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" To: dank@kegel.com Reply-To: "add" 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 - -#include - -__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