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

Comparing trunk/OOPSE/libmdtools/SimInfo.cpp (file contents):
Revision 574 by gezelter, Tue Jul 8 20:56:10 2003 UTC vs.
Revision 1214 by gezelter, Tue Jun 1 18:42:58 2004 UTC

# Line 1 | Line 1
1 < #include <cstdlib>
2 < #include <cstring>
3 < #include <cmath>
1 > #include <stdlib.h>
2 > #include <string.h>
3 > #include <math.h>
4  
5   #include <iostream>
6   using namespace std;
# Line 12 | Line 12 | using namespace std;
12  
13   #include "fortranWrappers.hpp"
14  
15 + #include "MatVec3.h"
16 +
17   #ifdef IS_MPI
18   #include "mpiSimulation.hpp"
19   #endif
# Line 20 | Line 22 | inline double roundMe( double x ){
22    return ( x >= 0 ) ? floor( x + 0.5 ) : ceil( x - 0.5 );
23   }
24            
25 + inline double min( double a, double b ){
26 +  return (a < b ) ? a : b;
27 + }
28  
29   SimInfo* currentInfo;
30  
31   SimInfo::SimInfo(){
32 <  excludes = NULL;
32 >
33    n_constraints = 0;
34 +  nZconstraints = 0;
35    n_oriented = 0;
36    n_dipoles = 0;
37    ndf = 0;
38    ndfRaw = 0;
39 +  nZconstraints = 0;
40    the_integrator = NULL;
41    setTemp = 0;
42    thermalTime = 0.0;
43 +  currentTime = 0.0;
44    rCut = 0.0;
45 +  rSw = 0.0;
46  
47 +  haveRcut = 0;
48 +  haveRsw = 0;
49 +  boxIsInit = 0;
50 +  
51 +  resetTime = 1e99;
52 +
53 +  orthoRhombic = 0;
54 +  orthoTolerance = 1E-6;
55 +  useInitXSstate = true;
56 +
57    usePBC = 0;
58    useLJ = 0;
59    useSticky = 0;
60 <  useDipole = 0;
60 >  useCharges = 0;
61 >  useDipoles = 0;
62    useReactionField = 0;
63    useGB = 0;
64    useEAM = 0;
65 +  useSolidThermInt = 0;
66 +  useLiquidThermInt = 0;
67  
68 <  wrapMeSimInfo( this );
47 < }
68 >  haveCutoffGroups = false;
69  
70 < void SimInfo::setBox(double newBox[3]) {
70 >  excludes = Exclude::Instance();
71  
72 <  double smallestBoxL, maxCutoff;
52 <  int status;
53 <  int i;
72 >  myConfiguration = new SimState();
73  
74 <  for(i=0; i<9; i++) Hmat[i] = 0.0;;
74 >  has_minimizer = false;
75 >  the_minimizer =NULL;
76  
77 <  Hmat[0] = newBox[0];
58 <  Hmat[4] = newBox[1];
59 <  Hmat[8] = newBox[2];
77 >  ngroup = 0;
78  
79 <  calcHmatI();
80 <  calcBoxL();
79 >  wrapMeSimInfo( this );
80 > }
81  
64  setFortranBoxSize(Hmat, HmatI, &orthoRhombic);
82  
83 <  smallestBoxL = boxLx;
67 <  if (boxLy < smallestBoxL) smallestBoxL = boxLy;
68 <  if (boxLz < smallestBoxL) smallestBoxL = boxLz;
83 > SimInfo::~SimInfo(){
84  
85 <  maxCutoff = smallestBoxL / 2.0;
85 >  delete myConfiguration;
86  
87 <  if (rList > maxCutoff) {
88 <    sprintf( painCave.errMsg,
89 <             "New Box size is forcing neighborlist radius down to %lf\n",
90 <             maxCutoff );
91 <    painCave.isFatal = 0;
92 <    simError();
87 >  map<string, GenericData*>::iterator i;
88 >  
89 >  for(i = properties.begin(); i != properties.end(); i++)
90 >    delete (*i).second;
91 >  
92 > }
93  
94 <    rList = maxCutoff;
94 > void SimInfo::setBox(double newBox[3]) {
95 >  
96 >  int i, j;
97 >  double tempMat[3][3];
98  
99 <    sprintf( painCave.errMsg,
100 <             "New Box size is forcing cutoff radius down to %lf\n",
83 <             maxCutoff - 1.0 );
84 <    painCave.isFatal = 0;
85 <    simError();
99 >  for(i=0; i<3; i++)
100 >    for (j=0; j<3; j++) tempMat[i][j] = 0.0;;
101  
102 <    rCut = rList - 1.0;
102 >  tempMat[0][0] = newBox[0];
103 >  tempMat[1][1] = newBox[1];
104 >  tempMat[2][2] = newBox[2];
105  
106 <    // list radius changed so we have to refresh the simulation structure.
