/* smpi_coll.c -- various optimized routing for collectives */
-/* Copyright (c) 2009-2018. The SimGrid Team. All rights reserved. */
+/* Copyright (c) 2009-2019. 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 "smpi_request.hpp"
#include "xbt/config.hpp"
-XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_coll, smpi, "Logging specific to SMPI (coll)");
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_coll, smpi, "Logging specific to SMPI collectives.");
#define COLL_SETTER(cat, ret, args, args2) \
- int(*Colls::cat) args; \
- void Colls::set_##cat(std::string name) \
+ void colls::_XBT_CONCAT(set_, cat)(const std::string& name) \
{ \
- int id = find_coll_description(mpi_coll_##cat##_description, name, #cat); \
- cat = reinterpret_cast<ret(*) args>(mpi_coll_##cat##_description[id].coll); \
+ int id = find_coll_description(_XBT_CONCAT3(mpi_coll_, cat, _description), name, _XBT_STRINGIFY(cat)); \
+ cat = reinterpret_cast<ret(*) args>(_XBT_CONCAT3(mpi_coll_, cat, _description)[id].coll); \
if (cat == nullptr) \
- xbt_die("Collective " #cat " set to nullptr!"); \
+ xbt_die("Collective " _XBT_STRINGIFY(cat) " set to nullptr!"); \
}
namespace simgrid{
namespace smpi{
-void (*Colls::smpi_coll_cleanup_callback)();
-
/* these arrays must be nullptr terminated */
-s_mpi_coll_description_t Colls::mpi_coll_gather_description[] = {
- COLL_GATHERS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_allgather_description[] = {
- COLL_ALLGATHERS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_allgatherv_description[] = {
- COLL_ALLGATHERVS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_allreduce_description[] ={
- COLL_ALLREDUCES(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_reduce_scatter_description[] = {
- COLL_REDUCE_SCATTERS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_scatter_description[] ={
- COLL_SCATTERS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_barrier_description[] ={
- COLL_BARRIERS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_alltoall_description[] = {
- COLL_ALLTOALLS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_alltoallv_description[] = {
- COLL_ALLTOALLVS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_bcast_description[] = {
- COLL_BCASTS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
-s_mpi_coll_description_t Colls::mpi_coll_reduce_description[] = {
- COLL_REDUCES(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr} };
+s_mpi_coll_description_t mpi_coll_gather_description[] = {COLL_GATHERS(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_allgather_description[] = {COLL_ALLGATHERS(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_allgatherv_description[] = {COLL_ALLGATHERVS(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_allreduce_description[] = {COLL_ALLREDUCES(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_reduce_scatter_description[] = {COLL_REDUCE_SCATTERS(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_scatter_description[] = {COLL_SCATTERS(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_barrier_description[] = {COLL_BARRIERS(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_alltoall_description[] = {COLL_ALLTOALLS(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_alltoallv_description[] = {COLL_ALLTOALLVS(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_bcast_description[] = {COLL_BCASTS(COLL_DESCRIPTION, COLL_COMMA), {"", "", nullptr}};
+s_mpi_coll_description_t mpi_coll_reduce_description[] = {COLL_REDUCES(COLL_DESCRIPTION, COLL_COMMA),
+ {"", "", nullptr}};
+
+// Needed by the automatic selector weird implementation
