Performance Profiling

A crucial component of the EuroHack is knowing which components to port to GPUs. Profiling the code (already on CPUs) will give a good indicator. Alistair Hart generously has provided some guidelines for profiling:

Getting Ready for EuroHack

Mandes Schoenherr of Cray has written an overview of profiling tools on an XC40 (albeit the installation at HLRS -- should be updated for CSCS)

Cray Performance Analysis Toolkits

Profiling OpenACC programs There are a number of ways to get simple profiling and debugging information, often only by turning on environment variables. Alistair and Mandes have put together a quick overview of these simple techniques for both Cray and PGI.

Getting Feedback at EuroHack

There are more complicated tools to profile OpenACC programs, which give you a nicer presentation of information. As with all technology, the devil is in the details. Jean-Guillaume Piccinali and Ben Cumming put together the following profiling example for various programming paradigms, including OpenACC:

OpenACC profiling examples for CSCS/USI Summer School

Cray CCE -h profile_generate

Cray Perftools API

Cray CRAY_ACC_DEBUG API

Nvidia nvprof