https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116155
--- Comment #5 from Dimitar Dimitrov <dimitar at gcc dot gnu.org> --- (In reply to Andrew Pinski from comment #1) > Unlike PR 116148, pru is little-endian but > > This test in the testcase: > if (sizeof (union with_fam_2) != __alignof__ (int)) > __builtin_abort (); > > > is wrong. If we can assume that "sizeof(char) <= __alignof__ (int)", then I think this case is valid and ok. To my understanding, idea is to ensure that the union size would be such that the "int b[]" variable array would be placed on aligned address. BTW, that check passes on pru. What fails on pru is: if (sizeof (struct only_fam_2) != sizeof (int)) __builtin_abort (); I think we need to separate it into a different file, and filter it for "target pcc_bitfield_type_matters".