https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81833
--- Comment #2 from Bill Schmidt <wschmidt at gcc dot gnu.org> --- The "altivec_vsumsws" pattern for little endian isn't working as intended. This needs to be rewritten. So does the "altivec_vsum2sws" pattern. I'll work on a patch early next week.