-#include "colls.h"
+#include "colls_private.h"
/* IMPLEMENTED BY PITCH PATARASUK
Non-topoloty-specific (however, number of cores/node need to be changed)
all-reduce operation designed for smp clusters
{
int comm_size, rank;
void *tmp_buf;
- int tag = 50;
+ int tag = COLL_TAG_ALLREDUCE;
int mask, src, dst;
int num_core = NUM_CORE;
MPI_Status status;
- /*
- #ifdef MPICH2_REDUCTION
- MPI_User_function * uop = MPIR_Op_table[op % 16 - 1];
- #else
- MPI_User_function *uop;
- struct MPIR_OP *op_ptr;
- op_ptr = MPIR_ToPointer(op);
- uop = op_ptr->op;
- #endif
- */
comm_size=smpi_comm_size(comm);
rank=smpi_comm_rank(comm);
MPI_Aint extent, lb;
smpi_datatype_extent(dtype, &lb, &extent);
- tmp_buf = (void *) malloc(count * extent);
+ tmp_buf = (void *) xbt_malloc(count * extent);
/* compute intra and inter ranking */
int intra_rank, inter_rank;
src = (inter_rank * num_core) + (intra_rank | mask);
if (src < comm_size) {
smpi_mpi_recv(tmp_buf, count, dtype, src, tag, comm, &status);
- star_reduction(op, tmp_buf, recv_buf, &count, &dtype);
+ smpi_op_apply(op, tmp_buf, recv_buf, &count, &dtype);
}
} else {
dst = (inter_rank * num_core) + (intra_rank & (~mask));
src = (inter_rank | mask) * num_core;
if (src < comm_size) {
smpi_mpi_recv(tmp_buf, count, dtype, src, tag, comm, &status);
- star_reduction(op, tmp_buf, recv_buf, &count, &dtype);
+ smpi_op_apply(op, tmp_buf, recv_buf, &count, &dtype);
}
} else {
dst = (inter_rank & (~mask)) * num_core;