mini-gmp "bug" missing mpz_fits_sint_p / mpz_fits_uint_p

[Marco Bodrato]

Il 2020-04-19 10:44 nisse at lysator.liu.se ha scritto:
> Marco Bodrato <bodrato at mail.dm.unipi.it> writes:
> 
>> +int
>> +mpz_fits_sint_p (const mpz_t u)
>> +{
>> +  return (INT_MAX + INT_MIN == 0 || mpz_cmp_ui (u, INT_MAX) <= 0) &&
>> +    mpz_cmpabs_ui (u, GMP_NEG_CAST (unsigned long int, INT_MIN)) <= 
>> 0;
>> +}
> 
> I think this and mpz_fits_sshort_p would be simpler using mpz_cmp_si,
> 
> int
> mpz_fits_sint_p (const mpz_t u)
> {
>   return mpz_cmp_si (u, INT_MAX) <= 0 && mpz_cmp_si (i, INT_MIN) >= 0;
> }

The current implementation of _cmp_si in mini- is:

mpz_cmp_si (const mpz_t u, long v)
{
   mp_size_t usize = u->_mp_size;

   if (v >= 0)
     return mpz_cmp_ui (u, v);
   else if (usize >= 0)
     return 1;
   else
     return - mpz_cmpabs_ui (u, GMP_NEG_CAST (unsigned long int, v));
}

So that the compiler may optimise your code to:

return mpz_cmp_ui (u, INT_MAX) <= 0 &&
        (u->_mp_size >= 0 ||
          mpz_cmpabs_ui (u, GMP_NEG_CAST (unsigned long int, INT_MIN)) <= 
0) ;

which basically is the condition I wrote, with one more branch.

> BTW, do we have any C implementation where INT_MAX + INT_MIN == 0, 
> i.e.,
> not using two's complement?

I'm almost sure the compiler can optimise that out at compile time.

Ĝis,
m