Change way replay is handled, to allow cohabitation between replay and "classic"...

[simgrid.git] / src / smpi / smpi_replay.c

diff --git a/src/smpi/smpi_replay.c b/src/smpi/smpi_replay.c
index 11f9d18..261aec6 100644 (file)
--- a/src/smpi/smpi_replay.c
+++ b/src/smpi/smpi_replay.c
@@ -18,6 +18,11 @@ xbt_dynar_t *reqq = NULL;
 MPI_Datatype MPI_DEFAULT_TYPE;
 MPI_Datatype MPI_CURRENT_TYPE;
 
+static int sendbuffer_size=0;
+char* sendbuffer=NULL;
+static int recvbuffer_size=0;
+char* recvbuffer=NULL;
+
 static void log_timed_action (const char *const *action, double clock){
   if (XBT_LOG_ISENABLED(smpi_replay, xbt_log_priority_verbose)){
     char *name = xbt_str_join_array(action, " ");
@@ -26,6 +31,32 @@ static void log_timed_action (const char *const *action, double clock){
   }
 }
 
+//allocate a single buffer for all sends, growing it if needed
+void* smpi_get_tmp_sendbuffer(int size){
+  if (!smpi_process_get_replaying())
+       return xbt_malloc(size);
+  if (sendbuffer_size<size){
+    sendbuffer=xbt_realloc(sendbuffer,size);
+    sendbuffer_size=size;
+  }
+  return sendbuffer;
+}
+//allocate a single buffer for all recv
+void* smpi_get_tmp_recvbuffer(int size){
+  if (!smpi_process_get_replaying())
+       return xbt_malloc(size);
+  if (recvbuffer_size<size){
+    recvbuffer=xbt_realloc(recvbuffer,size);
+    recvbuffer_size=size;
+  }
+  return sendbuffer;
+}
+
+void smpi_free_tmp_buffer(void* buf){
+  if (!smpi_process_get_replaying())
+    xbt_free(buf);
+}
+
 /* Helper function */
 static double parse_double(const char *string)
 {
@@ -591,8 +622,8 @@ static void action_allToAll(const char *const *action) {
     MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE;
     MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
   }
-  void *send = xbt_malloc(send_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE));  
-  void *recv = xbt_malloc(recv_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE2));  
+  void *send = smpi_get_tmp_sendbuffer(send_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE));
+  void *recv = smpi_get_tmp_recvbuffer(recv_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE2));
 
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
@@ -613,8 +644,7 @@ static void action_allToAll(const char *const *action) {
 #endif
 
   log_timed_action (action, clock);
-  xbt_free(send);
-  xbt_free(recv);
+
 }
 
 
@@ -643,14 +673,14 @@ static void action_gather(const char *const *action) {
     MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE;
     MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
   }
-  void *send = xbt_malloc(send_size* smpi_datatype_size(MPI_CURRENT_TYPE));
+  void *send = smpi_get_tmp_sendbuffer(send_size* smpi_datatype_size(MPI_CURRENT_TYPE));
   void *recv = NULL;
 
   int root=atoi(action[4]);
   int rank = smpi_process_index();
 
   if(rank==root)
-    recv = xbt_malloc(recv_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE2));
+    recv = smpi_get_tmp_recvbuffer(recv_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE2));
 
 #ifdef HAVE_TRACING
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
@@ -672,8 +702,7 @@ static void action_gather(const char *const *action) {
 #endif
 
   log_timed_action (action, clock);
-  xbt_free(send);
-  xbt_free(recv);
+
 }
 
 
@@ -706,7 +735,7 @@ static void action_gatherv(const char *const *action) {
     MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE;
     MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
   }
-  void *send = xbt_malloc(send_size* smpi_datatype_size(MPI_CURRENT_TYPE));
+  void *send = smpi_get_tmp_sendbuffer(send_size* smpi_datatype_size(MPI_CURRENT_TYPE));
   void *recv = NULL;
   for(i=0;i<comm_size;i++) {
     recvcounts[i] = atoi(action[i+3]);
@@ -718,7 +747,7 @@ static void action_gatherv(const char *const *action) {
   int rank = smpi_process_index();
 
   if(rank==root)
-    recv = xbt_malloc(recv_sum* smpi_datatype_size(MPI_CURRENT_TYPE2));
+    recv = smpi_get_tmp_recvbuffer(recv_sum* smpi_datatype_size(MPI_CURRENT_TYPE2));
 
