public inbox for libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
To: Siddhesh Poyarekar <siddhesh@sourceware.org>, libc-alpha@sourceware.org
Cc: fweimer@redhat.com, joseph@codesourcery.com
Subject: Re: [PATCH v2 4/5] x86 long double: Consider pseudo numbers as signaling
Date: Wed, 23 Dec 2020 15:14:37 -0300	[thread overview]
Message-ID: <c4b0ba22-832d-2fec-b61e-7dd823b080d1@linaro.org> (raw)
In-Reply-To: <20201223115602.3472627-5-siddhesh@sourceware.org>



On 23/12/2020 08:56, Siddhesh Poyarekar wrote:
> Add support to treat pseudo-numbers specially and implement x86
> version to consider all of them as signaling.
> ---
>  sysdeps/generic/nan-pseudo-number.h      | 27 ++++++++++++++++++++
>  sysdeps/ieee754/ldbl-96/s_issignalingl.c |  7 +++++-
>  sysdeps/x86/fpu/nan-pseudo-number.h      | 32 ++++++++++++++++++++++++
>  3 files changed, 65 insertions(+), 1 deletion(-)
>  create mode 100644 sysdeps/generic/nan-pseudo-number.h
>  create mode 100644 sysdeps/x86/fpu/nan-pseudo-number.h
> 
> diff --git a/sysdeps/generic/nan-pseudo-number.h b/sysdeps/generic/nan-pseudo-number.h
> new file mode 100644
> index 0000000000..ce1176a208
> --- /dev/null
> +++ b/sysdeps/generic/nan-pseudo-number.h
> @@ -0,0 +1,27 @@
> +/* Pseudo-normal number handling.  Generic 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/>.  */
> +
> +#ifndef NAN_PSEUDO_NUMBER_H
> +#define NAN_PSEUDO_NUMBER_H	1
> +
> +/* Define this macro to 1 and provide an definition for IS_PSEUDO_SIGNALING if
> +   pseudo-normal numbers need special handling.  */
> +#define HANDLE_PSEUDO_NUMBERS 0
> +#define IS_PSEUDO_SIGNALING(x) 0
> +
> +#endif /* nan-pseudo-number.h */
> diff --git a/sysdeps/ieee754/ldbl-96/s_issignalingl.c b/sysdeps/ieee754/ldbl-96/s_issignalingl.c
> index ec542ad468..fff6a85932 100644
> --- a/sysdeps/ieee754/ldbl-96/s_issignalingl.c
> +++ b/sysdeps/ieee754/ldbl-96/s_issignalingl.c
> @@ -19,12 +19,14 @@
>  #include <math.h>
>  #include <math_private.h>
>  #include <nan-high-order-bit.h>
> +#include <nan-pseudo-number.h>
>  
>  int
>  __issignalingl (long double x)
>  {
>    uint32_t exi, hxi, lxi;
>    GET_LDOUBLE_WORDS (exi, hxi, lxi, x);
> +  int ret = 0;
>  #if HIGH_ORDER_BIT_IS_SET_FOR_SNAN
>  # error not implemented
>  #else
> @@ -36,9 +38,12 @@ __issignalingl (long double x)
>    hxi |= (lxi | -lxi) >> 31;
>    /* We do not recognize a pseudo NaN as sNaN; they're invalid on 80387 and
>       later.  */
> +#if HANDLE_PSEUDO_NUMBERS
> +  ret = IS_PSEUDO_SIGNALING(x);
> +#endif
>    /* 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);
> +  return ret || (((exi & 0x7fff) == 0x7fff) && (hxi > 0xc0000000));
>  #endif
>  }
>  libm_hidden_def (__issignalingl)
> diff --git a/sysdeps/x86/fpu/nan-pseudo-number.h b/sysdeps/x86/fpu/nan-pseudo-number.h
> new file mode 100644
> index 0000000000..9cc37a3eee
> --- /dev/null
> +++ b/sysdeps/x86/fpu/nan-pseudo-number.h
> @@ -0,0 +1,32 @@
> +/* Pseudo-normal number handling.  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/>.  */
> +
> +#ifndef NAN_PSEUDO_NUMBER_H
> +#define NAN_PSEUDO_NUMBER_H	1
> +
> +#define HANDLE_PSEUDO_NUMBERS 1
> +/* A pseudo-normal number (that is not pseudo-denormal) has a non-zero exponent
> +   and the MSB of significand is clear.  All pseudo-normal numbers are deemed
> +   signaling.  */
> +#define IS_PSEUDO_SIGNALING(x) ({					      \
> +  uint32_t exi, hxi, lxi __attribute__ ((unused));			      \
> +  GET_LDOUBLE_WORDS (exi, hxi, lxi, x);					      \
> +  ((exi & 0x7fff) && ((hxi & 0x80000000) == 0));			      \
> +})
> +
> +#endif /* nan-pseudo-number.h */
> 

Maybe a static inline function would be better? Something like:

  static inline int pseudo_signaling (uint32_t exi, uint32_t hxi)
  {
    return (exi & 0x7fff) && (hxi & 0x80000000) == 0;
  }

  reply	other threads:[~2020-12-23 18:14 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-12-23 11:55 [PATCH v2 0/5] x86 pseudo-normal numbers Siddhesh Poyarekar
2020-12-23 11:55 ` [PATCH v2 1/5] x86 long double: Support pseudo numbers in fpclassifyl Siddhesh Poyarekar
2020-12-23 17:59   ` Adhemerval Zanella
2020-12-23 11:55 ` [PATCH v2 2/5] x86 long double: Support pseudo numbers in isnanl Siddhesh Poyarekar
2020-12-23 18:03   ` Adhemerval Zanella
2020-12-23 11:56 ` [PATCH v2 3/5] Partially revert 681900d29683722b1cb0a8e565a0585846ec5a61 Siddhesh Poyarekar
2020-12-23 12:09   ` Florian Weimer
2020-12-23 11:56 ` [PATCH v2 4/5] x86 long double: Consider pseudo numbers as signaling Siddhesh Poyarekar
2020-12-23 18:14   ` Adhemerval Zanella [this message]
2020-12-23 11:56 ` [PATCH v2 5/5] x86 long double: Add tests for pseudo normal numbers Siddhesh Poyarekar

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=c4b0ba22-832d-2fec-b61e-7dd823b080d1@linaro.org \
    --to=adhemerval.zanella@linaro.org \
    --cc=fweimer@redhat.com \
    --cc=joseph@codesourcery.com \
    --cc=libc-alpha@sourceware.org \
    --cc=siddhesh@sourceware.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).