public inbox for libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Siddhesh Poyarekar <siddhesh@sourceware.org>
To: libc-alpha@sourceware.org
Subject: [COMMITTED] x86 long double: Consider pseudo numbers as signaling
Date: Wed, 30 Dec 2020 11:01:26 +0530	[thread overview]
Message-ID: <20201230053126.311146-1-siddhesh@sourceware.org> (raw)
In-Reply-To: <66af299c-54d4-2163-2430-91a258afc892@linaro.org>

> Afaik, there is no need to __attribute__ ((unused)) on arguments for
> static inline functions.

You're right, this is what I finally committed.

Thanks,
Siddhesh

Add support to treat pseudo-numbers specially and implement x86
version to consider all of them as signaling.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
---
 sysdeps/generic/nan-pseudo-number.h      | 31 ++++++++++++++++++++++++
 sysdeps/ieee754/ldbl-96/s_issignalingl.c | 11 ++++++---
 sysdeps/x86/fpu/nan-pseudo-number.h      | 30 +++++++++++++++++++++++
 3 files changed, 69 insertions(+), 3 deletions(-)
 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..b39e347402
--- /dev/null
+++ b/sysdeps/generic/nan-pseudo-number.h
@@ -0,0 +1,31 @@
+/* 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
+
+/* Set this macro and override the definition of IS_PSEUDO_SIGNALING if pseudo
+   numbers need special handling.  */
+#define HANDLE_PSEUDO_NUMBERS 0
+static inline int
+is_pseudo_signaling (uint32_t exi, uint32_t hxi)
+{
+  return 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..37e05aa259 100644
--- a/sysdeps/ieee754/ldbl-96/s_issignalingl.c
+++ b/sysdeps/ieee754/ldbl-96/s_issignalingl.c
@@ -19,12 +19,19 @@
 #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);
+
+  /* By default we do not recognize a pseudo NaN as sNaN.  However on 80387 and
+     later all pseudo numbers including pseudo NaNs result in a signal and are
+     hence recognized as signaling.  */
+  int ret = is_pseudo_signaling (exi, hxi);
+
 #if HIGH_ORDER_BIT_IS_SET_FOR_SNAN
 # error not implemented
 #else
@@ -34,11 +41,9 @@ __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);
+  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..ea43b175a1
--- /dev/null
+++ b/sysdeps/x86/fpu/nan-pseudo-number.h
@@ -0,0 +1,30 @@
+/* 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
+/* Pseudo numbers on x86 are always signaling.  */
+static inline int
+is_pseudo_signaling (uint32_t exi, uint32_t hxi)
+{
+  return ((exi & 0x7fff) && ((hxi & 0x80000000) == 0));
+}
+
+#endif /* nan-pseudo-number.h */
-- 
2.29.2


  reply	other threads:[~2020-12-30  5:31 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-12-24  2:45 [PATCH v3 0/2] x86 pseudo-normal numbers Siddhesh Poyarekar
2020-12-24  2:45 ` [PATCH v3 1/2] x86 long double: Consider pseudo numbers as signaling Siddhesh Poyarekar
2020-12-28 16:57   ` Adhemerval Zanella
2020-12-30  5:31     ` Siddhesh Poyarekar [this message]
2020-12-31 22:58   ` Joseph Myers
2020-12-31 23:23     ` Andreas Schwab
2021-01-04  9:30     ` [COMMITTED] Move generic nan-pseudo-number.h to ldbl-96 Siddhesh Poyarekar
2020-12-24  2:45 ` [PATCH v3 2/2] x86 long double: Add tests for pseudo normal numbers Siddhesh Poyarekar
2020-12-28 17:04   ` Adhemerval Zanella
2020-12-31 23:09   ` Joseph Myers

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=20201230053126.311146-1-siddhesh@sourceware.org \
    --to=siddhesh@sourceware.org \
    --cc=libc-alpha@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).