An attempt to bridge petsc4py and PyCUDA.
At the time of writing,
there is no way to access
the underlying device memory
of a cusp vector
from petsc4py.
This project attempts to add this functionality,
making the memory accessible as a PyCUDA GPUArray.
To use, you should have the following environment variables
defined (for setup.py):
THRUST_DIRCUSP_DIRPETSC_DIRPETSC_ARCH
And then, just make build should create the extension GPUArray.so.
The tests in test_GPUArray.py
should give you a fairly good idea
of how things work.
This project is based on the following works; thanks to the respective authors for the amazing software they develop:
-
petsc4py: https://bitbucket.org/petsc/petsc4py -
PyCUDA: https://github.com/inducer/pycuda -
Cython: http://cython.org/ -
setup.pyis based on that from the projectnpcuda-example: https://github.com/rmcgibbo/npcuda-example