cmd/compile: omit unnecessary boolean zero extension on arm64

On arm64, all boolean-generating instructions (CSET, etc.) set the upper
63 bits of the destination register to zero, so there is no need
to zero-extend the lower 8 bits again.

Fixes #21445

Change-Id: I3b176baab706eb684105400bacbaa24175f721f3
Reviewed-on: https://go-review.googlesource.com/55671
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

src/cmd/compile/internal/ssa/gen/ARM64.rules

@@ -1033,6 +1033,10 @@
 (GreaterEqual (InvertFlags x)) -> (LessEqual x)
 (GreaterEqualU (InvertFlags x)) -> (LessEqualU x)
 
+// Boolean-generating instructions always
+// zero upper bit of the register; no need to zero-extend
+(MOVBUreg x) && x.Type.IsBoolean() -> (MOVDreg x)
+
 // absorb flag constants into conditional instructions
 (CSELULT _ y (FlagEQ)) -> y
 (CSELULT x _ (FlagLT_ULT)) -> x

src/cmd/compile/internal/ssa/rewriteARM64.go

@@ -3790,6 +3790,18 @@ func rewriteValueARM64_OpARM64MOVBUreg_0(v *Value) bool {
 		v.AuxInt = int64(uint8(c))
 		return true
 	}
+	// match: (MOVBUreg x)
+	// cond: x.Type.IsBoolean()
+	// result: (MOVDreg x)
+	for {
+		x := v.Args[0]
+		if !(x.Type.IsBoolean()) {
+			break
+		}
+		v.reset(OpARM64MOVDreg)
+		v.AddArg(x)
+		return true
+	}
 	return false
 }
 func rewriteValueARM64_OpARM64MOVBload_0(v *Value) bool {