Python library using the Futhark C backend via CFFI
Futhark provides several compiler backends, for example futhark opencl
which is a C backend, and futhark pyopencl
which is a Python
backend based on PyOpenCL. However, the host-side code of the Python
backend is quite slow, leading to a lot of overhead when small,
frequent kernels are used.
A solution to reduce this overhead is to use CFFI to used the C backend from Python, greatly reducing the calling overhead. The OpenCL code is the same, so this is not interesting for long-running kernels.
This library supports the following Futhark backends: c
, opencl
,
multicore
, and cuda
.
Futhark arrays are mapped to and from Numpy arrays. Multiple outputs and multi-dimensional arrays are supported.
Install Futhark, then simply
pip install futhark-ffi
Generate a C library, and build a Python binding for it
futhark opencl --library test.fut
build_futhark_ffi test
Use the Python wrapper
import numpy as np
import _test
from futhark_ffi import Futhark
test = Futhark(_test)
res = test.test3(np.arange(10))
test.from_futhark(res)