https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123631
--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> --- Thanks for the detailed response. I was indeed mostly wondering about the uniform vector case and as said I'm expecting such materializations to be hoisted out of loops (but that needs doing on RTL, so eventually register pressure issues might prevent it there). Ideally we'd be able to decide "late" how to materialize such constants, in particular not having to allocate the extra GPR early but only late when available. As for register files I'd have expected Intel to have a unified GPR and k-mask register file given their lack of split integer / FP domains. OTOH the weird setup of having separate k* instructions is a sign they don't.
