cmd/compile: use ANDconst to mask out leading/trailing bits on ARM64

For an AND that masks out leading or trailing bits, generic rules
rewrite it to a pair of shifts. On ARM64, the mask actually can
fit into an AND instruction. So we rewrite it back to AND.

Fixes #19857.

Change-Id: I479d7320ae4f29bb3f0056d5979bde4478063a8f
Reviewed-on: https://go-review.googlesource.com/39651
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>

3 files changed, 53 insertions, 12 deletions

diff --git a/src/cmd/compile/internal/gc/asm_test.go b/src/cmd/compile/internal/gc/asm_test.go
index dd96bec282..6c56b8d8e5 100644
--- a/src/cmd/compile/internal/gc/asm_test.go
+++ b/src/cmd/compile/internal/gc/asm_test.go
@@ -1327,6 +1327,22 @@ var linuxARM64Tests = []*asmTest{
 		`,
 		[]string{"\tCLZ\t"},
 	},
+	{
+		`
+		func f34(a uint64) uint64 {
+			return a & ((1<<63)-1)
+		}
+		`,
+		[]string{"\tAND\t"},
+	},
+	{
+		`
+		func f35(a uint64) uint64 {
+			return a & (1<<63)
+		}
+		`,
+		[]string{"\tAND\t"},
+	},
 }
 
 var linuxMIPSTests = []*asmTest{
diff --git a/src/cmd/compile/internal/ssa/gen/ARM64.rules b/src/cmd/compile/internal/ssa/gen/ARM64.rules
index 8f7680a347..189e739434 100644
--- a/src/cmd/compile/internal/ssa/gen/ARM64.rules
+++ b/src/cmd/compile/internal/ssa/gen/ARM64.rules
@@ -1148,8 +1148,11 @@
 ( ORshiftRL <t> [c] (SLLconst x [32-c]) (MOVWUreg x)) && c < 32 && t.Size() == 4 -> (RORWconst [   c] x)
 (XORshiftRL <t> [c] (SLLconst x [32-c]) (MOVWUreg x)) && c < 32 && t.Size() == 4 -> (RORWconst [   c] x)
 
-// Replace SRL-of-SLL with ROR-of-SLL to avoid hardware bug.
-(SRLconst [c] y:(SLLconst [c] _)) && c <= 8 -> (RORconst [c] y)
+// Generic rules rewrite certain AND to a pair of shifts.
+// However, on ARM64 the bitmask can fit into an instruction.
+// Rewrite it back to AND.
+(SRLconst [c] (SLLconst [c] x)) && 0 < c && c < 64 -> (ANDconst [1<<uint(64-c)-1] x) // mask out high bits
+(SLLconst [c] (SRLconst [c] x)) && 0 < c && c < 64 -> (ANDconst [^(1<<uint(c)-1)] x) // mask out low bits
 
 // do combined loads
 // little endian loads
diff --git a/src/cmd/compile/internal/ssa/rewriteARM64.go b/src/cmd/compile/internal/ssa/rewriteARM64.go
index c32bdf7a9e..16f3d06b27 100644
--- a/src/cmd/compile/internal/ssa/rewriteARM64.go
+++ b/src/cmd/compile/internal/ssa/rewriteARM64.go
@@ -8583,6 +8583,27 @@ func rewriteValueARM64_OpARM64SLLconst(v *Value) bool {
 		v.AuxInt = int64(d) << uint64(c)
 		return true
 	}
+	// match: (SLLconst [c] (SRLconst [c] x))
+	// cond: 0 < c && c < 64
+	// result: (ANDconst [^(1<<uint(c)-1)] x)
+	for {
+		c := v.AuxInt
+		v_0 := v.Args[0]
+		if v_0.Op != OpARM64SRLconst {
+			break
+		}
+		if v_0.AuxInt != c {
+			break
+		}
+		x := v_0.Args[0]
+		if !(0 < c && c < 64) {
+			break
+		}
+		v.reset(OpARM64ANDconst)
+		v.AuxInt = ^(1<<uint(c) - 1)
+		v.AddArg(x)
+		return true
+	}
 	return false
 }
 func rewriteValueARM64_OpARM64SRA(v *Value) bool {
@@ -8653,24 +8674,25 @@ func rewriteValueARM64_OpARM64SRLconst(v *Value) bool {
 		v.AuxInt = int64(uint64(d) >> uint64(c))
 		return true
 	}
-	// match: (SRLconst [c] y:(SLLconst [c] _))
-	// cond: c <= 8
-	// result: (RORconst [c] y)
+	// match: (SRLconst [c] (SLLconst [c] x))
+	// cond: 0 < c && c < 64
+	// result: (ANDconst [1<<uint(64-c)-1] x)
 	for {
 		c := v.AuxInt
-		y := v.Args[0]
-		if y.Op != OpARM64SLLconst {
+		v_0 := v.Args[0]
+		if v_0.Op != OpARM64SLLconst {
 			break
 		}
-		if y.AuxInt != c {
+		if v_0.AuxInt != c {
 			break
 		}
-		if !(c <= 8) {
+		x := v_0.Args[0]
+		if !(0 < c && c < 64) {
 			break
 		}
-		v.reset(OpARM64RORconst)
-		v.AuxInt = c
-		v.AddArg(y)
+		v.reset(OpARM64ANDconst)
+		v.AuxInt = 1<<uint(64-c) - 1
+		v.AddArg(x)
 		return true
 	}
 	return false