abelcarreras/cuda_functions

Python functions to calculate the FFT and autocorrelation function using GPU (Cuda)

CUDA functions

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.

Included functions

1. Fast Fourier transform 1D (Complex-to-complex)

• cuda_fft (single/double precision Fourier transfom)
• cuda_ifft (single/double precision inverse Fourier transfom)

2. Autocorrelation functions (Real-to-real and complex-to-complex)

• cuda_acorrelate (single/double precision autocorrelation function)

Installation

1. Requirements

• Python 2.7/3.4 or higher
• numpy
• cuda toolkit (http://docs.nvidia.com/cuda/)

2. Run setup.py script to compile only (testing)
   python setup.py build_ext --inplace

3. To install the module on your system use distutils setup.py as usual:
   python setup.py install --user

4. Now module is available on PyPI
   pip install cuda_functions

5. Run and check included python scripts as example

Contact info

Abel Carreras
abelcarreras83@gmail.com

Donostia International Physics Center (DIPC)
Donostia-San Sebastian (Spain)