13.3.5. Request-based RMA Communication Operations

Request-based RMA communication operations allow the user to associate a request handle with the RMA operations and test or wait for the completion of these requests using the functions described in Section Communication Completion. Request-based RMA operations are only valid within a passive target epoch (see Section Synchronization Calls).

Upon returning from a completion call in which an RMA operation completes, all fields of the status object, if any, and the results of status query functions (e.g., MPI_GET_COUNT) are undefined with the exception of MPI_ERROR if appropriate (see Section Return Status). It is valid to mix different request types (e.g., any combination of RMA requests, collective requests, I/O requests, generalized requests, or point-to-point requests) in functions that enable multiple completions (e.g., MPI_WAITALL). It is erroneous to call MPI_REQUEST_FREE or MPI_CANCEL for a request associated with an RMA operation. RMA requests are not persistent.

The closing of the epoch, or explicit bulk synchronization using MPI_WIN_FLUSH, MPI_WIN_FLUSH_ALL, MPI_WIN_FLUSH_LOCAL, or MPI_WIN_FLUSH_LOCAL_ALL, also indicates completion of request-based RMA operations on the specified window. However, users must still free the request by testing, waiting, or calling MPI_REQUEST_FREE on the request handle to allow the MPI implementation to release any resources associated with these requests.

MPI_RPUT is similar to MPI_PUT (Section Put), except that it allocates a communication request object and associates it with the request handle (the argument request). The completion of the operation at the origin (i.e., after the corresponding test or wait) indicates that the sender is now free to update the locations in the origin buffer. It does not indicate that the data is available at the target window. If remote completion is required, MPI_WIN_FLUSH, MPI_WIN_FLUSH_ALL, MPI_WIN_UNLOCK, or MPI_WIN_UNLOCK_ALL can be used.

MPI_RGET is similar to MPI_GET (Section Get), except that it allocates a communication request object and associates it with the request handle (the argument request) that can be used to wait or test for completion of the operation at the origin, which indicates that the data is available in the origin buffer. If origin_addr points to memory attached to a window, then the data becomes available in the private copy of this window.

MPI_RACCUMULATE is similar to MPI_ACCUMULATE (Section Accumulate Functions), except that it allocates a communication request object and associates it with the request handle (the argument request) that can be used to wait or test for completion. The completion of the operation at the origin (i.e., after the corresponding test or wait) indicates that the origin buffer is free to be updated. It does not indicate that the operation has completed at the target window.

MPI_RGET_ACCUMULATE is similar to MPI_GET_ACCUMULATE (Section Get Accumulate), except that it allocates a communication request object and associates it with the request handle (the argument request) that can be used to wait or test for completion. The completion of the operation at the origin (i.e., after the corresponding test or wait) indicates that the data is available in the result buffer and the origin buffer is free to be updated. It does not indicate that the operation has been completed at the target window.