Constants(3) Meaning of MPI

DATA TYPES

Note that the Fortran types should only be used in Fortran programs, and the C types should only be used in C programs. For example, it is in error to use MPI_INT for a Fortran INTEGER. Datatypes are of type MPI_Datatype in C and of tyep INTEGER in Fortran.

C DATATYPES

MPI_CHAR
- char
MPI_BYTE
- See standard; like unsigned char
MPI_SHORT
- short
MPI_INT
- int
MPI_LONG
- long
MPI_FLOAT
- float
MPI_DOUBLE
- double
MPI_UNSIGNED_CHAR
- unsigned char
MPI_UNSIGNED_SHORT
- unsigned short
MPI_UNSIGNED
- unsigned int
MPI_UNSIGNED_LONG
- unsigned long
MPI_LONG_DOUBLE
- long double (some systems may not implement)
MPI_LONG_LONG_INT
- long long (some systems may not implement)

The following are datatypes for the MPI functions MPI_MAXLOC and MPI_MINLOC .

MPI_FLOAT_INT
- struct { float, int }
MPI_LONG_INT
- struct { long, int }
MPI_DOUBLE_INT
- struct { double, int }
MPI_SHORT_INT
- struct { short, int }
MPI_2INT
- struct { int, int }
MPI_LONG_DOUBLE_INT
- struct { long double, int }; this is an OPTIONAL type, and may be set to NULL

Note that there is no MPI_LONG_LONG_INT type corresponding to a struct { long long, int } .

Special datatypes for C and Fortran

MPI_PACKED
- For MPI_Pack and MPI_Unpack

MPI_UB
- For MPI_Type_struct ; an upper-bound indicator
MPI_LB
- For MPI_Type_struct ; a lower-bound indicator

FORTRAN DATATYPES

MPI_REAL
- REAL

MPI_INTEGER
- INTEGER

MPI_LOGICAL
- LOGICAL

MPI_DOUBLE_PRECISION
- DOUBLE PRECISION

MPI_COMPLEX
- COMPLEX

MPI_DOUBLE_COMPLEX
- complex*16 (or complex*32 ) where supported.

The following datatypes are optional

MPI_INTEGER1
- integer*1 if supported
MPI_INTEGER2
- integer*2 if supported
MPI_INTEGER4
- integer*4 if supported
MPI_REAL4
- real*4 if supported
MPI_REAL8
- real*8 if supported

The following are datatypes for the MPI functions MPI_MAXLOC and MPI_MINLOC . In Fortran, these datatype always consist of two elements of the same Fortran type.

MPI_2INTEGER
- INTEGER,INTEGER

MPI_2REAL
- REAL, REAL

MPI_2DOUBLE_PRECISION
- DOUBLE PRECISION, DOUBLE PRECISION

MPI_2COMPLEX
- COMPLEX, COMPLEX

MPI_2DOUBLE_COMPLEX
- complex*16, complex*16

COMMUNICATORS

Communicators are of type MPI_Comm in C and INTEGER in Fortran
MPI_COMM_WORLD
- Contains all of the processes
MPI_COMM_SELF
- Contains only the calling process

GROUPS

Groups are of type MPI_Group in C and INTEGER in Fortran

MPI_GROUP_EMPTY
- A group containing no members.

RESULTS OF THE COMPARE OPERATIONS

MPI_IDENT
- Identical
MPI_CONGRUENT
- (only for MPI_COMM_COMPARE ) The groups are identical
MPI_SIMILAR
- Same members, but in a different order
MPI_UNEQUAL
- Different

COLLECTIVE OPERATIONS

The collective combination operations ( MPI_REDUCE , MPI_ALLREDUCE , MPI_REDUCE_SCATTER , and MPI_SCAN ) take a combination operation. This operation is of type MPI_Op in C and of type INTEGER in Fortran. The predefined operations are

MPI_MAX
- return the maximum
MPI_MIN
- return the minumum
MPI_SUM
- return the sum
MPI_PROD
- return the product
MPI_LAND
- return the logical and
MPI_BAND
- return the bitwise and
MPI_LOR
- return the logical or
MPI_BOR
- return the bitwise of
MPI_LXOR
- return the logical exclusive or
MPI_BXOR
- return the bitwise exclusive or
MPI_MINLOC
- return the minimum and the location (actually, the value of the second element of the structure where the minimum of the first is found)
MPI_MAXLOC
- return the maximum and the location

NOTES ON COLLECTIVE OPERATIONS

The reduction functions ( MPI_Op ) do not return an error value. As a result, if the functions detect an error, all they can do is either call MPI_Abort or silently skip the problem. Thus, if you change the error handler from MPI_ERRORS_ARE_FATAL to something else, for example, MPI_ERRORS_RETURN , then no error may be indicated.

