view libinterp/corefcn/pow2.cc @ 31605:e88a07dec498 stable

maint: Use macros to begin/end C++ namespaces. * oct-conf-post-public.in.h: Define two macros (OCTAVE_BEGIN_NAMESPACE, OCTAVE_END_NAMESPACE) that can be used to start/end a namespace. * mk-opts.pl, build-env.h, build-env.in.cc, __betainc__.cc, __contourc__.cc, __dsearchn__.cc, __eigs__.cc, __expint__.cc, __ftp__.cc, __gammainc__.cc, __ichol__.cc, __ilu__.cc, __isprimelarge__.cc, __lin_interpn__.cc, __magick_read__.cc, __pchip_deriv__.cc, __qp__.cc, amd.cc, auto-shlib.cc, auto-shlib.h, balance.cc, base-text-renderer.cc, base-text-renderer.h, besselj.cc, bitfcns.cc, bsxfun.cc, c-file-ptr-stream.cc, c-file-ptr-stream.h, call-stack.cc, call-stack.h, ccolamd.cc, cellfun.cc, chol.cc, colamd.cc, colloc.cc, conv2.cc, daspk.cc, dasrt.cc, dassl.cc, data.cc, data.h, debug.cc, defaults.cc, defaults.h, defun-int.h, defun.cc, det.cc, dirfns.cc, display.cc, display.h, dlmread.cc, dmperm.cc, dot.cc, dynamic-ld.cc, dynamic-ld.h, eig.cc, 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, fftn.cc, file-io.cc, filter.cc, find.cc, ft-text-renderer.cc, ft-text-renderer.h, gcd.cc, getgrent.cc, getpwent.cc, getrusage.cc, givens.cc, gl-render.cc, gl-render.h, gl2ps-print.cc, gl2ps-print.h, graphics-toolkit.cc, graphics-toolkit.h, graphics.cc, graphics.in.h, gsvd.cc, gtk-manager.cc, gtk-manager.h, hash.cc, help.cc, help.h, hess.cc, hex2num.cc, 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, kron.cc, latex-text-renderer.cc, latex-text-renderer.h, load-path.cc, load-path.h, load-save.cc, load-save.h, lookup.cc, ls-ascii-helper.cc, ls-ascii-helper.h, ls-oct-text.cc, ls-utils.cc, ls-utils.h, lsode.cc, lu.cc, mappers.cc, matrix_type.cc, max.cc, mex-private.h, mex.cc, mgorth.cc, nproc.cc, oct-fstrm.cc, oct-fstrm.h, oct-hdf5-types.cc, oct-hdf5-types.h, oct-hist.cc, oct-hist.h, oct-iostrm.cc, oct-iostrm.h, oct-opengl.h, oct-prcstrm.cc, oct-prcstrm.h, oct-procbuf.cc, oct-procbuf.h, oct-process.cc, oct-process.h, oct-stdstrm.h, oct-stream.cc, oct-stream.h, oct-strstrm.cc, oct-strstrm.h, oct-tex-lexer.in.ll, oct-tex-parser.yy, ordqz.cc, ordschur.cc, pager.cc, pager.h, pinv.cc, pow2.cc, pr-flt-fmt.cc, pr-output.cc, procstream.cc, procstream.h, psi.cc, qr.cc, quad.cc, quadcc.cc, qz.cc, rand.cc, rcond.cc, regexp.cc, schur.cc, settings.cc, settings.h, sighandlers.cc, sighandlers.h, sparse-xdiv.cc, sparse-xdiv.h, sparse-xpow.cc, sparse-xpow.h, sparse.cc, spparms.cc, sqrtm.cc, stack-frame.cc, stack-frame.h, stream-euler.cc, strfind.cc, strfns.cc, sub2ind.cc, svd.cc, sylvester.cc, symbfact.cc, syminfo.cc, syminfo.h, symrcm.cc, symrec.cc, symrec.h, symscope.cc, symscope.h, symtab.cc, symtab.h, syscalls.cc, sysdep.cc, sysdep.h, text-engine.cc, text-engine.h, text-renderer.cc, text-renderer.h, time.cc, toplev.cc, tril.cc, tsearch.cc, typecast.cc, url-handle-manager.cc, url-handle-manager.h, urlwrite.cc, utils.cc, utils.h, variables.cc, variables.h, xdiv.cc, xdiv.h, xnorm.cc, xnorm.h, xpow.cc, xpow.h, __delaunayn__.cc, __fltk_uigetfile__.cc, __glpk__.cc, __init_fltk__.cc, __init_gnuplot__.cc, __ode15__.cc, __voronoi__.cc, audiodevinfo.cc, audioread.cc, convhulln.cc, fftw.cc, gzip.cc, mk-build-env-features.sh, mk-builtins.pl, 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.cc, ov-base.h, ov-bool-mat.cc, ov-builtin.h, ov-cell.cc, ov-class.cc, ov-class.h, ov-classdef.cc, ov-classdef.h, ov-complex.cc, ov-fcn-handle.cc, ov-fcn-handle.h, ov-fcn.h, ov-java.cc, ov-java.h, ov-mex-fcn.h, ov-null-mat.cc, ov-oncleanup.cc, ov-struct.cc, ov-typeinfo.cc, ov-typeinfo.h, ov-usr-fcn.cc, ov-usr-fcn.h, ov.cc, ov.h, octave.cc, octave.h, mk-ops.sh, 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-fcdm-fcdm.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-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-m-cm.cc, op-m-cs.cc, op-m-m.cc, op-m-s.cc, op-m-scm.cc, op-m-sm.cc, op-mi.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, 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, lex.ll, oct-lvalue.cc, oct-lvalue.h, oct-parse.yy, 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-vm-eval.cc, 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 : Use new macros to begin/end C++ namespaces.
author Rik <rik@octave.org>
date Thu, 01 Dec 2022 14:23:45 -0800
parents a885bdb22374
children aac27ad79be6
line wrap: on
line source

