Can programs using PGI directives be profiled using computeprof (formerly known as cudaprof) or that only works for programs written in CUDA?
So far I am not able to get it to work, do I need special flags during compilation for the profiling to work?
Thank you for your help.
Hi Sindimo.
Yes you can profile the GPU code generated by PGI Accelerator Model directives using cudaprof, However, you can’t profile both the host and GPU code together. For this, use the PGI profiling utility ‘pgcollect’ and view the results with PGPROF. Please see the PGI Tool’s Guide (http://www.pgroup.com/doc/pgitools.pdf) for details.