view libinterp/corefcn/profiler.cc @ 17289:bc924baa2c4e

doc: Add new @qcode macro for code samples which are quoted. Macro handles options ("on") or properties ("position") more elegantly than @code{"text"}. * doc/interpreter/macros.texi: Add new @qcode macro. * doc/interpreter/tips.txi: Add documentation about @qcode macro. * doc/interpreter/basics.txi, doc/interpreter/container.txi, doc/interpreter/emacs.txi, doc/interpreter/errors.txi, doc/interpreter/eval.txi, doc/interpreter/expr.txi, doc/interpreter/external.txi, doc/interpreter/func.txi, doc/interpreter/grammar.txi, doc/interpreter/image.txi, doc/interpreter/install.txi, doc/interpreter/interp.txi, doc/interpreter/io.txi, doc/interpreter/matrix.txi, doc/interpreter/numbers.txi, doc/interpreter/oop.txi, doc/interpreter/package.txi, doc/interpreter/plot.txi, doc/interpreter/quad.txi, doc/interpreter/sparse.txi, doc/interpreter/strings.txi, doc/interpreter/system.txi, doc/interpreter/vectorize.txi, libinterp/corefcn/balance.cc, libinterp/corefcn/bitfcns.cc, libinterp/corefcn/cellfun.cc, libinterp/corefcn/conv2.cc, libinterp/corefcn/data.cc, libinterp/corefcn/debug.cc, libinterp/corefcn/defaults.cc, libinterp/corefcn/dirfns.cc, libinterp/corefcn/dlmread.cc, libinterp/corefcn/error.cc, libinterp/corefcn/file-io.cc, libinterp/corefcn/find.cc, libinterp/corefcn/gammainc.cc, libinterp/corefcn/graphics.cc, libinterp/corefcn/help.cc, libinterp/corefcn/hex2num.cc, libinterp/corefcn/input.cc, libinterp/corefcn/load-path.cc, libinterp/corefcn/load-save.cc, libinterp/corefcn/ls-oct-ascii.cc, libinterp/corefcn/lu.cc, libinterp/corefcn/luinc.cc, libinterp/corefcn/matrix_type.cc, libinterp/corefcn/oct-hist.cc, libinterp/corefcn/pager.cc, libinterp/corefcn/pr-output.cc, libinterp/corefcn/pt-jit.cc, libinterp/corefcn/qz.cc, libinterp/corefcn/rand.cc, libinterp/corefcn/regexp.cc, libinterp/corefcn/schur.cc, libinterp/corefcn/sighandlers.cc, libinterp/corefcn/sparse.cc, libinterp/corefcn/spparms.cc, libinterp/corefcn/str2double.cc, libinterp/corefcn/svd.cc, libinterp/corefcn/symtab.cc, libinterp/corefcn/syscalls.cc, libinterp/corefcn/toplev.cc, libinterp/corefcn/tril.cc, libinterp/corefcn/typecast.cc, libinterp/corefcn/utils.cc, libinterp/corefcn/variables.cc, libinterp/dldfcn/__init_fltk__.cc, libinterp/dldfcn/chol.cc, libinterp/dldfcn/colamd.cc, libinterp/dldfcn/fftw.cc, libinterp/dldfcn/qr.cc, libinterp/dldfcn/symbfact.cc, libinterp/octave-value/ov-base.cc, libinterp/octave-value/ov-fcn-handle.cc, libinterp/octave-value/ov-fcn-inline.cc, libinterp/octave-value/ov-java.cc, libinterp/octave-value/ov-range.cc, libinterp/octave-value/ov-struct.cc, libinterp/octave-value/ov-usr-fcn.cc, libinterp/parse-tree/oct-parse.in.yy, libinterp/parse-tree/pt-binop.cc, libinterp/parse-tree/pt-eval.cc, libinterp/parse-tree/pt-mat.cc, scripts/@ftp/ftp.m, scripts/deprecated/java_convert_matrix.m, scripts/deprecated/java_debug.m, scripts/deprecated/java_unsigned_conversion.m, scripts/deprecated/shell_cmd.m, scripts/general/dblquad.m, scripts/general/display.m, scripts/general/genvarname.m, scripts/general/idivide.m, scripts/general/interp1.m, scripts/general/interp2.m, scripts/general/interp3.m, scripts/general/interpn.m, scripts/general/isa.m, scripts/general/profexplore.m, scripts/general/profile.m, scripts/general/quadgk.m, scripts/general/randi.m, scripts/general/structfun.m, scripts/general/subsindex.m, scripts/general/triplequad.m, scripts/geometry/griddata.m, scripts/geometry/griddata3.m, scripts/geometry/griddatan.m, scripts/geometry/voronoi.m, scripts/help/help.m, scripts/help/lookfor.m, scripts/image/cmpermute.m, scripts/image/colormap.m, scripts/image/image.m, scripts/image/imagesc.m, scripts/image/imfinfo.m, scripts/image/imformats.m, scripts/image/imread.m, scripts/image/imshow.m, scripts/image/imwrite.m, scripts/image/ind2gray.m, scripts/image/lines.m, scripts/image/rgb2ind.m, scripts/image/spinmap.m, scripts/io/dlmwrite.m, scripts/io/strread.m, scripts/io/textread.m, scripts/io/textscan.m, scripts/java/javaclasspath.m, scripts/java/usejava.m, scripts/miscellaneous/bzip2.m, scripts/miscellaneous/computer.m, scripts/miscellaneous/copyfile.m, scripts/miscellaneous/debug.m, scripts/miscellaneous/dos.m, scripts/miscellaneous/edit.m, scripts/miscellaneous/gzip.m, scripts/miscellaneous/license.m, scripts/miscellaneous/mkoctfile.m, scripts/miscellaneous/movefile.m, scripts/miscellaneous/parseparams.m, scripts/miscellaneous/unix.m, scripts/optimization/fminbnd.m, scripts/optimization/fminsearch.m, scripts/optimization/fminunc.m, scripts/optimization/fsolve.m, scripts/optimization/fzero.m, scripts/optimization/glpk.m, scripts/optimization/lsqnonneg.m, scripts/optimization/optimset.m, scripts/optimization/pqpnonneg.m, scripts/pkg/pkg.m, scripts/plot/allchild.m, scripts/plot/ancestor.m, scripts/plot/area.m, scripts/plot/axis.m, scripts/plot/bar.m, scripts/plot/barh.m, scripts/plot/box.m, scripts/plot/caxis.m, scripts/plot/cla.m, scripts/plot/clabel.m, scripts/plot/clf.m, scripts/plot/close.m, scripts/plot/colorbar.m, scripts/plot/daspect.m, scripts/plot/ezmesh.m, scripts/plot/ezmeshc.m, scripts/plot/ezsurf.m, scripts/plot/ezsurfc.m, scripts/plot/findall.m, scripts/plot/findobj.m, scripts/plot/gcbo.m, scripts/plot/gcf.m, scripts/plot/gco.m, scripts/plot/grid.m, scripts/plot/guihandles.m, scripts/plot/hdl2struct.m, scripts/plot/hidden.m, scripts/plot/hold.m, scripts/plot/isonormals.m, scripts/plot/isosurface.m, scripts/plot/legend.m, scripts/plot/mesh.m, scripts/plot/meshc.m, scripts/plot/meshz.m, scripts/plot/newplot.m, scripts/plot/orient.m, scripts/plot/pareto.m, scripts/plot/patch.m, scripts/plot/pbaspect.m, scripts/plot/pcolor.m, scripts/plot/plot.m, scripts/plot/print.m, scripts/plot/private/__add_default_menu__.m, scripts/plot/quiver.m, scripts/plot/quiver3.m, scripts/plot/refreshdata.m, scripts/plot/saveas.m, scripts/plot/scatter.m, scripts/plot/scatter3.m, scripts/plot/shading.m, scripts/plot/shrinkfaces.m, scripts/plot/slice.m, scripts/plot/stem.m, scripts/plot/stem3.m, scripts/plot/struct2hdl.m, scripts/plot/subplot.m, scripts/plot/surf.m, scripts/plot/surfc.m, scripts/plot/surfl.m, scripts/plot/tetramesh.m, scripts/plot/uigetfile.m, scripts/plot/uimenu.m, scripts/plot/uiputfile.m, scripts/plot/waterfall.m, scripts/plot/whitebg.m, scripts/plot/xlim.m, scripts/plot/ylim.m, scripts/plot/zlim.m, scripts/polynomial/conv.m, scripts/polynomial/polyout.m, scripts/polynomial/splinefit.m, scripts/set/ismember.m, scripts/set/powerset.m, scripts/set/setdiff.m, scripts/set/union.m, scripts/set/unique.m, scripts/signal/detrend.m, scripts/signal/filter2.m, scripts/signal/freqz.m, scripts/signal/periodogram.m, scripts/signal/spectral_adf.m, scripts/signal/spectral_xdf.m, scripts/sparse/eigs.m, scripts/sparse/svds.m, scripts/specfun/legendre.m, scripts/special-matrix/gallery.m, scripts/statistics/base/mean.m, scripts/statistics/base/moment.m, scripts/statistics/tests/cor_test.m, scripts/statistics/tests/kolmogorov_smirnov_test.m, scripts/statistics/tests/kolmogorov_smirnov_test_2.m, scripts/statistics/tests/kruskal_wallis_test.m, scripts/statistics/tests/prop_test_2.m, scripts/statistics/tests/sign_test.m, scripts/statistics/tests/t_test.m, scripts/statistics/tests/t_test_2.m, scripts/statistics/tests/t_test_regression.m, scripts/statistics/tests/u_test.m, scripts/statistics/tests/var_test.m, scripts/statistics/tests/welch_test.m, scripts/statistics/tests/wilcoxon_test.m, scripts/statistics/tests/z_test.m, scripts/statistics/tests/z_test_2.m, scripts/strings/base2dec.m, scripts/strings/index.m, scripts/strings/isstrprop.m, scripts/strings/mat2str.m, scripts/strings/regexptranslate.m, scripts/strings/rindex.m, scripts/strings/str2num.m, scripts/strings/strcat.m, scripts/strings/strjust.m, scripts/strings/strmatch.m, scripts/strings/validatestring.m, scripts/testfun/demo.m, scripts/testfun/example.m, scripts/testfun/test.m, scripts/time/addtodate.m, scripts/time/asctime.m, scripts/time/datestr.m, scripts/time/datetick.m, scripts/time/weekday.m, scripts/ui/errordlg.m, scripts/ui/helpdlg.m, scripts/ui/inputdlg.m, scripts/ui/listdlg.m, scripts/ui/msgbox.m, scripts/ui/questdlg.m, scripts/ui/warndlg.m: Use new @qcode macro.
author Rik <rik@octave.org>
date Mon, 19 Aug 2013 20:46:38 -0700
parents 68fc671a9339
children
line wrap: on
line source

