matrix.cpp

/*
Copyright 2011. All rights reserved.
Institute of Measurement and Control Systems
Karlsruhe Institute of Technology, Germany

This file is part of libviso2.
Authors: Andreas Geiger

libviso2 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 any later version.

libviso2 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
libviso2; if not, write to the Free Software Foundation, Inc., 51 Franklin
Street, Fifth Floor, Boston, MA 02110-1301, USA 
*/

#include "matrix.h"
#include <math.h>

#define SWAP(a,b) {temp=a;a=b;b=temp;}
#define SIGN(a,b) ((b) >= 0.0 ? fabs(a) : -fabs(a))
static FLOAT sqrarg;
#define SQR(a) ((sqrarg=(a)) == 0.0 ? 0.0 : sqrarg*sqrarg)
static FLOAT maxarg1,maxarg2;
#define FMAX(a,b) (maxarg1=(a),maxarg2=(b),(maxarg1) > (maxarg2) ? (maxarg1) : (maxarg2))
static int32_t iminarg1,iminarg2;
#define IMIN(a,b) (iminarg1=(a),iminarg2=(b),(iminarg1) < (iminarg2) ? (iminarg1) : (iminarg2))


using namespace std;

Matrix::Matrix () {
  m   = 0;
  n   = 0;
  val = 0;
}

Matrix::Matrix (const int32_t m_,const int32_t n_) {
  allocateMemory(m_,n_);
}

Matrix::Matrix (const int32_t m_,const int32_t n_,const FLOAT* val_) {
  allocateMemory(m_,n_);
  int32_t k=0;
  for (int32_t i=0; i<m_; i++)
    for (int32_t j=0; j<n_; j++)
      val[i][j] = val_[k++];
}

Matrix::Matrix (const Matrix &M) {
  allocateMemory(M.m,M.n);
  for (int32_t i=0; i<M.m; i++)
    memcpy(val[i],M.val[i],M.n*sizeof(FLOAT));
}

Matrix::~Matrix () {
  releaseMemory();
}

Matrix& Matrix::operator= (const Matrix &M) {
  if (this!=&M) {
    if (M.m!=m || M.n!=n) {
      releaseMemory();
      allocateMemory(M.m,M.n);
    }
    if (M.n>0)
      for (int32_t i=0; i<M.m; i++)
        memcpy(val[i],M.val[i],M.n*sizeof(FLOAT));
  }
  return *this;
}

void Matrix::getData(FLOAT* val_,int32_t i1,int32_t j1,int32_t i2,int32_t j2) {
  if (i2==-1) i2 = m-1;
  if (j2==-1) j2 = n-1;
  int32_t k=0;
  for (int32_t i=i1; i<=i2; i++)
    for (int32_t j=j1; j<=j2; j++)
      val_[k++] = val[i][j];
}

Matrix Matrix::getMat(int32_t i1,int32_t j1,int32_t i2,int32_t j2) {
  if (i2==-1) i2 = m-1;
  if (j2==-1) j2 = n-1;
  if (i1<0 || i2>=m || j1<0 || j2>=n || i2<i1 || j2<j1) {
    cerr << "ERROR: Cannot get submatrix [" << i1 << ".." << i2 <<
        "] x [" << j1 << ".." << j2 << "]" <<
        " of a (" << m << "x" << n << ") matrix." << endl;
    exit(0);
  }
  Matrix M(i2-i1+1,j2-j1+1);
  for (int32_t i=0; i<M.m; i++)
    for (int32_t j=0; j<M.n; j++)
      M.val[i][j] = val[i1+i][j1+j];
  return M;
}

void Matrix::setMat(const Matrix &M,const int32_t i1,const int32_t j1) {
  if (i1<0 || j1<0 || i1+M.m>m || j1+M.n>n) {
    cerr << "ERROR: Cannot set submatrix [" << i1 << ".." << i1+M.m-1 <<
        "] x [" << j1 << ".." << j1+M.n-1 << "]" <<
        " of a (" << m << "x" << n << ") matrix." << endl;
    exit(0);
  }
  for (int32_t i=0; i<M.m; i++)
    for (int32_t j=0; j<M.n; j++)
      val[i1+i][j1+j] = M.val[i][j];
}

