OpenMP programming for parallel/vector computing
The latest generation of Intel Xeon SP (Scalable Processors)
CPUs is very powerful, with each CPU having up to 64 cores,
each with 1 or 2 AVX-512 vector units.
The purpose of this 1-day course is to introduce students,
postdocs and others to OpenMP programming which is often
the best way to obtain high performance on such systems.
The course consists of approximately 3.5 hours of lectures and
2.5 hours of practicals. All that will be assumed is some basic
proficiency with C. No prior experience with parallel computing
will be assumed.
In case you do not have a lot of experience with C, here are links to
a couple of introductory lectures on C
and an online resource.
Timetable
This is a possible timetable, but the timing is approximate
and might be changed.
- 10:00 - 11:15 lecture 1
- 11:15 - 11:30 break
- 11:30 - 12:45 lecture 2
- 12:45 - 13:30 lunch
- 13:30 - 15:00 practical 1
- 15:00 - 16:00 lecture 3
- 16:00 - 17:00 practical 2
References
Lectures
Practical 1
Application: a simple 2D finite difference method
OpenMP aspects: #pragma omd parallel for, shared and private variables,
reduction clause, run-time library, environment variables, avoiding
false sharing
Practical 2
Application: Monte Carlo simulation using Intel's MKL library
OpenMP aspects: #pragma omp parallel, #pragma omp threadprivate,
array reduction, MKL random number generation
Acknowledgements
Many thanks to:
- Wes Armour and Ian Bush for feedback on my original lectures and practicals
- Ian Bush, EPCC and NAG for their OpenMP training materials which
were very helpful in preparing my original materials