https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122215
--- Comment #5 from Jeffrey A. Law <law at gcc dot gnu.org> --- So one interesting thing I'm chasing elsewhere is another out of bounds memory reference in LRA. For the case I'm looking at the issue can be seen inside update_equiv where "regno" will be greater than or equal to "ira_reg_equiv_len". You could try something like b update_equiv if regno >= ira_reg_equiv_len If that triggers, then we're walking off the end of that array. That could result in this kind of failure, perhaps. You could even turn that into an assert if it's not that feasible to test it under a debugger. Anyway, I'm still poking at my issue, maybe we'll get lucky and ultimately they're the same underlying root cause.
