X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/eca1c9c7300542a49ee7c635dc10a678a87aca43..5c59b7149d98aa9375bc6d89bbfad43e46aaa7f0:/src/smpi/internals/smpi_replay.cpp diff --git a/src/smpi/internals/smpi_replay.cpp b/src/smpi/internals/smpi_replay.cpp index d4102d2c9f..0897ed1e98 100644 --- a/src/smpi/internals/smpi_replay.cpp +++ b/src/smpi/internals/smpi_replay.cpp @@ -17,6 +17,8 @@ #define KEY_SIZE (sizeof(int) * 2 + 1) +using simgrid::s4u::Actor; + XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_replay,smpi,"Trace Replay with SMPI"); int communicator_size = 0; @@ -41,12 +43,12 @@ static void log_timed_action (const char *const *action, double clock){ static std::vector* get_reqq_self() { - return reqq.at(smpi_process()->index()); + return reqq.at(Actor::self()->getPid()); } static void set_reqq_self(std::vector *mpi_request) { - reqq.insert({smpi_process()->index(), mpi_request}); + reqq.insert({Actor::self()->getPid(), mpi_request}); } //allocate a single buffer for all sends, growing it if needed @@ -87,6 +89,8 @@ static double parse_double(const char *string) return value; } + +//TODO: this logic should be moved inside the datatype class, to support all predefined types and get rid of is_replayable. static MPI_Datatype decode_datatype(const char *const action) { switch(atoi(action)) { @@ -196,15 +200,12 @@ static void action_compute(const char *const *action) CHECK_ACTION_PARAMS(action, 1, 0) double clock = smpi_process()->simulated_elapsed(); double flops= parse_double(action[2]); - int rank = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type=TRACING_COMPUTING; - extra->comp_size=flops; - TRACE_smpi_computing_in(rank, extra); + int rank = Actor::self()->getPid(); + TRACE_smpi_computing_in(rank, flops); smpi_execute_flops(flops); - TRACE_smpi_computing_out(rank); + log_timed_action (action, clock); } @@ -220,24 +221,19 @@ static void action_send(const char *const *action) else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; - int rank = smpi_process()->index(); + int rank = Actor::self()->getPid(); + int dst_traced = MPI_COMM_WORLD->group()->actor(to)->getPid(); - int dst_traced = MPI_COMM_WORLD->group()->rank(to); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_SEND; - extra->send_size = size; - extra->src = rank; - extra->dst = dst_traced; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - TRACE_smpi_ptp_in(rank, __FUNCTION__, extra); + TRACE_smpi_comm_in(rank, __FUNCTION__, + new simgrid::instr::Pt2PtTIData("send", to, size, encode_datatype(MPI_CURRENT_TYPE))); if (not TRACE_smpi_view_internals()) TRACE_smpi_send(rank, rank, dst_traced, 0, size*MPI_CURRENT_TYPE->size()); Request::send(nullptr, size, MPI_CURRENT_TYPE, to , 0, MPI_COMM_WORLD); - log_timed_action (action, clock); + TRACE_smpi_comm_out(rank); - TRACE_smpi_ptp_out(rank); + log_timed_action(action, clock); } static void action_Isend(const char *const *action) @@ -252,21 +248,16 @@ static void action_Isend(const char *const *action) else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; - int rank = smpi_process()->index(); - int dst_traced = MPI_COMM_WORLD->group()->rank(to); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_ISEND; - extra->send_size = size; - extra->src = rank; - extra->dst = dst_traced; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - TRACE_smpi_ptp_in(rank, __FUNCTION__, extra); + int rank = Actor::self()->getPid(); + int dst_traced = MPI_COMM_WORLD->group()->actor(to)->getPid(); + TRACE_smpi_comm_in(rank, __FUNCTION__, + new simgrid::instr::Pt2PtTIData("Isend", to, size, encode_datatype(MPI_CURRENT_TYPE))); if (not TRACE_smpi_view_internals()) TRACE_smpi_send(rank, rank, dst_traced, 0, size*MPI_CURRENT_TYPE->size()); MPI_Request request = Request::isend(nullptr, size, MPI_CURRENT_TYPE, to, 0, MPI_COMM_WORLD); - TRACE_smpi_ptp_out(rank); + TRACE_smpi_comm_out(rank); get_reqq_self()->push_back(request); @@ -285,26 +276,21 @@ static void action_recv(const char *const *action) { else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; - int rank = smpi_process()->index(); - int src_traced = MPI_COMM_WORLD->group()->rank(from); + int rank = Actor::self()->getPid(); + int src_traced = MPI_COMM_WORLD->group()->actor(from)->getPid(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_RECV; - extra->send_size = size; - extra->src = src_traced; - extra->dst = rank; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - TRACE_smpi_ptp_in(rank, __FUNCTION__, extra); + TRACE_smpi_comm_in(rank, __FUNCTION__, + new simgrid::instr::Pt2PtTIData("recv", from, size, encode_datatype(MPI_CURRENT_TYPE))); //unknown size from the receiver point of view - if(size<=0.0){ + if (size <= 0.0) { Request::probe(from, 0, MPI_COMM_WORLD, &status); size=status.count; } Request::recv(nullptr, size, MPI_CURRENT_TYPE, from, 0, MPI_COMM_WORLD, &status); - TRACE_smpi_ptp_out(rank); + TRACE_smpi_comm_out(rank); if (not TRACE_smpi_view_internals()) { TRACE_smpi_recv(src_traced, rank, 0); } @@ -324,31 +310,26 @@ static void action_Irecv(const char *const *action) else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; - int rank = smpi_process()->index(); - int src_traced = MPI_COMM_WORLD->group()->rank(from); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_IRECV; - extra->send_size = size; - extra->src = src_traced; - extra->dst = rank; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - TRACE_smpi_ptp_in(rank, __FUNCTION__, extra); + int rank = Actor::self()->getPid(); + TRACE_smpi_comm_in(rank, __FUNCTION__, + new simgrid::instr::Pt2PtTIData("Irecv", from, size, encode_datatype(MPI_CURRENT_TYPE))); MPI_Status status; //unknow size from the receiver pov - if(size<=0.0){ - Request::probe(from, 0, MPI_COMM_WORLD, &status); - size=status.count; + if (size <= 0.0) { + Request::probe(from, 0, MPI_COMM_WORLD, &status); + size = status.count; } MPI_Request request = Request::irecv(nullptr, size, MPI_CURRENT_TYPE, from, 0, MPI_COMM_WORLD); - TRACE_smpi_ptp_out(rank); + TRACE_smpi_comm_out(rank); get_reqq_self()->push_back(request); log_timed_action (action, clock); } -static void action_test(const char *const *action){ +static void action_test(const char* const* action) +{ CHECK_ACTION_PARAMS(action, 0, 0) double clock = smpi_process()->simulated_elapsed(); MPI_Status status; @@ -359,10 +340,8 @@ static void action_test(const char *const *action){ //Different times in traced application and replayed version may lead to this //In this case, ignore the extra calls. if(request!=nullptr){ - int rank = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type=TRACING_TEST; - TRACE_smpi_testing_in(rank, extra); + int rank = Actor::self()->getPid(); + TRACE_smpi_testing_in(rank); int flag = Request::test(&request, &status); @@ -396,13 +375,11 @@ static void action_wait(const char *const *action){ int src_traced = group->rank(request->src()); int dst_traced = group->rank(request->dst()); int is_wait_for_receive = (request->flags() & RECV); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_WAIT; - TRACE_smpi_ptp_in(rank, __FUNCTION__, extra); + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::NoOpTIData("wait")); Request::wait(&request, &status); - TRACE_smpi_ptp_out(rank); + TRACE_smpi_comm_out(rank); if (is_wait_for_receive) TRACE_smpi_recv(src_traced, dst_traced, 0); log_timed_action (action, clock); @@ -416,11 +393,8 @@ static void action_waitall(const char *const *action){ if (count_requests>0) { MPI_Status status[count_requests]; - int rank_traced = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_WAITALL; - extra->send_size=count_requests; - TRACE_smpi_ptp_in(rank_traced, __FUNCTION__,extra); + int rank_traced = Actor::self()->getPid(); + TRACE_smpi_comm_in(rank_traced, __FUNCTION__, new simgrid::instr::Pt2PtTIData("waitAll", -1, count_requests, "")); int recvs_snd[count_requests]; int recvs_rcv[count_requests]; for (unsigned int i = 0; i < count_requests; i++) { @@ -437,21 +411,19 @@ static void action_waitall(const char *const *action){ if (recvs_snd[i]!