ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/group/trunk/OOPSE/libmdtools/mpiSimulation.cpp
(Generate patch)

Comparing:
branches/mmeineke/OOPSE/libmdtools/mpiSimulation.cpp (file contents), Revision 377 by mmeineke, Fri Mar 21 17:42:12 2003 UTC vs.
trunk/OOPSE/libmdtools/mpiSimulation.cpp (file contents), Revision 422 by mmeineke, Thu Mar 27 19:21:42 2003 UTC

# Line 2 | Line 2
2  
3   #include <cstdlib>
4   #include <cstring>
5 + #include <cmath>
6   #include <mpi.h>
7   #include <mpi++.h>
8  
9   #include "mpiSimulation.hpp"
10   #include "simError.h"
11   #include "fortranWrappers.hpp"
12 + #include "randomSPRNG.hpp"
13  
14 + #define BASE_SEED 123456789
15  
13
14
16   mpiSimulation* mpiSim;
17  
18   mpiSimulation::mpiSimulation(SimInfo* the_entryPlug)
# Line 21 | Line 22 | mpiSimulation::mpiSimulation(SimInfo* the_entryPlug)
22    
23    mpiPlug->numberProcessors = MPI::COMM_WORLD.Get_size();
24    mpiPlug->myNode = worldRank;
25 <  
25 >
26 >  MolToProcMap = new int[entryPlug->n_mol];
27 >  MolComponentType = new int[entryPlug->n_mol];
28 >  AtomToProcMap = new int[entryPlug->n_atoms];
29 >
30    mpiSim = this;
31    wrapMeSimParallel( this );
32   }
# Line 29 | Line 34 | mpiSimulation::~mpiSimulation(){
34  
35   mpiSimulation::~mpiSimulation(){
36    
37 +  delete[] MolToProcMap;
38 +  delete[] MolComponentType;
39 +  delete[] AtomToProcMap;
40 +
41    delete mpiPlug;
42    // perhaps we should let fortran know the party is over.
43    
44   }
45  
37
38
46   int* mpiSimulation::divideLabor( void ){
47  
48    int* globalIndex;
49  
50    int nComponents;
51    MoleculeStamp** compStamps;
52 +  randomSPRNG *myRandom;
53    int* componentsNmol;
54 +  int* AtomsPerProc;
55  
56    double numerator;
57    double denominator;
58    double precast;
59 +  double x, y, a;
60 +  int old_atoms, add_atoms, new_atoms;
61  
62    int nTarget;
63    int molIndex, atomIndex, compIndex, compStart;
64    int done;
65    int nLocal, molLocal;
66 <  int i, index;
66 >  int i, j, loops, which_proc, nmol_local, natoms_local;
67 >  int nmol_global, natoms_global;
68 >  int local_index, index;
69    int smallDiff, bigDiff;
70 +  int baseSeed = BASE_SEED;
71  
72    int testSum;
73  
74    nComponents = entryPlug->nComponents;
75    compStamps = entryPlug->compStamps;
76    componentsNmol = entryPlug->componentsNmol;
77 <
77 >  AtomsPerProc = new int[mpiPlug->numberProcessors];
78 >  
79    mpiPlug->nAtomsGlobal = entryPlug->n_atoms;
80    mpiPlug->nBondsGlobal = entryPlug->n_bonds;
81    mpiPlug->nBendsGlobal = entryPlug->n_bends;
# Line 68 | Line 83 | int* mpiSimulation::divideLabor( void ){
83    mpiPlug->nSRIGlobal = entryPlug->n_SRI;
84    mpiPlug->nMolGlobal = entryPlug->n_mol;
85  
86 <  numerator = (double) entryPlug->n_atoms;
72 <  denominator = (double) mpiPlug->numberProcessors;
73 <  precast = numerator / denominator;
74 <  nTarget = (int)( precast + 0.5 );
75 <  
76 <  molIndex = 0;
77 <  atomIndex = 0;
78 <  compIndex = 0;
79 <  compStart = 0;
80 <  for( i=0; i<(mpiPlug->numberProcessors-1); i++){
81 <    
82 <    done = 0;
83 <    nLocal = 0;
84 <    molLocal = 0;
86 >  myRandom = new randomSPRNG( baseSeed );
87  
88 <    if( i == mpiPlug->myNode ){
87 <      mpiPlug->myMolStart = molIndex;
88 <      mpiPlug->myAtomStart = atomIndex;
89 <    }
90 <    
91 <    while( !done ){
92 <      
93 <      if( (molIndex-compStart) >= componentsNmol[compIndex] ){
94 <        compStart = molIndex;
95 <        compIndex++;
96 <        continue;
97 <      }
88 >  a = (double)mpiPlug->nMolGlobal / (double)mpiPlug->nAtomsGlobal;
89  
90 <      nLocal += compStamps[compIndex]->getNAtoms();
