.CU Files for MATLAB

I am trying to replicate an academic paper. The code from the paper includes a .cu file that generates files that are needed for the remainer of the paper.

I did some searching and I may need an additional toolkit for MATLAB to run this correctly

https://www.mathworks.com/matlabcentral/answers/35095-how-to-compile-cu-files-for-execution-in-matlab

From here, I tried the nvcc command, and my version of MATLAB does not recognize the function.

Is the following link the toolkit that I need to get this to run?

https://www.mathworks.com/solutions/gpu-computing.html

[Walter Roberson]

https://developer.nvidia.com/cuda-toolkit-archive at least

[Oli Shaw]

I've already dowloaded this. After doing so, I tried executing the following code

nvcc -ptx MCMC_CUDA.cu (this is the name of my file)

found from the site at the link below

https://www.mathworks.com/help/parallel-computing/run-cuda-or-ptx-code-on-gpu.html

Doing this, I get the following error

Is there something I did incorrectly with this download?

[Walter Roberson]

That toolkit itself does not install nvcc.m

https://www.mathworks.com/matlabcentral/fileexchange/29611-nvcc-cuda-compiler-wraper might help

Note: you might need VS 2017, or VS 2015 Professional, in order to use CUDA properly. On MS Windows, nvcc does not support mingw

[Oli Shaw]

I already downloaded the toolkit. Attached is where everything downloaded to. I'm very confused as far as what to do next. Clicking on the highlghted file does nothing.

[Walter Roberson]

After you installed that File Exchange contribution, go to the MATLAB command line and command

nvcc -config

What result do you get?

[Oli Shaw]

I keep getting this message

[Oli Shaw]

I am trying reinstalling the toolkit, and have received this message part-way through. Could this be part of the issue?

[Walter Roberson]

You need VS 2017, or VS 2015 Professional, in order to use CUDA properly.

Answer 1

[Joss Knight]

Hi Oli. You don't run nvcc in MATLAB, since it isn't a MATLAB feature. You run it at a Windows Command Prompt (or Powershell). Use it to generate a PTX file and then load that into MATLAB using the CUDAKernel feature. To use this properly is an Advanced skill because it requires you to understand a bit about CUDA code and CUDA kernels.

[Samuel Gray]

nvcc is a command-line tool that is installed when you install the CUDA toolkit.

https://www.mathworks.com/matlabcentral/fileexchange/41089-color-balance-demo-with-gpu-computing

the above download for an example of using nvcc with matlab