4.3.1. Type Matching Rules

One can think of message transfer as consisting of the following three phases.

1. Data is pulled out of the send buffer and a message is assembled.
2. A message is transferred from sender to receiver.
3. Data is pulled from the incoming message and disassembled into the receive buffer.

To define type matching more precisely, we need to deal with two issues: matching of types of the host language with types specified in communication operations; and matching of types at sender and receiver.

The types of a send and receive match (phase two) if both operations use identical names. That is, MPI_INTEGER matches MPI_INTEGER, MPI_REAL matches MPI_REAL, and so on. There is one exception to this rule, discussed in Section Pack and Unpack: the type MPI_PACKED can match any other type.

The type of a variable in a host program matches the type specified in the communication operation if the datatype name used by that operation corresponds to the basic type of the host program variable. For example, an entry with type name MPI_INTEGER matches a Fortran variable of type INTEGER. A table giving this correspondence for Fortran and C appears in Section Message Data. There are two exceptions to this last rule: an entry with type name MPI_BYTE or MPI_PACKED can be used to match any byte of storage (on a byte-addressable machine), irrespective of the datatype of the variable that contains this byte. The type MPI_PACKED is used to send data that has been explicitly packed, or receive data that will be explicitly unpacked, see Section Pack and Unpack. The type MPI_BYTE allows one to transfer the binary value of a byte in memory unchanged.

To summarize, the type matching rules fall into the three categories below.

• Communication of typed values (e.g., with datatype different from MPI_BYTE), where the datatypes of the corresponding entries in the sender program, in the send call, in the receive call and in the receiver program must all match.
• Communication of untyped values (e.g., of datatype MPI_BYTE), where both sender and receiver use the datatype MPI_BYTE. In this case, there are no requirements on the types of the corresponding entries in the sender and the receiver programs, nor is it required that they be the same.
• Communication involving packed data, where MPI_PACKED is used.

Example Sender and receiver specify matching types.

This code is correct if both a and b are real arrays of size . (In Fortran, it might be correct to use this code even if a or b have size < 10: e.g., when a(1) can be equivalenced to an array with ten reals.)

Example Sender and receiver do not specify matching types.

This code is erroneous, since sender and receiver do not provide matching datatype arguments.

Example Sender and receiver specify communication of untyped values.

This code is correct, irrespective of the type and size of a and b (unless this results in an out of bounds memory access).

Advice to users.

If a buffer of type MPI_BYTE is passed as an argument to MPI_SEND, then MPI will send the data stored at contiguous locations, starting from the address indicated by the buf argument. This may have unexpected results when the data layout is not as a casual user would expect it to be. For example, some Fortran compilers implement variables of type CHARACTER as a structure that contains the character length and a pointer to the actual string. In such an environment, sending and receiving a Fortran CHARACTER variable using the MPI_BYTE type will not have the anticipated result of transferring the character string. For this reason, the user is advised to use typed communication operations whenever possible. ( End of advice to users.)

4.3.1.1. Type MPI_CHARACTER

The type MPI_CHARACTER matches one character of a Fortran variable of type CHARACTER, rather than the entire character string stored in the variable. Fortran variables of type CHARACTER or substrings are transferred as if they were arrays of characters. This is illustrated in the example below.

Example Transfer of Fortran CHARACTERs.

The last five characters of string b at the MPI process with rank = 1 are replaced by the first five characters of string a at the MPI process with rank = 0.

Rationale.

The alternative choice would be for MPI_CHARACTER to match a character of arbitrary length. This runs into problems.

A Fortran character variable is a constant length string, with no special termination symbol. There is no fixed convention on how to represent characters, and how to store their length. Some compilers pass a character argument to a routine as a pair of arguments, one holding the address of the string and the other holding the length of string. Consider the case of an MPI communication call that is passed a communication buffer with type defined by a derived datatype (Section Derived Datatypes). If this communicator buffer contains variables of type CHARACTER then the information on their length will not be passed to the MPI routine.

This problem forces us to provide explicit information on character length with the MPI call. One could add a length parameter to the type MPI_CHARACTER, but this does not add much convenience and the same functionality can be achieved by defining a suitable derived datatype. ( End of rationale.)

Advice to implementors.

Some compilers pass Fortran CHARACTER arguments as a structure with a length and a pointer to the actual string. In such an environment, the MPI call needs to dereference the pointer in order to reach the string. ( End of advice to implementors.)