https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116979
--- Comment #19 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Uroš Bizjak from comment #18) > (In reply to Jakub Jelinek from comment #17) > > Not marking as fixed for GCC 15 yet, as there is the scalar cost computation > > issue unresolved. > There is also a issue how the final result for SFmode is constructed. It can > be seen when compiled with -ffast-math: > > vmovshdup %xmm0, %xmm4 > vmovss %xmm0, -8(%rsp) > vmovss %xmm4, -4(%rsp) > vmovq -8(%rsp), %xmm0 > > This will result in store forwarding stall. That looks like a backend issue? A missed vec_init patttern?
