https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122947
--- Comment #1 from Alexandre Oliva <aoliva at gcc dot gnu.org> --- Created attachment 62965 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62965&action=edit variant of pr117239.c that triggers the problem with pristine trunk We don't need the cselib improvement to hit the problem, after all. The difficulty was that the assignment to global d was invalidating the +16 part of the argument from the cselib table. Making d local and enforcing its later use was enough to get the corresponding arg store for the second call to be misoptimized out.
