xref: /dports/science/qmcpack/qmcpack-3.11.0/src/Numerics/MatrixOperators.h

//////////////////////////////////////////////////////////////////////////////////////
// This file is distributed under the University of Illinois/NCSA Open Source License.
// See LICENSE file in top directory for details.
//
// Copyright (c) 2016 Jeongnim Kim and QMCPACK developers.
//
// File developed by: Bryan Clark, bclark@Princeton.edu, Princeton University
//                    Jeremy McMinnis, jmcminis@gmail.com, University of Illinois at Urbana-Champaign
//                    Jeongnim Kim, jeongnim.kim@gmail.com, University of Illinois at Urbana-Champaign
//                    Jaron T. Krogel, krogeljt@ornl.gov, Oak Ridge National Laboratory
//                    Mark A. Berrill, berrillma@ornl.gov, Oak Ridge National Laboratory
//
// File created by: Jeongnim Kim, jeongnim.kim@gmail.com, University of Illinois at Urbana-Champaign
//////////////////////////////////////////////////////////////////////////////////////


#ifndef QMCPLUSPLUS_MATRIXOPERATORS_H
#define QMCPLUSPLUS_MATRIXOPERATORS_H

#include "OhmmsPETE/OhmmsMatrix.h"
#include "OhmmsPETE/OhmmsVector.h"
#include "OhmmsPETE/TinyVector.h"
#include "OhmmsPETE/Tensor.h"
#include "CPU/BLAS.hpp"
#include "CPU/SIMD/inner_product.hpp"

namespace qmcplusplus
{
template<typename T>
inline T TESTDOT(const T* restrict f, const T* restrict l, const T* restrict b)
{
  T s = 0;
  while (f != l)
    s += (*f++) * (*b++);
  return s;
}

namespace MatrixOperators
{
/** static function to perform C=AB for real matrices
   *
   * Call dgemm/zgemm/sgemm/cgemm via BLAS::gemm
   */
template<typename T>
inline void product(const Matrix<T>& A, const Matrix<T>& B, Matrix<T>& C)
{
  const char transa = 'N';
  const char transb = 'N';
  const T one(1.0);
  const T zero(0.0);
  BLAS::gemm(transa, transb, B.cols(), A.rows(), B.rows(), one, B.data(), B.cols(), A.data(), A.cols(), zero, C.data(),
             C.cols());
}

template<typename T>
inline void product_ABt(const Matrix<T>& A, const Matrix<T>& B, Matrix<T>& C)
{
  const char transa = 't';
  const char transb = 'n';
  const T one(1.0);
  const T zero(0.0);
  BLAS::gemm(transa, transb, B.rows(), A.rows(), B.cols(), one, B.data(), B.cols(), A.data(), A.cols(), zero, C.data(),
             C.cols());
}

template<typename T>
inline void product_AtB(const Matrix<T>& A, const Matrix<T>& B, Matrix<T>& C)
{
  const char transa = 'n';
  const char transb = 't';
  const T one(1.0);
  const T zero(0.0);
  BLAS::gemm(transa, transb, B.cols(), A.cols(), B.rows(), one, B.data(), B.cols(), A.data(), A.cols(), zero, C.data(),
             C.cols());
}


inline void half_outerProduct(const Matrix<double>& M, const Vector<double>& B, int iat, Matrix<double>& C)
{
  for (int i = 0; i < C.rows(); i++)
    for (int j = 0; j < C.cols(); j++)
      C(iat, i) += M(i, j) * B[j];
  //for (int i=0; i<C.rows(); i++)
  //  C(iat,i)+=simd::dot(M[i],B.data(),C.cols());
}

inline void other_half_outerProduct(const Matrix<double>& M, const Vector<double>& B, int iat, Matrix<double>& C)
{
  for (int i = 0; i < C.rows(); i++)
    for (int j = 0; j < C.cols(); j++)
      C(i, iat) += M(i, j) * B[j];
  //for (int i=0; i<C.rows(); i++)
  //  C(i,iat)+=simd::dot(M[i],B.data(),C.cols());
}

inline double innerProduct(const Vector<double>& A, const Vector<double>& B)
{
  double tot = 0.0;
  for (int i = 0; i < A.size(); i++)
    tot += A[i] * B[i];
  return tot;
  //return simd::dot(A.data(),B.data(),A.size());
}


template<typename T>
inline void transpose(Matrix<T>& A)
{
  for (int i = 0; i < A.extent(0); i++)
    for (int j = 0; j < i; j++)
      std::swap(A(i, j), A(j, i));
}

template<typename T>
inline void transpose(const Matrix<T>& A, Matrix<T>& B)
{
  simd::transpose(A.data(), A.rows(), A.cols(), B.data(), B.rows(), B.cols());
}


