Generating Object-Oriented C++ Code from MATLAB Algorithms

Intro

In this demonstration, I'm going to use MATLAB Coder to generate production C++ code from MATLAB algorithms. We've made a number of improvements to our C++ code generation and I want to highlight these in this video.

The MATLAB example I'm going to use represents oscillating systems. <Show a picture of Live Editor with the system's output>

Open app and setup

<Open App and select entry point in the background>

I'm going to generate some C++ classes from these MATLAB classes using the MATLAB Coder app. Everything I'll show today is also available from the command line using the codegen command.

The entry point is effectOfDamping. I need to tell MATLAB Coder the types of inputs the function needs. We can run the test script runEffectOfDamping to automatically detect my types.

Check for issues

I'll also use the same test script to check for run-time issues prior to generating my C++ code. <Check for issues in the background>

Code generation

I'm going to generate a DLL and C++ to share this code with my team. Choosing C++ shows an option added in R2019b called "Interface style" where you can choose an interface with free functions or an interface using a class and methods on that class. I'm going to choose "methods" to get more idiomatic C++ and I'll name the class OscillatorInterface so we can find it in the generated code. <Generate code now>

Looking at generated code

To explore the generated code, let's open the MATLAB Coder Report. On the left, we see our input MATLAB code and the generated C++ code.

Encapsulated class

Let's take a look at our interface class resulting from that "interface style" option. The class is exported from the DLL, handles initialization and cleanup, and has our entry-point function as a public method. Any MATLAB persistent and global variables are mapped to class properties to make the generated code multi-instance.

MATLAB classes to C++ classes

Starting in R2020a MATLAB classes are mapped to C++ classes. R2020b adds the ability to map MATLAB packages to C++ namespaces. The generated C++ class mimics as many of the MATLAB class's attributes as possible.

coder::array

In R2020a the generated code uses the coder::array class to manage dynamically allocated arrays. coder::array has an interface similar std::vector, provides indexing operator overloads, and ensures exception-safe memory deallocation.

SIL and debugging

Let's verify the behavior of our code. With Embedded Coder, <Select the test script and enable debugging> I can use SIL debugging to open the generated code in a debugger. <Run in the background> The code starts up in a separate process and that process is opened in Visual Studio for debugging. MATLAB Coder passes data from the script to the generated code.

Package

Once we're happy with the behavior of the generated code, I'll package the code for my team in a ZIP file.

Later, if you want to transition from the App to a command line workflow for automated builds, choose the "Convert to script" option to get a script version of your project.

Performance/Compatibility

We've been continually improving MATLAB Coder. Since R20xx, we've added support for numerous MATLAB capabilities, including Table, DateTime, Timetable, and support for machine learning and deep learning. We strive to generate fast code that approaches or even exceeds hand code.

Finalize

If you're interested in C++ code generation from MATLAB code I suggest taking a look at the new C++ capabilities in MATLAB Coder as of R2020b. Taken together, they allow for the generation of idiomatic and exception-safe C++ code using MATLAB Coder. The example code used in this video is available in the product documentation. Thanks for watching!