X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/b531f578ec2baa41a140c677217934de645723f8..762107b960e67cbb9713d0750147372d2178ccbd:/src/smpi/smpi_replay.c

diff --git a/src/smpi/smpi_replay.c b/src/smpi/smpi_replay.c
index 11f9d1846b..6cbe6ae2d7 100644
--- 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)
 {
@@ -320,15 +351,19 @@ static void action_test(const char *const *action){
   int flag = TRUE;
 
   request = xbt_dynar_pop_as(reqq[smpi_process_index()],MPI_Request);
-  xbt_assert(request != NULL, "found null request in reqq");
-
+  //if request is null here, this may mean that a previous test has succeeded 
+  //Different times in traced application and replayed version may lead to this 
+  //In this case, ignore the extra calls.
+  if(request){
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
   extra->type=TRACING_TEST;
   TRACE_smpi_testing_in(rank, extra);
 #endif
-  flag = smpi_mpi_test(&request, &status);
+
+    flag = smpi_mpi_test(&request, &status);
+
   XBT_DEBUG("MPI_Test result: %d", flag);
   /* push back request in dynar to be caught by a subsequent wait. if the test
    * did succeed, the request is now NULL.
@@ -338,7 +373,7 @@ static void action_test(const char *const *action){
 #ifdef HAVE_TRACING
   TRACE_smpi_testing_out(rank);
 #endif
-
+  }
   log_timed_action (action, clock);
 }
 
@@ -452,7 +487,9 @@ static void action_waitall(const char *const *action){
    xbt_dynar_free(&recvs);
   #endif
 
-   xbt_dynar_free_container(&(reqq[smpi_process_index()]));
+   int freedrank=smpi_process_index();
+   xbt_dynar_free_container(&(reqq[freedrank]));
+   reqq[freedrank]=xbt_dynar_new(sizeof(MPI_Request),&xbt_free_ref);
   }
   log_timed_action (action, clock);
 }
@@ -504,12 +541,11 @@ static void action_bcast(const char *const *action)
   TRACE_smpi_collective_in(rank, root_traced, __FUNCTION__, extra);
 
 #endif
-
-  mpi_coll_bcast_fun(NULL, size, MPI_CURRENT_TYPE, root, MPI_COMM_WORLD);
+    void *sendbuf = smpi_get_tmp_sendbuffer(size* smpi_datatype_size(MPI_CURRENT_TYPE));
+  mpi_coll_bcast_fun(sendbuf, size, MPI_CURRENT_TYPE, root, MPI_COMM_WORLD);
 #ifdef HAVE_TRACING
   TRACE_smpi_collective_out(rank, root_traced, __FUNCTION__);
 #endif
-
   log_timed_action (action, clock);
 }
 
@@ -527,6 +563,8 @@ static void action_reduce(const char *const *action)
       MPI_CURRENT_TYPE=decode_datatype(action[5]);
     }
   }
+  
+  
 
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
@@ -540,12 +578,13 @@ static void action_reduce(const char *const *action)
 
   TRACE_smpi_collective_in(rank, root_traced, __FUNCTION__,extra);
 #endif
-   mpi_coll_reduce_fun(NULL, NULL, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, root, MPI_COMM_WORLD);
+    void *recvbuf = smpi_get_tmp_sendbuffer(comm_size* smpi_datatype_size(MPI_CURRENT_TYPE));
+    void *sendbuf = smpi_get_tmp_sendbuffer(comm_size* smpi_datatype_size(MPI_CURRENT_TYPE));
+   mpi_coll_reduce_fun(sendbuf, recvbuf, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, root, MPI_COMM_WORLD);
    smpi_execute_flops(comp_size);
 #ifdef HAVE_TRACING
   TRACE_smpi_collective_out(rank, root_traced, __FUNCTION__);
 #endif
-
   log_timed_action (action, clock);
 }
 
@@ -567,12 +606,13 @@ static void action_allReduce(const char *const *action) {
 
   TRACE_smpi_collective_in(rank, -1, __FUNCTION__,extra);
 #endif
-  mpi_coll_allreduce_fun(NULL, NULL, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, MPI_COMM_WORLD);
+    void *recvbuf = smpi_get_tmp_sendbuffer(comm_size* smpi_datatype_size(MPI_CURRENT_TYPE));
+    void *sendbuf = smpi_get_tmp_sendbuffer(comm_size* smpi_datatype_size(MPI_CURRENT_TYPE));
+  mpi_coll_allreduce_fun(sendbuf, recvbuf, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, MPI_COMM_WORLD);
   smpi_execute_flops(comp_size);
 #ifdef HAVE_TRACING
   TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
 #endif
-
   log_timed_action (action, clock);
 }
 
@@ -591,8 +631,8 @@ 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();
@@ -611,10 +651,8 @@ static void action_allToAll(const char *const *action) {
 #ifdef HAVE_TRACING
   TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
 #endif
-
   log_timed_action (action, clock);
-  xbt_free(send);
-  xbt_free(recv);
+
 }
 
 
@@ -643,14 +681,15 @@ 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();
+  int root=0;
+  if(action[4])
+    root=atoi(action[4]);
+  int rank = smpi_comm_rank(MPI_COMM_WORLD);
 
   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);
@@ -661,19 +700,17 @@ static void action_gather(const char *const *action) {
   extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE);
   extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2);
 
-  TRACE_smpi_collective_in(rank, root, __FUNCTION__, extra);
+  TRACE_smpi_collective_in(smpi_process_index(), root, __FUNCTION__, extra);
 #endif
   mpi_coll_gather_fun(send, send_size, MPI_CURRENT_TYPE,
                 recv, recv_size, MPI_CURRENT_TYPE2,
                 root, MPI_COMM_WORLD);
 
 #ifdef HAVE_TRACING
-  TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+  TRACE_smpi_collective_out(smpi_process_index(), -1, __FUNCTION__);
 #endif
-
   log_timed_action (action, clock);
-  xbt_free(send);
-  xbt_free(recv);
+
 }
 
 
@@ -706,7 +743,7 @@ 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]);
@@ -715,10 +752,10 @@ static void action_gatherv(const char *const *action) {
   }
 
   int root=atoi(action[3+comm_size]);
-  int rank = smpi_process_index();
+  int rank = smpi_comm_rank(MPI_COMM_WORLD);;
 
   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);
@@ -732,22 +769,19 @@ static void action_gatherv(const char *const *action) {
   extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE);
   extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2);
 
-  TRACE_smpi_collective_in(rank, root, __FUNCTION__, extra);
+  TRACE_smpi_collective_in(smpi_process_index(), root, __FUNCTION__, extra);
 #endif
 smpi_mpi_gatherv(send, send_size, MPI_CURRENT_TYPE,
                 recv, recvcounts, disps, MPI_CURRENT_TYPE2,
                 root, MPI_COMM_WORLD);
 
 #ifdef HAVE_TRACING
-  TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+  TRACE_smpi_collective_out(smpi_process_index(), -1, __FUNCTION__);
 #endif
 
   log_timed_action (action, clock);
   xbt_free(recvcounts);
-  xbt_free(send);
-  xbt_free(recv);
   xbt_free(disps);
-
 }
 
 static void action_reducescatter(const char *const *action) {
@@ -773,7 +807,7 @@ static void action_reducescatter(const char *const *action) {
   int *disps = xbt_new0(int, comm_size);  
   int i=0;
   int rank = smpi_process_index();
-
+  int size = 0;
   if(action[3+comm_size])
     MPI_CURRENT_TYPE=decode_datatype(action[3+comm_size]);
   else
@@ -782,6 +816,7 @@ static void action_reducescatter(const char *const *action) {
   for(i=0;i<comm_size;i++) {
     recvcounts[i] = atoi(action[i+2]);
     disps[i] = 0;
+    size+=recvcounts[i];
   }
 
 #ifdef HAVE_TRACING
@@ -795,10 +830,12 @@ static void action_reducescatter(const char *const *action) {
   extra->comp_size = comp_size;
   extra->num_processes = comm_size;
 
-
   TRACE_smpi_collective_in(rank, -1, __FUNCTION__,extra);
 #endif
-   mpi_coll_reduce_scatter_fun(NULL, NULL, recvcounts, MPI_CURRENT_TYPE, MPI_OP_NULL,
+  void *sendbuf = smpi_get_tmp_sendbuffer(size* smpi_datatype_size(MPI_CURRENT_TYPE));
+   void *recvbuf = smpi_get_tmp_recvbuffer(size* smpi_datatype_size(MPI_CURRENT_TYPE));
+   
+   mpi_coll_reduce_scatter_fun(sendbuf, recvbuf, recvcounts, MPI_CURRENT_TYPE, MPI_OP_NULL,
        MPI_COMM_WORLD);
    smpi_execute_flops(comp_size);
 
@@ -844,13 +881,13 @@ 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 +911,6 @@ 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 +953,8 @@ 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 +989,6 @@ 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 +998,7 @@ 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 +1035,16 @@ 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);
 }
 
@@ -1023,18 +1066,25 @@ int smpi_replay_finalize(){
     active_processes--;
   }
 
-  xbt_dynar_free_container(&(reqq[smpi_process_index()]));
-
   if(!active_processes){
     /* Last process alive speaking */
     /* end the simulated timer */
     sim_time = smpi_process_simulated_elapsed();
+  }
+  
+
+  xbt_dynar_free_container(&(reqq[smpi_process_index()]));
+
+  if(!active_processes){
     XBT_INFO("Simulation time %f", sim_time);
     _xbt_replay_action_exit();
+    xbt_free(sendbuffer);
+    xbt_free(recvbuffer);
     xbt_free(reqq);
     reqq = NULL;
   }
-  mpi_coll_barrier_fun(MPI_COMM_WORLD);
+  
+
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);