=-100) TRACE_smpi_recv(recvs_snd[i], recvs_rcv[i],0); } - TRACE_smpi_ptp_out(rank_traced); + TRACE_smpi_comm_out(rank_traced); } log_timed_action (action, clock); } static void action_barrier(const char *const *action){ double clock = smpi_process()->simulated_elapsed(); - int rank = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_BARRIER; - TRACE_smpi_collective_in(rank, __FUNCTION__, extra); + int rank = Actor::self()->getPid(); + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::NoOpTIData("barrier")); Colls::barrier(MPI_COMM_WORLD); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -470,20 +442,16 @@ static void action_bcast(const char *const *action) MPI_CURRENT_TYPE=decode_datatype(action[4]); } - int rank = smpi_process()->index(); - int root_traced = MPI_COMM_WORLD->group()->index(root); + int rank = Actor::self()->getPid(); + TRACE_smpi_comm_in(rank, __FUNCTION__, + new simgrid::instr::CollTIData("bcast", MPI_COMM_WORLD->group()->actor(root)->getPid(), -1.0, size, -1, + encode_datatype(MPI_CURRENT_TYPE), "")); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_BCAST; - extra->send_size = size; - extra->root = root_traced; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - TRACE_smpi_collective_in(rank, __FUNCTION__, extra); void *sendbuf = smpi_get_tmp_sendbuffer(size* MPI_CURRENT_TYPE->size()); Colls::bcast(sendbuf, size, MPI_CURRENT_TYPE, root, MPI_COMM_WORLD); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -502,23 +470,17 @@ static void action_reduce(const char *const *action) MPI_CURRENT_TYPE=decode_datatype(action[5]); } - int rank = smpi_process()->index(); - int root_traced = MPI_COMM_WORLD->group()->rank(root); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_REDUCE; - extra->send_size = comm_size; - extra->comp_size = comp_size; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - extra->root = root_traced; - - TRACE_smpi_collective_in(rank, __FUNCTION__,extra); + int rank = Actor::self()->getPid(); + TRACE_smpi_comm_in(rank, __FUNCTION__, + new simgrid::instr::CollTIData("reduce", MPI_COMM_WORLD->group()->actor(root)->getPid(), comp_size, + comm_size, -1, encode_datatype(MPI_CURRENT_TYPE), "")); void *recvbuf = smpi_get_tmp_sendbuffer(comm_size* MPI_CURRENT_TYPE->size()); void *sendbuf = smpi_get_tmp_sendbuffer(comm_size* MPI_CURRENT_TYPE->size()); Colls::reduce(sendbuf, recvbuf, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, root, MPI_COMM_WORLD); smpi_execute_flops(comp_size); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -533,20 +495,16 @@ static void action_allReduce(const char *const *action) { MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; double clock = smpi_process()->simulated_elapsed(); - int rank = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_ALLREDUCE; - extra->send_size = comm_size; - extra->comp_size = comp_size; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - TRACE_smpi_collective_in(rank, __FUNCTION__,extra); + int rank = Actor::self()->getPid(); + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::CollTIData("allReduce", -1, comp_size, comm_size, -1, + encode_datatype(MPI_CURRENT_TYPE), "")); void *recvbuf = smpi_get_tmp_sendbuffer(comm_size* MPI_CURRENT_TYPE->size()); void *sendbuf = smpi_get_tmp_sendbuffer(comm_size* MPI_CURRENT_TYPE->size()); Colls::allreduce(sendbuf, recvbuf, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, MPI_COMM_WORLD); smpi_execute_flops(comp_size); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -568,19 +526,14 @@ static void action_allToAll(const char *const *action) { void *send = smpi_get_tmp_sendbuffer(send_size*comm_size* MPI_CURRENT_TYPE->size()); void *recv = smpi_get_tmp_recvbuffer(recv_size*comm_size* MPI_CURRENT_TYPE2->size()); - int rank = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_ALLTOALL; - extra->send_size = send_size; - extra->recv_size = recv_size; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2); - - TRACE_smpi_collective_in(rank, __FUNCTION__,extra); + int rank = Actor::self()->getPid(); + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::CollTIData("allToAll", -1, -1.0, send_size, recv_size, + encode_datatype(MPI_CURRENT_TYPE), + encode_datatype(MPI_CURRENT_TYPE2))); Colls::alltoall(send, send_size, MPI_CURRENT_TYPE, recv, recv_size, MPI_CURRENT_TYPE2, MPI_COMM_WORLD); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -616,22 +569,59 @@ static void action_gather(const char *const *action) { if(rank==root) recv = smpi_get_tmp_recvbuffer(recv_size*comm_size* MPI_CURRENT_TYPE2->size()); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_GATHER; - extra->send_size = send_size; - extra->recv_size = recv_size; - extra->root = root; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2); - - TRACE_smpi_collective_in(smpi_process()->index(), __FUNCTION__, extra); + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::CollTIData("gather", root, -1.0, send_size, recv_size, + encode_datatype(MPI_CURRENT_TYPE), + encode_datatype(MPI_CURRENT_TYPE2))); Colls::gather(send, send_size, MPI_CURRENT_TYPE, recv, recv_size, MPI_CURRENT_TYPE2, root, MPI_COMM_WORLD); - TRACE_smpi_collective_out(smpi_process()->index()); + TRACE_smpi_comm_out(Actor::self()->getPid()); log_timed_action (action, clock); } +static void action_scatter(const char* const* action) +{ + /* The structure of the scatter action for the rank 0 (total 4 processes) is the following: + 0 gather 68 68 0 0 0 + where: + 1) 68 is the sendcounts + 2) 68 is the recvcounts + 3) 0 is the root node + 4) 0 is the send datatype id, see decode_datatype() + 5) 0 is the recv datatype id, see decode_datatype() + */ + CHECK_ACTION_PARAMS(action, 2, 3) + double clock = smpi_process()->simulated_elapsed(); + int comm_size = MPI_COMM_WORLD->size(); + int send_size = parse_double(action[2]); + int recv_size = parse_double(action[3]); + MPI_Datatype MPI_CURRENT_TYPE2 = MPI_DEFAULT_TYPE; + if (action[4] && action[5]) { + MPI_CURRENT_TYPE = decode_datatype(action[5]); + MPI_CURRENT_TYPE2 = decode_datatype(action[6]); + } else { + MPI_CURRENT_TYPE = MPI_DEFAULT_TYPE; + } + void* send = smpi_get_tmp_sendbuffer(send_size * MPI_CURRENT_TYPE->size()); + void* recv = nullptr; + int root = 0; + if (action[4]) + root = atoi(action[4]); + int rank = MPI_COMM_WORLD->rank(); + + if (rank == root) + recv = smpi_get_tmp_recvbuffer(recv_size * comm_size * MPI_CURRENT_TYPE2->size()); + + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::CollTIData("gather", root, -1.0, send_size, recv_size, + encode_datatype(MPI_CURRENT_TYPE), + encode_datatype(MPI_CURRENT_TYPE2))); + + Colls::scatter(send, send_size, MPI_CURRENT_TYPE, recv, recv_size, MPI_CURRENT_TYPE2, root, MPI_COMM_WORLD); + + TRACE_smpi_comm_out(Actor::self()->getPid()); + log_timed_action(action, clock); +} + static void action_gatherv(const char *const *action) { /* 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 @@ -671,25 +661,74 @@ static void action_gatherv(const char *const *action) { if(rank==root) recv = smpi_get_tmp_recvbuffer(recv_sum* MPI_CURRENT_TYPE2->size()); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_GATHERV; - extra->send_size = send_size; - extra->recvcounts = new int[comm_size]; - for(int i=0; i< comm_size; i++)//copy data to avoid bad free - extra->recvcounts[i] = recvcounts[i]; - extra->root = root; - extra->num_processes = comm_size; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2); + std::vector* trace_recvcounts = new std::vector; + for (int i = 0; i < comm_size; i++) // copy data to avoid bad free + trace_recvcounts->push_back(recvcounts[i]); - TRACE_smpi_collective_in(smpi_process()->index(), __FUNCTION__, extra); + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::VarCollTIData( + "gatherV", root, send_size, nullptr, -1, trace_recvcounts, + encode_datatype(MPI_CURRENT_TYPE), encode_datatype(MPI_CURRENT_TYPE2))); Colls::gatherv(send, send_size, MPI_CURRENT_TYPE, recv, recvcounts, disps, MPI_CURRENT_TYPE2, root, MPI_COMM_WORLD); - TRACE_smpi_collective_out(smpi_process()->index()); + TRACE_smpi_comm_out(Actor::self()->getPid()); log_timed_action (action, clock); } +static void action_scatterv(const char* const* action) +{ + /* 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 + where: + 1) 68 10 10 10 is the sendcounts + 2) 68 is the recvcount + 3) 0 is the root node + 4) 0 is the send datatype id, see decode_datatype() + 5) 0 is the recv datatype id, see decode_datatype() + */ + double clock = smpi_process()->simulated_elapsed(); + int comm_size = MPI_COMM_WORLD->size(); + CHECK_ACTION_PARAMS(action, comm_size + 1, 2) + int recv_size = parse_double(action[2 + comm_size]); + int disps[comm_size]; + int sendcounts[comm_size]; + int send_sum = 0; + + MPI_Datatype MPI_CURRENT_TYPE2 = MPI_DEFAULT_TYPE; + if (action[4 + comm_size] && action[5 + comm_size]) { + MPI_CURRENT_TYPE = decode_datatype(action[4 + comm_size]); + MPI_CURRENT_TYPE2 = decode_datatype(action[5 + comm_size]); + } else + MPI_CURRENT_TYPE = MPI_DEFAULT_TYPE; + + void* send = nullptr; + void* recv = smpi_get_tmp_recvbuffer(recv_size * MPI_CURRENT_TYPE->size()); + for (int i = 0; i < comm_size; i++) { + sendcounts[i] = atoi(action[i + 2]); + send_sum += sendcounts[i]; + disps[i] = 0; + } + + int root = atoi(action[3 + comm_size]); + int rank = MPI_COMM_WORLD->rank(); + + if (rank == root) + send = smpi_get_tmp_sendbuffer(send_sum * MPI_CURRENT_TYPE2->size()); + + std::vector* trace_sendcounts = new std::vector; + for (int i = 0; i < comm_size; i++) // copy data to avoid bad free + trace_sendcounts->push_back(sendcounts[i]); + + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::VarCollTIData( + "gatherV", root, -1, trace_sendcounts, recv_size, nullptr, + encode_datatype(MPI_CURRENT_TYPE), encode_datatype(MPI_CURRENT_TYPE2))); + + Colls::scatterv(send, sendcounts, disps, MPI_CURRENT_TYPE, recv, recv_size, MPI_CURRENT_TYPE2, root, MPI_COMM_WORLD); + + TRACE_smpi_comm_out(Actor::self()->getPid()); + log_timed_action(action, clock); +} + static void action_reducescatter(const char *const *action) { /* The structure of the reducescatter action for the rank 0 (total 4 processes) is the following: 0 reduceScatter 275427 275427 275427 204020 11346849 0 @@ -703,8 +742,9 @@ static void action_reducescatter(const char *const *action) { CHECK_ACTION_PARAMS(action, comm_size+1, 1) int comp_size = parse_double(action[2+comm_size]); int recvcounts[comm_size]; - int rank = smpi_process()->index(); + int rank = Actor::self()->getPid(); int size = 0; + std::vector* trace_recvcounts = new std::vector; if(action[3+comm_size]) MPI_CURRENT_TYPE=decode_datatype(action[3+comm_size]); else @@ -712,20 +752,14 @@ static void action_reducescatter(const char *const *action) { for(int i=0;ipush_back(recvcounts[i]); size+=recvcounts[i]; } - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_REDUCE_SCATTER; - extra->send_size = 0; - extra->recvcounts = new int[comm_size]; - for(int i=0; i< comm_size; i++)//copy data to avoid bad free - extra->recvcounts[i] = recvcounts[i]; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - extra->comp_size = comp_size; - extra->num_processes = comm_size; - - TRACE_smpi_collective_in(rank, __FUNCTION__,extra); + TRACE_smpi_comm_in(rank, __FUNCTION__, + new simgrid::instr::VarCollTIData("reduceScatter", -1, 0, nullptr, -1, trace_recvcounts, + std::to_string(comp_size), /* ugly hack to print comp_size */ + encode_datatype(MPI_CURRENT_TYPE))); void *sendbuf = smpi_get_tmp_sendbuffer(size* MPI_CURRENT_TYPE->size()); void *recvbuf = smpi_get_tmp_recvbuffer(size* MPI_CURRENT_TYPE->size()); @@ -733,7 +767,7 @@ static void action_reducescatter(const char *const *action) { Colls::reduce_scatter(sendbuf, recvbuf, recvcounts, MPI_CURRENT_TYPE, MPI_OP_NULL, MPI_COMM_WORLD); smpi_execute_flops(comp_size); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -762,20 +796,15 @@ static void action_allgather(const char *const *action) { void *sendbuf = smpi_get_tmp_sendbuffer(sendcount* MPI_CURRENT_TYPE->size()); void *recvbuf = smpi_get_tmp_recvbuffer(recvcount* MPI_CURRENT_TYPE2->size()); - int rank = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_ALLGATHER; - extra->send_size = sendcount; - extra->recv_size= recvcount; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2); - extra->num_processes = MPI_COMM_WORLD->size(); + int rank = Actor::self()->getPid(); - TRACE_smpi_collective_in(rank, __FUNCTION__,extra); + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::CollTIData("allGather", -1, -1.0, sendcount, recvcount, + encode_datatype(MPI_CURRENT_TYPE), + encode_datatype(MPI_CURRENT_TYPE2))); Colls::allgather(sendbuf, sendcount, MPI_CURRENT_TYPE, recvbuf, recvcount, MPI_CURRENT_TYPE2, MPI_COMM_WORLD); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -812,23 +841,20 @@ static void action_allgatherv(const char *const *action) { } void *recvbuf = smpi_get_tmp_recvbuffer(recv_sum* MPI_CURRENT_TYPE2->size()); - int rank = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_ALLGATHERV; - extra->send_size = sendcount; - extra->recvcounts = new int[comm_size]; - for(int i=0; i< comm_size; i++)//copy data to avoid bad free - extra->recvcounts[i] = recvcounts[i]; - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2); - extra->num_processes = comm_size; + int rank = Actor::self()->getPid(); - TRACE_smpi_collective_in(rank, __FUNCTION__,extra); + std::vector* trace_recvcounts = new std::vector; + for (int i = 0; i < comm_size; i++) // copy data to avoid bad free + trace_recvcounts->push_back(recvcounts[i]); + + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::VarCollTIData( + "allGatherV", -1, sendcount, nullptr, -1, trace_recvcounts, + encode_datatype(MPI_CURRENT_TYPE), encode_datatype(MPI_CURRENT_TYPE2))); Colls::allgatherv(sendbuf, sendcount, MPI_CURRENT_TYPE, recvbuf, recvcounts, disps, MPI_CURRENT_TYPE2, MPI_COMM_WORLD); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -845,8 +871,12 @@ static void action_allToAllv(const char *const *action) { int comm_size = MPI_COMM_WORLD->size(); CHECK_ACTION_PARAMS(action, 2*comm_size+2, 2) + int send_size = 0; + int recv_size = 0; int sendcounts[comm_size]; + std::vector* trace_sendcounts = new std::vector; int recvcounts[comm_size]; + std::vector* trace_recvcounts = new std::vector; int senddisps[comm_size]; int recvdisps[comm_size]; @@ -861,38 +891,29 @@ static void action_allToAllv(const char *const *action) { else MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE; + int