vpGEMM.h

/*
 * ViSP, open source Visual Servoing Platform software.
 * Copyright (C) 2005 - 2025 by Inria. All rights reserved.
 *
 * This software is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 * See the file LICENSE.txt at the root directory of this source
 * distribution for additional information about the GNU GPL.
 *
 * For using ViSP with software that can not be combined with the GNU
 * GPL, please contact Inria about acquiring a ViSP Professional
 * Edition License.
 *
 * See https://visp.inria.fr for more information.
 *
 * This software was developed at:
 * Inria Rennes - Bretagne Atlantique
 * Campus Universitaire de Beaulieu
 * 35042 Rennes Cedex
 * France
 *
 * If you have questions regarding the use of this file, please contact
 * Inria at visp@inria.fr
 *
 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
 *
 * Description:
 * Matrix generalized multiplication.
 */
 
#ifndef VP_GEMM_H
#define VP_GEMM_H
 
#include <visp3/core/vpArray2D.h>
#include <visp3/core/vpException.h>
 
BEGIN_VISP_NAMESPACE
VP_ATTRIBUTE_NO_DESTROY const vpArray2D<double> null(0, 0);

typedef enum
{
  VP_GEMM_A_T = 1, 
  VP_GEMM_B_T = 2, 
  VP_GEMM_C_T = 4, 
} vpGEMMmethod;
 
template <unsigned int>
inline void GEMMsize(const vpArray2D<double> & /*A*/, const vpArray2D<double> & /*B*/, unsigned int & /*Arows*/,
                     unsigned int & /*Acols*/, unsigned int & /*Brows*/, unsigned int & /*Bcols*/)
{ }
 
template <>
void inline GEMMsize<0>(const vpArray2D<double> &A, const vpArray2D<double> &B, unsigned int &Arows,
                        unsigned int &Acols, unsigned int &Brows, unsigned int &Bcols)
{
  Arows = A.getRows();
  Acols = A.getCols();
  Brows = B.getRows();
  Bcols = B.getCols();
}
 
template <>
inline void GEMMsize<1>(const vpArray2D<double> &A, const vpArray2D<double> &B, unsigned int &Arows,
                        unsigned int &Acols, unsigned int &Brows, unsigned int &Bcols)
{
  Arows = A.getCols();
  Acols = A.getRows();
  Brows = B.getRows();
  Bcols = B.getCols();
}
template <>
inline void GEMMsize<2>(const vpArray2D<double> &A, const vpArray2D<double> &B, unsigned int &Arows,
                        unsigned int &Acols, unsigned int &Brows, unsigned int &Bcols)
{
  Arows = A.getRows();
  Acols = A.getCols();
  Brows = B.getCols();
  Bcols = B.getRows();
}
template <>
inline void GEMMsize<3>(const vpArray2D<double> &A, const vpArray2D<double> &B, unsigned int &Arows,
                        unsigned int &Acols, unsigned int &Brows, unsigned int &Bcols)
{
  Arows = A.getCols();
  Acols = A.getRows();
  Brows = B.getCols();
  Bcols = B.getRows();
}
 
template <>
inline void GEMMsize<4>(const vpArray2D<double> &A, const vpArray2D<double> &B, unsigned int &Arows,
                        unsigned int &Acols, unsigned int &Brows, unsigned int &Bcols)
{
  Arows = A.getRows();
  Acols = A.getCols();
  Brows = B.getRows();
  Bcols = B.getCols();
}
 
template <>
inline void GEMMsize<5>(const vpArray2D<double> &A, const vpArray2D<double> &B, unsigned int &Arows,
                        unsigned int &Acols, unsigned int &Brows, unsigned int &Bcols)
{
  Arows = A.getCols();
  Acols = A.getRows();
  Brows = B.getRows();
  Bcols = B.getCols();
}
 
template <>
inline void GEMMsize<6>(const vpArray2D<double> &A, const vpArray2D<double> &B, unsigned int &Arows,
                        unsigned int &Acols, unsigned int &Brows, unsigned int &Bcols)
{
  Arows = A.getRows();
  Acols = A.getCols();
  Brows = B.getCols();
  Bcols = B.getRows();
}
 
template <>
inline void GEMMsize<7>(const vpArray2D<double> &A, const vpArray2D<double> &B, unsigned int &Arows,
                        unsigned int &Acols, unsigned int &Brows, unsigned int &Bcols)
{
  Arows = A.getCols();
  Acols = A.getRows();
  Brows = B.getCols();
  Bcols = B.getRows();
}
 
template <unsigned int>
inline void GEMM1(const unsigned int & /*Arows*/, const unsigned int & /*Brows*/, const unsigned int & /*Bcols*/,
                  const vpArray2D<double> & /*A*/, const vpArray2D<double> & /*B*/, const double & /*alpha*/,
                  vpArray2D<double> & /*D*/)
{ }
 
template <>
inline void GEMM1<0>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[r][n] * B[n][c] * alpha;
      }
      D[r][c] = sum;
    }
  }
}
 
