[Bug target/99548] Help me! Lost the fight against the compiler.
unlvsur at live dot com
gcc-bugzilla@gcc.gnu.org
Thu Mar 11 23:35:52 GMT 2021
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99548
--- Comment #2 from cqwrteur <unlvsur at live dot com> ---
(In reply to Andrew Pinski from comment #1)
> For aarch64 we get:
> ldp x4, x3, [x1]
> mov x7, 38
> ldp x6, x5, [x2]
> ldr x8, [x1, 16]
> add x6, x4, x6
> cmp x4, x6
> adc x5, x3, x5
> cmp x3, x5
> ldr x3, [x1, 24]
> ldp x4, x1, [x2, 16]
> adc x4, x8, x4
> cmp x8, x4
> adc x1, x3, x1
> cmp x3, x1
> csetm x2, cs
> and x2, x2, x7
> add x2, x2, x6
> cmp x2, x6
> cinc x6, x5, ls
> cmp x6, x5
> cinc x5, x4, ls
> cmp x5, x4
> cinc x3, x1, ls
> cmp x3, x1
> csetm x1, ls
> and x1, x1, x7
> add x1, x1, x2
> str x1, [x0]
> cmp x1, x2
> cinc x1, x6, ls
> str x1, [x0, 8]
> cmp x1, x6
> cinc x1, x5, ls
> cmp x1, x5
> cinc x3, x3, ls
> stp x1, x3, [x0, 16]
> ret
> There is only one missing optimization there really:
> csetm x1, ls
> and x1, x1, x7
> Should be turned into:
> csel x1, x7, xzw, ls
> Which is repeated a few times.
>
> It looks like x86_64 has other issues.
my add carry implementation for none-x86_64 platforms is wrong on godbolt.
More information about the Gcc-bugs
mailing list