Quantized FC operator (#19497)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/19497
Implements a basic quantized FC (uint8 * int8 -> uint8) with FBGEMM APIs.
Related document:
https://fb.quip.com/rP5tAx56ApMM
https://fb.quip.com/MU7aAbzGDesu
Work Item List:
1. [DONE] currently we use prepack routines inside Quantized FC operator. Will separate it as a standalone operator soon.
2. [DONE] rebase to D14817809 and D14994781 (cpp custom types).
3. [DONE] correctness unit test.
4. [To Do] rebase to QTensor. Similar to D14565413, this will be implemented in the next Diff.
Differential Revision: D14761865
fbshipit-source-id: 031a39915fecd947afb4dd2719112b4ddc1082d3
