Closes the MPI state.

This function must be called by all processes before the program exits.

Returns true if the MPI state has been closed; otherwise returns false.

Returns the major and minor number of the MPI version.

Returns a string with the name of the processor on which the function was called. The string provides an identification for a specific piece of hardware, for example, the hostname of a node.

Predefined communicator that contains all processes.

Returns the rank of the process.

Returns the number of processes.

Performs a blocking send. A block of count elements with datatype datatype is read from buf and sent to the process with rank dest; unless dest is nil, in which case no send is performed. tag is a non-negative integer that identifies the message.

Performs a blocking receive. A block of count elements with datatype datatype is received from the process with rank source and stored in buf; unless source is nil, in which case no receive is performed. tag is a non-negative integer that identifies the message.

Performs a blocking send and receive. A block of sendcount elements with datatype sendtype is read from sendbuf and sent to the process with rank dest; unless dest is nil, in which case no send is performed. sendtag is a non-negative integer that identifies the sent message. A block of recvcount elements with datatype recvtype is received from the process with rank source and stored in recvbuf; unless source is nil, in which case no receive is performed. recvtag is a non-negative integer that identifies the received message.

Performs a blocking in-place send and receive. A block of count elements with datatype datatype is read from buf and sent to the process with rank dest; unless dest is nil, in which case no send is performed. sendtag is a non-negative integer that identifies the sent message. A block of count elements with datatype datatype is received from the process with rank source and stored in buf; unless source is nil, in which case no receive is performed.

Blocks the calling process until all processes have called the function.

Performs a broadcast operation. For each rank i in the communicator comm, the process with rank root sends a block of count elements with datatype datatype in buf to the process with rank i, which stores the values in buf.

Performs a gather operation. For each rank i in the communicator comm, the process with rank i sends a block of sendcount elements with datatype sendtype in sendbuf to the process with rank root, which stores the values in the i-th block of recvcount elements with datatype recvtype in recvbuf.

mpi.in_place may be passed as sendbuf at the process with rank root to perform the operation in place. In this case sendcount and sendtype are ignored, and the process does not gather elements from itself.

Performs a gather operation with varying block sizes. For each rank i in the communicator comm, the process with rank i sends a block of sendcount elements with datatype sendtype in sendbuf to the process with rank root, which stores the values in a block of recvcounts[i] elements at offset displs[i] with datatype recvtype in recvbuf.

mpi.in_place may be passed as sendbuf at the process with rank root to perform the operation in place. In this case sendcount and sendtype are ignored, and the process does not gather elements from itself.

Performs a scatter operation. For each rank i in the communicator comm, the process with rank root sends the i-th block of sendcount elements of datatype sendtype in sendbuf to the process with rank i, which stores the values in a block of recvcount elements with datatype recvtype in recvbuf.

mpi.in_place may be passed as recvbuf at the process with rank root to perform the operation in place. In this case recvcount and recvtype are ignored, and the process does not scatter elements to itself.

Performs a scatter operation with varying block sizes. For each rank i in the communicator comm, the process with rank root sends a block of sendcounts[i] elements at offset displs[i] with datatype sendtype in sendbuf to the process with rank i, which stores the values in a block of recvcount elements with datatype recvtype in recvbuf.

mpi.in_place may be passed as recvbuf at the process with rank root to perform the operation in place. In this case recvcount and recvtype are ignored, and the process does not scatter elements to itself.

Performs a gather-to-all operation. For each pair of ranks i and j in the communicator comm, the process with rank i sends a block of sendcount elements with datatype sendtype in sendbuf to the process with rank j, which stores the values in the i-th block of recvcount elements with datatype recvtype in recvbuf.

mpi.in_place may be passed as sendbuf at all processes to perform the operation in place. In this case sendcount and sendtype are ignored, and the process with rank i sends the i-th block of recvcount elements with datatype recvtype in recvbuf to the process with rank j for each pair of different ranks i and j.

Performs a gather-to-all operation with varying block sizes. For each pair of ranks i and j in the communicator comm, the process with rank i sends a block of sendcount elements with datatype sendtype in sendbuf to the process with rank j, which stores the values in a block of recvcounts[i] elements at offset displs[i] with datatype recvtype in recvbuf.

mpi.in_place may be passed as sendbuf at all processes to perform the operation in place. In this case sendcount and sendtype are ignored, and the process with rank i sends the i-th block of recvcounts[i] elements at offset displs[i] with datatype recvtype in recvbuf to the process with rank j for each pair of different ranks i and j.

Performs an all-to-all scatter/gather operation. For each pair of ranks i and j in the communicator comm, the process with rank i sends the j-th block of sendcount elements with datatype sendtype in sendbuf to the process with rank j, which stores the values in the i-th block of recvcount elements with datatype recvtype in recvbuf.

mpi.in_place may be passed as sendbuf at all processes to perform the operation in place. In this case sendcount and sendtype are ignored, and the process with rank i sends the j-th block of recvcount elements with datatype recvtype in recvbuf to the process with rank j for each pair of different ranks i and j.

The in-place variant of this function is available with MPI-2.2.

