viable/strict/1792837579: Use vectorized kernels for index_copy (#201814)
- PyTorch: 2750 events in the last 90 days
- PyTorch: 2707th Release in the last 90 days
- Previous: earlier the same day · viable/strict/1792834014
What happened
Linked issue or supporting maintainer Supported by @ngimel BC-breaking? No. Summary (human written only) Use vectorized kernel for scatter and index_copy ops, redispatch index_copy to scatter when profitable Cap the number of blocks because that leads to better perf (revisit later to maybe use CLC) Agent details (AI-generated) Authored with assistance from Codex. Adds a copy specialization to the existing native vec…
Summary assembled by rule from the sources below