Teuchos - Trilinos Tools Package Version of the Day
Public Member Functions
Teuchos::MaxLoc< Ordinal, ScalarType, IndexType > Class Template Reference

Teuchos version of MPI_MAXLOC. More...

Inheritance diagram for Teuchos::MaxLoc< Ordinal, ScalarType, IndexType >:
Teuchos::ValueTypeReductionOp< Ordinal, std::pair< ScalarType, IndexType > > Teuchos::Describable Teuchos::LabeledObject

List of all members.

Public Member Functions

void reduce (const Ordinal count, const std::pair< ScalarType, IndexType > inBuffer[], std::pair< ScalarType, IndexType > inoutBuffer[]) const
 

Detailed Description

template<class Ordinal, class ScalarType, class IndexType>
class Teuchos::MaxLoc< Ordinal, ScalarType, IndexType >

Teuchos version of MPI_MAXLOC.

Author:
Mark Hoemmen
Template Parameters:
OrdinalThe template parameter of Comm.
ScalarTypeType for which to find the maximum.
IndexTypeType indicating the index of the maximum.

MPI_MAXLOC is a standard reduction operator provided by the MPI standard. According to the standard, MPI_MAXLOC combines the (value, index) pairs (u,i) and (v,j) into (w,j), where $w = max(u,v)$, and

\[ k = \begin{cases} i & \text{if $u > v$}, \\ \min(i,j) & \text{if $u = v$}, \\ j & \text{if $u < v$}. \\ \end{cases} \]

This class implements the MPI_MAXLOC reduction operator for the Teuchos communication wrappers. We need to define this separately from MPI for two reasons. First, the Teuchos comm wrappers do not currently expose taking an MPI_Op. Second, we need MaxLoc to be templated on the scalar and index types, in order to match Teuchos' MPI interface.

What happens to NaN ("Not a Number")? A NaN is neither less than, greater than, or equal to any floating-point number or any NaN. We can alter the above definition slightly so that a MaxLoc reduction has a well-defined result in case the array contains a NaN:

\[ w = \begin{cases} u & \text{if $u > v$}, \\ v & \text{if $u < v$}. \\ u & \text{otherwise}. \\ \end{cases} \]

and

\[ k = \begin{cases} i & \text{if $u > v$}, \\ j & \text{if $u < v$}. \\ \min(i,j) & \text{otherwise}. \\ \end{cases} \]

Defining MaxLoc in this way ensures that for any array containing a NaN, the value (w) returned is the first NaN, and the index (k) returned is the index of the first NaN.

Definition at line 106 of file Teuchos_TimeMonitor.cpp.


Member Function Documentation

template<class Ordinal , class ScalarType , class IndexType >
void Teuchos::MaxLoc< Ordinal, ScalarType, IndexType >::reduce ( const Ordinal  count,
const std::pair< ScalarType, IndexType >  inBuffer[],
std::pair< ScalarType, IndexType >  inoutBuffer[] 
) const [virtual]

The documentation for this class was generated from the following file:
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Defines