Open
Description
masked_fill
is the point-wise operator so porting if from the TH code to Aten (and TensorIterator) expected to be easy. Such migration will help to clean up the code, simplify dispatch as well as provide immediate 2-3x operator performance gain.
Porting guide: https://github.com/pytorch/pytorch/wiki/TH-to-ATen-porting-guide
Example PR with porting of the adaptive_avg_pool2d: #14714
How to use TensorIterator: https://github.com/pytorch/pytorch/wiki/How-to-use-TensorIterator
Metadata
Metadata
Assignees
Labels
Relatively self-contained tasks for better engineering contributorsCPU specific problem (e.g., perf, algorithm)Related to torch.cuda, and CUDA support in generalIssues related to porting TH/THNN legacy to ATen nativeThis issue has been looked at a team member, and triaged and prioritized into an appropriate module