How to resolve unsupported functions in MATLAB HDL Coder?

Hello,

We have some simulation code written in MATLAB that we are trying to simulate and deploy on an FPGA, and so we are trying to use the HDL Coder tool for this task. However, we are receiving code generation issues indicating that the following functions are not supported:

• diff
• fft
• exp
• angle
• ifft
• filter

Is there perhaps a toolkit or package that replicates these functions with support for HDL coding? I tried referencing the list of supported functions found here:

https://www.mathworks.com/help/hdlcoder/ug/functions-supported-for-hdl-code-generation-alphabetical-list.html

But the links for supported functions don't seem to be working (at least for me).

Otherwise, is there a recommended way of resolving this?

Thank you,

James

P.S. I have successfully used the MATLAB Coder to generate C code for this project, if this helps.

[Walter Roberson]

fft is handled by dsphdl.FFT

exp is handled by https://www.mathworks.com/help/fixedpoint/ref/cordiccexp.html

angle is handled by cordicangle

ifft is handled by dsphdl.IFFT

filter() itself is not handled, but there are a variety of filters handled, such as dsphdl.BiquadFilter and dsp.HighpassFilter and dsp.FIRFilter

[James]

Hello Walter,

Thank you for the response! For the filter() function specifically, this dependence comes from a custom implementation of MATLAB's smooth() function, which we had to write as smooth() is not supported for C or HDL code generation. I'm not sure which of the supported filters would be most appropriate to replace this function, but I can try to dig into this a bit more.

Best,

James

Answer 1

[Kiran Kintali]

Happy to assist you with your MATLAB to HDL workflow. Attached is a sample zip file with the code for the attached functions (using R2025a pre-release)

Many customers use HDL Coder to create custom MATLAB algorithms to generate HDL IP. To tailor our assistance to your application needs, could you provide me some details on the following?

1. Development Approach: Are you planning to develop your MATLAB code in floating-point and utilize HDL Coder's integrated float2fixed workflow for fixed-point conversion before generating HDL? If so, what are your typical wordlength and fractionlength requirements?
2. Floating-Point Hardware: Do you need to generate floating-point hardware directly from your MATLAB code for any functions within the HDL Coder workflow?
3. Target Platform: Are you working with FPGA/SoC or ASIC? If it's FPGA/SoC, what is your target platform? For ASIC development, note that HDL Coder supports an HLS (SystemC) workflow compatible with the StratusHLS tool integrated into the Cadence Genus environment.
4. Performance Requirements: What are your frequency and area requirements? MATLAB to HDL supports VHDL, Verilog, and SystemVerilog, as well as SystemC and Synthesizable C++ for HLS workflows.

Thanks

[James]

Hello Kiran,

Unfortunately, many of the technical details are still undecided. Right now, we are trying to perform the conversion to run some initial simulations and estimate resource requirements until additional requirements come in. To answer your specific questions:

1. Unsure. For now, let's assume that we will not use the float2fixed workflow.
2. We may run some experiments to compare the performance of floating-point and fixed-point arithmetic, so I think the answer to this question is "Yes".
3. We are working with FPGA/SoC. Our target platform will be a Xilinx board, though the exact board is not yet decided.
4. As of now we do not have any frequency or area requirements. We are still in an exploratory stage and have some flexibility to maximize performance. We will be generating Verilog code.

Best,

James

[Kiran Kintali]

MathWorks provides a comprehensive suite of predefined IP modules for various functions:

• FFT: Use dsphdl.FFT. Learn more here.
• Exponential: Refer to the CORDIC Exponential documentation.
• Angle Calculation: Use cordicangle. Further details.
• IFFT: Utilize dsphdl.IFFT. Read more here.

Unsupported Functions

• Filter: The filter() function is not yet supported. Consider alternatives like dsphdl.BiquadFilter and dsp.FIRFilter.
• Differences and Approximate Derivatives: The diff() function is not supported; you may need to create a custom function.

Custom MATLAB Code

You can also implement all the above functions as custom MATLAB code. Refer to the mathfcns.m example in this folder, which defines these operators from first principles.

Project Configuration

The attached .prj file is configured for Verilog. If you set up Vivado in your path, you can perform synthesis and P&R steps directly within the viewer. A test bench is automatically generated based on the testbench. You can also write runme.m files and run the entire workflow (float2fixed + hdlcodegen + testbench + synthesis steps) from command line. See command line workflow

Mixed Code Support

You can mix floating-point and fixed-point code. By enabling the integrated fixed-point conversion step in HDL Coder, you can convert floating-point code to fixed-point. If the code is already in fixed-point, you can instruct the float2fixed step to skip certain functions. Additionally, the MATLAB to HDL product includes automatic function replacement and lookup table generation features.

Additional Steps

• Try the code generation and synthesis steps.
• For further assistance, do not hesitate to reach out to MathWorks support.

Useful Resources

• HDL Code Generation from MATLAB
• HDL Coder Examples
• Fixed-Point Conversion: Learn more about working with fixed-point code.

Examples and Demos

Several FFT-based examples are located in the design patterns folder (<matlabroot>/toolbox\hdlcoder\hdldesignpatterns\matlabhdl).

Use the following command to see fft based examples:

>> mlhdlc_demo_setup('fft')

To see the complete list of available sample code, try:

>> mlhdlc_demo_setup('list')

[James]

Hello Kiran,

Thank you for this very detailed information! I will take a look at these resources and the provided example project and let you know if we need any further assistance.

Best,

James