You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
coll/base: do not send and receive 0 byte message for linear alltoallv
Prior to this change, the linear algorithm of alltoallv will post
send and receive even when sendcount and recvcount is 0.
This patch make the algorithm to skip 0 byte send/receive.
This patch works because MPI standard has strict requirement on
collective message size.
According to MPI standard's Collectives Introduction and Overview
(6.1) section:
For collective operations, the amount of data sent must
exactly match the amount of data specified by
the receiver.
Signed-off-by: Wei Zhang <wzam@amazon.com>
0 commit comments