CUDA-based NumPy

CUDArray is a CUDA-accelerated subset of the NumPy library. The goal of CUDArray is to combine the easy of development from the NumPy with the computational power of Nvidia GPUs in a lightweight and extensible framework.

CUDArray currently imposes many limitations in order to span a manageable subset of the NumPy library. Nonetheless, it supports a neural network pipeline as demonstrated in the project deeppy.

Features

• Drop-in replacement for NumPy (limitations apply).
• Fast array operations based on cuBLAS, cuRAND and cuDNN.
• (somewhat) Simple C++/CUDA wrapper based on Cython.
• Extends NumPy with specialized functions for neural networks.
• CPU fall-back when CUDA is not available.

Installation

With CUDA back-end

First, you should consider specifying the following environment variables.

• INSTALL_PREFIX (default: /usr/local). Path where to install libcudarray. For the Anaconda Python distribution this should be /path/to/anaconda.
• CUDA_PREFIX (default: /usr/local/cuda). Path to the CUDA SDK organized in bin/, lib/, include/ folders.
• CUDNN_ENABLED. Set CUDNN_ENABLED to 1 to include cuDNN operations in libcudarray.

Then build and install libcudarray with

make
make install

Finally, install the cudarray Python package:

python setup.py install

Without CUDA back-end

Install the cudarray Python package:

python setup.py --without-cuda install

Documentation

Please consult the technical report for now. Proper documentation is on the TODO list.

Contact

Feel free to report an issue for feature requests and bug reports.

For a more informal chat, visit #cudarray on the freenode IRC network.

Citation

If you use CUDArray for research, please cite the technical report:

@techreport{larsen2014cudarray,
  author = "Larsen, Anders Boesen Lindbo",
  title = "{CUDArray}: {CUDA}-based {NumPy}",
  institution = "Department of Applied Mathematics and Computer Science, Technical University of Denmark",
  year = "2014",
  number = "DTU Compute 2014-21",
}

TODO

• Proper transpose support,
• Add functionality for copying from NumPy array to existing CUDArray array.
• FFT module based on cuFFT.
• Unit tests!
• Add documentation to wiki.
• Windows/OS X support.

Influences

Thanks to the following projects for inspiration.

• cudamat
• PyCUDA
• mshadow
• Caffe
• CUDPP