Performs an all-to-all scatter/gather operation with varying block sizes. For each pair of rank i and j in the communicator comm, the process with rank i sends a block of sendcounts[j] elements at offset sdispls[j] with datatype sendtype in sendbuf to the process with rank j, which stores the values in a block of recvcounts[i] elements at offset rdispls[i] with datatype recvtype in recvbuf.

mpi.in_place may be passed as sendbuf at all processes to perform the operation in place. In this case sendcounts, sdispls and sendtype are ignored, and the process with rank i sends a block of recvcounts[j] elements at offset rdispls[j] with datatype recvtype in recvbuf to the process with rank j for each pair of different ranks i and j.

The in-place variant of this function is available with MPI-2.2.

Performs an all-to-all scatter/gather operation with varying block sizes and datatypes. For each pair of rank i and j in the communicator comm, the process with rank i sends a block of sendcounts[j] elements at offset in bytes sdispls[j] with datatype sendtypes[j] in sendbuf to the process with rank j, which stores the values in a block of recvcounts[i] elements at offset in bytes rdispls[i] with datatype recvtypes[i] in recvbuf.

mpi.in_place may be passed as sendbuf at all processes to perform the operation in place. In this case sendcounts, sdispls and sendtypes are ignored, and the process with rank i sends a block of recvcounts[j] elements at offset in bytes rdispls[j] with datatype recvtypes[j] in recvbuf to the process with rank j for each pair of different ranks i and j.

The in-place variant of this function is available with MPI-2.2.

Performs a reduce operation. Blocks of count elements with datatype datatype in sendbuf of all processes in the communicator comm are combined element-wise using the reduction operator op, and the result is stored in a block of count elements with datatype datatype in recvbuf of the process with rank root.

mpi.in_place may be passed as sendbuf at the process with rank root to perform the operation in place. In this case the elements of the process are read from recvbuf.

Performs a reduce-to-all operation. Blocks of count elements with datatype datatype in sendbuf of all processes in the communicator comm are combined element-wise using the reduction operator op, and the result is stored in a block of count elements with datatype datatype in recvbuf of all processes.

mpi.in_place may be passed as sendbuf at all processes to perform the operation in place. In this case the elements of all processes are read from recvbuf.

Performs an inclusive scan operation. For each rank i in the communicator comm, blocks of count elements with datatype datatype in sendbuf of the processes with ranks 0 to i are combined element-wise using the reduction operator op, and the result is stored in a block of count elements with datatype datatype in recvbuf of the process with rank i.

mpi.in_place may be passed as sendbuf at all processes to perform the operation in place. In this case the elements of all processes are read from recvbuf.

Performs an exclusive scan operation. For each rank i > 0 in the communicator comm, blocks of count elements with datatype datatype in sendbuf of the processes with ranks 0 to i - 1 are combined element-wise using the reduction operator op, and the result is stored in a block of count elements with datatype datatype in recvbuf of the process with rank i.

mpi.in_place may be passed as sendbuf at all processes to perform the operation in place. In this case the elements of all processes are read from recvbuf.

Returns a new datatype constructed from a sequence of count elements of the given datatype.

Returns a new datatype constructed from a sequence of count blocks of elements of the given datatype. blocklength specifies the number of elements in each block. stride specifies the number of elements between the first elements of consecutive blocks.

Returns a new datatype constructed from a sequence of count blocks of elements of the given datatype. blocklength specifies the number of elements in each block. displacements specifies for each block the number of elements between the first elements of that block and of the first block.

Returns a new datatype constructed from a sequence of count blocks of elements of the given datatype. blocklengths specifies for each block the number of elements in that block. displacements specifies for each block the number of elements between the first elements of that block and of the first block.

Returns a new datatype constructed from a sequence of count blocks of elements. blocklengths specifies for each block the number of elements in that block. displacements specifies for each block the offset in bytes of that block relative to the start of the datatype. datatypes specifies for each block the datatype of the elements in that block.

Returns a new datatype constructed from the given datatype with lower bound lb and size extent in bytes.

This function may be used to adjust for padding at the beginning or end of a datatype.

Commits the datatype, after which it may be used for communications.

Returns the lower bound and size of the datatype in bytes.

Returns a new communicator that maps the processes in comm to a Cartesian grid. dims is a sequence that specifies the number of processes for each dimension of the grid. periods is a sequence that specifies for each dimension whether the grid is periodic (true) or non-periodic (false) along that dimension. If reorder is true, the ranks of the processes in the new communicator may be reordered to optimally map the Cartesian topology onto the machine topology; if reorder is false, the ranks of the processes remain the same as in comm.

If the size of the Cartesian grid is smaller than the number of processes in comm, the function returns nil on some processes.

Returns three sequences that specify for each dimension of the Cartesian grid the number of processes, whether the grid is periodic (true) or non-periodic (false) along that dimension, and the coordinate of the calling process along that dimension.

Returns a sequence with the Cartesian coordinates of the process with the given rank.

Returns the rank of the process with the given Cartesian coordinates.

Returns the ranks of the source and destination processes in a Cartesian shift, which involves a send to the destination process in the given direction and a receive from the source process in the opposite direction. direction specifies the coordinate dimension along which the shift is performed. disp specifies the displacement of the destination process relative to the calling process, either a positive integer for an upwards shift or a negative integer for a downwards shift.

If the shift to the calling process would cross a non-periodic boundary, the function returns nil inplace of a source rank; if the shift from the calling process would cross a non-periodic boundary, the function returns nil inplace of a destination rank.