view libinterp/corefcn/sqrtm.cc @ 31607:aac27ad79be6 stable

maint: Re-indent code after switch to using namespace macros. * build-env.h, build-env.in.cc, Cell.h, __betainc__.cc, __eigs__.cc, __ftp__.cc, __ichol__.cc, __ilu__.cc, __isprimelarge__.cc, __magick_read__.cc, __pchip_deriv__.cc, amd.cc, base-text-renderer.cc, base-text-renderer.h, besselj.cc, bitfcns.cc, bsxfun.cc, c-file-ptr-stream.h, call-stack.cc, call-stack.h, ccolamd.cc, cellfun.cc, chol.cc, colamd.cc, dasrt.cc, data.cc, debug.cc, defaults.cc, defaults.h, det.cc, display.cc, display.h, dlmread.cc, dynamic-ld.cc, dynamic-ld.h, ellipj.cc, environment.cc, environment.h, error.cc, error.h, errwarn.h, event-manager.cc, event-manager.h, event-queue.cc, event-queue.h, fcn-info.cc, fcn-info.h, fft.cc, fft2.cc, file-io.cc, filter.cc, find.cc, ft-text-renderer.cc, ft-text-renderer.h, gcd.cc, gl-render.cc, gl-render.h, gl2ps-print.cc, gl2ps-print.h, graphics-toolkit.cc, graphics-toolkit.h, graphics.cc, gsvd.cc, gtk-manager.cc, gtk-manager.h, help.cc, help.h, hook-fcn.cc, hook-fcn.h, input.cc, input.h, interpreter-private.cc, interpreter-private.h, interpreter.cc, interpreter.h, inv.cc, jsondecode.cc, jsonencode.cc, latex-text-renderer.cc, latex-text-renderer.h, load-path.cc, load-path.h, load-save.cc, load-save.h, lookup.cc, ls-hdf5.cc, ls-mat4.cc, ls-mat5.cc, lsode.cc, lu.cc, mappers.cc, matrix_type.cc, max.cc, mex.cc, mexproto.h, mxarray.h, mxtypes.in.h, oct-errno.in.cc, oct-hdf5-types.cc, oct-hist.cc, oct-hist.h, oct-map.cc, oct-map.h, oct-opengl.h, oct-prcstrm.h, oct-process.cc, oct-process.h, oct-stdstrm.h, oct-stream.cc, oct-stream.h, oct-strstrm.h, octave-default-image.h, ordqz.cc, ordschur.cc, pager.cc, pager.h, pinv.cc, pow2.cc, pr-output.cc, psi.cc, qr.cc, quadcc.cc, rand.cc, regexp.cc, settings.cc, settings.h, sighandlers.cc, sighandlers.h, sparse-xpow.cc, sqrtm.cc, stack-frame.cc, stack-frame.h, stream-euler.cc, strfns.cc, svd.cc, syminfo.cc, syminfo.h, symrcm.cc, symrec.cc, symrec.h, symscope.cc, symscope.h, symtab.cc, symtab.h, sysdep.cc, sysdep.h, text-engine.cc, text-engine.h, text-renderer.cc, text-renderer.h, time.cc, toplev.cc, typecast.cc, url-handle-manager.cc, url-handle-manager.h, urlwrite.cc, utils.cc, utils.h, variables.cc, variables.h, xdiv.cc, __delaunayn__.cc, __init_fltk__.cc, __init_gnuplot__.cc, __ode15__.cc, __voronoi__.cc, audioread.cc, convhulln.cc, gzip.cc, cdef-class.cc, cdef-class.h, cdef-fwd.h, cdef-manager.cc, cdef-manager.h, cdef-method.cc, cdef-method.h, cdef-object.cc, cdef-object.h, cdef-package.cc, cdef-package.h, cdef-property.cc, cdef-property.h, cdef-utils.cc, cdef-utils.h, ov-base-diag.cc, ov-base-int.cc, ov-base-mat.cc, ov-base-mat.h, ov-base-scalar.cc, ov-base.cc, ov-base.h, ov-bool-mat.cc, ov-bool-mat.h, ov-bool-sparse.cc, ov-bool.cc, ov-builtin.h, ov-cell.cc, ov-ch-mat.cc, ov-class.cc, ov-class.h, ov-classdef.cc, ov-classdef.h, ov-complex.cc, ov-cx-diag.cc, ov-cx-mat.cc, ov-cx-sparse.cc, ov-dld-fcn.cc, ov-dld-fcn.h, ov-fcn-handle.cc, ov-fcn-handle.h, ov-fcn.h, 