90 <    refreshSim();
91 <  }
106 >  setBoxM( tempMat );
107  
93  if (rCut > maxCutoff) {
94    sprintf( painCave.errMsg,
95             "New Box size is forcing cutoff radius down to %lf\n",
96             maxCutoff );
97    painCave.isFatal = 0;
98    simError();
99
100    status = 0;
101    LJ_new_rcut(&rCut, &status);
102    if (status != 0) {
103      sprintf( painCave.errMsg,
104               "Error in recomputing LJ shifts based on new rcut\n");
105      painCave.isFatal = 1;
106      simError();
107    }
108  }
108   }
109  
110 < void SimInfo::setBoxM( double theBox[9] ){
110 > void SimInfo::setBoxM( double theBox[3][3] ){
111    
112 <  int i, status;
113 <  double smallestBoxL, maxCutoff;
112 >  int i, j;
113 >  double FortranHmat[9]; // to preserve compatibility with Fortran the
114 >                         // ordering in the array is as follows:
115 >                         // [ 0 3 6 ]
116 >                         // [ 1 4 7 ]
117 >                         // [ 2 5 8 ]
118 >  double FortranHmatInv[9]; // the inverted Hmat (for Fortran);
119  
120 <  for(i=0; i<9; i++) Hmat[i] = theBox[i];
121 <  calcHmatI();
120 >  if( !boxIsInit ) boxIsInit = 1;
121 >
122 >  for(i=0; i < 3; i++)
123 >    for (j=0; j < 3; j++) Hmat[i][j] = theBox[i][j];
124 >  
125    calcBoxL();
126 +  calcHmatInv();
127  
128 <  setFortranBoxSize(Hmat, HmatI, &orthoRhombic);
129 <
130 <  smallestBoxL = boxLx;
131 <  if (boxLy < smallestBoxL) smallestBoxL = boxLy;
124 <  if (boxLz < smallestBoxL) smallestBoxL = boxLz;
125 <
126 <  maxCutoff = smallestBoxL / 2.0;
127 <
128 <  if (rList > maxCutoff) {
129 <    sprintf( painCave.errMsg,
130 <             "New Box size is forcing neighborlist radius down to %lf\n",
131 <             maxCutoff );
132 <    painCave.isFatal = 0;
133 <    simError();
134 <
135 <    rList = maxCutoff;
136 <
137 <    sprintf( painCave.errMsg,
138 <             "New Box size is forcing cutoff radius down to %lf\n",
139 <             maxCutoff - 1.0 );
140 <    painCave.isFatal = 0;
141 <    simError();
142 <
143 <    rCut = rList - 1.0;
144 <
145 <    // list radius changed so we have to refresh the simulation structure.
