Siddhardth, the NVIDIA hardware has only four counters for profilers, which is why in his email Andreas specified "For temp_cuda_profiler.conf, choose ***four*** from...". Just put four from the list in the .conf and run your PyCUDA app :)
On Sat, Mar 28, 2009 at 2:36 PM, Siddhardh chandra <[email protected]> wrote: > Andreas Klöckner wrote: >> >> The CUDA Profiler is supported just like in regular CUDA. >> >> Shell commands: >> >> export CUDA_PROFILE=1 >> export CUDA_PROFILE_CONFIG='temp_cuda_profiler.conf' >> >> For temp_cuda_profiler.conf, choose four from: >> >> gld_incoherent >> gld_coherent >> gst_incoherent >> gst_coherent >> warp_serialize >> divergent_branch >> branch >> >> Andreas >> >> _______________________________________________ >> PyCuda mailing list >> [email protected] >> http://tiker.net/mailman/listinfo/pycuda_tiker.net >> >> > > I tried this, but the log file I get is empty with only the headings in it( > method,gputime,cputime,occupancy,gld_coherent,gld_incoherent,gst_coherent,gst_incoherent). > Is there something else that is needed to be done? > > > _______________________________________________ > PyCuda mailing list > [email protected] > http://tiker.net/mailman/listinfo/pycuda_tiker.net > _______________________________________________ PyCuda mailing list [email protected] http://tiker.net/mailman/listinfo/pycuda_tiker.net
