cmd/compile: inline-able function with named results performs worse

[zigo101]

What version of Go are you using (go version)?

$ go version
go version go1.17.1 linux/amd64

Does this issue reproduce with the latest release?

Yes

What did you do?

package functions

import "testing"

const N = 1<<12
var buf = make([]byte, N)
var r [128][N]byte

func Benchmark_SliceToArray_Named(b *testing.B) {
	for i := 0; i < b.N; i++ {
		r[i&127] = SliceToArray_Named(buf)
	}
}

func Benchmark_SliceToArray_Unnamed(b *testing.B) {
	for i := 0; i < b.N; i++ {
		r[i&127] = SliceToArray_Unnamed(buf)
	}
}

func SliceToArray_Named(b []byte) (ret [N]byte) {
	ret = *(*[N]byte)(b)
	return
}

func SliceToArray_Unnamed(b []byte) [N]byte {
	return *(*[N]byte)(b)
}

What did you expect to see?

No performance difference

What did you see instead?

The named reult version is slower than the anonymous result version.

Benchmark_SliceToArray_Named-4     	 2983450	       397.7 ns/op
Benchmark_SliceToArray_Unnamed-4   	 4106510	       293.0 ns/op

It looks this is related to code inlining.

[randall77]

Related #20859

[zigo101]

Yes, some related, but not totally the same.
If we add the //go:noinline directive for the two functions,
then there is not performance difference.
That means the code generated for the two functions are the same
if they are not inlined, but not the same if they are inlined.

[randall77]

There is an extra zeroing of the named return value that doesn't exist in the unnamed one.