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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 3766 - (hide annotations)
Thu Jan 12 08:17:49 2012 UTC (7 years, 9 months ago) by caltinay
Original Path: branches/ripleygmg_from_3668/ripley/src/Rectangle.cpp
File size: 191592 byte(s)
Checkpoint. Getting weipa to like ripley again.

1 caltinay 3691
2     /*******************************************************
3     *
4 caltinay 3764 * Copyright (c) 2003-2012 by University of Queensland
5 caltinay 3691 * Earth Systems Science Computational Center (ESSCC)
6     * http://www.uq.edu.au/esscc
7     *
8     * Primary Business: Queensland, Australia
9     * Licensed under the Open Software License version 3.0
10     * http://www.opensource.org/licenses/osl-3.0.php
11     *
12     *******************************************************/
13    
14     #include <ripley/Rectangle.h>
15     extern "C" {
16 caltinay 3748 #include "paso/SystemMatrix.h"
17 caltinay 3691 }
18    
19     #if USE_SILO
20     #include <silo.h>
21     #ifdef ESYS_MPI
22     #include <pmpio.h>
23     #endif
24     #endif
25    
26     #include <iomanip>
27    
28     using namespace std;
29    
30     namespace ripley {
31    
32     Rectangle::Rectangle(int n0, int n1, double l0, double l1, int d0, int d1) :
33     RipleyDomain(2),
34     m_gNE0(n0),
35     m_gNE1(n1),
36     m_l0(l0),
37     m_l1(l1),
38     m_NX(d0),
39     m_NY(d1)
40     {
41     // ensure number of subdivisions is valid and nodes can be distributed
42     // among number of ranks
43     if (m_NX*m_NY != m_mpiInfo->size)
44     throw RipleyException("Invalid number of spatial subdivisions");
45    
46 caltinay 3752 if ((n0+1)%m_NX > 0 || (n1+1)%m_NY > 0)
47     throw RipleyException("Number of elements+1 must be separable into number of ranks in each dimension");
48 caltinay 3691
49 caltinay 3752 if ((m_NX > 1 && (n0+1)/m_NX<2) || (m_NY > 1 && (n1+1)/m_NY<2))
50     throw RipleyException("Too few elements for the number of ranks");
51    
52 caltinay 3764 // local number of elements (with and without overlap)
53     m_NE0 = m_ownNE0 = (m_NX>1 ? (n0+1)/m_NX : n0);
54 caltinay 3752 if (m_mpiInfo->rank%m_NX>0 && m_mpiInfo->rank%m_NX<m_NX-1)
55     m_NE0++;
56 caltinay 3764 else if (m_NX>1 && m_mpiInfo->rank%m_NX==m_NX-1)
57     m_ownNE0--;
58    
59     m_NE1 = m_ownNE1 = (m_NY>1 ? (n1+1)/m_NY : n1);
60 caltinay 3752 if (m_mpiInfo->rank/m_NX>0 && m_mpiInfo->rank/m_NX<m_NY-1)
61     m_NE1++;
62 caltinay 3764 else if (m_NY>1 && m_mpiInfo->rank/m_NX==m_NY-1)
63     m_ownNE1--;
64 caltinay 3752
65     // local number of nodes
66 caltinay 3691 m_N0 = m_NE0+1;
67     m_N1 = m_NE1+1;
68 caltinay 3752
69 caltinay 3691 // bottom-left node is at (offset0,offset1) in global mesh
70 caltinay 3752 m_offset0 = (n0+1)/m_NX*(m_mpiInfo->rank%m_NX);
71 caltinay 3753 if (m_offset0 > 0)
72 caltinay 3752 m_offset0--;
73     m_offset1 = (n1+1)/m_NY*(m_mpiInfo->rank/m_NX);
74 caltinay 3753 if (m_offset1 > 0)
75 caltinay 3752 m_offset1--;
76    
77 caltinay 3691 populateSampleIds();
78 caltinay 3756 createPattern();
79 caltinay 3766
80     if (m_gNE0>2*m_NX && m_gNE1>2*m_NY && m_gNE0%2==0 && m_gNE1%2==0) {
81     m_coarseMesh.reset(new Rectangle(n0/2, n1/2, l0, l1, d0, d1));
82     }
83 caltinay 3691 }
84    
85     Rectangle::~Rectangle()
86     {
87     }
88    
89     string Rectangle::getDescription() const
90     {
91     return "ripley::Rectangle";
92     }
93    
94     bool Rectangle::operator==(const AbstractDomain& other) const
95     {
96 caltinay 3744 const Rectangle* o=dynamic_cast<const Rectangle*>(&other);
97     if (o) {
98     return (RipleyDomain::operator==(other) &&
99     m_gNE0==o->m_gNE0 && m_gNE1==o->m_gNE1
100     && m_l0==o->m_l0 && m_l1==o->m_l1
101     && m_NX==o->m_NX && m_NY==o->m_NY);
102     }
103 caltinay 3691
104     return false;
105     }
106    
107     void Rectangle::dump(const string& fileName) const
108     {
109     #if USE_SILO
110     string fn(fileName);
