Taken from http://sources.redhat.com/ml/crossgcc/200402/msg00104.html
Author: addsub@eyou.com
Target: ARM
Fixes http://gcc.gnu.org/cgibin/cvsweb.cgi/gcc/gcc/testsuite/gcc.ctorture/execute/9205018.c
and makes printf("%f", 1.0) work.
Lennert Buytenhek wrote in http://sources.redhat.com/ml/crossgcc/200409/msg00115.html :
It ... fixes the 'printf("%f\n", 0.5); prints 0.000000' and general 'floating point
is broken' on my bigendian hardfloat FPA ARM platform. ...
It's definitely needed for hardfloat. So I'd think it's needed for
bigendian systems in any case, and for VFP on littleendian systems
too. Someone would have to verify that though.
Lennert Buytenhek wrote in http://sources.redhat.com/ml/crossgcc/200409/msg00123.html
I just had a look at glibc20040830, and [this patch] is still needed and useful
for this version. glibc20040830 outofthebox 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/gmpmparam.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/200409/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/gmpmparam.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/200405/msg00245.html)
MessageID: <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
MessageID: <20040217104238.8237.qmail@eyou.com>
From: "add"
To: dank@kegel.com
ReplyTo: "add"
Subject: Re: problem while building arm vfp softfloat gcc `
Hi, Dan, This is a patch I applied to my glibc2.3.2, 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 19990412 11:59:13.000000000 0400
+++ glibc.new/sysdeps/arm/bits/endian.h 20040212 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/gmpmparam.h glibc.new/sysdeps/arm/gmpmparam.h
 glibc.old/sysdeps/arm/gmpmparam.h 20010707 15:21:19.000000000 0400
+++ glibc.new/sysdeps/arm/gmpmparam.h 20040212 09:15:13.000000000 0500
@@ 26,5 +26,13 @@ MA 021111307, 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 20010707 15:21:19.000000000 0400
+++ glibc.new/sysdeps/arm/ieee754.h 19691231 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
 021111307 USA. */

#ifndef _IEEE754_H

#define _IEEE754_H 1
#include

#include

__BEGIN_DECLS

union ieee754_float
 {
 float f;

 /* This is the IEEE 754 singleprecision 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 doubleprecision 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 doubleextendedprecision 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 doubleextendedprecision 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 */