New actions for the time independent trace replay framework:

Gather
AllGatherV
ReduceScatter

* I changed some declarations from xbt_new0() to calloc(), I had to use an MPI datatype variable declared as MPI_CURRENT_TYPE and it was not possible with xbt_new0().
* Some new example files for the collectives calls with real values extracted from BigDFT.

diff --git a/examples/smpi/replay/actions_allgatherv.txt b/examples/smpi/replay/actions_allgatherv.txt
new file mode 100644 (file)
index 0000000..f0c0e58
--- /dev/null
+++ b/examples/smpi/replay/actions_allgatherv.txt
@@ -0,0 +1,14 @@
+0 init
+1 init
+2 init
+3 init
+
+0 allGatherV 275427 275427 275427 275427 204020 0 275427 550854 826281 0 0
+1 allGatherV 275427 275427 275427 275427 204020 0 275427 550854 826281 0 0
+2 allGatherV 275427 275427 275427 275427 204020 0 275427 550854 826281 0 0
+3 allGatherV 204020 275427 275427 275427 204020 0 275427 550854 826281 0 0
+
+0 finalize
+1 finalize
+2 finalize
+3 finalize

diff --git a/examples/smpi/replay/actions_gather.txt b/examples/smpi/replay/actions_gather.txt
new file mode 100644 (file)
index 0000000..33f7d0a
--- /dev/null
+++ b/examples/smpi/replay/actions_gather.txt
@@ -0,0 +1,14 @@
+0 init
+1 init
+2 init
+3 init
+
+0 gather 68 68 0 0 0
+1 gather 68 68 0 0 0
+2 gather 68 68 0 0 0
+3 gather 68 68 0 0 0
+
+0 finalize
+1 finalize
+2 finalize
+3 finalize

diff --git a/examples/smpi/replay/actions_reducescatter.txt b/examples/smpi/replay/actions_reducescatter.txt
new file mode 100644 (file)
index 0000000..6580e06
--- /dev/null
+++ b/examples/smpi/replay/actions_reducescatter.txt
@@ -0,0 +1,14 @@
+0 init
+1 init
+2 init
+3 init
+
+0 reduceScatter 275427 275427 275427 204020 11349173 0
+1 reduceScatter 275427 275427 275427 204020 12396024 0
+2 reduceScatter 275427 275427 275427 204020 12501522 0
+3 reduceScatter 275427 275427 275427 204020 12403123 0
+~                                                                    
+0 finalize
+1 finalize
+2 finalize
+3 finalize

diff --git a/src/smpi/smpi_replay.c b/src/smpi/smpi_replay.c
index a82925e..9dcf47b 100644 (file)
--- a/src/smpi/smpi_replay.c
+++ b/src/smpi/smpi_replay.c
@@ -16,7 +16,8 @@ int communicator_size = 0;
 static int active_processes = 0;
 xbt_dynar_t *reqq = NULL;
 
-MPI_Datatype MPI_DEFAULT_TYPE, MPI_CURRENT_TYPE;
+MPI_Datatype MPI_DEFAULT_TYPE;
+MPI_Datatype MPI_CURRENT_TYPE;
 
 static void log_timed_action (const char *const *action, double clock){
   if (XBT_LOG_ISENABLED(smpi_replay, xbt_log_priority_verbose)){
@@ -491,11 +492,18 @@ static void action_allToAll(const char *const *action) {
   int comm_size = smpi_comm_size(MPI_COMM_WORLD);
   int send_size = parse_double(action[2]);
   int recv_size = parse_double(action[3]);
-  void *send = xbt_new0(int, send_size*comm_size);  
-  void *recv = xbt_new0(int, send_size*comm_size);
+  MPI_Datatype MPI_CURRENT_TYPE2;
   
-  if(action[4]) MPI_CURRENT_TYPE=decode_datatype(action[4]);
-  else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE;
+  if(action[4]) {
+    MPI_CURRENT_TYPE=decode_datatype(action[4]);
+    MPI_CURRENT_TYPE2=decode_datatype(action[5]);
+  }
+  else {
+    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));  
 
