Automatic code generation of c++ code from matlab functions

Hi everyone,

I would like to generate c++ code from a automatically generated matlab function.

The matlab function header looks like this:

function fd = e7bd_nonlinsys_f(in1,in2,Ts)

with

fd and in1 being single 6x1 vectors, in2 being a single 2x1 vector and Ts being a single scalar.

The code for the code generator looks like this:

% Set configuration for Code Generation 
cfg = coder.config('lib');
cfg.GenCodeOnly = true; % Code generator produces C++ source code, but does not invoke make
                        % command or build object
%cfg.GenerateExampleMain = 'DoNotGenerate'; % Code Generator does not generate an example main function
cfg.TargetLang = 'C++';
cfg.CastingMode = 'Standards'; % Data type casting conforms to MISRA standards
%cfg.GenerateReport = true; 
%cfg.GenerateCodeMetricsReport = true; % Generate Metrics report for static code analyzation
cfg.CppInterfaceStyle = 'Functions'; % = 'Methods';
%cfg.CppInterfaceClassName = 'e7bd_linsys_class';
cfg.HighlightPotentialDataTypeIssues = true; % The code generation report highlights MATLAB code that 
                                             % results in single-precision or double-precision operations 
                                             % in the generated C/C++ code. If you have Fixed-Point Designer™, 
                                             % the report also highlights expressions in the MATLAB code that 
                                             % result in expensive fixed-point operations in the generated code.
cfg.PreserveArrayDimensions = true; % Generate code that uses N-dimensional indexing 
%cfg.PurelyIntegerCode = true; % The Code generator does not allow floating point data or operation. 
cfg.SaturateOnIntegerOverflow = true; % Code generator produces code to handle integer overflow 

% Generate C++-Code of function e7bd_nonlinsys_f
codegen e7bd_nonlinsys_f -config cfg -v...
    -args { single(zeros(6,1)), single(zeros(2,1)), single(0.02) }

As a result, I get several c++/h - header files, implementing checks for initialization of the vectors, checking data types and values of the input parameter and also a termination function.

The core c++ function itself, looks like this:

// Function Definitions

//

// E7BD_NONLINSYS_F

// FD = E7BD_NONLINSYS_F(IN1,IN2,TS)

// Arguments : const float in1[6]

// const float in2[2]

// float Ts

// float fd[6]

// Return Type : void

//

void e7bd_nonlinsys_f(const float in1[6], const float in2[2], float Ts, float

fd[6])

I have several questions regarding this generated code:

1. The resulting c++ function has the output vector fd as input parameter and void as return type. However, I would like to generate a function returning the vector fd just like the matlab function does. How can I change that?
2. Ideally, I would like to get a header file only, containing both the declaration and the definition of the function e7bd_nonlinsys_f. How can I omit the generation of the other helper files and compromise the e7bd_nonlinsys_f.cpp and e7bd_nonlinsys_f.h - file in one .h - file only?

Regarding question 2 - The ideal generated code (c++-header file only) should look somewhat like this:

e7bd_nonlinsys_f.h :

float[6] e7bd_nonlinsys_f(const float in1[6], const float in2[2], float Ts)

{

< calculation>

return fd;

}

I´m using Matlab R2020a, MATLAB Coder, Embedded Coder and the Symbolic math toolbox.

I furthermore also tried the ccode command which only generated the assignment of the mathematical term to the matrix without declaring/defining the function or any other part of the header file (e.g. including necessary libraries and so on).

Thanks for your help :)

Michael

Answer 1

[Ryan Livingston]

1. There is not an option to return arrays as a return value from the generated functions. The documentation describes the expected generated interfaces. In C, returning an array as a return parameter doesn't really make much sense either.
2. There is not a way to generate inline implementations in a header file. MATLAB Coder will always put the definitions in a C or CPP file. The FilePartitionMethod config settting will allow you package the code into fewer files.

[Michael Knörzer]

Hi Ryan,

thanks for your response!

I´m sorry for the question, but why does it not make sense to return an array as a return parameter?

Cool, I tried the FilePartitionMethod setting and as you said the coder generated one .cpp and one .h file. - Thank you very much!

Best regards,

Michael

[Ryan Livingston]

This answer:

https://stackoverflow.com/a/11656585/3297440

has a good discussion of returning arrays in C. Returning an array by value isn't supported in C. The common idioms are laid out there: allocate memory in the callee and return that or allocate memory in the caller and pass that to the callee. The generated code uses the second idiom.

[Michael Knörzer]

Ok, thanks for sharing!