[fortran,patch] Call mpfr_check_range after setting emin/emax

[Tobias Burnus]

FX wrote:
> The two patchlets below regtest fine on my system (x86_64-linux, both 
> -m32 and -m64) and fix the issue I am seeing with 
> gfortran.dg/nearest_4.f90. Is it OK to commit?
The arith.c part is OK. However, I wonder whether
>> @@ -2821,6 +2821,7 @@ gfc_simplify_nearest (gfc_expr *x, gfc_e
>>    mpfr_set_emin ((mp_exp_t) gfc_real_kinds[kind].min_exponent -
>>                 mpfr_get_prec(result->value.real) + 1);
>>    mpfr_set_emax ((mp_exp_t) gfc_real_kinds[kind].max_exponent - 1);
>> +  mpfr_check_range (result->value.real, 0, GMP_RNDU);
shouldn't be rather

  mpfr_set_emax ((mp_exp_t) gfc_real_kinds[kind].max_exponent - 1);

  if (mpfr_sgn (s->value.real) > 0)
    {
+  mpfr_check_range (result->value.real, 0, GMP_RNDU);
      mpfr_nextabove (result->value.real);
      mpfr_subnormalize (result->value.real, 0, GMP_RNDU);
    }
  else
    {
+  mpfr_check_range (result->value.real, 0, GMP_RNDD);
      mpfr_nextbelow (result->value.real);
      mpfr_subnormalize (result->value.real, 0, GMP_RNDD);
    }

But I have to admit that I have not thought much about this.

Tobias