- for (i = 0; i < comm->size; i++) {
- if (-1 == colormap[i]) {
- continue;
- }
- colortmp = colormap[i];
- keycount = 0;
- for (j = i; j < comm->size; j++) {
- if(colortmp == colormap[j]) {
- colormap[j] = -1;
- rankkeymap[keycount * 2] = j;
- rankkeymap[keycount * 2 + 1] = keymap[j];
- keycount++;
- }
- }
- qsort(rankkeymap, keycount, sizeof(int) * 2, &smpi_compare_rankkeys);
- tempcomm = xbt_new(s_smpi_mpi_communicator_t, 1);
- tempcomm->barrier_count = 0;
- tempcomm->size = keycount;
- tempcomm->barrier_mutex = SIMIX_mutex_init();
- tempcomm->barrier_cond = SIMIX_cond_init();
- tempcomm->rank_to_index_map = xbt_new(int, keycount);
- tempcomm->index_to_rank_map = xbt_new(int, smpi_global->host_count);
- for (j = 0; j < smpi_global->host_count; j++) {
- tempcomm->index_to_rank_map[j] = -1;
- }
- for (j = 0; j < keycount; j++) {
- indextmp = comm->rank_to_index_map[rankkeymap[j*2]];
- tempcomm->rank_to_index_map[j] = indextmp;
- tempcomm->index_to_rank_map[indextmp] = j;
- }
- for (j = 0; j < keycount; j++) {
- if (rankkeymap[j*2]) {
- retval = smpi_create_request(&j, 1, MPI_INT, 0,
- rankkeymap[j*2], 0, comm, &request);
- request->data = tempcomm;
- smpi_mpi_isend(request);
- smpi_mpi_wait(request, &status);
- xbt_mallocator_release(smpi_global->request_mallocator, request);
- } else {
- *comm_out = tempcomm;
- }
- }
- }
- } else {
- colorkey[0] = color;
- colorkey[1] = key;
- retval = smpi_create_request(colorkey, 2, MPI_INT, rank, 0, 0, comm, &request);
- smpi_mpi_isend(request);
- smpi_mpi_wait(request, &status);
- xbt_mallocator_release(smpi_global->request_mallocator, request);
- retval = smpi_create_request(colorkey, 1, MPI_INT, 0, rank, 0, comm, &request);
- smpi_mpi_irecv(request);
- smpi_mpi_wait(request, &status);
- *comm_out = request->data;
- }
-
- smpi_bench_begin();
-
- return retval;
+int SMPI_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 = MPI_SUCCESS;
+ int system_tag = 666;
+ int rank, size;
+
+ smpi_bench_end();
+ rank = smpi_mpi_comm_rank(comm);
+ size = comm->size;
+ if(rank != root) {
+ // Receive buffer from root
+ smpi_mpi_request_t request;
+
+ retval = smpi_create_request(recvbuf, recvcount, recvtype,
+ root, rank, system_tag, comm, &request);
+ smpi_mpi_isend(request);
+ smpi_mpi_wait(request, MPI_STATUS_IGNORE);
+ xbt_mallocator_release(smpi_global->request_mallocator, request);
+ } else {
+ // Send buffers to receivers
+ int dst;
+ smpi_mpi_request_t* requests;
+
+ requests = xbt_malloc((size-1) * sizeof(smpi_mpi_request_t));
+ for(dst = 0; dst < size; dst++) {
+ if(dst == root) {
+ // Local copy from root
+ memcpy(recvbuf, &((char*)sendbuf)[displs[dst]],
+ sendcounts[dst]*sendtype->size*sizeof(char));
+ } else {
+ int index = dst < root ? dst : dst - 1;
+ retval = smpi_create_request(&((char*)sendbuf)[displs[dst]], sendcounts[dst], sendtype,
+ root, dst, system_tag, comm, &requests[index]);
+ if(NULL != requests[index] && MPI_SUCCESS == retval) {
+ smpi_mpi_isend(requests[index]);
+ }
+ }
+ }
+ // Wait for completion of isend's.
+ for(dst = 0; dst < size - 1; dst++) {
+ int index = MPI_UNDEFINED;
+ smpi_mpi_waitany(size - 1, requests, &index, MPI_STATUS_IGNORE);
+ xbt_mallocator_release(smpi_global->request_mallocator, requests[index]);
+ }
+ xbt_free(requests);
+ }
+ smpi_bench_begin();
+ return retval;
+}
+
+int SMPI_MPI_Reduce_scatter(void* sendbuf, void* recvbuf, int *recvcounts,
+ MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
+{
+ // FIXME: Suboptimal implementation
+ int retval = MPI_SUCCESS;
+ int count = 0;
+ int root = 0;
+ int i, rank;
+ int* displs;
+
+ smpi_bench_end();
+ rank = smpi_mpi_comm_rank(comm);
+ displs = xbt_new(int, comm->size);
+ for(i = 0; i < comm->size; i++) {
+ count += recvcounts[i];
+ displs[i] = 0;
+ }
+ retval = smpi_mpi_reduce(sendbuf, recvbuf, count, datatype, op, root, comm);
+ retval = SMPI_MPI_Scatterv(recvbuf, recvcounts, displs, datatype, recvbuf, recvcounts[rank], datatype, root, comm);
+ xbt_free(displs);
+ smpi_bench_begin();
+ return retval;
+}
+
+int SMPI_MPI_Allgather(void* sendbuf, int sendcount, MPI_Datatype sendtype,
+ void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm)
+{
+ // FIXME: Suboptimal implementation
+ int root = 0;
+ int retval;
+
+ smpi_bench_end();
+ retval = SMPI_MPI_Gather(sendbuf, sendcount, sendtype,
+ recvbuf, recvcount, recvtype, root, comm);
+ if(retval == MPI_SUCCESS) {
+ retval = SMPI_MPI_Bcast(recvbuf, recvcount, recvtype, root, comm);
+ }
+ smpi_bench_begin();
+ return retval;
+}
+
+int SMPI_MPI_Allgatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype,
+ void* recvbuf, int *recvcounts, int *displs, MPI_Datatype recvtype,
+ MPI_Comm comm)
+{
+ // FIXME: Suboptimal implementation
+ int root = 0;
+ int last, retval;
+
+ smpi_bench_end();
+ retval = SMPI_MPI_Gatherv(sendbuf, sendcount, sendtype,
+ recvbuf, recvcounts, displs, recvtype, root, comm);
+ if(retval == MPI_SUCCESS) {
+ last = comm->size - 1;
+ retval = SMPI_MPI_Bcast(recvbuf, displs[last] + recvcounts[last], recvtype, root, comm);
+ }
+ smpi_bench_begin();
+ return retval;