This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Intermediate expression not rounded to float on x86


Summary: 
    gcc is not doing strict IEEE arithmetic for x86.  
    The various IEEE options do not help.

Here's the input program ("t.C"):
----------------------------------------------------------------------------
--
#include <stdio.h>
#include <math.h>

int   n = 200;
float f = 0.20000000298023224;    
float g;

int main () 
{

    int v1, v2;
    g = f*n;                    /* Put product in global float */
    v1 = (int) ceil(f*n);       /* Compute ceiling of product  */
    v2 = (int) ceil(g);
    if( v1==v2 ) {
        printf("Okay: v1=%d=v2\n", v1 );
    } else {
        printf("Bad: v1=%d!=%d=v2\n", v1, v2 );
    }
    return 0;
}
----------------------------------------------------------------------------
--
Here's a sample session, which shows the OS version (Red Hat Linux 6.2),
gcc version, options that I used, and the output that I got.
----------------------------------------------------------------------------
--
$ uname -a
Linux brick 2.2.14-5.0smp #1 SMP Tue Mar 7 21:01:40 EST 2000 i686 unknown
$ gcc -v -ieee -ffloat-store t.C -lm
Reading specs from /usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66/specs
gcc version egcs-2.91.66 19990314/Linux (egcs-1.1.2 release)
 /usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66/cpp -lang-c++ -v -undef
-D__GNUC__=2 -D__GNUG__=2 -D__cplusplus -D__GNUC_MINOR__=91 -D__ELF__ -Dunix
-Di386 -D__i386__ -Dlinux -D__ELF__ -D__unix__ -D__i386__ -D__i386__
-D__linux__ -D__unix -D__i386 -D__linux -Asystem(posix) -D__EXCEPTIONS
-Asystem(unix) -Acpu(i386) -Amachine(i386) -Di386 -D__i386 -D__i386__
-D__tune_i386__ -ieee t.C /usr/tmp/ccMGQMOB.ii
GNU CPP version egcs-2.91.66 19990314/Linux (egcs-1.1.2 release) (i386
Linux/ELF)
#include "..." search starts here:
#include <...> search starts here:
 /usr/include/g++-2
 /usr/i386-redhat-linux/include
 /usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66/include
 /usr/include
End of search list.
 /usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66/cc1plus
/usr/tmp/ccMGQMOB.ii -quiet -dumpbase t.cc -version -ffloat-store -o
/usr/tmp/ccCquk2f.s
GNU C++ version egcs-2.91.66 19990314/Linux (egcs-1.1.2 release)
(i386-redhat-linux) compiled by GNU C version egcs-2.91.66 19990314/Linux
(egcs-1.1.2 release).
 as -V -Qy -o /usr/tmp/ccW19es2.o /usr/tmp/ccCquk2f.s
GNU assembler version 2.9.5 (i386-redhat-linux) using BFD version 2.9.5.0.22
 /usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66/collect2 -m elf_i386
-dynamic-linker /lib/ld-linux.so.2 /usr/lib/crt1.o /usr/lib/crti.o
/usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66/crtbegin.o
-L/usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66
-L/usr/i386-redhat-linux/lib /usr/tmp/ccW19es2.o -lm -lgcc -lc -lgcc
/usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66/crtend.o /usr/lib/crtn.o
$ a.out
Bad: v1=41!=40=v2
----------------------------------------------------------------------------
----------
Section 6.2.1.4 of the ISO C standard permits excessive precision for values
of
floating-point expressions, but shouldn't -ffloat-store remove the excess
precision?
This is an issue for anyone trying to get reproducible results across IEEE 
platforms.

Arch D. Robison          	Kuck & Associates Inc.
Arch.Robison@intel.com   	1906 Fox Drive
217-356-2288 ext. 56     	Champaign IL 61820




Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]