+s_mpi_coll_description_t* colls::get_smpi_coll_description(const char* name, int rank)
+{
+ if (strcmp(name, "gather") == 0)
+ return &mpi_coll_gather_description[rank];
+ if (strcmp(name, "allgather") == 0)
+ return &mpi_coll_allgather_description[rank];
+ if (strcmp(name, "allgatherv") == 0)
+ return &mpi_coll_allgatherv_description[rank];
+ if (strcmp(name, "allreduce") == 0)
+ return &mpi_coll_allreduce_description[rank];
+ if (strcmp(name, "reduce_scatter") == 0)
+ return &mpi_coll_reduce_scatter_description[rank];
+ if (strcmp(name, "scatter") == 0)
+ return &mpi_coll_scatter_description[rank];
+ if (strcmp(name, "barrier") == 0)
+ return &mpi_coll_barrier_description[rank];
+ if (strcmp(name, "alltoall") == 0)
+ return &mpi_coll_alltoall_description[rank];
+ if (strcmp(name, "alltoallv") == 0)
+ return &mpi_coll_alltoallv_description[rank];
+ if (strcmp(name, "bcast") == 0)
+ return &mpi_coll_bcast_description[rank];
+ if (strcmp(name, "reduce") == 0)
+ return &mpi_coll_reduce_description[rank];
+ XBT_INFO("You requested an unknown collective: %s", name);
+ return nullptr;
+}
/** Displays the long description of all registered models, and quit */
-void Colls::coll_help(const char *category, s_mpi_coll_description_t * table)
+void colls::coll_help(const char* category, s_mpi_coll_description_t* table)
{
XBT_WARN("Long description of the %s models accepted by this simulator:\n", category);
for (int i = 0; not table[i].name.empty(); i++)
XBT_WARN(" %s: %s\n", table[i].name.c_str(), table[i].description.c_str());
}
-int Colls::find_coll_description(s_mpi_coll_description_t* table, std::string name, const char* desc)
+int colls::find_coll_description(s_mpi_coll_description_t* table, const std::string& name, const char* desc)
{
for (int i = 0; not table[i].name.empty(); i++)
if (name == table[i].name) {
return -1;
}
-COLL_APPLY(COLL_SETTER,COLL_GATHER_SIG,"");
+int (*colls::gather)(const void* send_buff, int send_count, MPI_Datatype send_type, void* recv_buff, int recv_count,
+ MPI_Datatype recv_type, int root, MPI_Comm comm);
+int (*colls::allgather)(const void* send_buff, int send_count, MPI_Datatype send_type, void* recv_buff, int recv_count,
+ MPI_Datatype recv_type, MPI_Comm comm);
+int (*colls::allgatherv)(const void* send_buff, int send_count, MPI_Datatype send_type, void* recv_buff,
+ const int* recv_count, const int* recv_disps, MPI_Datatype recv_type, MPI_Comm comm);
+int (*colls::alltoall)(const void* send_buff, int send_count, MPI_Datatype send_type, void* recv_buff, int recv_count,
+ MPI_Datatype recv_type, MPI_Comm comm);
+int (*colls::alltoallv)(const void* send_buff, const int* send_counts, const int* send_disps, MPI_Datatype send_type,
+ void* recv_buff, const int* recv_counts, const int* recv_disps, MPI_Datatype recv_type,
+ MPI_Comm comm);
+int (*colls::bcast)(void* buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm);
+int (*colls::reduce)(const void* buf, void* rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm);
+int (*colls::allreduce)(const void* sbuf, void* rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm);
+int (*colls::reduce_scatter)(const void* sbuf, void* rbuf, const int* rcounts, MPI_Datatype dtype, MPI_Op op,
+ MPI_Comm comm);
+int (*colls::scatter)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
+ MPI_Datatype recvtype, int root, MPI_Comm comm);
+int (*colls::barrier)(MPI_Comm comm);
+
+void (*colls::smpi_coll_cleanup_callback)();
+
+void colls::set_gather(const std::string& name)
+{
+ int id = find_coll_description(mpi_coll_gather_description, name, "gather");
+ gather = reinterpret_cast<int(*)(const void *send_buff, int send_count, MPI_Datatype send_type,
