Merge branch 'develop' of git.nic.uoregon.edu:/gitroot/xpress-apex in…

…to develop

CMakeLists.txt

@@ -784,6 +784,13 @@ CONFIGURE_FILE(${CMAKE_SOURCE_DIR}/pkgconfig/apex.pc.in
                ${CMAKE_BINARY_DIR}/pkgconfig/apex.pc @ONLY)
 INSTALL_FILES(/lib/pkgconfig FILES pkgconfig/apex.pc)
 
+if (APEX_USE_WEAK_SYMBOLS) 
+    add_definitions(-DAPEX_USE_WEAK_SYMBOLS)
+else()
+    find_library(DYNAMICLIB dl)
+    set(LIBS ${LIBS} ${DYNAMICLIB})
+endif (APEX_USE_WEAK_SYMBOLS)
+
 message(INFO " Getting processor count...")
 include(ProcessorCount) # how many cores at our disposal?
 if(PROCESSOR_COUNT)

src/apex/CMakeLists.hpx

@@ -304,6 +304,13 @@ if(APEX_WITH_MSR)
     hpx_libraries(${MSR_LIBRARIES})
 endif()
 
+if (APEX_USE_WEAK_SYMBOLS)
+    add_definitions(-DAPEX_USE_WEAK_SYMBOLS)
+else()
+    find_library(DYNAMICLIB dl)
+    hpx_libraries(${DYNAMICLIB})
+endif (APEX_USE_WEAK_SYMBOLS)
+
 set(apex_sources
     apex.cpp
     apex_options.cpp

src/apex/apex_ompt.cpp

@@ -41,8 +41,14 @@ class linked_timer {
             std::shared_ptr<apex::task_wrapper> &parent,
             bool auto_start) :
             prev(p), timing(auto_start) {
-            tw = apex::new_task(name, task_id, parent);
-            if (auto_start) { this->start(); }
+                // No GUIDs generated by the runtime? Generate our own.
+                if (task_id == 0ULL) {
+                    tw = apex::new_task(name);
+                } else {
+                    tw = apex::new_task(name, task_id, parent);
+                }
+                if (auto_start) { this->start(); 
+            }
         }
         /* destructor */
         ~linked_timer() {

src/apex/concurrency_handler.cpp

@@ -77,7 +77,7 @@ bool concurrency_handler::_handler(void) {
   apex* inst = apex::instance();
   if (inst == nullptr) return false; // running after finalization!
   if (apex_options::use_tau()) {
-    Tau_start("concurrency_handler::_handler");
+    tau_listener::Tau_start_wrapper("concurrency_handler::_handler");
   }
   map<task_identifier, unsigned int> *counts = new(map<task_identifier,
     unsigned int>);
@@ -119,7 +119,7 @@ bool concurrency_handler::_handler(void) {
     _power_samples.push_back(power);
   }
   if (apex_options::use_tau()) {
-    Tau_stop("concurrency_handler::_handler");
+    tau_listener::Tau_stop_wrapper("concurrency_handler::_handler");
   }
 #ifdef APEX_WITH_JUPYTER_SUPPORT
   // update the timeout, if the user changed it.

