SET: Stream-Event-Triggered Scheduling for Efficient CUDA Graph Pipelines

This feed contains summaries of multiple June 2026 arXiv papers across systems, ML, and crypto. Major items: SET introduces a CUDA runtime combining multi-stream event-chaining and graph-based per-stream buffers to reduce host-device sync and scheduling overheads for GPU pipelines; NVSHMEM is analyzed at the system level for GPU device-initiated one-sided communication and symmetric memory; several GPU/CUDA works explore integer-precision division and automated C++→CUDA porting using an LLM-driven Deopt-Reopt workflow. In ML/serving, SlidingServe proposes an SLO-aware sliding-window scheduler,