Add autograd for layer_norm on CPU (#20883)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20883
Add autograd for layer_norm on CPU, after this diff, both PyTorch and jit model can automatically benefit from performance improvement of nn.functional.layer_norm
Reviewed By: zheng-xq
Differential Revision: D15483790
fbshipit-source-id: 94ed3b16ab6d83ca6c254dbcfb224ff7d88837f3