This is a collection of python functions written with CUDA, using cuFFT and cuBLAS libraries. The interface with Python is written using the Python C API.
These functions intend to mimic the behavior of numpy functions: fft and correlate using the power of GPU.
- Fast Fourier transform 1D (Complex-to-complex)
- cuda_fft (single/double precision Fourier transfom)
- cuda_ifft (single/double precision inverse Fourier transfom)
- Autocorrelation functions (Real-to-real and complex-to-complex)
- cuda_acorrelate (single/double precision autocorrelation function)
- Requirements
- Python 2.7/3.4 or higher
- numpy
- cuda toolkit (http://docs.nvidia.com/cuda/)
-
Run setup.py script to compile only (testing)
python setup.py build_ext --inplace -
To install the module on your system use distutils setup.py as usual:
python setup.py install --user -
Now module is available on PyPI
pip install cuda_functions -
Run and check included python scripts as example
Abel Carreras
[email protected]
Donostia International Physics Center (DIPC)
Donostia-San Sebastian (Spain)