Add an option to have a sleep time before the start of a replay instance

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

diff --git a/src/smpi/smpi_replay.c b/src/smpi/smpi_replay.c
index 9e00ae97bd7737c4caa2aab94677556523701848..bf7fc596dbbff44fdb5bc67394f11c691bae5bc9 100644 (file)
--- a/src/smpi/smpi_replay.c
+++ b/src/smpi/smpi_replay.c
@@ -151,7 +151,7 @@ static void action_compute(const char *const *action)
   double clock = smpi_process_simulated_elapsed();
   double flops= parse_double(action[2]);
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
   extra->type=TRACING_COMPUTING;
   extra->comp_size=flops;
@@ -178,7 +178,7 @@ static void action_send(const char *const *action)
   }
 
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
 
   int dst_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), to);
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
@@ -212,7 +212,7 @@ static void action_Isend(const char *const *action)
   else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE;
 
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
   int dst_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), to);
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
   extra->type = TRACING_ISEND;
@@ -231,7 +231,7 @@ static void action_Isend(const char *const *action)
   request->send = 1;
 #endif
 
-  xbt_dynar_push(reqq[smpi_comm_rank(MPI_COMM_WORLD)],&request);
+  xbt_dynar_push(reqq[smpi_process_index()],&request);
 
   log_timed_action (action, clock);
 }
@@ -246,7 +246,7 @@ static void action_recv(const char *const *action) {
   else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE;
 
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
   int src_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), from);
 
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
@@ -285,7 +285,7 @@ static void action_Irecv(const char *const *action)
   else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE;
 
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
   int src_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), from);
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
   extra->type = TRACING_IRECV;
@@ -308,7 +308,36 @@ static void action_Irecv(const char *const *action)
   TRACE_smpi_ptp_out(rank, src_traced, rank, __FUNCTION__);
   request->recv = 1;
 #endif
-  xbt_dynar_push(reqq[smpi_comm_rank(MPI_COMM_WORLD)],&request);
+  xbt_dynar_push(reqq[smpi_process_index()],&request);
+
+  log_timed_action (action, clock);
+}
+
+static void action_test(const char *const *action){
+  double clock = smpi_process_simulated_elapsed();
+  MPI_Request request;
+  MPI_Status status;
+  int flag = TRUE;
+
+  request = xbt_dynar_pop_as(reqq[smpi_process_index()],MPI_Request);
+  xbt_assert(request != NULL, "found null request in reqq");
+
+#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);
+  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.
+   */
+  xbt_dynar_push_as(reqq[smpi_process_index()],MPI_Request, request);
+
+#ifdef HAVE_TRACING
+  TRACE_smpi_testing_out(rank);
+#endif
 
   log_timed_action (action, clock);
 }
@@ -318,11 +347,18 @@ static void action_wait(const char *const *action){
   MPI_Request request;
   MPI_Status status;
 
-  xbt_assert(xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)]),
+  xbt_assert(xbt_dynar_length(reqq[smpi_process_index()]),
       "action wait not preceded by any irecv or isend: %s",
       xbt_str_join_array(action," "));
-  request = xbt_dynar_pop_as(reqq[smpi_comm_rank(MPI_COMM_WORLD)],MPI_Request);
-  xbt_assert(request != NULL, "found null request in reqq");
+  request = xbt_dynar_pop_as(reqq[smpi_process_index()],MPI_Request);
+
+  if (!request){
+    /* Assuming that the trace is well formed, this mean the comm might have
+     * been caught by a MPI_test. Then just return.
+     */
+    return;
+  }
+
 #ifdef HAVE_TRACING
   int rank = request->comm != MPI_COMM_NULL
       ? smpi_comm_rank(request->comm)
@@ -352,7 +388,7 @@ static void action_waitall(const char *const *action){
   int count_requests=0;
   unsigned int i=0;
 
-  count_requests=xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)]);
+  count_requests=xbt_dynar_length(reqq[smpi_process_index()]);
 
   if (count_requests>0) {
     MPI_Request requests[count_requests];
@@ -360,7 +396,7 @@ static void action_waitall(const char *const *action){
 
     /*  The reqq is an array of dynars. Its index corresponds to the rank.
      Thus each rank saves its own requests to the array request. */
-    xbt_dynar_foreach(reqq[smpi_comm_rank(MPI_COMM_WORLD)],i,requests[i]); 
+    xbt_dynar_foreach(reqq[smpi_process_index()],i,requests[i]); 
 
   #ifdef HAVE_TRACING
    //save information from requests
@@ -416,7 +452,7 @@ static void action_waitall(const char *const *action){
    xbt_dynar_free(&recvs);
   #endif
 
-   xbt_dynar_free_container(&(reqq[smpi_comm_rank(MPI_COMM_WORLD)]));
+   xbt_dynar_free_container(&(reqq[smpi_process_index()]));
   }
   log_timed_action (action, clock);
 }
@@ -424,12 +460,12 @@ static void action_waitall(const char *const *action){
 static void action_barrier(const char *const *action){
   double clock = smpi_process_simulated_elapsed();
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
   extra->type = TRACING_BARRIER;
   TRACE_smpi_collective_in(rank, -1, __FUNCTION__, extra);
 #endif
-  smpi_mpi_barrier(MPI_COMM_WORLD);
+  mpi_coll_barrier_fun(MPI_COMM_WORLD);
 #ifdef HAVE_TRACING
   TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
 #endif
@@ -457,7 +493,7 @@ static void action_bcast(const char *const *action)
   }
 
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
   int root_traced = smpi_group_index(smpi_comm_group(MPI_COMM_WORLD), root);
 
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
@@ -493,7 +529,7 @@ static void action_reduce(const char *const *action)
   }
 
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
   int root_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), root);
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
   extra->type = TRACING_REDUCE;
@@ -522,7 +558,7 @@ static void action_allReduce(const char *const *action) {
 
   double clock = smpi_process_simulated_elapsed();
 #ifdef HAVE_TRACING
-  int rank = smpi_comm_rank(MPI_COMM_WORLD);
+  int rank = smpi_process_index();
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
   extra->type = TRACING_ALLREDUCE;
   extra->send_size = comm_size;
@@ -531,9 +567,8 @@ static void action_allReduce(const char *const *action) {
 
   TRACE_smpi_collective_in(rank, -1, __FUNCTION__,extra);
 #endif
-  mpi_coll_reduce_fun(NULL, NULL, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, 0, MPI_COMM_WORLD);
+  mpi_coll_allreduce_fun(NULL, NULL, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, MPI_COMM_WORLD);
   smpi_execute_flops(comp_size);
-  mpi_coll_bcast_fun(NULL, comm_size, MPI_CURRENT_TYPE, 0, MPI_COMM_WORLD);
 #ifdef HAVE_TRACING
   TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
 #endif
@@ -556,8 +591,8 @@ static void action_allToAll(const char *const *action) {
     MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE;
     MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
   }
-  void *send = calloc(send_size*comm_size, smpi_datatype_size(MPI_CURRENT_TYPE));  
-  void *recv = calloc(recv_size*comm_size, smpi_datatype_size(MPI_CURRENT_TYPE2));  
+  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));  
 
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
@@ -608,14 +643,14 @@ static void action_gather(const char *const *action) {
     MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE;
     MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
   }
-  void *send = calloc(send_size, smpi_datatype_size(MPI_CURRENT_TYPE));
+  void *send = xbt_malloc(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 = calloc(recv_size*comm_size, smpi_datatype_size(MPI_CURRENT_TYPE2));
+    recv = xbt_malloc(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);
@@ -628,7 +663,7 @@ static void action_gather(const char *const *action) {
 
   TRACE_smpi_collective_in(rank, root, __FUNCTION__, extra);
 #endif
-smpi_mpi_gather(send, send_size, MPI_CURRENT_TYPE,
+  mpi_coll_gather_fun(send, send_size, MPI_CURRENT_TYPE,
                 recv, recv_size, MPI_CURRENT_TYPE2,
                 root, MPI_COMM_WORLD);
 
@@ -671,7 +706,7 @@ static void action_gatherv(const char *const *action) {
     MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE;
     MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
   }
-  void *send = calloc(send_size, smpi_datatype_size(MPI_CURRENT_TYPE));
+  void *send = xbt_malloc(send_size* smpi_datatype_size(MPI_CURRENT_TYPE));
   void *recv = NULL;
   for(i=0;i<comm_size;i++) {
     recvcounts[i] = atoi(action[i+3]);
@@ -683,7 +718,7 @@ static void action_gatherv(const char *const *action) {
   int rank = smpi_process_index();
 
   if(rank==root)
-    recv = calloc(recv_sum, smpi_datatype_size(MPI_CURRENT_TYPE2));
+    recv = xbt_malloc(recv_sum* smpi_datatype_size(MPI_CURRENT_TYPE2));
 
 #ifdef HAVE_TRACING
   instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);
@@ -736,8 +771,7 @@ static void action_reducescatter(const char *const *action) {
   int comp_size = parse_double(action[2+comm_size]);
   int *recvcounts = xbt_new0(int, comm_size);  
   int *disps = xbt_new0(int, comm_size);  
-  int i=0,recv_sum=0;
-  int root=0;
+  int i=0;
   int rank = smpi_process_index();
 
   if(action[3+comm_size])
@@ -747,7 +781,6 @@ static void action_reducescatter(const char *const *action) {
 
   for(i=0;i<comm_size;i++) {
     recvcounts[i] = atoi(action[i+2]);
-    recv_sum=recv_sum+recvcounts[i];
     disps[i] = 0;
   }
 
@@ -765,10 +798,8 @@ static void action_reducescatter(const char *const *action) {
 
