« Return to documentation listing
NAME MPI_Reduce  Reduces values on all processes within a group.
SYNTAX
C Syntax #include <mpi.h>
int MPI_Reduce(void *sendbuf, void *recvbuf, int count,
MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm)
Fortran Syntax INCLUDE 'mpif.h'
MPI_REDUCE(SENDBUF, RECVBUF, COUNT, DATATYPE, OP, ROOT, COMM,
IERROR)
<type> SENDBUF(*), RECVBUF(*)
INTEGER COUNT, DATATYPE, OP, ROOT, COMM, IERROR
C++ Syntax #include <mpi.h>
void MPI::Intracomm::Reduce(const void* sendbuf, void* recvbuf,
int count, const MPI::Datatype& datatype, const MPI::Op& op,
int root) const
INPUT PARAMETERS sendbuf Address of send buffer (choice).
count Number of elements in send buffer (integer).
datatype Data type of elements of send buffer (handle).
op Reduce operation (handle).
root Rank of root process (integer).
comm Communicator (handle).
OUTPUT PARAMETERS recvbuf Address of receive buffer (choice, significant only at root).
IERROR Fortran only: Error status (integer).
DESCRIPTION The global reduce functions (MPI_Reduce, MPI_Op_create, MPI_Op_free,
MPI_Allreduce, MPI_Reduce_scatter, MPI_Scan) perform a global reduce
operation (such as sum, max, logical AND, etc.) across all the members
of a group. The reduction operation can be either one of a predefined
list of operations, or a userdefined operation. The global reduction
functions come in several flavors: a reduce that returns the result of
the reduction at one node, an allreduce that returns this result at
all nodes, and a scan (parallel prefix) operation. In addition, a
reducescatter operation combines the functionality of a reduce and a
scatter operation.
and output buffers of the same length, with elements of the same type.
Each process can provide one element, or a sequence of elements, in
which case the combine operation is executed elementwise on each entry
of the sequence. For example, if the operation is MPI_MAX and the send
buffer contains two elements that are floatingpoint numbers (count = 2
and datatype = MPI_FLOAT), then recvbuf(1) = global max (sendbuf(1))
and recvbuf(2) = global max(sendbuf(2)).
USE OF INPLACE OPTION When the communicator is an intracommunicator, you can perform a reduce
operation inplace (the output buffer is used as the input buffer).
Use the variable MPI_IN_PLACE as the value of the root process sendbuf.
In this case, the input data is taken at the root from the receive
buffer, where it will be replaced by the output data.
Note that MPI_IN_PLACE is a special kind of value; it has the same
restrictions on its use as MPI_BOTTOM.
Because the inplace option converts the receive buffer into a send
andreceive buffer, a Fortran binding that includes INTENT must mark
these as INOUT, not OUT.
WHEN COMMUNICATOR IS AN INTERCOMMUNICATOR When the communicator is an intercommunicator, the root process in the
first group combines data from all the processes in the second group
and then performs the op operation. The first group defines the root
process. That process uses MPI_ROOT as the value of its root argument.
The remaining processes use MPI_PROC_NULL as the value of their root
argument. All processes in the second group use the rank of that root
process in the first group as the value of their root argument. Only
the send buffer arguments are significant in the second group, and only
the receive buffer arguments are significant in the root process of the
first group.
PREDEFINED REDUCE OPERATIONS The set of predefined operations provided by MPI is listed below (Pre
defined Reduce Operations). That section also enumerates the datatypes
each operation can be applied to. In addition, users may define their
own operations that can be overloaded to operate on several datatypes,
either basic or derived. This is further explained in the description
of the userdefined operations (see the man pages for MPI_Op_create and
MPI_Op_free).
The operation op is always assumed to be associative. All predefined
operations are also assumed to be commutative. Users may define opera
tions that are assumed to be associative, but not commutative. The
``canonical'' evaluation order of a reduction is determined by the
ranks of the processes in the group. However, the implementation can
take advantage of associativity, or associativity and commutativity, in
order to change the order of evaluation. This may change the result of
the reduction for operations that are not strictly associative and com
mutative, such as floating point addition.
Predefined operators work only with the MPI types listed below (Prede
fined Reduce Operations, and the section MINLOC and MAXLOC, below).