////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 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 <cmath>

#include "lo-array-errwarn.h"

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

// FIXME: According to cppreference.com the implementation of `ldexp (f, e)`
// might be less efficient that the corresponding `f * exp2 (e)`.  Consider
// replacing our implementation with the latter.

template <typename T>
void
map_2_xldexp (Array<T>& y, const Array<T>& f, const Array<T>& e)
{
  if (f.numel () == e.numel () || e.numel () == 1)
    y = Array<T> (f.dims ());
  else if (f.numel () == 1)
    y = Array<T> (e.dims ());
  else
    octave::err_nonconformant ("pow2", f.dims (), e.dims ());

  octave_idx_type f_inc = (f.numel () == 1) ? 0 : 1;
  octave_idx_type e_inc = (e.numel () == 1) ? 0 : 1;

  for (octave_idx_type i = 0; i < y.numel (); i++)
    y.xelem (i) = std::ldexp (f.xelem (i * f_inc),
                              static_cast<int> (e.xelem (i * e_inc)));
}

void
map_2_xldexp_sparse (SparseMatrix& y, const SparseMatrix& f,
                                      const SparseMatrix& e)
{
  if (e.numel () == 1)
    {
      int ee = static_cast<int> (e.data (0));
      for (octave_idx_type i = 0; i < y.nnz (); i++)
        y.data (i) = std::ldexp (f.data (i), ee);
    }
  else if (f.numel () == e.numel ())
    {
      octave_idx_type col = 1;
      for (octave_idx_type i = 0; i < y.nnz (); i++)
        {
          // Determine current column.
          while (i >= f.cidx (col))
            col++;
          int ee = static_cast<int> (e.xelem (f.ridx (i), col - 1));
          y.data (i) = std::ldexp (f.data (i), ee);
        }
    }
  else
    octave::err_nonconformant ("pow2", f.dims (), e.dims ());
}

OCTAVE_BEGIN_NAMESPACE(octave)

DEFUN (pow2, args, ,
       doc: /* -*- texinfo -*-
@deftypefn  {} {@var{y} =} pow2 (@var{x})
@deftypefnx {} {@var{y} =} pow2 (@var{f}, @var{e})
With one input argument, compute
@tex
$y = 2^x$
@end tex
@ifnottex
y = 2 .^ x
@end ifnottex
for each element of @var{x}.

