cmd/compile: regalloc: handle desired registers of 2-output insns

Particularly with 2-word load instructions, this becomes important.
Classic example is:

    func f(p *string) string {
        return *p
    }

We want the two loads to put the return values directly into
the two ABI return registers.

At this point in the stack, cmd/go is 1.1% smaller.

Change-Id: I51fd1710238e81d15aab2bfb816d73c8e7c207b1
Reviewed-on: https://go-review.googlesource.com/c/go/+/631137
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Keith Randall <khr@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

1 files changed, 10 insertions, 0 deletions

diff --git a/test/codegen/memcombine.go b/test/codegen/memcombine.go
index 2a9cc68ab0..9345391b61 100644
--- a/test/codegen/memcombine.go
+++ b/test/codegen/memcombine.go
@@ -1004,6 +1004,16 @@ func dwloadArg(a [2]int64) int64 {
 	return a[0] + a[1]
 }
 
+func dwloadResult1(p *string) string {
+	// arm64:"LDP\t\\(R0\\), \\(R0, R1\\)"
+	return *p
+}
+
+func dwloadResult2(p *[2]int64) (int64, int64) {
+	// arm64:"LDP\t\\(R0\\), \\(R1, R0\\)"
+	return p[1], p[0]
+}
+
 // ---------------------------------- //
 //    Arm64 double-register stores    //
 // ---------------------------------- //