Why do my parfor loops run in parallel in matlab interactive mode, but in serial mode when I deployed, using Matlab compiler?

Question

William Smith 2018-2-14

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/382672-why-do-my-parfor-loops-run-in-parallel-in-matlab-interactive-mode-but-in-serial-mode-when-i-deploye

编辑： William Smith 2018-2-14

I have multiple programs that contain parfor loops. These parfor loops successfully execute the relevants sections in parallel when I run in interactive Matolab.

However when I deploy these programs using Matlab compiler, they only run in serial. It seems the Matlab compiler is not honouring my setting "Automatically create a parallel pool (if one doesn't exist) when parallel keywords are executed", which I have enabled in my Matlab settings. (I am compiling using the same settings, so these settings should be baked into the deployed app).

2 个评论
显示无隐藏无

Matt J 2018-2-14

编辑：Matt J 2018-2-14

I am compiling using the same settings, so these settings should be baked into the deployed app

I have a feeling that might not be a valid expectation. I don't see any such setting in the run-time options within the deploytool.

William Smith 2018-2-14

编辑：William Smith 2018-2-14

在 MATLAB Online 中打开

I'm compiling using mcc. I believe that the deployed program contains $PREFDIR and its contents, and unpacks these into $TEMPDIR/mcrCache9.1/ at runtime. I can see "parallel.settings" in that folder. And I can see

 <key name="AutoCreate">
                <bool>
                    <value>1</value>
                </bool>
 </key>

within that file.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

William Smith 2018-2-14

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/382672-why-do-my-parfor-loops-run-in-parallel-in-matlab-interactive-mode-but-in-serial-mode-when-i-deploye#answer_305132

Note, after 1 day messing around, I was able to work around this by explicitly calling parpool() just before my parfor. This should not be necessary because of the "Automatically create a parallel pool" settings.

Note, I am running Matlab 2016b. I have not verified this incorrect behaviour on a more recent version of Matlab.