With two input arguments, return
@tex
$y = f \cdot 2^e$,
@end tex
@ifnottex
y = f .* (2 .^ e).
@end ifnottex
where for complex inputs only the real part of both inputs is regarded
and from @var{e} only the real integer part.  This calling form corresponds
to C/C++ standard function @code{ldexp()}.
@seealso{log2, nextpow2, power}
@end deftypefn */)
{
  if (args.length () < 1 || args.length () > 2)
    print_usage ();

  if (! args(0).isfloat ())
    err_wrong_type_arg ("pow2", args(0));

  // Call exp2(f) where possible for numerical more accurate results.
  if (args.length () == 1)
    {
      if (args(0).iscomplex ())
        {
          // The C++ standard does not define exp2 for complex arguments.
          // Therefore call `2.^x`.
          octave_value retval = octave::binary_op (octave_value::op_el_pow,
                                                   2, args(0));

          // Preserve sparse datatype, but even for sparse input fill-up
          // is unavoidable `2^0 == 1` thus cast only.
          if (args(0).issparse ())
            retval = octave_value (retval.sparse_complex_matrix_value ());

          return ovl (retval);
        }
      else if (args(0).is_single_type ())
        {
          FloatNDArray x = args(0).float_array_value ();
          FloatNDArray y (x.dims ());
          for (octave_idx_type i = 0; i < y.numel (); i++)
            y.xelem (i) = std::exp2 (x.xelem (i));
          return ovl (y);
        }
      else
        {
          NDArray x = args(0).array_value ();
          NDArray y (x.dims ());
          for (octave_idx_type i = 0; i < y.numel (); i++)
            y.xelem (i) = std::exp2 (x.xelem (i));

          // Preserve sparse datatype, but even for sparse input fill-up
          // is unavoidable `2^0 == 1` thus cast only.
          if (args(0).issparse ())
            return ovl (SparseMatrix (y));
          else
            return ovl (y);
        }
    }

  // For Matlab compatibility, the two argument call `y = pow2 (f, e)`
  // corresponds to std::ldexp() (see bug #61968).  The resulting y is
  // computed quickly by adding the integer part of e to the floating-point
  // exponent of f.

  if (! args(1).isfloat ())
    err_wrong_type_arg ("pow2", args(1));

  if (args(0).iscomplex () || args(1).iscomplex ())
    warning_with_id ("Octave:pow2:imaginary-ignored",
                     "pow2: imaginary part is ignored");

  // Note: Matlab R2021a errors on `pow2 (sparse (f), single (e))`,
  //       but sparsity in f determines output and can significantly
  //       reduce computation, e.g. `N=1e5; pow2(speye(N),sparse(N,N))`.
  if (args(0).issparse ())
    {
      SparseMatrix f = args(0).sparse_matrix_value ();

      // Special case: return a sparse zero matrix in size of e.
      if ((f.numel () == 1) && (f.nnz () == 0))
        return ovl (SparseMatrix (args(1).rows (), args(1).columns ()));

      // Only do sparse computation, if it pays off.  For scalar f fill-up
      // is unavoidable even for sparse e because `f * 2^0 == f`.  Use dense
      // code below in this case.
      if (f.numel () > 1)
        {
          SparseMatrix e = args(1).sparse_matrix_value ();
          SparseMatrix y = SparseMatrix (f);
          map_2_xldexp_sparse (y, f, e);
          return ovl (y);
        }
    }

  if (args(0).is_single_type () || args(1).is_single_type ())
    {
      FloatNDArray f = args(0).float_array_value ();
      FloatNDArray e = args(1).float_array_value ();
      FloatNDArray y;
      map_2_xldexp (y, f, e);
      return ovl (y);
    }
  else
    {
      NDArray f = args(0).array_value ();
      NDArray e = args(1).array_value ();
      NDArray y;
      map_2_xldexp (y, f, e);

      // Preserve sparse datatype.
      // Cases for efficient use of sparsity were treated above already.
      if (args(0).issparse ())
        return ovl (SparseMatrix (y));
      else
        return ovl (y);
    }
}

