PARALLEL.RU - Информационно-аналитический центр по параллельным вычислениям

7.3.6. Exclusive Scan

MPI-1 provides an inclusive scan operation. The exclusive scan is described here.

MPI_EXSCAN(sendbuf, recvbuf, count, datatype, op, comm)
[ IN sendbuf] starting address of send buffer (choice)
[ OUT recvbuf] starting address of receive buffer (choice)
[ IN count] number of elements in input buffer (integer)
[ IN datatype] data type of elements of input buffer (handle)
[ IN op] operation (handle)
[ IN comm] intracommunicator (handle)

int MPI_Exscan(void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)

MPI_EXSCAN(SENDBUF, RECVBUF, COUNT, DATATYPE, OP, COMM, IERROR)
<type> SENDBUF(*), RECVBUF(*)
INTEGER COUNT, DATATYPE, OP, COMM, IERROR

void MPI::Intracomm::Exscan(const void* sendbuf, void* recvbuf, int count, const MPI::Datatype& datatype, const MPI::Op& op) const

MPI_EXSCAN is used to perform a prefix reduction on data distributed across the group. The value in recvbuf on the process with rank 0 is undefined, and recvbuf is not signficant on process 0. The value in recvbuf on the process with rank 1 is defined as the value in sendbuf on the process with rank 0. For processes with rank i > 1, the operation returns, in the receive buffer of the process with rank i, the reduction of the values in the send buffers of processes with ranks (inclusive). The type of operations supported, their semantics, and the constraints on send and receive buffers, are as for MPI_REDUCE.

No ``in place'' option is supported.

[] Advice to users.

As for MPI_SCAN, MPI does not specify which processes may call the operation, only that the result be correctly computed. In particular, note that the process with rank 1 need not call the MPI_Op, since all it needs to do is to receive the value from the process with rank 0. However, all processes, even the processes with ranks zero and one, must provide the same op. ( End of advice to users.)

[] Rationale.

The exclusive scan is more general than the inclusive scan provided in MPI-1 as MPI_SCAN. Any inclusive scan operation can be achieved by using the exclusive scan and then locally combining the local contribution. Note that for non-invertable operations such as MPI_MAX, the exclusive scan cannot be computed with the inclusive scan.

The reason that MPI-1 chose the inclusive scan is that the definition of behavior on processes zero and one was thought to offer too many complexities in definition, particularly for user-defined operations. ( End of rationale.)