Re: [PATCH 24/76] fpu: allow flushing of output denormals to be after rounding

[Richard Henderson]

On 1/24/25 08:27, Peter Maydell wrote:
> Currently we handle flushing of output denormals in uncanon_normal
> always before we deal with rounding.  This works for architectures
> that detect tininess before rounding, but is usually not the right
> place when the architecture detects tininess after rounding.  For
> example, for x86 the SDM states that the MXCSR FTZ control bit causes
> outputs to be flushed to zero "when it detects a floating-point
> underflow condition".  This means that we mustn't flush to zero if
> the input is such that after rounding it is no longer tiny.
>
> At least one of our guest architectures does underflow detection
> after rounding but flushing of denormals before rounding (MIPS MSA);

Whacky, but yes, I see that in the msa docs.

> Add an ftz_detection flag.  For consistency with
> tininess_before_rounding, we make it default to "detect ftz after
> rounding"; this means that we need to explicitly set the flag to
> "detect ftz before rounding" on every existing architecture that sets
> flush_to_zero, so that this commit has no behaviour change.
> (This means more code change here but for the long term a less
> confusing API.)

Do we really want flush_to_zero to be separate from ftz_detection?

E.g.

enum {
  float_ftz_disabled,
  float_ftz_after_rounding,
  float_ftz_before_rounding,
}

BTW, I'm not keen on your "detect_*" names, without "float_" prefix like (almost?) 
everything else.


> diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc
> index 0122b35008a..324e67de259 100644
> --- a/fpu/softfloat-parts.c.inc
> +++ b/fpu/softfloat-parts.c.inc
> @@ -334,7 +334,8 @@ static void partsN(uncanon_normal)(FloatPartsN *p, 
> float_status *s,
>               p->frac_lo &= ~round_mask;
>           }
>           frac_shr(p, frac_shift);
> -    } else if (s->flush_to_zero) {
> +    } else if (s->flush_to_zero &&
> +               s->ftz_detection == detect_ftz_before_rounding) {

else if (s->flush_to_zero == float_flush_to_zero_before_rounding)


>           flags |= float_flag_output_denormal_flushed;
>           p->cls = float_class_zero;
>           exp = 0;
> @@ -381,11 +382,19 @@ static void partsN(uncanon_normal)(FloatPartsN *p, 
> float_status *s,
>           exp = (p->frac_hi & DECOMPOSED_IMPLICIT_BIT) && !fmt->m68k_denormal;
>           frac_shr(p, frac_shift);
>   
> -        if (is_tiny && (flags & float_flag_inexact)) {
> -            flags |= float_flag_underflow;
> -        }
> -        if (exp == 0 && frac_eqz(p)) {
> -            p->cls = float_class_zero;
> +        if (is_tiny) {
> +            if (s->flush_to_zero) {
> +                assert(s->ftz_detection == detect_ftz_after_rounding);

if (s->flush_to_zero == float_flush_to_zero_after_rounding)

and no assert.

> +                flags |= float_flag_output_denormal_flushed;
> +                p->cls = float_class_zero;
> +                exp = 0;
> +                frac_clear(p);
> +            } else if (flags & float_flag_inexact) {
> +                flags |= float_flag_underflow;
> +            }
> +            if (exp == 0 && frac_eqz(p)) {
> +                p->cls = float_class_zero;
> +            }
>           }
>       }
>       p->exp = exp;