https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124180
--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> --- Created attachment 63766 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=63766&action=edit hack This hacks in non-convert cond_fixuns_trunc and friends and simply elides the if-conversion extra checking it does. If we want to have pattern fixups then I'd rather add the intermediate conversion there than elide it there. IMO we have to do without the premature check on vectorization in if-conversion - we might want to have a target macro indicating whether the target can do any kind of cond_* (or walk over all optabs and cache the answer) for compile-time issues. The vectorizable_call issue remains - maybe handling those in vectorizable_conversion will be easier, it already supports calls to some extent, but not masking or partial vectors.
