https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86948
--- Comment #2 from Alexander Monakov <amonakov at gcc dot gnu.org> --- Thanks for the heads-up. At present, BRIG frontend cannot rely on MULT_HIGHPART, as the pattern is optional and, as this bug shows, generic expansion is not implemented. BRIG frontend is already somewhat aware of the issue, as the comment in brig-basic-inst-handler.cc indicates: 554 /* MULT_HIGHPART_EXPR works only on target dependent vector sizes and 555 even the scalars do not seem to work at least for char elements. 556 557 Let's fall back to scalarization and promotion [...] So a quick-fix would be to never try to use MULT_HIGHPART_EXPR from BRIG FE. However, I think it would be nice to extend expand_mult_highpart to handle this, at least to avoid regressing here. I'll look into that. I'll also try to expose MULT_HIGHPART_EXPR via the GIMPLE FE so it can be tested in a more straightforward manner.