/* F06YA_A1W_F C++ Header Example Program.
 *
 * Copyright 2023 Numerical Algorithms Group.
 * Mark 29.1, 2023.
 */

#include <dco.hpp>
#include <iostream>
#include <math.h>
#include <nag.h>
#include <nagad.h>
#include <nagx04.h>
#include <stdio.h>
#include <string>
using namespace std;

int main()
{
  int               exit_status = 0;
  nag::ad::handle_t ad_handle;
  Integer           ifail = 0;

  cout << "F06YA_A1W_F C++ Header Example Program Results\n\n";
  // Skip heading in data file
  string mystr;
  getline(cin, mystr);

  // Read problem size and parameters
  Integer m, n, k;
  cin >> m >> n >> k;
  double alphar, betar;
  cin >> alphar >> betar;
  nagad_a1w_w_rtype alpha, beta;
  alpha = alphar;
  beta  = betar;

  // Allocate arrays containing A and its factorized form, B
  // and the solution X.
  nagad_a1w_w_rtype *a = 0, *b = 0, *c = 0;
  double *           ar = 0, *br = 0, *cr = 0, *da = 0, *db = 0;
  a  = new nagad_a1w_w_rtype[k * m];
  b  = new nagad_a1w_w_rtype[n * k];
  c  = new nagad_a1w_w_rtype[m * n];
  ar = new double[k * m];
  br = new double[n * k];
  cr = new double[m * n];
  da = new double[m * n];
  db = new double[m * n];

  // Read the matrix A
  double dd;
  for (int i = 0; i < k; ++i)
  {
    for (int j = 0; j < m; ++j)
    {
      cin >> dd;
      Integer l = i + j * k;
      ar[l]     = dd;
      a[l]      = dd;
    }
  }
  for (int i = 0; i < n; ++i)
  {
    for (int j = 0; j < k; ++j)
    {
      cin >> dd;
      Integer l = i + j * n;
      br[l]     = dd;
      b[l]      = dd;
    }
  }
  for (int i = 0; i < m * n; ++i)
  {
    cr[i] = 1.0;
    c[i]  = 1.0;
  }

  // Create AD tape
  dco::ga1s<double>::global_tape = dco::ga1s<double>::tape_t::create();

  // Create AD configuration data object
  ifail = 0;

  dco::ga1s<double>::global_tape->register_variable(alpha);
  dco::ga1s<double>::global_tape->register_variable(beta);

  //  y = beta*y + alpha*A^T*b
  ifail = 0;
  nag::ad::f06ya(ad_handle, "T", "T", m, n, k, alpha, a, k, b, n, beta, c, m,
                 ifail);

  for (int i = 0; i < m * n; ++i)
  {
    cr[i] = dco::value(c[i]);
  }

  cout << endl;
  NagError fail;
  INIT_FAIL(fail);
  x04cac(Nag_ColMajor, Nag_GeneralMatrix, Nag_NonUnitDiag, m, n, cr, m,
         "  Matrix C after update", 0, &fail);

  cout << "\n\n Derivatives calculated: First order adjoints\n";
  cout << " Computational mode    : algorithmic\n";
  cout << "\n Derivatives of C w.r.t alpha and beta:\n";

  for (int i = 0; i < m * n; ++i)
  {
    dco::ga1s<double>::global_tape->zero_adjoints();
    double inc = 1.0;
    dco::derivative(c[i]) += inc;
    ifail                                              = 0;
    dco::ga1s<double>::global_tape->sparse_interpret() = true;
    dco::ga1s<double>::global_tape->interpret_adjoint();
    da[i] = dco::derivative(alpha);
    db[i] = dco::derivative(beta);
  }

  // Print derivatives
  cout << endl;
  x04cac(Nag_ColMajor, Nag_GeneralMatrix, Nag_NonUnitDiag, m, n, da, m,
         "           dC/dalpha", 0, &fail);
  cout << endl;
  x04cac(Nag_ColMajor, Nag_GeneralMatrix, Nag_NonUnitDiag, m, n, db, m,
         "           dC/dbeta", 0, &fail);

  dco::ga1s<double>::tape_t::remove(dco::ga1s<double>::global_tape);

  return exit_status;
}