https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120138
--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to Thor Preimesberger from comment #4) > FWIW I bisected to commit d45ddc2c04e471d0dcee016b6edacc00b8341b16 - adding > this commit and compiling with -O3 causes two additional jumps to be > threaded. > > The gimple stmt > > _68 = r[1] > > in transpose_avx512_contiguous_thin<float> is triggering uninit1. > > It looks like, before the above commit, the above gimple has a couple more > nodes that are visited by walk_aliased_vdefs in maybe_warn_operand. I have > some nasty printf debugging logs but I'm not sure if they're useful enough > to bother attaching Right the warning seems correct based on the code even. Just the jump threading has been able to expose the warning. The code does look like in the case of M being between 0 and 7 will cause use of the uninitilized r[1].
