pytorch

c3550d83 - Add fast path for BF16 kernel if all the operations within the kernel support bf16 (#99814)

Add fast path for BF16 kernel if all the operations within the kernel support bf16 (#99814) Pull Request resolved: https://github.com/pytorch/pytorch/pull/99814 Approved by: https://github.com/jgong5, https://github.com/jansel