[PATCH FORTRAN]: Use mpc_t for representing complex numbers

[Tobias Burnus]

Kaveh R. GHAZI wrote:
> Tested on x86_64-unknown-linux-gnu with and without MPC, no fortran
> regressions.
>
> Okay for mainline?
>   
Yes, the patch is okay; for some small remarks see below. Thanks for the
patch.


>  	{
>  	  rc = ARITH_DIV0;
>  	  break;
>  	}
>
> -      gfc_set_model (op1->value.complex.r);
> +
> +      gfc_set_model (mpc_realref (op1->value.complex))
You are adding another empty line; I think one should be enough.


>  static void
>  complex_reciprocal (gfc_expr *op)
>  {
> +  gfc_set_model (mpc_realref (op->value.complex));
> +#ifdef HAVE_mpc
> +  mpc_ui_div (op->value.complex, 1, op->value.complex, GFC_MPC_RND_MODE);
> +#else
>   
I am wondering whether instead of keeping the "complex_reciprocal"
function, one should simply put "mpc_ui_div" into "arith_power", which
is the only user of that static function. (The question is what is
better: Calling a tiny static function with an obvious name and only
"op" arguments or to have the actual call. I leave this to you)

>    /* x = base */
> +#ifdef HAVE_mpc
> +  mpfr_set (x_r, mpc_realref (base->value.complex), GFC_RND_MODE);
> +  mpfr_set (x_i, mpc_imagref (base->value.complex), GFC_RND_MODE);
> +#else
>    mpfr_set (x_r, base->value.complex.r, GFC_RND_MODE);
>    mpfr_set (x_i, base->value.complex.i, GFC_RND_MODE);
> +#endif
>   
Here, the #else block is not needed as the #if block works for both.

> @@ -931,7 +994,6 @@ complex_pow (gfc_expr *result, gfc_expr
>    mpfr_clears (x_r, x_i, tmp, re, im, NULL);
>  }
>
> -
>  /* Raise a number to a power.  */
>
>  static arith
>   
I think the extra line can stay, having two empty lines before a new
function it more readable.

>  static int
>  compare_complex (gfc_expr *op1, gfc_expr *op2)
>  {
> +#ifdef HAVE_mpc
> +  return mpc_cmp (op1->value.complex, op2->value.complex) == 0;
> +#else
>    return (mpfr_equal_p (op1->value.complex.r, op2->value.complex.r)
>  	  && mpfr_equal_p (op1->value.complex.i, op2->value.complex.i));
> +#endif
>  }
>   
This would be also a candidate for moving to the two callers, but it can
equally well stay a separate function.


> -      return encode_complex (source->ts.kind, source->value.complex.r,
> -			     source->value.complex.i, buffer, buffer_size);
> +      return encode_complex (source->ts.kind,
> +			     mpc_realref (source->value.complex),
> +			     mpc_imagref (source->value.complex),
> +			     buffer, buffer_size);
>   
I would change the prototype of encode_complex to accepting mpc_t.

Tobias