trunk/0e653dc4c640a336dc77fb1171f01bf6ffabd2e1: [Inductor][AMD] Forward compiler options for Triton templates (#201352)
- PyTorch: 2582 events in the last 90 days
- PyTorch: 2547th Release in the last 90 days
- Previous: earlier the same day · trunk/f21bd8f7b7494c83a7f01da3a761c6c67865b0b6
What happened
Summary Make sure AMD compiler settings requested by Inductor reach Triton. Problem Inductor currently forwards these settings using a list supplied by optional TLX support. If that list is empty, the requested settings are dropped and Triton uses defaults instead. For example: Inductor requests: waves_per_eu=8 Triton receives: no waves_per_eu option Triton uses: waves_per_eu=0 (default) This PR Always forward matri…
Summary assembled by rule from the sources below