view libinterp/corefcn/sqrtm.cc @ 21301:40de9f8f23a6

Use '#include "config.h"' rather than <config.h>. * mk-octave-config-h.sh, mk-opts.pl, Backend.cc, BaseControl.cc, ButtonControl.cc, Canvas.cc, CheckBoxControl.cc, Container.cc, ContextMenu.cc, EditControl.cc, Figure.cc, FigureWindow.cc, GLCanvas.cc, KeyMap.cc, ListBoxControl.cc, Logger.cc, Menu.cc, MouseModeActionGroup.cc, Object.cc, ObjectFactory.cc, ObjectProxy.cc, Panel.cc, PopupMenuControl.cc, PushButtonControl.cc, PushTool.cc, QtHandlesUtils.cc, RadioButtonControl.cc, SliderControl.cc, TextControl.cc, TextEdit.cc, ToggleButtonControl.cc, ToggleTool.cc, ToolBar.cc, ToolBarButton.cc, __init_qt__.cc, annotation-dialog.cc, gl-select.cc, module.mk, kpty.cpp, color-picker.cc, dialog.cc, documentation-dock-widget.cc, files-dock-widget.cc, find-files-dialog.cc, find-files-model.cc, history-dock-widget.cc, file-editor-tab.cc, file-editor-tab.h, file-editor.cc, find-dialog.cc, marker.cc, octave-qscintilla.cc, octave-txt-lexer.cc, main-window.cc, octave-cmd.cc, octave-dock-widget.cc, octave-gui.cc, octave-interpreter.cc, octave-qt-link.cc, parser.cc, webinfo.cc, resource-manager.cc, settings-dialog.cc, shortcut-manager.cc, terminal-dock-widget.cc, thread-manager.cc, welcome-wizard.cc, workspace-model.cc, workspace-view.cc, build-env-features.sh, build-env.in.cc, Cell.cc, __contourc__.cc, __dispatch__.cc, __dsearchn__.cc, __ichol__.cc, __ilu__.cc, __lin_interpn__.cc, __pchip_deriv__.cc, __qp__.cc, balance.cc, besselj.cc, betainc.cc, bitfcns.cc, bsxfun.cc, c-file-ptr-stream.cc, cdisplay.c, cellfun.cc, coct-hdf5-types.c, colloc.cc, comment-list.cc, conv2.cc, daspk.cc, dasrt.cc, dassl.cc, data.cc, debug.cc, defaults.cc, defun.cc, det.cc, dirfns.cc, display.cc, dlmread.cc, dot.cc, dynamic-ld.cc, eig.cc, ellipj.cc, error.cc, errwarn.cc, event-queue.cc, fft.cc, fft2.cc, fftn.cc, file-io.cc, filter.cc, find.cc, ft-text-renderer.cc, gammainc.cc, gcd.cc, getgrent.cc, getpwent.cc, getrusage.cc, givens.cc, gl-render.cc, gl2ps-print.cc, graphics.cc, gripes.cc, hash.cc, help.cc, hess.cc, hex2num.cc, hook-fcn.cc, input.cc, inv.cc, jit-ir.cc, jit-typeinfo.cc, jit-util.cc, kron.cc, load-path.cc, load-save.cc, lookup.cc, ls-ascii-helper.cc, ls-hdf5.cc, ls-mat-ascii.cc, ls-mat4.cc, ls-mat5.cc, ls-oct-binary.cc, ls-oct-text.cc, ls-utils.cc, lsode.cc, lu.cc, luinc.cc, mappers.cc, matrix_type.cc, max.cc, mex.cc, mgorth.cc, nproc.cc, oct-errno.in.cc, oct-fstrm.cc, oct-hdf5-types.cc, oct-hist.cc, oct-iostrm.cc, oct-lvalue.cc, oct-map.cc, oct-prcstrm.cc, oct-procbuf.cc, oct-stream.cc, oct-strstrm.cc, oct-tex-lexer.in.ll, oct-tex-parser.in.yy, octave-link.cc, ordschur.cc, pager.cc, pinv.cc, pr-output.cc, procstream.cc, profiler.cc, psi.cc, pt-jit.cc, quad.cc, quadcc.cc, qz.cc, rand.cc, rcond.cc, regexp.cc, schur.cc, sighandlers.cc, siglist.c, sparse-xdiv.cc, sparse-xpow.cc, sparse.cc, spparms.cc, sqrtm.cc, str2double.cc, strfind.cc, strfns.cc, sub2ind.cc, svd.cc, sylvester.cc, symtab.cc, syscalls.cc, sysdep.cc, text-renderer.cc, time.cc, toplev.cc, tril.cc, tsearch.cc, txt-eng.cc, typecast.cc, urlwrite.cc, utils.cc, variables.cc, xdiv.cc, xgl2ps.c, xnorm.cc, xpow.cc, zfstream.cc, __delaunayn__.cc, __eigs__.cc, __fltk_uigetfile__.cc, __glpk__.cc, __init_fltk__.cc, __init_gnuplot__.cc, __magick_read__.cc, __osmesa_print__.cc, __voronoi__.cc, amd.cc, audiodevinfo.cc, audioread.cc, ccolamd.cc, chol.cc, colamd.cc, convhulln.cc, dmperm.cc, fftw.cc, qr.cc, symbfact.cc, symrcm.cc, mkbuiltins, mkops, ov-base-diag.cc, ov-base-int.cc, ov-base-mat.cc, ov-base-scalar.cc, ov-base-sparse.cc, ov-base.cc, ov-bool-mat.cc, ov-bool-sparse.cc, ov-bool.cc, ov-builtin.cc, ov-cell.cc, ov-ch-mat.cc, ov-class.cc, ov-classdef.cc, ov-colon.cc, ov-complex.cc, ov-cs-list.cc, ov-cx-diag.cc, ov-cx-mat.cc, ov-cx-sparse.cc, ov-dld-fcn.cc, ov-fcn-handle.cc, ov-fcn-inline.cc, ov-fcn.cc, ov-float.cc, ov-flt-complex.cc, ov-flt-cx-diag.cc, ov-flt-cx-mat.cc, ov-flt-re-diag.cc, ov-flt-re-mat.cc, ov-int16.cc, ov-int32.cc, ov-int64.cc, ov-int8.cc, ov-java.cc, ov-lazy-idx.cc, ov-mex-fcn.cc, ov-null-mat.cc, ov-oncleanup.cc, ov-perm.cc, ov-range.cc, ov-re-diag.cc, ov-re-mat.cc, ov-re-sparse.cc, ov-scalar.cc, ov-str-mat.cc, ov-struct.cc, ov-typeinfo.cc, ov-uint16.cc, ov-uint32.cc, ov-uint64.cc, ov-uint8.cc, ov-usr-fcn.cc, ov.cc, ovl.cc, octave.cc, op-b-b.cc, op-b-bm.cc, op-b-sbm.cc, op-bm-b.cc, op-bm-bm.cc, op-bm-sbm.cc, op-cdm-cdm.cc, op-cell.cc, op-chm.cc, op-class.cc, op-cm-cm.cc, op-cm-cs.cc, op-cm-m.cc, op-cm-s.cc, op-cm-scm.cc, op-cm-sm.cc, op-cs-cm.cc, op-cs-cs.cc, op-cs-m.cc, op-cs-s.cc, op-cs-scm.cc, op-cs-sm.cc, op-dm-dm.cc, op-dm-scm.cc, op-dm-sm.cc, op-dm-template.cc, op-dms-template.cc, op-double-conv.cc, op-fcdm-fcdm.cc, op-fcdm-fdm.cc, op-fcm-fcm.cc, op-fcm-fcs.cc, op-fcm-fm.cc, op-fcm-fs.cc, op-fcn.cc, op-fcs-fcm.cc, op-fcs-fcs.cc, op-fcs-fm.cc, op-fcs-fs.cc, op-fdm-fdm.cc, op-float-conv.cc, op-fm-fcm.cc, op-fm-fcs.cc, op-fm-fm.cc, op-fm-fs.cc, op-fs-fcm.cc, op-fs-fcs.cc, op-fs-fm.cc, op-fs-fs.cc, op-i16-i16.cc, op-i32-i32.cc, op-i64-i64.cc, op-i8-i8.cc, op-int-concat.cc, op-int-conv.cc, op-m-cm.cc, op-m-cs.cc, op-m-m.cc, op-m-s.cc, op-m-scm.cc, op-m-sm.cc, op-pm-pm.cc, op-pm-scm.cc, op-pm-sm.cc, op-pm-template.cc, op-range.cc, op-s-cm.cc, op-s-cs.cc, op-s-m.cc, op-s-s.cc, op-s-scm.cc, op-s-sm.cc, op-sbm-b.cc, op-sbm-bm.cc, op-sbm-sbm.cc, op-scm-cm.cc, op-scm-cs.cc, op-scm-m.cc, op-scm-s.cc, op-scm-scm.cc, op-scm-sm.cc, op-sm-cm.cc, op-sm-cs.cc, op-sm-m.cc, op-sm-s.cc, op-sm-scm.cc, op-sm-sm.cc, op-str-m.cc, op-str-s.cc, op-str-str.cc, op-struct.cc, op-ui16-ui16.cc, op-ui32-ui32.cc, op-ui64-ui64