cmd/compile: extend all the cmov into math generic rules with their contrary

If the bool comes from a local operation this is foldable into the comparison.
  if a == b {
  } else {
    x++
  }
becomes:
  x += !(a == b)
becomes:
  x += a != b

If the bool is passed in or loaded rather than being locally computed
this adds an extra XOR ^1 to invert it.

But at worst it should make the math equal to the compute + CMP + CMOV
which is a tie on modern CPUs which can execute CMOV on all int ALUs
and a win on the cheaper or older ones which can't.

Change-Id: Idd2566c7a3826ec432ebfbba7b3898aa0db4b812
Reviewed-on: https://go-review.googlesource.com/c/go/+/760922
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
Auto-Submit: Jorropo <jorropo.pgm@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Junyang Shao <shaojunyang@google.com>

1 files changed, 71 insertions, 0 deletions

diff --git a/test/codegen/condmove.go b/test/codegen/condmove.go
index 6d1bba4c70..ec1b971ed7 100644
--- a/test/codegen/condmove.go
+++ b/test/codegen/condmove.go
@@ -464,6 +464,16 @@ func cmovmathadd(a uint, b bool) uint {
 	// wasm:"I64Add" -"Select"
 	return a
 }
+func cmovmathaddelse(a uint, b bool) uint {
+	if !b {
+		a++
+	}
+	// amd64:"ADDQ" -"CMOV"
+	// arm64:"CSINC" -"CSEL"
+	// ppc64x:"ADD" -"ISEL"
+	// wasm:"I64Add" -"Select"
+	return a
+}
 
 func cmovmathsub(a uint, b bool) uint {
 	if b {
@@ -475,6 +485,16 @@ func cmovmathsub(a uint, b bool) uint {
 	// wasm:"I64Sub" -"Select"
 	return a
 }
+func cmovmathsubelse(a uint, b bool) uint {
+	if !b {
+		a--
+	}
+	// amd64:"SUBQ" -"CMOV"
+	// arm64:"SUB" -"CSEL"
+	// ppc64x:"SUB" -"ISEL"
+	// wasm:"I64Sub" -"Select"
+	return a
+}
 
 func cmovmathdouble(a uint, b bool) uint {
 	if b {
@@ -486,6 +506,16 @@ func cmovmathdouble(a uint, b bool) uint {
 	// wasm:"I64Shl" -"Select"
 	return a
 }
+func cmovmathdoubleelse(a uint, b bool) uint {
+	if !b {
+		a *= 2
+	}
+	// amd64:"SHL" -"CMOV"
+	// amd64/v3:"SHL" -"CMOV" -"MOV"
+	// arm64:"LSL" -"CSEL"
+	// wasm:"I64Shl" -"Select"
+	return a
+}
 
 func cmovmathhalvei(a int, b bool) int {
 	if b {
@@ -498,6 +528,17 @@ func cmovmathhalvei(a int, b bool) int {
 	// wasm:-"Select"
 	return a
 }
+func cmovmathhalveielse(a int, b bool) int {
+	if !b {
+		// For some reason the compiler attributes the shift to inside this block rather than where the Phi node is.
+		// arm64:"ASR" -"CSEL"
+		// wasm:"I64ShrS" -"Select"
+		a /= 2
+	}
+	// arm64:-"CSEL"
+	// wasm:-"Select"
+	return a
+}
 
 func cmovmathhalveu(a uint, b bool) uint {
 	if b {
@@ -509,6 +550,16 @@ func cmovmathhalveu(a uint, b bool) uint {
 	// wasm:"I64ShrU" -"Select"
 	return a
 }
+func cmovmathhalveuelse(a uint, b bool) uint {
+	if !b {
+		a /= 2
+	}
+	// amd64:"SHR" -"CMOV"
+	// amd64/v3:"SHR" -"CMOV" -"MOV"
+	// arm64:"LSR" -"CSEL"
+	// wasm:"I64ShrU" -"Select"
+	return a
+}
 
 func cmovmathor(a uint, b bool) uint {
 	if b {
@@ -520,6 +571,16 @@ func cmovmathor(a uint, b bool) uint {
 	// wasm:"I64Or" -"Select"
 	return a
 }
+func cmovmathorelse(a uint, b bool) uint {
+	if !b {
+		a |= 1
+	}
+	// amd64:"ORQ" -"CMOV"
+	// arm64:"ORR" -"CSEL"
+	// ppc64x:"OR" -"ISEL"
+	// wasm:"I64Or" -"Select"
+	return a
+}
 
 func cmovmathxor(a uint, b bool) uint {
 	if b {
@@ -531,6 +592,16 @@ func cmovmathxor(a uint, b bool) uint {
 	// wasm:"I64Xor" -"Select"
 	return a
 }
+func cmovmathxorelse(a uint, b bool) uint {
+	if !b {
+		a ^= 1
+	}
+	// amd64:"XORQ" -"CMOV"
+	// arm64:"EOR" -"CSEL"
+	// ppc64x:"XOR" -"ISEL"
+	// wasm:"I64Xor" -"Select"
+	return a
+}
 
 func branchlessBoolToUint8(b bool) (r uint8) {
 	if b {