[Bug tree-optimization/67577] Trivial float-vectorization foiled by a loop

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67577

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> ---
Trunk with -O3 -mavx generates:

Z1xv:
        .cfi_startproc
        vmovaps b(%rip), %ymm1
        vmovaps a(%rip), %ymm0
        vbroadcastss    scale(%rip), %ymm2
        vsubps  %ymm1, %ymm0, %ymm0
        vmulps  %ymm2, %ymm0, %ymm0
        vaddps  %ymm1, %ymm0, %ymm0
        vmovaps %ymm0, c(%rip)
        vzeroupper
        ret
        .cfi_endproc