cmd/compile: intrinsify RotateLeft32 on wasm

wasm has 32-bit versions of all integer operations. This change
lowers RotateLeft32 to i32.rotl on wasm and intrinsifies the math/bits
call.  Benchmarking on amd64 under node.js this is ~25% faster.

node v10.15.3/amd64
name          old time/op  new time/op  delta
RotateLeft    8.37ns ± 1%  8.28ns ± 0%   -1.05%  (p=0.029 n=4+4)
RotateLeft8   11.9ns ± 1%  11.8ns ± 0%     ~     (p=0.167 n=5+5)
RotateLeft16  11.8ns ± 0%  11.8ns ± 0%     ~     (all equal)
RotateLeft32  11.9ns ± 1%   8.7ns ± 0%  -26.32%  (p=0.008 n=5+5)
RotateLeft64  8.31ns ± 1%  8.43ns ± 2%     ~     (p=0.063 n=5+5)

Updates #31265

Change-Id: I5b8e155978faeea536c4f6427ac9564d2f096a46
Reviewed-on: https://go-review.googlesource.com/c/go/+/182359
Run-TryBot: Brian Kessler <brian.m.kessler@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Richard Musiol <neelance@gmail.com>

src/cmd/compile/internal/gc/ssa.go

@@ -3504,7 +3504,7 @@ func init() {
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue2(ssa.OpRotateLeft32, types.Types[TUINT32], args[0], args[1])
 		},
-		sys.AMD64, sys.ARM, sys.ARM64, sys.S390X, sys.PPC64)
+		sys.AMD64, sys.ARM, sys.ARM64, sys.S390X, sys.PPC64, sys.Wasm)
 	addF("math/bits", "RotateLeft64",
 		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
 			return s.newValue2(ssa.OpRotateLeft64, types.Types[TUINT64], args[0], args[1])

src/cmd/compile/internal/ssa/gen/Wasm.rules

@@ -147,7 +147,8 @@
 // Lowering rotates
 (RotateLeft8 <t> x (I64Const [c])) -> (Or8 (Lsh8x64 <t> x (I64Const [c&7])) (Rsh8Ux64 <t> x (I64Const [-c&7])))
 (RotateLeft16 <t> x (I64Const [c])) -> (Or16 (Lsh16x64 <t> x (I64Const [c&15])) (Rsh16Ux64 <t> x (I64Const [-c&15])))
-(RotateLeft32 <t> x (I64Const [c])) -> (Or32 (Lsh32x64 <t> x (I64Const [c&31])) (Rsh32Ux64 <t> x (I64Const [-c&31])))
+(RotateLeft32 x y) -> (I32Rotl x y)
+(RotateLeft64 x y) -> (I64Rotl x y)
 
 // Lowering comparisons
 (Less64  x y) -> (I64LtS x y)
@@ -362,8 +363,6 @@
 
 (BitLen64 x) -> (I64Sub (I64Const [64]) (I64Clz x))
 
-(RotateLeft64 x y) -> (I64Rotl x y)
-
 (PopCount64 x) -> (I64Popcnt x)
 (PopCount32 x) -> (I64Popcnt (ZeroExt32to64 x))
 (PopCount16 x) -> (I64Popcnt (ZeroExt16to64 x))

src/cmd/compile/internal/ssa/gen/WasmOps.go

@@ -206,6 +206,7 @@ func init() {
 
 		{name: "I64Ctz", asm: "I64Ctz", argLength: 1, reg: gp11, typ: "Int64"},       // ctz(arg0)
 		{name: "I64Clz", asm: "I64Clz", argLength: 1, reg: gp11, typ: "Int64"},       // clz(arg0)
+		{name: "I32Rotl", asm: "I32Rotl", argLength: 2, reg: gp21, typ: "Int32"},     // rotl(arg0, arg1)
 		{name: "I64Rotl", asm: "I64Rotl", argLength: 2, reg: gp21, typ: "Int64"},     // rotl(arg0, arg1)
 		{name: "I64Popcnt", asm: "I64Popcnt", argLength: 1, reg: gp11, typ: "Int64"}, // popcnt(arg0)
 	}

src/cmd/compile/internal/ssa/opGen.go

@@ -2168,6 +2168,7 @@ const (
 	OpWasmF64Copysign
 	OpWasmI64Ctz
 	OpWasmI64Clz
+	OpWasmI32Rotl
 	OpWasmI64Rotl
 	OpWasmI64Popcnt
 
