viable/strict/1791434203: [CUDA] Add fork/join execution on CUDA streams (#199128)
- PyTorch 近 90 天出现 2085 次
- PyTorch 近 90 天第 2056 次发版
- 上一次：同一天稍早 · ciflow/inductor/199800: [input_capture] Add an AOTAutograd compiled-graph hook (#199800)
发生了什么
Authored with an AI assistant. Add torch.cuda.execute_on_streams for fork/join execution with ordinary streams, green-context streams, or mixtures. Queued work is joined and caller state restored even when a callback raises. Same-device execution switches directly between streams and restores the caller once to reduce host-side overheads. Cross-device execution uses stream context managers to restore device and stre…
摘要按规则整理自下方来源原文