/*
## Call `y = pow2 (x)`

%!test
%! fcns = {@double, @single, @complex};
%! x = [3, 0, -3];
%! v = [8, 1, .125];
%! for i = 1:numel (fcns)
%!   fcn = fcns{i};
%!   assert (pow2 (fcn (x)), fcn (v), sqrt (eps));
%! endfor

%!test
%! fcns = {@double, @single, @complex, @sparse};
%! x = [3, 1, -3];
%! v = [8, 2, .125];
%! for i = 1:numel (fcns)
%!   fcn = fcns{i};
%!   assert (pow2 (fcn (x)), fcn (v), sqrt (eps));
%! endfor

%!test
%! fcns = {@double, @single, @complex, @sparse};
%! x = [1, 1+1i, 1i];
%! for i = 1:numel (fcns)
%!   fcn = fcns{i};
%!   assert (pow2 (fcn (x)), fcn (2) .^ fcn (x), sqrt (eps));
%! endfor

## Call `y = pow2 (f, e)`

%!test
%! fcns = {@double, @single, @complex, @sparse};
%! f = [2 2];
%! e = [2 2];
%! z = [8 8];
%! warning ("off", "Octave:pow2:imaginary-ignored", "local");
%! for i = 1:numel (fcns)
%!   fcn = fcns{i};
%!   assert (pow2 (fcn (f), fcn (e)), real (fcn (z)));
%! endfor

## Only integer part is taken into account.
%!test
%! f = 2;
%! e = [2, 2.1, 2.2, 2.4, 2.5, 2.8];
%! z = 8 .* ones (1, length (e));
%! assert (pow2 (f, e), z);

## Only real part is taken into account.
%!test
%! f = [1+1i, 1];
%! e = 2;
%! z = [4, 4];
%! warning ("off", "Octave:pow2:imaginary-ignored", "local");
%! assert (pow2 (f, e), z);

%!test
%! f = 1;
%! e = [1+1i, 1];
%! z = [2, 2];
%! warning ("off", "Octave:pow2:imaginary-ignored", "local");
%! assert (pow2 (f, e), z);

%!test
%! f = [1/2, pi/4, -3/4, 1/2, 1-eps()/2, 1/2];
%! e = [1, 2, 2, -51, 1024, -1021];
%! z = [1, pi, -3, eps(), realmax(), realmin()];
%! assert (pow2 (f, e), z);

## Tests for sparsity.
%!assert (pow2 (sparse (0), ones  (3)), sparse (3, 3));
%!assert (pow2 (sparse (1), ones  (3)), 2 .* sparse (ones (3)));
%!assert (pow2 (sparse (1), speye (3)), sparse (ones (3) + eye (3)));
%!assert (pow2 (sparse (3, 3), ones (3)), sparse (3, 3));
%!assert (pow2 (speye (3), ones (3)), 2 .* speye (3));
%!assert (pow2 (speye (3), 1),        2 .* speye (3));

%!test
%! f = speye (3);
%! e = sparse (3, 3);
%! e(1,1) = 1;
%! e(1,3) = 1;
%! z = f;
%! z(1,1) = 2;
%! assert (pow2 (f, e), z);

## Large sparse matrix (only few real elements).
%!test
%! ## FIXME: `N = 1e5` would be a better test, but `assert` fills-up somehow.
%! N = 1e3;
%! assert (pow2 (speye  (N), sparse (N,N)), speye (N));
%! assert (pow2 (sparse (0), speye  (N)),   sparse(N,N));

%!error <Invalid call> pow2 ()
%!error <Invalid call> pow2 (1,2,3)
%!error <wrong type argument> pow2 (int8 (1))
%!error <wrong type argument> pow2 (2, int8 (1))
%!warning <imaginary part is ignored> pow2 (i, 2);
%!warning <imaginary part is ignored> pow2 (2, i);
%!error <pow2: nonconformant arguments> pow2 ([1,2], [3,4,5])
%!error <pow2: nonconformant arguments> pow2 (sparse ([1,2]), sparse ([3,4,5]))
*/

OCTAVE_END_NAMESPACE(octave)