https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122620
--- Comment #4 from rguenther at suse dot de <rguenther at suse dot de> --- On Mon, 10 Nov 2025, jakub at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122620 > > --- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> --- > Far from having a usable testcase, but having a clobber with reg type (which > is > what the ICE is on when trying to regimplify it) is quite easy, e.g. Yes. Those are not canonical GIMPLE stmts but it seems we ignore that well enough ... apart from when trying to re-gimplify them. So I'd suggest to special-case those in the re-gimplifier somehow. I do wonder whether we want to disallow such clobbers in general or somehow make an exception to them. Like by making a clobber {} is_gimple_constant, when of register type (or in general?).
