--- /dev/null
+/* Copyright (c) 2013-2014. The SimGrid Team.
+ * All rights reserved. */
+
+/* This program is free software; you can redistribute it and/or modify it
+ * under the terms of the license (GNU LGPL) which comes with this package. */
+
+#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
+ It uses 2-layer communication: binomial for both intra-communication
+ inter-communication
+ The communication are done in a pipeline fashion */
+
+
+
+/* this is a default segment size for pipelining,
+ but it is typically passed as a command line argument */
+int allreduce_smp_binomial_pipeline_segment_size = 4096;
+
+/* ** NOTE **
+ This code is modified from allreduce-smp-binomial.c by wrapping the code with pipeline effect as follow
+ for (loop over pipelength) {
+ smp-binomial main code;
+ }
+*/
+
+/* ** NOTE **
+ Use -DMPICH2 if this code does not compile.
+ MPICH1 code also work on MPICH2 on our cluster and the performance are similar.
+ This code assume commutative and associative reduce operator (MPI_SUM, MPI_MAX, etc).
+*/
+
+/*
+This fucntion performs all-reduce operation as follow. ** in a pipeline fashion **
+1) binomial_tree reduce inside each SMP node
+2) binomial_tree reduce intra-communication between root of each SMP node
+3) binomial_tree bcast intra-communication between root of each SMP node
+4) binomial_tree bcast inside each SMP node
+*/
+int smpi_coll_tuned_allreduce_smp_binomial_pipeline(void *send_buf,
+ void *recv_buf, int count,
+ MPI_Datatype dtype,
+ MPI_Op op, MPI_Comm comm)
+{
+ int comm_size, rank;
+ void *tmp_buf;
+ int tag = COLL_TAG_ALLREDUCE;
+ int mask, src, dst;
+ MPI_Status status;
+ if(comm->get_leaders_comm()==MPI_COMM_NULL){
+ comm->init_smp();
+ }
+ int num_core=1;
+ if (comm->is_uniform()){
+ num_core = comm->get_intra_comm()->size();
+ }
+
+ comm_size = comm->size();
+ rank = comm->rank();
+ MPI_Aint extent;
+ extent = dtype->get_extent();
+ tmp_buf = (void *) smpi_get_tmp_sendbuffer(count * extent);
+
+ int intra_rank, inter_rank;
+ intra_rank = rank % num_core;
+ inter_rank = rank / num_core;
+
+ int phase;
+ int send_offset;
+ int recv_offset;
+ int pcount = allreduce_smp_binomial_pipeline_segment_size;
+ if (pcount > count) {
+ pcount = count;
+ }
+
+ /* size of processes participate in intra communications =>
+ should be equal to number of machines */
+ int inter_comm_size = (comm_size + num_core - 1) / num_core;
+
+ /* copy input buffer to output buffer */
+ Request::sendrecv(send_buf, count, dtype, rank, tag,
+ recv_buf, count, dtype, rank, tag, comm, &status);
+
+ /* compute pipe length */
+ int pipelength;
+ pipelength = count / pcount;
+
+ /* pipelining over pipelength (+3 is because we have 4 stages:
+ reduce-intra, reduce-inter, bcast-inter, bcast-intra */
+ for (phase = 0; phase < pipelength + 3; phase++) {
+
+ /* start binomial reduce intra communication inside each SMP node */
+ if (phase < pipelength) {
+ mask = 1;
+ while (mask < num_core) {
+ if ((mask & intra_rank) == 0) {
+ src = (inter_rank * num_core) + (intra_rank | mask);
+ if (src < comm_size) {
+ recv_offset = phase * pcount * extent;
+ Request::recv(tmp_buf, pcount, dtype, src, tag, comm, &status);
+ if(op!=MPI_OP_NULL) op->apply( tmp_buf, (char *)recv_buf + recv_offset, &pcount, dtype);
+ }
+ } else {
+ send_offset = phase * pcount * extent;
+ dst = (inter_rank * num_core) + (intra_rank & (~mask));
+ Request::send((char *)recv_buf + send_offset, pcount, dtype, dst, tag, comm);
+ break;
+ }
+ mask <<= 1;
+ }
+ }
+
+ /* start binomial reduce inter-communication between each SMP nodes:
+ each node only have one process that can communicate to other nodes */
+ if ((phase > 0) && (phase < (pipelength + 1))) {
+ if (intra_rank == 0) {
+
+ mask = 1;
+ while (mask < inter_comm_size) {
+ if ((mask & inter_rank) == 0) {
+ src = (inter_rank | mask) * num_core;
+ if (src < comm_size) {
+ recv_offset = (phase - 1) * pcount * extent;
+ Request::recv(tmp_buf, pcount, dtype, src, tag, comm, &status);
+ if(op!=MPI_OP_NULL) op->apply( tmp_buf, (char *)recv_buf + recv_offset, &pcount, dtype);
+ }
+ } else {
+ dst = (inter_rank & (~mask)) * num_core;
+ send_offset = (phase - 1) * pcount * extent;
+ Request::send((char *)recv_buf + send_offset, pcount, dtype, dst, tag, comm);
+ break;
+ }
+ mask <<= 1;
+ }
+ }
+ }
+
+ /* start binomial broadcast inter-communication between each SMP nodes:
+ each node only have one process that can communicate to other nodes */
+ if ((phase > 1) && (phase < (pipelength + 2))) {
+ if (intra_rank == 0) {
+ mask = 1;
+ while (mask < inter_comm_size) {
+ if (inter_rank & mask) {
+ src = (inter_rank - mask) * num_core;
+ recv_offset = (phase - 2) * pcount * extent;
+ Request::recv((char *)recv_buf + recv_offset, pcount, dtype, src, tag, comm,
+ &status);
+ break;
+ }
+ mask <<= 1;
+ }
+ mask >>= 1;
+
+ while (mask > 0) {
+ if (inter_rank < inter_comm_size) {
+ dst = (inter_rank + mask) * num_core;
+ if (dst < comm_size) {
+ //printf("Node %d send to node %d when mask is %d\n", rank, dst, mask);
+ send_offset = (phase - 2) * pcount * extent;
+ Request::send((char *)recv_buf + send_offset, pcount, dtype, dst, tag, comm);
+ }
+ }
+ mask >>= 1;
+ }
+ }
+ }
+
+ /* start binomial broadcast intra-communication inside each SMP nodes */
+ if (phase > 2) {
+ int num_core_in_current_smp = num_core;
+ if (inter_rank == (inter_comm_size - 1)) {
+ num_core_in_current_smp = comm_size - (inter_rank * num_core);
+ }
+ mask = 1;
+ while (mask < num_core_in_current_smp) {
+ if (intra_rank & mask) {
+ src = (inter_rank * num_core) + (intra_rank - mask);
+ recv_offset = (phase - 3) * pcount * extent;
+ Request::recv((char *)recv_buf + recv_offset, pcount, dtype, src, tag, comm,
+ &status);
+ break;
+ }
+ mask <<= 1;
+ }
+ mask >>= 1;
+
+ while (mask > 0) {
+ dst = (inter_rank * num_core) + (intra_rank + mask);
+ if (dst < comm_size) {
+ send_offset = (phase - 3) * pcount * extent;
+ Request::send((char *)recv_buf + send_offset, pcount, dtype, dst, tag, comm);
+ }
+ mask >>= 1;
+ }
+ }
+ } // for phase
+
+ smpi_free_tmp_buffer(tmp_buf);
+ return MPI_SUCCESS;
+}
