FEI Package Browser (Single Doxygen Collection)  Version of the Day
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
fei_CommUtils.cpp
Go to the documentation of this file.
1 
2 #include <fei_CommUtils.hpp>
3 #include <fei_TemplateUtils.hpp>
4 
5 #undef fei_file
6 #define fei_file "fei_CommUtils.cpp"
7 
8 #include <fei_ErrMacros.hpp>
9 
10 namespace fei {
11 
12 //------------------------------------------------------------------------
14 {
15 #ifdef FEI_SER
16  return 0;
17 #else
18  int local_proc = 0;
19  MPI_Comm_rank(comm, &local_proc);
20  return local_proc;
21 #endif
22 }
23 
24 //------------------------------------------------------------------------
25 int numProcs(MPI_Comm comm)
26 {
27 #ifdef FEI_SER
28  return 1;
29 #else
30  int num_procs = 1;
31  MPI_Comm_size(comm, &num_procs);
32  return num_procs;
33 #endif
34 }
35 
36 //------------------------------------------------------------------------
37 void Barrier(MPI_Comm comm)
38 {
39 #ifndef FEI_SER
40  MPI_Barrier(comm);
41 #endif
42 }
43 
44 //------------------------------------------------------------------------
45 int mirrorProcs(MPI_Comm comm, std::vector<int>& toProcs, std::vector<int>& fromProcs)
46 {
47  fromProcs.resize(0);
48 #ifdef FEI_SER
49  fromProcs.push_back(0);
50  return(0);
51 #else
52  int num_procs = fei::numProcs(comm);
53  std::vector<int> tmpIntData(num_procs*3, 0);
54 
55  int* buf = &tmpIntData[0];
56  int* recvbuf = buf+num_procs;
57 
58  for(unsigned i=0; i<toProcs.size(); ++i) {
59  buf[toProcs[i]] = 1;
60  }
61 
62  for(int ii=2*num_procs; ii<3*num_procs; ++ii) {
63  buf[ii] = 1;
64  }
65 
66  CHK_MPI( MPI_Reduce_scatter(buf, &(buf[num_procs]), &(buf[2*num_procs]),
67  MPI_INT, MPI_SUM, comm) );
68 
69  int numRecvProcs = buf[num_procs];
70 
71  int tag = 11116;
72  std::vector<MPI_Request> mpiReqs(numRecvProcs);
73 
74  int offset = 0;
75  for(int ii=0; ii<numRecvProcs; ++ii) {
76  CHK_MPI( MPI_Irecv(&(recvbuf[ii]), 1, MPI_INT, MPI_ANY_SOURCE, tag,
77  comm, &(mpiReqs[offset++])) );
78  }
79 
80  for(unsigned i=0; i<toProcs.size(); ++i) {
81  CHK_MPI( MPI_Send(&(toProcs[i]), 1, MPI_INT, toProcs[i], tag, comm) );
82  }
83 
84  MPI_Status status;
85  for(int ii=0; ii<numRecvProcs; ++ii) {
86  int index;
87  MPI_Waitany(numRecvProcs, &mpiReqs[0], &index, &status);
88  fromProcs.push_back(status.MPI_SOURCE);
89  }
90 
91  std::sort(fromProcs.begin(), fromProcs.end());
92 
93  return(0);
94 #endif
95 }
96 
97 //------------------------------------------------------------------------
98 int mirrorCommPattern(MPI_Comm comm, comm_map* inPattern, comm_map*& outPattern)
99 {
100 #ifdef FEI_SER
101  (void)inPattern;
102  (void)outPattern;
103 #else
104  int localP = localProc(comm);
105  int numP = numProcs(comm);
106 
107  if (numP < 2) return(0);
108 
109  std::vector<int> buf(numP*2, 0);
110 
111  int numInProcs = inPattern->getMap().size();
112  std::vector<int> inProcs(numInProcs);
113  fei::copyKeysToVector(inPattern->getMap(), inProcs);
114 
115  std::vector<int> outProcs;
116 
117  int err = mirrorProcs(comm, inProcs, outProcs);
118  if (err != 0) ERReturn(-1);
119 
120  std::vector<int> recvbuf(outProcs.size(), 0);
121 
122  outPattern = new comm_map(0,1);
123 
124  MPI_Datatype mpi_ttype = fei::mpiTraits<int>::mpi_type();
125 
126  //now recv a length (the contents of buf[i]) from each "out-proc", which
127  //will be the length of the equation data that will also be recvd from that
128  //proc.
