Richard Henderson <richard.hender...@linaro.org> writes:
> Because of FloatParts, there will only ever be one caller.
Isn't that admitting defeat - after all the logic here will be the same
as the login in the up coming float128_muladd code and we only seem to
need additional information:
> Inlining allows us to re-use abc_mask for the snan test.
couldn't we just pass the masks in?
<snip>
> - if (is_snan(a.cls)) {
> - return parts_silence_nan(a, s);
> - }
> - return a;
here.
> -}
> -
> /*
> * Returns the result of adding or subtracting the values of the
> * floating-point values `a' and `b'. The operation is performed
> @@ -1366,7 +1327,41 @@ static FloatParts muladd_floats(FloatParts a,
> FloatParts b, FloatParts c,
> * off to the target-specific pick-a-NaN routine.
> */
> if (unlikely(abc_mask & float_cmask_anynan)) {
> - return pick_nan_muladd(a, b, c, inf_zero, s);
> + int which;
> +
> + if (unlikely(abc_mask & float_cmask_snan)) {
> + float_raise(float_flag_invalid, s);
> + }
> +
> + which = pickNaNMulAdd(a.cls, b.cls, c.cls, inf_zero, s);
> +
> + if (s->default_nan_mode) {
> + /*
> + * Note that this check is after pickNaNMulAdd so that function
> + * has an opportunity to set the Invalid flag for inf_zero.
> + */
> + which = 3;
> + }
> +
> + switch (which) {
> + case 0:
> + break;
> + case 1:
> + a = b;
> + break;
> + case 2:
> + a = c;
> + break;
> + case 3:
> + return parts_default_nan(s);
> + default:
> + g_assert_not_reached();
> + }
> +
> + if (is_snan(a.cls)) {
> + return parts_silence_nan(a, s);
> + }
> + return a;
> }
>
> if (unlikely(inf_zero)) {
I'm not totally against it given it's fairly simple logic but it seems a
shame to loose the commonality of processing which makes the parts code
so much nicer.
--
Alex Bennée