void Matrix::setVal(FLOAT s,int32_t i1,int32_t j1,int32_t i2,int32_t j2) {
  if (i2==-1) i2 = m-1;
  if (j2==-1) j2 = n-1;
  if (i2<i1 || j2<j1) {
    cerr << "ERROR in setVal: Indices must be ordered (i1<=i2, j1<=j2)." << endl;
    exit(0);
  }
  for (int32_t i=i1; i<=i2; i++)
    for (int32_t j=j1; j<=j2; j++)
      val[i][j] = s;
}

void Matrix::setDiag(FLOAT s,int32_t i1,int32_t i2) {
  if (i2==-1) i2 = min(m-1,n-1);
  for (int32_t i=i1; i<=i2; i++)
    val[i][i] = s;
}

void Matrix::zero() {
  setVal(0);
}

Matrix Matrix::extractCols (vector<int> idx) {
  Matrix M(m,idx.size());
  for (int32_t j=0; j<M.n; j++)
    if (idx[j]<n)
      for (int32_t i=0; i<m; i++)
        M.val[i][j] = val[i][idx[j]];
  return M;
}

Matrix Matrix::eye (const int32_t m) {
  Matrix M(m,m);
  for (int32_t i=0; i<m; i++)
    M.val[i][i] = 1;
  return M;
}

void Matrix::eye () {
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      val[i][j] = 0;
  for (int32_t i=0; i<min(m,n); i++)
    val[i][i] = 1;
}

Matrix Matrix::diag (const Matrix &M) {
  if (M.m>1 && M.n==1) {
    Matrix D(M.m,M.m);
    for (int32_t i=0; i<M.m; i++)
      D.val[i][i] = M.val[i][0];
    return D;
  } else if (M.m==1 && M.n>1) {
    Matrix D(M.n,M.n);
    for (int32_t i=0; i<M.n; i++)
      D.val[i][i] = M.val[0][i];
    return D;
  }
  cout << "ERROR: Trying to create diagonal matrix from vector of size (" << M.m << "x" << M.n << ")" << endl;
  exit(0);
}

Matrix Matrix::reshape(const Matrix &M,int32_t m_,int32_t n_) {
  if (M.m*M.n != m_*n_) {
    cerr << "ERROR: Trying to reshape a matrix of size (" << M.m << "x" << M.n <<
            ") to size (" << m_ << "x" << n_ << ")" << endl;
    exit(0);
  }
  Matrix M2(m_,n_);
  for (int32_t k=0; k<m_*n_; k++) {
    int32_t i1 = k/M.n;
    int32_t j1 = k%M.n;
    int32_t i2 = k/n_;
    int32_t j2 = k%n_;
    M2.val[i2][j2] = M.val[i1][j1];
  }
  return M2;
}

Matrix Matrix::rotMatX (const FLOAT &angle) {
  FLOAT s = sin(angle);
  FLOAT c = cos(angle);
  Matrix R(3,3);
  R.val[0][0] = +1;
  R.val[1][1] = +c;
  R.val[1][2] = -s;
  R.val[2][1] = +s;
  R.val[2][2] = +c;
  return R;
}

Matrix Matrix::rotMatY (const FLOAT &angle) {
  FLOAT s = sin(angle);
  FLOAT c = cos(angle);
  Matrix R(3,3);
  R.val[0][0] = +c;
  R.val[0][2] = +s;
  R.val[1][1] = +1;
  R.val[2][0] = -s;
  R.val[2][2] = +c;
  return R;
}

Matrix Matrix::rotMatZ (const FLOAT &angle) {
  FLOAT s = sin(angle);
  FLOAT c = cos(angle);
  Matrix R(3,3);
  R.val[0][0] = +c;
  R.val[0][1] = -s;
  R.val[1][0] = +s;
  R.val[1][1] = +c;
  R.val[2][2] = +1;
  return R;
}

Matrix Matrix::operator+ (const Matrix &M) {
  const Matrix &A = *this;
  const Matrix &B = M;
  if (A.m!=B.m || A.n!=B.n) {
    cerr << "ERROR: Trying to add matrices of size (" << A.m << "x" << A.n <<
        ") and (" << B.m << "x" << B.n << ")" << endl;
    exit(0);
  }
  Matrix C(A.m,A.n);
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      C.val[i][j] = A.val[i][j]+B.val[i][j];
  return C;
}