src/apex/otf2_listener.cpp

@@ -293,7 +293,8 @@ namespace apex {
       if (evt_writer == nullptr && create) {
         // should already be locked by the "new thread" event.
         uint64_t my_node_id = my_saved_node_id;
-        my_node_id = (my_node_id << 32) + thread_instance::get_id();
+        //my_node_id = (my_node_id << 32) + thread_instance::get_id();
+        my_node_id = (my_node_id << 32) + _event_threads.size();
         evt_writer = OTF2_Archive_GetEvtWriter( archive, my_node_id );
         if (thread_instance::get_id() == 0) {
             comm_evt_writer = evt_writer;
@@ -570,11 +571,12 @@ namespace apex {
             uint64_t map_size = global_region_indices.size();
             OTF2_IdMap * my_map = OTF2_IdMap_CreateFromUint64Array(map_size,
                 mappings, false);
-            for (int i = 0 ; i < thread_instance::get_num_threads() ; i++) {
-                if (event_file_exists(i)) {
+            //for (int i = 0 ; i < thread_instance::get_num_threads() ; i++) {
+            for (size_t i = 0 ; i < _event_threads.size() ; i++) {
+                //if (event_file_exists(i)) {
                     OTF2_DefWriter_WriteMappingTable(getDefWriter(i),
                         OTF2_MAPPING_REGION, my_map);
-                }
+                //}
             }
             // free the map
             OTF2_IdMap_Free(my_map);
@@ -598,11 +600,12 @@ namespace apex {
             uint64_t map_size = global_metric_indices.size();
             OTF2_IdMap * my_map = OTF2_IdMap_CreateFromUint64Array(map_size,
                 mappings, false);
-            for (int i = 0 ; i < thread_instance::get_num_threads() ; i++) {
-                if (event_file_exists(i)) {
+            //for (int i = 0 ; i < thread_instance::get_num_threads() ; i++) {
+            for (size_t i = 0 ; i < _event_threads.size() ; i++) {
+                //if (event_file_exists(i)) {
                     OTF2_DefWriter_WriteMappingTable(getDefWriter(i),
                         OTF2_MAPPING_METRIC, my_map);
-                }
+                //}
             }
             // free the map
             OTF2_IdMap_Free(my_map);
@@ -717,7 +720,8 @@ namespace apex {
             /* if we are node 0, write the global definitions */
             if (my_saved_node_id == 0) {
                 // save my number of threads
-                rank_thread_map[0] = thread_instance::get_num_threads();
+                //rank_thread_map[0] = thread_instance::get_num_threads();
+                rank_thread_map[0] = _event_threads.size();
                 std::cout << "Writing OTF2 definition files..." << std::endl;
                 // make a common list of regions and metrics across all nodes...
                 reduce_regions();
@@ -800,12 +804,13 @@ namespace apex {
                 // write out the counter names we saw
                 reduce_metrics();
             }
-            for (int i = 0 ; i < thread_instance::get_num_threads() ; i++) {
+            //for (int i = 0 ; i < thread_instance::get_num_threads() ; i++) {
+            for (size_t i = 0 ; i < _event_threads.size() ; i++) {
                 /* close (and possibly create) the definition files */
-                if (event_file_exists(i)) {
+                //if (event_file_exists(i)) {
                     OTF2_EC(OTF2_Archive_CloseDefWriter( archive,
                         getDefWriter(i) ));
-                }
+                //}
             }
             if (my_saved_node_id == 0) {
                 std::cout << "Closing the archive..." << std::endl;
@@ -983,7 +988,8 @@ namespace apex {
             // create an attribute list
             OTF2_AttributeList * al = OTF2_AttributeList_New();
             // create an attribute
-            OTF2_AttributeList_AddUint64( al, 0, p->guid );
+            OTF2_AttributeList_AddUint64( al, 0, p->tt_ptr->guid );
+            OTF2_AttributeList_AddUint64( al, 1, p->tt_ptr->parent_guid );
             // unfortunately, we can't use the timestamp from the
             // profiler object. bummer. it has to be taken after
             // the lock is acquired, so that events happen on
@@ -1184,7 +1190,8 @@ namespace apex {
     std::string otf2_listener::write_my_regions(void) {
         stringstream region_file;
         // first, output our number of threads.
-        region_file << thread_instance::get_num_threads() << endl;
+        //region_file << thread_instance::get_num_threads() << endl;
+        region_file << _event_threads.size() << endl;
         // then iterate over the regions and write them out.
         for (auto const &i : global_region_indices) {
             task_identifier id = i.first;
@@ -1323,7 +1330,8 @@ namespace apex {
     std::string otf2_listener::write_my_metrics(void) {
         stringstream metric_file;
         // first, output our number of threads.
-        metric_file << thread_instance::get_num_threads() << endl;
+        //metric_file << thread_instance::get_num_threads() << endl;
+        metric_file << _event_threads.size() << endl;
         // then iterate over the metrics and write them out.
         for (auto const &i : global_metric_indices) {
             string id = i.first;
@@ -1512,7 +1520,8 @@ namespace apex {
         region_filename << region_filename_prefix << my_saved_node_id;
         ofstream region_file(region_filename.str(), ios::out | ios::trunc );
         // first, output our number of threads.
-        region_file << thread_instance::get_num_threads() << endl;
+        //region_file << thread_instance::get_num_threads() << endl;
+        region_file << _event_threads.size() << endl;
         // then iterate over the regions and write them out.
         for (auto const &i : global_region_indices) {
             task_identifier id = i.first;
@@ -1649,7 +1658,8 @@ namespace apex {
         metric_filename << metric_filename_prefix << my_saved_node_id;
         ofstream metric_file(metric_filename.str(), ios::out | ios::trunc );
         // first, output our number of threads.
-        metric_file << thread_instance::get_num_threads() << endl;
+        //metric_file << thread_instance::get_num_threads() << endl;
+        metric_file << _event_threads.size() << endl;
         // then iterate over the metrics and write them out.
         for (auto const &i : global_metric_indices) {
             string id = i.first;

