[common.profiler] split imp into source file

dune/stuff/common/profiler.cc

+#include <boost/foreach.hpp>
+#include <boost/format.hpp>
+
+namespace Dune {
+namespace Stuff {
+namespace Common {
+
+void Profiler::startTiming(const std::string section_name)
+{
+  if (m_cur_run_num >= m_timings.size()) {
+    m_timings.push_back(DataMap());
+    m_total_runs++;
+  }
+
+  KnownTimersMap::iterator section = known_timers_map_.find(section_name);
+  if (section != known_timers_map_.end()) {
+    if (section->second.first) // timer currently running
+      return;
+
+    section->second.first  = true; // set active, start with new
+    section->second.second = TimingData(section_name, clock());
+  } else {
+    // init new section
+    known_timers_map_[section_name] = std::make_pair(true, TimingData(section_name, clock()));
+  }
+} // StartTiming
+
+void Profiler::stopTiming(const std::string section_name)
+{
+  assert(m_cur_run_num < m_timings.size());
+  if (known_timers_map_.find(section_name) == known_timers_map_.end())
+    DUNE_THROW(Dune::RangeError, "trying to stop timer " << section_name << " that wasn't started\n");
+
+  known_timers_map_[section_name].first      = false;
+  known_timers_map_[section_name].second.end = clock();
+  DataMap& current_data = m_timings[m_cur_run_num];
+  if (current_data.find(section_name) == current_data.end())
+    current_data[section_name] = known_timers_map_[section_name].second.delta();
+  else
+    current_data[section_name] += known_timers_map_[section_name].second.delta();
+} // StopTiming
+
+long Profiler::getTiming(const std::string section_name) const
+{
+  assert(m_cur_run_num < m_timings.size());
+  return getTiming(section_name, m_cur_run_num);
+}
+
+long Profiler::getTiming(const std::string section_name, const int run_number) const
+{
+  assert(run_number < int(m_timings.size()));
+  const DataMap& data             = m_timings[run_number];
+  DataMap::const_iterator section = data.find(section_name);
+  if (section == data.end()) {
+    ASSERT_EXCEPTION(false, "no timer found: " + section_name);
+    return -1;
+  }
+  return long(section->second / double(CLOCKS_PER_SEC));
+} // GetTiming
+
+
+void Profiler::reset(const int numRuns)
+{
+  m_timings.clear();
+  m_timings     = MapVector(numRuns, DataMap());
+  m_total_runs  = numRuns;
+  m_cur_run_num = 0;
+  init_time_    = clock();
+} // Reset
+
+void Profiler::addCount(const int num)
+{
+  m_count[num] += 1;
+}
+
+void Profiler::nextRun()
+{
+  m_cur_run_num++;
+}
+
+template <class CollectiveCommunication>
+long Profiler::outputAveraged(CollectiveCommunication& comm, const int refineLevel, const long numDofs,
+                              const double scale_factor)
+{
+  const int numProce = comm.size();
+
+  std::ostringstream filename;
+
+  filename << "p" << numProce << "_refinelvl_" << refineLevel << ".csv";
+  filename.flush();
+
+  if (comm.rank() == 0)
+    std::cout << "Profiling info in: " << (filename.str()).c_str() << std::endl;
+
+#ifndef NDEBUG
+  for (std::map<int, int>::const_iterator it = m_count.begin(); it != m_count.end(); ++it) {
+    std::cout << "proc " << comm.rank() << " bId " << it->first << " count " << it->second << std::endl;
+  }
+#endif // ifndef NDEBUG
+
+  Dune::Stuff::Common::Filesystem::testCreateDirectory(Dune::Stuff::Common::Filesystem::pathOnly(filename.str()));
+  std::ofstream csv((filename.str()).c_str());
+
+  typedef std::map<std::string, long> AvgMap;
+  AvgMap averages;
+  for (MapVector::const_iterator vit = m_timings.begin(); vit != m_timings.end(); ++vit) {
+    for (DataMap::const_iterator it = vit->begin(); it != vit->end(); ++it) {
+      //! this used to be GetTiming( it->second ), which is only valid thru an implicit and wrong conversion..
+      averages[it->first] += getTiming(it->first);
+    }
+  }
+
+  // outputs column names
+  csv << "refine,"
+      << "processes,"
+      << "numDofs,"
