Performance Profiling on CPUs [Slides]
Sriram Swaminarayan, USDOE National Nuclear Security Administration (NNSA) · 2024
Measure before modification. Use CrayPat or Intel Advisor to find performance bottlenecks. Roofline analysis is easy with intel advisor. Running from the command line will allow you to submit to the job queue and allow you to analyze results later from the front end. It is possible to do collection only on a subset of ranks.