https://gcc.gnu.org/bugzilla/show_bug.cgi?id=24574
--- Comment #4 from rguenther at suse dot de <rguenther at suse dot de> --- On Tue, 12 Jul 2016, glisse at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=24574 > > --- Comment #3 from Marc Glisse <glisse at gcc dot gnu.org> --- > Makes sense, I don't know why I didn't add divisions in the original patch. > *_MOD_EXPR, shifts and rotates should work as well with 0 on the left, max > with > INT_MAX, etc, I was quite minimalist there. > > Of course, if I had gone with the simpler approach of folding the operation > after replacement (here 0 / 10) and comparing that with the PHI argument, we > wouldn't need the neutral/absorbing logic at all :-( Yeah, but folding would happily create undefined behavior from, say, if (x != INT_MIN) x = x * -1; as folding folds INT_MIN * -1 to INT_MIN (ok, maybe with TREE_OVERFLOW set, but I'm not sure we want to rely on that). Re-testing with mod and shift/rotates added.
