Pawel Salek wrote:
> I have got a problem: I have tried to run old large Fortran code
> (compiled with fort -fast) and have noticed that it is unexpectedly slow.
> Closer investigation revealed that it spends more than 50% of CPU time in
> kernel. I suspect that the program does some kind of ugly floating point
> operations that require kernel support. The final numbers appear to be
> correct (compared against other architecture).
It sounds like FP completion could be the culprit.
If you can disable ieee floating-point (I don't know how to do this with the
Fortran compiler) your code will trap on the offending lines, so you can
identify and fix them.
> Does anybody has ANY advice how to approach this problem? Is there any
> way to find out the offending part of code? I thought I could run gprof
> but I am not sure it show up in the output time spent in kernel..
grof won't tell you anything about kernel execution. However iprobe
(http://www.alphalinux.org/iprobe/) can profile user/kernel modes at once, if
that's what you're after.
-- Jeff Sturm jeff.sturm@commerceone.com_______________________________________________ Axp-list mailing list Axp-list@redhat.com https://listman.redhat.com/mailman/listinfo/axp-list
This archive was generated by hypermail version 2a22 on Fri Dec 1 08:00:05 2000 PST
Send any problems or questions about this archive to webmaster@alphalinux.org.