Taken from http://sources.redhat.com/ml/crossgcc/2004-02/msg00104.html
Author: addsub@eyou.com
Target: ARM
Fixes http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.c-torture/execute/920501-8.c
and makes printf("%f", 1.0) work.
Lennert Buytenhek wrote in http://sources.redhat.com/ml/crossgcc/2004-09/msg00115.html :
It ... fixes the 'printf("%f\n", 0.5); prints 0.000000' and general 'floating point
is broken' on my big-endian hardfloat FPA ARM platform. ...
It's definitely needed for hardfloat. So I'd think it's needed for
big-endian systems in any case, and for VFP on little-endian systems
too. Someone would have to verify that though.
Lennert Buytenhek wrote in http://sources.redhat.com/ml/crossgcc/2004-09/msg00123.html
I just had a look at glibc-20040830, and [this patch] is still needed and useful
for this version. glibc-20040830 out-of-the-box still contains the
following wrong assumptions:
- sysdeps/arm/bits/endian.h: float word order is big endian (which it is
not on vfp systems)
- sysdeps/arm/gmp-mparam.h: IEEE doubles are mixed endian (which they
are not on big endian systems, neither on vfp systems)
- sysdeps/arm/ieee754.h: IEEE doubles are in little endian byte order
(which they are not on big endian systems)
[This patch] seems the right solution for all of these issues.
Dimitry Andric wrote in http://sources.redhat.com/ml/crossgcc/2004-09/msg00132.html :
It's even needed for glibc CVS, AFAICS.
The patch hunk which modifies glibc.new/sysdeps/arm/bits/endian.h
(currently at version 1.4) is only needed for proper VFP operation.
But the hunk which modifies sysdeps/arm/gmp-mparam.h, and the hunk
that deletes sysdeps/arm/ieee754.h (yes, this IS correct), are needed
for proper operation of *any* FP model on big endian ARM.
See also discussion in followups to
http://sources.redhat.com/ml/crossgcc/2004-05/msg00245.html)
Message-ID: <276985760.37584@eyou.com>
Received: from unknown (HELO eyou.com) (172.16.2.2)
by 0.0.0.0 with SMTP; Tue, 17 Feb 2004 10:42:40 +0800
Received: (qmail 8238 invoked by uid 65534); 17 Feb 2004 10:42:38 +0800
Date: 17 Feb 2004 10:42:38 +0800
Message-ID: <20040217104238.8237.qmail@eyou.com>
From: "add" <addsub@eyou.com>
To: dank@kegel.com
Reply-To: "add" <addsub@eyou.com>
Subject: Re: problem while building arm vfp softfloat gcc `
Hi, Dan, This is a patch I applied to my glibc-2.3.2, then my softfloat
toolchain can printf("%f\n",1.0). So you may have a try of this
diff -uNrp glibc.old/sysdeps/arm/bits/endian.h glibc.new/sysdeps/arm/bits/endian.h
--- glibc.old/sysdeps/arm/bits/endian.h 1999-04-12 11:59:13.000000000 -0400
+++ glibc.new/sysdeps/arm/bits/endian.h 2004-02-12 09:15:13.000000000 -0500
@@ -9,4 +9,9 @@
#else
#define __BYTE_ORDER __LITTLE_ENDIAN
#endif
+
+#ifdef __VFP_FP__
+#define __FLOAT_WORD_ORDER __BYTE_ORDER
+#else
#define __FLOAT_WORD_ORDER __BIG_ENDIAN
+#endif
diff -uNrp glibc.old/sysdeps/arm/gmp-mparam.h glibc.new/sysdeps/arm/gmp-mparam.h
--- glibc.old/sysdeps/arm/gmp-mparam.h 2001-07-07 15:21:19.000000000 -0400
