[Bug target/80846] auto-vectorized AVX2 horizontal sum should narrow to 128b right away, to be more efficient for Ryzen and Intel

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80846

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Similar with AVX512F I get

.L2:
        vmovdqa64       -112(%rbp), %zmm3
        addq    $64, %rdi
        vpaddd  -64(%rdi), %zmm3, %zmm2
        cmpq    %rdi, %rax
        vmovdqa64       %zmm2, -112(%rbp)
        jne     .L2
        vmovdqa -80(%rbp), %ymm0
        vpaddd  -112(%rbp), %ymm0, %ymm5
        vmovdqa %ymm5, -112(%rbp)
        vmovdqa -96(%rbp), %xmm0
        vpaddd  -112(%rbp), %xmm0, %xmm0
...