BufferSystem.h 11.3 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
//======================================================================================================================
//
//  This file is part of waLBerla. waLBerla is free software: you can 
//  redistribute it and/or modify it under the terms of the GNU General Public
//  License as published by the Free Software Foundation, either version 3 of 
//  the License, or (at your option) any later version.
//  
//  waLBerla is distributed in the hope that it will be useful, but WITHOUT 
//  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 
//  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License 
//  for more details.
//  
//  You should have received a copy of the GNU General Public License along
//  with waLBerla (see COPYING.txt). If not, see <http://www.gnu.org/licenses/>.
//
//! \file BufferSystem.h
//! \ingroup core
//! \author Martin Bauer <martin.bauer@fau.de>
//! \brief Header file for BufferSystem
//
//======================================================================================================================

#pragma once

#include "BufferSystemHelper.h"
#include "core/DataTypes.h"
#include "core/debug/Debug.h"

#include <boost/range/counting_range.hpp>
#include <map>
#include <set>
#include <vector>


namespace walberla {
namespace mpi  {

class OpenMPBufferSystem;

//**********************************************************************************************************************
/*! Manages MPI Communication with a set of known communication partners.
*
*   Features / Restrictions:
*    - usable in every case where normal MPI_Send, MPI_Recv is needed
*    - communication partners have to be known, message size not necessarily
*    - unknown message sizes possible ( -> automatic extra message to exchange sizes )
*    - Implemented with non-blocking MPI calls, and MPI_Waitany to process a message as soon
*      as it was received while still waiting for other messages
*
*
* \ingroup mpi
*
*
*  Example:
*  --------
*
   \code
      BufferSystem bs ( MPI_COMM_WORLD );
      // we assume every process has exactly two neighbors ( periodic )
      // and the neighbor ranks are stored in 'neighborRank1' and 'neighborRank2'
      bs.sendBuffer( neighborRank1 ) << 42;
      bs.sendBuffer( neighborRank2 ) << 4242;

      // We expect to receive the same amount as we send:
      bs.setReceiverInfoFromSendBufferState( true, false );

      bs.sendAll();

      for ( auto i = bs.begin(); i != bs.end(); ++i ) {
         cout << "Message received from " << i.rank() << endl;
         int messageContent;
         i.buffer() >> messageContent;
         cout << "Received " << messageContent << endl;
      }
   \endcode
*
*
* Usage:
* ------
*
*    1. Setup:
*        - define receive information using setReceiverInfo() or setReceiverInfoFromSendBufferState()
*          With these functions one defines the processes that we receive from and optionally the size
*          of the received messages. If the message sizes are unknown, they have to be communicated first.
*          One also defines if the sizes stay constant or if they change in each communication step
*           (size message is then sent before every content message)
*        - The receiver and send information can be changed, if no communication is currently running.
*
*    2. Communication Step:
*        - Optionally call scheduleReceives() -> starts communication step and causes MPI_IRecv's to be called.
*          This is also automatically called on first send operation.
*        - fill send buffers
*        - call send() for each buffer after filling the buffer, or call sendAll() after filling all buffers.
*          The send*() functions return immediately, internally MPI_ISend's are called.
*          The send*() functions start the communication step if it was not already started by scheduleReceives()
*        - Receiving: iterate over incoming messages using begin() and end(). Internally a MPI_Waitany is executed that
*          returns as soon as a single message was received. Then this message can be processed while waiting
*          for the other messages.
*          \attention Even if the current process does not receive anything, call begin() and end()
*                     to finish the communication step
*        - When iteration has reached the end the communication step is finished.
*        - when communication has finished all Send- and RecvBuffers are automatically cleared
*
* When running multiple BufferSystems concurrently different MPI tags have to be used
* for the systems: the tag can be passed in the constructor.
*
*/
//**********************************************************************************************************************
class BufferSystem
{
public:
   class iterator;

   //**Construction and Destruction*************************************************************************************
   /*!\name Constructors */
   //@{
   explicit BufferSystem( const MPI_Comm & communicator, int tag = 0 );
118
119
   BufferSystem( const BufferSystem & other );
   BufferSystem & operator=( const BufferSystem & other );
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
   ~BufferSystem() {}
   //@}
   //*******************************************************************************************************************


   //** Receiver Registration        ***********************************************************************************
   /*! \name Receiver Registration  */
   //@{
   template<typename RankIter> void setReceiverInfo( RankIter begin, RankIter end,              bool changingSize );
   template<typename Range>    void setReceiverInfo( const Range & range,                       bool changingSize );
                               void setReceiverInfo( const std::set<MPIRank> & ranksToRecvFrom, bool changingSize );

   void setReceiverInfo( const std::map<MPIRank,MPISize> & ranksToRecvFrom );
   void setReceiverInfoFromSendBufferState( bool useSizeFromSendBuffers, bool changingSize );

   void sizeHasChanged( bool alwaysChangingSize = false );
   //@}
   //*******************************************************************************************************************


