X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/a1234c9c887de28ed21e81528886c20456c7b6cf..b9945098e2ba4e7b639e0c2245835d44800c1b29:/src/smpi/internals/smpi_bench.cpp

diff --git a/src/smpi/internals/smpi_bench.cpp b/src/smpi/internals/smpi_bench.cpp
index 9fbc6b0dc5..af6cf7817b 100644
--- a/src/smpi/internals/smpi_bench.cpp
+++ b/src/smpi/internals/smpi_bench.cpp
@@ -11,7 +11,6 @@
 #include "smpi_comm.hpp"
 #include "src/internal_config.h"
 #include "src/mc/mc_replay.hpp"
-#include "src/simix/ActorImpl.hpp"
 #include "xbt/config.hpp"
 
 #include "src/smpi/include/smpi_actor.hpp"
@@ -57,7 +56,7 @@ void smpi_execute_flops(double flops) {
   XBT_DEBUG("Handle real computation time: %f flops", flops);
   simgrid::s4u::this_actor::exec_init(flops)
       ->set_name("computation")
-      ->set_tracing_category(TRACE_internal_smpi_get_category())
+      ->set_tracing_category(smpi_process()->get_tracing_category())
       ->start()
       ->wait();
   smpi_switch_data_segment(simgrid::s4u::Actor::self());
@@ -337,9 +336,10 @@ std::unordered_map<SampleLocation, LocalData, std::hash<std::string>> samples;
 void smpi_sample_1(int global, const char *file, int line, int iters, double threshold)
 {
   SampleLocation loc(global, file, line);
-
-  smpi_bench_end();     /* Take time from previous, unrelated computation into account */
-  smpi_process()->set_sampling(1);
+  if (not smpi_process()->sampling()) { /* Only at first call when benchmarking, skip for next ones */
+    smpi_bench_end();     /* Take time from previous, unrelated computation into account */
+    smpi_process()->set_sampling(1);
+  }
 
   auto insert = samples.emplace(loc, LocalData{
                                          threshold, // threshold
@@ -372,12 +372,11 @@ void smpi_sample_1(int global, const char *file, int line, int iters, double thr
   }
 }
 
-int smpi_sample_2(int global, const char *file, int line)
+int smpi_sample_2(int global, const char *file, int line, int iter_count)
 {
   SampleLocation loc(global, file, line);
-  int res;
 
-  XBT_DEBUG("sample2 %s", loc.c_str());
+  XBT_DEBUG("sample2 %s %d", loc.c_str(), iter_count);
   auto sample = samples.find(loc);
   if (sample == samples.end())
     xbt_die("Y U NO use SMPI_SAMPLE_* macros? Stop messing directly with smpi_sample_* functions!");
@@ -385,24 +384,28 @@ int smpi_sample_2(int global, const char *file, int line)
 
   if (data.benching) {
     // we need to run a new bench
-    XBT_DEBUG("benchmarking: count:%d iter:%d stderr:%f thres:%f; mean:%f",
-              data.count, data.iters, data.relstderr, data.threshold, data.mean);
-    res = 1;
+    XBT_DEBUG("benchmarking: count:%d iter:%d stderr:%f thres:%f; mean:%f; total:%f",
+              data.count, data.iters, data.relstderr, data.threshold, data.mean, data.sum);
+    smpi_bench_begin();
   } else {
     // Enough data, no more bench (either we got enough data from previous visits to this benched nest, or we just
     //ran one bench and need to bail out now that our job is done). Just sleep instead
-    if (not data.need_more_benchs())
-      XBT_DEBUG("No benchmark (either no need, or just ran one): count >= iter (%d >= %d) or stderr<thres (%f<=%f)."
-              " apply the %fs delay instead",
-              data.count, data.iters, data.relstderr, data.threshold, data.mean);
-    else
-      XBT_DEBUG("Skipping - Benchmark already performed");
-    smpi_execute(data.mean);
-    smpi_process()->set_sampling(0);
-    res = 0; // prepare to capture future, unrelated computations
+    if (not data.need_more_benchs()){
+      XBT_DEBUG("No benchmark (either no need, or just ran one): count >= iter (%d >= %d) or stderr<thres (%f<=%f). "
+              "Mean is %f, will be injected %d times",
+              data.count, data.iters, data.relstderr, data.threshold, data.mean, iter_count);
+              
+      //we ended benchmarking, let's inject all the time, now, and fast forward out of the loop.
+      smpi_process()->set_sampling(0);
+      smpi_execute(data.mean*iter_count);
+      smpi_bench_begin();
+      return 0;
+    } else {
+      XBT_DEBUG("Skipping - Benchmark already performed - accumulating time");
+      xbt_os_threadtimer_start(smpi_process()->timer());
+    }
   }
-  smpi_bench_begin();
-  return res;
+  return 1;
 }
 
 void smpi_sample_3(int global, const char *file, int line)
@@ -429,10 +432,7 @@ void smpi_sample_3(int global, const char *file, int line)
   double n       = static_cast<double>(data.count);
   data.mean      = data.sum / n;
   data.relstderr = sqrt((data.sum_pow2 / n - data.mean * data.mean) / n) / data.mean;
-  if (data.need_more_benchs()) {
-    data.mean = period; // Still in benching process; We want sample_2 to simulate the exact time of this loop
-    // occurrence before leaving, not the mean over the history
-  }
+
   XBT_DEBUG("Average mean after %d steps is %f, relative standard error is %f (sample was %f)",
             data.count, data.mean, data.relstderr, period);
 
@@ -440,6 +440,25 @@ void smpi_sample_3(int global, const char *file, int line)
   data.benching = false;
 }
 
+int smpi_sample_exit(int global, const char *file, int line, int iter_count){
+  if (smpi_process()->sampling()){
+    SampleLocation loc(global, file, line);
+
+    XBT_DEBUG("sample exit %s", loc.c_str());
+    auto sample = samples.find(loc);
+    if (sample == samples.end())
+      xbt_die("Y U NO use SMPI_SAMPLE_* macros? Stop messing directly with smpi_sample_* functions!");
+    LocalData& data = sample->second;
+  
+    if (smpi_process()->sampling()){//end of loop, but still sampling needed
+        smpi_process()->set_sampling(0);
+        smpi_execute(data.mean*iter_count);
+        smpi_bench_begin();
+    }
+  }
+  return 0;
+}
+
 smpi_trace_call_location_t* smpi_trace_get_call_location()
 {
   return smpi_process()->call_location();