pytorch

925a3788 - [CUDA] Switch to `at::empty` in `max_pool3d_with_indices_backward_cuda` (#100138)

[CUDA] Switch to `at::empty` in `max_pool3d_with_indices_backward_cuda` (#100138) Looks like there's an extraneous call to `at::zero` as `gradInput` will always be zero'd by `max_pool3d_with_indices_backward_out_cuda_template`. CC @ptrblck @ngimel Pull Request resolved: https://github.com/pytorch/pytorch/pull/100138 Approved by: https://github.com/ngimel