Matrix Matrix::operator- (const Matrix &M) {
  const Matrix &A = *this;
  const Matrix &B = M;
  if (A.m!=B.m || A.n!=B.n) {
    cerr << "ERROR: Trying to subtract matrices of size (" << A.m << "x" << A.n <<
        ") and (" << B.m << "x" << B.n << ")" << endl;
    exit(0);
  }
  Matrix C(A.m,A.n);
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      C.val[i][j] = A.val[i][j]-B.val[i][j];
  return C;
}

Matrix Matrix::operator* (const Matrix &M) {
  const Matrix &A = *this;
  const Matrix &B = M;
  if (A.n!=B.m) {
    cerr << "ERROR: Trying to multiply matrices of size (" << A.m << "x" << A.n <<
        ") and (" << B.m << "x" << B.n << ")" << endl;
    exit(0);
  }
  Matrix C(A.m,B.n);
  for (int32_t i=0; i<A.m; i++)
    for (int32_t j=0; j<B.n; j++)
      for (int32_t k=0; k<A.n; k++)
        C.val[i][j] += A.val[i][k]*B.val[k][j];
  return C;
}

Matrix Matrix::operator* (const FLOAT &s) {
  Matrix C(m,n);
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      C.val[i][j] = val[i][j]*s;
  return C;
}

Matrix Matrix::operator/ (const Matrix &M) {
  const Matrix &A = *this;
  const Matrix &B = M;
  
  if (A.m==B.m && A.n==B.n) {
    Matrix C(A.m,A.n);
    for (int32_t i=0; i<A.m; i++)
      for (int32_t j=0; j<A.n; j++)
        if (B.val[i][j]!=0)
          C.val[i][j] = A.val[i][j]/B.val[i][j];
    return C;
    
  } else if (A.m==B.m && B.n==1) {
    Matrix C(A.m,A.n);
    for (int32_t i=0; i<A.m; i++)
      for (int32_t j=0; j<A.n; j++)
        if (B.val[i][0]!=0)
          C.val[i][j] = A.val[i][j]/B.val[i][0];
    return C;
    
  } else if (A.n==B.n && B.m==1) {
    Matrix C(A.m,A.n);
    for (int32_t i=0; i<A.m; i++)
      for (int32_t j=0; j<A.n; j++)
        if (B.val[0][j]!=0)
          C.val[i][j] = A.val[i][j]/B.val[0][j];
    return C;
    
  } else {
    cerr << "ERROR: Trying to divide matrices of size (" << A.m << "x" << A.n <<
        ") and (" << B.m << "x" << B.n << ")" << endl;
    exit(0);
  } 
}

Matrix Matrix::operator/ (const FLOAT &s) {
  if (fabs(s)<1e-20) {
    cerr << "ERROR: Trying to divide by zero!" << endl;
    exit(0);
  }
  Matrix C(m,n);
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      C.val[i][j] = val[i][j]/s;
  return C;
}

Matrix Matrix::operator- () {
  Matrix C(m,n);
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      C.val[i][j] = -val[i][j];
  return C;
}

Matrix Matrix::operator~ () {
  Matrix C(n,m);
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      C.val[j][i] = val[i][j];
  return C;
}

FLOAT Matrix::l2norm () {
  FLOAT norm = 0;
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      norm += val[i][j]*val[i][j];
  return sqrt(norm);
}

FLOAT Matrix::mean () {
  FLOAT mean = 0;
  for (int32_t i=0; i<m; i++)
    for (int32_t j=0; j<n; j++)
      mean += val[i][j];
  return mean/(FLOAT)(m*n);
}

Matrix Matrix::cross (const Matrix &a, const Matrix &b) {
  if (a.m!=3 || a.n!=1 || b.m!=3 || b.n!=1) {
    cerr << "ERROR: Cross product vectors must be of size (3x1)" << endl;
    exit(0);
  }
  Matrix c(3,1);
  c.val[0][0] = a.val[1][0]*b.val[2][0]-a.val[2][0]*b.val[1][0];
  c.val[1][0] = a.val[2][0]*b.val[0][0]-a.val[0][0]*b.val[2][0];
  c.val[2][0] = a.val[0][0]*b.val[1][0]-a.val[1][0]*b.val[0][0];
  return c;
}