146 <    refreshSim();
147 <  }
148 <
149 <  if (rCut > maxCutoff) {
150 <    sprintf( painCave.errMsg,
151 <             "New Box size is forcing cutoff radius down to %lf\n",
152 <             maxCutoff );
153 <    painCave.isFatal = 0;
154 <    simError();
155 <
156 <    status = 0;
157 <    LJ_new_rcut(&rCut, &status);
158 <    if (status != 0) {
159 <      sprintf( painCave.errMsg,
160 <               "Error in recomputing LJ shifts based on new rcut\n");
161 <      painCave.isFatal = 1;
162 <      simError();
128 >  for(i=0; i < 3; i++) {
129 >    for (j=0; j < 3; j++) {
130 >      FortranHmat[3*j + i] = Hmat[i][j];
131 >      FortranHmatInv[3*j + i] = HmatInv[i][j];
132      }
133    }
134 +
135 +  setFortranBoxSize(FortranHmat, FortranHmatInv, &orthoRhombic);
136 +
137   }
138  
139  
140 < void SimInfo::getBoxM (double theBox[9]) {
140 > void SimInfo::getBoxM (double theBox[3][3]) {
141  
142 <  int i;
143 <  for(i=0; i<9; i++) theBox[i] = Hmat[i];
142 >  int i, j;
143 >  for(i=0; i<3; i++)
144 >    for (j=0; j<3; j++) theBox[i][j] = Hmat[i][j];
145   }
146  
147  
148   void SimInfo::scaleBox(double scale) {
149 <  double theBox[9];
150 <  int i;
149 >  double theBox[3][3];
150 >  int i, j;
151  
152 <  for(i=0; i<9; i++) theBox[i] = Hmat[i]*scale;
152 >  // cerr << "Scaling box by " << scale << "\n";
153  
154 +  for(i=0; i<3; i++)
155 +    for (j=0; j<3; j++) theBox[i][j] = Hmat[i][j]*scale;
156 +
157    setBoxM(theBox);
158  
159   }
160  
161 < void SimInfo::calcHmatI( void ) {
162 <
163 <  double C[3][3];
164 <  double detHmat;
189 <  int i, j, k;
161 > void SimInfo::calcHmatInv( void ) {
162 >  
163 >  int oldOrtho;
164 >  int i,j;
165    double smallDiag;
166    double tol;
167    double sanity[3][3];
168  
169 <  // calculate the adjunct of Hmat;
169 >  invertMat3( Hmat, HmatInv );
170  
171 <  C[0][0] =  ( Hmat[4]*Hmat[8]) - (Hmat[7]*Hmat[5]);
197 <  C[1][0] = -( Hmat[1]*Hmat[8]) + (Hmat[7]*Hmat[2]);
198 <  C[2][0] =  ( Hmat[1]*Hmat[5]) - (Hmat[4]*Hmat[2]);
199 <
200 <  C[0][1] = -( Hmat[3]*Hmat[8]) + (Hmat[6]*Hmat[5]);
201 <  C[1][1] =  ( Hmat[0]*Hmat[8]) - (Hmat[6]*Hmat[2]);
202 <  C[2][1] = -( Hmat[0]*Hmat[5]) + (Hmat[3]*Hmat[2]);
203 <
204 <  C[0][2] =  ( Hmat[3]*Hmat[7]) - (Hmat[6]*Hmat[4]);
205 <  C[1][2] = -( Hmat[0]*Hmat[7]) + (Hmat[6]*Hmat[1]);
206 <  C[2][2] =  ( Hmat[0]*Hmat[4]) - (Hmat[3]*Hmat[1]);
207 <
208 <  // calcutlate the determinant of Hmat
171 >  // check to see if Hmat is orthorhombic
172    
173 <  detHmat = 0.0;
211 <  for(i=0; i<3; i++) detHmat += Hmat[i] * C[i][0];
173 >  oldOrtho = orthoRhombic;
174  
175 <  
176 <  // H^-1 = C^T / det(H)
177 <  
178 <  i=0;
217 <  for(j=0; j<3; j++){
218 <    for(k=0; k<3; k++){
175 >  smallDiag = fabs(Hmat[0][0]);
176 >  if(smallDiag > fabs(Hmat[1][1])) smallDiag = fabs(Hmat[1][1]);
