This fork is a stripped down version of neon, with only the convolution layer, no other layers or ... anything really. Just pure convolution. Also, no sass. Just Kepler convolutions.

The original neon, documentation etc is at https://github.com/nervanasystems/neon

Test code is at:

• test_perf.py
• test_correctness.py

It's going to contain a version of neon convolutions ported to OpenCL

This is started at:

https://github.com/hughperkins/winogradCl-underconstruction/blob/a6160f0fce407136e178df32417958a8ba77072a/neon/backends/kernels/cl/convolution_cl.py

Apache 2.0, per original neon