111     if (fileName.length() < 6 || fileName.compare(fileName.length()-5, 5, ".silo") != 0) {
112     fn+=".silo";
113     }
114    
115     const int NUM_SILO_FILES = 1;
116     const char* blockDirFmt = "/block%04d";
117     int driver=DB_HDF5;
118     DBfile* dbfile = NULL;
119    
120     #ifdef ESYS_MPI
121     PMPIO_baton_t* baton = NULL;
122     #endif
123    
124     if (m_mpiInfo->size > 1) {
125     #ifdef ESYS_MPI
126     baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
127     0x1337, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
128     PMPIO_DefaultClose, (void*)&driver);
129     // try the fallback driver in case of error
130     if (!baton && driver != DB_PDB) {
131     driver = DB_PDB;
132     baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
133     0x1338, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
134     PMPIO_DefaultClose, (void*)&driver);
135     }
136     if (baton) {
137 caltinay 3766 char siloPath[64];
138     snprintf(siloPath, 64, blockDirFmt, PMPIO_RankInGroup(baton, m_mpiInfo->rank));
139     dbfile = (DBfile*) PMPIO_WaitForBaton(baton, fn.c_str(), siloPath);
140 caltinay 3691 }
141     #endif
142     } else {
143     dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
144     getDescription().c_str(), driver);
145     // try the fallback driver in case of error
146     if (!dbfile && driver != DB_PDB) {
147     driver = DB_PDB;
148     dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
149     getDescription().c_str(), driver);
150     }
151 caltinay 3766 char siloPath[64];
152     snprintf(siloPath, 64, blockDirFmt, 0);
153     DBMkDir(dbfile, siloPath);
154     DBSetDir(dbfile, siloPath);
155 caltinay 3691 }
156    
157     if (!dbfile)
158     throw RipleyException("dump: Could not create Silo file");
159    
160     /*
161     if (driver==DB_HDF5) {
162     // gzip level 1 already provides good compression with minimal
163     // performance penalty. Some tests showed that gzip levels >3 performed
164     // rather badly on escript data both in terms of time and space
165     DBSetCompression("ERRMODE=FALLBACK METHOD=GZIP LEVEL=1");
166     }
167     */
168    
169     boost::scoped_ptr<double> x(new double[m_N0]);
170     boost::scoped_ptr<double> y(new double[m_N1]);
171     double* coords[2] = { x.get(), y.get() };
172 caltinay 3697 pair<double,double> xdx = getFirstCoordAndSpacing(0);
173     pair<double,double> ydy = getFirstCoordAndSpacing(1);
174 caltinay 3691 #pragma omp parallel
175     {
176 caltinay 3722 #pragma omp for nowait
177 caltinay 3691 for (dim_t i0 = 0; i0 < m_N0; i0++) {
178 caltinay 3697 coords[0][i0]=xdx.first+i0*xdx.second;
179 caltinay 3691 }
180 caltinay 3722 #pragma omp for nowait
181 caltinay 3691 for (dim_t i1 = 0; i1 < m_N1; i1++) {
182 caltinay 3697 coords[1][i1]=ydy.first+i1*ydy.second;
183 caltinay 3691 }
184     }
185 caltinay 3697 IndexVector dims = getNumNodesPerDim();
186    
187     // write mesh
188     DBPutQuadmesh(dbfile, "mesh", NULL, coords, &dims[0], 2, DB_DOUBLE,
189 caltinay 3691 DB_COLLINEAR, NULL);
190    
191 caltinay 3697 // write node ids
192     DBPutQuadvar1(dbfile, "nodeId", "mesh", (void*)&m_nodeId[0], &dims[0], 2,
193     NULL, 0, DB_INT, DB_NODECENT, NULL);
194    
195     // write element ids
196     dims = getNumElementsPerDim();
197     DBPutQuadvar1(dbfile, "elementId", "mesh", (void*)&m_elementId[0],
198     &dims[0], 2, NULL, 0, DB_INT, DB_ZONECENT, NULL);
199    
200     // rank 0 writes multimesh and multivar
201 caltinay 3691 if (m_mpiInfo->rank == 0) {
202     vector<string> tempstrings;
203 caltinay 3697 vector<char*> names;
204 caltinay 3691 for (dim_t i=0; i<m_mpiInfo->size; i++) {
205     stringstream path;
206     path << "/block" << setw(4) << setfill('0') << right << i << "/mesh";
207     tempstrings.push_back(path.str());
208 caltinay 3697 names.push_back((char*)tempstrings.back().c_str());
209 caltinay 3691 }
210 caltinay 3697 vector<int> types(m_mpiInfo->size, DB_QUAD_RECT);
