Possible bug in 128-bit integers implementation (gcc_checked_binop)

[antoyo]

It seems there's a mixup between the overflow flag and the result of the operation from what we could see here.

The code that might be wrong is here.

For instance __mulosi4 has the following signature:

si_int __mulosi4(si_int a, si_int b, int *overflow);

while __builtin_mul_overflow has:

bool __builtin_mul_overflow (type1 a, type2 b, type3 *res)

Yet, they seem to be called the same way (mixing up overflow with the return value of res with the other return value).