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
next prev parent 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).