NVSHMEM Best Practice Guide

• NVSHMEM Best Practices Guide
• NVSHMEM Initialization
  • Two-Stage Initialization
• Device APIs
  • Device APIs on Peer-to-Peer Transport
  • Device APIs on Proxy-Based Transport
  • Device APIs on IBGDA Transport
• On-stream APIs
• Host APIs
• CUDA NVSHMEM Interoperability
  • Using CUDA Streams APIs
• NVSHMEM Runtime Configuration
  • Environment behavior
• NVSHMEM Unsupported Operations
  • Toolchaining
  • Operations
• NVSHMEM Performance
  • Using 16-Byte Alignment Buffers in the Application
  • Using nvshmem_*block for Heterogenous Transports Configuration
  • Tuning the queue-pair Type and Configuration for IBGDA