Disable schema inference for unboxedOnly kernels (#27977)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27977
The only remaining reason why we couldn't move some ops from globalATenDispatch to the c10 dispatcher was that schema inference didn't support some use cases.
But actually, we don't need schema inference for these ops. By disabling it, we can move the remaining ops from globalATenDispatch to the c10 dispatcher.
ghstack-source-id: 92104807
Test Plan: waitforsandcastle
Differential Revision: D17929696
fbshipit-source-id: 05ec65b615487fde784293e3b533fa3ec09cf234