211 caltinay 3691 DBSetDir(dbfile, "/");
212 caltinay 3697 DBPutMultimesh(dbfile, "multimesh", m_mpiInfo->size, &names[0],
213     &types[0], NULL);
214     tempstrings.clear();
215     names.clear();
216     for (dim_t i=0; i<m_mpiInfo->size; i++) {
217     stringstream path;
218     path << "/block" << setw(4) << setfill('0') << right << i << "/nodeId";
219     tempstrings.push_back(path.str());
220     names.push_back((char*)tempstrings.back().c_str());
221     }
222     types.assign(m_mpiInfo->size, DB_QUADVAR);
223     DBPutMultivar(dbfile, "nodeId", m_mpiInfo->size, &names[0],
224     &types[0], NULL);
225     tempstrings.clear();
226     names.clear();
227     for (dim_t i=0; i<m_mpiInfo->size; i++) {
228     stringstream path;
229     path << "/block" << setw(4) << setfill('0') << right << i << "/elementId";
230     tempstrings.push_back(path.str());
231     names.push_back((char*)tempstrings.back().c_str());
232     }
233     DBPutMultivar(dbfile, "elementId", m_mpiInfo->size, &names[0],
234     &types[0], NULL);
235 caltinay 3691 }
236    
237     if (m_mpiInfo->size > 1) {
238     #ifdef ESYS_MPI
239     PMPIO_HandOffBaton(baton, dbfile);
240     PMPIO_Finish(baton);
241     #endif
242     } else {
243     DBClose(dbfile);
244     }
245    
246     #else // USE_SILO
247     throw RipleyException("dump(): no Silo support");
248     #endif
249     }
250    
251 caltinay 3697 const int* Rectangle::borrowSampleReferenceIDs(int fsType) const
252 caltinay 3691 {
253 caltinay 3697 switch (fsType) {
254 caltinay 3691 case Nodes:
255     return &m_nodeId[0];
256 caltinay 3766 case ReducedNodes:
257     if (m_coarseMesh)
258     return m_coarseMesh->borrowSampleReferenceIDs(Nodes);
259     break;
260 caltinay 3750 case DegreesOfFreedom:
261     return &m_dofId[0];
262 caltinay 3766 case ReducedDegreesOfFreedom:
263     if (m_coarseMesh)
264     return m_coarseMesh->borrowSampleReferenceIDs(DegreesOfFreedom);
265     break;
266 caltinay 3691 case Elements:
267 caltinay 3733 case ReducedElements:
268 caltinay 3691 return &m_elementId[0];
269     case FaceElements:
270 caltinay 3733 case ReducedFaceElements:
271 caltinay 3691 return &m_faceId[0];
272     default:
273     break;
274     }
275    
276     stringstream msg;
277     msg << "borrowSampleReferenceIDs() not implemented for function space type "
278 caltinay 3701 << functionSpaceTypeAsString(fsType);
279 caltinay 3691 throw RipleyException(msg.str());
280     }
281    
282 caltinay 3757 bool Rectangle::ownSample(int fsType, index_t id) const
283 caltinay 3691 {
284 caltinay 3759 if (getMPISize()==1)
285     return true;
286    
287 caltinay 3757 switch (fsType) {
288     case Nodes:
289     return (m_dofMap[id] < getNumDOF());
290 caltinay 3766 case ReducedNodes:
291     if (m_coarseMesh)
292     return m_coarseMesh->ownSample(Nodes, id);
293     break;
294 caltinay 3757 case DegreesOfFreedom:
295     case ReducedDegreesOfFreedom:
296     return true;
297     case Elements:
298     case ReducedElements:
299     // check ownership of element's bottom left node
300     return (m_dofMap[id%m_NE0+m_N0*(id/m_NE0)] < getNumDOF());
301     case FaceElements:
302     case ReducedFaceElements:
303 caltinay 3759 {
304 caltinay 3764 // determine which face the sample belongs to before
305     // checking ownership of face element's first node
306 caltinay 3759 const IndexVector faces = getNumFacesPerBoundary();
307     dim_t n=0;
308     for (size_t i=0; i<faces.size(); i++) {
309     n+=faces[i];
310     if (id<n) {
311     index_t k;
312     if (i==1)
313     k=m_N0-1;
314     else if (i==3)
315     k=m_N0*(m_N1-1);
316     else
317     k=0;
318     // determine whether to move right or up
319     const index_t delta=(i/2==0 ? m_N0 : 1);
320     return (m_dofMap[k+(id-n+faces[i])*delta] < getNumDOF());
321     }
322     }
323     return false;
324     }
325 caltinay 3757 default:
326     break;
327 caltinay 3702 }
328 caltinay 3757
329     stringstream msg;
330     msg << "ownSample() not implemented for "
