https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128225
--- Comment #3 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Richard Biener <[email protected]>: https://gcc.gnu.org/g:c06252ee5215bfb58ad97a8636234fc67318e052 commit r17-5667-gc06252ee5215bfb58ad97a8636234fc67318e052 Author: Richard Biener <[email protected]> Date: Wed Oct 28 11:40:37 2026 +0100 tree-optimization/128225 - ICE with OMP inbranch simd clone When the SIMD clone mask argument has a vector type with BLKmode we ICE when attempting to build the respective mask type. Use truth_type_for for !VECTOR_BOOLEAN_TYPE_P which handles that case just fine and as we are going to create a all-one constant mask is not an issue if that were to have BLKmode as well. For the VECTOR_BOOLEAN_TYPE_P case only re-construct mask types with a vector mode which matches the aarch64 case we do this dance for. I've also filled a hole of mask-mode verification for inbranch clone use for a not-inbranch use. PR tree-optimization/128225 * tree-vect-stmts.cc (vectorizable_simd_clone_call): Only use build_truth_vector_type_for_mode for vector mode vector boolean types. Check mask_mode of the simdclone is a supported one even if the inbranch clone is going to be used for a not-inbranch use. * gcc.dg/vect/vect-simd-clone-pr128225.c: New testcase.
