/[escript]/trunk/ripley/src/Rectangle.cpp
ViewVC logotype

Contents of /trunk/ripley/src/Rectangle.cpp

Parent Directory Parent Directory | Revision Log Revision Log


Revision 4378 - (show annotations)
Wed Apr 24 06:21:42 2013 UTC (6 years, 6 months ago) by caltinay
File size: 197080 byte(s)
ripley work: Almost done - one more round in Brick left....

1
2 /*****************************************************************************
3 *
4 * Copyright (c) 2003-2013 by University of Queensland
5 * http://www.uq.edu.au
6 *
7 * Primary Business: Queensland, Australia
8 * Licensed under the Open Software License version 3.0
9 * http://www.opensource.org/licenses/osl-3.0.php
10 *
11 * Development until 2012 by Earth Systems Science Computational Center (ESSCC)
12 * Development since 2012 by School of Earth Sciences
13 *
14 *****************************************************************************/
15
16 #include <ripley/Rectangle.h>
17 #include <paso/SystemMatrix.h>
18 #include <esysUtils/esysFileWriter.h>
19
20 #ifdef USE_NETCDF
21 #include <netcdfcpp.h>
22 #endif
23
24 #if USE_SILO
25 #include <silo.h>
26 #ifdef ESYS_MPI
27 #include <pmpio.h>
28 #endif
29 #endif
30
31 #include <iomanip>
32
33 using namespace std;
34 using esysUtils::FileWriter;
35
36 namespace ripley {
37
38 Rectangle::Rectangle(int n0, int n1, double x0, double y0, double x1,
39 double y1, int d0, int d1) :
40 RipleyDomain(2)
41 {
42 // ignore subdivision parameters for serial run
43 if (m_mpiInfo->size == 1) {
44 d0=1;
45 d1=1;
46 }
47
48 bool warn=false;
49 // if number of subdivisions is non-positive, try to subdivide by the same
50 // ratio as the number of elements
51 if (d0<=0 && d1<=0) {
52 warn=true;
53 d0=max(1, (int)(sqrt(m_mpiInfo->size*(n0+1)/(float)(n1+1))));
54 d1=m_mpiInfo->size/d0;
55 if (d0*d1 != m_mpiInfo->size) {
56 // ratios not the same so subdivide side with more elements only
57 if (n0>n1) {
58 d0=0;
59 d1=1;
60 } else {
61 d0=1;
62 d1=0;
63 }
64 }
65 }
66 if (d0<=0) {
67 // d1 is preset, determine d0 - throw further down if result is no good
68 d0=m_mpiInfo->size/d1;
69 } else if (d1<=0) {
70 // d0 is preset, determine d1 - throw further down if result is no good
71 d1=m_mpiInfo->size/d0;
72 }
73
74 // ensure number of subdivisions is valid and nodes can be distributed
75 // among number of ranks
76 if (d0*d1 != m_mpiInfo->size)
77 throw RipleyException("Invalid number of spatial subdivisions");
78
79 if (warn) {
80 cout << "Warning: Automatic domain subdivision (d0=" << d0 << ", d1="
81 << d1 << "). This may not be optimal!" << endl;
82 }
83
84 double l0 = x1-x0;
85 double l1 = y1-y0;
86 m_dx[0] = l0/n0;
87 m_dx[1] = l1/n1;
88
89 if ((n0+1)%d0 > 0) {
90 n0=(int)round((float)(n0+1)/d0+0.5)*d0-1;
91 l0=m_dx[0]*n0;
92 cout << "Warning: Adjusted number of elements and length. N0="
93 << n0 << ", l0=" << l0 << endl;
94 }
95 if ((n1+1)%d1 > 0) {
96 n1=(int)round((float)(n1+1)/d1+0.5)*d1-1;
97 l1=m_dx[1]*n1;
98 cout << "Warning: Adjusted number of elements and length. N1="
99 << n1 << ", l1=" << l1 << endl;
100 }
101
102 if ((d0 > 1 && (n0+1)/d0<2) || (d1 > 1 && (n1+1)/d1<2))
103 throw RipleyException("Too few elements for the number of ranks");
104
105 m_gNE[0] = n0;
106 m_gNE[1] = n1;
107 m_origin[0] = x0;
108 m_origin[1] = y0;
109 m_length[0] = l0;
110 m_length[1] = l1;
111 m_NX[0] = d0;
112 m_NX[1] = d1;
113
114 // local number of elements (with and without overlap)
115 m_NE[0] = m_ownNE[0] = (d0>1 ? (n0+1)/d0 : n0);
116 if (m_mpiInfo->rank%d0>0 && m_mpiInfo->rank%d0<d0-1)
117 m_NE[0]++;
118 else if (d0>1 && m_mpiInfo->rank%d0==d0-1)
119 m_ownNE[0]--;
120
121 m_NE[1] = m_ownNE[1] = (d1>1 ? (n1+1)/d1 : n1);
122 if (m_mpiInfo->rank/d0>0 && m_mpiInfo->rank/d0<d1-1)
123 m_NE[1]++;
124 else if (d1>1 && m_mpiInfo->rank/d0==d1-1)
125 m_ownNE[1]--;
126
127 // local number of nodes
128 m_NN[0] = m_NE[0]+1;
129 m_NN[1] = m_NE[1]+1;
130
131 // bottom-left node is at (offset0,offset1) in global mesh
132 m_offset[0] = (n0+1)/d0*(m_mpiInfo->rank%d0);
133 if (m_offset[0] > 0)
134 m_offset[0]--;
135 m_offset[1] = (n1+1)/d1*(m_mpiInfo->rank/d0);
136 if (m_offset[1] > 0)
137 m_offset[1]--;
138
139 populateSampleIds();
140 createPattern();
141 }
142
143 Rectangle::~Rectangle()
144 {
145 Paso_SystemMatrixPattern_free(m_pattern);
146 Paso_Connector_free(m_connector);
147 }
148
149 string Rectangle::getDescription() const
150 {
151 return "ripley::Rectangle";
152 }
153
154 bool Rectangle::operator==(const AbstractDomain& other) const
155 {
156 const Rectangle* o=dynamic_cast<const Rectangle*>(&other);
157 if (o) {
158 return (RipleyDomain::operator==(other) &&
159 m_gNE[0]==o->m_gNE[0] && m_gNE[1]==o->m_gNE[1]
160 && m_origin[0]==o->m_origin[0] && m_origin[1]==o->m_origin[1]
161 && m_length[0]==o->m_length[0] && m_length[1]==o->m_length[1]
162 && m_NX[0]==o->m_NX[0] && m_NX[1]==o->m_NX[1]);
163 }
164
165 return false;
166 }
167
168 void Rectangle::readNcGrid(escript::Data& out, string filename, string varname,
169 const vector<int>& first, const vector<int>& numValues,
170 const vector<int>& multiplier) const
171 {
172 #ifdef USE_NETCDF
173 // check destination function space
174 int myN0, myN1;
175 if (out.getFunctionSpace().getTypeCode() == Nodes) {
176 myN0 = m_NN[0];
177 myN1 = m_NN[1];
178 } else if (out.getFunctionSpace().getTypeCode() == Elements ||
179 out.getFunctionSpace().getTypeCode() == ReducedElements) {
180 myN0 = m_NE[0];
181 myN1 = m_NE[1];
182 } else
183 throw RipleyException("readNcGrid(): invalid function space for output data object");
184
185 if (first.size() != 2)
186 throw RipleyException("readNcGrid(): argument 'first' must have 2 entries");
187
188 if (numValues.size() != 2)
189 throw RipleyException("readNcGrid(): argument 'numValues' must have 2 entries");
190
191 if (multiplier.size() != 2)
192 throw RipleyException("readNcGrid(): argument 'multiplier' must have 2 entries");
193 for (size_t i=0; i<multiplier.size(); i++)
194 if (multiplier[i]<1)
195 throw RipleyException("readNcGrid(): all multipliers must be positive");
196
197 // check file existence and size
198 NcFile f(filename.c_str(), NcFile::ReadOnly);
199 if (!f.is_valid())
200 throw RipleyException("readNcGrid(): cannot open file");
201
202 NcVar* var = f.get_var(varname.c_str());
203 if (!var)
204 throw RipleyException("readNcGrid(): invalid variable");
205
206 // TODO: rank>0 data support
207 const int numComp = out.getDataPointSize();
208 if (numComp > 1)
209 throw RipleyException("readNcGrid(): only scalar data supported");
210
211 const int dims = var->num_dims();
212 const long *edges = var->edges();
213
214 // is this a slice of the data object (dims!=2)?
215 // note the expected ordering of edges (as in numpy: y,x)
216 if ( (dims==2 && (numValues[1] > edges[0] || numValues[0] > edges[1]))
217 || (dims==1 && numValues[1]>1) ) {
218 throw RipleyException("readNcGrid(): not enough data in file");
219 }
220
221 // check if this rank contributes anything
222 if (first[0] >= m_offset[0]+myN0 || first[0]+numValues[0]*multiplier[0] <= m_offset[0] ||
223 first[1] >= m_offset[1]+myN1 || first[1]+numValues[1]*multiplier[1] <= m_offset[1])
224 return;
225
226 // now determine how much this rank has to write
227
228 // first coordinates in data object to write to
229 const int first0 = max(0, first[0]-m_offset[0]);
230 const int first1 = max(0, first[1]-m_offset[1]);
231 // indices to first value in file
232 const int idx0 = max(0, m_offset[0]-first[0]);
233 const int idx1 = max(0, m_offset[1]-first[1]);
234 // number of values to read
235 const int num0 = min(numValues[0]-idx0, myN0-first0);
236 const int num1 = min(numValues[1]-idx1, myN1-first1);
237
238 vector<double> values(num0*num1);
239 if (dims==2) {
240 var->set_cur(idx1, idx0);
241 var->get(&values[0], num1, num0);
242 } else {
243 var->set_cur(idx0);
244 var->get(&values[0], num0);
245 }
246
247 const int dpp = out.getNumDataPointsPerSample();
248 out.requireWrite();
249
250 for (index_t y=0; y<num1; y++) {
251 #pragma omp parallel for
252 for (index_t x=0; x<num0; x++) {
253 const int baseIndex = first0+x*multiplier[0]
254 +(first1+y*multiplier[1])*myN0;
255 const int srcIndex = y*num0+x;
256 if (!isnan(values[srcIndex])) {
257 for (index_t m1=0; m1<multiplier[1]; m1++) {
258 for (index_t m0=0; m0<multiplier[0]; m0++) {
259 const int dataIndex = baseIndex+m0+m1*myN0;
260 double* dest = out.getSampleDataRW(dataIndex);
261 for (index_t q=0; q<dpp; q++) {
262 *dest++ = values[srcIndex];
263 }
264 }
265 }
266 }
267 }
268 }
269 #else
270 throw RipleyException("readNcGrid(): not compiled with netCDF support");
271 #endif
272 }
273
274 void Rectangle::readBinaryGrid(escript::Data& out, string filename,
275 const vector<int>& first,
276 const vector<int>& numValues,
277 const vector<int>& multiplier) const
278 {
279 // check destination function space
280 int myN0, myN1;
281 if (out.getFunctionSpace().getTypeCode() == Nodes) {
282 myN0 = m_NN[0];
283 myN1 = m_NN[1];
284 } else if (out.getFunctionSpace().getTypeCode() == Elements ||
285 out.getFunctionSpace().getTypeCode() == ReducedElements) {
286 myN0 = m_NE[0];
287 myN1 = m_NE[1];
288 } else
289 throw RipleyException("readBinaryGrid(): invalid function space for output data object");
290
291 // check file existence and size
292 ifstream f(filename.c_str(), ifstream::binary);
293 if (f.fail()) {
294 throw RipleyException("readBinaryGrid(): cannot open file");
295 }
296 f.seekg(0, ios::end);
297 const int numComp = out.getDataPointSize();
298 const int filesize = f.tellg();
299 const int reqsize = numValues[0]*numValues[1]*numComp*sizeof(float);
300 if (filesize < reqsize) {
301 f.close();
302 throw RipleyException("readBinaryGrid(): not enough data in file");
303 }
304
305 // check if this rank contributes anything
306 if (first[0] >= m_offset[0]+myN0 || first[0]+numValues[0] <= m_offset[0] ||
307 first[1] >= m_offset[1]+myN1 || first[1]+numValues[1] <= m_offset[1]) {
308 f.close();
309 return;
310 }
311
312 // now determine how much this rank has to write
313
314 // first coordinates in data object to write to
315 const int first0 = max(0, first[0]-m_offset[0]);
316 const int first1 = max(0, first[1]-m_offset[1]);
317 // indices to first value in file
318 const int idx0 = max(0, m_offset[0]-first[0]);
319 const int idx1 = max(0, m_offset[1]-first[1]);
320 // number of values to read
321 const int num0 = min(numValues[0]-idx0, myN0-first0);
322 const int num1 = min(numValues[1]-idx1, myN1-first1);
323
324 out.requireWrite();
325 vector<float> values(num0*numComp);
326 const int dpp = out.getNumDataPointsPerSample();
327
328 for (index_t y=0; y<num1; y++) {
329 const int fileofs = numComp*(idx0+(idx1+y)*numValues[0]);
330 f.seekg(fileofs*sizeof(float));
331 f.read((char*)&values[0], num0*numComp*sizeof(float));
332 for (index_t x=0; x<num0; x++) {
333 const int baseIndex = first0+x*multiplier[0]
334 +(first1+y*multiplier[1])*myN0;
335 for (index_t m1=0; m1<multiplier[1]; m1++) {
