https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107647
Richard Biener <rguenth at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot gnu.org Priority|P3 |P2 --- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> --- I think the SLP pattern detection doesn't even work with a FMADD, FMSUB combo. We don't generate FMSUB early, not sure if we could somehow trick us into folding one with __builtin_fma, maybe __builtin_fma (x, y, -z) would do, I'd have to check. So yes, the SLP pattern detection introduces a FMA where none was before (and then vectorizes it). I'll fix it.