With pgprof, I changed my serial program to GPU parallel.
I want to use pgprof to further find out the hotspots of the program. But when I open pgprof and select the executable I generated, pgprof prompts me:
Unable to profile application ,
“Unified Memory profiling failed”
I have almost nothing added to my program, just add some “acc kernels” in front of the loop.
Operating system: ubuntu16.04
If you disable ‘Unified Memory profiling’ (uncheck the box) when you create a new session are you then able to profile your application successfully?
Ohï¼ŒIt works!thq very much .
A new problem has arisen. The pgprof pops up halfway through the profile:
Unable to profile application,The application being profied received a signal."
In the console bar below:
Warning:some profileing data are not recorded Make sure cudaprofilerstop() or cuprofilerstop is called before…
Error:Application received signal 139
Can you tell from the console output whether the application is hit a exception or error? Which version of PGI and CUDA driver do you have installed?