336 for (index_t m0=0; m0<multiplier[0]; m0++) {
337 const int dataIndex = baseIndex+m0+m1*myN0;
338 double* dest = out.getSampleDataRW(dataIndex);
339 for (index_t c=0; c<numComp; c++) {
340 if (!std::isnan(values[x*numComp+c])) {
341 for (index_t q=0; q<dpp; q++) {
342 *dest++ = static_cast<double>(values[x*numComp+c]);
343 }
344 }
345 }
346 }
347 }
348 }
349 }
350
351 f.close();
352 }
353
354 void Rectangle::writeBinaryGrid(const escript::Data& in, string filename,
355 int byteOrder, int dataType) const
356 {
357 // the mapping is not universally correct but should work on our
358 // supported platforms
359 switch (dataType) {
360 case DATATYPE_INT32:
361 writeBinaryGridImpl<int>(in, filename, byteOrder);
362 break;
363 case DATATYPE_FLOAT32:
364 writeBinaryGridImpl<float>(in, filename, byteOrder);
365 break;
366 case DATATYPE_FLOAT64:
367 writeBinaryGridImpl<double>(in, filename, byteOrder);
368 break;
369 default:
370 throw RipleyException("writeBinaryGrid(): invalid or unsupported datatype");
371 }
372 }
373
374 template<typename ValueType>
375 void Rectangle::writeBinaryGridImpl(const escript::Data& in,
376 const string& filename, int byteOrder) const
377 {
378 // check function space and determine number of points
379 int myN0, myN1;
380 int totalN0, totalN1;
381 if (in.getFunctionSpace().getTypeCode() == Nodes) {
382 myN0 = m_NN[0];
383 myN1 = m_NN[1];
384 totalN0 = m_gNE[0]+1;
385 totalN1 = m_gNE[1]+1;
386 } else if (in.getFunctionSpace().getTypeCode() == Elements ||
387 in.getFunctionSpace().getTypeCode() == ReducedElements) {
388 myN0 = m_NE[0];
389 myN1 = m_NE[1];
390 totalN0 = m_gNE[0];
391 totalN1 = m_gNE[1];
392 } else
393 throw RipleyException("writeBinaryGrid(): invalid function space of data object");
394
395 const int numComp = in.getDataPointSize();
396 const int dpp = in.getNumDataPointsPerSample();
397
398 if (numComp > 1 || dpp > 1)
399 throw RipleyException("writeBinaryGrid(): only scalar, single-value data supported");
400
401 escript::Data* _in = const_cast<escript::Data*>(&in);
402 const int fileSize = sizeof(ValueType)*numComp*dpp*totalN0*totalN1;
403
404 // from here on we know that each sample consists of one value
405 FileWriter* fw = new FileWriter();
406 fw->openFile(filename, fileSize);
407 MPIBarrier();
408
409 for (index_t y=0; y<myN1; y++) {
410 const int fileofs = (m_offset[0]+(m_offset[1]+y)*totalN0)*sizeof(ValueType);
411 ostringstream oss;
412
413 for (index_t x=0; x<myN0; x++) {
414 const double* sample = _in->getSampleDataRO(y*myN0+x);
415 ValueType fvalue = static_cast<ValueType>(*sample);
416 if (byteOrder == BYTEORDER_NATIVE) {
417 oss.write((char*)&fvalue, sizeof(fvalue));
418 } else {
419 char* value = reinterpret_cast<char*>(&fvalue);
420 oss.write(byte_swap32(value), sizeof(fvalue));
421 }
422 }
423 fw->writeAt(oss, fileofs);
424 }
425 fw->close();
426 }
427
428 void Rectangle::dump(const string& fileName) const
429 {
430 #if USE_SILO
431 string fn(fileName);
432 if (fileName.length() < 6 || fileName.compare(fileName.length()-5, 5, ".silo") != 0) {
433 fn+=".silo";
434 }
435
436 int driver=DB_HDF5;
437 DBfile* dbfile = NULL;
438 const char* blockDirFmt = "/block%04d";
439
440 #ifdef ESYS_MPI
441 PMPIO_baton_t* baton = NULL;
442 const int NUM_SILO_FILES = 1;
443 #endif
444
445 if (m_mpiInfo->size > 1) {
446 #ifdef ESYS_MPI
447 baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
448 0x1337, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
449 PMPIO_DefaultClose, (void*)&driver);
450 // try the fallback driver in case of error
451 if (!baton && driver != DB_PDB) {
452 driver = DB_PDB;
453 baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
454 0x1338, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
455 PMPIO_DefaultClose, (void*)&driver);
456 }
457 if (baton) {
458 char siloPath[64];
459 snprintf(siloPath, 64, blockDirFmt, PMPIO_RankInGroup(baton, m_mpiInfo->rank));
460 dbfile = (DBfile*) PMPIO_WaitForBaton(baton, fn.c_str(), siloPath);
461 }
462 #endif
463 } else {
464 dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
465 getDescription().c_str(), driver);
466 // try the fallback driver in case of error
467 if (!dbfile && driver != DB_PDB) {
468 driver = DB_PDB;
469 dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
470 getDescription().c_str(), driver);
471 }
472 char siloPath[64];
473 snprintf(siloPath, 64, blockDirFmt, 0);
474 DBMkDir(dbfile, siloPath);
475 DBSetDir(dbfile, siloPath);
476 }
477
478 if (!dbfile)
479 throw RipleyException("dump: Could not create Silo file");
480
481 /*
482 if (driver==DB_HDF5) {
483 // gzip level 1 already provides good compression with minimal
484 // performance penalty. Some tests showed that gzip levels >3 performed
485 // rather badly on escript data both in terms of time and space
486 DBSetCompression("ERRMODE=FALLBACK METHOD=GZIP LEVEL=1");
487 }
488 */
489
490 boost::scoped_ptr<double> x(new double[m_NN[0]]);
491 boost::scoped_ptr<double> y(new double[m_NN[1]]);
492 double* coords[2] = { x.get(), y.get() };
493 #pragma omp parallel
494 {
495 #pragma omp for nowait
496 for (dim_t i0 = 0; i0 < m_NN[0]; i0++) {
497 coords[0][i0]=getLocalCoordinate(i0, 0);
498 }
499 #pragma omp for nowait
500 for (dim_t i1 = 0; i1 < m_NN[1]; i1++) {
501 coords[1][i1]=getLocalCoordinate(i1, 1);
502 }
503 }
504 int* dims = const_cast<int*>(getNumNodesPerDim());
505
506 // write mesh
507 DBPutQuadmesh(dbfile, "mesh", NULL, coords, dims, 2, DB_DOUBLE,
508 DB_COLLINEAR, NULL);
509
510 // write node ids
511 DBPutQuadvar1(dbfile, "nodeId", "mesh", (void*)&m_nodeId[0], dims, 2,
512 NULL, 0, DB_INT, DB_NODECENT, NULL);
513
514 // write element ids
515 dims = const_cast<int*>(getNumElementsPerDim());
516 DBPutQuadvar1(dbfile, "elementId", "mesh", (void*)&m_elementId[0],
517 dims, 2, NULL, 0, DB_INT, DB_ZONECENT, NULL);
518
519 // rank 0 writes multimesh and multivar
520 if (m_mpiInfo->rank == 0) {
521 vector<string> tempstrings;
522 vector<char*> names;
523 for (dim_t i=0; i<m_mpiInfo->size; i++) {
524 stringstream path;
525 path << "/block" << setw(4) << setfill('0') << right << i << "/mesh";
526 tempstrings.push_back(path.str());
527 names.push_back((char*)tempstrings.back().c_str());
528 }
529 vector<int> types(m_mpiInfo->size, DB_QUAD_RECT);
530 DBSetDir(dbfile, "/");
531 DBPutMultimesh(dbfile, "multimesh", m_mpiInfo->size, &names[0],
532 &types[0], NULL);
533 tempstrings.clear();
534 names.clear();
535 for (dim_t i=0; i<m_mpiInfo->size; i++) {
536 stringstream path;
537 path << "/block" << setw(4) << setfill('0') << right << i << "/nodeId";
538 tempstrings.push_back(path.str());
539 names.push_back((char*)tempstrings.back().c_str());
540 }
541 types.assign(m_mpiInfo->size, DB_QUADVAR);
542 DBPutMultivar(dbfile, "nodeId", m_mpiInfo->size, &names[0],
543 &types[0], NULL);
544 tempstrings.clear();
545 names.clear();
546 for (dim_t i=0; i<m_mpiInfo->size; i++) {
547 stringstream path;
548 path << "/block" << setw(4) << setfill('0') << right << i << "/elementId";
549 tempstrings.push_back(path.str());
550 names.push_back((char*)tempstrings.back().c_str());
551 }
552 DBPutMultivar(dbfile, "elementId", m_mpiInfo->size, &names[0],
553 &types[0], NULL);
554 }
555
556 if (m_mpiInfo->size > 1) {
557 #ifdef ESYS_MPI
558 PMPIO_HandOffBaton(baton, dbfile);
559 PMPIO_Finish(baton);
560 #endif
561 } else {
562 DBClose(dbfile);
563 }
564
565 #else // USE_SILO
566 throw RipleyException("dump: no Silo support");
567 #endif
568 }
569
570 const int* Rectangle::borrowSampleReferenceIDs(int fsType) const
571 {
572 switch (fsType) {
573 case Nodes:
574 case ReducedNodes: // FIXME: reduced
575 return &m_nodeId[0];
576 case DegreesOfFreedom:
577 case ReducedDegreesOfFreedom: // FIXME: reduced
578 return &m_dofId[0];
579 case Elements:
580 case ReducedElements:
581 return &m_elementId[0];
582 case FaceElements:
583 case ReducedFaceElements:
584 return &m_faceId[0];
585 default:
586 break;
587 }
588
589 stringstream msg;
590 msg << "borrowSampleReferenceIDs: invalid function space type " << fsType;
591 throw RipleyException(msg.str());
592 }
593
594 bool Rectangle::ownSample(int fsType, index_t id) const
595 {
596 if (getMPISize()==1)
597 return true;
598
599 switch (fsType) {
600 case Nodes:
601 case ReducedNodes: // FIXME: reduced
602 return (m_dofMap[id] < getNumDOF());
603 case DegreesOfFreedom:
604 case ReducedDegreesOfFreedom:
605 return true;
606 case Elements:
607 case ReducedElements:
608 // check ownership of element's bottom left node
609 return (m_dofMap[id%m_NE[0]+m_NN[0]*(id/m_NE[0])] < getNumDOF());
610 case FaceElements:
611 case ReducedFaceElements:
612 {
613 // determine which face the sample belongs to before
614 // checking ownership of corresponding element's first node
615 dim_t n=0;
616 for (size_t i=0; i<4; i++) {
617 n+=m_faceCount[i];
618 if (id<n) {
619 index_t k;
620 if (i==1)
621 k=m_NN[0]-2;
622 else if (i==3)
623 k=m_NN[0]*(m_NN[1]-2);
624 else
625 k=0;
626 // determine whether to move right or up
627 const index_t delta=(i/2==0 ? m_NN[0] : 1);
628 return (m_dofMap[k+(id-n+m_faceCount[i])*delta] < getNumDOF());
629 }
630 }
631 return false;
632 }
633 default:
634 break;
635 }
636
637 stringstream msg;
638 msg << "ownSample: invalid function space type " << fsType;
639 throw RipleyException(msg.str());
640 }
641
642 void Rectangle::setToNormal(escript::Data& out) const
643 {
644 if (out.getFunctionSpace().getTypeCode() == FaceElements) {
645 out.requireWrite();
646 #pragma omp parallel
647 {
648 if (m_faceOffset[0] > -1) {
649 #pragma omp for nowait
650 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
651 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
652 // set vector at two quadrature points
653 *o++ = -1.;
654 *o++ = 0.;
655 *o++ = -1.;
656 *o = 0.;
657 }
658 }
659
660 if (m_faceOffset[1] > -1) {
661 #pragma omp for nowait
662 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
663 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
664 // set vector at two quadrature points
665 *o++ = 1.;
666 *o++ = 0.;
667 *o++ = 1.;
668 *o = 0.;
669 }
670 }
671
672 if (m_faceOffset[2] > -1) {
673 #pragma omp for nowait
674 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
675 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
676 // set vector at two quadrature points
677 *o++ = 0.;
678 *o++ = -1.;
679 *o++ = 0.;
680 *o = -1.;
681 }
682 }
683
684 if (m_faceOffset[3] > -1) {
685 #pragma omp for nowait
686 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
687 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
688 // set vector at two quadrature points
689 *o++ = 0.;
690 *o++ = 1.;
691 *o++ = 0.;
692 *o = 1.;
693 }
694 }
695 } // end of parallel section
696 } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
697 out.requireWrite();
698 #pragma omp parallel
699 {
700 if (m_faceOffset[0] > -1) {
701 #pragma omp for nowait
702 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
703 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