177 >  if(smallDiag > fabs(Hmat[2][2])) smallDiag = fabs(Hmat[2][2]);
178 >  tol = smallDiag * orthoTolerance;
179  
180 <      HmatI[i] = C[j][k] / detHmat;
181 <      i++;
182 <    }
183 <  }
184 <
185 <  // sanity check
186 <
187 <  for(i=0; i<3; i++){
228 <    for(j=0; j<3; j++){
229 <      
230 <      sanity[i][j] = 0.0;
231 <      for(k=0; k<3; k++){
232 <        sanity[i][j] += Hmat[3*k+i] * HmatI[3*j+k];
180 >  orthoRhombic = 1;
181 >  
182 >  for (i = 0; i < 3; i++ ) {
183 >    for (j = 0 ; j < 3; j++) {
184 >      if (i != j) {
185 >        if (orthoRhombic) {
186 >          if ( fabs(Hmat[i][j]) >= tol) orthoRhombic = 0;
187 >        }        
188        }
189      }
190    }
191  
192 <  cerr << "sanity => \n"
238 <       << sanity[0][0] << "\t" << sanity[0][1] << "\t" << sanity [0][2] << "\n"
239 <       << sanity[1][0] << "\t" << sanity[1][1] << "\t" << sanity [1][2] << "\n"
240 <       << sanity[2][0] << "\t" << sanity[2][1] << "\t" << sanity [2][2]
241 <       << "\n";
192 >  if( oldOrtho != orthoRhombic ){
193      
194 <
195 <  // check to see if Hmat is orthorhombic
196 <  
197 <  smallDiag = Hmat[0];
198 <  if(smallDiag > Hmat[4]) smallDiag = Hmat[4];
199 <  if(smallDiag > Hmat[8]) smallDiag = Hmat[8];
200 <  tol = smallDiag * 1E-6;
201 <
202 <  orthoRhombic = 1;
252 <  for(i=0; (i<9) && orthoRhombic; i++){
253 <    
254 <    if( (i%4) ){ // ignore the diagonals (0, 4, and 8)
255 <      orthoRhombic = (Hmat[i] <= tol);
194 >    if( orthoRhombic ){
195 >      sprintf( painCave.errMsg,
196 >               "OOPSE is switching from the default Non-Orthorhombic\n"
197 >               "\tto the faster Orthorhombic periodic boundary computations.\n"
198 >               "\tThis is usually a good thing, but if you wan't the\n"
199 >               "\tNon-Orthorhombic computations, make the orthoBoxTolerance\n"
200 >               "\tvariable ( currently set to %G ) smaller.\n",
201 >               orthoTolerance);
202 >      simError();
203      }
204 +    else {
205 +      sprintf( painCave.errMsg,
206 +               "OOPSE is switching from the faster Orthorhombic to the more\n"
207 +               "\tflexible Non-Orthorhombic periodic boundary computations.\n"
208 +               "\tThis is usually because the box has deformed under\n"
209 +               "\tNPTf integration. If you wan't to live on the edge with\n"
210 +               "\tthe Orthorhombic computations, make the orthoBoxTolerance\n"
211 +               "\tvariable ( currently set to %G ) larger.\n",
212 +               orthoTolerance);
213 +      simError();
214 +    }
215    }
258    
216   }
217  
218   void SimInfo::calcBoxL( void ){
219  
220    double dx, dy, dz, dsq;
264  int i;
221  
222 <  // boxVol = h1 (dot) h2 (cross) h3
222 >  // boxVol = Determinant of Hmat
223  
224 <  boxVol = Hmat[0] * ( (Hmat[4]*Hmat[8]) - (Hmat[7]*Hmat[5]) )
