[1381] | 1 | #include "ep_lib.hpp" |
---|
| 2 | #include <mpi.h> |
---|
| 3 | #include "ep_declaration.hpp" |
---|
| 4 | #include "ep_mpi.hpp" |
---|
| 5 | |
---|
| 6 | using namespace std; |
---|
| 7 | |
---|
| 8 | namespace ep_lib |
---|
| 9 | { |
---|
| 10 | int MPI_Intercomm_create(MPI_Comm local_comm, int local_leader, MPI_Comm peer_comm, int remote_leader, int tag, MPI_Comm *newintercomm) |
---|
| 11 | { |
---|
[1500] | 12 | assert(local_comm->is_ep); |
---|
[1381] | 13 | |
---|
| 14 | int ep_rank, ep_rank_loc, mpi_rank; |
---|
| 15 | int ep_size, num_ep, mpi_size; |
---|
| 16 | |
---|
[1500] | 17 | ep_rank = local_comm->ep_comm_ptr->size_rank_info[0].first; |
---|
| 18 | ep_rank_loc = local_comm->ep_comm_ptr->size_rank_info[1].first; |
---|
| 19 | mpi_rank = local_comm->ep_comm_ptr->size_rank_info[2].first; |
---|
| 20 | ep_size = local_comm->ep_comm_ptr->size_rank_info[0].second; |
---|
| 21 | num_ep = local_comm->ep_comm_ptr->size_rank_info[1].second; |
---|
| 22 | mpi_size = local_comm->ep_comm_ptr->size_rank_info[2].second; |
---|
[1381] | 23 | |
---|
| 24 | |
---|
| 25 | MPI_Barrier(local_comm); |
---|
| 26 | |
---|
[1499] | 27 | |
---|
| 28 | |
---|
[1381] | 29 | int leader_ranks[6]; //! 0: rank in world, 1: mpi_size, 2: rank_in_peer. |
---|
| 30 | //! 3, 4, 5 : remote |
---|
| 31 | |
---|
| 32 | bool is_decider = false; |
---|
| 33 | |
---|
| 34 | |
---|
| 35 | if(ep_rank == local_leader) |
---|
| 36 | { |
---|
| 37 | MPI_Comm_rank(MPI_COMM_WORLD, &leader_ranks[0]); |
---|
| 38 | |
---|
| 39 | leader_ranks[1] = mpi_size; |
---|
| 40 | MPI_Comm_rank(peer_comm, &leader_ranks[2]); |
---|
| 41 | |
---|
| 42 | MPI_Request request[2]; |
---|
| 43 | MPI_Status status[2]; |
---|
| 44 | |
---|
| 45 | MPI_Isend(&leader_ranks[0], 3, MPI_INT, remote_leader, tag, peer_comm, &request[0]); |
---|
| 46 | MPI_Irecv(&leader_ranks[3], 3, MPI_INT, remote_leader, tag, peer_comm, &request[1]); |
---|
| 47 | |
---|
| 48 | MPI_Waitall(2, request, status); |
---|
| 49 | |
---|
| 50 | } |
---|
| 51 | |
---|
| 52 | |
---|
| 53 | MPI_Bcast(leader_ranks, 6, MPI_INT, local_leader, local_comm); |
---|
| 54 | |
---|
| 55 | |
---|
| 56 | MPI_Barrier(local_comm); |
---|
| 57 | |
---|
| 58 | |
---|
| 59 | if(leader_ranks[0] == leader_ranks[3]) |
---|
| 60 | { |
---|
| 61 | if( leader_ranks[1] * leader_ranks[4] == 1) |
---|
| 62 | { |
---|
| 63 | if(ep_rank == local_leader) Debug("calling MPI_Intercomm_create_unique_leader\n"); |
---|
[1500] | 64 | local_comm->ep_comm_ptr->comm_label = -99; |
---|
[1381] | 65 | |
---|
| 66 | return MPI_Intercomm_create_unique_leader(local_comm, local_leader, peer_comm, remote_leader, tag, newintercomm); |
---|
| 67 | } |
---|
| 68 | else // leader_ranks[1] * leader_ranks[4] != 1 |
---|
| 69 | { |
---|
| 70 | // change leader |
---|
| 71 | int new_local_leader; |
---|
| 72 | |
---|
| 73 | if(leader_ranks[2] < leader_ranks[5]) |
---|
| 74 | { |
---|
| 75 | if(leader_ranks[1] > 1) //! change leader |
---|
| 76 | { |
---|
| 77 | // change leader |
---|
| 78 | is_decider = true; |
---|
[1503] | 79 | int target = local_comm->ep_rank_map->at(local_leader).second; |
---|
[1381] | 80 | { |
---|
| 81 | for(int i=0; i<ep_size; i++) |
---|
| 82 | { |
---|
[1503] | 83 | if(local_comm->ep_rank_map->at(i).second != target && local_comm->ep_rank_map->at(i).first == 0) |
---|
[1381] | 84 | { |
---|
| 85 | new_local_leader = i; |
---|
| 86 | break; |
---|
| 87 | } |
---|
| 88 | } |
---|
| 89 | } |
---|
| 90 | } |
---|
| 91 | else |
---|
| 92 | { |
---|
| 93 | new_local_leader = local_leader; |
---|
| 94 | } |
---|
| 95 | } |
---|
| 96 | else |
---|
| 97 | { |
---|
| 98 | if(leader_ranks[4] == 1) |
---|
| 99 | { |
---|
| 100 | // change leader |
---|
| 101 | is_decider = true; |