The reason for this is the performance problems in ensuring that all collective routines return the same error value.

Note that not all datatypes are valid for these functions. For example, MPI_COMPLEX is not valid for MPI_MAX and MPI_MIN . In addition, the MPI 1.1 standard did not include the C types MPI_CHAR and MPI_UNSIGNED_CHAR among the lists of arithmetic types for operations like MPI_SUM . However, since the C type char is an integer type (like short ), it should have been included. The MPI Forum will probably include char and unsigned char as a clarification to MPI 1.1; until then, users are advised that MPI implementations may not accept MPI_CHAR and MPI_UNSIGNED_CHAR as valid datatypes for MPI_SUM , MPI_PROD , etc. MPICH does allow these datatypes.

PERMANENT KEY VALUES

These are the same in C and Fortran

MPI_TAG_UB
- Largest tag value
MPI_HOST
- Rank of process that is host, if any
MPI_IO
- Rank of process that can do I/O
MPI_WTIME_IS_GLOBAL
- Has value 1 if MPI_WTIME is globally synchronized.

NULL OBJECTS

MPI_COMM_NULL
- Null communicator
MPI_OP_NULL
- Null operation
MPI_GROUP_NULL
- Null group
MPI_DATATYPE_NULL
- Null datatype
MPI_REQUEST_NULL
- Null request
MPI_ERRHANDLER_NULL
- Null error handler

PREDEFINED CONSTANTS

MPI_MAX_PROCESSOR_NAME
- Maximum length of name returned by MPI_GET_PROCESSOR_NAME

MPI_MAX_ERROR_STRING
- Maximum length of string return by MPI_ERROR_STRING

MPI_UNDEFINED
- Used by many routines to indicated undefined or unknown integer value
MPI_UNDEFINED_RANK
- Unknown rank
MPI_KEYVAL_INVALID
- Special keyval that may be used to detect uninitialized keyvals.
MPI_BSEND_OVERHEAD
- Add this to the size of a MPI_BSEND buffer for each outstanding message
MPI_PROC_NULL
- This rank may be used to send or receive from no-one.
MPI_ANY_SOURCE
- In a receive, accept a message from anyone.
MPI_ANY_TAG
- In a receive, accept a message with any tag value.
MPI_BOTTOM
- May be used to indicate the bottom of the address space

TOPOLOGY TYPES

MPI_GRAPH
- General graph
MPI_CART
- Cartesian grid

MPI STATUS

The MPI_Status datatype is a structure. The three elements for use by programmers are
MPI_SOURCE
- Who sent the message
MPI_TAG
- What tag the message was sent with
MPI_ERROR
- Any error return

SPECIAL MPI TYPES AND FUNCTIONS

MPI_Aint
- C type that holds any valid address.
MPI_Handler_function
- C function for handling errors (see MPI_Errhandler_create ) .
MPI_User_function
- C function to combine values (see collective operations and MPI_Op_create )
MPI_Copy_function
- Function to copy attributes (see MPI_Keyval_create )
MPI_NULL_COPY_FN
- Predefined copy function
MPI_Delete_function
- Function to delete attributes (see MPI_Keyval_create )
MPI_NULL_DELETE_FN
- Predefined delete function
MPI_DUP_FN
- Predefined duplication function
MPI_ERRORS_ARE_FATAL
- Error handler that forces exit on error
MPI_ERRORS_RETURN
- Error handler that returns error codes (as value of MPI routine in C and through last argument in Fortran)

MPI ERROR CLASSES

MPI_SUCCESS
- Successful return code
MPI_ERR_BUFFER
- Invalid buffer pointer
MPI_ERR_COUNT
- Invalid count argument
MPI_ERR_TYPE
- Invalid datatype argument
MPI_ERR_TAG
- Invalid tag argument
MPI_ERR_COMM
- Invalid communicator
MPI_ERR_RANK
- Invalid rank
MPI_ERR_ROOT
- Invalid root
MPI_ERR_GROUP
- Null group passed to function
MPI_ERR_OP
- Invalid operation
MPI_ERR_TOPOLOGY
- Invalid topology
MPI_ERR_DIMS
- Illegal dimension argument
MPI_ERR_ARG
- Invalid argument
MPI_ERR_UNKNOWN
- Unknown error
MPI_ERR_TRUNCATE
- message truncated on receive
MPI_ERR_OTHER
- Other error; use Error_string
MPI_ERR_INTERN
- internal error code
MPI_ERR_IN_STATUS
- Look in status for error value
MPI_ERR_PENDING
- Pending request
MPI_ERR_REQUEST
- illegal mpi_request handle
MPI_ERR_LASTCODE
- Last error code -- always at end

LOCATION

/home/MPI/mansrc/mpiconsts.txt