Node Level Performance Optimization

Course material for the CSC course "Node Level Performance Optimization". The course is part of PRACE Training Center (PTC) activity at CSC. The most recent version of the course is found in the master branch, previous versions can be found in Tags.

Lecture slides

Lecture slides are found in the course web page. Lecture slides from 2022 are here.

Information about compilers

Selected options

Exercises

• General instructions

Getting started with Puhti and Mahti

• Getting started with CSC supercomputers

Instruction level parallelism

• Instruction level parallelism

Main memory performance

• Stream
• NUMA aware memory usage

Performance tools

• Using Intel VTune
• Using Intel Advisor

Vectorization

• Improving performance through vectorization

Cache optimization

• Loop interchange
• Cache blocking

Miscallenous optimizations

• Fast mathematics
• Branch prediction

Additional exercises

Threading optimization

• Simple thread affinity
• Thread affinity in heat equation