---|
[1503] | 102 | int target = local_comm->ep_rank_map->at(local_leader).second; |
---|
[1381] | 103 | { |
---|
| 104 | for(int i=0; i<ep_size; i++) |
---|
| 105 | { |
---|
[1503] | 106 | if(local_comm->ep_rank_map->at(i).second != target && local_comm->ep_rank_map->at(i).first == 0) |
---|
[1381] | 107 | { |
---|
| 108 | new_local_leader = i; |
---|
| 109 | break; |
---|
| 110 | } |
---|
| 111 | } |
---|
| 112 | } |
---|
| 113 | } |
---|
| 114 | else |
---|
| 115 | { |
---|
| 116 | new_local_leader = local_leader; |
---|
| 117 | } |
---|
| 118 | } |
---|
| 119 | |
---|
| 120 | |
---|
| 121 | int new_tag_in_world; |
---|
| 122 | |
---|
| 123 | int leader_in_world[2]; |
---|
| 124 | |
---|
| 125 | |
---|
| 126 | if(is_decider) |
---|
| 127 | { |
---|
| 128 | if(ep_rank == new_local_leader) |
---|
| 129 | { |
---|
| 130 | new_tag_in_world = TAG++; |
---|
| 131 | } |
---|
| 132 | MPI_Bcast(&new_tag_in_world, 1, MPI_INT, new_local_leader, local_comm); |
---|
| 133 | if(ep_rank == local_leader) MPI_Send(&new_tag_in_world, 1, MPI_INT, remote_leader, tag, peer_comm); |
---|
| 134 | } |
---|
| 135 | else |
---|
| 136 | { |
---|
| 137 | if(ep_rank == local_leader) |
---|
| 138 | { |
---|
| 139 | MPI_Status status; |
---|
| 140 | MPI_Recv(&new_tag_in_world, 1, MPI_INT, remote_leader, tag, peer_comm, &status); |
---|
| 141 | } |
---|
| 142 | MPI_Bcast(&new_tag_in_world, 1, MPI_INT, new_local_leader, local_comm); |
---|
| 143 | } |
---|
| 144 | |
---|
| 145 | |
---|
| 146 | if(ep_rank == new_local_leader) |
---|
| 147 | { |
---|
[1500] | 148 | ::MPI_Comm_rank(to_mpi_comm(MPI_COMM_WORLD->mpi_comm), &leader_in_world[0]); |
---|
[1381] | 149 | } |
---|
| 150 | |
---|
| 151 | MPI_Bcast(&leader_in_world[0], 1, MPI_INT, new_local_leader, local_comm); |
---|
| 152 | |
---|
| 153 | |
---|
| 154 | if(ep_rank == local_leader) |
---|
| 155 | { |
---|
| 156 | MPI_Request request[2]; |
---|
| 157 | MPI_Status status[2]; |
---|
| 158 | |
---|
| 159 | MPI_Isend(&leader_in_world[0], 1, MPI_INT, remote_leader, tag, peer_comm, &request[0]); |
---|
| 160 | MPI_Irecv(&leader_in_world[1], 1, MPI_INT, remote_leader, tag, peer_comm, &request[1]); |
---|
| 161 | |
---|
| 162 | MPI_Waitall(2, request, status); |
---|
| 163 | } |
---|
| 164 | |
---|
| 165 | MPI_Bcast(&leader_in_world[1], 1, MPI_INT, local_leader, local_comm); |
---|
| 166 | |
---|
[1500] | 167 | local_comm->ep_comm_ptr->comm_label = tag; |
---|
[1381] | 168 | |
---|
| 169 | if(ep_rank == local_leader) Debug("calling MPI_Intercomm_create_from_world\n"); |
---|
| 170 | |
---|
[1500] | 171 | return MPI_Intercomm_create_from_world(local_comm, new_local_leader, MPI_COMM_WORLD->mpi_comm, leader_in_world[1], new_tag_in_world, newintercomm); |
---|
[1381] | 172 | |
---|
| 173 | } |
---|
| 174 | } |
---|
| 175 | |
---|
| 176 | if(ep_rank == local_leader) Debug("calling MPI_Intercomm_create_kernel\n"); |
---|
| 177 | |
---|
| 178 | return MPI_Intercomm_create_kernel(local_comm, local_leader, peer_comm, remote_leader, tag, newintercomm); |
---|
| 179 | |
---|
| 180 | } |
---|
| 181 | |
---|
| 182 | int MPI_Comm_test_inter(MPI_Comm comm, int *flag) |
---|
| 183 | { |
---|
| 184 | *flag = false; |
---|
[1500] | 185 | if(comm->is_ep) |
---|
[1381] | 186 | { |
---|
[1500] | 187 | *flag = comm->is_intercomm; |
---|
[1381] | 188 | return 0; |
---|
| 189 | } |
---|
[1500] | 190 | else if(comm->mpi_comm != static_cast< ::MPI_Comm*>(MPI_COMM_NULL->mpi_comm)) |
---|
[1381] | 191 | { |
---|
[1500] | 192 | ::MPI_Comm mpi_comm = to_mpi_comm(comm->mpi_comm); |
---|
[1381] | 193 | |
---|
| 194 | ::MPI_Comm_test_inter(mpi_comm, flag); |
---|
| 195 | return 0; |
---|
| 196 | } |
---|
| 197 | return 0; |
---|
| 198 | } |
---|
| 199 | |
---|
| 200 | |
---|
| 201 | } |
---|