Fake Quantization Per Channel Kernel Core Implementation (CPU) (#41037)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/41037
This diff contains the core implementation for the fake quantizer per channel kernel that supports back propagation on the scale and zero point.
Test Plan:
On a devvm, use:
- `buck test //caffe2/test:quantization -- learnable_forward_per_channel`
- `buck test //caffe2/test:quantization -- learnable_backward_per_channel`
Reviewed By: z-a-f
Differential Revision: D22395665
fbshipit-source-id: 280c2405d04adfeda9fb9cfc94d89e8d868e0d41