/// C = A*diag(B)
template<typename T1, typename T2, typename T3>
inline void diag_product(const Matrix<T1>& A, const Vector<T2>& B, Matrix<T3>& C)
{
  for (int i = 0; i < C.rows(); ++i)
    for (int j = 0; j < C.cols(); ++j)
      C(i, j) = A(i, j) * B[j];
  //works?
  //const int ccols = C.cols();
  //const int ijmax = C.size();
  //for (int ij=0; ij<ijmax; ++ij)
  //  C(ij)=A(ij)*B(ij%ccols);
}


/// C = diag(A)*B
template<typename T1, typename T2, typename T3>
inline void diag_product(const Vector<T1>& A, const Matrix<T2>& B, Matrix<T3>& C)
{
  for (int i = 0; i < C.rows(); ++i)
    for (int j = 0; j < C.cols(); ++j)
      C(i, j) = A[i] * B(i, j);


  //const int crows = C.rows();
  //const int ccols = C.cols();
  //for (int i=0,ijb=0; i<crows; ++i,ijb+=ccols)
  //{
  //  const T1 a = A(i);
  //  for (int j=0; j<ccols; ++j)
  //  {
  //    int ij = ijb+j;
  //    C(ij)=a*B(ij);
  //  }
  //}
  //
  //const int crows = C.rows();
  //const int ijmax = C.size();
  //for (int ij=0; ij<ijmax; ++ij)
  //  C(ij)=A(ij%crows)*B(ij);
}


/** static function to perform C=AB for complex matrices
   *
   * Call zgemm
   */
inline void product(const Matrix<double>& A, const Matrix<std::complex<double>>& B, Matrix<double>& C)
{
  std::cerr << " Undefined C=AB with real A and complex B " << std::endl;
}

/** static function to perform y=Ax for generic matrix and vector
   */
template<typename T>
inline void product(const Matrix<T>& A, const Vector<T>& x, T* restrict yptr)
{
  const char transa = 'T';
  const T one       = 1.0;
  const T zero      = 0.0;
  BLAS::gemv(transa, A.cols(), A.rows(), one, A.data(), A.cols(), x.data(), 1, zero, yptr, 1);
}

/** static function to perform y = A^t x for generic matrix and vector
   */
template<typename T>
inline void product_Atx(const Matrix<T>& A, const Vector<T>& x, T* restrict yptr)
{
  const char transa = 'N';
  const T one       = 1.0;
  const T zero      = 0.0;
  BLAS::gemv(transa, A.cols(), A.rows(), one, A.data(), A.cols(), x.data(), 1, zero, yptr, 1);
}

/** static function to perform y=Ax for generic matrix and vector
   */
template<typename T>
inline void product(const Matrix<T>& A, const T* restrict xptr, T* restrict yptr)
{
  const char transa = 'T';
  const T one       = 1.0;
  const T zero      = 0.0;
  BLAS::gemv(transa, A.cols(), A.rows(), one, A.data(), A.cols(), xptr, 1, zero, yptr, 1);
}

/** static function to perform y = A^t x for generic matrix and vector
   */
template<typename T>
inline void product_Atx(const Matrix<T>& A, const T* restrict xptr, T* restrict yptr)
{
  const char transa = 'N';
  const T one       = 1.0;
  const T zero      = 0.0;
  BLAS::gemv(transa, A.cols(), A.rows(), one, A.data(), A.cols(), xptr, 1, zero, yptr, 1);
}

/** static function to perform y=Ax for generic matrix and vector
   */
template<typename T, unsigned D>
inline void product(const Matrix<T>& A, const TinyVector<T, D>* xvPtr, TinyVector<T, D>* restrict yptr)
{
  const T one       = 1.0;
  const T zero      = 0.0;
  const char transa = 'N';
  const char transb = 'N';
  BLAS::gemm(transa, transb, D, A.rows(), A.cols(), one, xvPtr->begin(), D, A.data(), A.cols(), zero, yptr->begin(), D);
}

template<typename T, unsigned D>
inline void product(const Matrix<T>& A, const Tensor<T, D>* xvPtr, Tensor<T, D>* restrict yptr)
{
  const T one       = 1.0;
  const T zero      = 0.0;
  const char transa = 'N';
  const char transb = 'N';
  BLAS::gemm(transa, transb, D * D, A.rows(), A.cols(), one, xvPtr->begin(), D * D, A.data(), A.cols(), zero,
             yptr->begin(), D * D);
}

