From 09212820ce67437a2ba2c7ce65c97cc0d721ca24 Mon Sep 17 00:00:00 2001
From: Faure Adrien <adrien.faure2@gmail.com>
Date: Tue, 20 Aug 2019 20:08:28 +0200
Subject: [PATCH] execute_flops now logs compute I created a internal function
 that execute flops without tracing for specific needs

---
 include/smpi/smpi.h                |  1 +
 src/smpi/internals/smpi_bench.cpp  | 14 ++++++++++++--
 src/smpi/internals/smpi_replay.cpp | 10 ++++------
 3 files changed, 17 insertions(+), 8 deletions(-)

diff --git a/include/smpi/smpi.h b/include/smpi/smpi.h
index c4e27b890a..f892ae76a8 100644
--- a/include/smpi/smpi.h
+++ b/include/smpi/smpi.h
@@ -979,6 +979,7 @@ XBT_PUBLIC MPI_Info smpi_process_info_env();
 XBT_PUBLIC void* smpi_process_get_user_data();
 XBT_PUBLIC void smpi_process_set_user_data(void*);
 
+XBT_PRIVATE void private_execute_flops(double flops);
 XBT_PUBLIC void smpi_execute_flops(double flops);
 XBT_PUBLIC void smpi_execute(double duration);
 XBT_PUBLIC void smpi_execute_benched(double duration);
diff --git a/src/smpi/internals/smpi_bench.cpp b/src/smpi/internals/smpi_bench.cpp
index ac32fa2ee6..8bf464a586 100644
--- a/src/smpi/internals/smpi_bench.cpp
+++ b/src/smpi/internals/smpi_bench.cpp
@@ -38,7 +38,8 @@ double smpi_host_speed;
 SharedMallocType smpi_cfg_shared_malloc = SharedMallocType::GLOBAL;
 double smpi_total_benched_time = 0;
 
-void smpi_execute_flops(double flops) {
+// Private execute_flops used by smpi_execute and spmi_execute benched
+void private_execute_flops(double flops) {
   xbt_assert(flops >= 0, "You're trying to execute a negative amount of flops (%f)!", flops);
   XBT_DEBUG("Handle real computation time: %f flops", flops);
   simgrid::s4u::this_actor::exec_init(flops)
@@ -49,6 +50,15 @@ void smpi_execute_flops(double flops) {
   smpi_switch_data_segment(simgrid::s4u::Actor::self());
 }
 
+void smpi_execute_flops(double flops) {
+  int rank     = simgrid::s4u::this_actor::get_pid();
+  TRACE_smpi_computing_in(rank, flops);
+
+  private_execute_flops(flops);
+
+  TRACE_smpi_computing_out(rank);
+}
+
 void smpi_execute(double duration)
 {
   if (duration >= smpi_cpu_threshold) {
@@ -57,7 +67,7 @@ void smpi_execute(double duration)
     int rank     = simgrid::s4u::this_actor::get_pid();
     TRACE_smpi_computing_in(rank, flops);
 
-    smpi_execute_flops(flops);
+    private_execute_flops(flops);
 
     TRACE_smpi_computing_out(rank);
 
diff --git a/src/smpi/internals/smpi_replay.cpp b/src/smpi/internals/smpi_replay.cpp
index 464ba2cfc3..1aad150545 100644
--- a/src/smpi/internals/smpi_replay.cpp
+++ b/src/smpi/internals/smpi_replay.cpp
@@ -477,9 +477,7 @@ void RecvAction::kernel(simgrid::xbt::ReplayAction&)
 
 void ComputeAction::kernel(simgrid::xbt::ReplayAction&)
 {
-  TRACE_smpi_computing_in(my_proc_id, args.flops);
   smpi_execute_flops(args.flops);
-  TRACE_smpi_computing_out(my_proc_id);
 }
 
 void TestAction::kernel(simgrid::xbt::ReplayAction&)
@@ -575,7 +573,7 @@ void ReduceAction::kernel(simgrid::xbt::ReplayAction&)
 
   Colls::reduce(send_buffer(args.comm_size * args.datatype1->size()),
       recv_buffer(args.comm_size * args.datatype1->size()), args.comm_size, args.datatype1, MPI_OP_NULL, args.root, MPI_COMM_WORLD);
-  smpi_execute_flops(args.comp_size);
+  private_execute_flops(args.comp_size);
 
   TRACE_smpi_comm_out(my_proc_id);
 }
@@ -587,7 +585,7 @@ void AllReduceAction::kernel(simgrid::xbt::ReplayAction&)
 
   Colls::allreduce(send_buffer(args.comm_size * args.datatype1->size()),
       recv_buffer(args.comm_size * args.datatype1->size()), args.comm_size, args.datatype1, MPI_OP_NULL, MPI_COMM_WORLD);
-  smpi_execute_flops(args.comp_size);
+  private_execute_flops(args.comp_size);
 
   TRACE_smpi_comm_out(my_proc_id);
 }
@@ -684,7 +682,7 @@ void ReduceScatterAction::kernel(simgrid::xbt::ReplayAction&)
       recv_buffer(args.recv_size_sum * args.datatype1->size()), args.recvcounts->data(),
       args.datatype1, MPI_OP_NULL, MPI_COMM_WORLD);
 
-  smpi_execute_flops(args.comp_size);
+  private_execute_flops(args.comp_size);
   TRACE_smpi_comm_out(my_proc_id);
 }
 
@@ -752,7 +750,7 @@ void smpi_replay_init(const char* instance_id, int rank, double start_delay_flop
   //if we have a delayed start, sleep here.
   if (start_delay_flops > 0) {
     XBT_VERB("Delayed start for instance - Sleeping for %f flops ", start_delay_flops);
-    smpi_execute_flops(start_delay_flops);
+    private_execute_flops(start_delay_flops);
   } else {
     // Wait for the other actors to initialize also
     simgrid::s4u::this_actor::yield();
-- 
2.20.1