https://hgpu.org/?p=16759
A Metaprogramming and Autotuning Framework for Deploying Deep Learning Applications