+ void *recv_buff, int recv_count, MPI_Datatype recv_type,
+ int root, MPI_Comm comm)>(mpi_coll_gather_description[id].coll);
+ if (gather == nullptr)
+ xbt_die("Collective gather set to nullptr!");
+}
+
+//COLL_APPLY(COLL_SETTER,COLL_GATHER_SIG,"");
COLL_APPLY(COLL_SETTER,COLL_ALLGATHER_SIG,"");
COLL_APPLY(COLL_SETTER,COLL_ALLGATHERV_SIG,"");
COLL_APPLY(COLL_SETTER,COLL_REDUCE_SIG,"");
COLL_APPLY(COLL_SETTER,COLL_ALLTOALL_SIG,"");
COLL_APPLY(COLL_SETTER,COLL_ALLTOALLV_SIG,"");
-void Colls::set_collectives(){
+void colls::set_collectives()
+{
std::string selector_name = simgrid::config::get_value<std::string>("smpi/coll-selector");
if (selector_name.empty())
selector_name = "default";
- std::map<std::string, std::function<void(std::string)>> setter_callbacks = {
- {"gather", &Colls::set_gather}, {"allgather", &Colls::set_allgather},
- {"allgatherv", &Colls::set_allgatherv}, {"allreduce", &Colls::set_allreduce},
- {"alltoall", &Colls::set_alltoall}, {"alltoallv", &Colls::set_alltoallv},
- {"reduce", &Colls::set_reduce}, {"reduce_scatter", &Colls::set_reduce_scatter},
- {"scatter", &Colls::set_scatter}, {"bcast", &Colls::set_bcast},
- {"barrier", &Colls::set_barrier}};
-
- // This only prevents code duplication
- std::function<void(std::string)> setup = [selector_name, &setter_callbacks](std::string coll) {
- std::string name = simgrid::config::get_value<std::string>(("smpi/" + coll).c_str());
+ std::pair<std::string, std::function<void(std::string)>> setter_callbacks[] = {
+ {"gather", &colls::set_gather}, {"allgather", &colls::set_allgather},
+ {"allgatherv", &colls::set_allgatherv}, {"allreduce", &colls::set_allreduce},
+ {"alltoall", &colls::set_alltoall}, {"alltoallv", &colls::set_alltoallv},
+ {"reduce", &colls::set_reduce}, {"reduce_scatter", &colls::set_reduce_scatter},
+ {"scatter", &colls::set_scatter}, {"bcast", &colls::set_bcast},
+ {"barrier", &colls::set_barrier}};
+
+ for (auto& elem : setter_callbacks) {
+ std::string name = simgrid::config::get_value<std::string>(("smpi/" + elem.first).c_str());
if (name.empty())
name = selector_name;
- setter_callbacks[coll](name);
- };
-
- setup("gather");
- setup("allgather");
- setup("allgatherv");
- setup("allreduce");
- setup("alltoall");
- setup("alltoallv");
- setup("reduce");
- setup("reduce_scatter");
- setup("scatter");
- setup("bcast");
- setup("barrier");
+ (elem.second)(name);
+ }
}
+//Implementations of the single algorithm collectives
-//Implementations of the single algorith collectives
-
-int Colls::gatherv(void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int *recvcounts, int *displs,
- MPI_Datatype recvtype, int root, MPI_Comm comm)
+int colls::gatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts,
+ const int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm)
{
- int system_tag = COLL_TAG_GATHERV;
- MPI_Aint lb = 0;
- MPI_Aint recvext = 0;
-
- int rank = comm->rank();
- int size = comm->size();
- if (rank != root) {
- // Send buffer to root
- Request::send(sendbuf, sendcount, sendtype, root, system_tag, comm);
- } else {
- recvtype->extent(&lb, &recvext);
- // Local copy from root
- Datatype::copy(sendbuf, sendcount, sendtype, static_cast<char*>(recvbuf) + displs[root] * recvext,
- recvcounts[root], recvtype);
- // Receive buffers from senders
- MPI_Request *requests = xbt_new(MPI_Request, size - 1);
- int index = 0;
- for (int src = 0; src < size; src++) {
- if(src != root) {
- requests[index] = Request::irecv_init(static_cast<char*>(recvbuf) + displs[src] * recvext,
- recvcounts[src], recvtype, src, system_tag, comm);
- index++;
- }
- }
- // Wait for completion of irecv's.