+      << "L1 error,";
+  for (AvgMap::const_iterator it = averages.begin(); it != averages.end(); ++it) {
+    csv << it->first << ",";
+  }
+  csv << "Speedup (total); Speedup (ohne Solver)" << std::endl;
+
+  // outputs column values
+  csv << refineLevel << "," << comm.size() << "," << numDofs << "," << 0 << ","; //! FIXME
+  for (AvgMap::const_iterator it = averages.begin(); it != averages.end(); ++it) {
+    long clock_count = it->second;
+    clock_count = long(comm.sum(clock_count) / double(scale_factor * numProce));
+    csv << clock_count / double(m_total_runs) << ",";
+  }
+  csv << "=I$2/I2,"
+      << "=SUM(E$2:G$2)/SUM(E2:G2)" << std::endl;
+
+  csv.close();
+
+  return long((clock() - init_time_) / double(CLOCKS_PER_SEC * scale_factor));
+} // OutputAveraged
+
+template <class CollectiveCommunication, class InfoContainer>
+long Profiler::output(CollectiveCommunication& comm, InfoContainer& run_infos, const double scale_factor)
+{
+  const int numProce = comm.size();
+
+  std::ostringstream filename;
+
+  filename << Dune::Stuff::Common::Parameter::Parameters().getParam("fem.io.datadir", std::string(".")) << "/prof_p"
+           << numProce << ".csv";
+  filename.flush();
+  return OutputCommon(comm, run_infos, filename.str(), scale_factor);
+} // Output
+
+template <class CollectiveCommunication, class InfoContainerMap>
+void Profiler::outputMap(CollectiveCommunication& comm, InfoContainerMap& run_infos_map, const double scale_factor)
+{
+  std::string dir(Dune::Stuff::Common::Parameter::Parameters().getParam("fem.io.datadir", std::string(".")));
+
+  BOOST_FOREACH (typename InfoContainerMap::value_type el, run_infos_map) {
+    OutputCommon(
+        comm, el.second, (boost::format("%s/prof_p%d_ref%s.csv") % dir % comm.size() % el.first).str(), scale_factor);
+  }
+} // OutputMap
+
+template <class CollectiveCommunication, class InfoContainer>
+long Profiler::outputCommon(CollectiveCommunication& comm, InfoContainer& run_infos, std::string filename,
+                            const double scale_factor)
+{
+  const int numProce = comm.size();
+
+  if (comm.rank() == 0)
+    std::cout << "Profiling info in: " << filename.c_str() << std::endl;
+
+#ifndef NDEBUG
+  for (std::map<int, int>::const_iterator it = m_count.begin(); it != m_count.end(); ++it) {
+    std::cout << "proc " << comm.rank() << " bId " << it->first << " count " << it->second << std::endl;
+  }
+#endif // ifndef NDEBUG
+
+  Dune::Stuff::Common::Filesystem::testCreateDirectory(Dune::Stuff::Common::Filesystem::pathOnly(filename));
+  std::ofstream csv(filename.c_str());
+
+  // outputs column names
+  csv << "refine,"
+      << "processes,"
+      << "numDofs,"
+      << "L2_error,";
+  for (DataMap::const_iterator it = m_timings[0].begin(); it != m_timings[0].end(); ++it) {
+    csv << it->first << ",";
+  }
+  csv << "Relative_total_time,compiler" << std::endl;
+
+  // outputs column values
+
+  MapVector::const_iterator ti_it = m_timings.begin();
+  int idx = 0;
+  assert(run_infos.size() >= m_timings.size());
+  for (; ti_it != m_timings.end(); ++ti_it) {
+    Dune::Stuff::Common::RunInfo info = run_infos[idx];
+    csv << boost::format("%d,%d,%d,%e,") % info.refine_level % comm.size() % info.codim0
+               % (info.L2Errors.size() ? info.L2Errors[0] : double(-1));
+
+    const DataMap& data_map = *ti_it;
+    for (DataMap::const_iterator it = data_map.begin(); it != data_map.end(); ++it) {
+      long clock_count = getTiming(it->first, idx);
+      clock_count      = long(comm.sum(clock_count) / double(scale_factor * numProce));
+      csv << clock_count << ",";
+    }
+    csv << boost::format("=1/I$2*I%d,%s\n") % (idx + 2) % BOOST_COMPILER;
+
+    idx++;
+  }
+
+  csv.close();
+
+  return long((clock() - init_time_) / double(CLOCKS_PER_SEC * scale_factor));
+} // OutputCommon
+
+} // namespace Common
+} // namespace Stuff
+} // namespace Dune