269 <         + Hmat[1] * ( (Hmat[5]*Hmat[6]) - (Hmat[8]*Hmat[3]) )
270 <         + Hmat[2] * ( (Hmat[3]*Hmat[7]) - (Hmat[6]*Hmat[4]) );
224 >  boxVol = matDet3( Hmat );
225  
272
226    // boxLx
227    
228 <  dx = Hmat[0]; dy = Hmat[1]; dz = Hmat[2];
228 >  dx = Hmat[0][0]; dy = Hmat[1][0]; dz = Hmat[2][0];
229    dsq = dx*dx + dy*dy + dz*dz;
230 <  boxLx = sqrt( dsq );
230 >  boxL[0] = sqrt( dsq );
231 >  //maxCutoff = 0.5 * boxL[0];
232  
233    // boxLy
234    
235 <  dx = Hmat[3]; dy = Hmat[4]; dz = Hmat[5];
235 >  dx = Hmat[0][1]; dy = Hmat[1][1]; dz = Hmat[2][1];
236    dsq = dx*dx + dy*dy + dz*dz;
237 <  boxLy = sqrt( dsq );
237 >  boxL[1] = sqrt( dsq );
238 >  //if( (0.5 * boxL[1]) < maxCutoff ) maxCutoff = 0.5 * boxL[1];
239  
240 +
241    // boxLz
242    
243 <  dx = Hmat[6]; dy = Hmat[7]; dz = Hmat[8];
243 >  dx = Hmat[0][2]; dy = Hmat[1][2]; dz = Hmat[2][2];
244    dsq = dx*dx + dy*dy + dz*dz;
245 <  boxLz = sqrt( dsq );
245 >  boxL[2] = sqrt( dsq );
246 >  //if( (0.5 * boxL[2]) < maxCutoff ) maxCutoff = 0.5 * boxL[2];
247 >
248 >  //calculate the max cutoff
249 >  maxCutoff =  calcMaxCutOff();
250    
251 +  checkCutOffs();
252 +
253   }
254  
255  
256 + double SimInfo::calcMaxCutOff(){
257 +
258 +  double ri[3], rj[3], rk[3];
259 +  double rij[3], rjk[3], rki[3];
260 +  double minDist;
261 +
262 +  ri[0] = Hmat[0][0];
263 +  ri[1] = Hmat[1][0];
264 +  ri[2] = Hmat[2][0];
265 +
266 +  rj[0] = Hmat[0][1];
267 +  rj[1] = Hmat[1][1];
268 +  rj[2] = Hmat[2][1];
269 +
270 +  rk[0] = Hmat[0][2];
271 +  rk[1] = Hmat[1][2];
272 +  rk[2] = Hmat[2][2];
273 +    
274 +  crossProduct3(ri, rj, rij);
275 +  distXY = dotProduct3(rk,rij) / norm3(rij);
276 +
277 +  crossProduct3(rj,rk, rjk);
278 +  distYZ = dotProduct3(ri,rjk) / norm3(rjk);
279 +
280 +  crossProduct3(rk,ri, rki);
281 +  distZX = dotProduct3(rj,rki) / norm3(rki);
282 +
283 +  minDist = min(min(distXY, distYZ), distZX);
284 +  return minDist/2;
285 +  
286 + }
287 +
288   void SimInfo::wrapVector( double thePos[3] ){
289  
290 <  int i, j, k;
290 >  int i;
291    double scaled[3];
292  
293    if( !orthoRhombic ){
294      // calc the scaled coordinates.
295 +  
296 +
297 +    matVecMul3(HmatInv, thePos, scaled);
298      
299      for(i=0; i<3; i++)
303      scaled[i] =
304        thePos[0]*HmatI[i] + thePos[1]*HmatI[i+3] + thePos[3]*HmatI[i+6];
305    
306    // wrap the scaled coordinates
307    
308    for(i=0; i<3; i++)
300        scaled[i] -= roundMe(scaled[i]);
301      
302      // calc the wrapped real coordinates from the wrapped scaled coordinates
303      
304 <    for(i=0; i<3; i++)
305 <      thePos[i] =
315 <        scaled[0]*Hmat[i] + scaled[1]*Hmat[i+3] + scaled[2]*Hmat[i+6];
304 >    matVecMul3(Hmat, scaled, thePos);
305 >
306    }
307    else{
308      // calc the scaled coordinates.