- Request::startall(size - 1, requests);
- Request::waitall(size - 1, requests, MPI_STATUS_IGNORE);
- for (int src = 0; src < size-1; src++) {
- Request::unref(&requests[src]);
- }
- xbt_free(requests);
- }
- return MPI_SUCCESS;
+ MPI_Request request;
+ colls::igatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root, comm, &request, 0);
+ return Request::wait(&request, MPI_STATUS_IGNORE);
}
-
-int Colls::scatterv(void *sendbuf, int *sendcounts, int *displs, MPI_Datatype sendtype, void *recvbuf, int recvcount,
- MPI_Datatype recvtype, int root, MPI_Comm comm)
+int colls::scatterv(const void* sendbuf, const int* sendcounts, const int* displs, MPI_Datatype sendtype, void* recvbuf,
+ int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm)
{
- int system_tag = COLL_TAG_SCATTERV;
- MPI_Aint lb = 0;
- MPI_Aint sendext = 0;
-
- int rank = comm->rank();
- int size = comm->size();
- if(rank != root) {
- // Recv buffer from root
- Request::recv(recvbuf, recvcount, recvtype, root, system_tag, comm, MPI_STATUS_IGNORE);
- } else {
- sendtype->extent(&lb, &sendext);
- // Local copy from root
- if(recvbuf!=MPI_IN_PLACE){
- Datatype::copy(static_cast<char *>(sendbuf) + displs[root] * sendext, sendcounts[root],
- sendtype, recvbuf, recvcount, recvtype);
- }
- // Send buffers to receivers
- MPI_Request *requests = xbt_new(MPI_Request, size - 1);
- int index = 0;
- for (int dst = 0; dst < size; dst++) {
- if (dst != root) {
- requests[index] = Request::isend_init(static_cast<char *>(sendbuf) + displs[dst] * sendext, sendcounts[dst],
- sendtype, dst, system_tag, comm);
- index++;
- }
- }
- // Wait for completion of isend's.
- Request::startall(size - 1, requests);
- Request::waitall(size - 1, requests, MPI_STATUS_IGNORE);
- for (int dst = 0; dst < size-1; dst++) {
- Request::unref(&requests[dst]);
- }
- xbt_free(requests);
- }
- return MPI_SUCCESS;
+ MPI_Request request;
+ colls::iscatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm, &request, 0);
+ return Request::wait(&request, MPI_STATUS_IGNORE);
}
-
-int Colls::scan(void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
+int colls::scan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
{
int system_tag = -888;
MPI_Aint lb = 0;
Datatype::copy(sendbuf, count, datatype, recvbuf, count, datatype);
// Send/Recv buffers to/from others
- MPI_Request *requests = xbt_new(MPI_Request, size - 1);
- void **tmpbufs = xbt_new(void *, rank);
+ MPI_Request* requests = new MPI_Request[size - 1];
+ unsigned char** tmpbufs = new unsigned char*[rank];
int index = 0;
for (int other = 0; other < rank; other++) {
tmpbufs[index] = smpi_get_tmp_sendbuffer(count * dataext);
for(index = 0; index < size-1; index++) {
Request::unref(&requests[index]);
}
- xbt_free(tmpbufs);
- xbt_free(requests);
+ delete[] tmpbufs;
+ delete[] requests;
return MPI_SUCCESS;
}
-int Colls::exscan(void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
+int colls::exscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
{
int system_tag = -888;
MPI_Aint lb = 0;
datatype->extent(&lb, &dataext);
// Send/Recv buffers to/from others
- MPI_Request *requests = xbt_new(MPI_Request, size - 1);
- void **tmpbufs = xbt_new(void *, rank);
+ MPI_Request* requests = new MPI_Request[size - 1];
+ unsigned char** tmpbufs = new unsigned char*[rank];
int index = 0;
for (int other = 0; other < rank; other++) {
tmpbufs[index] = smpi_get_tmp_sendbuffer(count * dataext);
for(index = 0; index < size-1; index++) {
Request::unref(&requests[index]);
}
- xbt_free(tmpbufs);
- xbt_free(requests);
+ delete[] tmpbufs;
+ delete[] requests;
return MPI_SUCCESS;
}
+int colls::alltoallw(const void* sendbuf, const int* sendcounts, const int* senddisps, const MPI_Datatype* sendtypes,
+ void* recvbuf, const int* recvcounts, const int* recvdisps, const MPI_Datatype* recvtypes,
+ MPI_Comm comm)
+{
+ MPI_Request request;
+ colls::ialltoallw(sendbuf, sendcounts, senddisps, sendtypes, recvbuf, recvcounts, recvdisps, recvtypes, comm,
+ &request, 0);
+ return Request::wait(&request, MPI_STATUS_IGNORE);
+}
+
}
}
