Improve legacy QuantizedLinear functions to reduce overhead (#29773)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29773
Improve legacy QuantizedLinear functions to reduce overhead.
Separate from the stack of D18381988.
Test Plan: buck test mode/dev-nosan //caffe2/test:jit -- "quant"
Reviewed By: lly-zero-one
Differential Revision: D18494988
fbshipit-source-id: 5627d7e8b0b7a750852eead9e28c5a9b3fa70559
