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
This aggregation scheme is intended to allow OpenMPI to transfer larger messages
if the user-reported partitions are too small or too high in number.
This is achieved by using an internal partitioning where each internal (transfer)
partition corresponds to one or multiple user-reported partitions.
The scheme provides a method to mark a user partition as ready, that optionally
returns a transfer partition that is ready.
This is implemented by associating each transfer partition with an atomic counter,
tracking the number of corresponding pready-calls already made.
If a counter reaches the number of corresponding user-partitions, the corresponding
transfer partition is returned.
This implementation is thread-safe.
Signed-off-by: Axel Schneewind <axel.schneewind@hlrs.de>
0 commit comments