src/colarray/test/getrf_test.cc

Go to the documentation of this file.

#include <vector>
#include <iomanip>

#include <sys/times.h>
#include <unistd.h>

#include <blas.h>
#include <matrix.h>
#include <vector.h>

using namespace std;

template <typename TYPE>
void fillMatrix(colarray::Matrix<TYPE>& A) {
  for (int i=0; i < A._m; i ++)
    for (int j=0; j < A._n; j ++)
      if (i == j)
        A(i,j) = 5*(A._n + A._m);
      else
        A(i,j) = i - j;
}

template <typename TYPE>
void bench() {
  colarray::Matrix<TYPE> *A;
  colarray::Vector<int> *piv;
  int nList[] = {10, 20, 50, 100, 200, 500, 1000, 1500, 2000, 2500, 3000};
  int nofTest = sizeof(nList) / sizeof(nList[0]);
  int n, count, info;
  clock_t t1, t2;
  double tWall, tTot, MFlops;
  struct tms timeBuf;
  
  long clk_tck = sysconf(_SC_CLK_TCK);
  
  for (int i = 0; i < nofTest; i ++) {
    n = nList[i];
    A = new colarray::Matrix< TYPE >(n, n);
    piv = new colarray::ColumnVector< blas::int_t >(n);
    
    count = 0; tTot = 0.0;
    do {
      fillMatrix(*A);

      t1 = times(&timeBuf);
      blas::getrf(*A, *piv, info);
      assert(info == 0);
      t2 = times(&timeBuf);

      tWall = static_cast< double >(t2-t1) / clk_tck;
      tTot += tWall;
      count ++;

    } while (tTot < 2.0);

    tTot /= count;

    delete A;
    delete piv;

    MFlops = 2.0/3.0 * n*n*n / tTot / 1e6;
    cout << "n=" << setw(6) << n << 
      ", Mflops=" << fixed << setw(8) << setprecision(1) << MFlops << endl;
  }

}

int main() {
  
  cout << "DGETRF performance:" << endl;
  bench< double >();
  cout << endl << "SGETRF performance:" << endl;
  bench< float >();
  
}

---