https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63262
--- Comment #14 from Francois-Xavier Coudert <fxcoudert at gcc dot gnu.org> --- (In reply to karsten from comment #13) > libgmp.10.dylib`__gmpn_popcount + 26, queue = 'com.apple.main-thread', stop > reason = EXC_BAD_INSTRUCTION (code=EXC_I386_INVOP, subcode=0x0) > * frame #0: 0x00000001415c747a libgmp.10.dylib`__gmpn_popcount + 26 > frame #1: 0x0000000141578f67 libmpfr.4.dylib`parsed_string_to_mpfr + 2567 > frame #2: 0x000000014157847b libmpfr.4.dylib`mpfr_strtofr + 2011 > frame #3: 0x00000001006f4d11 cc1plus`real_from_string(real_value*, char > const*) + 817 > frame #4: 0x00000001006f5741 cc1plus`real_from_string3(real_value*, char > const*, machine_mode) + 33 So the segfault happens in GMP. I also confirm that I don't see it after doing a clean build of the 4.9.2 sources you linked, which makes sense because I do my builds with an older GMP version. What I would advise is building GMP without too much optimization, or with no optimization at all, and see if it fixes your issue. PS: the attached short testcase might be useful. Could you confirm that it does trigger the ICE (internal compiler error)?