331     << functionSpaceTypeAsString(fsType);
332     throw RipleyException(msg.str());
333 caltinay 3691 }
334    
335 caltinay 3764 void Rectangle::setToNormal(escript::Data& out) const
336 caltinay 3691 {
337 caltinay 3764 if (out.getFunctionSpace().getTypeCode() == FaceElements) {
338     out.requireWrite();
339     #pragma omp parallel
340     {
341     if (m_faceOffset[0] > -1) {
342     #pragma omp for nowait
343     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
344     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
345     // set vector at two quadrature points
346     *o++ = -1.;
347     *o++ = 0.;
348     *o++ = -1.;
349     *o = 0.;
350     }
351     }
352    
353     if (m_faceOffset[1] > -1) {
354     #pragma omp for nowait
355     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
356     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
357     // set vector at two quadrature points
358     *o++ = 1.;
359     *o++ = 0.;
360     *o++ = 1.;
361     *o = 0.;
362     }
363     }
364    
365     if (m_faceOffset[2] > -1) {
366     #pragma omp for nowait
367     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
368     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
369     // set vector at two quadrature points
370     *o++ = 0.;
371     *o++ = -1.;
372     *o++ = 0.;
373     *o = -1.;
374     }
375     }
376    
377     if (m_faceOffset[3] > -1) {
378     #pragma omp for nowait
379     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
380     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
381     // set vector at two quadrature points
382     *o++ = 0.;
383     *o++ = 1.;
384     *o++ = 0.;
385     *o = 1.;
386     }
387     }
388     } // end of parallel section
389     } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
390     out.requireWrite();
391     #pragma omp parallel
392     {
393     if (m_faceOffset[0] > -1) {
394     #pragma omp for nowait
395     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
396     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
397     *o++ = -1.;
398     *o = 0.;
399     }
400     }
401    
402     if (m_faceOffset[1] > -1) {
403     #pragma omp for nowait
404     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
405     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
406     *o++ = 1.;
407     *o = 0.;
408     }
409     }
410    
411     if (m_faceOffset[2] > -1) {
412     #pragma omp for nowait
413     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
414     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
415     *o++ = 0.;
416     *o = -1.;
417     }
418     }
419    
420     if (m_faceOffset[3] > -1) {
421     #pragma omp for nowait
422     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
423     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
424     *o++ = 0.;
425     *o = 1.;
426     }
427     }
428     } // end of parallel section
429    
430     } else {
431     stringstream msg;
432     msg << "setToNormal() not implemented for "
433     << functionSpaceTypeAsString(out.getFunctionSpace().getTypeCode());
434     throw RipleyException(msg.str());
435     }
436     }
437    
438     void Rectangle::setToSize(escript::Data& out) const
439     {
440     if (out.getFunctionSpace().getTypeCode() == Elements
441     || out.getFunctionSpace().getTypeCode() == ReducedElements) {
442     out.requireWrite();
443     const dim_t numQuad=out.getNumDataPointsPerSample();
444     const double hSize=getFirstCoordAndSpacing(0).second;
445     const double vSize=getFirstCoordAndSpacing(1).second;
446     const double size=min(hSize,vSize);
447     #pragma omp parallel for
448     for (index_t k = 0; k < getNumElements(); ++k) {
449     double* o = out.getSampleDataRW(k);
450     fill(o, o+numQuad, size);
451     }
452     } else if (out.getFunctionSpace().getTypeCode() == FaceElements
453     || out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
454     out.requireWrite();