.cc, op-ui8-ui8.cc, lex.ll, oct-parse.in.yy, pt-arg-list.cc, pt-array-list.cc, pt-assign.cc, pt-binop.cc, pt-bp.cc, pt-cbinop.cc, pt-cell.cc, pt-check.cc, pt-classdef.cc, pt-cmd.cc, pt-colon.cc, pt-const.cc, pt-decl.cc, pt-eval.cc, pt-except.cc, pt-exp.cc, pt-fcn-handle.cc, pt-funcall.cc, pt-id.cc, pt-idx.cc, pt-jump.cc, pt-loop.cc, pt-mat.cc, pt-misc.cc, pt-pr-code.cc, pt-select.cc, pt-stmt.cc, pt-unop.cc, pt.cc, token.cc, Array-jit.cc, Array-os.cc, Array-sym.cc, Array-tc.cc, version.cc, Array-C.cc, Array-b.cc, Array-ch.cc, Array-d.cc, Array-f.cc, Array-fC.cc, Array-i.cc, Array-idx-vec.cc, Array-s.cc, Array-str.cc, Array-util.cc, Array-voidp.cc, Array.cc, CColVector.cc, CDiagMatrix.cc, CMatrix.cc, CNDArray.cc, CRowVector.cc, CSparse.cc, DiagArray2.cc, MArray-C.cc, MArray-d.cc, MArray-f.cc, MArray-fC.cc, MArray-i.cc, MArray-s.cc, MArray.cc, MDiagArray2.cc, MSparse-C.cc, MSparse-d.cc, MatrixType.cc, PermMatrix.cc, Range.cc, Sparse-C.cc, Sparse-b.cc, Sparse-d.cc, Sparse.cc, boolMatrix.cc, boolNDArray.cc, boolSparse.cc, chMatrix.cc, chNDArray.cc, dColVector.cc, dDiagMatrix.cc, dMatrix.cc, dNDArray.cc, dRowVector.cc, dSparse.cc, dim-vector.cc, fCColVector.cc, fCDiagMatrix.cc, fCMatrix.cc, fCNDArray.cc, fCRowVector.cc, fColVector.cc, fDiagMatrix.cc, fMatrix.cc, fNDArray.cc, fRowVector.cc, idx-vector.cc, int16NDArray.cc, int32NDArray.cc, int64NDArray.cc, int8NDArray.cc, intNDArray.cc, uint16NDArray.cc, uint32NDArray.cc, uint64NDArray.cc, uint8NDArray.cc, blaswrap.c, cquit.c, f77-extern.cc, f77-fcn.c, lo-error.c, quit.cc, CollocWt.cc, DASPK.cc, DASRT.cc, DASSL.cc, EIG.cc, LSODE.cc, ODES.cc, Quad.cc, aepbalance.cc, chol.cc, eigs-base.cc, fEIG.cc, gepbalance.cc, hess.cc, lo-mappers.cc, lo-specfun.cc, lu.cc, oct-convn.cc, oct-fftw.cc, oct-norm.cc, oct-rand.cc, oct-spparms.cc, qr.cc, qrp.cc, randgamma.c, randmtzig.c, randpoisson.c, schur.cc, sparse-chol.cc, sparse-dmsolve.cc, sparse-lu.cc, sparse-qr.cc, svd.cc, mk-ops.awk, dir-ops.cc, file-ops.cc, file-stat.cc, lo-sysdep.cc, mach-info.cc, oct-env.cc, oct-group.cc, oct-passwd.cc, oct-syscalls.cc, oct-time.cc, oct-uname.cc, cmd-edit.cc, cmd-hist.cc, data-conv.cc, f2c-main.c, glob-match.cc, kpse.cc, lo-array-errwarn.cc, lo-array-gripes.cc, lo-cutils.c, lo-ieee.cc, lo-regexp.cc, lo-utils.cc, oct-base64.cc, oct-glob.cc, oct-inttypes.cc, oct-locbuf.cc, oct-mutex.cc, oct-rl-edit.c, oct-rl-hist.c, oct-shlib.cc, oct-sort.cc, pathsearch.cc, singleton-cleanup.cc, sparse-sort.cc, sparse-util.cc, str-vec.cc, unwind-prot.cc, url-transfer.cc, display-available.c, main-cli.cc, main-gui.cc, main.in.cc, mkoctfile.in.cc, octave-config.in.cc: Use '#include "config.h"' rather than <config.h>.
author Rik <rik@octave.org>
date Thu, 18 Feb 2016 13:34:50 -0800
parents e69eaee28737
children bc9aa534bc29
line wrap: on
line source

