Closed zaikunzhang closed 2 years ago
Hi @zaikunzhang,
The approach we use to set up MATLAB on macOS GitHub-hosted runners is generally inferior to the approach we use on Linux (which is why macOS support remains undocumented). macOS also has an increasing number of security protections which can slow first launch and execution of newly installed apps.
I suspect if you ran these benchmarks locally, the results would be more consistent between the platforms.
Best, Mark
Hello @mcafaro Mark,
Thank you very much for the very quick response and explanation.
I do not have a Mac, so I cannot try the timing locally, but your explanation sounds reasonable.
Would you expect that the performance of MATLAB on macOS (concerning MEX particularly) will improve in the future?
Thank you very much again! Your work of making MATLAB available on GitHub Actions is life-changing for me. I am very grateful for it.
Best regards, Zaikun
Hi @zaikunzhang,
I believe the performance of MEX on macOS is comparable to other operating systems outside of GitHub Actions. I do not expect the performance on GitHub Actions to change significantly in the near future. It is something we are looking into for the next major release.
Best, Mark
Hello @mcafaro ,
Thank you very much for the explanation and updates!
Best regards, Zaikun
Hello!
I did some timing of
MEX
on macOS and Linux using GitHub Actions. It turns out that MEX is much slower on macOS than on Linux: slow to set up, slow to mexify, and the MEX function is slow to run. In particular, it is about 250 times slower to set up MEX for Fortran on macOS than on Linux.Below is the data (time in seconds). More details are available at the action. Is this a problem of macOS, the virtual environment of GitHub-hosted runners, Intel oneAPI, or MATLAB? Thank you very much for checking it.
See discussions on MATLAB Answers and StackOverflow.