From: Martin Quinson <martin.quinson@loria.fr>
Date: Mon, 10 Sep 2018 12:35:57 +0000 (+0200)
Subject: Improve option smpi/wtime
X-Git-Tag: v3_21~113
X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/commitdiff_plain/9fe49bf54cebc7c586e72cab932a175ed5b41ae7

Improve option smpi/wtime

- Set default value to 1ms instead of 0. This default settings may
  lead to slower simulation, but it works in more situations.
- Also apply this delay in gettimeofday() and clock_gettime()
- Improve the documentation.
---

diff --git a/doc/doxygen/options.doc b/doc/doxygen/options.doc
index 209de0b6af..200b0915a1 100644
--- a/doc/doxygen/options.doc
+++ b/doc/doxygen/options.doc
@@ -1248,28 +1248,33 @@ at least one huge page:
 Then, you can pass the option --cfg=smpi/shared-malloc-hugepage:/home/huge
 to smpirun to actually activate the huge page support in shared mallocs.
 
-@subsection options_model_smpi_wtime smpi/wtime: Inject constant times for calls to MPI_Wtime
+@subsection options_model_smpi_wtime smpi/wtime: Inject constant times for calls to MPI_Wtime, gettimeofday and clock_gettime
 
-@b Default value: 0
-
-By setting this option, you can control the amount of time a process sleeps
-when MPI_Wtime() is called; this is important, because SimGrid normally only
-advances the time while communication is happening and thus,
-MPI_Wtime will not add to the time, resulting in a deadlock if used as a
-break-condition.
+@b Default value: 1ms
 
-Here is an example:
+This option controls the amount of (simulated) time spent in calls to
+MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
+to 0, the simulated clock is not advanced in these calls, which leads
+to issue if your application contains such a loop:
 
 @code{.unparsed}
     while(MPI_Wtime() < some_time_bound) {
-        ...
+        /* some tests, with no communication nor computation */
     }
 @endcode
 
-If the time is never advanced, this loop will clearly never end as MPI_Wtime()
-always returns the same value. Hence, pass a (small) value to the smpi/wtime
-option to force a call to MPI_Wtime to advance the time as well.
-
+When the option smpi/wtime is set to 0, the time advances only on
+communications and computations, so the previous code results in an
+infinite loop: the current [simulated] time will never reach @c
+some_time_bound.  This infinite loop is avoided when that option is
+set to a small amount, as it is by default since SimGrid v3.21.
+
+Note that if your application does not contain any loop depending on
+the current time only, then setting this option to a non-zero value
+will slow down your simulations by a tiny bit: the simulation loop has
+to be broken and reset each time your code ask for the current time.
+If the simulation speed really matters to you, you can avoid this
+extra delay by setting smpi/wtime to 0.
 
 @section options_generic Configuring other aspects of SimGrid
 
diff --git a/src/smpi/internals/smpi_bench.cpp b/src/smpi/internals/smpi_bench.cpp
index 302050337c..c3d7e18893 100644
--- a/src/smpi/internals/smpi_bench.cpp
+++ b/src/smpi/internals/smpi_bench.cpp
@@ -28,8 +28,10 @@
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_bench, smpi, "Logging specific to SMPI (benchmarking)");
 
-static simgrid::config::Flag<double> smpi_wtime_sleep("smpi/wtime", "Minimum time to inject inside a call to MPI_Wtime",
-                                                      0.0);
+static simgrid::config::Flag<double>
+    smpi_wtime_sleep("smpi/wtime",
+                     "Minimum time to inject inside a call to MPI_Wtime(), gettimeofday() and clock_gettime()",
+                     1e-6 /* Documented to be 1ms */);
 
 double smpi_cpu_threshold = -1;
 double smpi_host_speed;
@@ -235,6 +237,8 @@ int smpi_gettimeofday(struct timeval* tv, struct timezone* tz)
     tv->tv_usec = static_cast<suseconds_t>((now - tv->tv_sec) * 1e6);
 #endif
   }
+  if (smpi_wtime_sleep > 0)
+    simcall_process_sleep(smpi_wtime_sleep);
   smpi_bench_begin();
   return 0;
 }
@@ -251,11 +255,28 @@ int smpi_clock_gettime(clockid_t clk_id, struct timespec* tp)
     tp->tv_sec = static_cast<time_t>(now);
     tp->tv_nsec = static_cast<long int>((now - tp->tv_sec) * 1e9);
   }
+  if (smpi_wtime_sleep > 0)
+    simcall_process_sleep(smpi_wtime_sleep);
   smpi_bench_begin();
   return 0;
 }
 #endif
 
+double smpi_mpi_wtime()
+{
+  double time;
+  if (smpi_process()->initialized() && not smpi_process()->finalized() && not smpi_process()->sampling()) {
+    smpi_bench_end();
+    time = SIMIX_get_clock();
+    if (smpi_wtime_sleep > 0)
+      simcall_process_sleep(smpi_wtime_sleep);
+    smpi_bench_begin();
+  } else {
+    time = SIMIX_get_clock();
+  }
+  return time;
+}
+
 extern double sg_surf_precision;
 unsigned long long smpi_rastro_resolution ()
 {
@@ -276,27 +297,6 @@ unsigned long long smpi_rastro_timestamp ()
   return static_cast<unsigned long long>(sec) * smpi_rastro_resolution() + pre;
 }
 
-double smpi_mpi_wtime()
-{
-  double time;
-  if (smpi_process()->initialized() && not smpi_process()->finalized() && not smpi_process()->sampling()) {
-    smpi_bench_end();
-    time = SIMIX_get_clock();
-    // to avoid deadlocks if used as a break condition, such as
-    //     while (MPI_Wtime(...) < time_limit) {
-    //       ....
-    //     }
-    // because the time will not normally advance when only calls to MPI_Wtime
-    // are made -> deadlock (MPI_Wtime never reaches the time limit)
-    if (smpi_wtime_sleep > 0)
-      simcall_process_sleep(smpi_wtime_sleep);
-    smpi_bench_begin();
-  } else {
-    time = SIMIX_get_clock();
-  }
-  return time;
-}
-
 /* ****************************** Functions related to the SMPI_SAMPLE_ macros ************************************/
 namespace {
 class SampleLocation : public std::string {