I have some code that compiles to .mex and to C++ fine in R2018a, but in R2024b, it errors after running out of memory.

I can't post my code on a public place like this. Is there any way this can be done privately?

I have looked at coder.config('mex') and coder.config('lib') for both versions and can't see anything obvious.

[Walter Roberson]

You could offer to send links to private repositories on dropbox or google; the question is whether you would get anyone interested who would be willing to sign a Non-Disclosure Agreement for free consultation.

[Dr W Kurt]

I went ahead and cut down the codebase to just what fails, then submitted another ASK that includes the testbench, and the subroutine that errors on compile.

Best,

Kurt

Answer 1

[Subhajyoti]

Hi @Dr W Kurt,

It is my understanding that you are able to generate the MEX files for your code. But when you are trying to generate C/C++ files, it is throwing out of memory error.

This error is due to implicit expansion operation occurring during run-time.

Refer to following MathWorks Documentation link to know more about ‘Optimize Implicit Expansion in Generated Code’ in MATLAB:

https://www.mathworks.com/help/releases/R2024b/coder/ug/optimizie-implicit-expansion-in-generated-code.html

While generating MEX file, the MATLAB Coder generates C or C++ files (as mentioned in code-generation parameter) and the MEX file acts like an API between MATLAB and the C/C++ file. You can also use these C/C++ files directly as a workaround for code-generation error. The C/C++ build files are generated in the following path:

./codegen/mex/OFDMRxR1/

Additionally, you can refer to the following resources to know more about Analysis and Generation of C/C++ Code from MATLAB Code Using "MATLAB Coder":

• https://www.mathworks.com/help/releases/R2024b/coder/ug/code-generation-workflow.html
• https://www.mathworks.com/help/releases/R2024b/coder/ug/preparing-matlab-code-for-c-c-code-generation-workflow.html

[Dr W Kurt]

Thank you, I will look this over and try the suggestions you mention. I am somewhat surprised because R2018 compiles the exact code to C/C++ without any issues whatsoever.

Best,

Kurt