91 <      atomIndex += compStamps[compIndex]->getNAtoms();
92 <      molIndex++;
93 <      molLocal++;
94 <      
95 <      if ( nLocal == nTarget ) done = 1;
96 <      
97 <      else if( nLocal < nTarget ){
98 <        smallDiff = nTarget - nLocal;
99 <      }
100 <      else if( nLocal > nTarget ){
101 <        bigDiff = nLocal - nTarget;
102 <        
112 <        if( bigDiff < smallDiff ) done = 1;
113 <        else{
114 <          molIndex--;
115 <          molLocal--;
116 <          atomIndex -= compStamps[compIndex]->getNAtoms();
117 <          nLocal -= compStamps[compIndex]->getNAtoms();
118 <          done = 1;
119 <        }
120 <      }
121 <    }
90 >  // Initialize things that we'll send out later:
91 >  for (i = 0; i < mpiPlug->numberProcessors; i++ ) {
92 >    AtomsPerProc[i] = 0;
93 >  }
94 >  for (i = 0; i < mpiPlug->nMolGlobal; i++ ) {
95 >    // default to an error condition:
96 >    MolToProcMap[i] = -1;
97 >    MolComponentType[i] = -1;
98 >  }
99 >  for (i = 0; i < mpiPlug->nAtomsGlobal; i++ ) {
100 >    // default to an error condition:
101 >    AtomToProcMap[i] = -1;
102 >  }
103      
104 <    if( i == mpiPlug->myNode ){
105 <      mpiPlug->myMolEnd = (molIndex - 1);
106 <      mpiPlug->myAtomEnd = (atomIndex - 1);
126 <      mpiPlug->myNlocal = nLocal;
127 <      mpiPlug->myMol = molLocal;
128 <    }
129 <    
130 <    numerator = (double)( entryPlug->n_atoms - atomIndex );
131 <    denominator = (double)( mpiPlug->numberProcessors - (i+1) );
104 >  if (mpiPlug->myNode == 0) {
105 >    numerator = (double) entryPlug->n_atoms;
106 >    denominator = (double) mpiPlug->numberProcessors;
107      precast = numerator / denominator;
108      nTarget = (int)( precast + 0.5 );
134  }
135  
136  if( mpiPlug->myNode == mpiPlug->numberProcessors-1 ){
137      mpiPlug->myMolStart = molIndex;
138      mpiPlug->myAtomStart = atomIndex;
109  
110 <      nLocal = 0;
111 <      molLocal = 0;
112 <      while( compIndex < nComponents ){
113 <        
114 <        if( (molIndex-compStart) >= componentsNmol[compIndex] ){
115 <          compStart = molIndex;
146 <          compIndex++;
147 <          continue;
148 <        }
149 <
150 <        nLocal += compStamps[compIndex]->getNAtoms();
151 <        atomIndex += compStamps[compIndex]->getNAtoms();
152 <        molIndex++;
153 <        molLocal++;
110 >    // Build the array of molecule component types first
111 >    molIndex = 0;
112 >    for (i=0; i < nComponents; i++) {
113 >      for (j=0; j < componentsNmol[i]; j++) {        
114 >        MolComponentType[molIndex] = i;
115 >        molIndex++;
116        }
117 <      
156 <      mpiPlug->myMolEnd = (molIndex - 1);
157 <      mpiPlug->myAtomEnd = (atomIndex - 1);
158 <      mpiPlug->myNlocal = nLocal;  
159 <      mpiPlug->myMol = molLocal;
160 <  }
117 >    }
118  
119 +    atomIndex = 0;
120  
121 <  MPI_Allreduce( &nLocal, &testSum, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD );
122 <  
123 <  if( mpiPlug->myNode == 0 ){
124 <    if( testSum != entryPlug->n_atoms ){
125 <      sprintf( painCave.errMsg,
126 <               "The summ of all nLocals, %d, did not equal the total number of atoms, %d.\n",
127 <               testSum, entryPlug->n_atoms );
128 <      painCave.isFatal = 1;
129 <      simError();
121 >    for (i = 0; i < molIndex; i++ ) {
122 >
123 >      done = 0;
124 >      loops = 0;
125 >
126 >      while( !done ){
127 >        loops++;
128 >        
129 >        // Pick a processor at random
130 >
131 >        which_proc = (int) (myRandom->getRandom() * mpiPlug->numberProcessors);
132 >
133 >        // How many atoms does this processor have?
