Accelerator Bindings for Partitioned Communication #4
Description
What
MPI language bindings for partitioned communication functions to be callable in an accelerator context.
Why
Performing MPI_Pready or MPI_Parrived operations from the accelerator context can provide an efficient/convenient way to overlap communication with computation on the accelerator.
Presentations
- Partitioned Communicaiton CUDA Bindings (Direct Link)
- 2021-11-03 -- Partitioned Communicaiton CUDA Bindings
- 2021-08-18 -- Partitioned Communicaiton CUDA Bindings