https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125746
--- Comment #4 from Dimitar Dimitrov <dimitar at gcc dot gnu.org> --- Created attachment 64747 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=64747&action=edit Fix I'm testing as of 20260615 I was wrong - decode_reg_name_and_count seems to be working correctly. The register mode is validated during expand: make_decl_rtl() -> targetm.hard_regno_mode_ok() Drea, I believe this particular PR is a distinct target-specific issue. Thanks for the pointers. I'm now testing the attached fix.
