I’m a new user for CUDA & matlab.
And I have a question now:
If I have a mex c function called by matlab, and I want to accelerate this function by CUDA.
Finally I want to use matlab to call this accelerated function, should I set up some special parameter for matlab or CUDA?
Or I just re-mex it and follow the normal procedure for c calling CUDA?
Thanks to everyone! And forgive me the poor English!
Google for nvmex.
Please try this document: http://developer.download.nvidia.com/compute/cuda/1_0/Accelerating%20Matlab%20with%20CUDA.pdf
You might also consider two other options:
- Using Jacket directly instead of writing your own CUDA kernels, http://accelereyes.com/jacket_tour
- If you definitely want to write your own CUDA, the Jacket SDK trumps standalone MEX, as described here: http://blog.accelereyes.com/blog/2010/10/29/jacket_sdk_trumps_mex/