Debugging GPU Coder Functions

I have a few matlab functions that I am converting to run on GPU using matlab's GPU Coder app. The functions convert fine but when I convert one of my functions in the check for issues stage I notice that the function produces NaN values and results in the results not being correct. I am assuming when the code is converted to GPU it is misinterpriting something and dividing by zero esecially since I am dealing with imaginary numbers. I want to add breakpoints to see where this happens but it seems that you cannot do that once the function is converted to .mexw64. How can I debug this function and pinpoint the location of the error?

[Justin Hontz]

Do you encounter incorrect results when generating plain C/C++ code, or just GPU code? Can you reproduce the issue for any simple example, or only for this particular appliation?

Resources for debugging generated MEX code can be found below:

• https://www.mathworks.com/help/coder/ug/running-debugging-mex-functions.html
• https://www.mathworks.com/help/gpucoder/gs/debug-cuda-mex-functions.html