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-flt-re-mat.h, ov-intx.h, ov-java.cc, ov-lazy-idx.cc, ov-legacy-range.cc, ov-magic-int.cc, ov-mex-fcn.cc, ov-mex-fcn.h, ov-null-mat.cc, ov-perm.cc, ov-range.cc, ov-re-diag.cc, ov-re-mat.cc, ov-re-mat.h, ov-re-sparse.cc, ov-scalar.cc, ov-str-mat.cc, ov-struct.cc, ov-typeinfo.cc, ov-typeinfo.h, ov-usr-fcn.cc, ov-usr-fcn.h, ov.cc, ov.h, ovl.h, octave.cc, octave.h, op-b-sbm.cc, op-bm-sbm.cc, op-cs-scm.cc, op-fm-fcm.cc, op-fs-fcm.cc, op-s-scm.cc, op-scm-cs.cc, op-scm-s.cc, op-sm-cs.cc, ops.h, anon-fcn-validator.cc, anon-fcn-validator.h, bp-table.cc, bp-table.h, comment-list.cc, comment-list.h, filepos.h, lex.h, oct-lvalue.cc, oct-lvalue.h, parse.h, profiler.cc, profiler.h, pt-anon-scopes.cc, pt-anon-scopes.h, pt-arg-list.cc, pt-arg-list.h, pt-args-block.cc, pt-args-block.h, pt-array-list.cc, pt-array-list.h, pt-assign.cc, pt-assign.h, pt-binop.cc, pt-binop.h, pt-bp.cc, pt-bp.h, pt-cbinop.cc, pt-cbinop.h, pt-cell.cc, pt-cell.h, pt-check.cc, pt-check.h, pt-classdef.cc, pt-classdef.h, pt-cmd.h, pt-colon.cc, pt-colon.h, pt-const.cc, pt-const.h, pt-decl.cc, pt-decl.h, pt-eval.cc, pt-eval.h, pt-except.cc, pt-except.h, pt-exp.cc, pt-exp.h, pt-fcn-handle.cc, pt-fcn-handle.h, pt-id.cc, pt-id.h, pt-idx.cc, pt-idx.h, pt-jump.h, pt-loop.cc, pt-loop.h, pt-mat.cc, pt-mat.h, pt-misc.cc, pt-misc.h, pt-pr-code.cc, pt-pr-code.h, pt-select.cc, pt-select.h, pt-spmd.cc, pt-spmd.h, pt-stmt.cc, pt-stmt.h, pt-tm-const.cc, pt-tm-const.h, pt-unop.cc, pt-unop.h, pt-walk.cc, pt-walk.h, pt.cc, pt.h, token.cc, token.h, Range.cc, Range.h, idx-vector.cc, idx-vector.h, range-fwd.h, CollocWt.cc, CollocWt.h, aepbalance.cc, aepbalance.h, chol.cc, chol.h, gepbalance.cc, gepbalance.h, gsvd.cc, gsvd.h, hess.cc, hess.h, lo-mappers.cc, lo-mappers.h, lo-specfun.cc, lo-specfun.h, lu.cc, lu.h, oct-convn.cc, oct-convn.h, oct-fftw.cc, oct-fftw.h, oct-norm.cc, oct-norm.h, oct-rand.cc, oct-rand.h, oct-spparms.cc, oct-spparms.h, qr.cc, qr.h, qrp.cc, qrp.h, randgamma.cc, randgamma.h, randmtzig.cc, randmtzig.h, randpoisson.cc, randpoisson.h, schur.cc, schur.h, sparse-chol.cc, sparse-chol.h, sparse-lu.cc, sparse-lu.h, sparse-qr.cc, sparse-qr.h, svd.cc, svd.h, child-list.cc, child-list.h, dir-ops.cc, dir-ops.h, file-ops.cc, file-ops.h, file-stat.cc, file-stat.h, lo-sysdep.cc, lo-sysdep.h, lo-sysinfo.cc, lo-sysinfo.h, mach-info.cc, mach-info.h, oct-env.cc, oct-env.h, oct-group.cc, oct-group.h, oct-password.cc, oct-password.h, oct-syscalls.cc, oct-syscalls.h, oct-time.cc, oct-time.h, oct-uname.cc, oct-uname.h, action-container.cc, action-container.h, base-list.h, cmd-edit.cc, cmd-edit.h, cmd-hist.cc, cmd-hist.h, f77-fcn.h, file-info.cc, file-info.h, lo-array-errwarn.cc, lo-array-errwarn.h, lo-hash.cc, lo-hash.h, lo-ieee.h, lo-regexp.cc, lo-regexp.h, lo-utils.cc, lo-utils.h, oct-base64.cc, oct-base64.h, oct-glob.cc, oct-glob.h, oct-inttypes.h, oct-mutex.cc, oct-mutex.h, oct-refcount.h, oct-shlib.cc, oct-shlib.h, oct-sparse.cc, oct-sparse.h, oct-string.h, octave-preserve-stream-state.h, pathsearch.cc, pathsearch.h, quit.cc, quit.h, unwind-prot.cc, unwind-prot.h, url-transfer.cc, url-transfer.h: Re-indent code after switch to using namespace macros.
author Rik <rik@octave.org>
date Thu, 01 Dec 2022 18:02:15 -0800
parents e88a07dec498
children 597f3ee61a48
line wrap: on
line source