309      
310      for(i=0; i<3; i++)
311 <      scaled[i] = thePos[i]*HmatI[i*4];
311 >      scaled[i] = thePos[i]*HmatInv[i][i];
312      
313      // wrap the scaled coordinates
314      
# Line 328 | Line 318 | void SimInfo::wrapVector( double thePos[3] ){
318      // calc the wrapped real coordinates from the wrapped scaled coordinates
319      
320      for(i=0; i<3; i++)
321 <      thePos[i] = scaled[i]*Hmat[i*4];
321 >      thePos[i] = scaled[i]*Hmat[i][i];
322    }
323      
334    
324   }
325  
326  
327   int SimInfo::getNDF(){
328 <  int ndf_local, ndf;
328 >  int ndf_local;
329 >
330 >  ndf_local = 0;
331    
332 <  ndf_local = 3 * n_atoms + 3 * n_oriented - n_constraints;
332 >  for(int i = 0; i < integrableObjects.size(); i++){
333 >    ndf_local += 3;
334 >    if (integrableObjects[i]->isDirectional()) {
335 >      if (integrableObjects[i]->isLinear())
336 >        ndf_local += 2;
337 >      else
338 >        ndf_local += 3;
339 >    }
340 >  }
341  
342 +  // n_constraints is local, so subtract them on each processor:
343 +
344 +  ndf_local -= n_constraints;
345 +
346   #ifdef IS_MPI
347    MPI_Allreduce(&ndf_local,&ndf,1,MPI_INT,MPI_SUM, MPI_COMM_WORLD);
348   #else
349    ndf = ndf_local;
350   #endif
351  
352 <  ndf = ndf - 3;
352 >  // nZconstraints is global, as are the 3 COM translations for the
353 >  // entire system:
354  
355 +  ndf = ndf - 3 - nZconstraints;
356 +
357    return ndf;
358   }
359  
360   int SimInfo::getNDFraw() {
361 <  int ndfRaw_local, ndfRaw;
361 >  int ndfRaw_local;
362  
363    // Raw degrees of freedom that we have to set
364 <  ndfRaw_local = 3 * n_atoms + 3 * n_oriented;
365 <  
364 >  ndfRaw_local = 0;
365 >
366 >  for(int i = 0; i < integrableObjects.size(); i++){
367 >    ndfRaw_local += 3;
368 >    if (integrableObjects[i]->isDirectional()) {
369 >       if (integrableObjects[i]->isLinear())
370 >        ndfRaw_local += 2;
371 >      else
372 >        ndfRaw_local += 3;
373 >    }
374 >  }
375 >    
376   #ifdef IS_MPI
377    MPI_Allreduce(&ndfRaw_local,&ndfRaw,1,MPI_INT,MPI_SUM, MPI_COMM_WORLD);
378   #else
# Line 365 | Line 381 | int SimInfo::getNDFraw() {
381  
382    return ndfRaw;
383   }
384 <
384 >
385 > int SimInfo::getNDFtranslational() {
386 >  int ndfTrans_local;
387 >
388 >  ndfTrans_local = 3 * integrableObjects.size() - n_constraints;
389 >
390 >
391 > #ifdef IS_MPI
392 >  MPI_Allreduce(&ndfTrans_local,&ndfTrans,1,MPI_INT,MPI_SUM, MPI_COMM_WORLD);
393 > #else
394 >  ndfTrans = ndfTrans_local;
395 > #endif
396 >
397 >  ndfTrans = ndfTrans - 3 - nZconstraints;
398 >
399 >  return ndfTrans;
400 > }
401 >
402 > int SimInfo::getTotIntegrableObjects() {
403 >  int nObjs_local;
404 >  int nObjs;
405 >
406 >  nObjs_local =  integrableObjects.size();
407 >
408 >
409 > #ifdef IS_MPI
410 >  MPI_Allreduce(&nObjs_local,&nObjs,1,MPI_INT,MPI_SUM, MPI_COMM_WORLD);
411 > #else
412 >  nObjs = nObjs_local;
413 > #endif
414 >
415 >
416 >  return nObjs;
417 > }
418 >
419   void SimInfo::refreshSim(){
420  
421    simtype fInfo;
422    int isError;
423    int n_global;
424    int* excl;
425 <  
376 <  fInfo.rrf = 0.0;
377 <  fInfo.rt = 0.0;
425 >
426    fInfo.dielect = 0.0;
427  
428 <  fInfo.rlist = rList;
381 <  fInfo.rcut = rCut;
382 <
383 <  if( useDipole ){
384 <    fInfo.rrf = ecr;
385 <    fInfo.rt = ecr - est;
428 >  if( useDipoles ){
429      if( useReactionField )fInfo.dielect = dielectric;
430    }
431  
# Line 391 | Line 434 | void SimInfo::refreshSim(){
434    fInfo.SIM_uses_LJ = useLJ;