These operations are invoked by placing the following in op:
Name Meaning
 
MPI_MAX maximum
MPI_MIN minimum
MPI_SUM sum
MPI_PROD product
MPI_LAND logical and
MPI_BAND bitwise and
MPI_LOR logical or
MPI_BOR bitwise or
MPI_LXOR logical xor
MPI_BXOR bitwise xor
MPI_MAXLOC max value and location
MPI_MINLOC min value and location
The two operations MPI_MINLOC and MPI_MAXLOC are discussed separately
below (MINLOC and MAXLOC). For the other predefined operations, we enu
merate below the allowed combinations of op and datatype arguments.
First, define groups of MPI basic datatypes in the following way:
C integer: MPI_INT, MPI_LONG, MPI_SHORT,
MPI_UNSIGNED_SHORT, MPI_UNSIGNED,
MPI_UNSIGNED_LONG
Fortran integer: MPI_INTEGER
Floatingpoint: MPI_FLOAT, MPI_DOUBLE, MPI_REAL,
MPI_DOUBLE_PRECISION, MPI_LONG_DOUBLE
Logical: MPI_LOGICAL
Complex: MPI_COMPLEX
Byte: MPI_BYTE
Now, the valid datatypes for each option is specified below.
Op Allowed Types
 
MPI_MAX, MPI_MIN C integer, Fortran integer,
floatingpoint
MPI_SUM, MPI_PROD C integer, Fortran integer,
floatingpoint, complex
MPI_LAND, MPI_LOR, C integer, logical
MPI_LXOR
MPI_BAND, MPI_BOR, C integer, Fortran integer, byte
MPI_BXOR
Example 1: A routine that computes the dot product of two vectors that
are distributed across a group of processes and returns the answer at
process zero.
SUBROUTINE PAR_BLAS1(m, a, b, c, comm)
REAL a(m), b(m) ! local slice of array
REAL c ! result (at process zero)
REAL sum
INTEGER m, comm, i, ierr
RETURN
Example 2: A routine that computes the product of a vector and an array
that are distributed across a group of processes and returns the
answer at process zero.
SUBROUTINE PAR_BLAS2(m, n, a, b, c, comm)
REAL a(m), b(m,n) ! local slice of array
REAL c(n) ! result
REAL sum(n)
INTEGER n, comm, i, j, ierr
! local sum
DO j= 1, n
sum(j) = 0.0
DO i = 1, m
sum(j) = sum(j) + a(i)*b(i,j)
END DO
END DO
! global sum
CALL MPI_REDUCE(sum, c, n, MPI_REAL, MPI_SUM, 0, comm, ierr)
! return result at process zero (and garbage at the other nodes)
RETURN
MINLOC AND MAXLOC The operator MPI_MINLOC is used to compute a global minimum and also an
index attached to the minimum value. MPI_MAXLOC similarly computes a
global maximum and index. One application of these is to compute a
global minimum (maximum) and the rank of the process containing this
value.
The operation that defines MPI_MAXLOC is
( u ) ( v ) ( w )
( ) o ( ) = ( )
( i ) ( j ) ( k )
where
w = max(u, v)
and
( i if u > v
(
k = ( min(i, j) if u = v
(
( j if u < v)
MPI_MINLOC is defined similarly:
( u ) ( v ) ( w )
( ) o ( ) = ( )
( i if u < v
(
k = ( min(i, j) if u = v
(
( j if u > v)
Both operations are associative and commutative. Note that if
MPI_MAXLOC is applied to reduce a sequence of pairs (u(0), 0), (u(1),
1), ..., (u(n1), n1), then the value returned is (u , r), where u=
max(i) u(i) and r is the index of the first global maximum in the
sequence. Thus, if each process supplies a value and its rank within
the group, then a reduce operation with op = MPI_MAXLOC will return the
maximum value and the rank of the first process with that value. Simi
larly, MPI_MINLOC can be used to return a minimum and its index. More
generally, MPI_MINLOC computes a lexicographic minimum, where elements
are ordered according to the first component of each pair, and ties are
resolved according to the second component.
The reduce operation is defined to operate on arguments that consist of
a pair: value and index. For both Fortran and C, types are provided to
describe the pair. The potentially mixedtype nature of such arguments
is a problem in Fortran. The problem is circumvented, for Fortran, by
having the MPIprovided type consist of a pair of the same type as
value, and coercing the index to this type also. In C, the MPIprovided
pair type has distinct types and the index is an int.
