X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/107a968d8f427b1cbf6aa5a5cbc35e01c4daee6d..98a4fb37cf38be98161ec6fc43ed0733eaefcbee:/src/smpi/internals/smpi_replay.cpp diff --git a/src/smpi/internals/smpi_replay.cpp b/src/smpi/internals/smpi_replay.cpp index ee61fe5712..2af987f919 100644 --- a/src/smpi/internals/smpi_replay.cpp +++ b/src/smpi/internals/smpi_replay.cpp @@ -1,4 +1,4 @@ -/* 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. */ @@ -61,7 +61,7 @@ public: }; } -typedef std::tuple req_key_t; +typedef std::tuple req_key_t; typedef std::unordered_map>> req_storage_t; void log_timed_action(simgrid::xbt::ReplayAction& action, double clock) @@ -73,7 +73,7 @@ void log_timed_action(simgrid::xbt::ReplayAction& action, double clock) } /* Helper function */ -static double parse_double(std::string string) +static double parse_double(const std::string& string) { return xbt_str_parse_double(string.c_str(), "%s is not a double"); } @@ -134,11 +134,14 @@ public: /* Sometimes we need to re-insert MPI_REQUEST_NULL but we still need src,dst and tag */ void addNullRequest(int src, int dst, int tag) { - store.insert({req_key_t(src, dst, tag), MPI_REQUEST_NULL}); + store.insert({req_key_t( + MPI_COMM_WORLD->group()->actor(src)->get_pid()-1, + MPI_COMM_WORLD->group()->actor(dst)->get_pid()-1, + tag), MPI_REQUEST_NULL}); } }; -void WaitTestParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void WaitTestParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { CHECK_ACTION_PARAMS(action, 3, 0) src = std::stoi(action[2]); @@ -146,7 +149,7 @@ void WaitTestParser::parse(simgrid::xbt::ReplayAction& action, std::string name) tag = std::stoi(action[4]); } -void SendRecvParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void SendRecvParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { CHECK_ACTION_PARAMS(action, 3, 1) partner = std::stoi(action[2]); @@ -156,13 +159,26 @@ void SendRecvParser::parse(simgrid::xbt::ReplayAction& action, std::string name) datatype1 = simgrid::smpi::Datatype::decode(action[5]); } -void ComputeParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void ComputeParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { CHECK_ACTION_PARAMS(action, 1, 0) flops = parse_double(action[2]); } -void BcastArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void SleepParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) +{ + CHECK_ACTION_PARAMS(action, 1, 0) + time = parse_double(action[2]); +} + +void LocationParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) +{ + CHECK_ACTION_PARAMS(action, 2, 0) + filename = std::string(action[2]); + line = std::stoi(action[3]); +} + +void BcastArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { CHECK_ACTION_PARAMS(action, 1, 2) size = parse_double(action[2]); @@ -171,7 +187,7 @@ void BcastArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) datatype1 = simgrid::smpi::Datatype::decode(action[4]); } -void ReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void ReduceArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { CHECK_ACTION_PARAMS(action, 2, 2) comm_size = parse_double(action[2]); @@ -181,7 +197,7 @@ void ReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name datatype1 = simgrid::smpi::Datatype::decode(action[5]); } -void AllReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void AllReduceArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { CHECK_ACTION_PARAMS(action, 2, 1) comm_size = parse_double(action[2]); @@ -190,7 +206,7 @@ void AllReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string n datatype1 = simgrid::smpi::Datatype::decode(action[4]); } -void AllToAllArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void AllToAllArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { CHECK_ACTION_PARAMS(action, 2, 1) comm_size = MPI_COMM_WORLD->size(); @@ -203,7 +219,7 @@ void AllToAllArgParser::parse(simgrid::xbt::ReplayAction& action, std::string na datatype2 = simgrid::smpi::Datatype::decode(action[5]); } -void GatherArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void GatherArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string& name) { /* The structure of the gather action for the rank 0 (total 4 processes) is the following: 0 gather 68 68 0 0 0 @@ -233,7 +249,7 @@ void GatherArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name } } -void GatherVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void GatherVArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string& name) { /* The structure of the gatherv action for the rank 0 (total 4 processes) is the following: 0 gather 68 68 10 10 10 0 0 0 @@ -257,13 +273,12 @@ void GatherVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string nam if (action.size() > 5 + comm_size) datatype2 = simgrid::smpi::Datatype::decode(action[5 + comm_size]); } else { - int datatype_index = 0; int disp_index = 0; /* The 3 comes from "0 gather ", which must always be present. * The + comm_size is the recvcounts array, which must also be present */ if (action.size() > 3 + comm_size + comm_size) { /* datatype + disp are specified */ - datatype_index = 3 + comm_size; + int datatype_index = 3 + comm_size; disp_index = datatype_index + 1; datatype1 = simgrid::smpi::Datatype::decode(action[datatype_index]); datatype2 = simgrid::smpi::Datatype::decode(action[datatype_index]); @@ -271,7 +286,7 @@ void GatherVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string nam 3 + comm_size + 2) { /* disps specified; datatype is not specified; use the default one */ disp_index = 3 + comm_size; } else if (action.size() > 3 + comm_size) { /* only datatype, no disp specified */ - datatype_index = 3 + comm_size; + int datatype_index = 3 + comm_size; datatype1 = simgrid::smpi::Datatype::decode(action[datatype_index]); datatype2 = simgrid::smpi::Datatype::decode(action[datatype_index]); } @@ -288,7 +303,7 @@ void GatherVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string nam recv_size_sum = std::accumulate(recvcounts->begin(), recvcounts->end(), 0); } -void ScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void ScatterArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { /* The structure of the scatter action for the rank 0 (total 4 processes) is the following: 0 gather 68 68 0 0 0 @@ -310,7 +325,7 @@ void ScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::string nam datatype2 = simgrid::smpi::Datatype::decode(action[6]); } -void ScatterVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void ScatterVArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { /* The structure of the scatterv action for the rank 0 (total 4 processes) is the following: 0 gather 68 10 10 10 68 0 0 0 @@ -338,7 +353,7 @@ void ScatterVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string na root = (action.size() > 3 + comm_size) ? std::stoi(action[3 + comm_size]) : 0; } -void ReduceScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void ReduceScatterArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { /* The structure of the reducescatter action for the rank 0 (total 4 processes) is the following: 0 reducescatter 275427 275427 275427 204020 11346849 0 @@ -360,10 +375,10 @@ void ReduceScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::stri recv_size_sum = std::accumulate(recvcounts->begin(), recvcounts->end(), 0); } -void AllToAllVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name) +void AllToAllVArgParser::parse(simgrid::xbt::ReplayAction& action, const std::string&) { - /* The structure of the allToAllV action for the rank 0 (total 4 processes) is the following: - 0 allToAllV 100 1 7 10 12 100 1 70 10 5 + /* The structure of the alltoallv action for the rank 0 (total 4 processes) is the following: + 0 alltoallv 100 1 7 10 12 100 1 70 10 5 where: 1) 100 is the size of the send buffer *sizeof(int), 2) 1 7 10 12 is the sendcounts array @@ -421,7 +436,7 @@ void WaitAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_recv(args.src, args.dst, args.tag); } -void SendAction::kernel(simgrid::xbt::ReplayAction& action) +void SendAction::kernel(simgrid::xbt::ReplayAction&) { int dst_traced = MPI_COMM_WORLD->group()->actor(args.partner)->get_pid(); @@ -432,7 +447,7 @@ void SendAction::kernel(simgrid::xbt::ReplayAction& action) if (name == "send") { Request::send(nullptr, args.size, args.datatype1, args.partner, args.tag, MPI_COMM_WORLD); - } else if (name == "Isend") { + } else if (name == "isend") { MPI_Request request = Request::isend(nullptr, args.size, args.datatype1, args.partner, args.tag, MPI_COMM_WORLD); req_storage.add(request); } else { @@ -442,10 +457,8 @@ void SendAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_comm_out(my_proc_id); } -void RecvAction::kernel(simgrid::xbt::ReplayAction& action) +void RecvAction::kernel(simgrid::xbt::ReplayAction&) { - int src_traced = MPI_COMM_WORLD->group()->actor(args.partner)->get_pid(); - TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::Pt2PtTIData(name, args.partner, args.size, args.tag, Datatype::encode(args.datatype1))); @@ -456,28 +469,46 @@ void RecvAction::kernel(simgrid::xbt::ReplayAction& action) args.size = status.count; } + bool is_recv = false; // Help analyzers understanding that status is not used unintialized if (name == "recv") { + is_recv = true; Request::recv(nullptr, args.size, args.datatype1, args.partner, args.tag, MPI_COMM_WORLD, &status); - } else if (name == "Irecv") { + } else if (name == "irecv") { MPI_Request request = Request::irecv(nullptr, args.size, args.datatype1, args.partner, args.tag, MPI_COMM_WORLD); req_storage.add(request); + } else { + THROW_IMPOSSIBLE; } TRACE_smpi_comm_out(my_proc_id); - // TODO: Check why this was only activated in the "recv" case and not in the "Irecv" case - if (name == "recv" && not TRACE_smpi_view_internals()) { + if (is_recv && not TRACE_smpi_view_internals()) { + int src_traced = MPI_COMM_WORLD->group()->actor(status.MPI_SOURCE)->get_pid(); TRACE_smpi_recv(src_traced, my_proc_id, args.tag); } } -void ComputeAction::kernel(simgrid::xbt::ReplayAction& action) +void ComputeAction::kernel(simgrid::xbt::ReplayAction&) { - TRACE_smpi_computing_in(my_proc_id, args.flops); - smpi_execute_flops(args.flops); - TRACE_smpi_computing_out(my_proc_id); + if (simgrid::config::get_value("smpi/simulate-computation")) { + smpi_execute_flops(args.flops/smpi_adjust_comp_speed()); + } } -void TestAction::kernel(simgrid::xbt::ReplayAction& action) +void SleepAction::kernel(simgrid::xbt::ReplayAction&) +{ + XBT_DEBUG("Sleep for: %lf secs", args.time); + int rank = simgrid::s4u::this_actor::get_pid(); + TRACE_smpi_sleeping_in(rank, args.time); + simgrid::s4u::this_actor::sleep_for(args.time/smpi_adjust_comp_speed()); + TRACE_smpi_sleeping_out(rank); +} + +void LocationAction::kernel(simgrid::xbt::ReplayAction&) +{ + smpi_trace_set_call_location(args.filename.c_str(), args.line); +} + +void TestAction::kernel(simgrid::xbt::ReplayAction&) { MPI_Request request = req_storage.find(args.src, args.dst, args.tag); req_storage.remove(request); @@ -488,7 +519,8 @@ void TestAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("test")); MPI_Status status; - int flag = Request::test(&request, &status); + int flag = 0; + Request::test(&request, &status, &flag); XBT_DEBUG("MPI_Test result: %d", flag); /* push back request in vector to be caught by a subsequent wait. if the test did succeed, the request is now @@ -512,12 +544,12 @@ void InitAction::kernel(simgrid::xbt::ReplayAction& action) smpi_process()->simulated_start(); } -void CommunicatorAction::kernel(simgrid::xbt::ReplayAction& action) +void CommunicatorAction::kernel(simgrid::xbt::ReplayAction&) { /* nothing to do */ } -void WaitAllAction::kernel(simgrid::xbt::ReplayAction& action) +void WaitAllAction::kernel(simgrid::xbt::ReplayAction&) { const unsigned int count_requests = req_storage.size(); @@ -542,14 +574,14 @@ void WaitAllAction::kernel(simgrid::xbt::ReplayAction& action) } } -void BarrierAction::kernel(simgrid::xbt::ReplayAction& action) +void BarrierAction::kernel(simgrid::xbt::ReplayAction&) { TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("barrier")); Colls::barrier(MPI_COMM_WORLD); TRACE_smpi_comm_out(my_proc_id); } -void BcastAction::kernel(simgrid::xbt::ReplayAction& action) +void BcastAction::kernel(simgrid::xbt::ReplayAction&) { TRACE_smpi_comm_in(my_proc_id, "action_bcast", new simgrid::instr::CollTIData("bcast", MPI_COMM_WORLD->group()->actor(args.root)->get_pid(), @@ -560,7 +592,7 @@ void BcastAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_comm_out(my_proc_id); } -void ReduceAction::kernel(simgrid::xbt::ReplayAction& action) +void ReduceAction::kernel(simgrid::xbt::ReplayAction&) { TRACE_smpi_comm_in(my_proc_id, "action_reduce", new simgrid::instr::CollTIData("reduce", MPI_COMM_WORLD->group()->actor(args.root)->get_pid(), @@ -569,27 +601,27 @@ void ReduceAction::kernel(simgrid::xbt::ReplayAction& action) Colls::reduce(send_buffer(args.comm_size * args.datatype1->size()), recv_buffer(args.comm_size * args.datatype1->size()), args.comm_size, args.datatype1, MPI_OP_NULL, args.root, MPI_COMM_WORLD); - smpi_execute_flops(args.comp_size); + private_execute_flops(args.comp_size); TRACE_smpi_comm_out(my_proc_id); } -void AllReduceAction::kernel(simgrid::xbt::ReplayAction& action) +void AllReduceAction::kernel(simgrid::xbt::ReplayAction&) { TRACE_smpi_comm_in(my_proc_id, "action_allreduce", new simgrid::instr::CollTIData("allreduce", -1, args.comp_size, args.comm_size, -1, Datatype::encode(args.datatype1), "")); Colls::allreduce(send_buffer(args.comm_size * args.datatype1->size()), recv_buffer(args.comm_size * args.datatype1->size()), args.comm_size, args.datatype1, MPI_OP_NULL, MPI_COMM_WORLD); - smpi_execute_flops(args.comp_size); + private_execute_flops(args.comp_size); TRACE_smpi_comm_out(my_proc_id); } -void AllToAllAction::kernel(simgrid::xbt::ReplayAction& action) +void AllToAllAction::kernel(simgrid::xbt::ReplayAction&) { - TRACE_smpi_comm_in(my_proc_id, "action_allToAll", - new simgrid::instr::CollTIData("allToAll", -1, -1.0, args.send_size, args.recv_size, + TRACE_smpi_comm_in(my_proc_id, "action_alltoall", + new simgrid::instr::CollTIData("alltoall", -1, -1.0, args.send_size, args.recv_size, Datatype::encode(args.datatype1), Datatype::encode(args.datatype2))); @@ -600,7 +632,7 @@ void AllToAllAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_comm_out(my_proc_id); } -void GatherAction::kernel(simgrid::xbt::ReplayAction& action) +void GatherAction::kernel(simgrid::xbt::ReplayAction&) { TRACE_smpi_comm_in(my_proc_id, name.c_str(), new simgrid::instr::CollTIData(name, (name == "gather") ? args.root : -1, -1.0, args.send_size, args.recv_size, Datatype::encode(args.datatype1), Datatype::encode(args.datatype2))); @@ -617,7 +649,7 @@ void GatherAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_comm_out(my_proc_id); } -void GatherVAction::kernel(simgrid::xbt::ReplayAction& action) +void GatherVAction::kernel(simgrid::xbt::ReplayAction&) { int rank = MPI_COMM_WORLD->rank(); @@ -639,7 +671,7 @@ void GatherVAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_comm_out(my_proc_id); } -void ScatterAction::kernel(simgrid::xbt::ReplayAction& action) +void ScatterAction::kernel(simgrid::xbt::ReplayAction&) { int rank = MPI_COMM_WORLD->rank(); TRACE_smpi_comm_in(my_proc_id, "action_scatter", new simgrid::instr::CollTIData(name, args.root, -1.0, args.send_size, args.recv_size, @@ -652,7 +684,7 @@ void ScatterAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_comm_out(my_proc_id); } -void ScatterVAction::kernel(simgrid::xbt::ReplayAction& action) +void ScatterVAction::kernel(simgrid::xbt::ReplayAction&) { int rank = MPI_COMM_WORLD->rank(); TRACE_smpi_comm_in(my_proc_id, "action_scatterv", new simgrid::instr::VarCollTIData(name, args.root, -1, args.sendcounts, args.recv_size, @@ -667,7 +699,7 @@ void ScatterVAction::kernel(simgrid::xbt::ReplayAction& action) TRACE_smpi_comm_out(my_proc_id); } -void ReduceScatterAction::kernel(simgrid::xbt::ReplayAction& action) +void ReduceScatterAction::kernel(simgrid::xbt::ReplayAction&) { TRACE_smpi_comm_in(my_proc_id, "action_reducescatter", new simgrid::instr::VarCollTIData("reducescatter", -1, 0, nullptr, -1, args.recvcounts, @@ -678,15 +710,15 @@ void ReduceScatterAction::kernel(simgrid::xbt::ReplayAction& action) recv_buffer(args.recv_size_sum * args.datatype1->size()), args.recvcounts->data(), args.datatype1, MPI_OP_NULL, MPI_COMM_WORLD); - smpi_execute_flops(args.comp_size); + private_execute_flops(args.comp_size); TRACE_smpi_comm_out(my_proc_id); } -void AllToAllVAction::kernel(simgrid::xbt::ReplayAction& action) +void AllToAllVAction::kernel(simgrid::xbt::ReplayAction&) { TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::VarCollTIData( - "allToAllV", -1, args.send_size_sum, args.sendcounts, args.recv_size_sum, args.recvcounts, + "alltoallv", -1, args.send_size_sum, args.sendcounts, args.recv_size_sum, args.recvcounts, Datatype::encode(args.datatype1), Datatype::encode(args.datatype2))); Colls::alltoallv(send_buffer(args.send_buf_size * args.datatype1->size()), args.sendcounts->data(), args.senddisps.data(), args.datatype1, @@ -697,39 +729,43 @@ void AllToAllVAction::kernel(simgrid::xbt::ReplayAction& action) } // Replay Namespace }} // namespace simgrid::smpi -static std::vector storage; +static std::unordered_map storage; /** @brief Only initialize the replay, don't do it for real */ -void smpi_replay_init(int* argc, char*** argv) +void smpi_replay_init(const char* instance_id, int rank, double start_delay_flops) { - simgrid::smpi::Process::init(argc, argv); + xbt_assert(not smpi_process()->initializing()); + + simgrid::s4u::Actor::self()->set_property("instance_id", instance_id); + simgrid::s4u::Actor::self()->set_property("rank", std::to_string(rank)); + simgrid::smpi::ActorExt::init(); + smpi_process()->mark_as_initialized(); smpi_process()->set_replaying(true); int my_proc_id = simgrid::s4u::this_actor::get_pid(); - storage.resize(smpi_process_count()); TRACE_smpi_init(my_proc_id); TRACE_smpi_computing_init(my_proc_id); TRACE_smpi_comm_in(my_proc_id, "smpi_replay_run_init", new simgrid::instr::NoOpTIData("init")); TRACE_smpi_comm_out(my_proc_id); xbt_replay_action_register("init", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::InitAction().execute(action); }); - xbt_replay_action_register("finalize", [](simgrid::xbt::ReplayAction& action) { /* nothing to do */ }); + xbt_replay_action_register("finalize", [](simgrid::xbt::ReplayAction&) { /* nothing to do */ }); xbt_replay_action_register("comm_size", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::CommunicatorAction().execute(action); }); xbt_replay_action_register("comm_split",[](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::CommunicatorAction().execute(action); }); xbt_replay_action_register("comm_dup", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::CommunicatorAction().execute(action); }); - xbt_replay_action_register("send", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SendAction("send", storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); }); - xbt_replay_action_register("Isend", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SendAction("Isend", storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); }); - xbt_replay_action_register("recv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::RecvAction("recv", storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); }); - xbt_replay_action_register("Irecv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::RecvAction("Irecv", storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); }); - xbt_replay_action_register("test", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::TestAction(storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); }); - xbt_replay_action_register("wait", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::WaitAction(storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); }); - xbt_replay_action_register("waitall", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::WaitAllAction(storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); }); + xbt_replay_action_register("send", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SendAction("send", storage[simgrid::s4u::this_actor::get_pid()]).execute(action); }); + xbt_replay_action_register("isend", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SendAction("isend", storage[simgrid::s4u::this_actor::get_pid()]).execute(action); }); + xbt_replay_action_register("recv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::RecvAction("recv", storage[simgrid::s4u::this_actor::get_pid()]).execute(action); }); + xbt_replay_action_register("irecv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::RecvAction("irecv", storage[simgrid::s4u::this_actor::get_pid()]).execute(action); }); + xbt_replay_action_register("test", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::TestAction(storage[simgrid::s4u::this_actor::get_pid()]).execute(action); }); + xbt_replay_action_register("wait", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::WaitAction(storage[simgrid::s4u::this_actor::get_pid()]).execute(action); }); + xbt_replay_action_register("waitall", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::WaitAllAction(storage[simgrid::s4u::this_actor::get_pid()]).execute(action); }); xbt_replay_action_register("barrier", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::BarrierAction().execute(action); }); xbt_replay_action_register("bcast", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::BcastAction().execute(action); }); xbt