Make the Tracy integration functional

Common/include/CConfig.hpp

@@ -46,6 +46,7 @@
 #include "option_structure.hpp"
 #include "containers/container_decorators.hpp"
 #include "toolboxes/printing_toolbox.hpp"
+#include "tracy_structure.hpp"
 
 #ifdef HAVE_CGNS
 #include "cgnslib.h"
@@ -9097,25 +9098,6 @@ class CConfig {
    */
   unsigned long GetNonphysical_Reconstr(void) const { return Nonphys_Reconstr; }
 
-  /*!
-   * \brief Start the timer for profiling subroutines.
-   * \param[in] val_start_time - the value of the start time.
-   */
-  void Tick(double *val_start_time);
-
-  /*!
-   * \brief Stop the timer for profiling subroutines and store results.
-   * \param[in] val_start_time - the value of the start time.
-   * \param[in] val_function_name - string for the name of the profiled subroutine.
-   * \param[in] val_group_id - string for the name of the profiled subroutine.
-   */
-  void Tock(double val_start_time, const string& val_function_name, int val_group_id);
-
-  /*!
-   * \brief Write a CSV file containing the results of the profiling.
-   */
-  void SetProfilingCSV(void);
-
   /*!
    * \brief Start the timer for profiling subroutines.
    * \param[in] val_start_time - the value of the start time.

Common/include/linear_algebra/CSysSolve.hpp

@@ -246,6 +246,7 @@ class CSysSolve {
    */
   template <class OtherType>
   void HandleTemporariesIn(const CSysVector<OtherType>& LinSysRes, CSysVector<OtherType>& LinSysSol) {
+    SU2_ZONE_SCOPED
     if constexpr (std::is_same_v<ScalarType, OtherType>) {
       /*--- Same type specialization, temporary variables are not required. ---*/
       BEGIN_SU2_OMP_SAFE_GLOBAL_ACCESS {
@@ -273,6 +274,7 @@ class CSysSolve {
    */
   template <class OtherType>
   void HandleTemporariesOut(CSysVector<OtherType>& LinSysSol) {
+    SU2_ZONE_SCOPED
     if constexpr (std::is_same_v<ScalarType, OtherType>) {
       /*--- Same type specialization, temporary variables are not required. ---*/
       BEGIN_SU2_OMP_SAFE_GLOBAL_ACCESS {

Common/include/tracy_structure.hpp

@@ -6,7 +6,6 @@
  *        they can be completely "disabled" when compiling without tracy.
  * \note Do not include tracy headers explicitly anywhere, use this header instead.
  * \note To enable tracy, define the TRACY_ENABLE macro during compilation.
- * \author Divyaprakash
  * \version 8.4.0 "Harrier"
  *
  * SU2 Project Website: https://su2code.github.io
@@ -32,11 +31,16 @@
 
 #pragma once
 
-#ifdef HAVE_TRACY
+#ifdef TRACY_ENABLE
 #include "tracy/Tracy.hpp"
-#define SU2_ZONE_SCOPED ZoneScoped
-#define SU2_ZONE_SCOPED_N(name) ZoneScopedN(name)
+#define SU2_ZONE_SCOPED ZoneScoped;
+#define SU2_ZONE_SCOPED_N(name) ZoneScopedN(name);
 #else
 #define SU2_ZONE_SCOPED
 #define SU2_ZONE_SCOPED_N(name)
 #endif
+
+#define BEGIN_SU2_ZONE_N(name) \
+  {                            \
+    SU2_ZONE_SCOPED_N(name)
+#define END_SU2_ZONE }

Common/src/CConfig.cpp

@@ -46,11 +46,6 @@ using namespace PrintingToolbox;
 #endif
 #endif
 
-vector<string> Profile_Function_tp;       /*!< \brief Vector of string names for profiled functions. */
-vector<double> Profile_Time_tp;           /*!< \brief Vector of elapsed time for profiled functions. */
-vector<double> Profile_ID_tp;             /*!< \brief Vector of group ID number for profiled functions. */
-map<string, vector<int> > Profile_Map_tp; /*!< \brief Map containing the final results for profiled functions. */
-
 map<CLong3T, int> GEMM_Profile_MNK;       /*!< \brief Map, which maps the GEMM size to the index where
                                                       the data for this GEMM is stored in several vectors. */
 vector<long>   GEMM_Profile_NCalls;       /*!< \brief Vector, which stores the number of calls to this
@@ -59,8 +54,6 @@ vector<double> GEMM_Profile_TotTime;      /*!< \brief Total time spent for this
 vector<double> GEMM_Profile_MinTime;      /*!< \brief Minimum time spent for this GEMM size. */
 vector<double> GEMM_Profile_MaxTime;      /*!< \brief Maximum time spent for this GEMM size. */
 
