[PATCH] sysdeps/ieee754/flt-32: Fix remainderf() sign of zero for FE_DOWNWARD

[Adhemerval Zanella Netto]

On 17/02/25 06:40, Sergei Zimmerman wrote:
> Single-precision remainderf() implementation derived from Sun is
> affected by an issue when the result is +-0. IEEE754 requires that
> if remainder(x, y) = 0, its sign shall be that of x regardless of the
> rounding direction.
> 
> The implementation seems to have assumed that x - x = +0 in all
> rounding modes, which is not the case. When rounding direction is
> roundTowardNegative the sign of an exact zero sum (or difference) is −0.
> 
> Regression tests that triggered this erroneous behavior are added to
> math/libm-test-remainder.inc.
> 
> Original fix by: Bruce Evans <bde@FreeBSD.org> in FreeBSD's
> a2ddfa5ea726c56dbf825763ad371c261b89b7c7.
> ---
>  math/libm-test-remainder.inc          | 4 ++++
>  sysdeps/ieee754/flt-32/e_remainderf.c | 3 +++
>  2 files changed, 7 insertions(+)
> 
> diff --git a/math/libm-test-remainder.inc b/math/libm-test-remainder.inc
> index 1a1d3e9937..f2b0104087 100644
> --- a/math/libm-test-remainder.inc
> +++ b/math/libm-test-remainder.inc
> @@ -173,6 +173,10 @@ static const struct test_ff_f_data remainder_test_data[] =
>      TEST_ff_f (remainder, 2, -1, plus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
>      TEST_ff_f (remainder, -2, 1, minus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
>      TEST_ff_f (remainder, -2, -1, minus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
> +
> +    TEST_ff_f (remainder, 0x1.08001cp-2, 0x1p-24, plus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
> +    TEST_ff_f (remainder, -0x1.003ffep-126, -0x1.8p-148, minus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
> +    TEST_ff_f (remainder, 0x1.2c3224p+17, 0x1p-5, plus_zero, NO_INEXACT_EXCEPTION|ERRNO_UNCHANGED),
>    };
>  
>  static void

This new test added a regression for other implementation (dbl-128
and ldbl-128ibm), and I have opened a bug report to track it [1]

I think we will need to either fix on all implementation (which
should be easy), or just enable the tests for float.

[1] https://sourceware.org/bugzilla/show_bug.cgi?id=32711 

> diff --git a/sysdeps/ieee754/flt-32/e_remainderf.c b/sysdeps/ieee754/flt-32/e_remainderf.c
> index 81781363af..7d773ab36d 100644
> --- a/sysdeps/ieee754/flt-32/e_remainderf.c
> +++ b/sysdeps/ieee754/flt-32/e_remainderf.c
> @@ -56,6 +56,9 @@ __ieee754_remainderf(float x, float p)
>  	    }
>  	}
>  	GET_FLOAT_WORD(hx,x);
> +	/* Make sure x is not -0. This can occur only when x = p
> +	   and rounding direction is towards negative infinity. */
> +	if ((hx&0x7fffffff)==0) hx = 0;
>  	SET_FLOAT_WORD(x,hx^sx);
>  	return x;
>  }