| 1 |
// Copyright 2002 - 2008, 2010, 2011 National Technology Engineering |
| 2 |
// Solutions of Sandia, LLC (NTESS). Under the terms of Contract |
| 3 |
// DE-NA0003525 with NTESS, the U.S. Government retains certain rights |
| 4 |
// in this software. |
| 5 |
// |
| 6 |
// Redistribution and use in source and binary forms, with or without |
| 7 |
// modification, are permitted provided that the following conditions are |
| 8 |
// met: |
| 9 |
// |
| 10 |
// * Redistributions of source code must retain the above copyright |
| 11 |
// notice, this list of conditions and the following disclaimer. |
| 12 |
// |
| 13 |
// * Redistributions in binary form must reproduce the above |
| 14 |
// copyright notice, this list of conditions and the following |
| 15 |
// disclaimer in the documentation and/or other materials provided |
| 16 |
// with the distribution. |
| 17 |
// |
| 18 |
// * Neither the name of NTESS nor the names of its contributors |
| 19 |
// may be used to endorse or promote products derived from this |
| 20 |
// software without specific prior written permission. |
| 21 |
// |
| 22 |
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 23 |
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 24 |
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 25 |
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 26 |
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 27 |
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 28 |
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 29 |
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 30 |
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 31 |
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 32 |
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 33 |
// |
| 34 |
|
| 35 |
#include "gtest/gtest.h" |
| 36 |
#include "stk_util/parallel/CommBufferV.hpp" // for CommBufferV |
| 37 |
#include "stk_util/parallel/CommNeighbors.hpp" // for CommNeighbors, STK_MPI_SUPPORTS_NEIGHBOR_... |
| 38 |
#include "stk_util/parallel/Parallel.hpp" // for parallel_machine_rank, parallel_machine_size |
| 39 |
#include "stk_util/stk_config.h" // for STK_HAS_MPI |
| 40 |
#include <cstddef> // for size_t |
| 41 |
#include <vector> // for vector |
| 42 |
|
| 43 |
#if defined ( STK_HAS_MPI ) |
| 44 |
|
| 45 |
std::vector<int> get_neighbor_procs(int numAllProcs, int localProc, int numNeighbors) |
| 46 |
{ |
| 47 |
std::vector<int> neighbors; |
| 48 |
for(int proc=localProc-numNeighbors/2; proc<=localProc+numNeighbors/2; ++proc) { |
| 49 |
if (proc >= 0 && proc != localProc && proc < numAllProcs) { |
| 50 |
neighbors.push_back(proc); |
| 51 |
} |
| 52 |
} |
| 53 |
return neighbors; |
| 54 |
} |
| 55 |
|
| 56 |
TEST(Parallel, CommNeighborsOneSided_Raw_MPI) |
| 57 |
{ |
| 58 |
#ifdef OMPI_MAJOR_VERSION |
| 59 |
#if OMPI_MAJOR_VERSION < 2 |
| 60 |
#undef STK_MPI_SUPPORTS_NEIGHBOR_COMM |
| 61 |
#endif |
| 62 |
#endif |
| 63 |
|
| 64 |
#if !defined(STK_MPI_SUPPORTS_NEIGHBOR_COMM) |
| 65 |
return; |
| 66 |
#else |
| 67 |
|
| 68 |
MPI_Comm comm = MPI_COMM_WORLD; |
| 69 |
int numProcs = 1; |
| 70 |
MPI_Comm_size(comm, &numProcs); |
| 71 |
if (numProcs != 2) { |
| 72 |
return; |
| 73 |
} |
| 74 |
|
| 75 |
//this test will send from proc 0 to proc 1. |
| 76 |
//proc 0 will not recv anything and proc 1 will not send anything. |
