view libinterp/corefcn/profiler.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 fcac5dbbf9ed
children aba2e6293dd8
line wrap: on
line source

/*

Copyright (C) 2014-2015 Julien Bect
Copyright (C) 2012-2015 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::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 */)
{
  // FIXME: These assert statements don't make sense if profile() is called
  //        from within a function hierarchy to begin with.  See bug #39587.
  //  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 (0), 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 (0, 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);
  // FIXME: This assert statements doesn't make sense if profile() is called
  //        from within a function hierarchy to begin with.  See bug #39587.
  //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);
  // FIXME: This assert statements doesn't make sense if profile() is called
  //        from within a function hierarchy to begin with.  See bug #39587.
  //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.");

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

  if (call_tree)
    {
      delete call_tree;
      call_tree = 0;
    }

  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 {} {} __profiler_enable__ ()\n\
Undocumented internal function.\n\
@end deftypefn")
{
  int nargin = args.length ();

  if (nargin > 1)
    print_usage ();

  if (nargin > 0)
    profiler.set_active (args(0).bool_value ());

  return ovl (profiler.is_active ());
}

// Clear all collected profiling data.
DEFUN (__profiler_reset__, args, ,
       "-*- texinfo -*-\n\
@deftypefn {} {} __profiler_reset__ ()\n\
Undocumented internal function.\n\
@end deftypefn")
{
  if (args.length () > 0)
    warning ("profiler_reset: ignoring extra arguments");

  profiler.reset ();

  return ovl ();
}

// Query the timings collected by the profiler.
DEFUN (__profiler_data__, args, nargout,
       "-*- texinfo -*-\n\
@deftypefn {} {} __profiler_data__ ()\n\
Undocumented internal function.\n\
@end deftypefn")
{
  if (args.length () > 0)
    warning ("profiler_data: ignoring extra arguments");

  if (nargout > 1)
    return ovl (profiler.get_flat (), profiler.get_hierarchical ());
  else
    return ovl (profiler.get_flat ());
}