[iOS GPU][BE][2/n] - Use dispatcher in MPSCNNTests.mm (#53429)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/53429
Call the testing ops through dispatcher instead of calling them through `at::native`. Some metal ops can't be called through dispatcher yet. For example, `at::t` will call `at::as_strided` which hasn't been implemented on metal yet. For those ops, we'll skip and call `mpscnn::`directly. We'll convert those ops once we have implemented the missing ops.
ghstack-source-id: 123038068
Test Plan:
- Sandcastle CI
- Circle CI
- AIBench/Mobilelab
Reviewed By: SS-JIA, AshkanAliabadi
Differential Revision: D26683366
fbshipit-source-id: bf130b191046f5d9ac9b544d512bc6cb94f08c09