704 *o++ = -1.;
705 *o = 0.;
706 }
707 }
708
709 if (m_faceOffset[1] > -1) {
710 #pragma omp for nowait
711 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
712 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
713 *o++ = 1.;
714 *o = 0.;
715 }
716 }
717
718 if (m_faceOffset[2] > -1) {
719 #pragma omp for nowait
720 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
721 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
722 *o++ = 0.;
723 *o = -1.;
724 }
725 }
726
727 if (m_faceOffset[3] > -1) {
728 #pragma omp for nowait
729 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
730 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
731 *o++ = 0.;
732 *o = 1.;
733 }
734 }
735 } // end of parallel section
736
737 } else {
738 stringstream msg;
739 msg << "setToNormal: invalid function space type "
740 << out.getFunctionSpace().getTypeCode();
741 throw RipleyException(msg.str());
742 }
743 }
744
745 void Rectangle::setToSize(escript::Data& out) const
746 {
747 if (out.getFunctionSpace().getTypeCode() == Elements
748 || out.getFunctionSpace().getTypeCode() == ReducedElements) {
749 out.requireWrite();
750 const dim_t numQuad=out.getNumDataPointsPerSample();
751 const double size=sqrt(m_dx[0]*m_dx[0]+m_dx[1]*m_dx[1]);
752 #pragma omp parallel for
753 for (index_t k = 0; k < getNumElements(); ++k) {
754 double* o = out.getSampleDataRW(k);
755 fill(o, o+numQuad, size);
756 }
757 } else if (out.getFunctionSpace().getTypeCode() == FaceElements
758 || out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
759 out.requireWrite();
760 const dim_t numQuad=out.getNumDataPointsPerSample();
761 #pragma omp parallel
762 {
763 if (m_faceOffset[0] > -1) {
764 #pragma omp for nowait
765 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
766 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
767 fill(o, o+numQuad, m_dx[1]);
768 }
769 }
770
771 if (m_faceOffset[1] > -1) {
772 #pragma omp for nowait
773 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
774 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
775 fill(o, o+numQuad, m_dx[1]);
776 }
777 }
778
779 if (m_faceOffset[2] > -1) {
780 #pragma omp for nowait
781 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
782 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
783 fill(o, o+numQuad, m_dx[0]);
784 }
785 }
786
787 if (m_faceOffset[3] > -1) {
788 #pragma omp for nowait
789 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
790 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
791 fill(o, o+numQuad, m_dx[0]);
792 }
793 }
794 } // end of parallel section
795
796 } else {
797 stringstream msg;
798 msg << "setToSize: invalid function space type "
799 << out.getFunctionSpace().getTypeCode();
800 throw RipleyException(msg.str());
801 }
802 }
803
804 void Rectangle::Print_Mesh_Info(const bool full) const
805 {
806 RipleyDomain::Print_Mesh_Info(full);
807 if (full) {
808 cout << " Id Coordinates" << endl;
809 cout.precision(15);
810 cout.setf(ios::scientific, ios::floatfield);
811 for (index_t i=0; i < getNumNodes(); i++) {
812 cout << " " << setw(5) << m_nodeId[i]
813 << " " << getLocalCoordinate(i%m_NN[0], 0)
814 << " " << getLocalCoordinate(i/m_NN[0], 1) << endl;
815 }
816 }
817 }
818
819
820 //protected
821 void Rectangle::assembleCoordinates(escript::Data& arg) const
822 {
823 escriptDataC x = arg.getDataC();
824 int numDim = m_numDim;
825 if (!isDataPointShapeEqual(&x, 1, &numDim))
826 throw RipleyException("setToX: Invalid Data object shape");
827 if (!numSamplesEqual(&x, 1, getNumNodes()))
828 throw RipleyException("setToX: Illegal number of samples in Data object");
829
830 arg.requireWrite();
831 #pragma omp parallel for
832 for (dim_t i1 = 0; i1 < m_NN[1]; i1++) {
833 for (dim_t i0 = 0; i0 < m_NN[0]; i0++) {
834 double* point = arg.getSampleDataRW(i0+m_NN[0]*i1);
835 point[0] = getLocalCoordinate(i0, 0);
836 point[1] = getLocalCoordinate(i1, 1);
837 }
838 }
839 }
840
841 //protected
842 void Rectangle::assembleGradient(escript::Data& out, escript::Data& in) const
843 {
844 const dim_t numComp = in.getDataPointSize();
845 const double cx0 = .21132486540518711775/m_dx[0];
846 const double cx1 = .78867513459481288225/m_dx[0];
847 const double cx2 = 1./m_dx[0];
848 const double cy0 = .21132486540518711775/m_dx[1];
849 const double cy1 = .78867513459481288225/m_dx[1];
850 const double cy2 = 1./m_dx[1];
851
852 if (out.getFunctionSpace().getTypeCode() == Elements) {
853 out.requireWrite();
854 #pragma omp parallel
855 {
856 vector<double> f_00(numComp);
857 vector<double> f_01(numComp);
858 vector<double> f_10(numComp);
859 vector<double> f_11(numComp);
860 #pragma omp for
861 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
862 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
863 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
864 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
865 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
866 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
867 double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
868 for (index_t i=0; i < numComp; ++i) {
869 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx1 + (f_11[i]-f_01[i])*cx0;
870 o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy1 + (f_11[i]-f_10[i])*cy0;
871 o[INDEX3(i,0,1,numComp,2)] = (f_10[i]-f_00[i])*cx1 + (f_11[i]-f_01[i])*cx0;
872 o[INDEX3(i,1,1,numComp,2)] = (f_01[i]-f_00[i])*cy0 + (f_11[i]-f_10[i])*cy1;
873 o[INDEX3(i,0,2,numComp,2)] = (f_10[i]-f_00[i])*cx0 + (f_11[i]-f_01[i])*cx1;
874 o[INDEX3(i,1,2,numComp,2)] = (f_01[i]-f_00[i])*cy1 + (f_11[i]-f_10[i])*cy0;
875 o[INDEX3(i,0,3,numComp,2)] = (f_10[i]-f_00[i])*cx0 + (f_11[i]-f_01[i])*cx1;
876 o[INDEX3(i,1,3,numComp,2)] = (f_01[i]-f_00[i])*cy0 + (f_11[i]-f_10[i])*cy1;
877 } // end of component loop i
878 } // end of k0 loop
879 } // end of k1 loop
880 } // end of parallel section
881 } else if (out.getFunctionSpace().getTypeCode() == ReducedElements) {
882 out.requireWrite();
883 #pragma omp parallel
884 {
885 vector<double> f_00(numComp);
886 vector<double> f_01(numComp);
887 vector<double> f_10(numComp);
888 vector<double> f_11(numComp);
889 #pragma omp for
890 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
891 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
892 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
893 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
894 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
895 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
896 double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
897 for (index_t i=0; i < numComp; ++i) {
898 o[INDEX3(i,0,0,numComp,2)] = (f_10[i] + f_11[i] - f_00[i] - f_01[i])*cx2/2;
899 o[INDEX3(i,1,0,numComp,2)] = (f_01[i] + f_11[i] - f_00[i] - f_10[i])*cy2/2;
900 } // end of component loop i
901 } // end of k0 loop
902 } // end of k1 loop
903 } // end of parallel section
904 } else if (out.getFunctionSpace().getTypeCode() == FaceElements) {
905 out.requireWrite();
906 #pragma omp parallel
907 {
908 vector<double> f_00(numComp);
909 vector<double> f_01(numComp);
910 vector<double> f_10(numComp);
911 vector<double> f_11(numComp);
912 if (m_faceOffset[0] > -1) {
913 #pragma omp for nowait
914 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
915 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
916 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
917 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(1,k1, m_NN[0])), numComp*sizeof(double));
918 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(1,k1+1, m_NN[0])), numComp*sizeof(double));
919 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
920 for (index_t i=0; i < numComp; ++i) {
921 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx1 + (f_11[i]-f_01[i])*cx0;
922 o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy2;
923 o[INDEX3(i,0,1,numComp,2)] = (f_10[i]-f_00[i])*cx0 + (f_11[i]-f_01[i])*cx1;
924 o[INDEX3(i,1,1,numComp,2)] = (f_01[i]-f_00[i])*cy2;
925 } // end of component loop i
926 } // end of k1 loop
927 } // end of face 0
928 if (m_faceOffset[1] > -1) {
929 #pragma omp for nowait
930 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
931 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1, m_NN[0])), numComp*sizeof(double));
932 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1+1, m_NN[0])), numComp*sizeof(double));
933 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
934 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
935 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
936 for (index_t i=0; i < numComp; ++i) {
937 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx1 + (f_11[i]-f_01[i])*cx0;
938 o[INDEX3(i,1,0,numComp,2)] = (f_11[i]-f_10[i])*cy2;
939 o[INDEX3(i,0,1,numComp,2)] = (f_10[i]-f_00[i])*cx0 + (f_11[i]-f_01[i])*cx1;
940 o[INDEX3(i,1,1,numComp,2)] = (f_11[i]-f_10[i])*cy2;
941 } // end of component loop i
942 } // end of k1 loop
943 } // end of face 1
944 if (m_faceOffset[2] > -1) {
945 #pragma omp for nowait
946 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
947 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
948 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,1, m_NN[0])), numComp*sizeof(double));
949 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
950 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,1, m_NN[0])), numComp*sizeof(double));
951 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
952 for (index_t i=0; i < numComp; ++i) {
953 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx2;
954 o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy1 + (f_11[i]-f_10[i])*cy0;
955 o[INDEX3(i,0,1,numComp,2)] = (f_10[i]-f_00[i])*cx2;
956 o[INDEX3(i,1,1,numComp,2)] = (f_01[i]-f_00[i])*cy0 + (f_11[i]-f_10[i])*cy1;
957 } // end of component loop i
958 } // end of k0 loop
959 } // end of face 2
960 if (m_faceOffset[3] > -1) {
961 #pragma omp for nowait
962 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
963 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
964 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
965 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
966 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
967 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
968 for (index_t i=0; i < numComp; ++i) {
969 o[INDEX3(i,0,0,numComp,2)] = (f_11[i]-f_01[i])*cx2;
970 o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy1 + (f_11[i]-f_10[i])*cy0;
971 o[INDEX3(i,0,1,numComp,2)] = (f_11[i]-f_01[i])*cx2;
972 o[INDEX3(i,1,1,numComp,2)] = (f_01[i]-f_00[i])*cy0 + (f_11[i]-f_10[i])*cy1;
973 } // end of component loop i
974 } // end of k0 loop
975 } // end of face 3
976 } // end of parallel section
977
978 } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
979 out.requireWrite();
980 #pragma omp parallel
981 {
982 vector<double> f_00(numComp);
983 vector<double> f_01(numComp);
984 vector<double> f_10(numComp);
985 vector<double> f_11(numComp);
986 if (m_faceOffset[0] > -1) {
987 #pragma omp for nowait
