Skip to content

WIP: CUDA implementation for computing a distance-1 parallel maximal independent sets

Hong Zhang requested to merge hongzh/cuda-pmis into main

Current status:

  • It works on SUMMIT.
  • It does not work on systems that do not have cuda-aware MPI support (e.g. without openmpi), but merging with !2205 (merged) can make the MPI communication across GPUs fall back to using CPU as a hub.
Edited by Satish Balay

Merge request reports