@@ -29140,6 +29141,20 @@ var opcodeTable = [...]opInfo{
 			},
 		},
 	},
+	{
+		name:   "I32Rotl",
+		argLen: 2,
+		asm:    wasm.AI32Rotl,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 4295032831}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 SP
+				{1, 4295032831}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 SP
+			},
+			outputs: []outputInfo{
+				{0, 65535}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15
+			},
+		},
+	},
 	{
 		name:   "I64Rotl",
 		argLen: 2,

src/cmd/compile/internal/ssa/rewriteWasm.go

@@ -3826,36 +3826,17 @@ func rewriteValueWasm_OpRotateLeft16_0(v *Value) bool {
 	return false
 }
 func rewriteValueWasm_OpRotateLeft32_0(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
-	// match: (RotateLeft32 <t> x (I64Const [c]))
+	// match: (RotateLeft32 x y)
 	// cond:
-	// result: (Or32 (Lsh32x64 <t> x (I64Const [c&31])) (Rsh32Ux64 <t> x (I64Const [-c&31])))
+	// result: (I32Rotl x y)
 	for {
-		t := v.Type
-		_ = v.Args[1]
+		y := v.Args[1]
 		x := v.Args[0]
-		v_1 := v.Args[1]
-		if v_1.Op != OpWasmI64Const {
-			break
-		}
-		c := v_1.AuxInt
-		v.reset(OpOr32)
-		v0 := b.NewValue0(v.Pos, OpLsh32x64, t)
-		v0.AddArg(x)
-		v1 := b.NewValue0(v.Pos, OpWasmI64Const, typ.Int64)
-		v1.AuxInt = c & 31
-		v0.AddArg(v1)
-		v.AddArg(v0)
-		v2 := b.NewValue0(v.Pos, OpRsh32Ux64, t)
-		v2.AddArg(x)
-		v3 := b.NewValue0(v.Pos, OpWasmI64Const, typ.Int64)
-		v3.AuxInt = -c & 31
-		v2.AddArg(v3)
-		v.AddArg(v2)
+		v.reset(OpWasmI32Rotl)
+		v.AddArg(x)
+		v.AddArg(y)
 		return true
 	}
-	return false
 }
 func rewriteValueWasm_OpRotateLeft64_0(v *Value) bool {
 	// match: (RotateLeft64 x y)

src/cmd/compile/internal/wasm/ssa.go

@@ -299,6 +299,12 @@ func ssaGenValueOnStack(s *gc.SSAGenState, v *ssa.Value, extend bool) {
 		getValue64(s, v.Args[1])
 		s.Prog(v.Op.Asm())
 
+	case ssa.OpWasmI32Rotl:
+		getValue32(s, v.Args[0])
+		getValue32(s, v.Args[1])
+		s.Prog(wasm.AI32Rotl)
+		s.Prog(wasm.AI64ExtendI32U)
+
 	case ssa.OpWasmI64DivS:
 		getValue64(s, v.Args[0])
 		getValue64(s, v.Args[1])

test/codegen/mathbits.go

@@ -213,6 +213,7 @@ func RotateLeft32(n uint32) uint32 {
 	// ppc64:"ROTLW"
 	// ppc64le:"ROTLW"
 	// s390x:"RLL"
+	// wasm:"I32Rotl"
 	return bits.RotateLeft32(n, 9)
 }
 
@@ -232,6 +233,7 @@ func RotateLeftVariable(n uint, m int) uint {
 	// ppc64:"ROTL"
 	// ppc64le:"ROTL"
 	// s390x:"RLLG"
+	// wasm:"I64Rotl"
 	return bits.RotateLeft(n, m)
 }
 
@@ -241,6 +243,7 @@ func RotateLeftVariable64(n uint64, m int) uint64 {
 	// ppc64:"ROTL"
 	// ppc64le:"ROTL"
 	// s390x:"RLLG"
+	// wasm:"I64Rotl"
 	return bits.RotateLeft64(n, m)
 }
 
@@ -251,6 +254,7 @@ func RotateLeftVariable32(n uint32, m int) uint32 {
 	// ppc64:"ROTLW"
 	// ppc64le:"ROTLW"
 	// s390x:"RLL"
+	// wasm:"I32Rotl"
 	return bits.RotateLeft32(n, m)
 }