/*

Copyright (C) 2001-2015 Ross Lippert and Paul Kienzle
Copyright (C) 2010 VZLU Prague

This file is part of Octave.

Octave 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 3 of the License, or (at your
option) any later version.

Octave 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 Octave; see the file COPYING.  If not, see
<http://www.gnu.org/licenses/>.

*/

#ifdef HAVE_CONFIG_H
#  include "config.h"
#endif

#include <float.h>

#include "schur.h"
#include "lo-ieee.h"
#include "lo-mappers.h"
#include "oct-norm.h"

#include "defun.h"
#include "error.h"
#include "errwarn.h"
#include "utils.h"
#include "xnorm.h"

template <typename Matrix>
static void
sqrtm_utri_inplace (Matrix& T)
{
  typedef typename Matrix::element_type element_type;

  const element_type zero = element_type ();

  bool singular = false;

  // The following code is equivalent to this triple loop:
  //
  //   n = rows (T);
  //   for j = 1:n
  //     T(j,j) = sqrt (T(j,j));
  //     for i = j-1:-1:1
  //       T(i,j) /= (T(i,i) + T(j,j));
  //       k = 1:i-1;
  //       T(k,j) -= T(k,i) * T(i,j);
  //     endfor
  //   endfor
  //
  // this is an in-place, cache-aligned variant of the code
  // given in Higham's paper.

  const octave_idx_type n = T.rows ();
  element_type *Tp = T.fortran_vec ();
  for (octave_idx_type j = 0; j < n; j++)
    {
      element_type *colj = Tp + n*j;
      if (colj[j] != zero)
        colj[j] = sqrt (colj[j]);
      else
        singular = true;

      for (octave_idx_type i = j-1; i >= 0; i--)
        {
          const element_type *coli = Tp + n*i;
          const element_type colji = colj[i] /= (coli[i] + colj[j]);
          for (octave_idx_type k = 0; k < i; k++)
            colj[k] -= coli[k] * colji;
        }
    }

  if (singular)
    warning_with_id ("Octave:sqrtm:SingularMatrix",
                     "sqrtm: matrix is singular, may not have a square root");
}

template <typename Matrix, typename ComplexMatrix, typename ComplexSCHUR>
static octave_value
do_sqrtm (const octave_value& arg)
{

  octave_value retval;

  MatrixType mt = arg.matrix_type ();

  bool iscomplex = arg.is_complex_type ();

  typedef typename Matrix::element_type real_type;

  real_type cutoff = 0;
  real_type one = 1;
  real_type eps = std::numeric_limits<real_type>::epsilon ();

  if (! iscomplex)
    {
      Matrix x = octave_value_extract<Matrix> (arg);

      if (mt.is_unknown ()) // if type is not known, compute it now.
        arg.matrix_type (mt = MatrixType (x));

      switch (mt.type ())
        {
        case MatrixType::Upper:
        case MatrixType::Diagonal:
          if (! x.diag ().any_element_is_negative ())
            {
              // Do it in real arithmetic.
              sqrtm_utri_inplace (x);
              retval = x;
              retval.matrix_type (mt);
            }
          else
            iscomplex = true;
          break;

