https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118476
--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> --- So backporting the fix for PR 116412 will also fix the issue here too. But still leave the latent bug in phiopt. The reason is that EVRP does the folding of: Folding statement: iftmp.1_9 = VIEW_CONVERT_EXPR<long unsigned int[1]>(a)[src_6]; into: Folded into: iftmp.1_9 = BIT_FIELD_REF <a, 64, 0>; And the match and simplify changes that into: Folded into: iftmp.1_9 = VIEW_CONVERT_EXPR<long unsigned int>(a); BUT with the patch for PR 116412, the conversion of BFR to VCE will no longer happen since a is a SSA_NAME. But this also means there might be some missed optimizations now on the trunk due to that. Let me find a testcase. I still think a patch like in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116412#c4 is the right way forward for backporting ...
