trunk/3f83700b397fdb04beb392d6b55fde076fe75083: Fix CUDA SDPA grid overflow for large batches and heads (#199594)
- PyTorch: 1899 events in the last 90 days
- PyTorch: 1875th Release in the last 90 days
- Previous: earlier the same day · ciflow/inductor/199719
What happened
Human Note For batch and head dims that exceed the grid cap, we just launch multiple kernels used some features found in #177985 Agent Report Agent details Prepared with AI assistance. CUDA grid.y/grid.z cannot exceed 65,535 blocks. Chunk the memory-efficient forward/backward launch grids while retaining full tensor dimensions and logical batch/head offsets. Kernels write directly into final outputs, preserving GQA…
Summary assembled by rule from the sources below