   TRACE_smpi_collective_in(rank, -1, __FUNCTION__,extra);
 #endif
-   mpi_coll_reduce_fun(NULL, NULL, recv_sum, MPI_CURRENT_TYPE, MPI_OP_NULL,
-       root, MPI_COMM_WORLD);
-   smpi_mpi_scatterv(NULL, recvcounts, disps, MPI_CURRENT_TYPE, NULL,
-                      recvcounts[rank], MPI_CURRENT_TYPE, 0, MPI_COMM_WORLD);
+   mpi_coll_reduce_scatter_fun(NULL, NULL, recvcounts, MPI_CURRENT_TYPE, MPI_OP_NULL,
+       MPI_COMM_WORLD);
    smpi_execute_flops(comp_size);
 
 
@@ -813,13 +844,13 @@ static void action_allgatherv(const char *const *action) {
     MPI_CURRENT_TYPE = MPI_DEFAULT_TYPE;
     MPI_CURRENT_TYPE2 = MPI_DEFAULT_TYPE;    
   }
-  void *sendbuf = calloc(sendcount, smpi_datatype_size(MPI_CURRENT_TYPE));    
+  void *sendbuf = xbt_malloc(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 = calloc(recv_sum, smpi_datatype_size(MPI_CURRENT_TYPE2));  
+  void *recvbuf = xbt_malloc(recv_sum* smpi_datatype_size(MPI_CURRENT_TYPE2));  
 
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
@@ -836,7 +867,7 @@ static void action_allgatherv(const char *const *action) {
   TRACE_smpi_collective_in(rank, -1, __FUNCTION__,extra);
 #endif
 
-mpi_coll_allgatherv_fun(sendbuf, sendcount, MPI_CURRENT_TYPE, recvbuf, recvcounts, disps, MPI_CURRENT_TYPE2, MPI_COMM_WORLD);
+  mpi_coll_allgatherv_fun(sendbuf, sendcount, MPI_CURRENT_TYPE, recvbuf, recvcounts, disps, MPI_CURRENT_TYPE2, MPI_COMM_WORLD);
 
 #ifdef HAVE_TRACING
   TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
@@ -887,8 +918,8 @@ static void action_allToAllv(const char *const *action) {
       MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
   }
 
-  void *sendbuf = calloc(send_buf_size, smpi_datatype_size(MPI_CURRENT_TYPE));  
-  void *recvbuf = calloc(recv_buf_size, smpi_datatype_size(MPI_CURRENT_TYPE2));  
+  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));  
 
   for(i=0;i<comm_size;i++) {
     sendcounts[i] = atoi(action[i+3]);
@@ -915,7 +946,7 @@ static void action_allToAllv(const char *const *action) {
 
   TRACE_smpi_collective_in(rank, -1, __FUNCTION__,extra);
 #endif
-    mpi_coll_alltoallv_fun(sendbuf, sendcounts, senddisps, MPI_CURRENT_TYPE,
+  mpi_coll_alltoallv_fun(sendbuf, sendcounts, senddisps, MPI_CURRENT_TYPE,
                                recvbuf, recvcounts, recvdisps, MPI_CURRENT_TYPE,
                                MPI_COMM_WORLD);
 #ifdef HAVE_TRACING
@@ -955,6 +986,7 @@ void smpi_replay_init(int *argc, char***argv){
     xbt_replay_action_register("Isend",      action_Isend);
     xbt_replay_action_register("recv",       action_recv);
     xbt_replay_action_register("Irecv",      action_Irecv);
+    xbt_replay_action_register("test",       action_test);
     xbt_replay_action_register("wait",       action_wait);
     xbt_replay_action_register("waitAll",    action_waitall);
     xbt_replay_action_register("barrier",    action_barrier);
@@ -969,7 +1001,16 @@ 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);
 }
 
@@ -977,21 +1018,21 @@ int smpi_replay_finalize(){
   double sim_time= 1.;
   /* One active process will stop. Decrease the counter*/
   XBT_DEBUG("There are %lu elements in reqq[*]",
-            xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)]));
-  if (!xbt_dynar_is_empty(reqq[smpi_comm_rank(MPI_COMM_WORLD)])){
-    int count_requests=xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)]);
+            xbt_dynar_length(reqq[smpi_process_index()]));
+  if (!xbt_dynar_is_empty(reqq[smpi_process_index()])){
+    int count_requests=xbt_dynar_length(reqq[smpi_process_index()]);
     MPI_Request requests[count_requests];
     MPI_Status status[count_requests];
     unsigned int i;
 
-    xbt_dynar_foreach(reqq[smpi_comm_rank(MPI_COMM_WORLD)],i,requests[i]);
+    xbt_dynar_foreach(reqq[smpi_process_index()],i,requests[i]);
     smpi_mpi_waitall(count_requests, requests, status);
     active_processes--;
   } else {
     active_processes--;
   }
 
-  xbt_dynar_free_container(&(reqq[smpi_comm_rank(MPI_COMM_WORLD)]));
+  xbt_dynar_free_container(&(reqq[smpi_process_index()]));
 
   if(!active_processes){
     /* Last process alive speaking */
@@ -1002,7 +1043,7 @@ int smpi_replay_finalize(){
     xbt_free(reqq);
     reqq = NULL;
   }
-  smpi_mpi_barrier(MPI_COMM_WORLD);
+  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);