/** static function to perform y=Ax for generic matrix and vector
   */
template<unsigned D>
inline void product(const Matrix<double>& A,
                    const Vector<TinyVector<double, D>>& x,
                    TinyVector<double, D>* restrict yptr)
{
  const double one  = 1.0;
  const double zero = 0.0;
  const char transa = 'N';
  const char transb = 'N';
  dgemm(transa, transb, D, A.rows(), x.size(), one, x.data()->begin(), D, A.data(), A.cols(), zero, yptr->begin(), D);
}

/** static function to perform y=Ax for generic matrix and vector
   */
template<unsigned D>
inline void product(const Matrix<std::complex<double>>& A,
                    const Vector<TinyVector<std::complex<double>, D>>& x,
                    TinyVector<std::complex<double>, D>* restrict yptr)
{
  const char transa = 'N';
  const char transb = 'N';
  const std::complex<double> zone(1.0, 0.0);
  const std::complex<double> zero(0.0, 0.0);
  zgemm(transa, transb, D, A.rows(), x.size(), zone, x.data()->begin(), D, A.data(), A.cols(), zero, yptr->begin(), D);
}


/** static function to perform y=Ax for generic matrix and vector
   */
inline void product(const Matrix<std::complex<double>>& A,
                    const Vector<std::complex<double>>& x,
                    std::complex<double>* restrict yptr)
{
  const char transa = 'T';
  const std::complex<double> zone(1.0, 0.0);
  const std::complex<double> zero(0.0, 0.0);
  zgemv(transa, A.cols(), A.rows(), zone, A.data(), A.cols(), x.data(), 1, zero, yptr, 1);
}

/** static function to perform y=Ax for generic matrix and vector
   */
inline void product(const Matrix<std::complex<double>>& A, const Vector<double>& x, std::complex<double>* restrict yptr)
{
  const int n                               = A.rows();
  const int m                               = A.cols();
  const std::complex<double>* restrict aptr = A.data();
  for (int i = 0, ij = 0; i < n; ++i)
  {
    std::complex<double> t = 0.0;
    for (int j = 0; j < m; ++j, ++ij)
      t += aptr[ij] * x[j];
    yptr[i] = t;
  }
}

inline void product(const Matrix<std::complex<double>>& A,
                    const std::complex<double>* restrict x,
                    std::complex<double>* restrict yptr)
{
  const char transa = 'T';
  const std::complex<double> zone(1.0, 0.0);
  const std::complex<double> zero(0.0, 0.0);
  zgemv(transa, A.cols(), A.rows(), zone, A.data(), A.cols(), x, 1, zero, yptr, 1);
}

/** static function to perform y=Ax for generic matrix and vector
   */
inline void product(const Matrix<double>& A, const Vector<std::complex<double>>& x, double* restrict yptr)
{
  std::cerr << " Undefined C=AB with real A and complex x " << std::endl;
}

template<typename T>
inline void product(const Matrix<T>& A, const Matrix<T>& B, Matrix<T>& C, std::vector<int>& offset)
{
  int nr = C.rows();
  int nb = offset.size() - 1;
  for (int i = 0; i < nr; i++)
  {
    for (int b = 0; b < nb; b++)
    {
      int firstK               = offset[b];
      int lastK                = offset[b + 1];
      const T* restrict firstY = A[i] + firstK;
      const T* restrict lastY  = A[i] + lastK;
      for (int k = firstK; k < lastK; k++)
      {
        C(i, k) = TESTDOT(firstY, lastY, B[k] + firstK);
      }
    }
  }
}

//    template<typename T>
//      inline statis void product(const Matrix<T>& A, const T* restrict x, T* restrict y)
//      {
//        GEMV<T,0>::apply(A.data(),x,y,A.rows(),A.cols());
//      }
} // namespace MatrixOperators

/** API to handle gemv */
namespace simd
{
template<typename T>
inline void gemv(const Matrix<T>& a, const T* restrict v, T* restrict b)
{
  MatrixOperators::product(a, v, b);
}

template<typename T, unsigned D>
inline void gemv(const Matrix<T>& a, const TinyVector<T, D>* restrict v, TinyVector<T, D>* restrict b)
{
  MatrixOperators::product(a, v, b);
}

template<typename T, unsigned D>
inline void gemv(const Matrix<T>& a, const Tensor<T, D>* restrict v, Tensor<T, D>* restrict b)
{
  MatrixOperators::product(a, v, b);
}

} // namespace simd

} // namespace qmcplusplus
#endif