MATLAB compatibility?

[maciejka3000]

Hey, does anyone had success with launching ZLUDA with MATLAB on Windows? After executing command, it immediately halts.

[maciejka3000]

Okay, i managed to launch it with -wait parameter. GPU is seen properly after using gpuDevice command. However, deep learning in MATLAB is not working. I've got this error message:

Call to cublasDgemm in CUBLAS failed with error status: CUBLAS_STATUS_NOT_SUPPORTED.

[pegeac]

Hi, I'm sorry, I'm not good with this stuff. How did you manage to make it work?
It was windows/Linux?
Can you write the modified lines you used to make it work?
Moreover, did you use the gpuDevice on MATLAB?

[maciejka3000]

Hi, sorry for late reply.
I've tried using ZLUDA on Windows by launching it using following command:
.\zluda.exe -- matlab -wait

I had used gpuDevice on matlab r2023a and it showed proper information about GPU. However after updating to r2024a matlab immediately crashes.

[lshqqytiger]

Could you try my fork?

[maciejka3000]

Hey, thanks for answer. Unfortunately, your solution doesn't work. I've tested your fork with those commands:

• trainNetwork gives error An unexpected error occurred trying to launch a kernel. The CUDA error was: named symbol not found
• classify gives error Unexpected error calling cuDNN: CUDNN_STATUS_EXECUTION_FAILED.
• trainAutoencoder gives error Call to cublasDgemm in CUBLAS failed with error status: CUBLAS_STATUS_NOT_SUPPORTED.

I've also tried multiplying 2 gpu arrays together. Creating gpu variables went fine, but multiplying them gave an error associated with CUBLAS.