @ 2500.00MHz
          |  CL 624576   |               this CL                |
          |    sec/op    |    sec/op     vs base                |
Reverse     4.0050n ± 0%   0.8011n ± 0%  -80.00% (p=0.000 n=20)
Reverse8    0.8010n ± 0%   0.5210n ± 1%  -34.96% (p=0.000 n=20)
Reverse16   1.6160n ± 0%   0.6008n ± 0%  -62.82% (p=0.000 n=20)
Reverse32   3.8550n ± 0%   0.5179n ± 0%  -86.57% (p=0.000 n=20)
Reverse64   3.8050n ± 0%   0.5177n ± 0%  -86.40% (p=0.000 n=20)
geomean      2.378n        0.5828n       -75.49%

Updates #59120

This patch is a copy of CL 483656.
Co-authored-by: WANG Xuerui <git@xen0n.name>

Change-Id: I98681091763279279c8404bd0295785f13ea1c8e
Reviewed-on: https://go-review.googlesource.com/c/go/+/624276
Reviewed-by: abner chenc <chenguoqi@loongson.cn>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>

7 files changed, 89 insertions, 5 deletions

diff --git a/src/cmd/compile/internal/loong64/ssa.go b/src/cmd/compile/internal/loong64/ssa.go
index e7cb82a280..4c9bcfe46e 100644
--- a/src/cmd/compile/internal/loong64/ssa.go
+++ b/src/cmd/compile/internal/loong64/ssa.go
@@ -516,6 +516,9 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		ssa.OpLOONG64REVB2H,
 		ssa.OpLOONG64REVB2W,
 		ssa.OpLOONG64REVBV,
+		ssa.OpLOONG64BITREV4B,
+		ssa.OpLOONG64BITREVW,
+		ssa.OpLOONG64BITREVV,
 		ssa.OpLOONG64ABSD:
 		p := s.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG
diff --git a/src/cmd/compile/internal/ssa/_gen/LOONG64.rules b/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
index bac1f27b1d..6ff98a46f7 100644
--- a/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
+++ b/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
@@ -148,6 +148,10 @@
 (BitLen64 <t> x) => (NEGV <t> (SUBVconst <t> [64] (CLZV <t> x)))
 (BitLen32 <t> x) => (NEGV <t> (SUBVconst <t> [32] (CLZW <t> x)))
 (Bswap(16|32|64) ...) => (REVB(2H|2W|V) ...)
+(BitRev8 ...) => (BITREV4B ...)
+(BitRev16 <t> x) => (REVB2H (BITREV4B <t> x))
+(BitRev32 ...) => (BITREVW ...)
+(BitRev64 ...) => (BITREVV ...)
 
 // math package intrinsics
 (Sqrt ...) => (SQRTD ...)
diff --git a/src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go b/src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go
index 0da0bb8227..a8a38ee7b8 100644
--- a/src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go
+++ b/src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go
@@ -216,6 +216,10 @@ func init() {
 		{name: "REVB2W", argLength: 1, reg: gp11, asm: "REVB2W"}, // Swap bytes: 0x1122334455667788 -> 0x4433221188776655
 		{name: "REVBV", argLength: 1, reg: gp11, asm: "REVBV"},   // Swap bytes: 0x1122334455667788 -> 0x8877665544332211
 