template <>
inline void GEMM1<1>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[n][r] * B[n][c] * alpha;
      }
      D[r][c] = sum;
    }
  }
}
 
template <>
inline void GEMM1<2>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[r][n] * B[c][n] * alpha;
      }
      D[r][c] = sum;
    }
  }
}
 
template <>
inline void GEMM1<3>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[n][r] * B[c][n] * alpha;
      }
      D[r][c] = sum;
    }
  }
}
 
template <unsigned int>
inline void GEMM2(const unsigned int & /*Arows*/, const unsigned int & /*Brows*/, const unsigned int & /*Bcols*/,
                  const vpArray2D<double> & /*A*/, const vpArray2D<double> & /*B*/, const double & /*alpha*/,
                  const vpArray2D<double> & /*C*/, const double & /*beta*/, vpArray2D<double> & /*D*/)
{ }
 
template <>
inline void GEMM2<0>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                     const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[r][n] * B[n][c] * alpha;
      }
      D[r][c] = sum + (C[r][c] * beta);
    }
  }
}
 
template <>
inline void GEMM2<1>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                     const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[n][r] * B[n][c] * alpha;
      }
      D[r][c] = sum + (C[r][c] * beta);
    }
  }
}
 
template <>
inline void GEMM2<2>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                     const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[r][n] * B[c][n] * alpha;
      }
      D[r][c] = sum + (C[r][c] * beta);
    }
  }
}
 
template <>
inline void GEMM2<3>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                     const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[n][r] * B[c][n] * alpha;
      }
      D[r][c] = sum + (C[r][c] * beta);
    }
  }
}
 
template <>
inline void GEMM2<4>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                     const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[r][n] * B[n][c] * alpha;
      }
      D[r][c] = sum + (C[c][r] * beta);
    }
  }
}
 
template <>
inline void GEMM2<5>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                     const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[n][r] * B[n][c] * alpha;
      }
      D[r][c] = sum + (C[c][r] * beta);
    }
  }
}
 
template <>
inline void GEMM2<6>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                     const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[r][n] * B[c][n] * alpha;
      }
      D[r][c] = sum + (C[c][r] * beta);
    }
  }
}
 
template <>
inline void GEMM2<7>(const unsigned int &Arows, const unsigned int &Brows, const unsigned int &Bcols,
                     const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                     const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  for (unsigned int r = 0; r < Arows; ++r) {
    for (unsigned int c = 0; c < Bcols; ++c) {
      double sum = 0;
      for (unsigned int n = 0; n < Brows; ++n) {
        sum += A[n][r] * B[c][n] * alpha;
      }
      D[r][c] = sum + (C[c][r] * beta);
    }
  }
}
 
template <unsigned int T>
inline void vpTGEMM(const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                    const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D)
{
  unsigned int Arows;
  unsigned int Acols;
  unsigned int Brows;
  unsigned int Bcols;
 
  GEMMsize<T>(A, B, Arows, Acols, Brows, Bcols);
 
  try {
    if ((Arows != D.getRows()) || (Bcols != D.getCols())) {
      D.resize(Arows, Bcols);
    }
  }
  catch (...) {
    throw;
  }
 
  if (Acols != Brows) {
    throw(vpException(vpException::dimensionError, "In vpGEMM, cannot multiply (%dx%d) matrix by (%dx%d) matrix", Arows,
                      Acols, Brows, Bcols));
  }
 
  if (C.getRows() != 0 && C.getCols() != 0) {
    if ((Arows != C.getRows()) || (Bcols != C.getCols())) {
      throw(vpException(vpException::dimensionError, "In vpGEMM, cannot add resulting (%dx%d) matrix to (%dx%d) matrix",
                        Arows, Bcols, C.getRows(), C.getCols()));
    }
 
    GEMM2<T>(Arows, Brows, Bcols, A, B, alpha, C, beta, D);
  }
  else {
    GEMM1<T>(Arows, Brows, Bcols, A, B, alpha, D);
  }
}

inline void vpGEMM(const vpArray2D<double> &A, const vpArray2D<double> &B, const double &alpha,
                   const vpArray2D<double> &C, const double &beta, vpArray2D<double> &D, const unsigned int &ops = 0)
{
  switch (ops) {
  case 0:
    vpTGEMM<0>(A, B, alpha, C, beta, D);
    break;
  case 1:
    vpTGEMM<1>(A, B, alpha, C, beta, D);
    break;
  case 2:
    vpTGEMM<2>(A, B, alpha, C, beta, D);
    break;
  case 3:
    vpTGEMM<3>(A, B, alpha, C, beta, D);
    break;
  case 4:
    vpTGEMM<4>(A, B, alpha, C, beta, D);
    break;
  case 5:
    vpTGEMM<5>(A, B, alpha, C, beta, D);
    break;
  case 6:
    vpTGEMM<6>(A, B, alpha, C, beta, D);
    break;
  case 7:
    vpTGEMM<7>(A, B, alpha, C, beta, D);
    break;
  default:
    throw(vpException(vpException::functionNotImplementedError, "Operation on vpGEMM not implemented"));
    break;
  }
}
END_VISP_NAMESPACE
#endif