trunk/5e822a11ee228141bfa86331e9c3a34bdee4b486: [xpu][inductor] Add reduction heuristic for xpu INNER hint (#190986)
- PyTorch: 1081 events in the last 90 days
- PyTorch: 1068th Release in the last 90 days
- Previous: earlier the same day · trunk/1968f5b843b9b695588a871f82e3bcb76c667b0e
What happened
Re-opening #176049 (closed) rebased onto latest main . We see performance regression on xpu in some models because reduction hint changed from default to inner in the 2.11 release by bc4d0bf . Since the original PR was closed, torch/_inductor/runtime/triton_heuristics.py 's monolithic reduction/pointwise config logic was refactored into a per-device heuristic registry ( torch/_inductor/heuristics/triton_codegen/{red…
Summary assembled by rule from the sources below