Matrix Matrix::inv (const Matrix &M) {
  if (M.m!=M.n) {
    cerr << "ERROR: Trying to invert matrix of size (" << M.m << "x" << M.n << ")" << endl;
    exit(0);
  }
  Matrix A(M);
  Matrix B = eye(M.m);
  B.solve(A);
  return B;
}

bool Matrix::inv () {
  if (m!=n) {
    cerr << "ERROR: Trying to invert matrix of size (" << m << "x" << n << ")" << endl;
    exit(0);
  }
  Matrix A(*this);
  eye();
  solve(A);
  return true;
}

FLOAT Matrix::det () {
  
  if (m != n) {
    cerr << "ERROR: Trying to compute determinant of a matrix of size (" << m << "x" << n << ")" << endl;
    exit(0);
  }
    
  Matrix A(*this);
  int32_t *idx = (int32_t*)malloc(m*sizeof(int32_t));
  FLOAT d = 1;
  A.lu(idx,d);
  for( int32_t i=0; i<m; i++)
    d *= A.val[i][i];
  free(idx);
  return d;
}

bool Matrix::solve (const Matrix &M, FLOAT eps) {
  
  // substitutes
  const Matrix &A = M;
  Matrix &B       = *this;
  
  if (A.m != A.n || A.m != B.m || A.m<1 || B.n<1) {
    cerr << "ERROR: Trying to eliminate matrices of size (" << A.m << "x" << A.n <<
            ") and (" << B.m << "x" << B.n << ")" << endl;
    exit(0);
  }
  
  // index vectors for bookkeeping on the pivoting
  int32_t* indxc = new int32_t[m];
  int32_t* indxr = new int32_t[m];
  int32_t* ipiv  = new int32_t[m];
  
  // loop variables
  int32_t i, icol=0, irow=0, j, k, l, ll;
  FLOAT big, dum, pivinv, temp;
  
  // initialize pivots to zero
  for (j=0;j<m;j++) ipiv[j]=0;
  
  // main loop over the columns to be reduced
  for (i=0;i<m;i++) {
    
    big=0.0;
    
    // search for a pivot element
    for (j=0;j<m;j++)
      if (ipiv[j]!=1)
        for (k=0;k<m;k++)
          if (ipiv[k]==0)
            if (fabs(A.val[j][k])>=big) {
      big=fabs(A.val[j][k]);
      irow=j;
      icol=k;
            }
    ++(ipiv[icol]);
    
    // We now have the pivot element, so we interchange rows, if needed, to put the pivot
    // element on the diagonal. The columns are not physically interchanged, only relabeled.
    if (irow != icol) {
      for (l=0;l<m;l++) SWAP(A.val[irow][l], A.val[icol][l])
      for (l=0;l<n;l++) SWAP(B.val[irow][l], B.val[icol][l])
    }
    
    indxr[i]=irow; // We are now ready to divide the pivot row by the
    indxc[i]=icol; // pivot element, located at irow and icol.
    
    // check for singularity
    if (fabs(A.val[icol][icol]) < eps) {
      delete[] indxc;
      delete[] indxr;
      delete[] ipiv;
      return false;
    }
    
    pivinv=1.0/A.val[icol][icol];
    A.val[icol][icol]=1.0;
    for (l=0;l<m;l++) A.val[icol][l] *= pivinv;
    for (l=0;l<n;l++) B.val[icol][l] *= pivinv;
    
    // Next, we reduce the rows except for the pivot one
    for (ll=0;ll<m;ll++)
      if (ll!=icol) {
      dum = A.val[ll][icol];
      A.val[ll][icol] = 0.0;
      for (l=0;l<m;l++) A.val[ll][l] -= A.val[icol][l]*dum;
      for (l=0;l<n;l++) B.val[ll][l] -= B.val[icol][l]*dum;
      }
  }
  
  // This is the end of the main loop over columns of the reduction. It only remains to unscramble
  // the solution in view of the column interchanges. We do this by interchanging pairs of
  // columns in the reverse order that the permutation was built up.
  for (l=m-1;l>=0;l--) {
    if (indxr[l]!=indxc[l])
      for (k=0;k<m;k++)
        SWAP(A.val[k][indxr[l]], A.val[k][indxc[l]])
  }
  
  // success
  delete[] indxc;
  delete[] indxr;
  delete[] ipiv;
  return true;
}

// Given a matrix a[1..n][1..n], this routine replaces it by the LU decomposition of a rowwise
// permutation of itself. a and n are input. a is output, arranged as in equation (2.3.14) above;
// indx[1..n] is an output vector that records the row permutation effected by the partial
// pivoting; d is output as ±1 depending on whether the number of row interchanges was even
// or odd, respectively. This routine is used in combination with lubksb to solve linear equations
// or invert a matrix.

bool Matrix::lu(int32_t *idx, FLOAT &d, FLOAT eps) {
  
  if (m != n) {
    cerr << "ERROR: Trying to LU decompose a matrix of size (" << m << "x" << n << ")" << endl;
    exit(0);
  }
  
  int32_t i,imax=0,j,k;
  FLOAT   big,dum,sum,temp;
  FLOAT* vv = (FLOAT*)malloc(n*sizeof(FLOAT)); // vv stores the implicit scaling of each row.
  d = 1.0;
  for (i=0; i<n; i++) { // Loop over rows to get the implicit scaling information.
    big = 0.0;
    for (j=0; j<n; j++)
      if ((temp=fabs(val[i][j]))>big)
        big = temp;
    if (big == 0.0) { // No nonzero largest element.
      free(vv);
      return false;
    }
    vv[i] = 1.0/big; // Save the scaling.
  }
  for (j=0; j<n; j++) { // This is the loop over columns of Crout’s method.
    for (i=0; i<j; i++) { // This is equation (2.3.12) except for i = j.
      sum = val[i][j];
      for (k=0; k<i; k++)
        sum -= val[i][k]*val[k][j];
      val[i][j] = sum;
    }
    big = 0.0; // Initialize the search for largest pivot element.
    for (i=j; i<n; i++) {
      sum = val[i][j];
      for (k=0; k<j; k++)
        sum -= val[i][k]*val[k][j];
      val[i][j] = sum;
      if ( (dum=vv[i]*fabs(sum))>=big) {
        big  = dum;
        imax = i;
      }
    }
    if (j!=imax) { // Do we need to interchange rows?
      for (k=0; k<n; k++) { // Yes, do so...
        dum          = val[imax][k];
        val[imax][k] = val[j][k];
        val[j][k]    = dum;
      }
      d = -d;     // ...and change the parity of d.
      vv[imax]=vv[j]; // Also interchange the scale factor.
    }
    idx[j] = imax;
    if (j!=n-1) { // Now, finally, divide by the pivot element.
      dum = 1.0/val[j][j];
      for (i=j+1; i<n; i++)
        val[i][j] *= dum;
    }
  } // Go back for the next column in the reduction.
  
  // success
  free(vv);
  return true;
}

// Given a matrix M/A[1..m][1..n], this routine computes its singular value decomposition, M/A =
// U·W·V T. Thematrix U replaces a on output. The diagonal matrix of singular values W is output
// as a vector w[1..n]. Thematrix V (not the transpose V T ) is output as v[1..n][1..n].
void Matrix::svd(Matrix &U2,Matrix &W,Matrix &V) {

  Matrix U = Matrix(*this);
  U2 = Matrix(m,m);
  V  = Matrix(n,n);

  FLOAT* w   = (FLOAT*)malloc(n*sizeof(FLOAT));
  FLOAT* rv1 = (FLOAT*)malloc(n*sizeof(FLOAT));

  int32_t flag,i,its,j,jj,k,l,nm;
  FLOAT   anorm,c,f,g,h,s,scale,x,y,z;