129  std::vector<MPI_Request> mpiReqs(outProcs.size());
130  std::vector<MPI_Status> mpiStss(outProcs.size());
131  MPI_Request* requests = &mpiReqs[0];
132  MPI_Status* statuses = &mpiStss[0];
133 
134  int firsttag = 11117;
135  int offset = 0;
136  int* outProcsPtr = &outProcs[0];
137  for(unsigned i=0; i<outProcs.size(); ++i) {
138  if (MPI_Irecv(&(recvbuf[i]), 1, MPI_INT, outProcsPtr[i], firsttag,
139  comm, &requests[offset++]) != MPI_SUCCESS) ERReturn(-1);
140  }
141 
142  comm_map::map_type& in_row_map = inPattern->getMap();
144  in_iter = in_row_map.begin(),
145  in_end = in_row_map.end();
146 
147  int* inProcsPtr = &inProcs[0];
148  for(int ii=0; in_iter!= in_end; ++in_iter, ++ii) {
149  comm_map::row_type* in_row = in_iter->second;
150  buf[ii] = in_row->size();
151  if (MPI_Send(&(buf[ii]), 1, MPI_INT, inProcsPtr[ii], firsttag,
152  comm) != MPI_SUCCESS) ERReturn(-1);
153  }
154 
155  int numOutProcs = outProcs.size();
156 
157  MPI_Waitall(numOutProcs, requests, statuses);
158  std::vector<int> lengths(numOutProcs);
159  int totalRecvLen = 0;
160  offset = 0;
161  for(int ii=0; ii<numOutProcs; ++ii) {
162  if (recvbuf[ii] > 0) {
163  lengths[offset++] = recvbuf[ii];
164  totalRecvLen += recvbuf[ii];
165  }
166  }
167 
168  //now we need to create the space into which we'll receive the
169  //lists that other procs send to us.
170  std::vector<int> recvData(totalRecvLen, 999999);
171 
172  int tag2 = 11118;
173  offset = 0;
174  for(int ii=0; ii<numOutProcs; ++ii) {
175  CHK_MPI(MPI_Irecv(&(recvData[offset]), lengths[ii], mpi_ttype,
176  outProcs[ii], tag2, comm, &requests[ii]) );
177  offset += lengths[ii];
178  }
179 
180  std::vector<int> sendList;
181 
182  in_iter = in_row_map.begin();
183 
184  for(int ii=0; in_iter != in_end; ++in_iter,++ii) {
185  if (inProcs[ii] == localP) {
186  continue;
187  }
188  sendList.resize(in_iter->second->size());
189  fei::copySetToArray(*(in_iter->second), sendList.size(), &sendList[0]);
190 
191  CHK_MPI(MPI_Send(&sendList[0], sendList.size(), mpi_ttype,
192  inProcs[ii], tag2, comm) );
193  }
194 
195  //our final communication operation is to catch the Irecvs we started above.
196  for(int ii=0; ii<numOutProcs; ++ii) {
197  MPI_Wait(&requests[ii], &statuses[ii]);
198  }
199 
200  //now we've completed all the communication, so we're ready to put the data
201  //we received into the outPattern object.
