https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128869
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Ah, so we have # y_3 = PHI <y_7(3), 5(2), 0(4)> <L6>: _8 = .SUB_OVERFLOW (0x8000000000000000000000000000000000000000000000000000000000000000, y_3); _1 = IMAGPART_EXPR <_8>; _2 = (bool) _1; _9 = (int) _2; return _9; in the last bb, but we note there is a large/huge _BitInt switch and expand that one. And end up with s_5 = (_BitInt(257)) c_4(D); _8 = __complex__ (0, 1); _1 = IMAGPART_EXPR <_8>; _2 = (bool) _1; _9 = (int) _2; return _9; as the sole bb, note the COMPLEX_CST initializer of _8, and _8 and _1 and _5 are the only large/huge _BitInt SSA_NAMEs. And we don't handle COMPLEX_CST in bitint lowering.