        case MatrixType::Lower:
          if (! x.diag ().any_element_is_negative ())
            {
              x = x.transpose ();
              sqrtm_utri_inplace (x);
              retval = x.transpose ();
              retval.matrix_type (mt);
            }
          else
            iscomplex = true;
          break;

        default:
          iscomplex = true;
          break;
        }

      if (iscomplex)
        cutoff = 10 * x.rows () * eps * xnorm (x, one);
    }

  if (iscomplex)
    {
      ComplexMatrix x = octave_value_extract<ComplexMatrix> (arg);

      if (mt.is_unknown ()) // if type is not known, compute it now.
        arg.matrix_type (mt = MatrixType (x));

      switch (mt.type ())
        {
        case MatrixType::Upper:
        case MatrixType::Diagonal:
          sqrtm_utri_inplace (x);
          retval = x;
          retval.matrix_type (mt);
          break;

        case MatrixType::Lower:
          x = x.transpose ();
          sqrtm_utri_inplace (x);
          retval = x.transpose ();
          retval.matrix_type (mt);
          break;

        default:
          {
            ComplexMatrix u;

            do
              {
                ComplexSCHUR schur_fact (x, "", true);
                x = schur_fact.schur_matrix ();
                u = schur_fact.unitary_matrix ();
              }
            while (0); // schur no longer needed.

            sqrtm_utri_inplace (x);

            x = u * x; // original x no longer needed.
            ComplexMatrix res = xgemm (x, u, blas_no_trans, blas_conj_trans);

            if (cutoff > 0 && xnorm (imag (res), one) <= cutoff)
              retval = real (res);
            else
              retval = res;
          }
          break;
        }
    }

  return retval;
}

DEFUN (sqrtm, args, nargout,
       "-*- texinfo -*-\n\
@deftypefn  {} {@var{s} =} sqrtm (@var{A})\n\
@deftypefnx {} {[@var{s}, @var{error_estimate}] =} sqrtm (@var{A})\n\
Compute the matrix square root of the square matrix @var{A}.\n\
\n\
Ref: @nospell{N.J. Higham}.  @cite{A New sqrtm for @sc{matlab}}.  Numerical\n\
Analysis Report No. 336, Manchester @nospell{Centre} for Computational\n\
Mathematics, Manchester, England, January 1999.\n\
@seealso{expm, logm}\n\
@end deftypefn")
{
  if (args.length () != 1)
    print_usage ();

  octave_value arg = args(0);

  octave_idx_type n = arg.rows ();
  octave_idx_type nc = arg.columns ();

  if (n != nc || arg.ndims () > 2)
    err_square_matrix_required ("sqrtm", "A");

  octave_value_list retval (nargout > 1 ? 3 : 1);

  if (nargout > 1)
    {
      // FIXME: Octave does not calculate a condition number with respect to
      //        sqrtm.  Should this return NaN instead of -1?
      retval(2) = -1.0;
    }

  if (arg.is_diag_matrix ())
    // sqrtm of a diagonal matrix is just sqrt.
    retval(0) = arg.sqrt ();
  else if (arg.is_single_type ())
    retval(0) = do_sqrtm<FloatMatrix, FloatComplexMatrix,
                         schur<FloatComplexMatrix> > (arg);
  else if (arg.is_numeric_type ())
    retval(0) = do_sqrtm<Matrix, ComplexMatrix, schur<ComplexMatrix> > (arg);

  if (nargout > 1)
    {
      // This corresponds to generic code
      //
      //   norm (s*s - x, "fro") / norm (x, "fro");

      octave_value s = retval(0);
      retval(1) = xfrobnorm (s*s - arg) / xfrobnorm (arg);
    }

  return retval;
}

/*
%!assert (sqrtm (2*ones (2)), ones (2), 3*eps)

## The following two tests are from the reference in the docstring above.
%!test
%! x = [0 1; 0 0];
%! assert (any (isnan (sqrtm (x))(:)));

%!test
%! x = eye (4);  x(2,2) = x(3,3) = 2^-26;  x(1,4) = 1;
%! z = eye (4);  z(2,2) = z(3,3) = 2^-13;  z(1,4) = 0.5;
%! [y, err] = sqrtm (x);
%! assert (y, z);
%! assert (err, 0);   # Yes, this one has to hold exactly
*/