Short circuits DistributedDataParallel._recursive_to's copy and stream syncs if input is already on the right device (#55624)
Summary:
^
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55624
Reviewed By: pbelevich, agolynski
Differential Revision: D27836170
Pulled By: rohan-varma
fbshipit-source-id: 954bf336d70f9e80c045a6a96c1d8843c7f1cf2c