134 >        
135 >        old_atoms = AtomsPerProc[which_proc];
136 >
137 >        // If the processor already had too many atoms, just skip this
138 >        // processor and try again.
139 >
140 >        if (old_atoms >= nTarget) continue;
141 >
142 >        add_atoms = compStamps[MolComponentType[i]]->getNAtoms();
143 >        new_atoms = old_atoms + add_atoms;
144 >    
145 >        // If we can add this molecule to this processor without sending
146 >        // it above nTarget, then go ahead and do it:
147 >    
148 >        if (new_atoms <= nTarget) {
149 >          MolToProcMap[i] = which_proc;
150 >          AtomsPerProc[which_proc] += add_atoms;
151 >          for (j = 0 ; j < add_atoms; j++ ) {
152 >            AtomToProcMap[atomIndex] = which_proc;
153 >            atomIndex++;
154 >          }
155 >          done = 1;
156 >          continue;
157 >        }
158 >
159 >        // If we've been through this loop too many times, we need
160 >        // to just give up and assign the molecule to this processor
161 >        // and be done with it.
162 >        
163 >        if (loops > 100) {          
164 >          sprintf( painCave.errMsg,
165 >                   "I've tried 100 times to assign molecule %d to a "
166 >                   " processor, but can't find a good spot.\n"  
167 >                   "I'm assigning it at random to processor %d.\n",
168 >                   i, which_proc);
169 >          painCave.isFatal = 0;
170 >          simError();
171 >          
172 >          MolToProcMap[i] = which_proc;
173 >          AtomsPerProc[which_proc] += add_atoms;
174 >          for (j = 0 ; j < add_atoms; j++ ) {
175 >            AtomToProcMap[atomIndex] = which_proc;
176 >            atomIndex++;
177 >          }
178 >          done = 1;
179 >          continue;
180 >        }
181 >
182 >        // The only situation left is where old_atoms < nTarget, but
183 >        // new_atoms > nTarget.   We want to accept this with some
184 >        // probability that dies off the farther we are from nTarget
185 >
186 >        // roughly:  x = new_atoms - nTarget
187 >        //           Pacc(x) = exp(- a * x)
188 >        // where a = 1 / (average atoms per molecule)
189 >
190 >        x = (double) (new_atoms - nTarget);
191 >        y = myRandom->getRandom();
192 >        
193 >        if (exp(- a * x) > y) {
194 >          MolToProcMap[i] = which_proc;
195 >          AtomsPerProc[which_proc] += add_atoms;
196 >          for (j = 0 ; j < add_atoms; j++ ) {
197 >            AtomToProcMap[atomIndex] = which_proc;
198 >            atomIndex++;
199 >           }
200 >          done = 1;
201 >          continue;
202 >        } else {
203 >          continue;
204 >        }      
205 >        
206 >      }
207      }
208 +
209 +    // Spray out this nonsense to all other processors:
210 +
211 +    MPI::COMM_WORLD.Bcast(MolToProcMap, mpiPlug->nMolGlobal,
212 +                          MPI_INT, 0);
213 +