| 77 |
//'sendProcs' means 'procs that I will send to' |
| 78 |
//'recvprocs' means 'procs that I will recv from' |
| 79 |
|
| 80 |
int localProc = 0; |
| 81 |
MPI_Comm_rank(comm, &localProc); |
| 82 |
int otherProc = 1 - localProc; |
| 83 |
std::vector<int> neighbors = {otherProc}; |
| 84 |
std::vector<int> sendProcs = {otherProc}; |
| 85 |
std::vector<int> recvProcs = {otherProc}; |
| 86 |
if (localProc==0) { |
| 87 |
recvProcs.clear(); |
| 88 |
} |
| 89 |
else { |
| 90 |
sendProcs.clear(); |
| 91 |
} |
| 92 |
|
| 93 |
MPI_Info info; |
| 94 |
MPI_Info_create(&info); |
| 95 |
int reorder = 0; |
| 96 |
const int* weights = (int*)MPI_UNWEIGHTED; |
| 97 |
|
| 98 |
MPI_Comm neighborComm; |
| 99 |
MPI_Dist_graph_create_adjacent(comm, |
| 100 |
neighbors.size(), neighbors.data(), weights, |
| 101 |
neighbors.size(), neighbors.data(), weights, |
| 102 |
info, reorder, &neighborComm); |
| 103 |
MPI_Info_free(&info); |
| 104 |
|
| 105 |
int numItems = 5; |
| 106 |
std::vector<double> sendBuf(numItems, 0), recvBuf; |
| 107 |
std::vector<int> sendCounts(neighbors.size(), 0), recvCounts(neighbors.size()); |
| 108 |
std::vector<int> sendDispls(neighbors.size(), 0), recvDispls(neighbors.size()); |
| 109 |
|
| 110 |
for(size_t i=0; i<sendProcs.size(); ++i) { |
| 111 |
sendCounts[i] = numItems; |
| 112 |
} |
| 113 |
|
| 114 |
if (localProc == 0) { |
| 115 |
for(int i=0; i<numItems; ++i) { |
| 116 |
sendBuf[i] = (localProc + 1 + i); |
| 117 |
} |
| 118 |
} |
| 119 |
|
| 120 |
MPI_Neighbor_alltoall((void*)sendCounts.data(), 1, MPI_INT, |
| 121 |
(void*)recvCounts.data(), 1, MPI_INT, neighborComm); |
| 122 |
|
| 123 |
int totalRecv = 0; |
| 124 |
recvDispls.resize(recvCounts.size()); |
| 125 |
for(size_t i=0; i<recvCounts.size(); ++i) { |
| 126 |
recvDispls[i] = totalRecv; |
| 127 |
totalRecv += recvCounts[i]; |
| 128 |
} |
| 129 |
recvBuf.resize(totalRecv); |
| 130 |
|
| 131 |
MPI_Neighbor_alltoallv( |
| 132 |
(void*)sendBuf.data(), sendCounts.data(), sendDispls.data(), MPI_DOUBLE, |
| 133 |
(void*)recvBuf.data(), recvCounts.data(), recvDispls.data(), MPI_DOUBLE, neighborComm); |
| 134 |
|
| 135 |
for(unsigned i=0; i<recvProcs.size(); ++i) { |
| 136 |
EXPECT_EQ(1, localProc);//should only recv on proc 1 |
| 137 |
|
| 138 |
const double* recvdFromProcI = recvBuf.data() + recvDispls[i]; |
| 139 |
EXPECT_EQ(numItems, recvCounts[i]); |
| 140 |
|
| 141 |
for(int item=0; item<numItems; ++item) { |
| 142 |
double val = recvdFromProcI[item]; |
| 143 |
double expected = recvProcs[i] + 1 + item; |
| 144 |
EXPECT_EQ(expected, val); |
| 145 |
} |
| 146 |
} |
| 147 |
#endif |
| 148 |
} |
| 149 |
|
| 150 |
TEST(Parallel, CommNeighborsOneSided) |
| 151 |
{ |
| 152 |
stk::ParallelMachine comm = MPI_COMM_WORLD; |
| 153 |
int numProcs = stk::parallel_machine_size(comm); |
| 154 |
if (numProcs != 2) { |
| 155 |
return; |
| 156 |
} |
| 157 |
|
| 158 |
int localProc = stk::parallel_machine_rank(comm); |
| 159 |
int otherProc = 1 - localProc; |
| 160 |
std::vector<int> sendProcs = {otherProc}; |
| 161 |
std::vector<int> recvProcs = {otherProc}; |
| 162 |
if (localProc==0) { |
| 163 |
recvProcs.clear(); |
| 164 |
} |
| 165 |
else { |
| 166 |
sendProcs.clear(); |
| 167 |
} |
| 168 |
|
| 169 |
stk::CommNeighbors commNeighbors(comm, sendProcs, recvProcs); |
| 170 |
|
| 171 |
int numItems = 5; |
| 172 |
