From: Michael Meissner <meissner@linux.ibm.com>
To: Michael Meissner <meissner@linux.ibm.com>,
gcc-patches@gcc.gnu.org,
Segher Boessenkool <segher@kernel.crashing.org>,
David Edelsohn <dje.gcc@gmail.com>,
Bill Schmidt <wschmidt@linux.ibm.com>,
Peter Bergner <bergner@linux.ibm.com>, Jeff Law <law@redhat.com>,
Jonathan Wakely <jwakely@redhat.com>
Subject: [PATCH 3/9] PowerPC: Update IEEE <-> IBM 128-bit floating point conversions.
Date: Thu, 24 Sep 2020 16:34:35 -0400 [thread overview]
Message-ID: <20200924203435.GC31597@ibm-toto.the-meissners.org> (raw)
In-Reply-To: <20200924202036.GA28437@ibm-toto.the-meissners.org>
PowerPC: Update IEEE <-> IBM 128-bit floating point conversions.
This patch changes the code for doing conversions between IEEE 128-bit floating
point and IBM 128-bit extended double floating point. It moves the conversion
functions to a separate file. It uses explicit __ibm128 instead of long
double to allow the long double type to be set to IEEE 128-bit.
libgcc/
2020-09-23 Michael Meissner <meissner@linux.ibm.com>
* config/rs6000/extendkftf2-sw.c: Move __float128 to __ibm128
conversion into float128-convert.h.
* config/rs6000/float128-convert.h: New file.
* config/rs6000/float128-hw.c: Move conversions between __float128
and __ibm128 into float128-convert.h.
* config/rs6000/quad-float128.h: Move conversions between
__float128 and __ibm128 into float128-convert.h.
* config/rs6000/trunctfkf2-sw.c: Move __ibm128 to __float128
conversion to float128-convert.h.
---
libgcc/config/rs6000/extendkftf2-sw.c | 6 +-
libgcc/config/rs6000/float128-convert.h | 77 +++++++++++++++++++++++++
libgcc/config/rs6000/float128-hw.c | 11 +---
libgcc/config/rs6000/quad-float128.h | 58 -------------------
libgcc/config/rs6000/trunctfkf2-sw.c | 6 +-
5 files changed, 84 insertions(+), 74 deletions(-)
create mode 100644 libgcc/config/rs6000/float128-convert.h
diff --git a/libgcc/config/rs6000/extendkftf2-sw.c b/libgcc/config/rs6000/extendkftf2-sw.c
index f0de1784c43..80b48c20d9c 100644
--- a/libgcc/config/rs6000/extendkftf2-sw.c
+++ b/libgcc/config/rs6000/extendkftf2-sw.c
@@ -38,6 +38,7 @@
#include "soft-fp.h"
#include "quad-float128.h"
+#include "float128-convert.h"
#ifndef FLOAT128_HW_INSNS
#define __extendkftf2_sw __extendkftf2
@@ -46,8 +47,5 @@
IBM128_TYPE
__extendkftf2_sw (TFtype value)
{
- IBM128_TYPE ret;
-
- CVT_FLOAT128_TO_IBM128 (ret, value);
- return ret;
+ return convert_float128_to_ibm128 (value);
}
diff --git a/libgcc/config/rs6000/float128-convert.h b/libgcc/config/rs6000/float128-convert.h
new file mode 100644
index 00000000000..bb6b3d71889
--- /dev/null
+++ b/libgcc/config/rs6000/float128-convert.h
@@ -0,0 +1,77 @@
+/* Convert between IEEE 128-bit and IBM 128-bit floating point types.
+ Copyright (C) 2016-2020 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+ Contributed by Michael Meissner (meissner@linux.ibm.com).
+
+ 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.
+
+ In addition to the permissions in the GNU Lesser General Public
+ License, the Free Software Foundation gives you unlimited
+ permission to link the compiled version of this file into
+ combinations with other programs, and to distribute those
+ combinations without any restriction coming from the use of this
+ file. (The Lesser General Public License restrictions do apply in
+ other respects; for example, they cover modification of the file,
+ and distribution when not linked into a combine executable.)
+
+ 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
+ <http://www.gnu.org/licenses/>. */
+
+/* Implementation of conversions between __ibm128 and __float128, to allow the
+ same code to be used on systems with IEEE 128-bit emulation and with IEEE
+ 128-bit hardware support.
