Epetra_MultiVector Class Reference

Epetra_MultiVector: A class for constructing and using dense multi-vectors, vectors and matrices in parallel. More...

#include <Epetra_MultiVector.h>

Inheritance diagram for Epetra_MultiVector:

[legend]Collaboration diagram for Epetra_MultiVector:[legend]List of all members.

Public Member Functions
int	ReplaceMap (const Epetra_BlockMap &map)
int	Reduce ()
Constructors/destructors
	Epetra_MultiVector (const Epetra_BlockMap &Map, int NumVectors, bool zeroOut=true)
	Basic Epetra_MultiVector constuctor.
	Epetra_MultiVector (const Epetra_MultiVector &Source)
	Epetra_MultiVector copy constructor.
	Epetra_MultiVector (Epetra_DataAccess CV, const Epetra_BlockMap &Map, double *A, int MyLDA, int NumVectors)
	Set multi-vector values from two-dimensional array.
	Epetra_MultiVector (Epetra_DataAccess CV, const Epetra_BlockMap &Map, double **ArrayOfPointers, int NumVectors)
	Set multi-vector values from array of pointers.
	Epetra_MultiVector (Epetra_DataAccess CV, const Epetra_MultiVector &Source, int *Indices, int NumVectors)
	Set multi-vector values from list of vectors in an existing Epetra_MultiVector.
	Epetra_MultiVector (Epetra_DataAccess CV, const Epetra_MultiVector &Source, int StartIndex, int NumVectors)
	Set multi-vector values from range of vectors in an existing Epetra_MultiVector.
virtual	~Epetra_MultiVector ()
	Epetra_MultiVector destructor.
Post-construction modification routines
int	ReplaceGlobalValue (int GlobalRow, int VectorIndex, double ScalarValue)
	Replace current value at the specified (GlobalRow, VectorIndex) location with ScalarValue.
int	ReplaceGlobalValue (int GlobalBlockRow, int BlockRowOffset, int VectorIndex, double ScalarValue)
	Replace current value at the specified (GlobalBlockRow, BlockRowOffset, VectorIndex) location with ScalarValue.
int	SumIntoGlobalValue (int GlobalRow, int VectorIndex, double ScalarValue)
	Adds ScalarValue to existing value at the specified (GlobalRow, VectorIndex) location.
int	SumIntoGlobalValue (int GlobalBlockRow, int BlockRowOffset, int VectorIndex, double ScalarValue)
	Adds ScalarValue to existing value at the specified (GlobalBlockRow, BlockRowOffset, VectorIndex) location.
int	ReplaceMyValue (int MyRow, int VectorIndex, double ScalarValue)
	Replace current value at the specified (MyRow, VectorIndex) location with ScalarValue.
int	ReplaceMyValue (int MyBlockRow, int BlockRowOffset, int VectorIndex, double ScalarValue)
	Replace current value at the specified (MyBlockRow, BlockRowOffset, VectorIndex) location with ScalarValue.
int	SumIntoMyValue (int MyRow, int VectorIndex, double ScalarValue)
	Adds ScalarValue to existing value at the specified (MyRow, VectorIndex) location.
int	SumIntoMyValue (int MyBlockRow, int BlockRowOffset, int VectorIndex, double ScalarValue)
	Adds ScalarValue to existing value at the specified (MyBlockRow, BlockRowOffset, VectorIndex) location.
int	PutScalar (double ScalarConstant)
	Initialize all values in a multi-vector with constant value.
int	Random ()
	Set multi-vector values to random numbers.
Extraction methods
int	ExtractCopy (double *A, int MyLDA) const
	Put multi-vector values into user-provided two-dimensional array.
int	ExtractCopy (double **ArrayOfPointers) const
	Put multi-vector values into user-provided array of pointers.
int	ExtractView (double **A, int *MyLDA) const
	Set user-provided addresses of A and MyLDA.
int	ExtractView (double ***ArrayOfPointers) const
	Set user-provided addresses of ArrayOfPointers.
Mathematical methods
int	Dot (const Epetra_MultiVector &A, double *Result) const
	Computes dot product of each corresponding pair of vectors.
int	Abs (const Epetra_MultiVector &A)
	Puts element-wise absolute values of input Multi-vector in target.
int	Reciprocal (const Epetra_MultiVector &A)
	Puts element-wise reciprocal values of input Multi-vector in target.
int	Scale (double ScalarValue)
	Scale the current values of a multi-vector, this = ScalarValue*this.
int	Scale (double ScalarA, const Epetra_MultiVector &A)
	Replace multi-vector values with scaled values of A, this = ScalarA*A.
int	Update (double ScalarA, const Epetra_MultiVector &A, double ScalarThis)
	Update multi-vector values with scaled values of A, this = ScalarThis*this + ScalarA*A.
int	Update (double ScalarA, const Epetra_MultiVector &A, double ScalarB, const Epetra_MultiVector &B, double ScalarThis)
	Update multi-vector with scaled values of A and B, this = ScalarThis*this + ScalarA*A + ScalarB*B.
int	Norm1 (double *Result) const
	Compute 1-norm of each vector in multi-vector.
int	Norm2 (double *Result) const
	Compute 2-norm of each vector in multi-vector.
int	NormInf (double *Result) const
	Compute Inf-norm of each vector in multi-vector.
int	NormWeighted (const Epetra_MultiVector &Weights, double *Result) const
	Compute Weighted 2-norm (RMS Norm) of each vector in multi-vector.
int	MinValue (double *Result) const
	Compute minimum value of each vector in multi-vector.
int	MaxValue (double *Result) const
	Compute maximum value of each vector in multi-vector.
int	MeanValue (double *Result) const
	Compute mean (average) value of each vector in multi-vector.
int	Multiply (char TransA, char TransB, double ScalarAB, const Epetra_MultiVector &A, const Epetra_MultiVector &B, double ScalarThis)
	Matrix-Matrix multiplication, this = ScalarThis*this + ScalarAB*A*B.
int	Multiply (double ScalarAB, const Epetra_MultiVector &A, const Epetra_MultiVector &B, double ScalarThis)
	Multiply a Epetra_MultiVector with another, element-by-element.
int	ReciprocalMultiply (double ScalarAB, const Epetra_MultiVector &A, const Epetra_MultiVector &B, double ScalarThis)
	Multiply a Epetra_MultiVector by the reciprocal of another, element-by-element.
Random number utilities
int	SetSeed (unsigned int Seed)
	Set seed for Random function.
unsigned int	Seed ()
	Get seed from Random function.
Overloaded operators
Epetra_MultiVector &	operator= (const Epetra_MultiVector &Source)
	= Operator.
double *&	operator[] (int i)
	Vector access function.
double *const &	operator[] (int i) const
	Vector access function.
Epetra_Vector *&	operator() (int i)
	Vector access function.
const Epetra_Vector *&	operator() (int i) const
	Vector access function.
Attribute access functions
int	NumVectors () const
	Returns the number of vectors in the multi-vector.
int	MyLength () const
	Returns the local vector length on the calling processor of vectors in the multi-vector.
int	GlobalLength () const
	Returns the global vector length of vectors in the multi-vector.
int	Stride () const
	Returns the stride between vectors in the multi-vector (only meaningful if ConstantStride() is true).
bool	ConstantStride () const
	Returns true if this multi-vector has constant stride between vectors.
I/O methods
virtual void	Print (ostream &os) const
	Print method.
Expert-only unsupported methods
int	ResetView (double **ArrayOfPointers)
	Reset the view of an existing multivector to point to new user data.
double *	Values () const
	Get pointer to MultiVector values.
double **	Pointers () const
	Get pointer to individual vector pointers.
Protected Member Functions
void	Assign (const Epetra_MultiVector &rhs)
int	CheckInput ()
Protected Attributes
double *	Values_

---

Detailed Description

Epetra_MultiVector: A class for constructing and using dense multi-vectors, vectors and matrices in parallel.

The Epetra_MultiVector class enables the construction and use of real-valued, double-precision dense vectors, multi-vectors, and matrices in a distributed memory environment. The dimensions and distribution of the dense multi-vectors is determined in part by a Epetra_Comm object, a Epetra_Map (or Epetra_LocalMap or Epetra_BlockMap) and the number of vectors passed to the constructors described below.

There are several concepts that important for understanding the Epetra_MultiVector class:

• Multi-vectors, Vectors and Matrices.
  • Vector - A list of real-valued, double-precision numbers. Also a multi-vector with one vector.
  • Multi-Vector - A collection of one or more vectors, all having the same length and distribution.
  • (Dense) Matrix - A special form of multi-vector such that stride in memory between any two consecutive vectors in the multi-vector is the same for all vectors. This is identical to a two-dimensional array in Fortran and plays an important part in high performance computations.
• Distributed Global vs. Replicated Local.
  • Distributed Global Multi-vectors - In most instances, a multi-vector will be partitioned across multiple memory images associated with multiple processors. In this case, there is a unique copy of each element and elements are spread across all processors specified by the Epetra_Comm communicator.
  • Replicated Local Multi-vectors - Some algorithms use multi-vectors that are too small to be distributed across all processors, the Hessenberg matrix in a GMRES computation. In other cases, such as with block iterative methods, block dot product functions produce small dense matrices that are required by all processors. Replicated local multi-vectors handle these types of situation.
• Multi-vector Functions vs. Dense Matrix Functions.
  • Multi-vector functions - These functions operate simultaneously but independently on each vector in the multi-vector and produce individual results for each vector.
  • Dense matrix functions - These functions operate on the multi-vector as a matrix, providing access to selected dense BLAS and LAPACK operations.

Constructing Epetra_MultiVectors

Except for the basic constructor and copy constructor, Epetra_MultiVector constructors have two data access modes:

1. Copy mode - Allocates memory and makes a copy of the user-provided data. In this case, the user data is not needed after construction.
2. View mode - Creates a "view" of the user data. In this case, the user data is required to remain intact for the life of the multi-vector.

Warning:

View mode is extremely dangerous from a data hiding perspective. Therefore, we strongly encourage users to develop code using Copy mode first and only use the View mode in a secondary optimization phase.

All Epetra_MultiVector constructors require a map argument that describes the layout of elements on the parallel machine. Specifically, map is a Epetra_Map, Epetra_LocalMap or Epetra_BlockMap object describing the desired memory layout for the multi-vector.

There are six different Epetra_MultiVector constructors:

• Basic - All values are zero.
• Copy - Copy an existing multi-vector.
• Copy from or make view of two-dimensional Fortran style array.
• Copy from or make view of an array of pointers.
• Copy or make view of a list of vectors from another Epetra_MultiVector object.
• Copy or make view of a range of vectors from another Epetra_MultiVector object.

Extracting Data from Epetra_MultiVectors

Once a Epetra_MultiVector is constructed, it is possible to extract a copy of the values or create a view of them.

Warning:

ExtractView functions are extremely dangerous from a data hiding perspective. For both ExtractView fuctions, there is a corresponding ExtractCopy function. We strongly encourage users to develop code using ExtractCopy functions first and only use the ExtractView functions in a secondary optimization phase.

There are four Extract functions:

• ExtractCopy - Copy values into a user-provided two-dimensional array.
• ExtractCopy - Copy values into a user-provided array of pointers.
• ExtractView - Set user-provided two-dimensional array parameters to point to Epetra_MultiVector data.
• ExtractView - Set user-provided array of pointer parameters to point to Epetra_MultiVector data.

Vector, Matrix and Utility Functions

Once a Epetra_MultiVector is constructed, a variety of mathematical functions can be applied to the individual vectors. Specifically:

• Dot Products.
• Vector Updates.
• p Norms.
• Weighted Norms.
• Minimum, Maximum and Average Values.

In addition, a matrix-matrix multiply function supports a variety of operations on any viable combination of global distributed and local replicated multi-vectors using calls to DGEMM, a high performance kernel for matrix operations. In the near future we will add support for calls to other selected BLAS and LAPACK functions.

Counting Floating Point Operations

Each Epetra_MultiVector object keep track of the number of serial floating point operations performed using the specified object as the this argument to the function. The Flops() function returns this number as a double precision number. Using this information, in conjunction with the Epetra_Time class, one can get accurate parallel performance numbers. The ResetFlops() function resets the floating point counter.

Warning:

A Epetra_Map, Epetra_LocalMap or Epetra_BlockMap object is required for all Epetra_MultiVector constructors.

---

Constructor & Destructor Documentation

Epetra_MultiVector::Epetra_MultiVector (  const Epetra_BlockMap &  Map, int  NumVectors, bool  zeroOut = true )

Basic Epetra_MultiVector constuctor. Creates a Epetra_MultiVector object and, by default, fills with zero values. Parameters: In  Map - A Epetra_LocalMap, Epetra_Map or Epetra_BlockMap. Warning: Note that, because Epetra_LocalMap derives from Epetra_Map and Epetra_Map derives from Epetra_BlockMap, this constructor works for all three types of Epetra map classes. Parameters: In  NumVectors - Number of vectors in multi-vector. In  zeroOut - If true then the allocated memory will be zeroed out initialy. If false then this memory will not be touched which can be significantly faster. Returns: Pointer to a Epetra_MultiVector.

Epetra_MultiVector::Epetra_MultiVector (  Epetra_DataAccess  CV, const Epetra_BlockMap &  Map, double *  A, int  MyLDA, int  NumVectors )

Set multi-vector values from two-dimensional array. Parameters: In  Epetra_DataAccess - Enumerated type set to Copy or View. In  Map - A Epetra_LocalMap, Epetra_Map or Epetra_BlockMap. In  A - Pointer to an array of double precision numbers. The first vector starts at A. The second vector starts at A+MyLDA, the third at A+2*MyLDA, and so on. In  MyLDA - The "Leading Dimension", or stride between vectors in memory. Warning: This value refers to the stride on the calling processor. Thus it is a local quantity, not a global quantity. Parameters: In  NumVectors - Number of vectors in multi-vector. Returns: Integer error code, set to 0 if successful. See Detailed Description section for further discussion.

Epetra_MultiVector::Epetra_MultiVector (  Epetra_DataAccess  CV, const Epetra_BlockMap &  Map, double **  ArrayOfPointers, int  NumVectors )

Set multi-vector values from array of pointers. Parameters: In  Epetra_DataAccess - Enumerated type set to Copy or View. In  Map - A Epetra_LocalMap, Epetra_Map or Epetra_BlockMap. In  ArrayOfPointers - An array of pointers such that ArrayOfPointers[i] points to the memory location containing ith vector to be copied. In  NumVectors - Number of vectors in multi-vector. Returns: Integer error code, set to 0 if successful. See Detailed Description section for further discussion.

Epetra_MultiVector::Epetra_MultiVector (  Epetra_DataAccess  CV, const Epetra_MultiVector &  Source, int *  Indices, int  NumVectors )

Set multi-vector values from list of vectors in an existing Epetra_MultiVector. Parameters: In  Epetra_DataAccess - Enumerated type set to Copy or View. In  Source - An existing fully constructed Epetra_MultiVector. In  Indices - Integer list of the vectors to copy. In  NumVectors - Number of vectors in multi-vector. Returns: Integer error code, set to 0 if successful. See Detailed Description section for further discussion.

Epetra_MultiVector::Epetra_MultiVector (  Epetra_DataAccess  CV, const Epetra_MultiVector &  Source, int  StartIndex, int  NumVectors )

Set multi-vector values from range of vectors in an existing Epetra_MultiVector. Parameters: In  Epetra_DataAccess - Enumerated type set to Copy or View. In  Source - An existing fully constructed Epetra_MultiVector. In  StartIndex - First of the vectors to copy. In  NumVectors - Number of vectors in multi-vector. Returns: Integer error code, set to 0 if successful. See Detailed Description section for further discussion.

---

Member Function Documentation

int Epetra_MultiVector::Abs (  const Epetra_MultiVector &  A  )

Puts element-wise absolute values of input Multi-vector in target. Parameters: In  A - Input Multi-vector. Out  this will contain the absolute values of the entries of A. Returns: Integer error code, set to 0 if successful. Note: It is possible to use the same argument for A and this.

int Epetra_MultiVector::Dot (  const Epetra_MultiVector &  A, double *  Result )  const

Computes dot product of each corresponding pair of vectors. Parameters: In  A - Multi-vector to be used with the "\e this" multivector. Out  Result - Result[i] will contain the ith dot product result. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::ExtractCopy (  double **  ArrayOfPointers  )  const

Put multi-vector values into user-provided array of pointers. Parameters: Out  ArrayOfPointers - An array of pointers to memory space that will contain the multi-vector values, such that ArrayOfPointers[i] points to the memory location where the ith vector to be copied. Returns: Integer error code, set to 0 if successful. See Detailed Description section for further discussion.

int Epetra_MultiVector::ExtractCopy (  double *  A, int  MyLDA )  const

Put multi-vector values into user-provided two-dimensional array. Parameters: Out  A - Pointer to memory space that will contain the multi-vector values. The first vector will be copied to the memory pointed to by A. The second vector starts at A+MyLDA, the third at A+2*MyLDA, and so on. In  MyLDA - The "Leading Dimension", or stride between vectors in memory. Warning: This value refers to the stride on the calling processor. Thus it is a local quantity, not a global quantity. Returns: Integer error code, set to 0 if successful. See Detailed Description section for further discussion.

int Epetra_MultiVector::ExtractView (  double ***  ArrayOfPointers  )  const

Set user-provided addresses of ArrayOfPointers. Parameters: ArrayOfPointers  (Out) - Address of array of pointers to memory space that will set to the multi-vector array of pointers, such that ArrayOfPointers[i] points to the memory location where the ith vector is located. Returns: Integer error code, set to 0 if successful. See Detailed Description section for further discussion.

int Epetra_MultiVector::ExtractView (  double **  A, int *  MyLDA )  const

Set user-provided addresses of A and MyLDA. Parameters: A  (Out) - Address of a pointer to that will be set to point to the values of the multi-vector. The first vector will be at the memory pointed to by A. The second vector starts at A+MyLDA, the third at A+2*MyLDA, and so on. MyLDA  (Out) - Address of the "Leading Dimension", or stride between vectors in memory. Warning: This value refers to the stride on the calling processor. Thus it is a local quantity, not a global quantity. Returns: Integer error code, set to 0 if successful. See Detailed Description section for further discussion.

int Epetra_MultiVector::MaxValue (  double *  Result  )  const

Compute maximum value of each vector in multi-vector. Note that the vector contents must be already initialized for this function to compute a well-defined result. The length of the vector need not be greater than zero on all processors. If length is greater than zero on any processor then a valid result will be computed. Parameters: Out  Result - Result[i] contains maximum value of ith vector. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::MeanValue (  double *  Result  )  const

Compute mean (average) value of each vector in multi-vector. Parameters: Out  Result - Result[i] contains mean value of ith vector. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::MinValue (  double *  Result  )  const

Compute minimum value of each vector in multi-vector. Note that the vector contents must be already initialized for this function to compute a well-defined result. The length of the vector need not be greater than zero on all processors. If length is greater than zero on any processor then a valid result will be computed. Parameters: Out  Result - Result[i] contains minimum value of ith vector. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::Multiply (  double  ScalarAB, const Epetra_MultiVector &  A, const Epetra_MultiVector &  B, double  ScalarThis )

Multiply a Epetra_MultiVector with another, element-by-element. This function supports diagonal matrix multiply. A is usually a single vector while B and this may have one or more columns. Note that B and this must have the same shape. A can be one vector or have the same shape as B. The actual computation is this = ScalarThis * this + ScalarAB * B @ A where @ denotes element-wise multiplication.

int Epetra_MultiVector::Multiply (  char  TransA, char  TransB, double  ScalarAB, const Epetra_MultiVector &  A, const Epetra_MultiVector &  B, double  ScalarThis )

Matrix-Matrix multiplication, this = ScalarThis*this + ScalarAB*A*B. This function performs a variety of matrix-matrix multiply operations, interpreting the Epetra_MultiVectors (this-aka C , A and B) as 2D matrices. Variations are due to the fact that A, B and C can be local replicated or global distributed Epetra_MultiVectors and that we may or may not operate with the transpose of A and B. Possible cases are: Total of 32 case (2^5). Num OPERATIONS case Notes 1) C(local) = A^X(local) * B^X(local) 4 (X=Transpose or Not, No comm needed) 2) C(local) = A^T(distr) * B (distr) 1 (2D dot product, replicate C) 3) C(distr) = A (distr) * B^X(local) 2 (2D vector update, no comm needed) Note that the following operations are not meaningful for 1D distributions: 1) C(local) = A^T(distr) * B^T(distr) 1 2) C(local) = A (distr) * B^X(distr) 2 3) C(distr) = A^X(local) * B^X(local) 4 4) C(distr) = A^X(local) * B^X(distr) 4 5) C(distr) = A^T(distr) * B^X(local) 2 6) C(local) = A^X(distr) * B^X(local) 4 7) C(distr) = A^X(distr) * B^X(local) 4 8) C(local) = A^X(local) * B^X(distr) 4 Parameters: In  TransA - Operate with the transpose of A if = 'T', else no transpose if = 'N'. In  TransB - Operate with the transpose of B if = 'T', else no transpose if = 'N'. In  ScalarAB - Scalar to multiply with A*B. In  A - Multi-vector. In  B - Multi-vector. In  ScalarThis - Scalar to multiply with this. Returns: Integer error code, set to 0 if successful. Warning: {Each multi-vector A, B and this is checked if it has constant stride using the ConstantStride() query function. If it does not have constant stride, a temporary copy is made and used for the computation. This activity is transparent to the user, except that there is memory and computation overhead. All temporary space is deleted prior to exit.}

int Epetra_MultiVector::Norm1 (  double *  Result  )  const

Compute 1-norm of each vector in multi-vector. Parameters: Out  Result - Result[i] contains 1-norm of ith vector. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::Norm2 (  double *  Result  )  const

Compute 2-norm of each vector in multi-vector. Parameters: Out  Result - Result[i] contains 2-norm of ith vector. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::NormInf (  double *  Result  )  const

Compute Inf-norm of each vector in multi-vector. Parameters: Out  Result - Result[i] contains Inf-norm of ith vector. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::NormWeighted (  const Epetra_MultiVector &  Weights, double *  Result )  const

Compute Weighted 2-norm (RMS Norm) of each vector in multi-vector. Parameters: In  Weights - Multi-vector of weights. If Weights contains a single vector, that vector will be used as the weights for all vectors of this. Otherwise, Weights should have the same number of vectors as this. Out  Result - Result[i] contains the weighted 2-norm of ith vector. Specifically if we denote the ith vector in the multivector by , and the ith weight vector by and let j represent the jth entry of each vector, on return Result[i] will contain the following result: , where is the global length of the vectors. Returns: Integer error code, set to 0 if successful.

const Epetra_Vector* & Epetra_MultiVector::operator() (  int  i  )  const

Vector access function. Returns: An Epetra_Vector pointer to the ith vector in the multi-vector.

Epetra_Vector* & Epetra_MultiVector::operator() (  int  i  )

Vector access function. Returns: An Epetra_Vector pointer to the ith vector in the multi-vector.

Epetra_MultiVector& Epetra_MultiVector::operator= (  const Epetra_MultiVector &  Source  )

= Operator. Parameters: In  A - Epetra_MultiVector to copy. Returns: Epetra_MultiVector.

double* const& Epetra_MultiVector::operator[] (  int  i  )  const [inline]

Vector access function. Returns: Pointer to the array of doubles containing the local values of the ith vector in the multi-vector. Reimplemented in Epetra_Vector.

double*& Epetra_MultiVector::operator[] (  int  i  )  [inline]

Vector access function. Returns: Pointer to the array of doubles containing the local values of the ith vector in the multi-vector. Reimplemented in Epetra_Vector.

int Epetra_MultiVector::PutScalar (  double  ScalarConstant  )

Initialize all values in a multi-vector with constant value. Parameters: In  ScalarConstant - Value to use. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::Random (   )

Set multi-vector values to random numbers. MultiVector uses the random number generator provided by Epetra_Util. The multi-vector values will be set to random values on the interval (-1.0, 1.0). Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::Reciprocal (  const Epetra_MultiVector &  A  )

Puts element-wise reciprocal values of input Multi-vector in target. Parameters: In  A - Input Multi-vector. Out  this will contain the element-wise reciprocal values of the entries of A. Returns: Integer error code, set to 0 if successful. Returns 2 if some entry is too small, but not zero. Returns 1 if some entry is zero. Note: It is possible to use the same argument for A and this. Also, if a given value of A is smaller than Epetra_DoubleMin (defined in Epetra_Epetra.h), but nonzero, then the return code is 2. If an entry is zero, the return code is 1. However, in all cases the reciprocal value is still used, even if a NaN is the result.

int Epetra_MultiVector::ReciprocalMultiply (  double  ScalarAB, const Epetra_MultiVector &  A, const Epetra_MultiVector &  B, double  ScalarThis )

Multiply a Epetra_MultiVector by the reciprocal of another, element-by-element. This function supports diagonal matrix scaling. A is usually a single vector while B and this may have one or more columns. Note that B and this must have the same shape. A can be one vector or have the same shape as B. The actual computation is this = ScalarThis * this + ScalarAB * B @ A where @ denotes element-wise division.

int Epetra_MultiVector::ReplaceGlobalValue (  int  GlobalBlockRow, int  BlockRowOffset, int  VectorIndex, double  ScalarValue )

Replace current value at the specified (GlobalBlockRow, BlockRowOffset, VectorIndex) location with ScalarValue. Replaces the existing value for a single entry in the multivector. The specified global block row and block row offset must correspond to a GID owned by the map of the multivector on the calling processor. In other words, this method does not perform cross-processor communication. Parameters: In  GlobalBlockRow - BlockRow of Multivector to modify in global index space. In  BlockRowOffset - Offset into BlockRow of Multivector to modify in global index space. In  VectorIndex - Vector within MultiVector that should to modify. In  ScalarValue - Value to add to existing value. Returns: Integer error code, set to 0 if successful, set to 1 if GlobalRow not associated with calling processor set to -1 if VectorIndex >= NumVectors(), set to -2 if BlockRowOffset is out-of-range.

int Epetra_MultiVector::ReplaceGlobalValue (  int  GlobalRow, int  VectorIndex, double  ScalarValue )

Replace current value at the specified (GlobalRow, VectorIndex) location with ScalarValue. Replaces the existing value for a single entry in the multivector. The specified global row must correspond to a GID owned by the map of the multivector on the calling processor. In other words, this method does not perform cross-processor communication. If the map associated with this multivector is an Epetra_BlockMap, only the first point entry associated with the global row will be modified. To modify a different point entry, use the other version of this method Parameters: In  GlobalRow - Row of Multivector to modify in global index space. In  VectorIndex - Vector within MultiVector that should to modify. In  ScalarValue - Value to add to existing value. Returns: Integer error code, set to 0 if successful, set to 1 if GlobalRow not associated with calling processor set to -1 if VectorIndex >= NumVectors().

int Epetra_MultiVector::ReplaceMap (  const Epetra_BlockMap &  map  )

Replace map, only if new map has same point-structure as current map. return 0 if map is replaced, -1 if not.

int Epetra_MultiVector::ReplaceMyValue (  int  MyBlockRow, int  BlockRowOffset, int  VectorIndex, double  ScalarValue )

Replace current value at the specified (MyBlockRow, BlockRowOffset, VectorIndex) location with ScalarValue. Replaces the existing value for a single entry in the multivector. The specified local block row and block row offset must correspond to a GID owned by the map of the multivector on the calling processor. In other words, this method does not perform cross-processor communication. Parameters: In  MyBlockRow - BlockRow of Multivector to modify in local index space. In  BlockRowOffset - Offset into BlockRow of Multivector to modify in local index space. In  VectorIndex - Vector within MultiVector that should to modify. In  ScalarValue - Value to add to existing value. Returns: Integer error code, set to 0 if successful, set to 1 if MyRow not associated with calling processor set to -1 if VectorIndex >= NumVectors(), set to -2 if BlockRowOffset is out-of-range.

int Epetra_MultiVector::ReplaceMyValue (  int  MyRow, int  VectorIndex, double  ScalarValue )

Replace current value at the specified (MyRow, VectorIndex) location with ScalarValue. Replaces the existing value for a single entry in the multivector. The specified local row must correspond to a GID owned by the map of the multivector on the calling processor. In other words, this method does not perform cross-processor communication. This method is intended for use with vectors based on an Epetra_Map. If used on a vector based on a non-trivial Epetra_BlockMap, this will update only block row 0, i.e. Epetra_MultiVector::ReplaceMyValue ( MyRow, VectorIndex, ScalarValue ) is equivalent to: Epetra_MultiVector::ReplaceMyValue ( 0, MyRow, VectorIndex, ScalarValue ) Parameters: In  MyRow - Row of Multivector to modify in local index space. In  VectorIndex - Vector within MultiVector that should to modify. In  ScalarValue - Value to add to existing value. Returns: Integer error code, set to 0 if successful, set to 1 if MyRow not associated with calling processor set to -1 if VectorIndex >= NumVectors().

int Epetra_MultiVector::ResetView (  double **  ArrayOfPointers  )

Reset the view of an existing multivector to point to new user data. Allows the (very) light-weight replacement of multivector values for an existing multivector that was constructed using an Epetra_DataAccess mode of View. No checking is performed to see if the array of values passed in contains valid data. It is assumed that the user has verified the integrity of data before calling this method. This method is useful for situations where a multivector is needed for use with an Epetra operator or matrix and the user is not passing in a multivector, or the multivector is being passed in with another map that is not exactly compatible with the operator, but has the correct number of entries. This method is used by AztecOO and Ifpack in the matvec, and solve methods to improve performance and reduce repeated calls to constructors and destructors. Parameters: ArrayOfPointers  Contains the array of pointers containing the multivector data. Returns: Integer error code, set to 0 if successful, -1 if the multivector was not created as a View. Warning: This method is extremely dangerous and should only be used by experts.

int Epetra_MultiVector::Scale (  double  ScalarA, const Epetra_MultiVector &  A )

Replace multi-vector values with scaled values of A, this = ScalarA*A. Parameters: In  ScalarA - Scale value. In  A - Multi-vector to copy. Out  This - Multi-vector with values overwritten by scaled values of A. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::Scale (  double  ScalarValue  )

Scale the current values of a multi-vector, this = ScalarValue*this. Parameters: In  ScalarValue - Scale value. Out  This - Multi-vector with scaled values. Returns: Integer error code, set to 0 if successful.

unsigned int Epetra_MultiVector::Seed (   )  [inline]

Get seed from Random function. Returns: Current random number seed.

int Epetra_MultiVector::SetSeed (  unsigned int  Seed  )  [inline]

Set seed for Random function. Parameters: In  Seed - Should be an integer on the interval (0, 2^31-1). Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::SumIntoGlobalValue (  int  GlobalBlockRow, int  BlockRowOffset, int  VectorIndex, double  ScalarValue )

Adds ScalarValue to existing value at the specified (GlobalBlockRow, BlockRowOffset, VectorIndex) location. Sums the given value into the existing value for a single entry in the multivector. The specified global block row and block row offset must correspond to a GID owned by the map of the multivector on the calling processor. In other words, this method does not perform cross-processor communication. Parameters: In  GlobalBlockRow - BlockRow of Multivector to modify in global index space. In  BlockRowOffset - Offset into BlockRow of Multivector to modify in global index space. In  VectorIndex - Vector within MultiVector that should to modify. In  ScalarValue - Value to add to existing value. Returns: Integer error code, set to 0 if successful, set to 1 if GlobalRow not associated with calling processor set to -1 if VectorIndex >= NumVectors(), set to -2 if BlockRowOffset is out-of-range.

int Epetra_MultiVector::SumIntoGlobalValue (  int  GlobalRow, int  VectorIndex, double  ScalarValue )

Adds ScalarValue to existing value at the specified (GlobalRow, VectorIndex) location. Sums the given value into the existing value for a single entry in the multivector. The specified global row must correspond to a GID owned by the map of the multivector on the calling processor. In other words, this method does not perform cross-processor communication. If the map associated with this multivector is an Epetra_BlockMap, only the first point entry associated with the global row will be modified. To modify a different point entry, use the other version of this method Parameters: In  GlobalRow - Row of Multivector to modify in global index space. In  VectorIndex - Vector within MultiVector that should to modify. In  ScalarValue - Value to add to existing value. Returns: Integer error code, set to 0 if successful, set to 1 if GlobalRow not associated with calling processor set to -1 if VectorIndex >= NumVectors().

int Epetra_MultiVector::SumIntoMyValue (  int  MyBlockRow, int  BlockRowOffset, int  VectorIndex, double  ScalarValue )

Adds ScalarValue to existing value at the specified (MyBlockRow, BlockRowOffset, VectorIndex) location. Sums the given value into the existing value for a single entry in the multivector. The specified local block row and block row offset must correspond to a GID owned by the map of the multivector on the calling processor. In other words, this method does not perform cross-processor communication. Parameters: In  MyBlockRow - BlockRow of Multivector to modify in local index space. In  BlockRowOffset - Offset into BlockRow of Multivector to modify in local index space. In  VectorIndex - Vector within MultiVector that should to modify. In  ScalarValue - Value to add to existing value. Returns: Integer error code, set to 0 if successful, set to 1 if MyRow not associated with calling processor set to -1 if VectorIndex >= NumVectors(), set to -2 if BlockRowOffset is out-of-range.

int Epetra_MultiVector::SumIntoMyValue (  int  MyRow, int  VectorIndex, double  ScalarValue )

Adds ScalarValue to existing value at the specified (MyRow, VectorIndex) location. Sums the given value into the existing value for a single entry in the multivector. The specified local row must correspond to a GID owned by the map of the multivector on the calling processor. In other words, this method does not perform cross-processor communication. If the map associated with this multivector is an Epetra_BlockMap, only the first point entry associated with the local row will be modified. To modify a different point entry, use the other version of this method Parameters: In  MyRow - Row of Multivector to modify in local index space. In  VectorIndex - Vector within MultiVector that should to modify. In  ScalarValue - Value to add to existing value. Returns: Integer error code, set to 0 if successful, set to 1 if MyRow not associated with calling processor set to -1 if VectorIndex >= NumVectors().

int Epetra_MultiVector::Update (  double  ScalarA, const Epetra_MultiVector &  A, double  ScalarB, const Epetra_MultiVector &  B, double  ScalarThis )

Update multi-vector with scaled values of A and B, this = ScalarThis*this + ScalarA*A + ScalarB*B. Parameters: In  ScalarA - Scale value for A. In  A - Multi-vector to add. In  ScalarB - Scale value for B. In  B - Multi-vector to add. In  ScalarThis - Scale value for this. Out  This - Multi-vector with updatede values. Returns: Integer error code, set to 0 if successful.

int Epetra_MultiVector::Update (  double  ScalarA, const Epetra_MultiVector &  A, double  ScalarThis )

Update multi-vector values with scaled values of A, this = ScalarThis*this + ScalarA*A. Parameters: In  ScalarA - Scale value for A. In  A - Multi-vector to add. In  ScalarThis - Scale value for this. Out  This - Multi-vector with updatede values. Returns: Integer error code, set to 0 if successful.

---

The documentation for this class was generated from the following file:

• Epetra_MultiVector.h

---