bench/eig33.cpp

// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.

// The computeRoots function included in this is based on materials
// covered by the following copyright and license:
//
// Geometric Tools, LLC
// Copyright (c) 1998-2010
// Distributed under the Boost Software License, Version 1.0.
//
// Permission is hereby granted, free of charge, to any person or organization
// obtaining a copy of the software and accompanying documentation covered by
// this license (the "Software") to use, reproduce, display, distribute,
// execute, and transmit the Software, and to prepare derivative works of the
// Software, and to permit third-parties to whom the Software is furnished to
// do so, all subject to the following:
//
// The copyright notices in the Software and this entire statement, including
// the above license grant, this restriction and the following disclaimer,
// must be included in all copies of the Software, in whole or in part, and
// all derivative works of the Software, unless such copies or derivative
// works are solely in the form of machine-executable object code generated by
// a source language processor.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
// SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
// FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
// ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
// DEALINGS IN THE SOFTWARE.

#include <iostream>
#include <Eigen/Core>
#include <Eigen/Eigenvalues>
#include <Eigen/Geometry>
#include <bench/BenchTimer.h>

using namespace Eigen;
using namespace std;

template <typename Matrix, typename Roots>
inline void computeRoots(const Matrix& m, Roots& roots) {
  typedef typename Matrix::Scalar Scalar;
  const Scalar s_inv3 = 1.0 / 3.0;
  const Scalar s_sqrt3 = std::sqrt(Scalar(3.0));

  // The characteristic equation is x^3 - c2*x^2 + c1*x - c0 = 0.  The
  // eigenvalues are the roots to this equation, all guaranteed to be
  // real-valued, because the matrix is symmetric.
  Scalar c0 = m(0, 0) * m(1, 1) * m(2, 2) + Scalar(2) * m(0, 1) * m(0, 2) * m(1, 2) - m(0, 0) * m(1, 2) * m(1, 2) -
              m(1, 1) * m(0, 2) * m(0, 2) - m(2, 2) * m(0, 1) * m(0, 1);
  Scalar c1 = m(0, 0) * m(1, 1) - m(0, 1) * m(0, 1) + m(0, 0) * m(2, 2) - m(0, 2) * m(0, 2) + m(1, 1) * m(2, 2) -
              m(1, 2) * m(1, 2);
  Scalar c2 = m(0, 0) + m(1, 1) + m(2, 2);

  // Construct the parameters used in classifying the roots of the equation
  // and in solving the equation for the roots in closed form.
  Scalar c2_over_3 = c2 * s_inv3;
  Scalar a_over_3 = (c1 - c2 * c2_over_3) * s_inv3;
  if (a_over_3 > Scalar(0)) a_over_3 = Scalar(0);

  Scalar half_b = Scalar(0.5) * (c0 + c2_over_3 * (Scalar(2) * c2_over_3 * c2_over_3 - c1));

  Scalar q = half_b * half_b + a_over_3 * a_over_3 * a_over_3;
  if (q > Scalar(0)) q = Scalar(0);

  // Compute the eigenvalues by solving for the roots of the polynomial.
  Scalar rho = std::sqrt(-a_over_3);
  Scalar theta = std::atan2(std::sqrt(-q), half_b) * s_inv3;
  Scalar cos_theta = std::cos(theta);
  Scalar sin_theta = std::sin(theta);
  roots(2) = c2_over_3 + Scalar(2) * rho * cos_theta;
  roots(0) = c2_over_3 - rho * (cos_theta + s_sqrt3 * sin_theta);
  roots(1) = c2_over_3 - rho * (cos_theta - s_sqrt3 * sin_theta);
}

template <typename Matrix, typename Vector>
void eigen33(const Matrix& mat, Matrix& evecs, Vector& evals) {
  typedef typename Matrix::Scalar Scalar;
  // Scale the matrix so its entries are in [-1,1].  The scaling is applied
  // only when at least one matrix entry has magnitude larger than 1.

