cmd/compile: fix sign-extension merging rules

If we have

  y = <int16> (MOVBQSX x)
  z = <int32> (MOVWQSX y)

We used to use this rewrite rule:

(MOVWQSX x:(MOVBQSX _)) -> x

But that resulted in replacing z with a value whose type
is only int16.  Then if z is spilled and restored, it gets
zero extended instead of sign extended.

Instead use the rule

(MOVWQSX (MOVBQSX x)) -> (MOVBQSX x)

The result is has the correct type, so it can be spilled
and restored correctly.  It might mean that a few more extension
ops might not be eliminated, but that's the price for correctness.

Fixes #21963

Change-Id: I6ec82c3d2dbe43cc1fee6fb2bd6b3a72fca3af00
Reviewed-on: https://go-review.googlesource.com/65290
Reviewed-by: Cherry Zhang <[email protected]>
Run-TryBot: Cherry Zhang <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>

Author: randall77

src/cmd/compile/internal/ssa/gen/AMD64.rules

@@ -2512,18 +2512,20 @@
 (BSFQ (ORQconst <t> [1<<16] (MOVWQZX x))) -> (BSFQ (ORQconst <t> [1<<16] x))
 
 // Redundant sign/zero extensions
-(MOVLQSX x:(MOVLQSX _)) -> x
-(MOVLQSX x:(MOVWQSX _)) -> x
-(MOVLQSX x:(MOVBQSX _)) -> x
-(MOVWQSX x:(MOVWQSX _)) -> x
-(MOVWQSX x:(MOVBQSX _)) -> x
-(MOVBQSX x:(MOVBQSX _)) -> x
-(MOVLQZX x:(MOVLQZX _)) -> x
-(MOVLQZX x:(MOVWQZX _)) -> x
-(MOVLQZX x:(MOVBQZX _)) -> x
-(MOVWQZX x:(MOVWQZX _)) -> x
-(MOVWQZX x:(MOVBQZX _)) -> x
-(MOVBQZX x:(MOVBQZX _)) -> x
+// Note: see issue 21963. We have to make sure we use the right type on
+// the resulting extension (the outer type, not the inner type).
+(MOVLQSX (MOVLQSX x)) -> (MOVLQSX x)
+(MOVLQSX (MOVWQSX x)) -> (MOVWQSX x)
+(MOVLQSX (MOVBQSX x)) -> (MOVBQSX x)
+(MOVWQSX (MOVWQSX x)) -> (MOVWQSX x)
+(MOVWQSX (MOVBQSX x)) -> (MOVBQSX x)
+(MOVBQSX (MOVBQSX x)) -> (MOVBQSX x)
+(MOVLQZX (MOVLQZX x)) -> (MOVLQZX x)
+(MOVLQZX (MOVWQZX x)) -> (MOVWQZX x)
+(MOVLQZX (MOVBQZX x)) -> (MOVBQZX x)
+(MOVWQZX (MOVWQZX x)) -> (MOVWQZX x)
+(MOVWQZX (MOVBQZX x)) -> (MOVBQZX x)
+(MOVBQZX (MOVBQZX x)) -> (MOVBQZX x)
 
 (MOVQstore [off] {sym} ptr a:(ADDQconst [c] l:(MOVQload [off] {sym} ptr2 mem)) mem)
 	&& isSamePtr(ptr, ptr2) && a.Uses == 1 && l.Uses == 1 && validValAndOff(c,off) ->

src/cmd/compile/internal/ssa/rewriteAMD64.go

@@ -0,0 +1,27 @@
+// run
+
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import (
+	"fmt"
+	"runtime"
+)
+
+//go:noinline
+func f(x []int32, y *int8) int32 {
+	c := int32(int16(*y))
+	runtime.GC()
+	return x[0] * c
+}
+
+func main() {
+	var x = [1]int32{5}
+	var y int8 = -1
+	if got, want := f(x[:], &y), int32(-5); got != want {
+		panic(fmt.Sprintf("wanted %d, got %d", want, got))
+	}
+}