+++ glibc.new/sysdeps/arm/gmp-mparam.h 2004-02-12 09:15:13.000000000 -0500
@@ -26,5 +26,13 @@ MA 02111-1307, USA. */
#define BITS_PER_SHORTINT 16
#define BITS_PER_CHAR 8
-#define IEEE_DOUBLE_BIG_ENDIAN 0
-#define IEEE_DOUBLE_MIXED_ENDIAN 1
+#if defined(__ARMEB__)
+# define IEEE_DOUBLE_MIXED_ENDIAN 0
+# define IEEE_DOUBLE_BIG_ENDIAN 1
+#elif defined(__VFP_FP__)
+# define IEEE_DOUBLE_MIXED_ENDIAN 0
+# define IEEE_DOUBLE_BIG_ENDIAN 0
+#else
+# define IEEE_DOUBLE_BIG_ENDIAN 0
+# define IEEE_DOUBLE_MIXED_ENDIAN 1
+#endif
diff -uNrp glibc.old/sysdeps/arm/ieee754.h glibc.new/sysdeps/arm/ieee754.h
--- glibc.old/sysdeps/arm/ieee754.h 2001-07-07 15:21:19.000000000 -0400
+++ glibc.new/sysdeps/arm/ieee754.h 1969-12-31 19:00:00.000000000 -0500
@@ -1,115 +0,0 @@
-/* Copyright (C) 1992, 1995, 1996, 1998 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, write to the Free
- Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
- 02111-1307 USA. */
-
-#ifndef _IEEE754_H
-
-#define _IEEE754_H 1
-#include <features.h>
-
-#include <endian.h>
-
-__BEGIN_DECLS
-
-union ieee754_float
- {
- float f;
-
- /* This is the IEEE 754 single-precision format. */
- struct
- {
- unsigned int mantissa:23;
- unsigned int exponent:8;
- unsigned int negative:1;
- } ieee;
-
- /* This format makes it easier to see if a NaN is a signalling NaN. */
- struct
- {
- unsigned int mantissa:22;
- unsigned int quiet_nan:1;
- unsigned int exponent:8;
- unsigned int negative:1;
- } ieee_nan;
- };
-
-#define IEEE754_FLOAT_BIAS 0x7f /* Added to exponent. */
-
-
-union ieee754_double
- {
- double d;
-
- /* This is the IEEE 754 double-precision format. */
- struct
- {
- unsigned int mantissa0:20;
- unsigned int exponent:11;
- unsigned int negative:1;
- unsigned int mantissa1:32;
- } ieee;
-
- /* This format makes it easier to see if a NaN is a signalling NaN. */
- struct
- {
- unsigned int mantissa0:19;
- unsigned int quiet_nan:1;
- unsigned int exponent:11;
- unsigned int negative:1;
- unsigned int mantissa1:32;
- } ieee_nan;
- };
-
-#define IEEE754_DOUBLE_BIAS 0x3ff /* Added to exponent. */
-
-
-/* The following two structures are correct for `new' floating point systems but
- wrong for the old FPPC. The only solution seems to be to avoid their use on
- old hardware. */
-
-union ieee854_long_double
- {
- long double d;
-
- /* This is the IEEE 854 double-extended-precision format. */
- struct
- {
- unsigned int exponent:15;
- unsigned int empty:16;
- unsigned int negative:1;
- unsigned int mantissa1:32;
- unsigned int mantissa0:32;
- } ieee;
-
- /* This is for NaNs in the IEEE 854 double-extended-precision format. */
- struct
- {
- unsigned int exponent:15;
- unsigned int empty:16;
- unsigned int negative:1;
- unsigned int mantissa1:32;
- unsigned int mantissa0:30;
- unsigned int quiet_nan:1;
- unsigned int one:1;
- } ieee_nan;
- };
-
-#define IEEE854_LONG_DOUBLE_BIAS 0x3fff
-
-__END_DECLS
-
-#endif /* ieee754.h */
Signed-off-by: Robert P. J. Day <rpjday@mindspring.com>