Copying a static value in an array with -Oz far from optimal size

[AZero13]

https://godbolt.org/z/o6cf889oG

Even ignoring the fact GCC's code is smaller, there is an instruction that is not being used that can make this happen faster:
repstosd

[rilysh]

-Oz also enables -O2, and stosd is 3 uops, where inc or add to REX is 1 uops. Both GCC and Clang generates very similar output.

GCC:

main:
        xor     eax, eax
.L2:
        mov     DWORD PTR h[0+rax*4], 23321332
        inc     rax
        cmp     rax, 32
        jne     .L2
        xor     eax, eax
        ret
h:
        .zero   128

And Clang:

main:
        xor     eax, eax
        lea     rcx, [rip + h]
.LBB0_1:
        cmp     rax, 128
        je      .LBB0_2
        mov     dword ptr [rax + rcx], 23321332
        add     rax, 4
        jmp     .LBB0_1
.LBB0_2:
        xor     eax, eax
        ret

h:
        .zero   128

GCC's generated assembly is increasing rax, with each iteration (total 32 iterations (if increasing by 1) or 8 (if increasing by 4)), whereas Clang is adding 4, with each iteration (total 8 iterations). Both seems nearly identical in terms of size, so I don't see any problem here.

[AZero13]

The point is that you can use repstosd this is something GCC ALSO gets wrong

[rilysh]

The point is that you can use repstosd this is something GCC ALSO gets wrong

Neither one is wrong, I think you didn't read my previous message. -Oz also enables a few optimizations from -O2 and stosd is 3 uops.