This is a follow-up of a question a place a while ago.
For double precision calculation it seems that consumer AMD hardware is a much cost effective solution than CUDA-based hardware. However MATLAB (perhaps also Octave?) lacks the capability to interface with OpenCL .
I would like to know if someone found a consistent solution to interface OpenCL with matlab.
Best
Marco
There exists a opencl-toolbox by Radford Ray Juang available at [http://code.google.com/p/opencl-toolbox/]. The tools are working, yet I experienced some stability problems. However, I suspect that those were related to my own bugs in kernel code.
Be also warned that you have to remember that c/c++ and Matlab use different memory indexing (ie row vs column major...) and you may have to take that in consideration when running code from Matlab (see example of simple center symmetric local binary pattern kernel attached)...
The zip contains an example running the kernel (and the toolbox) ..
There exists a opencl-toolbox by Radford Ray Juang available at [http://code.google.com/p/opencl-toolbox/]. The tools are working, yet I experienced some stability problems. However, I suspect that those were related to my own bugs in kernel code.
Be also warned that you have to remember that c/c++ and Matlab use different memory indexing (ie row vs column major...) and you may have to take that in consideration when running code from Matlab (see example of simple center symmetric local binary pattern kernel attached)...
The zip contains an example running the kernel (and the toolbox) ..
If you can't get anything working directly, but have a basic understanding of both C/C++ and OpenCL you can make yourself a DIY solution. Matlab supports arbitrary code in MEX files. There you can write regular C or C++ code to interface with Matlab. This code can then also contain functions setting up OpenCL and calling kernels.
http://in.mathworks.com/matlabcentral/fileexchange/30109-opencl-toolbox-v0-17
- Badges
- Science topic