view libinterp/parse-tree/profiler.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 32d2b6604a9f
children aac27ad79be6
line wrap: on
line source

////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2012-2022 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 "defun.h"
#include "event-manager.h"
#include "interpreter.h"
#include "oct-time.h"
#include "ov-struct.h"
#include "pager.h"
#include "profiler.h"

OCTAVE_BEGIN_NAMESPACE(octave)

  profiler::stats::stats (void)
    : m_time (0.0), m_calls (0), m_recursive (false),
      m_parents (), m_children ()
  { }

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

    RowVector retval (n);
    octave_idx_type i = 0;
    for (const auto& nm : list)
      retval(i++) = nm;

    return retval;
  }

  profiler::tree_node::tree_node (tree_node *p, octave_idx_type f)
    : m_parent (p), m_fcn_id (f), m_children (), m_time (0.0), m_calls (0)
  { }

  profiler::tree_node::~tree_node ()
  {
    for (auto& idx_tnode : m_children)
      delete idx_tnode.second;
  }

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

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

    ++retval->m_calls;
    return retval;
  }

  profiler::tree_node*
  profiler::tree_node::exit (octave_idx_type /* fcn */)
  {
    // FIXME: These panic_unless statements don't make sense if profile() is
    //  called from within a function hierarchy to begin with.  See bug #39587.
    //  panic_unless (m_parent);
    //  panic_unless (m_fcn_id == fcn);

    return m_parent;
  }

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

        entry.m_time += m_time;
        entry.m_calls += m_calls;

        panic_unless (m_parent);
        if (m_parent->m_fcn_id != 0)
          {
            entry.m_parents.insert (m_parent->m_fcn_id);
            data[m_parent->m_fcn_id - 1].m_children.insert (m_fcn_id);
          }

        if (! entry.m_recursive)
          for (const tree_node *i = m_parent; i; i = i->m_parent)
            if (i->m_fcn_id == m_fcn_id)
              {
                entry.m_recursive = true;
                break;
              }
      }

    // Recurse on children.
    for (const auto& idx_tnode : m_children)
      idx_tnode.second->build_flat (data);
  }

  octave_value
  profiler::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 = m_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 (const auto& idx_tnode : m_children)
      {
        const tree_node& entry = *idx_tnode.second;
        double child_total = entry.m_time;

        rv_indices(i) = octave_value (idx_tnode.first);
        rv_times(i) = octave_value (entry.m_time);
        rv_calls(i) = octave_value (entry.m_calls);
        rv_children(i) = entry.get_hierarchical (&child_total);
        rv_totals(i) = octave_value (child_total);

        if (total)
          *total += child_total;

        ++i;
      }

    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;
  }

  profiler::profiler (void)
    : m_known_functions (), m_fcn_index (),
      m_enabled (false), m_call_tree (new tree_node (nullptr, 0)),
      m_active_fcn (nullptr), m_last_time (-1.0)
  { }

  profiler::~profiler (void)
  {
    delete m_call_tree;
  }

  void
  profiler::set_active (bool value)
  {
    m_enabled = value;
  }

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

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

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

    if (! m_active_fcn)
      m_active_fcn = m_call_tree;

    m_active_fcn = m_active_fcn->enter (fcn_idx);

    m_last_time = query_time ();

  }

  void
  profiler::exit_function (const std::string& fcn)
  {
    if (m_active_fcn)
      {
        panic_unless (m_call_tree);
        // FIXME: This panic_unless statements doesn't make sense if profile()
        //        is called from within a function hierarchy to begin with.
        //        See bug #39587.
        // panic_unless (m_active_fcn != m_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 (enabled ())
          add_current_time ();

        fcn_index_map::iterator pos = m_fcn_index.find (fcn);
        // FIXME: This panic_unless statements doesn't make sense if profile()
        //        is called from within a function hierarchy to begin with.
        //        See bug #39587.
        // panic_unless (pos != m_fcn_index.end ());
        m_active_fcn = m_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!
        m_last_time = query_time ();
      }
  }

  void
  profiler::reset (void)
  {
    if (enabled ())
      error ("profile: can't reset active profiler");

    m_known_functions.clear ();
    m_fcn_index.clear ();

    if (m_call_tree)
      {
        delete m_call_tree;
        m_call_tree = new tree_node (nullptr, 0);
        m_active_fcn = nullptr;
      }

    m_last_time = -1.0;
  }

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

    const octave_idx_type n = m_known_functions.size ();

    flat_profile flat (n);

    if (m_call_tree)
      {
        m_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 (m_known_functions[i]);
            rv_times(i) = octave_value (flat[i].m_time);
            rv_calls(i) = octave_value (flat[i].m_calls);
            rv_recursive(i) = octave_value (flat[i].m_recursive);
            rv_parents(i) = stats::function_set_value (flat[i].m_parents);
            rv_children(i) = stats::function_set_value (flat[i].m_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",
            nullptr
          };

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

        retval = m;
      }

    return retval;
  }

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

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

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

        retval = m;
      }

    return retval;
  }

  double
  profiler::query_time (void) const
  {
    sys::time now;

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

    return dnow;
  }

  void
  profiler::add_current_time (void)
  {
    if (m_active_fcn)
      {
        const double t = query_time ();

        m_active_fcn->add_time (t - m_last_time);
      }
  }

// Enable or disable the profiler data collection.
DEFMETHOD (__profiler_enable__, interp, args, ,
           doc: /* -*- texinfo -*-
@deftypefn {} {@var{state} =} __profiler_enable__ ()
Undocumented internal function.
@end deftypefn */)
{
  int nargin = args.length ();

  if (nargin > 1)
    print_usage ();

  profiler& profiler = interp.get_profiler ();

  if (nargin == 1)
    {
      profiler.set_active (args(0).bool_value ());

      std::string status = "off";
      if (args(0).bool_value ())
        status = "on";

      event_manager& evmgr = interp.get_event_manager ();
      evmgr.gui_status_update ("profiler", status);  // tell GUI
    }

  return ovl (profiler.enabled ());
}

// Clear all collected profiling data.
DEFMETHOD (__profiler_reset__, interp, args, ,
           doc: /* -*- texinfo -*-
@deftypefn {} {} __profiler_reset__ ()
Undocumented internal function.
@end deftypefn */)
{
  if (args.length () != 0)
    print_usage ();

  profiler& profiler = interp.get_profiler ();

  profiler.reset ();

  return ovl ();
}

// Query the timings collected by the profiler.
DEFMETHOD (__profiler_data__, interp, args, nargout,
           doc: /* -*- texinfo -*-
@deftypefn {} {@var{data} =} __profiler_data__ ()
Undocumented internal function.
@end deftypefn */)
{
  if (args.length () != 0)
    print_usage ();

  profiler& profiler = interp.get_profiler ();

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

OCTAVE_END_NAMESPACE(octave)