src/apex/otf2_listener.hpp

@@ -9,7 +9,7 @@
 #include "event_listener.hpp"
 #include <otf2/otf2.h>
 #include <map>
-#include <unordered_set>
+#include <set>
 #include <string>
 #include <tuple>
 #include <memory>
@@ -29,7 +29,7 @@ namespace apex {
         std::mutex _metric_mutex;
         std::mutex _comm_mutex;
         std::mutex _event_set_mutex;
-        std::unordered_set<int> _event_threads;
+        std::set<int> _event_threads;
         /* this is a reader/writer lock. Don't close the archive
          * if other threads are writing to it. but allow concurrent
          * access from the writer threads. */

src/apex/policy_handler.cpp

@@ -71,12 +71,12 @@ namespace apex {
         this->_reset();
         if (_terminate) return true;
         if (apex_options::use_tau()) {
-            Tau_start("policy_handler::_handler");
+            tau_listener::Tau_start_wrapper("policy_handler::_handler");
         }
         periodic_event_data data;
         this->on_periodic(data);
         if (apex_options::use_tau()) {
-            Tau_stop("policy_handler::_handler");
+            tau_listener::Tau_stop_wrapper("policy_handler::_handler");
         }
         return true;
     }

src/apex/proc_read.cpp

@@ -583,7 +583,7 @@ void* proc_data_reader::read_proc(void * _ptw) {
       _initialized = true;
   }
   if (apex_options::use_tau()) {
-    Tau_start("proc_data_reader::read_proc");
+    tau_listener::Tau_start_wrapper("proc_data_reader::read_proc");
   }
   if (done) { return nullptr; }
 #ifdef APEX_HAVE_LM_SENSORS
@@ -605,7 +605,7 @@ void* proc_data_reader::read_proc(void * _ptw) {
   while(ptw->wait()) {
     if (done) break;
     if (apex_options::use_tau()) {
-      Tau_start("proc_data_reader::read_proc: main loop");
+      tau_listener::Tau_start_wrapper("proc_data_reader::read_proc: main loop");
     }
     if (apex_options::use_proc_stat()) {
         // take a reading
@@ -630,15 +630,15 @@ void* proc_data_reader::read_proc(void * _ptw) {
 #endif
 
     if (apex_options::use_tau()) {
-      Tau_stop("proc_data_reader::read_proc: main loop");
+      tau_listener::Tau_stop_wrapper("proc_data_reader::read_proc: main loop");
     }
   }
 #ifdef APEX_HAVE_LM_SENSORS
   delete(mysensors);
 #endif
 
   if (apex_options::use_tau()) {
-    Tau_stop("proc_data_reader::read_proc");
+    tau_listener::Tau_stop_wrapper("proc_data_reader::read_proc");
   }
   delete(oldData);
   thread_instance::delete_instance();

src/apex/profiler_listener.cpp

@@ -587,7 +587,7 @@ std::unordered_set<profile*> free_profiles;
    * file, or both. */
   void profiler_listener::finalize_profiles(dump_event_data &data) {
     if (apex_options::use_tau()) {
-      Tau_start("profiler_listener::finalize_profiles");
+      tau_listener::Tau_start_wrapper("profiler_listener::finalize_profiles");
     }
     // our TOTAL available time is the elapsed * the number of threads, or cores
     int num_worker_threads = thread_instance::get_num_threads();
@@ -718,7 +718,7 @@ std::unordered_set<profile*> free_profiles;
         csvfile.close();
     }
     if (apex_options::use_tau()) {
-      Tau_stop("profiler_listener::finalize_profiles");
+      tau_listener::Tau_stop_wrapper("profiler_listener::finalize_profiles");
     }
   }
 