for(int proc : sendProcs) { |
| 173 |
stk::CommBufferV& buf = commNeighbors.send_buffer(proc); |
| 174 |
for(int i=0; i<numItems; ++i) { |
| 175 |
buf.pack<double>(localProc+1+i); |
| 176 |
} |
| 177 |
} |
| 178 |
|
| 179 |
commNeighbors.communicate(); |
| 180 |
|
| 181 |
for(int proc : recvProcs) { |
| 182 |
stk::CommBufferV& buf = commNeighbors.recv_buffer(proc); |
| 183 |
|
| 184 |
int remaining = buf.size_in_bytes()/sizeof(double); |
| 185 |
EXPECT_EQ(numItems, remaining); |
| 186 |
|
| 187 |
for(int i=0; i<numItems; ++i) { |
| 188 |
double val = 0; |
| 189 |
buf.unpack<double>(val); |
| 190 |
double expected = proc+1+i; |
| 191 |
EXPECT_EQ(expected, val); |
| 192 |
} |
| 193 |
} |
| 194 |
} |
| 195 |
|
| 196 |
TEST(Parallel, CommNeighbors) |
| 197 |
{ |
| 198 |
stk::ParallelMachine comm = MPI_COMM_WORLD; |
| 199 |
int numProcs = stk::parallel_machine_size(comm); |
| 200 |
if (numProcs == 1) { |
| 201 |
return; |
| 202 |
} |
| 203 |
|
| 204 |
int myProc = stk::parallel_machine_rank(comm); |
| 205 |
int numNeighbors = 5; |
| 206 |
std::vector<int> neighborProcs = get_neighbor_procs(numProcs, myProc, numNeighbors); |
| 207 |
int numItems = 20; |
| 208 |
|
| 209 |
stk::parallel_machine_barrier(comm); |
| 210 |
|
| 211 |
stk::CommNeighbors commNeighbors(comm, neighborProcs); |
| 212 |
|
| 213 |
EXPECT_EQ(numProcs, commNeighbors.parallel_size()); |
| 214 |
EXPECT_EQ(myProc, commNeighbors.parallel_rank()); |
| 215 |
|
| 216 |
for(int proc : neighborProcs) { |
| 217 |
stk::CommBufferV& proc_buff = commNeighbors.send_buffer(proc); |
| 218 |
for(int i=0; i<numItems; ++i) { |
| 219 |
proc_buff.pack<double>(myProc+1+i); |
| 220 |
} |
| 221 |
} |
| 222 |
|
| 223 |
commNeighbors.communicate(); |
| 224 |
|
| 225 |
for(int proc : neighborProcs) { |
| 226 |
stk::CommBufferV& proc_buff = commNeighbors.recv_buffer(proc); |
| 227 |
|
| 228 |
int remaining = proc_buff.size_in_bytes()/sizeof(double); |
| 229 |
EXPECT_EQ(numItems, remaining); |
| 230 |
|
| 231 |
for(int i=0; i<numItems; ++i) { |
| 232 |
double val = 0; |
| 233 |
proc_buff.unpack<double>(val); |
| 234 |
double expected = proc+1+i; |
| 235 |
EXPECT_EQ(expected, val); |
| 236 |
} |
| 237 |
} |
| 238 |
} |
| 239 |
|
| 240 |
TEST(Parallel, CommNeighborsResetBuffers) { |
| 241 |
stk::ParallelMachine comm = MPI_COMM_WORLD; |
| 242 |
std::vector<int> send, recv; |
| 243 |
int numProcs = stk::parallel_machine_size(comm); |
| 244 |
int rank= stk::parallel_machine_rank(comm); |
| 245 |
if (numProcs == 1) { |
| 246 |
return; |
| 247 |
} |
| 248 |
if(rank == 0) { |
| 249 |
for(int p = 1; p < numProcs; ++p) { |
| 250 |
send.push_back(p); |
| 251 |
} |
| 252 |
} else { |
| 253 |
recv.push_back(0); |
| 254 |
} |
| 255 |
|
| 256 |
stk::CommNeighbors commNeighbors(comm, send, recv); |
| 257 |
|
| 258 |
for(int i = 0; i < 3; ++i) { |
| 259 |
commNeighbors.reset_buffers(); |
| 260 |
for(int p : commNeighbors.send_procs()) { |
| 261 |
commNeighbors.send_buffer(p).pack<int>(i); |
| 262 |
} |
| 263 |
|
| 264 |
commNeighbors.communicate(); |
| 265 |
|
| 266 |
for(int p : commNeighbors.recv_procs()) { |
| 267 |
int val = 0; |
| 268 |
commNeighbors.recv_buffer(p).unpack<int>(val); |
| 269 |
EXPECT_EQ(i, val); |
| 270 |
EXPECT_EQ(0u, commNeighbors.recv_buffer(p).size_in_bytes()); |
| 271 |
} |
| 272 |
} |
| 273 |
} |
| 274 |
|
| 275 |
#endif |