  Scalar shift = mat.trace() / 3;
  Matrix scaledMat = mat;
  scaledMat.diagonal().array() -= shift;
  Scalar scale = scaledMat.cwiseAbs() /*.template triangularView<Lower>()*/.maxCoeff();
  scale = std::max(scale, Scalar(1));
  scaledMat /= scale;

  // Compute the eigenvalues
  //   scaledMat.setZero();
  computeRoots(scaledMat, evals);

  // compute the eigen vectors
  // **here we assume 3 different eigenvalues**

  // "optimized version" which appears to be slower with gcc!
  //     Vector base;
  //     Scalar alpha, beta;
  //     base <<   scaledMat(1,0) * scaledMat(2,1),
  //               scaledMat(1,0) * scaledMat(2,0),
  //              -scaledMat(1,0) * scaledMat(1,0);
  //     for(int k=0; k<2; ++k)
  //     {
  //       alpha = scaledMat(0,0) - evals(k);
  //       beta  = scaledMat(1,1) - evals(k);
  //       evecs.col(k) = (base + Vector(-beta*scaledMat(2,0), -alpha*scaledMat(2,1), alpha*beta)).normalized();
  //     }
  //     evecs.col(2) = evecs.col(0).cross(evecs.col(1)).normalized();

  //   // naive version
  //   Matrix tmp;
  //   tmp = scaledMat;
  //   tmp.diagonal().array() -= evals(0);
  //   evecs.col(0) = tmp.row(0).cross(tmp.row(1)).normalized();
  //
  //   tmp = scaledMat;
  //   tmp.diagonal().array() -= evals(1);
  //   evecs.col(1) = tmp.row(0).cross(tmp.row(1)).normalized();
  //
  //   tmp = scaledMat;
  //   tmp.diagonal().array() -= evals(2);
  //   evecs.col(2) = tmp.row(0).cross(tmp.row(1)).normalized();

  // a more stable version:
  if ((evals(2) - evals(0)) <= Eigen::NumTraits<Scalar>::epsilon()) {
    evecs.setIdentity();
  } else {
    Matrix tmp;
    tmp = scaledMat;
    tmp.diagonal().array() -= evals(2);
    evecs.col(2) = tmp.row(0).cross(tmp.row(1)).normalized();

    tmp = scaledMat;
    tmp.diagonal().array() -= evals(1);
    evecs.col(1) = tmp.row(0).cross(tmp.row(1));
    Scalar n1 = evecs.col(1).norm();
    if (n1 <= Eigen::NumTraits<Scalar>::epsilon())
      evecs.col(1) = evecs.col(2).unitOrthogonal();
    else
      evecs.col(1) /= n1;

    // make sure that evecs[1] is orthogonal to evecs[2]
    evecs.col(1) = evecs.col(2).cross(evecs.col(1).cross(evecs.col(2))).normalized();
    evecs.col(0) = evecs.col(2).cross(evecs.col(1));
  }

  // Rescale back to the original size.
  evals *= scale;
  evals.array() += shift;
}

int main() {
  BenchTimer t;
  int tries = 10;
  int rep = 400000;
  typedef Matrix3d Mat;
  typedef Vector3d Vec;
  Mat A = Mat::Random(3, 3);
  A = A.adjoint() * A;
  //   Mat Q = A.householderQr().householderQ();
  //   A = Q * Vec(2.2424567,2.2424566,7.454353).asDiagonal() * Q.transpose();

  SelfAdjointEigenSolver<Mat> eig(A);
  BENCH(t, tries, rep, eig.compute(A));
  std::cout << "Eigen iterative:  " << t.best() << "s\n";

  BENCH(t, tries, rep, eig.computeDirect(A));
  std::cout << "Eigen direct   :  " << t.best() << "s\n";

  Mat evecs;
  Vec evals;
  BENCH(t, tries, rep, eigen33(A, evecs, evals));
  std::cout << "Direct: " << t.best() << "s\n\n";

  //   std::cerr << "Eigenvalue/eigenvector diffs:\n";
  //   std::cerr << (evals - eig.eigenvalues()).transpose() << "\n";
  //   for(int k=0;k<3;++k)
  //     if(evecs.col(k).dot(eig.eigenvectors().col(k))<0)
  //       evecs.col(k) = -evecs.col(k);
  //   std::cerr << evecs - eig.eigenvectors() << "\n\n";
}