You can use hpctoolkit to profile the cpu portion, it does exactly what you are asking. To profile the cuda portion, you will have to use nvidia's visual profiler. John On Wed, Jun 20, 2012 at 4:32 AM, Luigia Ambrosio <[email protected]> wrote:

Hi all I'm interested in using PETSC with CUDA. I just downloaded and installed the developer version and I was testing the example #1 ($HOME/petsc-dev/src/vec/vec/examples/tutorials) to compute the norm of a vector. I would like to compare PETSC "without cuda" and "with cuda". Is there a way to profile a PETSC executable, obtain the stack of the called functions and methods and access to the function body? Can anybody make an exmple? Thanks a million