https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114041
--- Comment #4 from rguenther at suse dot de <rguenther at suse dot de> --- On Thu, 22 Feb 2024, jakub at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114041 > > --- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> --- > I thought graphite is done only after bitint lowering. > At that point, there should be just <= MAX_FIXED_MODE_SIZE BITINT_TYPEs around > in the IL, with the exception of loading SSA_NAMEs from memory in order to > pass > them to function calls, or undefined SSA_NAMEs, or usually TREE_ADDRESSABLE > PARM_DECLs being converted to something else. > Would be probably wise to punt on graphite optimizations on loops which have > those larger SSA_NAMEs anywhere in the loop bodies. > That said, ub4_0_17(D) is unsigned _BitInt(4). I think we want to understand what goes wrong, not just disable graphite on any BitInt.