988 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
989 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
990 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
991 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(1,k1, m_NN[0])), numComp*sizeof(double));
992 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(1,k1+1, m_NN[0])), numComp*sizeof(double));
993 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
994 for (index_t i=0; i < numComp; ++i) {
995 o[INDEX3(i,0,0,numComp,2)] = (f_10[i] + f_11[i] - f_00[i] - f_01[i])*cx2/2;
996 o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy2;
997 } // end of component loop i
998 } // end of k1 loop
999 } // end of face 0
1000 if (m_faceOffset[1] > -1) {
1001 #pragma omp for nowait
1002 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1003 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1, m_NN[0])), numComp*sizeof(double));
1004 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1+1, m_NN[0])), numComp*sizeof(double));
1005 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
1006 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
1007 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1008 for (index_t i=0; i < numComp; ++i) {
1009 o[INDEX3(i,0,0,numComp,2)] = (f_10[i] + f_11[i] - f_00[i] - f_01[i])*cx2/2;
1010 o[INDEX3(i,1,0,numComp,2)] = (f_11[i]-f_10[i])*cy2;
1011 } // end of component loop i
1012 } // end of k1 loop
1013 } // end of face 1
1014 if (m_faceOffset[2] > -1) {
1015 #pragma omp for nowait
1016 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1017 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
1018 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,1, m_NN[0])), numComp*sizeof(double));
1019 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
1020 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,1, m_NN[0])), numComp*sizeof(double));
1021 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1022 for (index_t i=0; i < numComp; ++i) {
1023 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx2;
1024 o[INDEX3(i,1,0,numComp,2)] = (f_01[i] + f_11[i] - f_00[i] - f_10[i])*cy2/2;
1025 } // end of component loop i
1026 } // end of k0 loop
1027 } // end of face 2
1028 if (m_faceOffset[3] > -1) {
1029 #pragma omp for nowait
1030 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1031 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
1032 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1033 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
1034 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1035 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1036 for (index_t i=0; i < numComp; ++i) {
1037 o[INDEX3(i,0,0,numComp,2)] = (f_11[i]-f_01[i])*cx2;
1038 o[INDEX3(i,1,0,numComp,2)] = (f_01[i] + f_11[i] - f_00[i] - f_10[i])*cy2/2;
1039 } // end of component loop i
1040 } // end of k0 loop
1041 } // end of face 3
1042 } // end of parallel section
1043 }
1044 }
1045
1046 //protected
1047 void Rectangle::assembleIntegrate(vector<double>& integrals, escript::Data& arg) const
1048 {
1049 const dim_t numComp = arg.getDataPointSize();
1050 const index_t left = (m_offset[0]==0 ? 0 : 1);
1051 const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1052 const int fs=arg.getFunctionSpace().getTypeCode();
1053 if (fs == Elements && arg.actsExpanded()) {
1054 #pragma omp parallel
1055 {
1056 vector<double> int_local(numComp, 0);
1057 const double w = m_dx[0]*m_dx[1]/4.;
1058 #pragma omp for nowait
1059 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1060 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1061 const double* f = arg.getSampleDataRO(INDEX2(k0, k1, m_NE[0]));
1062 for (index_t i=0; i < numComp; ++i) {
1063 const double f0 = f[INDEX2(i,0,numComp)];
1064 const double f1 = f[INDEX2(i,1,numComp)];
1065 const double f2 = f[INDEX2(i,2,numComp)];
1066 const double f3 = f[INDEX2(i,3,numComp)];
1067 int_local[i]+=(f0+f1+f2+f3)*w;
1068 } // end of component loop i
1069 } // end of k0 loop
1070 } // end of k1 loop
1071 #pragma omp critical
1072 for (index_t i=0; i<numComp; i++)
1073 integrals[i]+=int_local[i];
1074 } // end of parallel section
1075
1076 } else if (fs==ReducedElements || (fs==Elements && !arg.actsExpanded())) {
1077 const double w = m_dx[0]*m_dx[1];
1078 #pragma omp parallel
1079 {
1080 vector<double> int_local(numComp, 0);
1081 #pragma omp for nowait
1082 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1083 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1084 const double* f = arg.getSampleDataRO(INDEX2(k0, k1, m_NE[0]));
1085 for (index_t i=0; i < numComp; ++i) {
1086 int_local[i]+=f[i]*w;
1087 }
1088 }
1089 }
1090 #pragma omp critical
1091 for (index_t i=0; i<numComp; i++)
1092 integrals[i]+=int_local[i];
1093 } // end of parallel section
1094
1095 } else if (fs == FaceElements && arg.actsExpanded()) {
1096 #pragma omp parallel
1097 {
1098 vector<double> int_local(numComp, 0);
1099 const double w0 = m_dx[0]/2.;
1100 const double w1 = m_dx[1]/2.;
1101 if (m_faceOffset[0] > -1) {
1102 #pragma omp for nowait
1103 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1104 const double* f = arg.getSampleDataRO(m_faceOffset[0]+k1);
1105 for (index_t i=0; i < numComp; ++i) {
1106 const double f0 = f[INDEX2(i,0,numComp)];
1107 const double f1 = f[INDEX2(i,1,numComp)];
1108 int_local[i]+=(f0+f1)*w1;
1109 } // end of component loop i
1110 } // end of k1 loop
1111 }
1112
1113 if (m_faceOffset[1] > -1) {
1114 #pragma omp for nowait
1115 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1116 const double* f = arg.getSampleDataRO(m_faceOffset[1]+k1);
1117 for (index_t i=0; i < numComp; ++i) {
1118 const double f0 = f[INDEX2(i,0,numComp)];
1119 const double f1 = f[INDEX2(i,1,numComp)];
1120 int_local[i]+=(f0+f1)*w1;
1121 } // end of component loop i
1122 } // end of k1 loop
1123 }
1124
1125 if (m_faceOffset[2] > -1) {
1126 #pragma omp for nowait
1127 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1128 const double* f = arg.getSampleDataRO(m_faceOffset[2]+k0);
1129 for (index_t i=0; i < numComp; ++i) {
1130 const double f0 = f[INDEX2(i,0,numComp)];
1131 const double f1 = f[INDEX2(i,1,numComp)];
1132 int_local[i]+=(f0+f1)*w0;
1133 } // end of component loop i
1134 } // end of k0 loop
1135 }
1136
1137 if (m_faceOffset[3] > -1) {
1138 #pragma omp for nowait
1139 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1140 const double* f = arg.getSampleDataRO(m_faceOffset[3]+k0);
1141 for (index_t i=0; i < numComp; ++i) {
1142 const double f0 = f[INDEX2(i,0,numComp)];
1143 const double f1 = f[INDEX2(i,1,numComp)];
1144 int_local[i]+=(f0+f1)*w0;
1145 } // end of component loop i
1146 } // end of k0 loop
1147 }
1148 #pragma omp critical
1149 for (index_t i=0; i<numComp; i++)
1150 integrals[i]+=int_local[i];
1151 } // end of parallel section
1152
1153 } else if (fs==ReducedFaceElements || (fs==FaceElements && !arg.actsExpanded())) {
1154 #pragma omp parallel
1155 {
1156 vector<double> int_local(numComp, 0);
1157 if (m_faceOffset[0] > -1) {
1158 #pragma omp for nowait
1159 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1160 const double* f = arg.getSampleDataRO(m_faceOffset[0]+k1);
1161 for (index_t i=0; i < numComp; ++i) {
1162 int_local[i]+=f[i]*m_dx[1];
1163 }
1164 }
1165 }
1166
1167 if (m_faceOffset[1] > -1) {
1168 #pragma omp for nowait
1169 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1170 const double* f = arg.getSampleDataRO(m_faceOffset[1]+k1);
1171 for (index_t i=0; i < numComp; ++i) {
1172 int_local[i]+=f[i]*m_dx[1];
1173 }
1174 }
1175 }
1176
1177 if (m_faceOffset[2] > -1) {
1178 #pragma omp for nowait
1179 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1180 const double* f = arg.getSampleDataRO(m_faceOffset[2]+k0);
1181 for (index_t i=0; i < numComp; ++i) {
1182 int_local[i]+=f[i]*m_dx[0];
1183 }
1184 }
1185 }
1186
1187 if (m_faceOffset[3] > -1) {
1188 #pragma omp for nowait
1189 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1190 const double* f = arg.getSampleDataRO(m_faceOffset[3]+k0);
1191 for (index_t i=0; i < numComp; ++i) {
1192 int_local[i]+=f[i]*m_dx[0];
1193 }
1194 }
1195 }
1196
1197 #pragma omp critical
1198 for (index_t i=0; i<numComp; i++)
1199 integrals[i]+=int_local[i];
1200 } // end of parallel section
1201 } // function space selector
1202 }
1203
1204 //protected
1205 dim_t Rectangle::insertNeighbourNodes(IndexVector& index, index_t node) const
1206 {
1207 const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1208 const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1209 const int x=node%nDOF0;
1210 const int y=node/nDOF0;
1211 dim_t num=0;
1212 // loop through potential neighbours and add to index if positions are
1213 // within bounds
1214 for (int i1=-1; i1<2; i1++) {
1215 for (int i0=-1; i0<2; i0++) {
1216 // skip node itself
1217 if (i0==0 && i1==0)
1218 continue;
1219 // location of neighbour node
1220 const int nx=x+i0;
1221 const int ny=y+i1;
1222 if (nx>=0 && ny>=0 && nx<nDOF0 && ny<nDOF1) {
1223 index.push_back(ny*nDOF0+nx);
1224 num++;
1225 }
1226 }
1227 }
1228
1229 return num;
1230 }
1231
1232 //protected
1233 void Rectangle::nodesToDOF(escript::Data& out, escript::Data& in) const
1234 {
1235 const dim_t numComp = in.getDataPointSize();
1236 out.requireWrite();
1237
1238 const index_t left = (m_offset[0]==0 ? 0 : 1);
1239 const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1240 const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1241 const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1242 #pragma omp parallel for
1243 for (index_t i=0; i<nDOF1; i++) {
1244 for (index_t j=0; j<nDOF0; j++) {
1245 const index_t n=j+left+(i+bottom)*m_NN[0];
1246 const double* src=in.getSampleDataRO(n);
1247 copy(src, src+numComp, out.getSampleDataRW(j+i*nDOF0));
1248 }
1249 }
1250 }
1251
1252 //protected
1253 void Rectangle::dofToNodes(escript::Data& out, escript::Data& in) const
1254 {
1255 const dim_t numComp = in.getDataPointSize();
1256 Paso_Coupler* coupler = Paso_Coupler_alloc(m_connector, numComp);
1257 in.requireWrite();
1258 Paso_Coupler_startCollect(coupler, in.getSampleDataRW(0));
1259
1260 const dim_t numDOF = getNumDOF();
1261 out.requireWrite();
1262 const double* buffer = Paso_Coupler_finishCollect(coupler);
1263
1264 #pragma omp parallel for
1265 for (index_t i=0; i<getNumNodes(); i++) {
1266 const double* src=(m_dofMap[i]<numDOF ?
1267 in.getSampleDataRO(m_dofMap[i])
1268 : &buffer[(m_dofMap[i]-numDOF)*numComp]);
1269 copy(src, src+numComp, out.getSampleDataRW(i));
1270 }
1271 Paso_Coupler_free(coupler);
1272 }
1273
1274 //private
1275 void Rectangle::populateSampleIds()
1276 {
1277 // degrees of freedom are numbered from left to right, bottom to top in
1278 // each rank, continuing on the next rank (ranks also go left-right,
1279 // bottom-top).
1280 // This means rank 0 has id 0...n0-1, rank 1 has id n0...n1-1 etc. which
1281 // helps when writing out data rank after rank.
1282
1283 // build node distribution vector first.
