https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124643
--- Comment #8 from Richard Sandiford <rsandifo at gcc dot gnu.org> --- Created attachment 64228 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=64228&action=edit POC patch for cse_local FWIW, here's a proof-of-concept patch to cse_local that makes it use validate_subreg when replacing subregs. It fixes the ICE, but I get the impression that it isn't what the port wants to happen.
