trunk/bf06d76b78b5735e759f6f2f4668c4baca964eee: [inductor] Make CUDA graph autotuning measurements consistent (#194649)
- PyTorch: 1835 events in the last 90 days
- PyTorch: 1811th Release in the last 90 days
- Previous: earlier the same day · ciflow/trunk/199479: Enable bugprone-unused-local-non-trivial-variable and remove hits
What happened
Human commentary: I want every candidate in an autotune decision measured under the same launch policy so the selected kernel reflects production execution. AI-assisted content: Problem Fixes #198171 Autotune requests can be created under one Inductor configuration and executed later in another process. Without carrying the effective CUDA-graph policy on the request, candidates from the same decision can be timed un…
Summary assembled by rule from the sources below