fused int8 static (#73452)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/73452
Added a fused Int8FC path using PackAWithQuantRowOffset like the INT8 dynamic path. There are two ways to enable it
(1) set an positive "X_scale" value in the arg list of Int8FC op
(2) send both "Qparam" (for output requantization, could be dummy values) and "in_Qparam" (for fused input quantization)
Differential Revision: D34034681
fbshipit-source-id: f25ca8a2b783ea597389d31c110448d19610218e
(cherry picked from commit 6fa10ba0e3be2d46298b439fba0fe9ae7e329f3a)