-//#pragma omp threadprivate(Profile_Function_tp, Profile_Time_tp, Profile_ID_tp, Profile_Map_tp)
-
 
 CConfig::CConfig(char case_filename[MAX_STRING_SIZE], SU2_COMPONENT val_software, bool verb_high) {
 
@@ -9953,191 +9946,6 @@ short CConfig::FindInterfaceMarker(unsigned short iInterface) const {
   return -1;
 }
 
-void CConfig::Tick(double *val_start_time) {
-
-#ifdef PROFILE
-  *val_start_time = SU2_MPI::Wtime();
-#endif
-
-}
-
-void CConfig::Tock(double val_start_time, const string& val_function_name, int val_group_id) {
-
-#ifdef PROFILE
-
-  double val_stop_time = 0.0, val_elapsed_time = 0.0;
-
-  val_stop_time = SU2_MPI::Wtime();
-
-  /*--- Compute the elapsed time for this subroutine ---*/
-  val_elapsed_time = val_stop_time - val_start_time;
-
-  /*--- Store the subroutine name and the elapsed time ---*/
-  Profile_Function_tp.push_back(val_function_name);
-  Profile_Time_tp.push_back(val_elapsed_time);
-  Profile_ID_tp.push_back(val_group_id);
-
-#endif
-
-}
-
-void CConfig::SetProfilingCSV() {
-
-#ifdef PROFILE
-
-  int rank = MASTER_NODE;
-  int size = SINGLE_NODE;
-#ifdef HAVE_MPI
-  SU2_MPI::Comm_rank(SU2_MPI::GetComm(), &rank);
-  SU2_MPI::Comm_size(SU2_MPI::GetComm(), &size);
-#endif
-
-  /*--- Each rank has the same stack trace, so the they have the same
-   function calls and ordering in the vectors. We're going to reduce
-   the timings from each rank and extract the avg, min, and max timings. ---*/
-
-  /*--- First, create a local mapping, so that we can extract the
-   min and max values for each function. ---*/
-
-  for (unsigned int i = 0; i < Profile_Function_tp.size(); i++) {
-
-    /*--- Add the function and initialize if not already stored (the ID
-     only needs to be stored the first time).---*/
-    if (Profile_Map_tp.find(Profile_Function_tp[i]) == Profile_Map_tp.end()) {
-
-      vector<int> profile; profile.push_back(i);
-      Profile_Map_tp.insert(pair<string,vector<int> >(Profile_Function_tp[i],profile));
-
-    } else {
-
-      /*--- This function has already been added, so simply increment the
-       number of calls and total time for this function. ---*/
-
-      Profile_Map_tp[Profile_Function_tp[i]].push_back(i);
-
-    }
-  }
-
-  /*--- We now have everything gathered by function name, so we can loop over
-   each function and store the min/max times. ---*/
-
-  int map_size = 0;
-  for (map<string,vector<int> >::iterator it=Profile_Map_tp.begin(); it!=Profile_Map_tp.end(); ++it) {
-    map_size++;
-  }
-
-  /*--- Allocate and initialize memory ---*/
-
-  double *l_min_red = nullptr, *l_max_red = nullptr, *l_tot_red = nullptr, *l_avg_red = nullptr;
-  int *n_calls_red = nullptr;
-  double* l_min = new double[map_size];
-  double* l_max = new double[map_size];
-  double* l_tot = new double[map_size];
-  double* l_avg = new double[map_size];
-  int* n_calls  = new int[map_size];
-  for (int i = 0; i < map_size; i++)
-  {
-    l_min[i]   = 1e10;
-    l_max[i]   = 0.0;
-    l_tot[i]   = 0.0;
-    l_avg[i]   = 0.0;
-    n_calls[i] = 0;
-  }
-
-  /*--- Collect the info for each function from the current rank ---*/
-
-  int func_counter = 0;
-  for (map<string,vector<int> >::iterator it=Profile_Map_tp.begin(); it!=Profile_Map_tp.end(); ++it) {
