https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68861
--- Comment #1 from Jeffrey A. Law <law at redhat dot com> --- What a PITA to debug. I'm making some progress here in that I can see a hunk of memory that doesn't have the expected value. Unfortunately mapping that back to something done in optimizers is proving difficult. It doesn't help that the test is amazingly sensitive to reduction techniques -- to the point where I'm wondering if we've got a missing barrier/synchronization problem. Anyway I'm still plugging away.