/*

Copyright (C) 2012 Daniel Kraft

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 <iostream>

#include "defun.h"
#include "oct-time.h"
#include "ov-struct.h"
#include "pager.h"
#include "profiler.h"

profile_data_accumulator::enter::enter (profile_data_accumulator& a,
                                        const std::string& f)
  : acc (a)
{
  if (acc.is_active ())
    {
      fcn = f;
      acc.enter_function (fcn);
    }
  else
    fcn = "";
}

profile_data_accumulator::enter::~enter ()
{
  if (fcn != "")
    acc.exit_function (fcn);
}

profile_data_accumulator::stats::stats ()
  : time (0.0), calls (0), recursive (false),
    parents (), children ()
{}

octave_value
profile_data_accumulator::stats::function_set_value (const function_set& list)
{
  const octave_idx_type n = list.size ();

  RowVector retval (n);
  octave_idx_type i = 0;
  for (function_set::const_iterator p = list.begin (); p != list.end (); ++p)
    {
      retval(i) = *p;
      ++i;
    }
  assert (i == n);

  return retval;
}

profile_data_accumulator::tree_node::tree_node (tree_node* p, octave_idx_type f)
  : parent (p), fcn_id (f), children (), time (0.0), calls (0)
{}

profile_data_accumulator::tree_node::~tree_node ()
{
  for (child_map::iterator i = children.begin (); i != children.end (); ++i)
    delete i->second;
}

profile_data_accumulator::tree_node*
profile_data_accumulator::tree_node::enter (octave_idx_type fcn)
{
  tree_node* retval;

  child_map::iterator pos = children.find (fcn);
  if (pos == children.end ())
    {
      retval = new tree_node (this, fcn);
      children[fcn] = retval;
    }
  else
    retval = pos->second;