@@ -1038,7 +1038,7 @@ node_color * get_node_color(double v,double vmin,double vmax)
       _initialized = true;
     }
     if (apex_options::use_tau()) {
-      Tau_start("profiler_listener::process_profiles");
+      tau_listener::Tau_start_wrapper("profiler_listener::process_profiles");
     }
 
     std::shared_ptr<profiler> p;
@@ -1076,7 +1076,7 @@ node_color * get_node_color(double v,double vmin,double vmax)
     while (!_done) {
         queue_signal.wait();
         if (apex_options::use_tau()) {
-            Tau_start("profiler_listener::process_profiles: main loop");
+            tau_listener::Tau_start_wrapper("profiler_listener::process_profiles: main loop");
         }
         {
             std::unique_lock<std::mutex> queue_lock(queue_mtx);
@@ -1095,7 +1095,7 @@ node_color * get_node_color(double v,double vmin,double vmax)
             }
         }
         if (apex_options::use_tau()) {
-            Tau_stop("profiler_listener::process_profiles: main loop");
+            tau_listener::Tau_stop_wrapper("profiler_listener::process_profiles: main loop");
         }
         // release the flag, and wait for the queue_signal
         consumer_task_running.clear(memory_order_release);
@@ -1109,7 +1109,7 @@ node_color * get_node_color(double v,double vmin,double vmax)
 #endif
 
     if (apex_options::use_tau()) {
-      Tau_stop("profiler_listener::process_profiles");
+      tau_listener::Tau_stop_wrapper("profiler_listener::process_profiles");
     }
   }
 
@@ -1276,11 +1276,11 @@ if (rc != 0) cout << "PAPI error! " << name << ": " << PAPI_strerror(rc) << endl
             std::unique_lock<std::mutex> queue_lock(queue_mtx);
             for (unsigned int i=0; i<allqueues.size(); ++i) {
                 if (apex_options::use_tau()) {
-                    Tau_start("profiler_listener::concurrent_cleanup");
+                    tau_listener::Tau_start_wrapper("profiler_listener::concurrent_cleanup");
                 }
                 concurrent_cleanup(i);
                 if (apex_options::use_tau()) {
-                    Tau_stop("profiler_listener::concurrent_cleanup");
+                    tau_listener::Tau_stop_wrapper("profiler_listener::concurrent_cleanup");
                 }
             }
         }
@@ -1425,7 +1425,7 @@ if (rc != 0) cout << "PAPI error! " << name << ": " << PAPI_strerror(rc) << endl
       //is_resume);
       // get the right task identifier, based on whether there are aliases
       profiler * p = new profiler(tt_ptr, is_resume);
-      p->guid = thread_instance::get_guid();
+      p->guid = tt_ptr->guid;
       thread_instance::instance().set_current_profiler(p);
 #if APEX_HAVE_PAPI
       if (num_papi_counters > 0 && !apex_options::papi_suspend()) {

src/apex/tau_dummy.cpp

@@ -8,6 +8,8 @@
 
 using namespace std;
 
+#ifdef APEX_USE_WEAK_SYMBOLS
+
 /* Weak symbols that are redefined if we use TAU at runtime */
 extern "C" {
 APEX_EXPORT APEX_WEAK_PRE int Tau_init(int, char**) APEX_WEAK_POST {
@@ -24,7 +26,7 @@ APEX_EXPORT APEX_WEAK_PRE int
 APEX_EXPORT APEX_WEAK_PRE int Tau_start(const char *) APEX_WEAK_POST {return 0;}
 APEX_EXPORT APEX_WEAK_PRE int Tau_stop(const char *) APEX_WEAK_POST {return 0;}
 APEX_EXPORT APEX_WEAK_PRE int Tau_exit(const char*) APEX_WEAK_POST {return 0;}
-APEX_EXPORT APEX_WEAK_PRE int Tau_dump(void) APEX_WEAK_POST {return 0;}
+APEX_EXPORT APEX_WEAK_PRE int Tau_dump_prefix(const char*) APEX_WEAK_POST {return 0;}
 APEX_EXPORT APEX_WEAK_PRE int Tau_set_node(int) APEX_WEAK_POST {return 0;}
 APEX_EXPORT APEX_WEAK_PRE int
     Tau_profile_exit_all_threads(void) APEX_WEAK_POST {return 0;}
@@ -35,3 +37,5 @@ APEX_EXPORT APEX_WEAK_PRE int Tau_global_stop(void) APEX_WEAK_POST {return 0;}
 APEX_EXPORT APEX_WEAK_PRE int
     Tau_trigger_context_event_thread(char*, double, int) APEX_WEAK_POST {return 0;}
 }
+
+#endif // APEX_USE_WEAK_SYMBOLS