 #ifdef HAVE_TRACING
   int rank = smpi_process_index();
@@ -503,7 +511,64 @@ static void action_allToAll(const char *const *action) {
   TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
 #endif
   
-  mpi_coll_alltoall_fun(send, send_size, MPI_CURRENT_TYPE, recv, recv_size, MPI_CURRENT_TYPE, MPI_COMM_WORLD);
+  mpi_coll_alltoall_fun(send, send_size, MPI_CURRENT_TYPE, recv, recv_size, MPI_CURRENT_TYPE2, MPI_COMM_WORLD);
+
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+  TRACE_smpi_computing_in(rank);
+#endif
+
+  log_timed_action (action, clock);
+  xbt_free(send);
+  xbt_free(recv);
+}
+
+
+static void action_gather(const char *const *action) {
+  
+  
+  /*
+ The structure of the gather action for the rank 0 (total 4 processes) 
+ is the following:   
+ 0 gather 68 68 0 0 0
+
+  where: 
+  1) 68 is the sendcounts
+  2) 68 is the recvcounts
+  3) 0 is the root node
+  4) 0 is the send datatype id, see decode_datatype()
+  5) 0 is the recv datatype id, see decode_datatype()
+    
+  */
+  double clock = smpi_process_simulated_elapsed();
+  int comm_size = smpi_comm_size(MPI_COMM_WORLD);
+  int send_size = parse_double(action[2]);
+  int recv_size = parse_double(action[3]);
+  MPI_Datatype MPI_CURRENT_TYPE2;
+  if(action[5]) {
+    MPI_CURRENT_TYPE=decode_datatype(action[5]);
+    MPI_CURRENT_TYPE2=decode_datatype(action[6]);
+  } else {
+    MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE;
+    MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE;
+  }
+  void *send = calloc(send_size, smpi_datatype_size(MPI_CURRENT_TYPE));  
+  void *recv = calloc(recv_size, smpi_datatype_size(MPI_CURRENT_TYPE2));  
+
+  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));  
+
+   
+#ifdef HAVE_TRACING
+  int rank = smpi_process_index();
+  TRACE_smpi_computing_out(rank);
+  TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#endif
+smpi_mpi_gather(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__);
@@ -515,6 +580,124 @@ static void action_allToAll(const char *const *action) {
   xbt_free(recv);
 }
 
+
+static void action_reducescatter(const char *const *action) {
+  
+    /*
+ The structure of the reducescatter action for the rank 0 (total 4 processes) 
+ is the following:   
+0 reduceScatter 275427 275427 275427 204020 11346849 0
+
+  where: 
+  1) The first four values after the name of the action declare the recvcounts array
+  2) The value 11346849 is the amount of instructions
+  3) The last value corresponds to the datatype, see decode_datatype().
+  
+  We analyze a MPI_Reduce_scatter call to one MPI_Reduce and one MPI_Scatterv.
+  
+   */
+
+  double clock = smpi_process_simulated_elapsed();
+  int comm_size = smpi_comm_size(MPI_COMM_WORLD);
+  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 rank = smpi_process_index();
+
+  if(action[3+comm_size]) MPI_CURRENT_TYPE=decode_datatype(action[3+comm_size]);
+  else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE;
+
+  for(i=0;i<comm_size;i++) {
+    recvcounts[i] = atoi(action[i+2]);
+    recv_sum=recv_sum+recvcounts[i];
+    disps[i] = 0;
+  }
+ 
+#ifdef HAVE_TRACING
+  int rank = smpi_process_index();
+  TRACE_smpi_computing_out(rank);
+  TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#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);
+   smpi_execute_flops(comp_size);
+    
+    
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+  TRACE_smpi_computing_in(rank);
+#endif
+
+  log_timed_action (action, clock);
+}
+
+
+static void action_allgatherv(const char *const *action) {
+  
+  /*
+ The structure of the allgatherv action for the rank 0 (total 4 processes) 
+ is the following:   
+0 allGatherV 275427 275427 275427 275427 204020 0 275427 550854 826281 
+
+  where: 
+  1) 275427 is the sendcount
+  2) The next four elements declare the recvcounts array
+  3) The next four values declare the disps array
+  4) No more values mean that the datatype for sent and receive buffer
+  is the default one, see decode_datatype().
+
+   */
+
+  double clock = smpi_process_simulated_elapsed();
+  
+  int comm_size = smpi_comm_size(MPI_COMM_WORLD);
+  int i=0;
+  int sendcount=atoi(action[2]);
+  int *recvcounts = xbt_new0(int, comm_size);  
+  int *disps = xbt_new0(int, comm_size);  
+  int recv_sum=0;  
+  MPI_Datatype MPI_CURRENT_TYPE2;
+
+  if(action[3+2*comm_size]) {
+    MPI_CURRENT_TYPE = decode_datatype(action[3+2*comm_size]);
+    MPI_CURRENT_TYPE2 = decode_datatype(action[4+2*comm_size]);
+  } else {
+    MPI_CURRENT_TYPE = MPI_DEFAULT_TYPE;
+    MPI_CURRENT_TYPE2 = MPI_DEFAULT_TYPE;    
+  }
+  void *sendbuf = calloc(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];
+    disps[i] = atoi(action[i+3+comm_size]);
+  }
+  void *recvbuf = calloc(recv_sum, smpi_datatype_size(MPI_CURRENT_TYPE2));  
+
+#ifdef HAVE_TRACING
+  int rank = MPI_COMM_WORLD != MPI_COMM_NULL ? smpi_process_index() : -1;
+  TRACE_smpi_computing_out(rank);
+  TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#endif
+  
+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__);
+  TRACE_smpi_computing_in(rank);
+#endif
+   
+  log_timed_action (action, clock);
+  xbt_free(sendbuf);
+  xbt_free(recvbuf);
+  xbt_free(recvcounts);
+  xbt_free(disps);
+}
+
+
 static void action_allToAllv(const char *const *action) {
   /*
  The structure of the allToAllV action for the rank 0 (total 4 processes) 
@@ -540,15 +723,22 @@ static void action_allToAllv(const char *const *action) {
   int *recvcounts = xbt_new0(int, comm_size);  
   int *senddisps = xbt_new0(int, comm_size);  
   int *recvdisps = xbt_new0(int, comm_size);  
+
+  MPI_Datatype MPI_CURRENT_TYPE2;
   
   send_buf_size=parse_double(action[2]);
   recv_buf_size=parse_double(action[3+2*comm_size]);
-
-  int *sendbuf = xbt_new0(int, send_buf_size);  
-  int *recvbuf = xbt_new0(int, recv_buf_size);  
-
-  if(action[4+4*comm_size]) MPI_CURRENT_TYPE=decode_datatype(action[4+4*comm_size]);    
-  else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE;
+  if(action[4+4*comm_size]) {
+    MPI_CURRENT_TYPE=decode_datatype(action[4+4*comm_size]);    
+    MPI_CURRENT_TYPE2=decode_datatype(action[5+4*comm_size]);    
+  }
+  else {
+      MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE;
+      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));  
 
   for(i=0;i<comm_size;i++) {
     sendcounts[i] = atoi(action[i+3]);
@@ -603,6 +793,9 @@ void smpi_replay_init(int *argc, char***argv){
     xbt_replay_action_register("allReduce",  action_allReduce);
     xbt_replay_action_register("allToAll",   action_allToAll);
     xbt_replay_action_register("allToAllV",  action_allToAllv);
+    xbt_replay_action_register("gather",  action_gather);
+    xbt_replay_action_register("allGatherV",  action_allgatherv);
+    xbt_replay_action_register("reduceScatter",  action_reducescatter);
     xbt_replay_action_register("compute",    action_compute);
   }