[AArch64] Fold `and` and `cmp` into `tst`

[Kmeakin]

https://godbolt.org/z/Mh7q8TY99

InstCombine is able to fold (x & 0xFF) < C) into (x & -C) == 0

eg:

bool ult32_u32(u32 x) { return (x & 0xFF) < 32; }

produces

ult32_u32:
        tst     w0, #0xe0
        cset    w0, eq
        ret

But the same transform is not done in later stages, so if the and is introduced due to eg passing a u8 in a u32 register, the fold is not performed:

bool ult32_u8(u8 x) { return x < 32; }

ult32_u8:
        and     w8, w0, #0xff
        cmp     w8, #32
        cset    w0, lo
        ret

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: Karl Meakin (Kmeakin)

https://godbolt.org/z/Mh7q8TY99

InstCombine is able to fold (x & 0xFF) < C) into (x & -C) != 0

eg:

bool ult32_u32(u32 x) { return (x & 0xFF) < 32; }

produces

ult32_u32:
        tst     w0, #<!-- -->0xe0
        cset    w0, eq
        ret

But the same transform is not done in later stages, so if the and is introduced due to eg passing a u8 in a u32 register, the fold is not performed:

bool ult32_u8(u8 x) { return x < 32; }

ult32_u8:
        and     w8, w0, #<!-- -->0xff
        cmp     w8, #<!-- -->32
        cset    w0, lo
        ret

[pvimal816]

Hi @Kmeakin
Orthogonal to the root issue, shouldn't (x & -C) != 0 be (x & -C) == 0 instead?

Because, as per my understanding, tst a, b will set Z flag only if a&b is zero, and unless Z flag is set cset a, eq will not set a.

[jf-botto]

I'd like to work on this issue, if possible? Please assign me?

[Kmeakin]

Hi @Kmeakin Orthogonal to the root issue, shouldn't (x & -C) != 0 be (x & -C) == 0 instead?

Because, as per my understanding, tst a, b will set Z flag only if a&b is zero, and unless Z flag is set cset a, eq will not set a.

Yes, that was a typo, thanks