How do I upload my deeplearning network to previous version of Simulink?

I'm trying to upload my deep learning model which is trained at R2021a, to Simulink R2018b.

I've been trying this for over a month, but I didn't succeed.

If you know any Idea, It will be huge gift for me.

I have tried these things:

- at R2018b Simulink, Use matlab function block

-> It causes error(coder dot indexing problem)

- at R2018b, generate C++ code for my model

-> It doesn't work(dot indexing problem)

- at R2021a, generate C++ code for my model, and upload to R2018b Simulink via S-function

-> Code generation was successful, but I have no idea what to do next.

Answer 1

[Hariprasad Ravishankar]

Hello,

I believe step 2 of this MATLAB Answers post will help you take your generated code to a previous release of MATLAB via S-function.

https://www.mathworks.com/matlabcentral/answers/832938-how-to-deploy-deep-learning-networks-for-hardware-in-the-loop-hil-simulation-with-for-example-sp#answer_702348

Please let us know if you found this helpful.

Hari

[재원]

You are my savior. Thank you so much.

Instead of using cuDNN, I used MKL_DNN as my target library. and It works well, too.

To use MKL_DNN, I should build mkldnn library first. The link below will help.

MATLAB Coder: How do I build the Intel MKL-DNN library for Deep Learning C++ code generation and deployment? - MATLAB Answers - MATLAB Central (mathworks.com)

this is the code I used.

% Function to DLL
setenv('INTEL_MKLDNN', 'C:\Program Files\mkl-dnn')
setenv('PATH', [getenv('INTEL_MKLDNN') filesep 'lib' pathsep getenv('PATH')])

cfg = coder.config('dll');
cfg.TargetLang = 'C++';
cfg.GenerateReport = true;
cfg.DeepLearningConfig = coder.DeepLearningConfig('mkldnn');
codegen -args {ones(1,200,1,8,'uint16')} -config cfg CNN_as_Func

copyfile(fullfile(pwd, 'codegen','dll', 'CNN_as_Func','CNN_as_Func.dll'), pwd);


% DLL to S-Function
srcPath = fullfile(pwd, 'codegen', 'dll', 'CNN_as_Func');

mkldnnPath = getenv('INTEL_MKLDNN');
mkldnnLibPath = fullfile(mkldnnPath,'lib');
mkldnnIncludePath = fullfile(mkldnnPath,'include');

libs = {'CNN_as_Func.lib','dnnl.lib'};

headerPath = {srcPath;mkldnnIncludePath};
libPath = {srcPath;mkldnnLibPath};

def = legacy_code('initialize');
def.SFunctionName = 'High_Classification_S_Func';
def.OutputFcnSpec = 'void CNN_as_Func(uint16 u1[1600],single y1[16])';
def.IncPaths = headerPath;
def.HeaderFiles = {'CNN_as_Func.h'};
def.LibPaths = libPath;
def.HostLibFiles = libs;
def.Options.useTlcWithAccel = false;
def.Options.language = 'C++';

legacy_code('sfcn_cmex_generate', def);
status = evalc("legacy_code('compile', def)");

Thanks again, Hariprasad Ravishankar.