2020 - mpiwg-hybrid/hybrid-issues GitHub Wiki
Possible Future Topics
- HIP Streams, Stream triggered comms (AMD Speaker)
- OneAPI, libfabric (HMEM) (Intel Speaker)
- Olga Pearce – GPU communication bottlenecks @LLNL (e.g. COMB)
- Ryan Grant – Challenges for applications
- E.g. Determining how MPI library will handle GPU communications
- Jim Dinan – NCCL and NVSHMEM: GPU Comms Best Practices
- Pavan Balaji – Proposal to pass stream to MPI
- User Defined Memory Regions (UMR) (Mellanox Speaker)
Past Topics
- Kickoff and Stream-Based Communication Discussion (Jim Dinan) [slides]
- Callback-based Completion Notification (Joseph Schuchart and Joachim Protze)
- Fibers are not (P)Threads: The Case for Loose Coupling of Asynchronous Programming Models and MPI Through Continuations [EuroMPI ‘20]
- MPI Detach - Asynchronous Local Completion [EuroMPI ‘20]
- CUDA Streams and Graphs, GPUDirect Async, GPUDirect Verbs (Stephen Jones, NVIDIA)
- ECP Recap - Feedback from ECP discussions (Ryan Grant and others)