dune/stuff/common/profiler.hh

@@ -80,60 +80,14 @@ protected:
 
 public:
   //! set this to begin a named section
-  void startTiming(const std::string section_name)
-  {
-    if (m_cur_run_num >= m_timings.size()) {
-      m_timings.push_back(DataMap());
-      m_total_runs++;
-    }
-
-    KnownTimersMap::iterator section = known_timers_map_.find(section_name);
-    if (section != known_timers_map_.end()) {
-      if (section->second.first) // timer currently running
-        return;
-
-      section->second.first  = true; // set active, start with new
-      section->second.second = TimingData(section_name, clock());
-    } else {
-      // init new section
-      known_timers_map_[section_name] = std::make_pair(true, TimingData(section_name, clock()));
-    }
-  } // StartTiming
+  void startTiming(const std::string section_name);
 
   //! stop named section's counter
-  void stopTiming(const std::string section_name)
-  {
-    assert(m_cur_run_num < m_timings.size());
-    if (known_timers_map_.find(section_name) == known_timers_map_.end())
-      DUNE_THROW(Dune::RangeError, "trying to stop timer " << section_name << " that wasn't started\n");
-
-    known_timers_map_[section_name].first      = false;
-    known_timers_map_[section_name].second.end = clock();
-    DataMap& current_data = m_timings[m_cur_run_num];
-    if (current_data.find(section_name) == current_data.end())
-      current_data[section_name] = known_timers_map_[section_name].second.delta();
-    else
-      current_data[section_name] += known_timers_map_[section_name].second.delta();
-  } // StopTiming
+  void stopTiming(const std::string section_name);
 
   //! get runtime of section in seconds
-  long getTiming(const std::string section_name) const
-  {
-    assert(m_cur_run_num < m_timings.size());
-    return getTiming(section_name, m_cur_run_num);
-  }
-
-  long getTiming(const std::string section_name, const int run_number) const
-  {
-    assert(run_number < int(m_timings.size()));
-    const DataMap& data             = m_timings[run_number];
-    DataMap::const_iterator section = data.find(section_name);
-    if (section == data.end()) {
-      ASSERT_EXCEPTION(false, "no timer found: " + section_name);
-      return -1;
-    }
-    return long(section->second / double(CLOCKS_PER_SEC));
-  } // GetTiming
+  long getTiming(const std::string section_name) const;
+  long getTiming(const std::string section_name, const int run_number) const;
 
   /** output to currently pre-defined (csv) file, does not output individual run results, but average over all recorded
    * results
@@ -164,27 +118,14 @@ public:
      *  if you're planning on doing more than one iteration of your code
      *  called once fromm ctor with numRuns=1
      **/
-  void reset(const int numRuns)
-  {
-    m_timings.clear();
-    m_timings     = MapVector(numRuns, DataMap());
-    m_total_runs  = numRuns;
-    m_cur_run_num = 0;
-    init_time_    = clock();
-  } // Reset
+  void reset(const int numRuns);
 
   //! simple counter, usable to count how often a single piece of code is called
-  void addCount(const int num)
-  {
-    m_count[num] += 1;
-  }
+  void addCount(const int num);
 
   //! call this after one iteration of your code has finished. increments current run number and puts new timing data
   // into the vector
-  void nextRun()
-  {
-    m_cur_run_num++;
-  }
+  void nextRun();
 
   //! a utility class to time a limited scope of code
   class ScopedTiming : public boost::noncopyable
@@ -220,142 +161,6 @@ protected:
   }
 };
 
