Skip to content
J.L Stevens edited this page Apr 17, 2015 · 1 revision

Stonesoup setup:

First add CUDA to the path:

export CUDA_ROOT=/opt/cuda/
export PATH=$PATH:/opt/cuda/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/cuda/lib

Having added CUDA to the path we can install the Python dependencies:

pip install scipy pycuda cffi
pip install https://github.com/lebedov/scikits.cuda/tree/cffi

Temporary hack to add includes such that cffi can find them, file scikits-cuda/cuda/_cusparse_cffi_autogen.py

    ffi_lib = ffi.verify(__verify_scr, libraries=['cusparse'],
                         include_dirs=['/usr/local/cuda/include',
                                       '/opt/cuda/include'],
                         library_dirs=['/usr/local/cuda/lib64/',
                                       '/opt/cuda/lib64/'])

To get weights from the GPU into CF form:

lgnon = topo.sim.V1.LGNOnAfferent.weights_gpu.todense().get()
Image(lgnon[1296, :].reshape(60,60))