+
+ These functions are called by the actual conversion functions called by the
+ compiler. This code is here to allow being built at power8 (no hardware
+ float128) and power9 (hardware float128) varients that is selected by an
+ IFUNC function. */
+
+static inline __ibm128 convert_float128_to_ibm128 (__float128);
+static inline __float128 convert_ibm128_to_float128 (__ibm128);
+
+static inline __ibm128
+convert_float128_to_ibm128 (__float128 value)
+{
+ double high, high_temp, low;
+
+ high = (double) value;
+ if (__builtin_isnan (high) || __builtin_isinf (high))
+ low = 0.0;
+
+ else
+ {
+ low = (double) (value - (__float128) high);
+ /* Renormalize low/high and move them into canonical IBM long
+ double form. */
+ high_temp = high + low;
+ low = (high - high_temp) + low;
+ high = high_temp;
+ }
+
+ return __builtin_pack_ibm128 (high, low);
+}
+
+static inline __float128
+convert_ibm128_to_float128 (__ibm128 value)
+{
+ double high = __builtin_unpack_ibm128 (value, 0);
+ double low = __builtin_unpack_ibm128 (value, 1);
+
+ /* Handle the special cases of NAN and infinity. Similarly, if low is 0.0,
+ there no need to do the add. In addition, avoiding the add produces the
+ correct sign if high is -0.0. */
+ if (__builtin_isnan (high) || __builtin_isinf (high) || low == 0.0)
+ return (__float128) high;
+
+ else
+ return ((__float128) high) + ((__float128) low);
+}
diff --git a/libgcc/config/rs6000/float128-hw.c b/libgcc/config/rs6000/float128-hw.c
index 8705b53e22a..b9b04cd2feb 100644
--- a/libgcc/config/rs6000/float128-hw.c
+++ b/libgcc/config/rs6000/float128-hw.c
@@ -33,6 +33,7 @@
#include <soft-fp.h>
#include <quad-float128.h>
+#include "float128-convert.h"
#ifndef __FLOAT128_HARDWARE__
#error "This module must be compiled with IEEE 128-bit hardware support"
@@ -190,18 +191,12 @@ __unordkf2_hw (TFtype a, TFtype b)
IBM128_TYPE
__extendkftf2_hw (TFtype value)
{
- IBM128_TYPE ret;
-
- CVT_FLOAT128_TO_IBM128 (ret, value);
- return ret;
+ return convert_float128_to_ibm128 (value);
}
/* Convert __ibm128 to __float128. */
TFtype
__trunctfkf2_hw (IBM128_TYPE value)
{
- TFtype ret;
-
- CVT_IBM128_TO_FLOAT128 (ret, value);
- return ret;
+ return convert_ibm128_to_float128 (value);
}
diff --git a/libgcc/config/rs6000/quad-float128.h b/libgcc/config/rs6000/quad-float128.h
index 32ef328a8ea..15880c2ab30 100644
--- a/libgcc/config/rs6000/quad-float128.h
+++ b/libgcc/config/rs6000/quad-float128.h
@@ -170,61 +170,3 @@ extern TFtype __trunctfkf2 (IBM128_TYPE);
/* Complex __float128 built on __float128 interfaces. */
extern TCtype __mulkc3 (TFtype, TFtype, TFtype, TFtype);
extern TCtype __divkc3 (TFtype, TFtype, TFtype, TFtype);
-
-/* Implementation of conversions between __ibm128 and __float128, to allow the
- same code to be used on systems with IEEE 128-bit emulation and with IEEE
- 128-bit hardware support. */
-
-union ibm128_union {
- IBM128_TYPE ibm128;
- double dbl[2];
-};
-
-#define CVT_FLOAT128_TO_IBM128(RESULT, VALUE) \
-{ \
- double __high, __low; \
- TFtype __value = (VALUE); \
- union ibm128_union u; \
- \
- __high = (double) __value; \
- if (__builtin_isnan (__high) || __builtin_isinf (__high)) \
- __low = 0.0; \
- \
- else \
- { \
- double __high_temp; \
- \
- __low = (double) (__value - (TFtype) __high); \
- /* Renormalize low/high and move them into canonical IBM long \
- double form. */ \
- __high_temp = __high + __low; \
- __low = (__high - __high_temp) + __low; \
- __high = __high_temp; \
- } \
- \
- u.dbl[0] = __high; \
- u.dbl[1] = __low; \
- RESULT = u.ibm128; \