1284 // rank i owns m_nodeDistribution[i+1]-nodeDistribution[i] nodes which is
1285 // constant for all ranks in this implementation
1286 m_nodeDistribution.assign(m_mpiInfo->size+1, 0);
1287 const dim_t numDOF=getNumDOF();
1288 for (dim_t k=1; k<m_mpiInfo->size; k++) {
1289 m_nodeDistribution[k]=k*numDOF;
1290 }
1291 m_nodeDistribution[m_mpiInfo->size]=getNumDataPointsGlobal();
1292 m_nodeId.resize(getNumNodes());
1293 m_dofId.resize(numDOF);
1294 m_elementId.resize(getNumElements());
1295
1296 // populate face element counts
1297 //left
1298 if (m_offset[0]==0)
1299 m_faceCount[0]=m_NE[1];
1300 else
1301 m_faceCount[0]=0;
1302 //right
1303 if (m_mpiInfo->rank%m_NX[0]==m_NX[0]-1)
1304 m_faceCount[1]=m_NE[1];
1305 else
1306 m_faceCount[1]=0;
1307 //bottom
1308 if (m_offset[1]==0)
1309 m_faceCount[2]=m_NE[0];
1310 else
1311 m_faceCount[2]=0;
1312 //top
1313 if (m_mpiInfo->rank/m_NX[0]==m_NX[1]-1)
1314 m_faceCount[3]=m_NE[0];
1315 else
1316 m_faceCount[3]=0;
1317
1318 m_faceId.resize(getNumFaceElements());
1319
1320 const index_t left = (m_offset[0]==0 ? 0 : 1);
1321 const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1322 const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1323 const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1324
1325 #define globalNodeId(x,y) \
1326 ((m_offset[0]+x)/nDOF0)*nDOF0*nDOF1+(m_offset[0]+x)%nDOF0 \
1327 + ((m_offset[1]+y)/nDOF1)*nDOF0*nDOF1*m_NX[0]+((m_offset[1]+y)%nDOF1)*nDOF0
1328
1329 // set corner id's outside the parallel region
1330 m_nodeId[0] = globalNodeId(0, 0);
1331 m_nodeId[m_NN[0]-1] = globalNodeId(m_NN[0]-1, 0);
1332 m_nodeId[m_NN[0]*(m_NN[1]-1)] = globalNodeId(0, m_NN[1]-1);
1333 m_nodeId[m_NN[0]*m_NN[1]-1] = globalNodeId(m_NN[0]-1,m_NN[1]-1);
1334 #undef globalNodeId
1335
1336 #pragma omp parallel
1337 {
1338 // populate degrees of freedom and own nodes (identical id)
1339 #pragma omp for nowait
1340 for (dim_t i=0; i<nDOF1; i++) {
1341 for (dim_t j=0; j<nDOF0; j++) {
1342 const index_t nodeIdx=j+left+(i+bottom)*m_NN[0];
1343 const index_t dofIdx=j+i*nDOF0;
1344 m_dofId[dofIdx] = m_nodeId[nodeIdx]
1345 = m_nodeDistribution[m_mpiInfo->rank]+dofIdx;
1346 }
1347 }
1348
1349 // populate the rest of the nodes (shared with other ranks)
1350 if (m_faceCount[0]==0) { // left column
1351 #pragma omp for nowait
1352 for (dim_t i=0; i<nDOF1; i++) {
1353 const index_t nodeIdx=(i+bottom)*m_NN[0];
1354 const index_t dofId=(i+1)*nDOF0-1;
1355 m_nodeId[nodeIdx]
1356 = m_nodeDistribution[m_mpiInfo->rank-1]+dofId;
1357 }
1358 }
1359 if (m_faceCount[1]==0) { // right column
1360 #pragma omp for nowait
1361 for (dim_t i=0; i<nDOF1; i++) {
1362 const index_t nodeIdx=(i+bottom+1)*m_NN[0]-1;
1363 const index_t dofId=i*nDOF0;
1364 m_nodeId[nodeIdx]
1365 = m_nodeDistribution[m_mpiInfo->rank+1]+dofId;
1366 }
1367 }
1368 if (m_faceCount[2]==0) { // bottom row
1369 #pragma omp for nowait
1370 for (dim_t i=0; i<nDOF0; i++) {
1371 const index_t nodeIdx=i+left;
1372 const index_t dofId=nDOF0*(nDOF1-1)+i;
1373 m_nodeId[nodeIdx]
1374 = m_nodeDistribution[m_mpiInfo->rank-m_NX[0]]+dofId;
1375 }
1376 }
1377 if (m_faceCount[3]==0) { // top row
1378 #pragma omp for nowait
1379 for (dim_t i=0; i<nDOF0; i++) {
1380 const index_t nodeIdx=m_NN[0]*(m_NN[1]-1)+i+left;
1381 const index_t dofId=i;
1382 m_nodeId[nodeIdx]
1383 = m_nodeDistribution[m_mpiInfo->rank+m_NX[0]]+dofId;
1384 }
1385 }
1386
1387 // populate element id's
1388 #pragma omp for nowait
1389 for (dim_t i1=0; i1<m_NE[1]; i1++) {
1390 for (dim_t i0=0; i0<m_NE[0]; i0++) {
1391 m_elementId[i0+i1*m_NE[0]]=(m_offset[1]+i1)*m_gNE[0]+m_offset[0]+i0;
1392 }
1393 }
1394
1395 // face elements
1396 #pragma omp for
1397 for (dim_t k=0; k<getNumFaceElements(); k++)
1398 m_faceId[k]=k;
1399 } // end parallel section
1400
1401 m_nodeTags.assign(getNumNodes(), 0);
1402 updateTagsInUse(Nodes);
1403
1404 m_elementTags.assign(getNumElements(), 0);
1405 updateTagsInUse(Elements);
1406
1407 // generate face offset vector and set face tags
1408 const index_t LEFT=1, RIGHT=2, BOTTOM=10, TOP=20;
1409 const index_t faceTag[] = { LEFT, RIGHT, BOTTOM, TOP };
1410 m_faceOffset.assign(4, -1);
1411 m_faceTags.clear();
1412 index_t offset=0;
1413 for (size_t i=0; i<4; i++) {
1414 if (m_faceCount[i]>0) {
1415 m_faceOffset[i]=offset;
1416 offset+=m_faceCount[i];
1417 m_faceTags.insert(m_faceTags.end(), m_faceCount[i], faceTag[i]);
1418 }
1419 }
1420 setTagMap("left", LEFT);
1421 setTagMap("right", RIGHT);
1422 setTagMap("bottom", BOTTOM);
1423 setTagMap("top", TOP);
1424 updateTagsInUse(FaceElements);
1425 }
1426
1427 //private
1428 void Rectangle::createPattern()
1429 {
1430 const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1431 const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1432 const index_t left = (m_offset[0]==0 ? 0 : 1);
1433 const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1434
1435 // populate node->DOF mapping with own degrees of freedom.
1436 // The rest is assigned in the loop further down
1437 m_dofMap.assign(getNumNodes(), 0);
1438 #pragma omp parallel for
1439 for (index_t i=bottom; i<bottom+nDOF1; i++) {
1440 for (index_t j=left; j<left+nDOF0; j++) {
1441 m_dofMap[i*m_NN[0]+j]=(i-bottom)*nDOF0+j-left;
1442 }
1443 }
1444
1445 // build list of shared components and neighbours by looping through
1446 // all potential neighbouring ranks and checking if positions are
1447 // within bounds
1448 const dim_t numDOF=nDOF0*nDOF1;
1449 vector<IndexVector> colIndices(numDOF); // for the couple blocks
1450 RankVector neighbour;
1451 IndexVector offsetInShared(1,0);
1452 IndexVector sendShared, recvShared;
1453 int numShared=0;
1454 const int x=m_mpiInfo->rank%m_NX[0];
1455 const int y=m_mpiInfo->rank/m_NX[0];
1456 for (int i1=-1; i1<2; i1++) {
1457 for (int i0=-1; i0<2; i0++) {
1458 // skip this rank
1459 if (i0==0 && i1==0)
1460 continue;
1461 // location of neighbour rank
1462 const int nx=x+i0;
1463 const int ny=y+i1;
1464 if (nx>=0 && ny>=0 && nx<m_NX[0] && ny<m_NX[1]) {
1465 neighbour.push_back(ny*m_NX[0]+nx);
1466 if (i0==0) {
1467 // sharing top or bottom edge
1468 const int firstDOF=(i1==-1 ? 0 : numDOF-nDOF0);
1469 const int firstNode=(i1==-1 ? left : m_NN[0]*(m_NN[1]-1)+left);
1470 offsetInShared.push_back(offsetInShared.back()+nDOF0);
1471 for (dim_t i=0; i<nDOF0; i++, numShared++) {
1472 sendShared.push_back(firstDOF+i);
1473 recvShared.push_back(numDOF+numShared);
1474 if (i>0)
1475 colIndices[firstDOF+i-1].push_back(numShared);
1476 colIndices[firstDOF+i].push_back(numShared);
1477 if (i<nDOF0-1)
1478 colIndices[firstDOF+i+1].push_back(numShared);
1479 m_dofMap[firstNode+i]=numDOF+numShared;
1480 }
1481 } else if (i1==0) {
1482 // sharing left or right edge
1483 const int firstDOF=(i0==-1 ? 0 : nDOF0-1);
1484 const int firstNode=(i0==-1 ? bottom*m_NN[0] : (bottom+1)*m_NN[0]-1);
1485 offsetInShared.push_back(offsetInShared.back()+nDOF1);
1486 for (dim_t i=0; i<nDOF1; i++, numShared++) {
1487 sendShared.push_back(firstDOF+i*nDOF0);
1488 recvShared.push_back(numDOF+numShared);
1489 if (i>0)
1490 colIndices[firstDOF+(i-1)*nDOF0].push_back(numShared);
1491 colIndices[firstDOF+i*nDOF0].push_back(numShared);
1492 if (i<nDOF1-1)
1493 colIndices[firstDOF+(i+1)*nDOF0].push_back(numShared);
1494 m_dofMap[firstNode+i*m_NN[0]]=numDOF+numShared;
1495 }
1496 } else {
1497 // sharing a node
1498 const int dof=(i0+1)/2*(nDOF0-1)+(i1+1)/2*(numDOF-nDOF0);
1499 const int node=(i0+1)/2*(m_NN[0]-1)+(i1+1)/2*m_NN[0]*(m_NN[1]-1);
1500 offsetInShared.push_back(offsetInShared.back()+1);
1501 sendShared.push_back(dof);
1502 recvShared.push_back(numDOF+numShared);
1503 colIndices[dof].push_back(numShared);
1504 m_dofMap[node]=numDOF+numShared;
1505 ++numShared;
1506 }
1507 }
1508 }
1509 }
1510
1511 // create connector
1512 Paso_SharedComponents *snd_shcomp = Paso_SharedComponents_alloc(
1513 numDOF, neighbour.size(), &neighbour[0], &sendShared[0],
1514 &offsetInShared[0], 1, 0, m_mpiInfo);
1515 Paso_SharedComponents *rcv_shcomp = Paso_SharedComponents_alloc(
1516 numDOF, neighbour.size(), &neighbour[0], &recvShared[0],
1517 &offsetInShared[0], 1, 0, m_mpiInfo);
1518 m_connector = Paso_Connector_alloc(snd_shcomp, rcv_shcomp);
1519 Paso_SharedComponents_free(snd_shcomp);
1520 Paso_SharedComponents_free(rcv_shcomp);
1521
1522 // create main and couple blocks
1523 Paso_Pattern *mainPattern = createMainPattern();
1524 Paso_Pattern *colPattern, *rowPattern;
1525 createCouplePatterns(colIndices, numShared, &colPattern, &rowPattern);
1526
1527 // allocate paso distribution
1528 Paso_Distribution* distribution = Paso_Distribution_alloc(m_mpiInfo,
1529 const_cast<index_t*>(&m_nodeDistribution[0]), 1, 0);
1530
1531 // finally create the system matrix
1532 m_pattern = Paso_SystemMatrixPattern_alloc(MATRIX_FORMAT_DEFAULT,
1533 distribution, distribution, mainPattern, colPattern, rowPattern,
1534 m_connector, m_connector);
1535
1536 Paso_Distribution_free(distribution);
1537
1538 // useful debug output
1539 /*
1540 cout << "--- rcv_shcomp ---" << endl;
1541 cout << "numDOF=" << numDOF << ", numNeighbors=" << neighbour.size() << endl;
