This library uses the cuda kernels from Alex Khrizevsky's cuda-convnet, so there's no point in benchmarking it.

Kernels located here: https://github.com/TorontoDeepLearning/convnet/blob/master/src/cudamat_conv_kernels.cuh