https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66420
--- Comment #2 from Vladimir Makarov <vmakarov at gcc dot gnu.org> --- I reproduced the problem with --eanble-checking=valgrind. The valgrind message refers for if ((hregno = ALLOCNO_HARD_REGNO (a)) < 0) base_cost = ALLOCNO_UPDATED_MEMORY_COST (a); I checked all ALL_ALLOCNO_HARD_REGNO in gdb (the test is not big) and I see that hard regnos are always setup before their usage. I also built the same GCC modifying Makefile the same way but *without* --eanble-checking=valgrind and then used valgrind directly. I have no valgrind messages. So I don't know is it IRA, GCC in whole, or valgrind problem. Probably, it is not an IRA problem.