1542 for (size_t i=0; i<neighbour.size(); i++) {
1543 cout << "neighbor[" << i << "]=" << neighbour[i]
1544 << " offsetInShared[" << i+1 << "]=" << offsetInShared[i+1] << endl;
1545 }
1546 for (size_t i=0; i<recvShared.size(); i++) {
1547 cout << "shared[" << i << "]=" << recvShared[i] << endl;
1548 }
1549 cout << "--- snd_shcomp ---" << endl;
1550 for (size_t i=0; i<sendShared.size(); i++) {
1551 cout << "shared[" << i << "]=" << sendShared[i] << endl;
1552 }
1553 cout << "--- dofMap ---" << endl;
1554 for (size_t i=0; i<m_dofMap.size(); i++) {
1555 cout << "m_dofMap[" << i << "]=" << m_dofMap[i] << endl;
1556 }
1557 cout << "--- colIndices ---" << endl;
1558 for (size_t i=0; i<colIndices.size(); i++) {
1559 cout << "colIndices[" << i << "].size()=" << colIndices[i].size() << endl;
1560 }
1561 */
1562
1563 /*
1564 cout << "--- main_pattern ---" << endl;
1565 cout << "M=" << mainPattern->numOutput << ", N=" << mainPattern->numInput << endl;
1566 for (size_t i=0; i<mainPattern->numOutput+1; i++) {
1567 cout << "ptr[" << i << "]=" << mainPattern->ptr[i] << endl;
1568 }
1569 for (size_t i=0; i<mainPattern->ptr[mainPattern->numOutput]; i++) {
1570 cout << "index[" << i << "]=" << mainPattern->index[i] << endl;
1571 }
1572 */
1573
1574 /*
1575 cout << "--- colCouple_pattern ---" << endl;
1576 cout << "M=" << colPattern->numOutput << ", N=" << colPattern->numInput << endl;
1577 for (size_t i=0; i<colPattern->numOutput+1; i++) {
1578 cout << "ptr[" << i << "]=" << colPattern->ptr[i] << endl;
1579 }
1580 for (size_t i=0; i<colPattern->ptr[colPattern->numOutput]; i++) {
1581 cout << "index[" << i << "]=" << colPattern->index[i] << endl;
1582 }
1583 */
1584
1585 /*
1586 cout << "--- rowCouple_pattern ---" << endl;
1587 cout << "M=" << rowPattern->numOutput << ", N=" << rowPattern->numInput << endl;
1588 for (size_t i=0; i<rowPattern->numOutput+1; i++) {
1589 cout << "ptr[" << i << "]=" << rowPattern->ptr[i] << endl;
1590 }
1591 for (size_t i=0; i<rowPattern->ptr[rowPattern->numOutput]; i++) {
1592 cout << "index[" << i << "]=" << rowPattern->index[i] << endl;
1593 }
1594 */
1595
1596 Paso_Pattern_free(mainPattern);
1597 Paso_Pattern_free(colPattern);
1598 Paso_Pattern_free(rowPattern);
1599 }
1600
1601 //private
1602 void Rectangle::addToMatrixAndRHS(Paso_SystemMatrix* S, escript::Data& F,
1603 const vector<double>& EM_S, const vector<double>& EM_F, bool addS,
1604 bool addF, index_t firstNode, dim_t nEq, dim_t nComp) const
1605 {
1606 IndexVector rowIndex;
1607 rowIndex.push_back(m_dofMap[firstNode]);
1608 rowIndex.push_back(m_dofMap[firstNode+1]);
1609 rowIndex.push_back(m_dofMap[firstNode+m_NN[0]]);
1610 rowIndex.push_back(m_dofMap[firstNode+m_NN[0]+1]);
1611 if (addF) {
1612 double *F_p=F.getSampleDataRW(0);
1613 for (index_t i=0; i<rowIndex.size(); i++) {
1614 if (rowIndex[i]<getNumDOF()) {
1615 for (index_t eq=0; eq<nEq; eq++) {
1616 F_p[INDEX2(eq, rowIndex[i], nEq)]+=EM_F[INDEX2(eq,i,nEq)];
1617 }
1618 }
1619 }
1620 }
1621 if (addS) {
1622 addToSystemMatrix(S, rowIndex, nEq, rowIndex, nComp, EM_S);
1623 }
1624 }
1625
1626 //protected
1627 void Rectangle::interpolateNodesOnElements(escript::Data& out,
1628 escript::Data& in, bool reduced) const
1629 {
1630 const dim_t numComp = in.getDataPointSize();
1631 if (reduced) {
1632 out.requireWrite();
1633 const double c0 = 0.25;
1634 #pragma omp parallel
1635 {
1636 vector<double> f_00(numComp);
1637 vector<double> f_01(numComp);
1638 vector<double> f_10(numComp);
1639 vector<double> f_11(numComp);
1640 #pragma omp for
1641 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1642 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1643 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
1644 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
1645 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
1646 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
1647 double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
1648 for (index_t i=0; i < numComp; ++i) {
1649 o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i] + f_10[i] + f_11[i]);
1650 } /* end of component loop i */
1651 } /* end of k0 loop */
1652 } /* end of k1 loop */
1653 } /* end of parallel section */
1654 } else {
1655 out.requireWrite();
1656 const double c0 = 0.16666666666666666667;
1657 const double c1 = 0.044658198738520451079;
1658 const double c2 = 0.62200846792814621559;
1659 #pragma omp parallel
1660 {
1661 vector<double> f_00(numComp);
1662 vector<double> f_01(numComp);
1663 vector<double> f_10(numComp);
1664 vector<double> f_11(numComp);
1665 #pragma omp for
1666 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1667 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1668 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
1669 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
1670 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
1671 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
1672 double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
1673 for (index_t i=0; i < numComp; ++i) {
1674 o[INDEX2(i,numComp,0)] = c0*(f_01[i] + f_10[i]) + c1*f_11[i] + c2*f_00[i];
1675 o[INDEX2(i,numComp,1)] = c0*(f_00[i] + f_11[i]) + c1*f_01[i] + c2*f_10[i];
1676 o[INDEX2(i,numComp,2)] = c0*(f_00[i] + f_11[i]) + c1*f_10[i] + c2*f_01[i];
1677 o[INDEX2(i,numComp,3)] = c0*(f_01[i] + f_10[i]) + c1*f_00[i] + c2*f_11[i];
1678 } /* end of component loop i */
1679 } /* end of k0 loop */
1680 } /* end of k1 loop */
1681 } /* end of parallel section */
1682 }
1683 }
1684
1685 //protected
1686 void Rectangle::interpolateNodesOnFaces(escript::Data& out, escript::Data& in,
1687 bool reduced) const
1688 {
1689 const dim_t numComp = in.getDataPointSize();
1690 if (reduced) {
1691 out.requireWrite();
1692 #pragma omp parallel
1693 {
1694 vector<double> f_00(numComp);
1695 vector<double> f_01(numComp);
1696 vector<double> f_10(numComp);
1697 vector<double> f_11(numComp);
1698 if (m_faceOffset[0] > -1) {
1699 #pragma omp for nowait
1700 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1701 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
1702 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
1703 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1704 for (index_t i=0; i < numComp; ++i) {
1705 o[INDEX2(i,numComp,0)] = (f_00[i] + f_01[i])/2;
1706 } /* end of component loop i */
1707 } /* end of k1 loop */
1708 } /* end of face 0 */
1709 if (m_faceOffset[1] > -1) {
1710 #pragma omp for nowait
1711 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1712 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
1713 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
1714 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1715 for (index_t i=0; i < numComp; ++i) {
1716 o[INDEX2(i,numComp,0)] = (f_10[i] + f_11[i])/2;
1717 } /* end of component loop i */
1718 } /* end of k1 loop */
1719 } /* end of face 1 */
1720 if (m_faceOffset[2] > -1) {
1721 #pragma omp for nowait
1722 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1723 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
1724 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
1725 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1726 for (index_t i=0; i < numComp; ++i) {
1727 o[INDEX2(i,numComp,0)] = (f_00[i] + f_10[i])/2;
1728 } /* end of component loop i */
1729 } /* end of k0 loop */
1730 } /* end of face 2 */
1731 if (m_faceOffset[3] > -1) {
1732 #pragma omp for nowait
1733 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1734 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1735 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1736 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1737 for (index_t i=0; i < numComp; ++i) {
1738 o[INDEX2(i,numComp,0)] = (f_01[i] + f_11[i])/2;
1739 } /* end of component loop i */
1740 } /* end of k0 loop */
1741 } /* end of face 3 */
1742 } /* end of parallel section */
1743 } else {
1744 out.requireWrite();
1745 const double c0 = 0.21132486540518711775;
1746 const double c1 = 0.78867513459481288225;
1747 #pragma omp parallel
1748 {
1749 vector<double> f_00(numComp);
1750 vector<double> f_01(numComp);
1751 vector<double> f_10(numComp);
1752 vector<double> f_11(numComp);
1753 if (m_faceOffset[0] > -1) {
1754 #pragma omp for nowait
1755 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1756 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
1757 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
1758 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1759 for (index_t i=0; i < numComp; ++i) {
1760 o[INDEX2(i,numComp,0)] = c0*f_01[i] + c1*f_00[i];
1761 o[INDEX2(i,numComp,1)] = c0*f_00[i] + c1*f_01[i];
1762 } /* end of component loop i */
1763 } /* end of k1 loop */
1764 } /* end of face 0 */
1765 if (m_faceOffset[1] > -1) {
1766 #pragma omp for nowait
1767 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1768 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
1769 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
