Matlab Codegen Deploy

MATLAB Codegen Deploy

Use this skill when MATLAB or Simulink algorithms need to become generated code or deployable artifacts.

Workflow

1. Identify target: MEX, C/C++, embedded C, CUDA, HDL, PLC, or packaged app.
2. Verify required products with ver and license.
3. Isolate the algorithm into a codegen-friendly function.
4. Define input types with coder.typeof or representative test vectors.
5. Generate the smallest artifact first, usually MEX.
6. Compare generated result against MATLAB golden output.
7. Save generated code, logs, reports, and validation metrics.

Self-Compile and Verify Preference

The user expects generated MATLAB/Simulink work to compile and verify itself after writing. Do not stop at source generation when a build or smoke check is feasible:

1. For MATLAB Coder, run codegen and then compare generated/MEX output against MATLAB golden output when compiler support exists.
2. For Simulink Coder, update/compile the model first, run simulation, then call slbuild only after simulation passes.
3. For embedded targets such as STM32, first try the configured hardware target and toolchain; if it fails, capture the exact missing package/compiler message and generate the closest portable C fallback if possible.
4. Verify generated .c, .h, project, library, elf/axf/hex/bin, or fallback artifacts exist and are nonempty.
5. Report clearly which stage passed: model update, simulation, code generation, toolchain build, binary generation, or only portable fallback.

MATLAB Coder Pattern

Use this pattern before larger deployment:

cfg = coder.config("mex");
codegen -config cfg myFunction -args {coder.typeof(0,[100 1],[1 0])}
gold = myFunction(x);
actual = myFunction_mex(x);
assert(norm(gold-actual) < 1e-9)

Embedded Readiness

Check:

• Fixed-size vs variable-size arrays.
• Dynamic allocation and recursion.
• Unsupported functions.
• Numeric overflow and fixed-point scaling.
• Stack and heap impact.
• Timing budget and hardware target assumptions.

Simulink Codegen

For models:

• Update/compile the model after editing and before slbuild.
• Build only after simulation passes.
• Set solver and sample times explicitly.
• Run Model Advisor or relevant checks when available.
• Compare simulation outputs before and after code generation.

STM32 Embedded Coder Readiness

For STM32 targets, do not treat "an executable exists on disk" as enough. Verify the full MATLAB-to-ST chain before blaming the model:

• MATLAB products: Simulink, Simulink Coder, Embedded Coder, STM32 support package.
• MATLAB registration: stm32cube.hwsetup.stm32Tools.getInstalledSTM32CubeMX() returns the CubeMX directory MATLAB will use.
• ST tools: GNU Tools for STM32, CMSIS, CMSIS-DSP, STM32CubeMX, and the matching STM32Cube firmware package such as STM32Cube_FW_F1_*.
• Model binding: the model has a real .ioc path in STM32CubeMX.ProjectFile.
• Target hardware: codertarget.targethardware.getRegisteredTargetHardwareNames includes the required STM32 family, for example STM32F1xx Based.

Prefer target data APIs over editing opaque structures:

tools = stm32cube.hwsetup.stm32Tools();
tools.updateSTM32CubeMXPath("C:\Users\me\AppData\Local\Programs\STM32CubeMX");

codertarget.data.setParameterValue(model, "STM32CubeMX.ProjectFile", iocPath);
codertarget.data.setParameterValue(model, "STM32CubeMX.DeviceId", "STM32F103RBTx");
codertarget.data.setParameterValue(model, "STM32CubeMX.Family", "STM32F1");

If slbuild hangs around "Generating code from STM32CubeMX project" or "Starting compilation process":

• Check STM32CubeMX.log, generated hardware scripts, and background java.exe / STM32CubeMX.exe processes.
• Compare direct STM32CubeMX.exe -q script with MATLAB's java -jar STM32CubeMX.exe -q script; some CubeMX versions can behave differently.
• Prefer the MathWorks-recommended CubeMX version shown by message("stm32:setup:CubeMXReqVersion").getString.
• If non-ASCII paths are involved, copy the model and .ioc to an ASCII scratch path and retry once to isolate path encoding from model errors.
• Only use a local CubeMX wrapper as a documented workaround after recording the direct failure and the real CubeMX path.

Output

Report target, generated artifact path, report path, golden comparison result, and limitations.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: wzyn20051216
• Source: wzyn20051216/matlab-agent-skills
• License: MIT
• Homepage: https://github.com/wzyn20051216/matlab-agent-skills

Install and usage instructions live in the source repository linked above.