显而易见但效率不高的解决方案是使用MPI\u Neighbor\u alltoall 以增加缓冲区大小为代价,你有什么建议吗?
没有这样的事情 MPI_Neighbor_allreduce 如果你想要它,请随时在 http://mpi-forum.org/
MPI_Neighbor_allreduce
顺便说一句,你是想用 MPI_Neighbor_allgather MPI_Neighbor_alltoall 为了实现邻居allreduce?
MPI_Neighbor_allgather
MPI_Neighbor_alltoall