  g = scale = anorm = 0.0; // Householder reduction to bidiagonal form.
  for (i=0;i<n;i++) {
    l = i+1;
    rv1[i] = scale*g;
    g = s = scale = 0.0;
    if (i < m) {
      for (k=i;k<m;k++) scale += fabs(U.val[k][i]);
      if (scale) {
        for (k=i;k<m;k++) {
          U.val[k][i] /= scale;
          s += U.val[k][i]*U.val[k][i];
        }
        f = U.val[i][i];
        g = -SIGN(sqrt(s),f);
        h = f*g-s;
        U.val[i][i] = f-g;
        for (j=l;j<n;j++) {
          for (s=0.0,k=i;k<m;k++) s += U.val[k][i]*U.val[k][j];
          f = s/h;
          for (k=i;k<m;k++) U.val[k][j] += f*U.val[k][i];
        }
        for (k=i;k<m;k++) U.val[k][i] *= scale;
      }
    }
    w[i] = scale*g;
    g = s = scale = 0.0;
    if (i<m && i!=n-1) {
      for (k=l;k<n;k++) scale += fabs(U.val[i][k]);
      if (scale) {
        for (k=l;k<n;k++) {
          U.val[i][k] /= scale;
          s += U.val[i][k]*U.val[i][k];
        }
        f = U.val[i][l];
        g = -SIGN(sqrt(s),f);
        h = f*g-s;
        U.val[i][l] = f-g;
        for (k=l;k<n;k++) rv1[k] = U.val[i][k]/h;
        for (j=l;j<m;j++) {
          for (s=0.0,k=l;k<n;k++) s += U.val[j][k]*U.val[i][k];
          for (k=l;k<n;k++) U.val[j][k] += s*rv1[k];
        }
        for (k=l;k<n;k++) U.val[i][k] *= scale;
      }
    }
    anorm = FMAX(anorm,(fabs(w[i])+fabs(rv1[i])));
  }
  for (i=n-1;i>=0;i--) { // Accumulation of right-hand transformations.
    if (i<n-1) {
      if (g) {
        for (j=l;j<n;j++) // Double division to avoid possible underflow.
          V.val[j][i]=(U.val[i][j]/U.val[i][l])/g;
        for (j=l;j<n;j++) {
          for (s=0.0,k=l;k<n;k++) s += U.val[i][k]*V.val[k][j];
          for (k=l;k<n;k++) V.val[k][j] += s*V.val[k][i];
        }
      }
      for (j=l;j<n;j++) V.val[i][j] = V.val[j][i] = 0.0;
    }
    V.val[i][i] = 1.0;
    g = rv1[i];
    l = i;
  }
  for (i=IMIN(m,n)-1;i>=0;i--) { // Accumulation of left-hand transformations.
    l = i+1;
    g = w[i];
    for (j=l;j<n;j++) U.val[i][j] = 0.0;
    if (g) {
      g = 1.0/g;
      for (j=l;j<n;j++) {
        for (s=0.0,k=l;k<m;k++) s += U.val[k][i]*U.val[k][j];
        f = (s/U.val[i][i])*g;
        for (k=i;k<m;k++) U.val[k][j] += f*U.val[k][i];
      }
      for (j=i;j<m;j++) U.val[j][i] *= g;
    } else for (j=i;j<m;j++) U.val[j][i]=0.0;
    ++U.val[i][i];
  }
  for (k=n-1;k>=0;k--) { // Diagonalization of the bidiagonal form: Loop over singular values,
    for (its=0;its<30;its++) { // and over allowed iterations.
      flag = 1;
      for (l=k;l>=0;l--) { // Test for splitting.
        nm = l-1;
        if ((FLOAT)(fabs(rv1[l])+anorm) == anorm) { flag = 0; break; }
        if ((FLOAT)(fabs( w[nm])+anorm) == anorm) { break; }
      }
      if (flag) {
        c = 0.0; // Cancellation of rv1[l], if l > 1.
        s = 1.0;
        for (i=l;i<=k;i++) {
          f = s*rv1[i];
          rv1[i] = c*rv1[i];
          if ((FLOAT)(fabs(f)+anorm) == anorm) break;
          g = w[i];
          h = pythag(f,g);
          w[i] = h;
          h = 1.0/h;
          c = g*h;
          s = -f*h;
          for (j=0;j<m;j++) {
            y = U.val[j][nm];
            z = U.val[j][i];
            U.val[j][nm] = y*c+z*s;
            U.val[j][i]  = z*c-y*s;
          }
        }
      }
      z = w[k];
      if (l==k) { // Convergence.
        if (z<0.0) { // Singular value is made nonnegative.
          w[k] = -z;
          for (j=0;j<n;j++) V.val[j][k] = -V.val[j][k];
        }
        break;
      }
      if (its == 29)
        cerr << "ERROR in SVD: No convergence in 30 iterations" << endl;
      x = w[l]; // Shift from bottom 2-by-2 minor.
      nm = k-1;
      y = w[nm];
      g = rv1[nm];
      h = rv1[k];
      f = ((y-z)*(y+z)+(g-h)*(g+h))/(2.0*h*y);
      g = pythag(f,1.0);
      f = ((x-z)*(x+z)+h*((y/(f+SIGN(g,f)))-h))/x;
      c = s = 1.0; // Next QR transformation:
      for (j=l;j<=nm;j++) {
        i = j+1;
        g = rv1[i];
        y = w[i];
        h = s*g;
        g = c*g;
        z = pythag(f,h);
        rv1[j] = z;
        c = f/z;
        s = h/z;
        f = x*c+g*s;
        g = g*c-x*s;
        h = y*s;
        y *= c;
        for (jj=0;jj<n;jj++) {
          x = V.val[jj][j];
          z = V.val[jj][i];
          V.val[jj][j] = x*c+z*s;
          V.val[jj][i] = z*c-x*s;
        }
        z = pythag(f,h);
        w[j] = z; // Rotation can be arbitrary if z = 0.
        if (z) {
          z = 1.0/z;
          c = f*z;
          s = h*z;
        }
        f = c*g+s*y;
        x = c*y-s*g;
        for (jj=0;jj<m;jj++) {
          y = U.val[jj][j];
          z = U.val[jj][i];
          U.val[jj][j] = y*c+z*s;
          U.val[jj][i] = z*c-y*s;
        }
      }
      rv1[l] = 0.0;
      rv1[k] = f;
      w[k] = x;
    }
  }
  
