https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115777
--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Hongtao Liu from comment #8) > > in backend costing we do anticipate the vector construction to happen > > by loading from memory though, so we don't account for the extra > > GPR->xmm move penalty. > Yes, I saw something similar before and had a patch, for ssa_name with > multiple uses, there should be a GPR->XMM even it's from memory. That's probably the conservative answer for BB vectorization, for loop vect we know all those uses will be also in vector code. For BB vectorization there is currently no easly reliable check to ensure this.