455     const dim_t numQuad=out.getNumDataPointsPerSample();
456     const double hSize=getFirstCoordAndSpacing(0).second;
457     const double vSize=getFirstCoordAndSpacing(1).second;
458     #pragma omp parallel
459     {
460     if (m_faceOffset[0] > -1) {
461     #pragma omp for nowait
462     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
463     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
464     fill(o, o+numQuad, vSize);
465     }
466     }
467    
468     if (m_faceOffset[1] > -1) {
469     #pragma omp for nowait
470     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
471     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
472     fill(o, o+numQuad, vSize);
473     }
474     }
475    
476     if (m_faceOffset[2] > -1) {
477     #pragma omp for nowait
478     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
479     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
480     fill(o, o+numQuad, hSize);
481     }
482     }
483    
484     if (m_faceOffset[3] > -1) {
485     #pragma omp for nowait
486     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
487     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
488     fill(o, o+numQuad, hSize);
489     }
490     }
491     } // end of parallel section
492    
493     } else {
494     stringstream msg;
495     msg << "setToSize() not implemented for "
496     << functionSpaceTypeAsString(out.getFunctionSpace().getTypeCode());
497     throw RipleyException(msg.str());
498     }
499     }
500    
501     Paso_SystemMatrixPattern* Rectangle::getPattern(bool reducedRowOrder,
502     bool reducedColOrder) const
503     {
504     if (reducedRowOrder || reducedColOrder)
505     throw RipleyException("getPattern() not implemented for reduced order");
506    
507     return m_pattern;
508     }
509    
510     void Rectangle::Print_Mesh_Info(const bool full) const
511     {
512     RipleyDomain::Print_Mesh_Info(full);
513     if (full) {
514     cout << " Id Coordinates" << endl;
515     cout.precision(15);
516     cout.setf(ios::scientific, ios::floatfield);
517     pair<double,double> xdx = getFirstCoordAndSpacing(0);
518     pair<double,double> ydy = getFirstCoordAndSpacing(1);
519     for (index_t i=0; i < getNumNodes(); i++) {
520     cout << " " << setw(5) << m_nodeId[i]
521     << " " << xdx.first+(i%m_N0)*xdx.second
522     << " " << ydy.first+(i/m_N0)*ydy.second << endl;
523     }
524     }
525     }
526    
527     IndexVector Rectangle::getNumNodesPerDim() const
528     {
529     IndexVector ret;
530     ret.push_back(m_N0);
531     ret.push_back(m_N1);
532     return ret;
533     }
534    
535     IndexVector Rectangle::getNumElementsPerDim() const
536     {
537     IndexVector ret;
538     ret.push_back(m_NE0);
539     ret.push_back(m_NE1);
540     return ret;
541     }
542    
543     IndexVector Rectangle::getNumFacesPerBoundary() const
544     {
545     IndexVector ret(4, 0);
546     //left
547     if (m_offset0==0)
548     ret[0]=m_NE1;
549     //right
550     if (m_mpiInfo->rank%m_NX==m_NX-1)
551     ret[1]=m_NE1;
552     //bottom
553     if (m_offset1==0)
554     ret[2]=m_NE0;
555     //top
556     if (m_mpiInfo->rank/m_NX==m_NY-1)
557     ret[3]=m_NE0;
558     return ret;
559     }
560    
561 caltinay 3766 IndexVector Rectangle::getNumSubdivisionsPerDim() const
562     {
563     IndexVector ret;
564     ret.push_back(m_NX);
565     ret.push_back(m_NY);
566     return ret;
567     }
568    
569 caltinay 3764 pair<double,double> Rectangle::getFirstCoordAndSpacing(dim_t dim) const
570     {
571     if (dim==0) {
572     return pair<double,double>((m_l0*m_offset0)/m_gNE0, m_l0/m_gNE0);
573     } else if (dim==1) {
574     return pair<double,double>((m_l1*m_offset1)/m_gNE1, m_l1/m_gNE1);
575     }
576     throw RipleyException("getFirstCoordAndSpacing(): invalid argument");
577     }
578    
579     //protected
580     dim_t Rectangle::getNumDOF() const
581     {
582     return (m_gNE0+1)/m_NX*(m_gNE1+1)/m_NY;
583     }
584    