435    fInfo.SIM_uses_sticky = useSticky;
436    //fInfo.SIM_uses_sticky = 0;
437 <  fInfo.SIM_uses_dipoles = useDipole;
437 >  fInfo.SIM_uses_charges = useCharges;
438 >  fInfo.SIM_uses_dipoles = useDipoles;
439    //fInfo.SIM_uses_dipoles = 0;
440 <  //fInfo.SIM_uses_RF = useReactionField;
441 <  fInfo.SIM_uses_RF = 0;
440 >  fInfo.SIM_uses_RF = useReactionField;
441 >  //fInfo.SIM_uses_RF = 0;
442    fInfo.SIM_uses_GB = useGB;
443    fInfo.SIM_uses_EAM = useEAM;
444  
445 <  excl = Exclude::getArray();
446 <
445 >  n_exclude = excludes->getSize();
446 >  excl = excludes->getFortranArray();
447 >  
448   #ifdef IS_MPI
449 <  n_global = mpiSim->getTotAtoms();
449 >  n_global = mpiSim->getNAtomsGlobal();
450   #else
451    n_global = n_atoms;
452   #endif
453 <
453 >  
454    isError = 0;
455 <
455 >  
456 >  getFortranGroupArrays(this, FglobalGroupMembership, mfact);
457 >  //it may not be a good idea to pass the address of first element in vector
458 >  //since c++ standard does not require vector to be stored continuously in meomory
459 >  //Most of the compilers will organize the memory of vector continuously
460    setFsimulation( &fInfo, &n_global, &n_atoms, identArray, &n_exclude, excl,
461 <                  &nGlobalExcludes, globalExcludes, molMembershipArray,
462 <                  &isError );
463 <
461 >                  &nGlobalExcludes, globalExcludes, molMembershipArray,
462 >                  &mfact[0], &ngroup, &FglobalGroupMembership[0], &isError);
463 >  
464    if( isError ){
465 <
465 >    
466      sprintf( painCave.errMsg,
467 <             "There was an error setting the simulation information in fortran.\n" );
467 >             "There was an error setting the simulation information in fortran.\n" );
468      painCave.isFatal = 1;
469      simError();
470    }
471 <
471 >  
472   #ifdef IS_MPI
473    sprintf( checkPointMsg,
474             "succesfully sent the simulation information to fortran.\n");
475    MPIcheckPoint();
476   #endif // is_mpi
477 <
477 >  
478    this->ndf = this->getNDF();
479    this->ndfRaw = this->getNDFraw();
480 +  this->ndfTrans = this->getNDFtranslational();
481 + }
482  
483 + void SimInfo::setDefaultRcut( double theRcut ){
484 +  
485 +  haveRcut = 1;
486 +  rCut = theRcut;
487 +  rList = rCut + 1.0;
488 +  
489 +  notifyFortranCutOffs( &rCut, &rSw, &rList );
490   }
491  
492 + void SimInfo::setDefaultRcut( double theRcut, double theRsw ){
493 +
494 +  rSw = theRsw;
495 +  setDefaultRcut( theRcut );
496 + }
497 +
498 +
499 + void SimInfo::checkCutOffs( void ){
500 +  
501 +  if( boxIsInit ){
502 +    
503 +    //we need to check cutOffs against the box
504 +    
505 +    if( rCut > maxCutoff ){
506 +      sprintf( painCave.errMsg,
507 +               "cutoffRadius is too large for the current periodic box.\n"
508 +               "\tCurrent Value of cutoffRadius = %G at time %G\n "
509 +               "\tThis is larger than half of at least one of the\n"
510 +               "\tperiodic box vectors.  Right now, the Box matrix is:\n"
511 +               "\n"
512 +               "\t[ %G %G %G ]\n"
513 +               "\t[ %G %G %G ]\n"
514 +               "\t[ %G %G %G ]\n",
515 +               rCut, currentTime,
516 +               Hmat[0][0], Hmat[0][1], Hmat[0][2],
517 +               Hmat[1][0], Hmat[1][1], Hmat[1][2],
518 +               Hmat[2][0], Hmat[2][1], Hmat[2][2]);
519 +      painCave.isFatal = 1;
520 +      simError();
521 +    }    
522 +  } else {
523 +    // initialize this stuff before using it, OK?
