https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116955
--- Comment #2 from Andrew Stubbs <ams at gcc dot gnu.org> --- Compared to gfx908, gfx1100 lacks 64-lane permute and vector reductions. Permute works with 32 lanes or fewer, but reductions are unimplemented in the backend. Otherwise it should vectorize the same. That might explain some additional failures, but probably doesn't explain a regression.
