https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94734
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> --- (In reply to Jeffrey A. Law from comment #6) > THe whole point of that change is to not require a dominating load if the > object comes from the stack. Yeah, but I find that flawed. One can do it after performing get_ref_base_and_extent and verifying the offset is const and the whole access must be within the object, or one could do it with VR verification again verifying the access is fully within the object, but otherwise it can't be done and one needs to find dominating load, just the noload needs to handle also ARRAY_REFs etc., not only MEM_REFs.
