Hi, I am developing an mpi-based library. We are heavily relied on the one-sided operations and GPUs. If I understand correctly, according the FAQ page https://www.open-mpi.org/faq/?category=runcuda <https://www.open-mpi.org/faq/?category=runcuda>: * OpenMPI>2.0 supports CUDA-aware MPI through UCX * UCX + OpenMPI does not support ANY one-sided operations yet. * 1.8 < OpenMPI < 2.0 DO supports three basic One-Sided Operations * OpenMPI does plan to support CUDA-aware One-sided operations in the future.
However, through the web-site I cannot find any timeline for that. So I want to ask is there any plan recently to Support CUDA-aware One-sided operations? If yes, there is any estimated time-line for it? Thanks, Bicheng