1770 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1771 for (index_t i=0; i < numComp; ++i) {
1772 o[INDEX2(i,numComp,0)] = c1*f_10[i] + c0*f_11[i];
1773 o[INDEX2(i,numComp,1)] = c1*f_11[i] + c0*f_10[i];
1774 } /* end of component loop i */
1775 } /* end of k1 loop */
1776 } /* end of face 1 */
1777 if (m_faceOffset[2] > -1) {
1778 #pragma omp for nowait
1779 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1780 memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
1781 memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
1782 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1783 for (index_t i=0; i < numComp; ++i) {
1784 o[INDEX2(i,numComp,0)] = c0*f_10[i] + c1*f_00[i];
1785 o[INDEX2(i,numComp,1)] = c0*f_00[i] + c1*f_10[i];
1786 } /* end of component loop i */
1787 } /* end of k0 loop */
1788 } /* end of face 2 */
1789 if (m_faceOffset[3] > -1) {
1790 #pragma omp for nowait
1791 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1792 memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1793 memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1794 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1795 for (index_t i=0; i < numComp; ++i) {
1796 o[INDEX2(i,numComp,0)] = c0*f_11[i] + c1*f_01[i];
1797 o[INDEX2(i,numComp,1)] = c0*f_01[i] + c1*f_11[i];
1798 } /* end of component loop i */
1799 } /* end of k0 loop */
1800 } /* end of face 3 */
1801 } /* end of parallel section */
1802 }
1803 }
1804
1805 //protected
1806 void Rectangle::assemblePDESingle(Paso_SystemMatrix* mat,
1807 escript::Data& rhs, const escript::Data& A, const escript::Data& B,
1808 const escript::Data& C, const escript::Data& D,
1809 const escript::Data& X, const escript::Data& Y) const
1810 {
1811 const double SQRT3 = 1.73205080756887719318;
1812 const double w1 = 1.0/24.0;
1813 const double w5 = -SQRT3/24 + 1.0/12;
1814 const double w2 = -SQRT3/24 - 1.0/12;
1815 const double w19 = -m_dx[0]/12;
1816 const double w11 = w19*(SQRT3 + 3)/12;
1817 const double w14 = w19*(-SQRT3 + 3)/12;
1818 const double w16 = w19*(5*SQRT3 + 9)/12;
1819 const double w17 = w19*(-5*SQRT3 + 9)/12;
1820 const double w27 = w19*(-SQRT3 - 3)/2;
1821 const double w28 = w19*(SQRT3 - 3)/2;
1822 const double w18 = -m_dx[1]/12;
1823 const double w12 = w18*(5*SQRT3 + 9)/12;
1824 const double w13 = w18*(-5*SQRT3 + 9)/12;
1825 const double w10 = w18*(SQRT3 + 3)/12;
1826 const double w15 = w18*(-SQRT3 + 3)/12;
1827 const double w25 = w18*(-SQRT3 - 3)/2;
1828 const double w26 = w18*(SQRT3 - 3)/2;
1829 const double w22 = m_dx[0]*m_dx[1]/144;
1830 const double w20 = w22*(SQRT3 + 2);
1831 const double w21 = w22*(-SQRT3 + 2);
1832 const double w23 = w22*(4*SQRT3 + 7);
1833 const double w24 = w22*(-4*SQRT3 + 7);
1834 const double w3 = m_dx[0]/(24*m_dx[1]);
1835 const double w7 = w3*(SQRT3 + 2);
1836 const double w8 = w3*(-SQRT3 + 2);
1837 const double w6 = -m_dx[1]/(24*m_dx[0]);
1838 const double w0 = w6*(SQRT3 + 2);
1839 const double w4 = w6*(-SQRT3 + 2);
1840
1841 rhs.requireWrite();
1842 #pragma omp parallel
1843 {
1844 for (index_t k1_0=0; k1_0<2; k1_0++) { // colouring
1845 #pragma omp for
1846 for (index_t k1=k1_0; k1<m_NE[1]; k1+=2) {
1847 for (index_t k0=0; k0<m_NE[0]; ++k0) {
1848 bool add_EM_S=false;
1849 bool add_EM_F=false;
1850 vector<double> EM_S(4*4, 0);
1851 vector<double> EM_F(4, 0);
1852 const index_t e = k0 + m_NE[0]*k1;
1853 ///////////////
1854 // process A //
1855 ///////////////
1856 if (!A.isEmpty()) {
1857 add_EM_S = true;
1858 const double* A_p = const_cast<escript::Data*>(&A)->getSampleDataRO(e);
1859 if (A.actsExpanded()) {
1860 const double A_00_0 = A_p[INDEX3(0,0,0,2,2)];
1861 const double A_01_0 = A_p[INDEX3(0,1,0,2,2)];
1862 const double A_10_0 = A_p[INDEX3(1,0,0,2,2)];
1863 const double A_11_0 = A_p[INDEX3(1,1,0,2,2)];
1864 const double A_00_1 = A_p[INDEX3(0,0,1,2,2)];
1865 const double A_01_1 = A_p[INDEX3(0,1,1,2,2)];
1866 const double A_10_1 = A_p[INDEX3(1,0,1,2,2)];
1867 const double A_11_1 = A_p[INDEX3(1,1,1,2,2)];
1868 const double A_00_2 = A_p[INDEX3(0,0,2,2,2)];
1869 const double A_01_2 = A_p[INDEX3(0,1,2,2,2)];
1870 const double A_10_2 = A_p[INDEX3(1,0,2,2,2)];
1871 const double A_11_2 = A_p[INDEX3(1,1,2,2,2)];
1872 const double A_00_3 = A_p[INDEX3(0,0,3,2,2)];
1873 const double A_01_3 = A_p[INDEX3(0,1,3,2,2)];
1874 const double A_10_3 = A_p[INDEX3(1,0,3,2,2)];
1875 const double A_11_3 = A_p[INDEX3(1,1,3,2,2)];
1876 const double tmp0 = w3*(A_11_0 + A_11_1 + A_11_2 + A_11_3);
1877 const double tmp1 = w1*(A_01_0 + A_01_3 - A_10_1 - A_10_2);
1878 const double tmp2 = w4*(A_00_2 + A_00_3);
1879 const double tmp3 = w0*(A_00_0 + A_00_1);
1880 const double tmp4 = w5*(A_01_2 - A_10_3);
1881 const double tmp5 = w2*(-A_01_1 + A_10_0);
1882 const double tmp6 = w5*(A_01_3 + A_10_0);
1883 const double tmp7 = w3*(-A_11_0 - A_11_1 - A_11_2 - A_11_3);
1884 const double tmp8 = w6*(A_00_0 + A_00_1 + A_00_2 + A_00_3);
1885 const double tmp9 = w1*(A_01_1 + A_01_2 + A_10_1 + A_10_2);
1886 const double tmp10 = w2*(-A_01_0 - A_10_3);
1887 const double tmp11 = w4*(A_00_0 + A_00_1);
1888 const double tmp12 = w0*(A_00_2 + A_00_3);
1889 const double tmp13 = w5*(A_01_1 - A_10_0);
1890 const double tmp14 = w2*(-A_01_2 + A_10_3);
1891 const double tmp15 = w7*(A_11_0 + A_11_2);
1892 const double tmp16 = w4*(-A_00_2 - A_00_3);
1893 const double tmp17 = w0*(-A_00_0 - A_00_1);
1894 const double tmp18 = w5*(A_01_3 + A_10_3);
1895 const double tmp19 = w8*(A_11_1 + A_11_3);
1896 const double tmp20 = w2*(-A_01_0 - A_10_0);
1897 const double tmp21 = w7*(A_11_1 + A_11_3);
1898 const double tmp22 = w4*(-A_00_0 - A_00_1);
1899 const double tmp23 = w0*(-A_00_2 - A_00_3);
1900 const double tmp24 = w5*(A_01_0 + A_10_0);
1901 const double tmp25 = w8*(A_11_0 + A_11_2);
1902 const double tmp26 = w2*(-A_01_3 - A_10_3);
1903 const double tmp27 = w5*(-A_01_1 - A_10_2);
1904 const double tmp28 = w1*(-A_01_0 - A_01_3 - A_10_0 - A_10_3);
1905 const double tmp29 = w2*(A_01_2 + A_10_1);
1906 const double tmp30 = w7*(-A_11_1 - A_11_3);
1907 const double tmp31 = w1*(-A_01_1 - A_01_2 + A_10_0 + A_10_3);
1908 const double tmp32 = w5*(-A_01_0 + A_10_2);
1909 const double tmp33 = w8*(-A_11_0 - A_11_2);
1910 const double tmp34 = w6*(-A_00_0 - A_00_1 - A_00_2 - A_00_3);
1911 const double tmp35 = w2*(A_01_3 - A_10_1);
1912 const double tmp36 = w5*(A_01_0 + A_10_3);
1913 const double tmp37 = w2*(-A_01_3 - A_10_0);
1914 const double tmp38 = w7*(-A_11_0 - A_11_2);
1915 const double tmp39 = w5*(-A_01_3 + A_10_1);
1916 const double tmp40 = w8*(-A_11_1 - A_11_3);
1917 const double tmp41 = w2*(A_01_0 - A_10_2);
1918 const double tmp42 = w5*(A_01_1 - A_10_3);
1919 const double tmp43 = w2*(-A_01_2 + A_10_0);
1920 const double tmp44 = w5*(A_01_2 - A_10_0);
1921 const double tmp45 = w2*(-A_01_1 + A_10_3);
1922 const double tmp46 = w5*(-A_01_0 + A_10_1);
1923 const double tmp47 = w2*(A_01_3 - A_10_2);
1924 const double tmp48 = w5*(-A_01_1 - A_10_1);
1925 const double tmp49 = w2*(A_01_2 + A_10_2);
1926 const double tmp50 = w5*(-A_01_3 + A_10_2);
1927 const double tmp51 = w2*(A_01_0 - A_10_1);
1928 const double tmp52 = w5*(-A_01_2 - A_10_1);
1929 const double tmp53 = w2*(A_01_1 + A_10_2);
1930 const double tmp54 = w5*(-A_01_2 - A_10_2);
1931 const double tmp55 = w2*(A_01_1 + A_10_1);
1932 EM_S[INDEX2(0,0,4)]+=tmp15 + tmp16 + tmp17 + tmp18 + tmp19 + tmp20 + tmp9;
1933 EM_S[INDEX2(0,1,4)]+=tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5;
1934 EM_S[INDEX2(0,2,4)]+=tmp31 + tmp34 + tmp38 + tmp39 + tmp40 + tmp41;
1935 EM_S[INDEX2(0,3,4)]+=tmp28 + tmp52 + tmp53 + tmp7 + tmp8;
1936 EM_S[INDEX2(1,0,4)]+=tmp0 + tmp2 + tmp3 + tmp31 + tmp50 + tmp51;
1937 EM_S[INDEX2(1,1,4)]+=tmp16 + tmp17 + tmp21 + tmp25 + tmp28 + tmp54 + tmp55;
1938 EM_S[INDEX2(1,2,4)]+=tmp10 + tmp6 + tmp7 + tmp8 + tmp9;
1939 EM_S[INDEX2(1,3,4)]+=tmp1 + tmp30 + tmp33 + tmp34 + tmp44 + tmp45;
1940 EM_S[INDEX2(2,0,4)]+=tmp1 + tmp34 + tmp38 + tmp40 + tmp42 + tmp43;
1941 EM_S[INDEX2(2,1,4)]+=tmp36 + tmp37 + tmp7 + tmp8 + tmp9;
1942 EM_S[INDEX2(2,2,4)]+=tmp15 + tmp19 + tmp22 + tmp23 + tmp28 + tmp48 + tmp49;
1943 EM_S[INDEX2(2,3,4)]+=tmp0 + tmp11 + tmp12 + tmp31 + tmp46 + tmp47;
1944 EM_S[INDEX2(3,0,4)]+=tmp27 + tmp28 + tmp29 + tmp7 + tmp8;
1945 EM_S[INDEX2(3,1,4)]+=tmp30 + tmp31 + tmp32 + tmp33 + tmp34 + tmp35;
1946 EM_S[INDEX2(3,2,4)]+=tmp0 + tmp1 + tmp11 + tmp12 + tmp13 + tmp14;
1947 EM_S[INDEX2(3,3,4)]+=tmp21 + tmp22 + tmp23 + tmp24 + tmp25 + tmp26 + tmp9;
1948 } else { // constant data
1949 const double A_00 = A_p[INDEX2(0,0,2)];
1950 const double A_01 = A_p[INDEX2(0,1,2)];
1951 const double A_10 = A_p[INDEX2(1,0,2)];
1952 const double A_11 = A_p[INDEX2(1,1,2)];
1953 const double tmp0 = 6*w1*(A_01 - A_10);
1954 const double tmp1 = 6*w1*(A_01 + A_10);
1955 const double tmp2 = 6*w1*(-A_01 - A_10);
1956 const double tmp3 = 6*w1*(-A_01 + A_10);
1957 EM_S[INDEX2(0,0,4)]+=-8*A_00*w6 + 8*A_11*w3 + tmp1;
1958 EM_S[INDEX2(0,1,4)]+=8*A_00*w6 + 4*A_11*w3 + tmp0;
1959 EM_S[INDEX2(0,2,4)]+=-4*A_00*w6 - 8*A_11*w3 + tmp3;
1960 EM_S[INDEX2(0,3,4)]+=4*A_00*w6 - 4*A_11*w3 + tmp2;
1961 EM_S[INDEX2(1,0,4)]+=8*A_00*w6 + 4*A_11*w3 + tmp3;
