trunk/298a1c806b63a44afe7e5d120f281dda30daf851: flex_gemm: rename the tuned kernel option to autotune (#201088)
- PyTorch: 2454 events in the last 90 days
- PyTorch: 2421th Release in the last 90 days
- Previous: earlier the same day · trunk/09673565957c49fb174ece4a84187cd98f9d5962
What happened
Human Note Change to better name Agent note kernel_options={"tuned": True} becomes kernel_options={"autotune": True} . The option names the action (benchmark candidate configs), matching Inductor's max_autotune vocabulary. FlexGEMM is a prototype, so there is no alias: the existing unknown-option check rejects tuned with a clear error. NVGEMM now follows the same contract as QUACK. Previously NVGEMM always benchmark…
Summary assembled by rule from the sources below