I know there was a way to setup an environment variable or a text file which told the CUDA driver to profile the code, but I can’t seem to remember how that worked. If anyone could point me in the rignt direction, I would be grateful. I know that the visual profiler does this automatically, but I’ve had problems using the visual profiler before when running cuda code on multiple cards, so I would like to do this manually. I’m running Fedora linux right now. Thanks in advance.
Read the documentation in $CUDA_INSTALL_DIR/doc/CUDA_Profiler_2.0.txt
Ah, thanks. That is the file I was looking for. :)