pytorch

f3e014f3 - [iOS GPU][Perf][3/n] Cache the compuation pipeline state object (#57666)

[iOS GPU][Perf][3/n] Cache the compuation pipeline state object (#57666) Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/57666 Got 5% improvement on mobilenetv2 and Unet 1. `std::unordered_map` is faster than `NSMutableDictionary` 2. `std::string` is cheaper than `NSString` ghstack-source-id: 128338531 Test Plan: CI Reviewed By: kimishpatel, SS-JIA Differential Revision: D28048992 fbshipit-source-id: fc4f7e41928c524acde48947d2cd6b9f6ef7cbc8