Summary

Overview

Teaching: 10 min
Exercises: 0 min

Questions

• Which techniques should I use for which types of problem?

• What should I do next?

Objectives

Summary

We have leant that:

• MATLAB PCT provides a platform for utilising the power of supercomputers.
• parfor is a parallelised version of a for-loop.
• spmd offers a very flexible environment to develop parallel codes in MATLAB.
• We can use profile and mpiprofile commands to obtain the performance statistics of our code.

We also have learnt how to run MATLAB without the GUI and how to access MATLAB from HPC clusters.

What next?

We have covered a lot of topics in this course. However, this course is by no means exhaustive. This course covers only the basics of some essential features of MATLAB PCT that will help in improving the performance of your code by exploiting the parallel computers. It must be super-exciting to try all these techniques and tools to improve the performance of your MATLAB codes.

There are many features that we have not covered. We can also explore the use of GPUs in accelerating the performance. The topic of GPU computing in MATLAB is big enough to be a one-day workshop. See

• https://blogs.mathworks.com/loren/2012/02/06/using-gpus-in-matlab/
• https://uk.mathworks.com/help/parallel-computing/gpu-computing-in-matlab.html for details.

Some tips

For helping you in improving the performance of your code, we have the following tips:

• Avoid premature optimisations.
• Profile the code before jumping on parallelising. It is important to know what to parallelise.
• Follow incremental approach to parallelisation - parallelise your code step by step, starting with the most critical function or block of code.
• To track the changes to your code and revert to the working version, in case the modifications are not satisfactory, or if you make a mistake, use a version control system. We recommend Git. If you are not familiar with version control and Git, then we suggest this Software Carpentry course.
• Prepare tests that will be useful to test your code once you make some major changes, so that modifications do not affect the code developed thus far; to make sure that the code still works the way it should. These tests can be unit tests or integration tests and some program that can verify the output of your code.
• You don’t have to run the entire simulation. Always use a small data set when testing and optimising your code.

Additional resources

For additional information, we suggest the following resources:

On MATLAB Parallel Computing Toolbox

• https://uk.mathworks.com/help/parallel-computing/getting-started-with-parallel-computing-toolbox.html
• http://cacs.usc.edu/education/cs596/ParallelComputingWithMATLAB.pdf
• https://www.glue.umd.edu/hpcc/help/software/matlab.html

On some (best) practices for improving the performance of your code

• http://www-h.eng.cam.ac.uk/help/tpl/programs/Matlab/tricks.html
• https://undocumentedmatlab.com/books/matlab-performance
• https://uk.mathworks.com/help/matlab/matlab_prog/techniques-for-improving-performance.html
• https://www.slideshare.net/jbhuang/writing-fast-matlab-code
• https://uk.mathworks.com/company/newsletters/articles/tips-for-accelerating-matlab-performance.html
• http://www.mit.edu/~pwb/cssm/GMPP.pdf
• https://vp.research.msu.edu/sites/default/files/2018-08/Speeding_up_MATLAB_Applications.pdf
• http://www.csc.kth.se/utbildning/kth/kurser/DN2255/ndiff13/matopt.pdf
• http://www.walkingrandomly.com/?cat=53

Key Points