524 +    sprintf( painCave.errMsg,
525 +             "Trying to check cutoffs without a box.\n"
526 +             "\tOOPSE should have better programmers than that.\n" );
527 +    painCave.isFatal = 1;
528 +    simError();      
529 +  }
530 +  
531 + }
532 +
533 + void SimInfo::addProperty(GenericData* prop){
534 +
535 +  map<string, GenericData*>::iterator result;
536 +  result = properties.find(prop->getID());
537 +  
538 +  //we can't simply use  properties[prop->getID()] = prop,
539 +  //it will cause memory leak if we already contain a propery which has the same name of prop
540 +  
541 +  if(result != properties.end()){
542 +    
543 +    delete (*result).second;
544 +    (*result).second = prop;
545 +      
546 +  }
547 +  else{
548 +
549 +    properties[prop->getID()] = prop;
550 +
551 +  }
552 +    
553 + }
554 +
555 + GenericData* SimInfo::getProperty(const string& propName){
556 +
557 +  map<string, GenericData*>::iterator result;
558 +  
559 +  //string lowerCaseName = ();
560 +  
561 +  result = properties.find(propName);
562 +  
563 +  if(result != properties.end())
564 +    return (*result).second;  
565 +  else  
566 +    return NULL;  
567 + }
568 +
569 +
570 + void SimInfo::getFortranGroupArrays(SimInfo* info,
571 +                                    vector<int>& FglobalGroupMembership,
572 +                                    vector<double>& mfact){
573 +  
574 +  Molecule* myMols;
575 +  Atom** myAtoms;
576 +  int numAtom;
577 +  double mtot;
578 +  int numMol;
579 +  int numCutoffGroups;
580 +  CutoffGroup* myCutoffGroup;
581 +  vector<CutoffGroup*>::iterator iterCutoff;
582 +  Atom* cutoffAtom;
583 +  vector<Atom*>::iterator iterAtom;
584 +  int atomIndex;
585 +  double totalMass;
586 +  
587 +  mfact.clear();
588 +  FglobalGroupMembership.clear();
589 +  
590 +
591 +  // Fix the silly fortran indexing problem
592 + #ifdef IS_MPI
593 +  numAtom = mpiSim->getNAtomsGlobal();
594 + #else
595 +  numAtom = n_atoms;
596 + #endif
597 +  for (int i = 0; i < numAtom; i++)
598 +    FglobalGroupMembership.push_back(globalGroupMembership[i] + 1);
599 +
600 +  myMols = info->molecules;
601 +  numMol = info->n_mol;
602 +  for(int i  = 0; i < numMol; i++){
603 +    numCutoffGroups = myMols[i].getNCutoffGroups();
604 +    for(myCutoffGroup =myMols[i].beginCutoffGroup(iterCutoff);
605 +        myCutoffGroup != NULL;
606 +        myCutoffGroup =myMols[i].nextCutoffGroup(iterCutoff)){
607 +
608 +      totalMass = myCutoffGroup->getMass();
609 +      
610 +      for(cutoffAtom = myCutoffGroup->beginAtom(iterAtom);
611 +          cutoffAtom != NULL;
612 +          cutoffAtom = myCutoffGroup->nextAtom(iterAtom)){
613 +        mfact.push_back(cutoffAtom->getMass()/totalMass);
614 +      }  
615 +    }
616 +  }
617 +
618 + }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines