+int MPI_Waitall(int count, MPI_Request requests[], MPI_Status status[])
+{
+
+ smpi_bench_end(-1, NULL); //FIXME
+#ifdef HAVE_TRACING
+ //save information from requests
+ int i;
+ xbt_dynar_t srcs = xbt_dynar_new(sizeof(int), xbt_free);
+ xbt_dynar_t dsts = xbt_dynar_new(sizeof(int), xbt_free);
+ xbt_dynar_t recvs = xbt_dynar_new(sizeof(int), xbt_free);
+ for (i = 0; i < count; i++) {
+ MPI_Request req = requests[i]; //all req should be valid in Waitall
+ int *asrc = xbt_new(int, 1);
+ int *adst = xbt_new(int, 1);
+ int *arecv = xbt_new(int, 1);
+ *asrc = req->src;
+ *adst = req->dst;
+ *arecv = req->recv;
+ xbt_dynar_insert_at(srcs, i, asrc);
+ xbt_dynar_insert_at(dsts, i, adst);
+ xbt_dynar_insert_at(recvs, i, arecv);
+ }
+ int rank_traced = smpi_comm_rank (MPI_COMM_WORLD);
+ TRACE_smpi_ptp_in(rank_traced, -1, -1, __FUNCTION__);
+#endif
+ smpi_mpi_waitall(count, requests, status);
+#ifdef HAVE_TRACING
+ for (i = 0; i < count; i++) {
+ int src_traced, dst_traced, is_wait_for_receive;
+ xbt_dynar_get_cpy(srcs, i, &src_traced);
+ xbt_dynar_get_cpy(dsts, i, &dst_traced);
+ xbt_dynar_get_cpy(recvs, i, &is_wait_for_receive);
+ if (is_wait_for_receive) {
+ TRACE_smpi_recv(rank_traced, src_traced, dst_traced);
+ }
+ }
+ TRACE_smpi_ptp_out(rank_traced, -1, -1, __FUNCTION__);
+ //clean-up of dynars
+ xbt_free(srcs);
+ xbt_free(dsts);
+ xbt_free(recvs);
+#endif
+ smpi_bench_begin(-1, NULL);
+ return MPI_SUCCESS;
+}
+
+int MPI_Waitsome(int incount, MPI_Request requests[], int *outcount,
+ int *indices, MPI_Status status[])
+{
+ int retval;
+
+ smpi_bench_end(-1, NULL); //FIXME
+ if (outcount == NULL || indices == NULL) {
+ retval = MPI_ERR_ARG;
+ } else {
+ *outcount = smpi_mpi_waitsome(incount, requests, indices, status);
+ retval = MPI_SUCCESS;
+ }
+ smpi_bench_begin(-1, NULL);
+ return retval;
+}
+
+int MPI_Bcast(void *buf, int count, MPI_Datatype datatype, int root,
+ MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Bcast");
+#ifdef HAVE_TRACING
+ int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+ TRACE_smpi_collective_in(rank, root_traced, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else {
+ smpi_mpi_bcast(buf, count, datatype, root, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, root_traced, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Bcast");
+ return retval;
+}
+
+int MPI_Barrier(MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Barrier");
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else {
+ smpi_mpi_barrier(comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Barrier");
+ return retval;
+}
+
+int MPI_Gather(void *sendbuf, int sendcount, MPI_Datatype sendtype,
+ void *recvbuf, int recvcount, MPI_Datatype recvtype,
+ int root, MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Gather");
+#ifdef HAVE_TRACING
+ int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+ TRACE_smpi_collective_in(rank, root_traced, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (sendtype == MPI_DATATYPE_NULL
+ || recvtype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else {
+ smpi_mpi_gather(sendbuf, sendcount, sendtype, recvbuf, recvcount,
+ recvtype, root, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, root_traced, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Gather");
+ return retval;
+}
+
+int MPI_Gatherv(void *sendbuf, int sendcount, MPI_Datatype sendtype,
+ void *recvbuf, int *recvcounts, int *displs,
+ MPI_Datatype recvtype, int root, MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Gatherv");
+#ifdef HAVE_TRACING
+ int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+ TRACE_smpi_collective_in(rank, root_traced, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (sendtype == MPI_DATATYPE_NULL
+ || recvtype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else if (recvcounts == NULL || displs == NULL) {
+ retval = MPI_ERR_ARG;
+ } else {
+ smpi_mpi_gatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts,
+ displs, recvtype, root, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, root_traced, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Gatherv");
+ return retval;
+}
+
+int MPI_Allgather(void *sendbuf, int sendcount, MPI_Datatype sendtype,
+ void *recvbuf, int recvcount, MPI_Datatype recvtype,
+ MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Allgather");
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (sendtype == MPI_DATATYPE_NULL
+ || recvtype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else {
+ smpi_mpi_allgather(sendbuf, sendcount, sendtype, recvbuf, recvcount,
+ recvtype, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Allgather");
+ return retval;
+}
+
+int MPI_Allgatherv(void *sendbuf, int sendcount, MPI_Datatype sendtype,
+ void *recvbuf, int *recvcounts, int *displs,
+ MPI_Datatype recvtype, MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Allgatherv");
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (sendtype == MPI_DATATYPE_NULL
+ || recvtype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else if (recvcounts == NULL || displs == NULL) {
+ retval = MPI_ERR_ARG;
+ } else {
+ smpi_mpi_allgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts,
+ displs, recvtype, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Allgatherv");
+ return retval;
+}
+
+int MPI_Scatter(void *sendbuf, int sendcount, MPI_Datatype sendtype,
+ void *recvbuf, int recvcount, MPI_Datatype recvtype,
+ int root, MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Scatter");
+#ifdef HAVE_TRACING
+ int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+ TRACE_smpi_collective_in(rank, root_traced, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (sendtype == MPI_DATATYPE_NULL
+ || recvtype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else {
+ smpi_mpi_scatter(sendbuf, sendcount, sendtype, recvbuf, recvcount,
+ recvtype, root, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, root_traced, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Scatter");
+ return retval;
+}
+
+int MPI_Scatterv(void *sendbuf, int *sendcounts, int *displs,
+ MPI_Datatype sendtype, void *recvbuf, int recvcount,
+ MPI_Datatype recvtype, int root, MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Scatterv");
+#ifdef HAVE_TRACING
+ int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+ TRACE_smpi_collective_in(rank, root_traced, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (sendtype == MPI_DATATYPE_NULL
+ || recvtype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else if (sendcounts == NULL || displs == NULL) {
+ retval = MPI_ERR_ARG;
+ } else {
+ smpi_mpi_scatterv(sendbuf, sendcounts, displs, sendtype, recvbuf,
+ recvcount, recvtype, root, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, root_traced, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Scatterv");
+ return retval;
+}
+
+int MPI_Reduce(void *sendbuf, void *recvbuf, int count,
+ MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Reduce");
+#ifdef HAVE_TRACING
+ int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+ TRACE_smpi_collective_in(rank, root_traced, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (datatype == MPI_DATATYPE_NULL || op == MPI_OP_NULL) {
+ retval = MPI_ERR_ARG;
+ } else {
+ smpi_mpi_reduce(sendbuf, recvbuf, count, datatype, op, root, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, root_traced, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Reduce");
+ return retval;
+}
+
+int MPI_Allreduce(void *sendbuf, void *recvbuf, int count,
+ MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Allreduce");
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (datatype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else if (op == MPI_OP_NULL) {
+ retval = MPI_ERR_OP;
+ } else {
+ smpi_mpi_allreduce(sendbuf, recvbuf, count, datatype, op, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Allreduce");
+ return retval;
+}
+
+int MPI_Scan(void *sendbuf, void *recvbuf, int count,
+ MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
+{
+ int retval;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Scan");
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (datatype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else if (op == MPI_OP_NULL) {
+ retval = MPI_ERR_OP;
+ } else {
+ smpi_mpi_scan(sendbuf, recvbuf, count, datatype, op, comm);
+ retval = MPI_SUCCESS;
+ }
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_out(rank, -1, __FUNCTION__);
+#endif
+ smpi_bench_begin(rank, "Scan");
+ return retval;
+}
+
+int MPI_Reduce_scatter(void *sendbuf, void *recvbuf, int *recvcounts,
+ MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
+{
+ int retval, i, size, count;
+ int *displs;
+ int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+ smpi_bench_end(rank, "Reduce_scatter");
+#ifdef HAVE_TRACING
+ TRACE_smpi_collective_in(rank, -1, __FUNCTION__);
+#endif
+ if (comm == MPI_COMM_NULL) {
+ retval = MPI_ERR_COMM;
+ } else if (datatype == MPI_DATATYPE_NULL) {
+ retval = MPI_ERR_TYPE;
+ } else if (op == MPI_OP_NULL) {
+ retval = MPI_ERR_OP;
+ } else if (recvcounts == NULL) {
+ retval = MPI_ERR_ARG;
+ } else {
+ /* arbitrarily choose root as rank 0 */
+ /* TODO: faster direct implementation ? */
+ size = smpi_comm_size(comm);
+ count = 0;
+ displs = xbt_new(int, size);
+ for (i = 0; i < size; i++) {
+ count += recvcounts[i];
+ displs[i] = 0;