In order to use MPI_MINLOC and MPI_MAXLOC in a reduce operation, one
must provide a datatype argument that represents a pair (value and
index). MPI provides nine such predefined datatypes. The operations
MPI_MAXLOC and MPI_MINLOC can be used with each of the following
datatypes:
Fortran:
Name Description
MPI_2REAL pair of REALs
MPI_2DOUBLE_PRECISION pair of DOUBLEPRECISION variables
MPI_2INTEGER pair of INTEGERs
C:
Name Description
MPI_FLOAT_INT float and int
MPI_DOUBLE_INT double and int
MPI_LONG_INT long and int
MPI_2INT pair of ints
MPI_SHORT_INT short and int
MPI_LONG_DOUBLE_INT long double and int
The data type MPI_2REAL is equivalent to:
MPI_TYPE_CONTIGUOUS(2, MPI_REAL, MPI_2REAL)
Similar statements apply for MPI_2INTEGER, MPI_2DOUBLE_PRECISION, and
MPI_2INT.
The datatype MPI_FLOAT_INT is as if defined by the following sequence
of instructions.
Similar statements apply for MPI_LONG_INT and MPI_DOUBLE_INT.
Example 3: Each process has an array of 30 doubles, in C. For each of
the 30 locations, compute the value and rank of the process containing
the largest value.
...
/* each process has an array of 30 double: ain[30]
*/
double ain[30], aout[30];
int ind[30];
struct {
double val;
int rank;
} in[30], out[30];
int i, myrank, root;
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
for (i=0; i<30; ++i) {
in[i].val = ain[i];
in[i].rank = myrank;
}
MPI_Reduce( in, out, 30, MPI_DOUBLE_INT, MPI_MAXLOC, root, comm );
/* At this point, the answer resides on process root
*/
if (myrank == root) {
/* read ranks out
*/
for (i=0; i<30; ++i) {
aout[i] = out[i].val;
ind[i] = out[i].rank;
}
}
Example 4: Same example, in Fortran.
...
! each process has an array of 30 double: ain(30)
DOUBLE PRECISION ain(30), aout(30)
INTEGER ind(30);
DOUBLE PRECISION in(2,30), out(2,30)
INTEGER i, myrank, root, ierr;
MPI_COMM_RANK(MPI_COMM_WORLD, myrank);
DO I=1, 30
in(1,i) = ain(i)
in(2,i) = myrank ! myrank is coerced to a double
END DO
MPI_REDUCE( in, out, 30, MPI_2DOUBLE_PRECISION, MPI_MAXLOC, root,
comm, ierr );
! At this point, the answer resides on process root
IF (myrank .EQ. root) THEN
! read ranks out
DO I= 1, 30
#define LEN 1000
float val[LEN]; /* local array of values */
int count; /* local number of values */
int myrank, minrank, minindex;
float minval;
struct {
float value;
int index;
} in, out;
/* local minloc */
in.value = val[0];
in.index = 0;
for (i=1; i < count; i++)
if (in.value > val[i]) {
in.value = val[i];
in.index = i;
}
/* global minloc */
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
in.index = myrank*LEN + in.index;
MPI_Reduce( in, out, 1, MPI_FLOAT_INT, MPI_MINLOC, root, comm );
/* At this point, the answer resides on process root
*/
if (myrank == root) {
/* read answer out
*/
minval = out.value;
minrank = out.index / LEN;
minindex = out.index % LEN;
All MPI objects (e.g., MPI_Datatype, MPI_Comm) are of type INTEGER in
Fortran.
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.
ERRORS Almost all MPI routines return an error value; C routines as the value
of the function and Fortran routines in the last argument. C++ func
tions do not return errors. If the default error handler is set to
MPI::ERRORS_THROW_EXCEPTIONS, then on error the C++ exception mechanism
will be used to throw an MPI:Exception object.
Before the error value is returned, the current MPI error handler is
called. By default, this error handler aborts the MPI job, except for
MPI_Reduce_scatter
MPI_Scan
MPI_Op_create
MPI_Op_free
Open MPI 1.2 September 2006 MPI_Reduce(3OpenMPI)
« Return to documentation listing