214 +    MPI::COMM_WORLD.Bcast(AtomToProcMap, mpiPlug->nAtomsGlobal,
215 +                          MPI_INT, 0);
216 +
217 +    MPI::COMM_WORLD.Bcast(MolComponentType, mpiPlug->nMolGlobal,
218 +                          MPI_INT, 0);
219 +
220 +    MPI::COMM_WORLD.Bcast(AtomsPerProc, mpiPlug->numberProcessors,
221 +                          MPI_INT, 0);    
222 +  } else {
223 +
224 +    // Listen to your marching orders from processor 0:
225 +    
226 +    MPI::COMM_WORLD.Bcast(MolToProcMap, mpiPlug->nMolGlobal,
227 +                          MPI_INT, 0);
228 +    
229 +    MPI::COMM_WORLD.Bcast(AtomToProcMap, mpiPlug->nAtomsGlobal,
230 +                          MPI_INT, 0);
231 +
232 +    MPI::COMM_WORLD.Bcast(MolComponentType, mpiPlug->nMolGlobal,
233 +                          MPI_INT, 0);
234 +    
235 +    MPI::COMM_WORLD.Bcast(AtomsPerProc, mpiPlug->numberProcessors,
236 +                          MPI_INT, 0);
237    }
238  
239 +
240 +  // Let's all check for sanity:
241 +
242 +  nmol_local = 0;
243 +  for (i = 0 ; i < mpiPlug->nMolGlobal; i++ ) {
244 +    if (MolToProcMap[i] == mpiPlug->myNode) {
245 +      nmol_local++;
246 +    }
247 +  }
248 +
249 +  natoms_local = 0;
250 +  for (i = 0; i < mpiPlug->nAtomsGlobal; i++) {
251 +    if (AtomToProcMap[i] == mpiPlug->myNode) {
252 +      natoms_local++;      
253 +    }
254 +  }
255 +
256 +  MPI::COMM_WORLD.Allreduce(&nmol_local,&nmol_global,1,MPI_INT,MPI_SUM);
257 +  MPI::COMM_WORLD.Allreduce(&natoms_local,&natoms_global,1,MPI_INT,MPI_SUM);
258 +  
259 +  if( nmol_global != entryPlug->n_mol ){
260 +    sprintf( painCave.errMsg,
261 +             "The sum of all nmol_local, %d, did not equal the "
262 +             "total number of molecules, %d.\n",
263 +             nmol_global, entryPlug->n_mol );
264 +    painCave.isFatal = 1;
265 +    simError();
266 +  }
267 +  
268 +  if( natoms_global != entryPlug->n_atoms ){
269 +    sprintf( painCave.errMsg,
270 +             "The sum of all natoms_local, %d, did not equal the "
271 +             "total number of atoms, %d.\n",
272 +             natoms_global, entryPlug->n_atoms );
273 +    painCave.isFatal = 1;
274 +    simError();
275 +  }
276 +
277    sprintf( checkPointMsg,
278             "Successfully divided the molecules among the processors.\n" );
279    MPIcheckPoint();
280  
281 <  // lets create the identity array
281 >  mpiPlug->myNMol = nmol_local;
282 >  mpiPlug->myNlocal = natoms_local;
283  
284    globalIndex = new int[mpiPlug->myNlocal];
285 <  index = mpiPlug->myAtomStart;
286 <  for( i=0; i<mpiPlug->myNlocal; i++){
287 <    globalIndex[i] = index;
288 <    index++;
285 >  local_index = 0;
286 >  for (i = 0; i < mpiPlug->nAtomsGlobal; i++) {
287 >    if (AtomToProcMap[i] == mpiPlug->myNode) {
288 >      globalIndex[local_index] = i;
289 >      local_index++;
290 >    }
291    }
292 <
292 >
293    return globalIndex;
294   }
295  

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines