Files
unison/src/core/model/matrix-array.h
Biljana Bojovic b7c3ba83fb core: Add MatrixArray and MatrixArrayTestSuite
MatrixArray class is introduced for efficient operations on arrays of matrices
core: Allow using GetPagePtr as public function of MatrixArray
core: Remove Eigen include from matrix-array.h
core: Fix MatrixArray Doxygen
core: Fix MatrixArrayTestSuite Doxygen
core: Make HermitianTranspose member of MatrixArray
fix matrix-array spell check
core: Improve constructor Doxygen description
core: Move out MatrixArray inline function implementation from the class declaration (Thanks to Peter Barnes)
Use auto in MatrixArrayTestCase to be consistent
2023-03-03 21:15:50 +01:00

309 lines
14 KiB
C++

/*
* Copyright (c) 2022 Centre Tecnologic de Telecomunicacions de Catalunya (CTTC)
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation;
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
* Author: Biljana Bojovic <bbojovic@cttc.es>
*/
#ifndef MATRIX_ARRAY_H
#define MATRIX_ARRAY_H
#include "val-array.h"
#include <valarray>
namespace ns3
{
/**
* \ingroup Matrices
*
* \brief MatrixArray class inherits ValArray class and provides additional
* interfaces to ValArray which enable page-wise linear algebra operations for
* arrays of matrices. While ValArray class is focused on efficient storage,
* access and basic algebra element-wise operations on 3D numerical arrays,
* MatrixArray is focused on the efficient algebra operations on arrays of
* matrices.
*
* Each page (2-dimensional array) within a 3D array is considered to
* be a matrix, and operations are performed separately for each page/matrix.
* For example, a multiplication of two MatrixArrays means that a matrix
* multiplication is performed between each page in the first MatrixArray with
* the corresponding page in the second MatrixArray. Some of the operations can
* use the Eigen3 matrix library for speedup. In addition, this class can be
* extended to provide interfaces to the Eigen3 matrix library and enable
* additional linear algebra operations. The objective of this class is to two-fold:
*
* - First, it provides simple interfaces to perform matrix operations in parallel
* on all pages without the need to write an outer loop over all the pages.
*
* - Second, it can improve the performance of such parallel matrix operations,
* as this class uses a single memory allocation and one common set of size
* variables (rows, cols) for all pages, which can be an advantage compared to
* having separate memory allocations and size variables for each page/matrix,
* especially when those matrices are very small.
*
* Important characteristics:
*
* - Matrices are in column-major order.
*
* - Matrices can be 1-dimensional, in the sense that either the number of rows
* or columns is equal to 1.
*
* - The array of matrices can contain just one matrix or more matrices.
*
* - For binary operators, both MatrixArrays must have an equal number of pages,
* as well compatible row and column dimensions between the matrices.
*
* - Binary operators are performed among matrices with the same index from lhs
* and rhs matrix.
*
* - MatrixArray improves computationally complex matrix operations by using
* Eigen library when it is available at compile time. MatrixArray class makes
* use of Eigen Map class, which allows re-using already existing portion of
* memory as an Eigen type, e.g. ValArray can return a pointer to a memory
* where are placed elements of a specific matrix, and this can be passed to
* Eigen Map class which does not do a copy of the elements, but is using these
* elements directly in linear algebra operations.
*/
template <class T>
class MatrixArray : public ValArray<T>
{
public:
// instruct the compiler to generate the default constructor
MatrixArray<T>() = default;
/**
* \brief Constructor "pages" number of matrices that are of size "numRows"x"numCols",
* and are initialized with all-zero elements.
* If only 1 parameter, numRows, is provided then a single 1D array is being created.
* \param numRows the number of rows
* \param numCols the number of columns
* \param numPages the number of pages
*/
MatrixArray<T>(uint16_t numRows, uint16_t numCols = 1, uint16_t numPages = 1);
/**
* \brief Constructor creates a single array of values.size () elements and 1 column,
* and uses std::valarray<T> values to initialize the elements.
* \param values std::valarray<T> that will be used to initialize elements of 1D array
*/
explicit MatrixArray<T>(const std::valarray<T>& values);
/**
* \brief Constructor creates a single array of values.size () elements and 1 column,
* and moves std::valarray<T> values to initialize the elements.
* \param values std::valarray<T> that will be moved to initialize elements of 1D array
*/
MatrixArray<T>(std::valarray<T>&& values);
/**
* \brief Constructor creates a single array of values.size () elements and 1 column,
* and uses std::vector<T> values to initialize the elements.
* \param values std::vector<T> that will be used to initialize elements of 1D array
*/
explicit MatrixArray<T>(const std::vector<T>& values);
/**
* \brief Constructor creates a single matrix of numRows and numCols, and uses
* std::valarray<T> values to initialize the elements.
* \param numRows the number of rows
* \param numCols the number of columns
* \param values std::valarray<T> that will be used to initialize elements of 3D array
*/
MatrixArray<T>(uint16_t numRows, uint16_t numCols, const std::valarray<T>& values);
/**
* \brief Constructor creates a single matrix of numRows and numCols, and moves
* std::valarray<T> values to initialize the elements.
* \param numRows the number of rows
* \param numCols the number of columns
* \param values std::valarray<T> that will be moved to initialize elements of 3D array
*/
MatrixArray<T>(uint16_t numRows, uint16_t numCols, std::valarray<T>&& values);
/**
* \brief Constructor creates the array of numPages matrices of numRows x numCols dimensions,
* and uses std::valarray<T> values to initialize all the elements.
* \param numRows the number of rows
* \param numCols the number of columns
* \param numPages the number of pages
* \param values std::valarray<T> that will be used to initialize elements of 3D array
*/
MatrixArray<T>(uint16_t numRows,
uint16_t numCols,
uint16_t numPages,
const std::valarray<T>& values);
/**
* \brief Constructor creates the array of numPages matrices of numRows x numCols dimensions,
* and moves std::valarray<T> values to initialize all the elements.
* \param numRows the number of rows
* \param numCols the number of columns
* \param numPages the number of pages
* \param values std::valarray<T> that will be used to initialize elements of 3D array
*/
MatrixArray<T>(uint16_t numRows,
uint16_t numCols,
uint16_t numPages,
std::valarray<T>&& values);
/** instruct the compiler to generate the implicitly declared destructor*/
~MatrixArray<T>() override = default;
/** instruct the compiler to generate the implicitly declared copy constructor*/
MatrixArray<T>(const MatrixArray<T>&) = default;
/**
* \brief Copy assignment operator.
* Instruct the compiler to generate the implicitly declared copy assignment operator.
* \return a reference to the result of the assignment
*/
MatrixArray<T>& operator=(const MatrixArray<T>&) = default;
/** instruct the compiler to generate the implicitly declared move constructor*/
MatrixArray<T>(MatrixArray<T>&&) noexcept = default;
/**
* \brief Move assignment operator.
* Instruct the compiler to generate the implicitly declared move assignment operator.
* \return a reference to the result of the assignment
*/
MatrixArray<T>& operator=(MatrixArray<T>&&) noexcept = default;
/**
* \brief Element-wise multiplication with a scalar value.
* \param rhs is a scalar value of type T
* \returns The matrix in which each element has been multiplied by the given
* scalar value.
*/
MatrixArray<T> operator*(const T& rhs) const;
/**
* \brief operator+ definition for MatrixArray<T>.
* \param rhs The rhs MatrixArray object
* \returns The result of operator+ of this MatrixArray and rhs MatrixAray
*/
MatrixArray<T> operator+(const MatrixArray<T>& rhs) const;
/**
* \brief binary operator- definition for MatrixArray<T>.
* \param rhs The rhs MatrixArray object
* \returns The result of operator- of this MatrixArray and rhs MatrixAray
*/
MatrixArray<T> operator-(const MatrixArray<T>& rhs) const;
/**
* \brief unary operator- definition for MatrixArray<T>.
* \return the result of the operator-
*/
MatrixArray<T> operator-() const;
/**
* \brief Page-wise matrix multiplication.
* This operator interprets the 3D array as an array of matrices,
* and performs a linear algebra operation on each of the matrices (pages),
* i.e., matrices from this MatrixArray and rhs with the same page indices are
* multiplied.
* The number of columns of this MatrixArray must be equal to the number of
* rows in rhs, and rhs must have the same number of pages as this MatrixArray.
* \param rhs is another MatrixArray instance
* \returns The array of results of matrix multiplications.
*/
MatrixArray<T> operator*(const MatrixArray<T>& rhs) const;
/**
* \brief This operator interprets the 3D array as an array of matrices,
* and performs a linear algebra operation on each of the matrices (pages),
* i.e., transposes the matrix or array of matrices definition for MatrixArray<T>.
* \return The resulting MatrixArray composed of the array of transposed matrices.
*/
MatrixArray<T> Transpose() const;
/**
*\brief Multiply each matrix in the array by the left and the right matrix.
* For each page of this MatrixArray the operation performed is
* lMatrix * matrix(pageIndex) * rMatrix, and the resulting MatrixArray<T>
* contains the array of the results per page. If "this" has dimensions
* M x N x P, lMatrix must have dimensions M number of columns, and a single
* page, and rMatrix must have N number of rows, and also a single page.
*
* The result of this function will have the dimensions J x K x P. Where J is
* the number of rows of the lMatrix, and K is the number of
* columns of rMatrix. Dimensions are:
*
* lMatrix(JxMx1) * this (MxNxP) * rMatrix(NxKx1) -> result(JxKxP)
*
* This operation is not possible when using the multiplication operator because
* the number of pages does not match.
*
* \param lMatrix the left matrix in the multiplication
* \param rMatrix the right matrix in the multiplication
* \returns Returns the result of the multiplication which is a 3D MatrixArray
* with dimensions J x K x P as explained previously.
*/
MatrixArray<T> MultiplyByLeftAndRightMatrix(const MatrixArray<T>& lMatrix,
const MatrixArray<T>& rMatrix) const;
using ValArray<T>::GetPagePtr;
using ValArray<T>::EqualDims;
using ValArray<T>::AssertEqualDims;
/**
* \brief Function that performs the Hermitian transpose of this MatrixArray
* and returns a new matrix that is the result of the operation.
* This function assumes that the matrix contains complex numbers, because of that
* it is only available for the <std::complex<double>> specialization of MatrixArray.
* \return Returns a new matrix that is the result of the Hermitian transpose
*/
template <bool EnableBool = true,
typename = typename std::enable_if<(std::is_same<T, std::complex<double>>::value &&
EnableBool)>::type>
MatrixArray<T> HermitianTranspose() const;
private:
// To simplify functions in MatrixArray that are using members from the template base class
using ValArray<T>::m_numRows;
using ValArray<T>::m_numCols;
using ValArray<T>::m_numPages;
using ValArray<T>::m_values;
};
using IntMatrixArray = MatrixArray<int>; //!< Create an alias for MatrixArray using int type
using DoubleMatrixArray =
MatrixArray<double>; //!< Create an alias for MatrixArray using double type
using ComplexMatrixArray =
MatrixArray<std::complex<double>>; //!< Create an alias for MatrixArray using complex type
/*************************************************
** Class MatrixArray inline implementations
************************************************/
template <class T>
inline MatrixArray<T>
MatrixArray<T>::operator*(const T& rhs) const
{
return MatrixArray<T>(m_numRows,
m_numCols,
m_numPages,
m_values * std::valarray<T>(rhs, m_numRows * m_numCols * m_numPages));
}
template <class T>
inline MatrixArray<T>
MatrixArray<T>::operator+(const MatrixArray<T>& rhs) const
{
AssertEqualDims(rhs);
return MatrixArray<T>(m_numRows, m_numCols, m_numPages, m_values + rhs.m_values);
}
template <class T>
inline MatrixArray<T>
MatrixArray<T>::operator-(const MatrixArray<T>& rhs) const
{
AssertEqualDims(rhs);
return MatrixArray<T>(m_numRows, m_numCols, m_numPages, m_values - rhs.m_values);
}
template <class T>
inline MatrixArray<T>
MatrixArray<T>::operator-() const
{
return MatrixArray<T>(m_numRows, m_numCols, m_numPages, -m_values);
}
} // namespace ns3
#endif // MATRIX_ARRAY_H