  ++retval->calls;
  return retval;
}

profile_data_accumulator::tree_node*
profile_data_accumulator::tree_node::exit (octave_idx_type fcn)
{
  assert (parent);
  assert (fcn_id == fcn);

  return parent;
}

void
profile_data_accumulator::tree_node::build_flat (flat_profile& data) const
{
  // If this is not the top-level node, update profile entry for this function.
  if (fcn_id != 0)
    {
      stats& entry = data[fcn_id - 1];

      entry.time += time;
      entry.calls += calls;

      assert (parent);
      if (parent->fcn_id != 0)
        {
          entry.parents.insert (parent->fcn_id);
          data[parent->fcn_id - 1].children.insert (fcn_id);
        }

      if (!entry.recursive)
        for (const tree_node* i = parent; i; i = i->parent)
          if (i->fcn_id == fcn_id)
            {
              entry.recursive = true;
              break;
            }
    }

  // Recurse on children.
  for (child_map::const_iterator i = children.begin ();
       i != children.end (); ++i)
    i->second->build_flat (data);
}

octave_value
profile_data_accumulator::tree_node::get_hierarchical (double* total) const
{
  /* Note that we don't generate the entry just for this node, but rather
     a struct-array with entries for all children.  This way, the top-node
     (for which we don't want a real entry) generates already the final
     hierarchical profile data.  */

  const octave_idx_type n = children.size ();

  Cell rv_indices (n, 1);
  Cell rv_times (n, 1);
  Cell rv_totals (n, 1);
  Cell rv_calls (n, 1);
  Cell rv_children (n, 1);

  octave_idx_type i = 0;
  for (child_map::const_iterator p = children.begin ();
       p != children.end (); ++p)
    {
      const tree_node& entry = *p->second;
      double child_total = entry.time;

      rv_indices(i) = octave_value (p->first);
      rv_times(i) = octave_value (entry.time);
      rv_calls(i) = octave_value (entry.calls);
      rv_children(i) = entry.get_hierarchical (&child_total);
      rv_totals(i) = octave_value (child_total);

      if (total)
        *total += child_total;

      ++i;
    }
  assert (i == n);

  octave_map retval;

  retval.assign ("Index", rv_indices);
  retval.assign ("SelfTime", rv_times);
  retval.assign ("TotalTime", rv_totals);
  retval.assign ("NumCalls", rv_calls);
  retval.assign ("Children", rv_children);

  return retval;
}

profile_data_accumulator::profile_data_accumulator ()
  : known_functions (), fcn_index (),
    enabled (false), call_tree (NULL), last_time (-1.0)
{}

profile_data_accumulator::~profile_data_accumulator ()
{
  if (call_tree)
    delete call_tree;
}

void
profile_data_accumulator::set_active (bool value)
{
  if (value)
    {
      // Create a call-tree top-node if there isn't yet one.
      if (!call_tree)
        call_tree = new tree_node (NULL, 0);

      // Let the top-node be the active one.  This ensures we have a clean
      // fresh start collecting times.
      active_fcn = call_tree;
    }
  else
    {
      // Make sure we start with fresh timing if we're re-enabled later.
      last_time = -1.0;
    }

  enabled = value;
}

void
profile_data_accumulator::enter_function (const std::string& fcn)
{
  // The enter class will check and only call us if the profiler is active.
  assert (is_active ());
  assert (call_tree);

  // If there is already an active function, add to its time before
  // pushing the new one.
  if (active_fcn != call_tree)
    add_current_time ();

  // Map the function's name to its index.
  octave_idx_type fcn_idx;
  fcn_index_map::iterator pos = fcn_index.find (fcn);
  if (pos == fcn_index.end ())
    {
      known_functions.push_back (fcn);
      fcn_idx = known_functions.size ();
      fcn_index[fcn] = fcn_idx;
    }
  else
    fcn_idx = pos->second;

  active_fcn = active_fcn->enter (fcn_idx);
  last_time = query_time ();
}

void
profile_data_accumulator::exit_function (const std::string& fcn)
{
  assert (call_tree);
  assert (active_fcn != call_tree);

  // Usually, if we are disabled this function is not even called.  But the
  // call disabling the profiler is an exception.  So also check here
  // and only record the time if enabled.
  if (is_active ())
    add_current_time ();

  fcn_index_map::iterator pos = fcn_index.find (fcn);
  assert (pos != fcn_index.end ());
  active_fcn = active_fcn->exit (pos->second);