_replay_action_register("reduce", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ReduceAction().execute(action); }); xbt_replay_action_register("allreduce", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllReduceAction().execute(action); }); - xbt_replay_action_register("allToAll", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllToAllAction().execute(action); }); - xbt_replay_action_register("allToAllV", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllToAllVAction().execute(action); }); + xbt_replay_action_register("alltoall", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllToAllAction().execute(action); }); + xbt_replay_action_register("alltoallv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllToAllVAction().execute(action); }); xbt_replay_action_register("gather", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherAction("gather").execute(action); }); xbt_replay_action_register("scatter", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ScatterAction().execute(action); }); xbt_replay_action_register("gatherv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherVAction("gatherv").execute(action); }); @@ -738,36 +774,38 @@ void smpi_replay_init(int* argc, char*** argv) xbt_replay_action_register("allgatherv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherVAction("allgatherv").execute(action); }); xbt_replay_action_register("reducescatter", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ReduceScatterAction().execute(action); }); xbt_replay_action_register("compute", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ComputeAction().execute(action); }); + xbt_replay_action_register("sleep", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SleepAction().execute(action); }); + xbt_replay_action_register("location", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::LocationAction().execute(action); }); //if we have a delayed start, sleep here. - if(*argc>2){ - double value = xbt_str_parse_double((*argv)[2], "%s is not a double"); - XBT_VERB("Delayed start for instance - Sleeping for %f flops ",value ); - smpi_execute_flops(value); + if (start_delay_flops > 0) { + XBT_VERB("Delayed start for instance - Sleeping for %f flops ", start_delay_flops); + private_execute_flops(start_delay_flops); } else { - //UGLY: force a context switch to be sure that all MSG_processes begin initialization - XBT_DEBUG("Force context switch by smpi_execute_flops - Sleeping for 0.0 flops "); - smpi_execute_flops(0.0); + // Wait for the other actors to initialize also + simgrid::s4u::this_actor::yield(); } } /** @brief actually run the replay after initialization */ -void smpi_replay_main(int* argc, char*** argv) +void smpi_replay_main(int rank, const char* trace_filename) { static int active_processes = 0; active_processes++; - simgrid::xbt::replay_runner(*argc, *argv); + storage[simgrid::s4u::this_actor::get_pid()] = simgrid::smpi::replay::RequestStorage(); + std::string rank_string = std::to_string(rank); + simgrid::xbt::replay_runner(rank_string.c_str(), trace_filename); /* and now, finalize everything */ /* One active process will stop. Decrease the counter*/ - unsigned int count_requests = storage[simgrid::s4u::this_actor::get_pid() - 1].size(); + unsigned int count_requests = storage[simgrid::s4u::this_actor::get_pid()].size(); XBT_DEBUG("There are %ud elements in reqq[*]", count_requests); if (count_requests > 0) { MPI_Request requests[count_requests]; MPI_Status status[count_requests]; unsigned int i=0; - for (auto const& pair : storage[simgrid::s4u::this_actor::get_pid() - 1].get_store()) { + for (auto const& pair : storage[simgrid::s4u::this_actor::get_pid()].get_store()) { requests[i] = pair.second; i++; } @@ -787,12 +825,11 @@ void smpi_replay_main(int* argc, char*** argv) smpi_process()->finalize(); TRACE_smpi_comm_out(simgrid::s4u::this_actor::get_pid()); - TRACE_smpi_finalize(simgrid::s4u::this_actor::get_pid()); } /** @brief chain a replay initialization and a replay start */ -void smpi_replay_run(int* argc, char*** argv) +void smpi_replay_run(const char* instance_id, int rank, double start_delay_flops, const char* trace_filename) { - smpi_replay_init(argc, argv); - smpi_replay_main(argc, argv); + smpi_replay_init(instance_id, rank, start_delay_flops); + smpi_replay_main(rank, trace_filename); }