tree 49286a35d776f64d5300e937817a238c8585cb8f
parent 18eb4bbdf9563c0620bbc93daa045c2258b63bde
author Sam Gross <sgross@fb.com> 1507588738 -0700
committer Zachary DeVito <zdevito@gmail.com> 1507771687 -0700

Generate PyTorch-style NN bindings

This generates NN bindings with a similar interface to PyTorch's
torch.nn.functional package. The file nn.yaml specifies function
signatures and THNN implementations.

Each NN operation generates three functions. For example:

  - conv2d
  - conv2d_forward
  - conv2d_backward

The conv2d and conv2d_forward functions differ in how they handle
buffers that need to be passed to the backward function. conv2d_forward
takes the buffers as parameters. conv2d creates the buffers internally
and discards them.