 #ifdef HAVE_TRACING
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
@@ -744,8 +773,6 @@ smpi_mpi_gatherv(send, send_size, MPI_CURRENT_TYPE,
 
   log_timed_action (action, clock);
   xbt_free(recvcounts);
-  xbt_free(send);
-  xbt_free(recv);
   xbt_free(disps);
 
 }
@@ -844,13 +871,13 @@ static void action_allgatherv(const char *const *action) {
     MPI_CURRENT_TYPE = MPI_DEFAULT_TYPE;
     MPI_CURRENT_TYPE2 = MPI_DEFAULT_TYPE;    
   }
-  void *sendbuf = xbt_malloc(sendcount* smpi_datatype_size(MPI_CURRENT_TYPE));    
+  void *sendbuf = smpi_get_tmp_sendbuffer(sendcount* smpi_datatype_size(MPI_CURRENT_TYPE));
 
   for(i=0;i<comm_size;i++) {
     recvcounts[i] = atoi(action[i+3]);
     recv_sum=recv_sum+recvcounts[i];
   }
-  void *recvbuf = xbt_malloc(recv_sum* smpi_datatype_size(MPI_CURRENT_TYPE2));  
+  void *recvbuf = smpi_get_tmp_recvbuffer(recv_sum* smpi_datatype_size(MPI_CURRENT_TYPE2));
 
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
@@ -874,8 +901,6 @@ static void action_allgatherv(const char *const *action) {
 #endif
 
   log_timed_action (action, clock);
-  xbt_free(sendbuf);
-  xbt_free(recvbuf);
   xbt_free(recvcounts);
   xbt_free(disps);
 }
@@ -918,8 +943,8 @@ static void action_allToAllv(const char *const *action) {
       MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
   }
 
-  void *sendbuf = xbt_malloc(send_buf_size* smpi_datatype_size(MPI_CURRENT_TYPE));  
-  void *recvbuf = xbt_malloc(recv_buf_size* smpi_datatype_size(MPI_CURRENT_TYPE2));  
+  void *sendbuf = smpi_get_tmp_sendbuffer(send_buf_size* smpi_datatype_size(MPI_CURRENT_TYPE));
+  void *recvbuf  = smpi_get_tmp_recvbuffer(recv_buf_size* smpi_datatype_size(MPI_CURRENT_TYPE2));
 
   for(i=0;i<comm_size;i++) {
     sendcounts[i] = atoi(action[i+3]);
@@ -954,8 +979,6 @@ static void action_allToAllv(const char *const *action) {
 #endif
 
   log_timed_action (action, clock);
-  xbt_free(sendbuf);
-  xbt_free(recvbuf);
   xbt_free(sendcounts);
   xbt_free(recvcounts);
   xbt_free(senddisps);
@@ -965,6 +988,7 @@ static void action_allToAllv(const char *const *action) {
 void smpi_replay_init(int *argc, char***argv){
   smpi_process_init(argc, argv);
   smpi_process_mark_as_initialized();
+  smpi_process_set_replaying(1);
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
   TRACE_smpi_init(rank);
@@ -1001,7 +1025,16 @@ void smpi_replay_init(int *argc, char***argv){
     xbt_replay_action_register("reduceScatter",  action_reducescatter);
     xbt_replay_action_register("compute",    action_compute);
   }
-
+  
+  //if we have a delayed start, sleep here.
+  if(*argc>2){
+    char *endptr;
+    double value = strtod((*argv)[2], &endptr);
+    if (*endptr != '\0')
+      THROWF(unknown_error, 0, "%s is not a double", (*argv)[2]);
+    XBT_VERB("Delayed start for instance - Sleeping for %f flops ",value );
+    smpi_execute_flops(value);
+  }
   xbt_replay_action_runner(*argc, *argv);
 }
 
@@ -1031,6 +1064,8 @@ int smpi_replay_finalize(){
     sim_time = smpi_process_simulated_elapsed();
     XBT_INFO("Simulation time %f", sim_time);
     _xbt_replay_action_exit();
+    xbt_free(sendbuffer);
+    xbt_free(recvbuffer);
     xbt_free(reqq);
     reqq = NULL;
   }