[4/5] x86 long double: Consider pseudo numbers as signaling

Implement a separate x86 version of issignalingl that returns true for
pseudo-normal numbers.

Also drop comment from the generic version of s_issignalingl since it
is no longer true or relevant.
---
 sysdeps/ieee754/ldbl-96/s_issignalingl.c |  2 --
 sysdeps/x86/fpu/s_issignalingl.c         | 39 ++++++++++++++++++++++++
 2 files changed, 39 insertions(+), 2 deletions(-)
 create mode 100644 sysdeps/x86/fpu/s_issignalingl.c

On 15/12/2020 11:13, Siddhesh Poyarekar via Libc-alpha wrote:
> Implement a separate x86 version of issignalingl that returns true for
> pseudo-normal numbers.
> 
> Also drop comment from the generic version of s_issignalingl since it
> is no longer true or relevant.
> ---
>  sysdeps/ieee754/ldbl-96/s_issignalingl.c |  2 --
>  sysdeps/x86/fpu/s_issignalingl.c         | 39 ++++++++++++++++++++++++
>  2 files changed, 39 insertions(+), 2 deletions(-)
>  create mode 100644 sysdeps/x86/fpu/s_issignalingl.c
> 
> diff --git a/sysdeps/ieee754/ldbl-96/s_issignalingl.c b/sysdeps/ieee754/ldbl-96/s_issignalingl.c
> index ec542ad468..2aa0ffae0e 100644
> --- a/sysdeps/ieee754/ldbl-96/s_issignalingl.c
> +++ b/sysdeps/ieee754/ldbl-96/s_issignalingl.c
> @@ -34,8 +34,6 @@ __issignalingl (long double x)
>    hxi ^= 0x40000000;
>    /* If lxi != 0, then set any suitable bit of the significand in hxi.  */
>    hxi |= (lxi | -lxi) >> 31;
> -  /* We do not recognize a pseudo NaN as sNaN; they're invalid on 80387 and
> -     later.  */
>    /* We have to compare for greater (instead of greater or equal), because x's
>       significand being all-zero designates infinity not NaN.  */
>    return ((exi & 0x7fff) == 0x7fff) && (hxi > 0xc0000000);
> diff --git a/sysdeps/x86/fpu/s_issignalingl.c b/sysdeps/x86/fpu/s_issignalingl.c

We have multiple internal defines that try to use the common implementation
instead of pushing for arch-specific ones (for instance nan-high-order-bit.h, 
fix-int-fp-convert-zero.h, fix-fp-int-convert-overflow.h, etc).  Maybe it
would be simpler to just add a new header (nan-handle-pseudo-number.h or
something related) and adapt the generic implementation.

> new file mode 100644
> index 0000000000..df273b881d
> --- /dev/null
> +++ b/sysdeps/x86/fpu/s_issignalingl.c
> @@ -0,0 +1,39 @@
> +/* Test for signaling NaN - x86 version.
> +   Copyright (C) 2020 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <https://www.gnu.org/licenses/>.  */
> +
> +#include <math.h>
> +#include <math_private.h>
> +#include <nan-high-order-bit.h>
> +#include "sysdeps/x86/fpu/isnanl_common.h"
> +
> +int
> +__issignalingl (long double x)
> +{
> +  uint32_t exi, hxi, lxi, nan, pn;
> +  GET_LDOUBLE_WORDS (exi, hxi, lxi, x);
> +#if HIGH_ORDER_BIT_IS_SET_FOR_SNAN
> +# error not implemented
> +#else
> +  nan = x86_isnanl (exi, hxi, lxi);
> +  pn = (~hxi & 0x80000000) >> 31;
> +  hxi = (~hxi & 0x40000000) >> 30;
> +
> +  return nan & (pn | hxi);
> +#endif
> +}
> +libm_hidden_def (__issignalingl)
>

On 12/23/20 1:43 AM, Adhemerval Zanella via Libc-alpha wrote:
> 
> 
> On 15/12/2020 11:13, Siddhesh Poyarekar via Libc-alpha wrote:
>> Implement a separate x86 version of issignalingl that returns true for
>> pseudo-normal numbers.
>>
>> Also drop comment from the generic version of s_issignalingl since it
>> is no longer true or relevant.
>> ---
>>   sysdeps/ieee754/ldbl-96/s_issignalingl.c |  2 --
>>   sysdeps/x86/fpu/s_issignalingl.c         | 39 ++++++++++++++++++++++++
>>   2 files changed, 39 insertions(+), 2 deletions(-)
>>   create mode 100644 sysdeps/x86/fpu/s_issignalingl.c
>>
>> diff --git a/sysdeps/ieee754/ldbl-96/s_issignalingl.c b/sysdeps/ieee754/ldbl-96/s_issignalingl.c
>> index ec542ad468..2aa0ffae0e 100644
>> --- a/sysdeps/ieee754/ldbl-96/s_issignalingl.c
>> +++ b/sysdeps/ieee754/ldbl-96/s_issignalingl.c
>> @@ -34,8 +34,6 @@ __issignalingl (long double x)
>>     hxi ^= 0x40000000;
>>     /* If lxi != 0, then set any suitable bit of the significand in hxi.  */
>>     hxi |= (lxi | -lxi) >> 31;
>> -  /* We do not recognize a pseudo NaN as sNaN; they're invalid on 80387 and
>> -     later.  */
>>     /* We have to compare for greater (instead of greater or equal), because x's
>>        significand being all-zero designates infinity not NaN.  */
>>     return ((exi & 0x7fff) == 0x7fff) && (hxi > 0xc0000000);
>> diff --git a/sysdeps/x86/fpu/s_issignalingl.c b/sysdeps/x86/fpu/s_issignalingl.c
> 
> We have multiple internal defines that try to use the common implementation
> instead of pushing for arch-specific ones (for instance nan-high-order-bit.h,
> fix-int-fp-convert-zero.h, fix-fp-int-convert-overflow.h, etc).  Maybe it
> would be simpler to just add a new header (nan-handle-pseudo-number.h or
> something related) and adapt the generic implementation.

OK I'll try to do that.

Thanks,
Siddhesh