+		{name: "BITREV4B", argLength: 1, reg: gp11, asm: "BITREV4B"}, // Reverse the bits of each byte inside a 32-bit arg[0]
+		{name: "BITREVW", argLength: 1, reg: gp11, asm: "BITREVW"},   // Reverse the bits in a 32-bit arg[0]
+		{name: "BITREVV", argLength: 1, reg: gp11, asm: "BITREVV"},   // Reverse the bits in a 64-bit arg[0]
+
 		{name: "FMINF", argLength: 2, reg: fp21, resultNotInArgs: true, asm: "FMINF", commutative: true, typ: "Float32"}, // min(arg0, arg1), float32
 		{name: "FMIND", argLength: 2, reg: fp21, resultNotInArgs: true, asm: "FMIND", commutative: true, typ: "Float64"}, // min(arg0, arg1), float64
 		{name: "FMAXF", argLength: 2, reg: fp21, resultNotInArgs: true, asm: "FMAXF", commutative: true, typ: "Float32"}, // max(arg0, arg1), float32
diff --git a/src/cmd/compile/internal/ssa/opGen.go b/src/cmd/compile/internal/ssa/opGen.go
index f1006a3f3c..ae0e87702a 100644
--- a/src/cmd/compile/internal/ssa/opGen.go
+++ b/src/cmd/compile/internal/ssa/opGen.go
@@ -1801,6 +1801,9 @@ const (
 	OpLOONG64REVB2H
 	OpLOONG64REVB2W
 	OpLOONG64REVBV
+	OpLOONG64BITREV4B
+	OpLOONG64BITREVW
+	OpLOONG64BITREVV
 	OpLOONG64FMINF
 	OpLOONG64FMIND
 	OpLOONG64FMAXF
@@ -24202,6 +24205,45 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "BITREV4B",
+		argLen: 1,
+		asm:    loong64.ABITREV4B,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 1073741816}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 g R23 R24 R25 R26 R27 R28 R29 R31
+			},
+			outputs: []outputInfo{
+				{0, 1071644664}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R23 R24 R25 R26 R27 R28 R29 R31
+			},
+		},
+	},
+	{
+		name:   "BITREVW",
+		argLen: 1,
+		asm:    loong64.ABITREVW,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 1073741816}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 g R23 R24 R25 R26 R27 R28 R29 R31
+			},
+			outputs: []outputInfo{
+				{0, 1071644664}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R23 R24 R25 R26 R27 R28 R29 R31
+			},
+		},
+	},
+	{
+		name:   "BITREVV",
+		argLen: 1,
+		asm:    loong64.ABITREVV,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 1073741816}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 g R23 R24 R25 R26 R27 R28 R29 R31
+			},
+			outputs: []outputInfo{
+				{0, 1071644664}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R23 R24 R25 R26 R27 R28 R29 R31
+			},
+		},
+	},
+	{
 		name:            "FMINF",
 		argLen:          2,
 		commutative:     true,
diff --git a/src/cmd/compile/internal/ssa/rewriteLOONG64.go b/src/cmd/compile/internal/ssa/rewriteLOONG64.go
index 88c5036b54..779ec89134 100644
--- a/src/cmd/compile/internal/ssa/rewriteLOONG64.go
+++ b/src/cmd/compile/internal/ssa/rewriteLOONG64.go
@@ -125,6 +125,17 @@ func rewriteValueLOONG64(v *Value) bool {
 		return rewriteValueLOONG64_OpBitLen32(v)
 	case OpBitLen64:
 		return rewriteValueLOONG64_OpBitLen64(v)
+	case OpBitRev16:
+		return rewriteValueLOONG64_OpBitRev16(v)
+	case OpBitRev32:
+		v.Op = OpLOONG64BITREVW
+		return true
+	case OpBitRev64:
+		v.Op = OpLOONG64BITREVV
+		return true
+	case OpBitRev8:
+		v.Op = OpLOONG64BITREV4B
+		return true
 	case OpBswap16:
 		v.Op = OpLOONG64REVB2H
 		return true
@@ -975,6 +986,21 @@ func rewriteValueLOONG64_OpBitLen64(v *Value) bool {
 		return true
 	}
 }
+func rewriteValueLOONG64_OpBitRev16(v *Value) bool {
+	v_0 := v.Args[0]
+	b := v.Block
+	// match: (BitRev16 <t> x)
+	// result: (REVB2H (BITREV4B <t> x))
+	for {
+		t := v.Type
+		x := v_0
+		v.reset(OpLOONG64REVB2H)
+		v0 := b.NewValue0(v.Pos, OpLOONG64BITREV4B, t)
+		v0.AddArg(x)
+		v.AddArg(v0)
+		return true
+	}
+}
 func rewriteValueLOONG64_OpCom16(v *Value) bool {
 	v_0 := v.Args[0]
 	b := v.Block
diff --git a/src/cmd/compile/internal/ssagen/intrinsics.go b/src/cmd/compile/internal/ssagen/intrinsics.go
index 33345e9296..db335ee8b3 100644
--- a/src/cmd/compile/internal/ssagen/intrinsics.go
+++ b/src/cmd/compile/internal/ssagen/intrinsics.go
@@ -946,27 +946,27 @@ func initIntrinsics(cfg *intrinsicBuildConfig) {
 		func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpBitRev64, types.Types[types.TINT], args[0])
 		},
-		sys.ARM64)
+		sys.ARM64, sys.Loong64)
 	addF("math/bits", "Reverse32",
 		func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpBitRev32, types.Types[types.TINT], args[0])
 		},
-		sys.ARM64)
+		sys.ARM64, sys.Loong64)
 	addF("math/bits", "Reverse16",
 		func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpBitRev16, types.Types[types.TINT], args[0])
 		},
-		sys.ARM64)
+		sys.ARM64, sys.Loong64)
 	addF("math/bits", "Reverse8",
 		func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpBitRev8, types.Types[types.TINT], args[0])
 		},
-		sys.ARM64)
+		sys.ARM64, sys.Loong64)
 	addF("math/bits", "Reverse",
 		func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
 			return s.newValue1(ssa.OpBitRev64, types.Types[types.TINT], args[0])
 		},
-		sys.ARM64)
+		sys.ARM64, sys.Loong64)
 	addF("math/bits", "RotateLeft8",
 		func(s *state, n *ir.CallExpr, args []*ssa.Value) *ssa.Value {
 			return s.newValue2(ssa.OpRotateLeft8, types.Types[types.TUINT8], args[0], args[1])
diff --git a/src/cmd/compile/internal/ssagen/intrinsics_test.go b/src/cmd/compile/internal/ssagen/intrinsics_test.go
index a8656cc3d4..ca9e1c9e0a 100644
--- a/src/cmd/compile/internal/ssagen/intrinsics_test.go
+++ b/src/cmd/compile/internal/ssagen/intrinsics_test.go
@@ -421,6 +421,11 @@ var wantIntrinsics = map[testIntrinsicKey]struct{}{
 	{"loong64", "math/bits", "Len16"}:                          struct{}{},
 	{"loong64", "math/bits", "Len32"}:                          struct{}{},
 	{"loong64", "math/bits", "Len64"}:                          struct{}{},
+	{"loong64", "math/bits", "Reverse"}:                        struct{}{},
+	{"loong64", "math/bits", "Reverse8"}:                       struct{}{},
+	{"loong64", "math/bits", "Reverse16"}:                      struct{}{},
+	{"loong64", "math/bits", "Reverse32"}:                      struct{}{},
+	{"loong64", "math/bits", "Reverse64"}:                      struct{}{},
 	{"loong64", "math/bits", "RotateLeft"}:                     struct{}{},
 	{"loong64", "math/bits", "RotateLeft32"}:                   struct{}{},
 	{"loong64", "math/bits", "RotateLeft64"}:                   struct{}{},