  // If this was an "inner call", we resume executing the parent function
  // up the stack.  So note the start-time for this!
  last_time = query_time ();
}

void
profile_data_accumulator::reset (void)
{
  if (is_active ())
    {
      error ("Can't reset active profiler.");
      return;
    }

  known_functions.clear ();
  fcn_index.clear ();

  if (call_tree)
    {
      delete call_tree;
      call_tree = NULL;
    }

  last_time = -1.0;
}

octave_value
profile_data_accumulator::get_flat (void) const
{
  octave_value retval;

  const octave_idx_type n = known_functions.size ();

  flat_profile flat (n);

  if (call_tree)
    {
      call_tree->build_flat (flat);

      Cell rv_names (n, 1);
      Cell rv_times (n, 1);
      Cell rv_calls (n, 1);
      Cell rv_recursive (n, 1);
      Cell rv_parents (n, 1);
      Cell rv_children (n, 1);

      for (octave_idx_type i = 0; i != n; ++i)
        {
          rv_names(i) = octave_value (known_functions[i]);
          rv_times(i) = octave_value (flat[i].time);
          rv_calls(i) = octave_value (flat[i].calls);
          rv_recursive(i) = octave_value (flat[i].recursive);
          rv_parents(i) = stats::function_set_value (flat[i].parents);
          rv_children(i) = stats::function_set_value (flat[i].children);
        }

      octave_map m;

      m.assign ("FunctionName", rv_names);
      m.assign ("TotalTime", rv_times);
      m.assign ("NumCalls", rv_calls);
      m.assign ("IsRecursive", rv_recursive);
      m.assign ("Parents", rv_parents);
      m.assign ("Children", rv_children);

      retval = m;
    }
  else
    {
      static const char *fn[] =
        {
          "FunctionName",
          "TotalTime",
          "NumCalls",
          "IsRecursive",
          "Parents",
          "Children",
          0
        };

      static octave_map m (dim_vector (0, 1), string_vector (fn));

      retval = m;
    }

  return retval;
}

octave_value
profile_data_accumulator::get_hierarchical (void) const
{
  octave_value retval;

  if (call_tree)
    retval = call_tree->get_hierarchical ();
  else
    {
      static const char *fn[] =
        {
          "Index",
          "SelfTime",
          "NumCalls",
          "Children",
          0
        };

      static octave_map m (dim_vector (0, 1), string_vector (fn));

      retval = m;
    }

  return retval;
}

double
profile_data_accumulator::query_time (void) const
{
  octave_time now;

  // FIXME -- is this volatile declaration really needed?
  // See bug #34210 for additional details.
  volatile double dnow = now.double_value ();

  return dnow;
}

void
profile_data_accumulator::add_current_time (void)
{
  const double t = query_time ();
  assert (last_time >= 0.0 && last_time <= t);

  assert (call_tree && active_fcn != call_tree);
  active_fcn->add_time (t - last_time);
}

profile_data_accumulator profiler;

// Enable or disable the profiler data collection.
DEFUN (__profiler_enable__, args, ,
  "-*- texinfo -*-\n\
@deftypefn {Function File} __profiler_enable ()\n\
Undocumented internal function.\n\
@end deftypefn")
{
  octave_value_list retval;

  const int nargin = args.length ();
  if (nargin > 0)
    {
      if (nargin > 1)
        {
          print_usage ();
          return retval;
        }

      profiler.set_active (args(0).bool_value ());
    }

  retval(0) = profiler.is_active ();

  return retval;
}

// Clear all collected profiling data.
DEFUN (__profiler_reset__, args, ,
  "-*- texinfo -*-\n\
@deftypefn {Function File} __profiler_reset ()\n\
Undocumented internal function.\n\
@end deftypefn")
{
  octave_value_list retval;
  const int nargin = args.length ();

  if (nargin > 0)
    warning ("profiler_reset: ignoring extra arguments");

  profiler.reset ();

  return retval;
}

// Query the timings collected by the profiler.
DEFUN (__profiler_data__, args, nargout,
  "-*- texinfo -*-\n\
@deftypefn {Function File} __profiler_data ()\n\
Undocumented internal function.\n\
@end deftypefn")
{
  octave_value_list retval;
  const int nargin = args.length ();

  if (nargin > 0)
    warning ("profiler_data: ignoring extra arguments");

  retval(0) = profiler.get_flat ();
  if (nargout > 1)
    retval(1) = profiler.get_hierarchical ();

  return retval;
}