all-reduce operation designed for smp clusters
It uses 2-layer communication: binomial for intra-communication
and rdb for inter-communication*/
all-reduce operation designed for smp clusters
It uses 2-layer communication: binomial for intra-communication
and rdb for inter-communication*/
/* ** NOTE **
Use -DMPICH2 if this code does not compile.
MPICH1 code also work on MPICH2 on our cluster and the performance are similar.
/* ** NOTE **
Use -DMPICH2 if this code does not compile.
MPICH1 code also work on MPICH2 on our cluster and the performance are similar.