   //** Executing Communication Step   *********************************************************************************
   /*! \name Executing Communication Step */
   //@{
   void scheduleReceives() { startCommunication(); }

   SendBuffer & sendBuffer ( MPIRank rank );
   SendBuffer & sendBuffer ( uint_t  rank ) { return sendBuffer( int_c( rank ) ); }
   inline size_t size() const;


   void sendAll();
   void send( MPIRank rank );

   iterator begin() { WALBERLA_ASSERT( communicationRunning_); return iterator( *this, true ); }
   iterator end()   {                                          return iterator( *this, false); }
   //@}
   //*******************************************************************************************************************


   //** Iterator        ************************************************************************************************
   /*! \name Iterator  */
   //@{
   class iterator
   {
   public:
      MPIRank      rank()   { return currentSenderRank_;  }
      RecvBuffer & buffer() { return *currentRecvBuffer_; }

      void operator++();
      bool operator==( const iterator & other );
      bool operator!=( const iterator & other );

   private:
      iterator( BufferSystem & bufferSystem, bool begin );

      BufferSystem & bufferSystem_;

      RecvBuffer * currentRecvBuffer_;
      MPIRank      currentSenderRank_;

      friend class BufferSystem;
   };
   friend class iterator;
   //@}
   //*******************************************************************************************************************


   //** Status Queries        ******************************************************************************************
   /*! \name Status Queries  */
   //@{
   bool isSizeCommunicatedInNextStep() const { return (currentComm_ == &unknownSizeComm_); }
   bool isCommunciationRunning() const       { return communicationRunning_;               }
   bool isReceiverInformationSet() const     { return currentComm_ != NULL;                }
   //@}
   //*******************************************************************************************************************

196
197
198
   int64_t getBytesSent() const { return bytesSent_; }
   int64_t getBytesReceived() const { return bytesReceived_; }

199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245

   //* Rank Ranges     *************************************************************************************************
   /*! \name Rank Ranges  */
   //@{
   typedef boost::counting_iterator<MPIRank> RankCountIter;
   typedef boost::iterator_range< RankCountIter > RankRange;
   static RankRange noRanks();
   static RankRange allRanks();
   static RankRange allRanksButRoot();
   static RankRange onlyRoot();
   static RankRange onlyRank( MPIRank includedRank );
   //@}
   //*******************************************************************************************************************

protected:

   friend class OpenMPBufferSystem;

   void startCommunication();
   void endCommunication();
   RecvBuffer * waitForNext( MPIRank & fromRank );
   void setCommunicationType( const bool knownSize );

   internal::KnownSizeCommunication   knownSizeComm_;
   internal::UnknownSizeCommunication unknownSizeComm_;
   internal::NoMPICommunication       noMPIComm_;
   internal::AbstractCommunication *  currentComm_;  //< after receiver setup, this points to unknown- or knownSizeComm_

   bool sizeChangesEverytime_; //< if set to true, the receiveSizeUnknown_ is set to true before communicating
   bool communicationRunning_; //< indicates if a communication step is currently running


   /// Info about the message to be received from a certain rank:
   /// information holds the buffer and, if known, the message size
   std::map<MPIRank, internal::AbstractCommunication::ReceiveInfo> recvInfos_;


   struct SendInfo {
      SendInfo() : alreadySent(false) {}
      SendBuffer buffer;
      bool alreadySent;
   };
   std::map<MPIRank, SendInfo> sendInfos_;

   //stores tags of running communications in debug mode to ensure that
   //each concurrently running communication uses different tags
   static std::set<int> activeTags_;
246
247
248

   int64_t bytesSent_     = 0; ///< number of bytes sent during last communication
   int64_t bytesReceived_ = 0; ///< number of bytes received during last communication
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
};







//======================================================================================================================
//
//  Template function definitions
//
//======================================================================================================================



template<typename Range>
void BufferSystem::setReceiverInfo( const Range & range, bool changingSize )
{
   setReceiverInfo( range.begin(), range.end(), changingSize );
}

template<typename RankIter>
void BufferSystem::setReceiverInfo( RankIter rankBegin, RankIter rankEnd, bool changingSize )
{
   WALBERLA_ASSERT( ! communicationRunning_ );

   recvInfos_.clear();
   for ( auto it = rankBegin; it != rankEnd; ++it )
   {
      const MPIRank sender = *it;
      recvInfos_[ sender ].size = INVALID_SIZE;
   }

   sizeChangesEverytime_ = changingSize;
   setCommunicationType( false );
}

inline size_t BufferSystem::size() const
{
   size_t sum = 0;
   for( auto iter = sendInfos_.begin(); iter != sendInfos_.end(); ++iter )
   {
      sum += iter->second.buffer.size();
   }
   return sum;
}



} // namespace mpi
} // namespace walberla