have smp-aware algorithms use number of cores on the node as basis for their computat...

[simgrid.git] / src / smpi / colls / allgather-loosely-lr.c

diff --git a/src/smpi/colls/allgather-loosely-lr.c b/src/smpi/colls/allgather-loosely-lr.c
index 6455db1..242990f 100644 (file)
--- a/src/smpi/colls/allgather-loosely-lr.c
+++ b/src/smpi/colls/allgather-loosely-lr.c
@@ -10,12 +10,21 @@ int smpi_coll_tuned_allgather_loosely_lr(void *sbuf, int scount,
                                          MPI_Comm comm)
 {
   int comm_size, rank;
-  int tag = 50;
+  int tag = COLL_TAG_ALLGATHER;
   int i, j, send_offset, recv_offset;
   int intra_rank, inter_rank, inter_comm_size, intra_comm_size;
   int inter_dst, inter_src;
 
   comm_size = smpi_comm_size(comm);
+
+  int num_core = simcall_host_get_core(SIMIX_host_self());
+  // do we use the default one or the number of cores in the platform ?
+  // if the number of cores is one, the platform may be simulated with 1 node = 1 core
+  if (num_core == 1) num_core = NUM_CORE;
+
+  if(comm_size%num_core)
+    THROWF(arg_error,0, "allgather loosely lr algorithm can't be used with non multiple of NUM_CORE=%d number of processes ! ",num_core);
+
   rank = smpi_comm_rank(comm);
   MPI_Aint rextent, sextent;
   rextent = smpi_datatype_get_extent(rtype);
@@ -32,10 +41,10 @@ int smpi_coll_tuned_allgather_loosely_lr(void *sbuf, int scount,
 
   MPI_Status status;
 
-  intra_rank = rank % NUM_CORE;
-  inter_rank = rank / NUM_CORE;
-  inter_comm_size = (comm_size + NUM_CORE - 1) / NUM_CORE;
-  intra_comm_size = NUM_CORE;
+  intra_rank = rank % num_core;
+  inter_rank = rank / num_core;
+  inter_comm_size = (comm_size + num_core - 1) / num_core;
+  intra_comm_size = num_core;
 
   int src_seg, dst_seg;
 
@@ -104,7 +113,7 @@ int smpi_coll_tuned_allgather_loosely_lr(void *sbuf, int scount,
     }                           // intra loop
 
 
-    // wait for inter communication to finish for these rounds (# of round equals NUM_CORE)
+    // wait for inter communication to finish for these rounds (# of round equals num_core)
     if (i != inter_comm_size - 1) {
       smpi_mpi_wait(&inter_rrequest, &status);
     }