cmd/compile: rewrite condselects into doublings and halvings

For performance see CL 685676.

This allows something like:
  if y { x *= 2 }

To be compiled to:
  SHLXQ BX, AX, AX

Instead of:
  MOVQ    AX, CX
  SHLQ    $1, CX
  MOVBLZX BL, DX
  TESTQ   DX, DX
  CMOVQNE CX, AX

While ./make.bash uniqued per LOC, there is 2 doublings and 4 halvings.

Change-Id: Ic0727cbf429528a2dbf17cbfc3b0121db8387444
Reviewed-on: https://go-review.googlesource.com/c/go/+/685695
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Keith Randall <khr@golang.org>

2 files changed, 247 insertions, 0 deletions

diff --git a/src/cmd/compile/internal/ssa/_gen/generic.rules b/src/cmd/compile/internal/ssa/_gen/generic.rules
index 89657bdabb..b98dfae2d5 100644
--- a/src/cmd/compile/internal/ssa/_gen/generic.rules
+++ b/src/cmd/compile/internal/ssa/_gen/generic.rules
@@ -2855,3 +2855,10 @@
 // if b { x-- } => x -= b
 (CondSelect (Add8 <t> x (Const8 [-1])) x bool) => (Sub8 x (CvtBoolToUint8 <t> bool))
 (CondSelect (Add(64|32|16) <t> x (Const(64|32|16) [-1])) x bool) => (Sub(64|32|16) x (ZeroExt8to(64|32|16) <t> (CvtBoolToUint8 <types.Types[types.TUINT8]> bool)))
+
+// if b { x <<= 1 } => x <<= b
+(CondSelect (Lsh(64|32|16|8)x64 x (Const64 [1])) x bool) => (Lsh(64|32|16|8)x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+
+// if b { x >>= 1 } => x >>= b
+(CondSelect (Rsh(64|32|16|8)x64 x (Const64 [1])) x bool) => (Rsh(64|32|16|8)x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+(CondSelect (Rsh(64|32|16|8)Ux64 x (Const64 [1])) x bool) => (Rsh(64|32|16|8)Ux8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
diff --git a/src/cmd/compile/internal/ssa/rewritegeneric.go b/src/cmd/compile/internal/ssa/rewritegeneric.go
index a8c3373e40..0b9f9c09f9 100644
--- a/src/cmd/compile/internal/ssa/rewritegeneric.go
+++ b/src/cmd/compile/internal/ssa/rewritegeneric.go
@@ -5942,6 +5942,246 @@ func rewriteValuegeneric_OpCondSelect(v *Value) bool {
 		}
 		break
 	}
+	// match: (CondSelect (Lsh64x64 x (Const64 [1])) x bool)
+	// result: (Lsh64x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpLsh64x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpLsh64x8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Lsh32x64 x (Const64 [1])) x bool)
+	// result: (Lsh32x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpLsh32x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpLsh32x8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Lsh16x64 x (Const64 [1])) x bool)
+	// result: (Lsh16x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpLsh16x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpLsh16x8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Lsh8x64 x (Const64 [1])) x bool)
+	// result: (Lsh8x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpLsh8x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpLsh8x8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Rsh64x64 x (Const64 [1])) x bool)
+	// result: (Rsh64x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpRsh64x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpRsh64x8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Rsh32x64 x (Const64 [1])) x bool)
+	// result: (Rsh32x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpRsh32x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpRsh32x8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Rsh16x64 x (Const64 [1])) x bool)
+	// result: (Rsh16x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpRsh16x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpRsh16x8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Rsh8x64 x (Const64 [1])) x bool)
+	// result: (Rsh8x8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpRsh8x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpRsh8x8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Rsh64Ux64 x (Const64 [1])) x bool)
+	// result: (Rsh64Ux8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpRsh64Ux64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpRsh64Ux8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Rsh32Ux64 x (Const64 [1])) x bool)
+	// result: (Rsh32Ux8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpRsh32Ux64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpRsh32Ux8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Rsh16Ux64 x (Const64 [1])) x bool)
+	// result: (Rsh16Ux8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpRsh16Ux64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpRsh16Ux8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
+	// match: (CondSelect (Rsh8Ux64 x (Const64 [1])) x bool)
+	// result: (Rsh8Ux8 [true] x (CvtBoolToUint8 <types.Types[types.TUINT8]> bool))
+	for {
+		if v_0.Op != OpRsh8Ux64 {
+			break
+		}
+		_ = v_0.Args[1]
+		x := v_0.Args[0]
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || auxIntToInt64(v_0_1.AuxInt) != 1 || x != v_1 {
+			break
+		}
+		bool := v_2
+		v.reset(OpRsh8Ux8)
+		v.AuxInt = boolToAuxInt(true)
+		v0 := b.NewValue0(v.Pos, OpCvtBoolToUint8, types.Types[types.TUINT8])
+		v0.AddArg(bool)
+		v.AddArg2(x, v0)
+		return true
+	}
 	return false
 }
 func rewriteValuegeneric_OpConstInterface(v *Value) bool {