g77 in gcc3.4 20030827 miscompiles
Jack Howarth
howarth@bromo.msbb.uc.edu
Wed Oct 22 08:00:00 GMT 2003
In doing a few quick benchmarks of the current g77 3.4
from 20030827 on my dual processor G5 (fink is having
trouble autobuilding 20031015 but thats another story),
I ran across a problem with the recommended optimization
flags...
-mcpu=970 -mpowerpc64 -mtune=970 -mpowerpc-gpopt -force_cpusubtype_ALL
If I download the MDBNCH fortran benchmark and second.f timing
routine from...
http://www.fisica.uniud.it/~ercolessi/Mdbnch/info.html#sec21
and build it with...
g77 -O3 -g -mcpu=970 -mpowerpc64 -mtune=970 -mpowerpc-gpopt -force_cpusubtype_ALL A.f
I get a segfault during the run. Unfortunately the current gdb we have
in fink doesn't seem able to backtrace this code. However it seems
that the flag that is triggering the crash is -mpowerpc64. If I
recompile the code with...
g77 -O3 -g -mcpu=970 -mtune=970 -mpowerpc-gpopt -force_cpusubtype_ALL A.f
I don't get the crash when running MDBNCH.
Jack
More information about the Gcc
mailing list