An x86 bug in egcs 1.0.2?

[Byeong-ryeol Kim]

On Mon, 13 Apr 1998, H.J. Lu wrote:

> Can someone please tell me if it is an egcs 1.0.2 bug?
....
>        Did you see my example (posted in March egcs-bugs)
> 
>   No problem under glibc-2.1  Trouble with glibc-2.0.6 & glibc-2.0.7pre2
>   The problen is not numerical precision (Jim Wilson), but in-lining asm.
> 
>   gcc -o tst -O2 -mpentiumpro tst.c     Works
>   gcc -o tst -O2 -mpentium    tst.c     Fails
>   gcc -o tst -O2 -m486        tst.c       "
> 
>   Regards,
> 
>           Roger Brown
> 
....
> int main ()
> {
>        printf(" Gamma(1.2): %16.9e  %16.9e\n", gamma(1.2), exp(lgamma(1.2)));
> }
....
 I tested above code on my rh5.0 system(kernel-2.1.96pre1, 
glibc-2.0.7-6+wcsmbs-patch, egcs-1.0.2+latest-i386.md-patch+libio.patch, 
binutils-2.9.0.2). If necessary, I use ASUS P/I-P55T2P4 with 
AMD-PR2/MMX-200(stepping 1), 64M EDO RAM, 512K pipelined burst cache RAM. 
 There were problems according to optimization level. 
 In short, using '-m386', '-m486', '-mpentium' with any optimization level
yields erroneous results. Using '-mpentiumpro' and any '-O?' except '-O' 
yileds seems to yields accurate results.

results:

1) no otimization

$ gcc -o test -m486 test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01

$ gcc -o test -mpentium test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01

$ gcc -o test -mpentiumpro test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01


2) various optimization

 2-1) '-m386' & '-m486'

$ /usr/bin/gcc -O -o test -m386 test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01

$ /usr/bin/gcc -O3 -o test test.c -lm
$ ./test
 Gamma(1.2):  1.234298459e-55   9.181687424e-01

$ /usr/bin/gcc -O2 -o test test.c -lm
$ ./test
 Gamma(1.2):  1.234298459e-55   9.181687424e-01

$ /usr/bin/gcc -O9 -o test -m386 test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01

$ /usr/bin/gcc -O -o test -m486 test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01

$ /usr/bin/gcc -O2 -o test -m486 test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01

$ /usr/bin/gcc -O3 -o test -m486 test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01

$ /usr/bin/gcc -O6 -o test -m486 test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01

$ /usr/bin/gcc -O9 -o test -m486 test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01


 2-2) '-mpentium'

$ /usr/bin/gcc -O -o test -mcpu=pentium -march=pentium test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01

$ /usr/bin/gcc -O2 -o test -mcpu=pentium -march=pentium test.c -lm
$ ./test
 Gamma(1.2):  1.234298459e-55   9.181687424e-01

$ /usr/bin/gcc -O3 -o test -mcpu=pentium -march=pentium test.c -lm
$ ./test
 Gamma(1.2):  1.234298459e-55   9.181687424e-01

$ /usr/bin/gcc -O6 -o test -mcpu=pentium -march=pentium test.c -lm
$ ./test
 Gamma(1.2):  1.234298459e-55   9.181687424e-01

$ /usr/bin/gcc -O9 -o test -mcpu=pentium -march=pentium test.c -lm
$ ./test
 Gamma(1.2):  1.234298459e-55   9.181687424e-01


 2-3) '-mpentiumpro'

$ /usr/bin/gcc -O -o test -mcpu=pentiumpro -march=pentiumpro test.c -lm
$ ./test
 Gamma(1.2):  6.235220524e+00   9.181687424e-01

$ /usr/bin/gcc -O2 -o test -mcpu=pentiumpro -march=pentiumpro test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01

$ /usr/bin/gcc -O3 -o test -mcpu=pentiumpro -march=pentiumpro test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01

$ /usr/bin/gcc -O4 -o test -mcpu=pentiumpro -march=pentiumpro test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01

$ /usr/bin/gcc -O6 -o test -mcpu=pentiumpro -march=pentiumpro test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01

$ /usr/bin/gcc -O9 -o test -mcpu=pentiumpro -march=pentiumpro test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01

$ /usr/bin/gcc -O2 -o test -mcpu=pentiumpro -march=pentiumpro test.c -lm
$ ./test
 Gamma(1.2):  9.181687424e-01   9.181687424e-01

 "Where there is a will, there is a way."  jinbo21@soback.kornet.nm.kr 
                                           kbeyl@kids.kotel.co.kr      
  For the future of you and me!            hitel: jinbo21