We explain how to make optimal use of modern multi-core computer architectures with vectorization capabilities, how to profile your code, and how to determine maximum parallel speed up. We use a convolution code and an n-body code as central examples to work on in this and subsequent workshops.

Registration

• Academic participant : 10$
• Non-academic participant : $200

Prerequisites

• Linux command line
• Experience with C, C++ or Fortran

Course plan

9am to 12pm

• Where most time is spent
• Best practices
• Amdahl's law and the Karp-Flatt metric
• Profilers, such as perf
• Computer architecture, NUMA, caches, and data proximity
• Vectorization
• Examples: convolution and n-body

13h30 to 15h (Discord)

• Question and Answers
• Extra exercices and help

Instructor

Bart Oldeman, analyst in advanced research computing at Calcul Québec.

Language

English

Technical prerequisites

We will use the Zoom platform. Because this event is a practical workshop, it is very useful having a secondary screen where you would get the instructor window on one screen and your own window on your main screen.

We will use the Jupyter Lab interface. Make sure you have a modern Web browser like Google Chrome, Firefox, Edge or Safari.

Contact

For any question, please write to training@calculquebec.ca.