Efficient GPU kernels for block-sparse matrix multiplication and convolution

Automatically exported from code.google.com/p/cuda-convnet2

A GPU implementation of Convolutional Neural Nets in C++

Torch7 bindings for cuda-convnet2 kernels!