https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107642
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> --- At some point one of the ideas was to avoid deciding this "early" and leave it entirely to RTL expansion what to do to have more similar GIMPLE through the pipeline. That eventually means prefering if (a & b) to if (a && b) on GIMPLE, RTL expansion can only change it in one direction.
