cmd/compile: special case slice consuming sliding window pattern

[Jorropo]

go version go1.26-devel_e7c0b74a97 Tue Jan 6 20:29:01 2026 +0100 linux/amd64

I don't know what the name is for this pattern so I've made one up, but basically the assembly for this function looks awful:

func ConstantTimeCompare(x, y []byte) int {
	if len(x) != len(y) { panic("...") }

	var v byte

	for ; len(x) > 0; x, y = x[1:], y[1:] {
		a := y[0]
		b := x[0]

		v |= a ^ b
	}

	return constanttime.Eq(v, 0)
}

• after early loop rotation cmd/compile: early loop rotation #77089 we could move s[1:] operations into if len(x) > 0 { /* here */; goto loop } second loop check, this would allow us to remove the slice mask
• we keep track of the length of both slices, even tho it is identical

[gopherbot]

Change https://go.dev/cl/733845 mentions this issue: crypto/internal/fips140/subtle: use pseudo-SIMD uint wide loops for []byte ops

[randall77]

I think my CL 688017 would also get rid of the slice masks.

[gabyhelp]

Related Issues

• cmd/compile: shorter slicing instruction sequence on x86/amd64 #47969
• cmd/compile: early loop rotation #77089
• cmd/compile: loads/constants not lifted out of loop #15808
• cmd/compile: optimize slice access with constant upper/lower bound #14266 (closed)
• cmd/compile: x86-64: sub-optimal codegen less-than-int sized index from structured slice/array #74489
• cmd/compile: range loop over []byte, []int can avoid base pointer increments #15809 (closed)
• cmd/compile: BCE of fixed width blocks during iteration #74474
• cmd/compile: slicing can be improved on ARM #23006
• cmd/compile: handle subslices in range -> memclear optimization #18908 (closed)
• cmd/compile: missed opportunity to coalesce loads that compare with constants #43248

_{(Emoji vote if this was helpful or unhelpful; more detailed feedback welcome in this discussion.)}

[Jorropo]

I think my CL 688017 would also get rid of the slice masks.

Indeed, thanks for reminding me of this one.