-}
-
-#define CVT_IBM128_TO_FLOAT128(RESULT, VALUE) \
-{ \
- union ibm128_union u; \
- double __high, __low; \
- \
- u.ibm128 = (VALUE); \
- __high = u.dbl[0]; \
- __low = u.dbl[1]; \
- \
- /* Handle the special cases of NAN and infinity. */ \
- if (__builtin_isnan (__high) || __builtin_isinf (__high)) \
- RESULT = (TFtype) __high; \
- \
- /* If low is 0.0, there no need to do the add. In addition, \
- avoiding the add produces the correct sign if high is -0.0. */ \
- else if (__low == 0.0) \
- RESULT = (TFtype) __high; \
- \
- else \
- RESULT = ((TFtype) __high) + ((TFtype) __low); \
-}
diff --git a/libgcc/config/rs6000/trunctfkf2-sw.c b/libgcc/config/rs6000/trunctfkf2-sw.c
index 3e8af25a585..af52cff5954 100644
--- a/libgcc/config/rs6000/trunctfkf2-sw.c
+++ b/libgcc/config/rs6000/trunctfkf2-sw.c
@@ -38,6 +38,7 @@
#include "soft-fp.h"
#include "quad-float128.h"
+#include "float128-convert.h"
#ifndef FLOAT128_HW_INSNS
#define __trunctfkf2_sw __trunctfkf2
@@ -46,8 +47,5 @@
TFtype
__trunctfkf2_sw (IBM128_TYPE value)
{
- TFtype ret;
-
- CVT_IBM128_TO_FLOAT128 (ret, value);
- return ret;
+ return convert_ibm128_to_float128 (value);
}
--
2.22.0
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.ibm.com, phone: +1 (978) 899-4797
next prev parent reply other threads:[~2020-09-24 20:34 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-09-24 20:20 [PATCH 0/9] PowerPC: Patches to enable changing the long double default to IEEE 128-bit on little endian PowerPC 64-bit Linux systems Michael Meissner
2020-09-24 20:31 ` [PATCH 1/9] PowerPC: Map long double built-in functions if IEEE 128-bit long double Michael Meissner
2020-10-01 23:05 ` Joseph Myers
2020-10-08 5:20 ` Michael Meissner
2020-09-24 20:33 ` [PATCH 2/9] PowerPC: Update __float128 and __ibm128 error messages Michael Meissner
2020-09-24 20:34 ` Michael Meissner [this message]
2020-09-24 20:35 ` [PATCH 4/9] PowerPC: Add IEEE 128-bit <-> Decimal conversions Michael Meissner
2020-09-24 20:36 ` [PATCH 5/9] PowerPC: Update tests to run if long double is IEEE 128-bit Michael Meissner
2020-09-24 20:40 ` [PATCH 6/9] PowerPC: If long double is IEEE 128-bit, map q built-ins to *l instead of *f128 Michael Meissner
2020-09-24 20:42 ` [PATCH 7/9] PowerPC: Update IEEE 128-bit built-in functions to work if long double is IEEE 128-bit Michael Meissner
2020-09-24 20:45 ` [PATCH 8/9] PowerPC: Change tests to use __float128 instead of __ieee128 Michael Meissner
2020-09-24 20:47 ` [PATCH 9/9] PowerPC: Use __builtin_pack_ieee128 if long double is IEEE 128-bit Michael Meissner
2020-10-09 4:35 ` [PATCH 1/9, revised] PowerPC: Map long double built-in functions if IEEE 128-bit long double Michael Meissner
2020-10-16 20:30 ` Segher Boessenkool
2020-10-19 19:30 ` Michael Meissner
2020-10-19 21:24 ` Michael Meissner
2020-10-12 23:16 ` Ping: [PATCH 0/9] PowerPC: Patches to enable changing the long double default to IEEE 128-bit on little endian PowerPC 64-bit Linux systems Michael Meissner
2020-10-16 14:37 ` Ping #2: " Michael Meissner
2020-10-16 19:14 ` Segher Boessenkool
2020-10-19 19:12 ` Michael Meissner
2020-10-19 21:50 ` Segher Boessenkool
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=20200924203435.GC31597@ibm-toto.the-meissners.org \
--to=meissner@linux.ibm.com \
--cc=bergner@linux.ibm.com \
--cc=dje.gcc@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jwakely@redhat.com \
--cc=law@redhat.com \
--cc=segher@kernel.crashing.org \
--cc=wschmidt@linux.ibm.com \
/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).