https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109462
--- Comment #5 from Andrew Macleod <amacleod at redhat dot com> --- Created attachment 54835 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54835&action=edit in progress patch THe fix for PR 108139 disallowed an equivalences with a PHI because it may be a one way equivalence. The PHI may be an equivalence via UNDEFINED arguments so PHIDEF == name, but name may not be == PHIDEF. We were checking if the equivalence was a PHI node, buit we were not checking if the current NAME was a phi node. SO in this case, # Result$16_552 = PHI <_143(160), Result$16_453(D)(158)> we are evaluating Result$16_552, so the check for _143 from that PR was no triggering. I think we probably need to also check if the name we are evaluating is a PHI node as well. Patch is in testing. I do not know if it fixes the execution test or not, but it removes the problematic code we were looking at.
