cmd/compile: enable disjoint memmove inlining on amd64

Memmove can use AVX/prefetches/other optional instructions, so
only do it for small sizes, when call overhead dominates.

Change-Id: Ice5e93deb11462217f7fb5fc350b703109bb4090
Reviewed-on: https://go-review.googlesource.com/112517
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <mike.munday@ibm.com>

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

@@ -905,7 +905,7 @@ func isInlinableMemmove(dst, src *Value, sz int64, c *Config) bool {
 	// have fast Move ops.
 	switch c.arch {
 	case "amd64", "amd64p32":
-		return sz <= 16
+		return sz <= 16 || (sz < 1024 && disjoint(dst, sz, src, sz))
 	case "386", "ppc64", "ppc64le", "arm64":
 		return sz <= 8
 	case "s390x":

test/codegen/copy.go

@@ -40,6 +40,7 @@ var x [256]byte
 func moveDisjointStack() {
 	var s [256]byte
 	// s390x:-".*memmove"
+	// amd64:-".*memmove"
 	copy(s[:], x[:])
 	runtime.KeepAlive(&s)
 }
@@ -47,12 +48,14 @@ func moveDisjointStack() {
 func moveDisjointArg(b *[256]byte) {
 	var s [256]byte
 	// s390x:-".*memmove"
+	// amd64:-".*memmove"
 	copy(s[:], b[:])
 	runtime.KeepAlive(&s)
 }
 
 func moveDisjointNoOverlap(a *[256]byte) {
 	// s390x:-".*memmove"
+	// amd64:-".*memmove"
 	copy(a[:], a[128:])
 }