Porting Autogenerated C code from Deep learning toolbox to embedded environment

In ANN Auto C code all matrices are appearing as Static const variables which means that they will be RAM. These variables need to be in FLASH and always read from FLASH due to limitations in RAM size. How is it possible to customise the autocode from Matlab scripts to achieve this ?

Are there are any code generation configuration which allows us to achieve this ?

Please note that I am using MATLAB coder and M-scripts (not SIMULINK)

[Walter Roberson]

Huh, I would have thought that static const variables would be the most likely to be kept in FLASH.

Answer 1

[Ram Kokku]

@Nitin Skandan, mapping constants to hardware depends on the backend C/C++ compiler. MATLAB Coder allows customization of deep learning constant generation using the config option LargeConstantGeneration ( https://www.mathworks.com/help/coder/ref/coder.codeconfig.html ). If this does not meet your requirements, please provide an example code snippet that describes your requirements.

[Nitin Skandan]

Will try with LargeConstantGeneration options. But looks like this will help us seggregate the weights from the C Source files .