-template <class CollectiveCommunication>
-long Profiler::outputAveraged(CollectiveCommunication& comm, const int refineLevel, const long numDofs,
-                              const double scale_factor)
-{
-  const int numProce = comm.size();
-
-  std::ostringstream filename;
-
-  filename << "p" << numProce << "_refinelvl_" << refineLevel << ".csv";
-  filename.flush();
-
-  if (comm.rank() == 0)
-    std::cout << "Profiling info in: " << (filename.str()).c_str() << std::endl;
-
-#ifndef NDEBUG
-  for (std::map<int, int>::const_iterator it = m_count.begin(); it != m_count.end(); ++it) {
-    std::cout << "proc " << comm.rank() << " bId " << it->first << " count " << it->second << std::endl;
-  }
-#endif // ifndef NDEBUG
-
-  Dune::Stuff::Common::Filesystem::testCreateDirectory(Dune::Stuff::Common::Filesystem::pathOnly(filename.str()));
-  std::ofstream csv((filename.str()).c_str());
-
-  typedef std::map<std::string, long> AvgMap;
-  AvgMap averages;
-  for (MapVector::const_iterator vit = m_timings.begin(); vit != m_timings.end(); ++vit) {
-    for (DataMap::const_iterator it = vit->begin(); it != vit->end(); ++it) {
-      //! this used to be GetTiming( it->second ), which is only valid thru an implicit and wrong conversion..
-      averages[it->first] += getTiming(it->first);
-    }
-  }
-
-  // outputs column names
-  csv << "refine,"
-      << "processes,"
-      << "numDofs,"
-      << "L1 error,";
-  for (AvgMap::const_iterator it = averages.begin(); it != averages.end(); ++it) {
-    csv << it->first << ",";
-  }
-  csv << "Speedup (total); Speedup (ohne Solver)" << std::endl;
-
-  // outputs column values
-  csv << refineLevel << "," << comm.size() << "," << numDofs << "," << 0 << ","; //! FIXME
-  for (AvgMap::const_iterator it = averages.begin(); it != averages.end(); ++it) {
-    long clock_count = it->second;
-    clock_count = long(comm.sum(clock_count) / double(scale_factor * numProce));
-    csv << clock_count / double(m_total_runs) << ",";
-  }
-  csv << "=I$2/I2,"
-      << "=SUM(E$2:G$2)/SUM(E2:G2)" << std::endl;
-
-  csv.close();
-
-  return long((clock() - init_time_) / double(CLOCKS_PER_SEC * scale_factor));
-} // OutputAveraged
-
-template <class CollectiveCommunication, class InfoContainer>
-long Profiler::output(CollectiveCommunication& comm, InfoContainer& run_infos, const double scale_factor)
-{
-  const int numProce = comm.size();
-
-  std::ostringstream filename;
-
-  filename << Dune::Stuff::Common::Parameter::Parameters().getParam("fem.io.datadir", std::string(".")) << "/prof_p"
-           << numProce << ".csv";
-  filename.flush();
-  return OutputCommon(comm, run_infos, filename.str(), scale_factor);
-} // Output
-
-template <class CollectiveCommunication, class InfoContainerMap>
-void Profiler::outputMap(CollectiveCommunication& comm, InfoContainerMap& run_infos_map, const double scale_factor)
-{
-  std::string dir(Dune::Stuff::Common::Parameter::Parameters().getParam("fem.io.datadir", std::string(".")));
-
-  BOOST_FOREACH (typename InfoContainerMap::value_type el, run_infos_map) {
-    OutputCommon(
-        comm, el.second, (boost::format("%s/prof_p%d_ref%s.csv") % dir % comm.size() % el.first).str(), scale_factor);
-  }
-} // OutputMap
-
-template <class CollectiveCommunication, class InfoContainer>
-long Profiler::outputCommon(CollectiveCommunication& comm, InfoContainer& run_infos, std::string filename,
-                            const double scale_factor)
-{
-  const int numProce = comm.size();
-
-  if (comm.rank() == 0)
-    std::cout << "Profiling info in: " << filename.c_str() << std::endl;
-
-#ifndef NDEBUG
-  for (std::map<int, int>::const_iterator it = m_count.begin(); it != m_count.end(); ++it) {
-    std::cout << "proc " << comm.rank() << " bId " << it->first << " count " << it->second << std::endl;
-  }
-#endif // ifndef NDEBUG
-
-  Dune::Stuff::Common::Filesystem::testCreateDirectory(Dune::Stuff::Common::Filesystem::pathOnly(filename));
-  std::ofstream csv(filename.c_str());
-
-  // outputs column names
-  csv << "refine,"
-      << "processes,"
-      << "numDofs,"
-      << "L2_error,";
-  for (DataMap::const_iterator it = m_timings[0].begin(); it != m_timings[0].end(); ++it) {
-    csv << it->first << ",";
-  }
-  csv << "Relative_total_time,compiler" << std::endl;
-
-  // outputs column values
-
-  MapVector::const_iterator ti_it = m_timings.begin();
-  int idx = 0;
-  assert(run_infos.size() >= m_timings.size());
-  for (; ti_it != m_timings.end(); ++ti_it) {
-    Dune::Stuff::Common::RunInfo info = run_infos[idx];
-    csv << boost::format("%d,%d,%d,%e,") % info.refine_level % comm.size() % info.codim0
-               % (info.L2Errors.size() ? info.L2Errors[0] : double(-1));
-
-    const DataMap& data_map = *ti_it;
-    for (DataMap::const_iterator it = data_map.begin(); it != data_map.end(); ++it) {
-      long clock_count = getTiming(it->first, idx);
-      clock_count      = long(comm.sum(clock_count) / double(scale_factor * numProce));
-      csv << clock_count << ",";
-    }
-    csv << boost::format("=1/I$2*I%d,%s\n") % (idx + 2) % BOOST_COMPILER;
-
-    idx++;
-  }
-
-  csv.close();
-
-  return long((clock() - init_time_) / double(CLOCKS_PER_SEC * scale_factor));
-} // OutputCommon
-
-
 //! global profiler object
 Profiler& profiler()
 {
@@ -368,6 +173,8 @@ Profiler& profiler()
 
 #define DSC_PROF Dune::Stuff::Common::profiler()
 
+#include "profiler.cc"
+
 #endif // DUNE_STUFF_PROFILER_HH_INCLUDED
 /** Copyright (c) 2012, Rene Milk
    * All rights reserved.