1962 EM_S[INDEX2(1,1,4)]+=-8*A_00*w6 + 8*A_11*w3 + tmp2;
1963 EM_S[INDEX2(1,2,4)]+=4*A_00*w6 - 4*A_11*w3 + tmp1;
1964 EM_S[INDEX2(1,3,4)]+=-4*A_00*w6 - 8*A_11*w3 + tmp0;
1965 EM_S[INDEX2(2,0,4)]+=-4*A_00*w6 - 8*A_11*w3 + tmp0;
1966 EM_S[INDEX2(2,1,4)]+=4*A_00*w6 - 4*A_11*w3 + tmp1;
1967 EM_S[INDEX2(2,2,4)]+=-8*A_00*w6 + 8*A_11*w3 + tmp2;
1968 EM_S[INDEX2(2,3,4)]+=8*A_00*w6 + 4*A_11*w3 + tmp3;
1969 EM_S[INDEX2(3,0,4)]+=4*A_00*w6 - 4*A_11*w3 + tmp2;
1970 EM_S[INDEX2(3,1,4)]+=-4*A_00*w6 - 8*A_11*w3 + tmp3;
1971 EM_S[INDEX2(3,2,4)]+=8*A_00*w6 + 4*A_11*w3 + tmp0;
1972 EM_S[INDEX2(3,3,4)]+=-8*A_00*w6 + 8*A_11*w3 + tmp1;
1973 }
1974 }
1975 ///////////////
1976 // process B //
1977 ///////////////
1978 if (!B.isEmpty()) {
1979 add_EM_S=true;
1980 const double* B_p=const_cast<escript::Data*>(&B)->getSampleDataRO(e);
1981 if (B.actsExpanded()) {
1982 const double B_0_0 = B_p[INDEX2(0,0,2)];
1983 const double B_1_0 = B_p[INDEX2(1,0,2)];
1984 const double B_0_1 = B_p[INDEX2(0,1,2)];
1985 const double B_1_1 = B_p[INDEX2(1,1,2)];
1986 const double B_0_2 = B_p[INDEX2(0,2,2)];
1987 const double B_1_2 = B_p[INDEX2(1,2,2)];
1988 const double B_0_3 = B_p[INDEX2(0,3,2)];
1989 const double B_1_3 = B_p[INDEX2(1,3,2)];
1990 const double tmp0 = w11*(B_1_0 + B_1_1);
1991 const double tmp1 = w14*(B_1_2 + B_1_3);
1992 const double tmp2 = w15*(-B_0_1 - B_0_3);
1993 const double tmp3 = w10*(-B_0_0 - B_0_2);
1994 const double tmp4 = w11*(B_1_2 + B_1_3);
1995 const double tmp5 = w14*(B_1_0 + B_1_1);
1996 const double tmp6 = w11*(-B_1_2 - B_1_3);
1997 const double tmp7 = w14*(-B_1_0 - B_1_1);
1998 const double tmp8 = w11*(-B_1_0 - B_1_1);
1999 const double tmp9 = w14*(-B_1_2 - B_1_3);
2000 const double tmp10 = w10*(-B_0_1 - B_0_3);
2001 const double tmp11 = w15*(-B_0_0 - B_0_2);
2002 const double tmp12 = w15*(B_0_0 + B_0_2);
2003 const double tmp13 = w10*(B_0_1 + B_0_3);
2004 const double tmp14 = w10*(B_0_0 + B_0_2);
2005 const double tmp15 = w15*(B_0_1 + B_0_3);
2006 EM_S[INDEX2(0,0,4)]+=B_0_0*w12 + B_0_1*w10 + B_0_2*w15 + B_0_3*w13 + B_1_0*w16 + B_1_1*w14 + B_1_2*w11 + B_1_3*w17;
2007 EM_S[INDEX2(0,1,4)]+=B_0_0*w10 + B_0_1*w12 + B_0_2*w13 + B_0_3*w15 + tmp0 + tmp1;
2008 EM_S[INDEX2(0,2,4)]+=B_1_0*w11 + B_1_1*w17 + B_1_2*w16 + B_1_3*w14 + tmp14 + tmp15;
2009 EM_S[INDEX2(0,3,4)]+=tmp12 + tmp13 + tmp4 + tmp5;
2010 EM_S[INDEX2(1,0,4)]+=-B_0_0*w12 - B_0_1*w10 - B_0_2*w15 - B_0_3*w13 + tmp0 + tmp1;
2011 EM_S[INDEX2(1,1,4)]+=-B_0_0*w10 - B_0_1*w12 - B_0_2*w13 - B_0_3*w15 + B_1_0*w14 + B_1_1*w16 + B_1_2*w17 + B_1_3*w11;
2012 EM_S[INDEX2(1,2,4)]+=tmp2 + tmp3 + tmp4 + tmp5;
2013 EM_S[INDEX2(1,3,4)]+=B_1_0*w17 + B_1_1*w11 + B_1_2*w14 + B_1_3*w16 + tmp10 + tmp11;
2014 EM_S[INDEX2(2,0,4)]+=-B_1_0*w16 - B_1_1*w14 - B_1_2*w11 - B_1_3*w17 + tmp14 + tmp15;
2015 EM_S[INDEX2(2,1,4)]+=tmp12 + tmp13 + tmp8 + tmp9;
2016 EM_S[INDEX2(2,2,4)]+=B_0_0*w15 + B_0_1*w13 + B_0_2*w12 + B_0_3*w10 - B_1_0*w11 - B_1_1*w17 - B_1_2*w16 - B_1_3*w14;
2017 EM_S[INDEX2(2,3,4)]+=B_0_0*w13 + B_0_1*w15 + B_0_2*w10 + B_0_3*w12 + tmp6 + tmp7;
2018 EM_S[INDEX2(3,0,4)]+=tmp2 + tmp3 + tmp8 + tmp9;
2019 EM_S[INDEX2(3,1,4)]+=-B_1_0*w14 - B_1_1*w16 - B_1_2*w17 - B_1_3*w11 + tmp10 + tmp11;
2020 EM_S[INDEX2(3,2,4)]+=-B_0_0*w15 - B_0_1*w13 - B_0_2*w12 - B_0_3*w10 + tmp6 + tmp7;
2021 EM_S[INDEX2(3,3,4)]+=-B_0_0*w13 - B_0_1*w15 - B_0_2*w10 - B_0_3*w12 - B_1_0*w17 - B_1_1*w11 - B_1_2*w14 - B_1_3*w16;
2022 } else { // constant data
2023 const double B_0 = B_p[0];
2024 const double B_1 = B_p[1];
2025 EM_S[INDEX2(0,0,4)]+= 2*B_0*w18 + 2*B_1*w19;
2026 EM_S[INDEX2(0,1,4)]+= 2*B_0*w18 + B_1*w19;
2027 EM_S[INDEX2(0,2,4)]+= B_0*w18 + 2*B_1*w19;
2028 EM_S[INDEX2(0,3,4)]+= B_0*w18 + B_1*w19;
2029 EM_S[INDEX2(1,0,4)]+=-2*B_0*w18 + B_1*w19;
2030 EM_S[INDEX2(1,1,4)]+=-2*B_0*w18 + 2*B_1*w19;
2031 EM_S[INDEX2(1,2,4)]+= -B_0*w18 + B_1*w19;
2032 EM_S[INDEX2(1,3,4)]+= -B_0*w18 + 2*B_1*w19;
2033 EM_S[INDEX2(2,0,4)]+= B_0*w18 - 2*B_1*w19;
2034 EM_S[INDEX2(2,1,4)]+= B_0*w18 - B_1*w19;
2035 EM_S[INDEX2(2,2,4)]+= 2*B_0*w18 - 2*B_1*w19;
2036 EM_S[INDEX2(2,3,4)]+= 2*B_0*w18 - B_1*w19;
2037 EM_S[INDEX2(3,0,4)]+= -B_0*w18 - B_1*w19;
2038 EM_S[INDEX2(3,1,4)]+= -B_0*w18 - 2*B_1*w19;
2039 EM_S[INDEX2(3,2,4)]+=-2*B_0*w18 - B_1*w19;
2040 EM_S[INDEX2(3,3,4)]+=-2*B_0*w18 - 2*B_1*w19;
2041 }
2042 }
2043 ///////////////
2044 // process C //
2045 ///////////////
2046 if (!C.isEmpty()) {
2047 add_EM_S=true;
2048 const double* C_p=const_cast<escript::Data*>(&C)->getSampleDataRO(e);
2049 if (C.actsExpanded()) {
2050 const double C_0_0 = C_p[INDEX2(0,0,2)];
2051 const double C_1_0 = C_p[INDEX2(1,0,2)];
2052 const double C_0_1 = C_p[INDEX2(0,1,2)];
2053 const double C_1_1 = C_p[INDEX2(1,1,2)];
2054 const double C_0_2 = C_p[INDEX2(0,2,2)];
2055 const double C_1_2 = C_p[INDEX2(1,2,2)];
2056 const double C_0_3 = C_p[INDEX2(0,3,2)];
2057 const double C_1_3 = C_p[INDEX2(1,3,2)];
2058 const double tmp0 = w11*(C_1_0 + C_1_1);
2059 const double tmp1 = w14*(C_1_2 + C_1_3);
2060 const double tmp2 = w15*(C_0_0 + C_0_2);
2061 const double tmp3 = w10*(C_0_1 + C_0_3);
2062 const double tmp4 = w11*(-C_1_0 - C_1_1);
2063 const double tmp5 = w14*(-C_1_2 - C_1_3);
2064 const double tmp6 = w11*(-C_1_2 - C_1_3);
2065 const double tmp7 = w14*(-C_1_0 - C_1_1);
2066 const double tmp8 = w11*(C_1_2 + C_1_3);
2067 const double tmp9 = w14*(C_1_0 + C_1_1);
2068 const double tmp10 = w10*(-C_0_1 - C_0_3);
2069 const double tmp11 = w15*(-C_0_0 - C_0_2);
2070 const double tmp12 = w15*(-C_0_1 - C_0_3);
2071 const double tmp13 = w10*(-C_0_0 - C_0_2);
2072 const double tmp14 = w10*(C_0_0 + C_0_2);
2073 const double tmp15 = w15*(C_0_1 + C_0_3);
2074 EM_S[INDEX2(0,0,4)]+=C_0_0*w12 + C_0_1*w10 + C_0_2*w15 + C_0_3*w13 + C_1_0*w16 + C_1_1*w14 + C_1_2*w11 + C_1_3*w17;
2075 EM_S[INDEX2(0,1,4)]+=-C_0_0*w12 - C_0_1*w10 - C_0_2*w15 - C_0_3*w13 + tmp0 + tmp1;
2076 EM_S[INDEX2(0,2,4)]+=-C_1_0*w16 - C_1_1*w14 - C_1_2*w11 - C_1_3*w17 + tmp14 + tmp15;
2077 EM_S[INDEX2(0,3,4)]+=tmp12 + tmp13 + tmp4 + tmp5;
2078 EM_S[INDEX2(1,0,4)]+=C_0_0*w10 + C_0_1*w12 + C_0_2*w13 + C_0_3*w15 + tmp0 + tmp1;
2079 EM_S[INDEX2(1,1,4)]+=-C_0_0*w10 - C_0_1*w12 - C_0_2*w13 - C_0_3*w15 + C_1_0*w14 + C_1_1*w16 + C_1_2*w17 + C_1_3*w11;
2080 EM_S[INDEX2(1,2,4)]+=tmp2 + tmp3 + tmp4 + tmp5;
2081 EM_S[INDEX2(1,3,4)]+=-C_1_0*w14 - C_1_1*w16 - C_1_2*w17 - C_1_3*w11 + tmp10 + tmp11;
2082 EM_S[INDEX2(2,0,4)]+=C_1_0*w11 + C_1_1*w17 + C_1_2*w16 + C_1_3*w14 + tmp14 + tmp15;
2083 EM_S[INDEX2(2,1,4)]+=tmp12 + tmp13 + tmp8 + tmp9;
2084 EM_S[INDEX2(2,2,4)]+=C_0_0*w15 + C_0_1*w13 + C_0_2*w12 + C_0_3*w10 - C_1_0*w11 - C_1_1*w17 - C_1_2*w16 - C_1_3*w14;
2085 EM_S[INDEX2(2,3,4)]+=-C_0_0*w15 - C_0_1*w13 - C_0_2*w12 - C_0_3*w10 + tmp6 + tmp7;
2086 EM_S[INDEX2(3,0,4)]+=tmp2 + tmp3 + tmp8 + tmp9;
2087 EM_S[INDEX2(3,1,4)]+=C_1_0*w17 + C_1_1*w11 + C_1_2*w14 + C_1_3*w16 + tmp10 + tmp11;
2088 EM_S[INDEX2(3,2,4)]+=C_0_0*w13 + C_0_1*w15 + C_0_2*w10 + C_0_3*w12 + tmp6 + tmp7;
2089 EM_S[INDEX2(3,3,4)]+=-C_0_0*w13 - C_0_1*w15 - C_0_2*w10 - C_0_3*w12 - C_1_0*w17 - C_1_1*w11 - C_1_2*w14 - C_1_3*w16;
2090 } else { // constant data
2091 const double C_0 = C_p[0];
2092 const double C_1 = C_p[1];
2093 EM_S[INDEX2(0,0,4)]+= 2*C_0*w18 + 2*C_1*w19;
2094 EM_S[INDEX2(0,1,4)]+=-2*C_0*w18 + C_1*w19;
2095 EM_S[INDEX2(0,2,4)]+= C_0*w18 - 2*C_1*w19;
2096 EM_S[INDEX2(0,3,4)]+= -C_0*w18 - C_1*w19;
2097 EM_S[INDEX2(1,0,4)]+= 2*C_0*w18 + C_1*w19;
2098 EM_S[INDEX2(1,1,4)]+=-2*C_0*w18 + 2*C_1*w19;
2099 EM_S[INDEX2(1,2,4)]+= C_0*w18 - C_1*w19;
2100 EM_S[INDEX2(1,3,4)]+= -C_0*w18 - 2*C_1*w19;
2101 EM_S[INDEX2(2,0,4)]+= C_0*w18 + 2*C_1*w19;
2102 EM_S[INDEX2(2,1,4)]+= -C_0*w18 + C_1*w19;
2103 EM_S[INDEX2(2,2,4)]+= 2*C_0*w18 - 2*C_1*w19;
2104 EM_S[INDEX2(2,3,4)]+=-2*C_0*w18 - C_1*w19;
2105 EM_S[INDEX2(3,0,4)]+= C_0*w18 + C_1*w19;
2106 EM_S[INDEX2(3,1,4)]+= -C_0*w18 + 2*C_1*w19;
2107 EM_S[INDEX2(3,2,4)]+= 2*C_0*w18 - C_1*w19;
2108 EM_S[INDEX2(3,3,4)]+=-2*C_0*w18 - 2*C_1*w19;
2109 }
2110 }
2111 ///////////////
2112 // process D //
2113 ///////////////
2114 if (!D.isEmpty()) {
2115 add_EM_S=true;
2116 const double* D_p=const_cast<escript::Data*>(&D)->getSampleDataRO(e);
2117 if (D.actsExpanded()) {
2118 const double D_0 = D_p[0];
2119 const double D_1 = D_p[1];
2120 const double D_2 = D_p[2];
2121 const double D_3 = D_p[3];
2122 const double tmp0 = w21*(D_2 + D_3);
2123 const double tmp1 = w20*(D_0 + D_1);
2124 const double tmp2 = w22*(D_0 + D_1 + D_2 + D_3);
2125 const double tmp3 = w21*(D_0 + D_1);
2126 const double tmp4 = w20*(D_2 + D_3);
2127 const double tmp5 = w22*(D_1 + D_2);
2128 const double tmp6 = w21*(D_0 + D_2);
2129 const double tmp7 = w20*(D_1 + D_3);
2130 const double tmp8 = w21*(D_1 + D_3);
2131 const double tmp9 = w20*(D_0 + D_2);
2132 const double tmp10 = w22*(D_0 + D_3);
2133 EM_S[INDEX2(0,0,4)]+=D_0*w23 + D_3*w24 + tmp5;
2134 EM_S[INDEX2(0,1,4)]+=tmp0 + tmp1;
2135 EM_S[INDEX2(0,2,4)]+=tmp8 + tmp9;
2136 EM_S[INDEX2(0,3,4)]+=tmp2;
2137 EM_S[INDEX2(1,0,4)]+=tmp0 + tmp1;
2138 EM_S[INDEX2(1,1,4)]+=D_1*w23 + D_2*w24 + tmp10;
2139 EM_S[INDEX2(1,2,4)]+=tmp2;
2140 EM_S[INDEX2(1,3,4)]+=tmp6 + tmp7;