  // sort singular values and corresponding columns of u and v
  // by decreasing magnitude. Also, signs of corresponding columns are
  // flipped so as to maximize the number of positive elements.
  int32_t s2,inc=1;
  FLOAT   sw;
  FLOAT* su = (FLOAT*)malloc(m*sizeof(FLOAT));
  FLOAT* sv = (FLOAT*)malloc(n*sizeof(FLOAT));
  do { inc *= 3; inc++; } while (inc <= n);
  do {
    inc /= 3;
    for (i=inc;i<n;i++) {
      sw = w[i];
      for (k=0;k<m;k++) su[k] = U.val[k][i];
      for (k=0;k<n;k++) sv[k] = V.val[k][i];
      j = i;
      while (w[j-inc] < sw) {
        w[j] = w[j-inc];
        for (k=0;k<m;k++) U.val[k][j] = U.val[k][j-inc];
        for (k=0;k<n;k++) V.val[k][j] = V.val[k][j-inc];
        j -= inc;
        if (j < inc) break;
      }
      w[j] = sw;
      for (k=0;k<m;k++) U.val[k][j] = su[k];
      for (k=0;k<n;k++) V.val[k][j] = sv[k];
    }
  } while (inc > 1);
  for (k=0;k<n;k++) { // flip signs
    s2=0;
    for (i=0;i<m;i++) if (U.val[i][k] < 0.0) s2++;
    for (j=0;j<n;j++) if (V.val[j][k] < 0.0) s2++;
    if (s2 > (m+n)/2) {
      for (i=0;i<m;i++) U.val[i][k] = -U.val[i][k];
      for (j=0;j<n;j++) V.val[j][k] = -V.val[j][k];
    }
  }

  // create vector and copy singular values
  W = Matrix(min(m,n),1,w);
  
  // extract mxm submatrix U
  U2.setMat(U.getMat(0,0,m-1,min(m-1,n-1)),0,0);

  // release temporary memory
  free(w);
  free(rv1);
  free(su);
  free(sv);
}

ostream& operator<< (ostream& out,const Matrix& M) {
  if (M.m==0 || M.n==0) {
    out << "[empty matrix]";
  } else {
    char buffer[1024];
    for (int32_t i=0; i<M.m; i++) {
      for (int32_t j=0; j<M.n; j++) {
        sprintf(buffer,"%12.7f ",M.val[i][j]);
        out << buffer;
      }
      if (i<M.m-1)
        out << endl;
    }
  }
  return out;
}

void Matrix::allocateMemory (const int32_t m_,const int32_t n_) {
  m = abs(m_); n = abs(n_);
  if (m==0 || n==0) {
    val = 0;
    return;
  }
  val    = (FLOAT**)malloc(m*sizeof(FLOAT*));
  val[0] = (FLOAT*)calloc(m*n,sizeof(FLOAT));
  for(int32_t i=1; i<m; i++)
    val[i] = val[i-1]+n;
}

void Matrix::releaseMemory () {
  if (val!=0) {
    free(val[0]);
    free(val);
  }
}

FLOAT Matrix::pythag(FLOAT a,FLOAT b) {
  FLOAT absa,absb;
  absa = fabs(a);
  absb = fabs(b);
  if (absa > absb)
    return absa*sqrt(1.0+SQR(absb/absa));
  else
    return (absb == 0.0 ? 0.0 : absb*sqrt(1.0+SQR(absa/absb)));
}