This document shows how to prepare your system for GPU accelerated DisVis for MacOSX, using the OpenCL framework. Note though, that it requires a high-end graphics card to get a proper acceleration, else the speedup will be neglible. Practically this means that it should be either an AMD or NVIDIA GPU; if it is a power efficient Intel card, usually present on laptops, installing the GPU version will most likely not help much.
- An OpenCL1.1+ enabled GPU (this is for current Macs usually the case)
- brew (MacOSX package manager)
- git
- pip
First install Python bindings to OpenCL with the pyopencl Python package
pip install --upgrade pyopencl
Note that this will upgrade all dependencies required for pyopencl, such as NumPy.
Next comes the somewhat more tedious part of installing a high-performance FFT library for GPU using the OpenCL framework clFFT. This requires the cmake program first
brew install cmake
Next download the source code of clFFT and build the library
git clone https://github.com/clMathLibraries/clFFT.git
cd clFFT/src && cmake CMakeLists.txt && make && cd ../../
Python bindings to to the clFFT library are provided with the gpyfft package
git clone https://github.com/geggo/gpyfft.git
cd gpyfft
In the folder of gpyfft there is a file called setup.py. Open it with a text editor and change the directory that points to the clFFT directory. Then type
sudo python setup.py install
If it all worked out, congratulations, your system is now ready for GPU calculations!