-
-    for (unsigned int i = 0; i < (it->second).size(); i++) {
-      n_calls[func_counter]++;
-      l_tot[func_counter] += Profile_Time_tp[(it->second)[i]];
-      if (Profile_Time_tp[(it->second)[i]] < l_min[func_counter])
-        l_min[func_counter] = Profile_Time_tp[(it->second)[i]];
-      if (Profile_Time_tp[(it->second)[i]] > l_max[func_counter])
-        l_max[func_counter] = Profile_Time_tp[(it->second)[i]];
-
-    }
-    l_avg[func_counter] = l_tot[func_counter]/((double)n_calls[func_counter]);
-    func_counter++;
-  }
-
-  /*--- Now reduce the data ---*/
-
-  if (rank == MASTER_NODE) {
-    l_min_red = new double[map_size];
-    l_max_red = new double[map_size];
-    l_tot_red = new double[map_size];
-    l_avg_red = new double[map_size];
-    n_calls_red  = new int[map_size];
-  }
-
-#ifdef HAVE_MPI
-  MPI_Reduce(n_calls, n_calls_red, map_size, MPI_INT, MPI_SUM, MASTER_NODE, SU2_MPI::GetComm());
-  MPI_Reduce(l_tot, l_tot_red, map_size, MPI_DOUBLE, MPI_SUM, MASTER_NODE, SU2_MPI::GetComm());
-  MPI_Reduce(l_avg, l_avg_red, map_size, MPI_DOUBLE, MPI_SUM, MASTER_NODE, SU2_MPI::GetComm());
-  MPI_Reduce(l_min, l_min_red, map_size, MPI_DOUBLE, MPI_MIN, MASTER_NODE, SU2_MPI::GetComm());
-  MPI_Reduce(l_max, l_max_red, map_size, MPI_DOUBLE, MPI_MAX, MASTER_NODE, SU2_MPI::GetComm());
-#else
-  memcpy(n_calls_red, n_calls, map_size*sizeof(int));
-  memcpy(l_tot_red,   l_tot,   map_size*sizeof(double));
-  memcpy(l_avg_red,   l_avg,   map_size*sizeof(double));
-  memcpy(l_min_red,   l_min,   map_size*sizeof(double));
-  memcpy(l_max_red,   l_max,   map_size*sizeof(double));
-#endif
-
-  /*--- The master rank will write the file ---*/
-
-  if (rank == MASTER_NODE) {
-
-    /*--- Take averages over all ranks on the master ---*/
-
-    for (int i = 0; i < map_size; i++) {
-      l_tot_red[i]   = l_tot_red[i]/(double)size;
-      l_avg_red[i]   = l_avg_red[i]/(double)size;
-      n_calls_red[i] = n_calls_red[i]/size;
-    }
-
-    /*--- Now write a CSV file with the processed results ---*/
-
-    ofstream Profile_File;
-    Profile_File.precision(15);
-    Profile_File.open("profiling.csv");
-
-    /*--- Create the CSV header ---*/
-
-    Profile_File << "\"Function_Name\", \"N_Calls\", \"Avg_Total_Time\", \"Avg_Time\", \"Min_Time\", \"Max_Time\", \"Function_ID\"" << endl;
-
-    /*--- Loop through the map and write the results to the file ---*/
-
-    func_counter = 0;
-    for (map<string,vector<int> >::iterator it=Profile_Map_tp.begin(); it!=Profile_Map_tp.end(); ++it) {
-
-      Profile_File << scientific << it->first << ", " << n_calls_red[func_counter] << ", " << l_tot_red[func_counter] << ", " << l_avg_red[func_counter] << ", " << l_min_red[func_counter] << ", " << l_max_red[func_counter] << ", " << (int)Profile_ID_tp[(it->second)[0]] << endl;
-      func_counter++;
-    }
-
-    Profile_File.close();
-
-  }
-
-  delete [] l_min;
-  delete [] l_max;
-  delete [] l_avg;
-  delete [] l_tot;
-  delete [] n_calls;
-  if (rank == MASTER_NODE) {
-    delete [] l_min_red;
-    delete [] l_max_red;
-    delete [] l_avg_red;
-    delete [] l_tot_red;
-    delete [] n_calls_red;
-  }
-
-#endif
-
-}
-
 void CConfig::GEMM_Tick(double *val_start_time) const {
 
 #ifdef PROFILE