https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127274
--- Comment #4 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The releases/gcc-16 branch has been updated by jeevitha <[email protected]>: https://gcc.gnu.org/g:096da3d6e3814aeab097a091efb66908f1789821 commit r16-9895-g096da3d6e3814aeab097a091efb66908f1789821 Author: Jeevitha <[email protected]> Date: Tue Nov 3 23:30:27 2026 -0500 rs6000: Fix _mm_movemask intrinsics on Power10BE [PR127274] The Power10 implementation of the SSE movemask intrinsics uses vec_extractm(). On big-endian PowerPC, the vector element order causes vec_extractm() to produce the mask bits in a different order than required by the SSE movemask intrinsics. Reverse the vector element order with vec_reve() on big-endian Power10 before calling vec_extractm(). This ensures that _mm_movemask_pd(), _mm_movemask_ps(), and _mm_movemask_epi8() produce the same results on both big and little endian targets. 2026-11-02 Jeevitha Palanisamy <[email protected]> gcc/ PR target/127274 * config/rs6000/emmintrin.h (_mm_movemask_pd): Reverse vector element order on big-endian Power10 before calling vec_extractm(). (_mm_movemask_epi8): Likewise. * config/rs6000/xmmintrin.h (_mm_movemask_ps): Likewise. (cherry picked from commit b51e191dfdb9ae0bb98e7926498cc103670ff928)
