Split IGamma cuda kernel into it's own file to speed up compilation times. (#47401)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/47401
Test Plan: Imported from OSS
Reviewed By: mruberry
Differential Revision: D24740657
Pulled By: gchanan
fbshipit-source-id: 78244dba8624ca7be8761a8f4bf1aa078602e5cc