////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2001-2022 The Octave Project Developers
//
// See the file COPYRIGHT.md in the top-level directory of this
// distribution or <https://octave.org/copyright/>.
//
// 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
// <https://www.gnu.org/licenses/>.
//
////////////////////////////////////////////////////////////////////////

#if defined (HAVE_CONFIG_H)
#  include "config.h"
#endif

#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"

OCTAVE_BEGIN_NAMESPACE(octave)

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
  //       if T(i,j) != 0
  //         T(i,j) /= (T(i,i) + T(j,j));
  //       endif
  //       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;
          if (colj[i] != zero)
            colj[i] /= (coli[i] + colj[j]);
          const element_type colji = colj[i];
          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.iscomplex ();

  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_schur_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,
       doc: /* -*- texinfo -*-
@deftypefn  {} {@var{s} =} sqrtm (@var{A})
@deftypefnx {} {[@var{s}, @var{error_estimate}] =} sqrtm (@var{A})
Compute the matrix square root of the square matrix @var{A}.

Ref: @nospell{N.J. Higham}.  @cite{A New sqrtm for @sc{matlab}}.  Numerical
Analysis Report No.@: 336, Manchester @nospell{Centre} for Computational
Mathematics, Manchester, England, January 1999.
@seealso{expm, logm}
@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,
    math::schur<FloatComplexMatrix>> (arg);
  else if (arg.isnumeric ())
    retval(0) = do_sqrtm<Matrix, ComplexMatrix,
    math::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)
%!assert <*60797> (sqrtm (ones (4))^2, ones (4), 5*eps)

## The following two tests are from the reference in the docstring above.
%!test
%! warning ("off", "Octave:sqrtm:SingularMatrix", "local");
%! 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
*/

OCTAVE_END_NAMESPACE(octave)