Skip to content

WIP: Add CUDA streams to GPU communication PackInfo

Proposed modifications:

  • Add support for generic allocators in MPI buffers. Using allocators enables allocation of page-locked memory in MPI buffers, which is required for asynchronous CUDA memory copies (which use streams).
  • Add support for asynchronous operations using CUDA streams in GPUPackInfo

Issue: #27 (closed)

Edited by João Victor Tozatti Risso

Merge request reports