rank = Actor::self()->getPid(); void *sendbuf = smpi_get_tmp_sendbuffer(send_buf_size* MPI_CURRENT_TYPE->size()); void *recvbuf = smpi_get_tmp_recvbuffer(recv_buf_size* MPI_CURRENT_TYPE2->size()); for(int i=0;ipush_back(sendcounts[i]); + send_size += sendcounts[i]; recvcounts[i] = atoi(action[i+4+comm_size]); + trace_recvcounts->push_back(recvcounts[i]); + recv_size += recvcounts[i]; senddisps[i] = 0; recvdisps[i] = 0; } - int rank = smpi_process()->index(); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_ALLTOALLV; - extra->recvcounts = new int[comm_size]; - extra->sendcounts = new int[comm_size]; - extra->num_processes = comm_size; - - for(int i=0; i< comm_size; i++){//copy data to avoid bad free - extra->send_size += sendcounts[i]; - extra->sendcounts[i] = sendcounts[i]; - extra->recv_size += recvcounts[i]; - extra->recvcounts[i] = recvcounts[i]; - } - extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); - extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2); - - TRACE_smpi_collective_in(rank, __FUNCTION__,extra); + TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::VarCollTIData( + "allToAllV", -1, send_size, trace_sendcounts, recv_size, trace_recvcounts, + encode_datatype(MPI_CURRENT_TYPE), encode_datatype(MPI_CURRENT_TYPE2))); Colls::alltoallv(sendbuf, sendcounts, senddisps, MPI_CURRENT_TYPE,recvbuf, recvcounts, recvdisps, MPI_CURRENT_TYPE, MPI_COMM_WORLD); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_out(rank); log_timed_action (action, clock); } @@ -905,13 +926,11 @@ void smpi_replay_init(int* argc, char*** argv) smpi_process()->mark_as_initialized(); smpi_process()->set_replaying(true); - int rank = smpi_process()->index(); + int rank = Actor::self()->getPid(); TRACE_smpi_init(rank); TRACE_smpi_computing_init(rank); - instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); - extra->type = TRACING_INIT; - TRACE_smpi_collective_in(rank, "smpi_replay_run_init", extra); - TRACE_smpi_collective_out(rank); + TRACE_smpi_comm_in(rank, "smpi_replay_run_init", new simgrid::instr::NoOpTIData("init")); + TRACE_smpi_comm_out(rank); xbt_replay_action_register("init", simgrid::smpi::action_init); xbt_replay_action_register("finalize", simgrid::smpi::action_finalize); xbt_replay_action_register("comm_size", simgrid::smpi::action_comm_size); @@ -931,7 +950,9 @@ void smpi_replay_init(int* argc, char*** argv) xbt_replay_action_register("allToAll", simgrid::smpi::action_allToAll); xbt_replay_action_register("allToAllV", simgrid::smpi::action_allToAllv); xbt_replay_action_register("gather", simgrid::smpi::action_gather); + xbt_replay_action_register("scatter", simgrid::smpi::action_scatter); xbt_replay_action_register("gatherV", simgrid::smpi::action_gatherv); + xbt_replay_action_register("scatterV", simgrid::smpi::action_scatterv); xbt_replay_action_register("allGather", simgrid::smpi::action_allgather); xbt_replay_action_register("allGatherV", simgrid::smpi::action_allgatherv); xbt_replay_action_register("reduceScatter", simgrid::smpi::action_reducescatter); @@ -979,14 +1000,12 @@ void smpi_replay_main(int* argc, char*** argv) xbt_free(recvbuffer); } - instr_extra_data extra_fin = xbt_new0(s_instr_extra_data_t,1); - extra_fin->type = TRACING_FINALIZE; - TRACE_smpi_collective_in(smpi_process()->index(), "smpi_replay_run_finalize", extra_fin); + TRACE_smpi_comm_in(Actor::self()->getPid(), "smpi_replay_run_finalize", new simgrid::instr::NoOpTIData("finalize")); smpi_process()->finalize(); - TRACE_smpi_collective_out(smpi_process()->index()); - TRACE_smpi_finalize(smpi_process()->index()); + TRACE_smpi_comm_out(Actor::self()->getPid()); + TRACE_smpi_finalize(Actor::self()->getPid()); } /** @brief chain a replay initialization and a replay start */