PGI User Forum

[Nate Crawford]

When running a fortran program compiled with PGF95 6.1-3 as:

pgf95 -O3 -fastsse -Mprof=func -tp amd64 -Mlarge_arrays -mcmodel=medium -Bstatic -o foobar foobar.f90

the generated pgprof.out file looks like:

PROF NODALL 0 a.out 1143462776 0
h tcholx1 9949 0 1
t 1 2
p 0
f foobar.f90
r foobar 1 1 1 nan nan 0 0
r wrtpisc 1 19 50 nan nan 0 0
z


Removing -mcmodel=medium gives a more reasonable output file:

PROF NODALL 0 a.out 1143462711 0
h tcholx1 9919 0 1
t 1 2
p 0
f foobar.f90
r foobar 1 1 1 0.0299295 0.0111448 0 0
r wrtpisc 1 19 50 0.0187847 0.0187847 0 0
z

The actual program output is identical.

About 5% of the runs of the -mcmodel=medium version will replace the "nan" with "0", or with a multiple of 0.0658832. I've tried running a profiled binary on a 20 minute test with similar results.

The user's guide for 6.1 has some contradictory statements regarding the combination of static compilation and -mcmodel=medium. Do I need to link dynamically for profiling?

Thanks,
Nate

[mleair]

Hi Nate,

Sorry about the delay in getting back to you. Profiling programs compiled with -mcmodel=medium is not formally supported yet. We hope to address this limitation in a future release.

-Mark