https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114040
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Ah, there is one difference, for the
unsigned
foo (unsigned _BitInt(8671) x, unsigned y, unsigned _BitInt(512) z)
{
unsigned _BitInt (8671) r
= x * __builtin_sub_overflow_p (y * z, 0, (unsigned _BitInt(255)) 0);
return r;
}
int
main ()
{
if (foo (1, 1, 0xfffa46471e7c2dd60000000000000000wb))
__builtin_abort ();
}
case bitint lowering chooses the same source and destination array, as source
it is ulong[8] - 8 limbs of unsigned _BitInt(512), as destination ulong[8] as
well, 4 limbs for the __real__ part and 4 limbs for the overflow flag (0/1).
But I don't see anything wrong with that in the bitintlower1 dump, all the
bitint.3 loads are done before the stores to the same location.