202  offset = 0;
203  for(int ii=0; ii<numOutProcs; ii++) {
204  outPattern->addIndices(outProcs[ii], lengths[ii],
205  &(recvData[offset]));
206  offset += lengths[ii];
207  }
208 
209 #endif
210  return(0);
211 }
212 
213 
214 //------------------------------------------------------------------------
216  const std::vector<int>& sendProcs,
217  std::vector<int>& sendData,
218  const std::vector<int>& recvProcs,
219  std::vector<int>& recvData)
220 {
221  if (sendProcs.size() == 0 && recvProcs.size() == 0) return(0);
222  if (sendProcs.size() != sendData.size()) return(-1);
223 #ifndef FEI_SER
224  recvData.resize(recvProcs.size());
225  std::vector<MPI_Request> mpiReqs;
226  mpiReqs.resize(recvProcs.size());
227 
228  int tag = 11114;
229  MPI_Datatype mpi_dtype = MPI_INT;
230 
231  //launch Irecv's for recvData:
232 
233  int localProc = fei::localProc(comm);
234  int numRecvProcs = recvProcs.size();
235  int req_offset = 0;
236  for(unsigned i=0; i<recvProcs.size(); ++i) {
237  if (recvProcs[i] == localProc) {--numRecvProcs; continue; }
238 
239  CHK_MPI( MPI_Irecv(&(recvData[i]), 1, mpi_dtype, recvProcs[i], tag,
240  comm, &mpiReqs[req_offset++]) );
241  }
242 
243  //send the sendData:
244 
245  for(unsigned i=0; i<sendProcs.size(); ++i) {
246  if (sendProcs[i] == localProc) continue;
247 
248  CHK_MPI( MPI_Send(&(sendData[i]), 1, mpi_dtype,
249  sendProcs[i], tag, comm) );
250  }
251 
252  //complete the Irecv's:
253 
254  for(int ii=0; ii<numRecvProcs; ++ii) {
255  int index;
256  MPI_Status status;
257  CHK_MPI( MPI_Waitany(numRecvProcs, &mpiReqs[0], &index, &status) );
258  }
259 
260 #endif
261  return(0);
262 }
263 
264 //------------------------------------------------------------------------
265 int Allreduce(MPI_Comm comm, bool localBool, bool& globalBool)
266 {
267 #ifndef FEI_SER
268  int localInt = localBool ? 1 : 0;
269  int globalInt = 0;
270 
271  CHK_MPI( MPI_Allreduce(&localInt, &globalInt, 1, MPI_INT, MPI_MAX, comm) );
272 
273  globalBool = globalInt==1 ? true : false;
274 #else
275  globalBool = localBool;
276 #endif
277 
278  return(0);
279 }
280 
281 }//namespace fei
282 
int mirrorCommPattern(MPI_Comm comm, comm_map *inPattern, comm_map *&outPattern)
void Barrier(MPI_Comm comm)
void copySetToArray(const SET_TYPE &set_obj, int lenList, int *list)
#define MPI_Comm
Definition: fei_mpi.h:56
#define MPI_SUCCESS
Definition: fei_mpi.h:62
#define MPI_Request
Definition: fei_mpi.h:57
void copyKeysToVector(const MAP_TYPE &map_obj, std::vector< int > &keyvector)
int mirrorProcs(MPI_Comm comm, std::vector< int > &toProcs, std::vector< int > &fromProcs)
#define ERReturn(a)
int Allreduce(MPI_Comm comm, bool localBool, bool &globalBool)
#define MPI_Barrier(a)
Definition: fei_mpi.h:61
int exchangeIntData(MPI_Comm comm, const std::vector< int > &sendProcs, std::vector< int > &sendData, const std::vector< int > &recvProcs, std::vector< int > &recvData)
snl_fei::RaggedTable< std::map< int, std::set< int > * >, std::set< int > > comm_map
int localProc(MPI_Comm comm)
#define CHK_MPI(a)
Definition: fei_chk_mpi.hpp:21
MAP_TYPE::iterator iterator
int numProcs(MPI_Comm comm)