[Aarch64] `lsr v, v, #56`+`ldr d0, [x9, x8, lsl #3]` optimized to `lsr v, v, #53`+`and v, v, #0x7f8`+`ldr d0, [x9, x8]`

[Validark]

Godbolt link

export fn foo(bitstring: u64, dest: [*]u8, src: [*][8]u8) void {
    dest[0..8].* = src[bitstring >> 56];
}

Compiled for the Apple M3:

foo:
        lsr     x8, x0, #53
        and     x8, x8, #0x7f8
        ldr     x8, [x2, x8]
        str     x8, [x1]
        ret

I think it should be:

foo:
        lsr     x8, x0, #56
        ldr     x8, [x2, x8, lsl #3]
        str     x8, [x1]
        ret

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: Niles Salter (Validark)

[Godbolt link](https://zig.godbolt.org/z/sbefWdEcf)

export fn foo(bitstring: u64, dest: [*]u8, src: [*][8]u8) void {
    dest[0..8].* = src[bitstring >> 56];
}

Compiled for the Apple M3:

foo:
        lsr     x8, x0, #<!-- -->53
        and     x8, x8, #<!-- -->0x7f8
        ldr     x8, [x2, x8]
        str     x8, [x1]
        ret

I think it should be:

foo:
        lsr     x8, x0, #<!-- -->56
        ldr     x8, [x2, x8, lsl #<!-- -->3]
        str     x8, [x1]
        ret

[davemgreen]

I believe this might have been fixed by #89531.
https://godbolt.org/z/sxT1dsxed

[Validark]

Looks like it was! Job well done!