585     //protected
586     dim_t Rectangle::getNumFaceElements() const
587     {
588     const IndexVector faces = getNumFacesPerBoundary();
589     dim_t n=0;
590     for (size_t i=0; i<faces.size(); i++)
591     n+=faces[i];
592     return n;
593     }
594    
595     //protected
596     void Rectangle::assembleCoordinates(escript::Data& arg) const
597     {
598     escriptDataC x = arg.getDataC();
599     int numDim = m_numDim;
600     if (!isDataPointShapeEqual(&x, 1, &numDim))
601     throw RipleyException("setToX: Invalid Data object shape");
602     if (!numSamplesEqual(&x, 1, getNumNodes()))
603     throw RipleyException("setToX: Illegal number of samples in Data object");
604    
605     pair<double,double> xdx = getFirstCoordAndSpacing(0);
606     pair<double,double> ydy = getFirstCoordAndSpacing(1);
607     arg.requireWrite();
608     #pragma omp parallel for
609     for (dim_t i1 = 0; i1 < m_N1; i1++) {
610     for (dim_t i0 = 0; i0 < m_N0; i0++) {
611     double* point = arg.getSampleDataRW(i0+m_N0*i1);
612     point[0] = xdx.first+i0*xdx.second;
613     point[1] = ydy.first+i1*ydy.second;
614     }
615     }
616     }
617    
618     //protected
619     void Rectangle::assembleGradient(escript::Data& out, escript::Data& in) const
620     {
621 caltinay 3701 const dim_t numComp = in.getDataPointSize();
622 caltinay 3702 const double h0 = m_l0/m_gNE0;
623     const double h1 = m_l1/m_gNE1;
624 caltinay 3724 const double cx0 = -1./h0;
625     const double cx1 = -.78867513459481288225/h0;
626     const double cx2 = -.5/h0;
627     const double cx3 = -.21132486540518711775/h0;
628     const double cx4 = .21132486540518711775/h0;
629     const double cx5 = .5/h0;
630     const double cx6 = .78867513459481288225/h0;
631     const double cx7 = 1./h0;
632     const double cy0 = -1./h1;
633     const double cy1 = -.78867513459481288225/h1;
634     const double cy2 = -.5/h1;
635     const double cy3 = -.21132486540518711775/h1;
636     const double cy4 = .21132486540518711775/h1;
637     const double cy5 = .5/h1;
638     const double cy6 = .78867513459481288225/h1;
639     const double cy7 = 1./h1;
640    
641 caltinay 3702 if (out.getFunctionSpace().getTypeCode() == Elements) {
642 caltinay 3760 out.requireWrite();
643 caltinay 3697 #pragma omp parallel for
644 caltinay 3707 for (index_t k1=0; k1 < m_NE1; ++k1) {
645     for (index_t k0=0; k0 < m_NE0; ++k0) {
646 caltinay 3702 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
647     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
648     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
649     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
650     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
651     for (index_t i=0; i < numComp; ++i) {
652 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
653     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
654     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
655     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
656     o[INDEX3(i,0,2,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
657     o[INDEX3(i,1,2,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
658     o[INDEX3(i,0,3,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
659     o[INDEX3(i,1,3,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
660 caltinay 3702 } /* end of component loop i */
661     } /* end of k0 loop */
662     } /* end of k1 loop */
663 caltinay 3711 } else if (out.getFunctionSpace().getTypeCode() == ReducedElements) {
664 caltinay 3760 out.requireWrite();
665 caltinay 3711 #pragma omp parallel for
666     for (index_t k1=0; k1 < m_NE1; ++k1) {
667     for (index_t k0=0; k0 < m_NE0; ++k0) {
668     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
669     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
670     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
671     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
672     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
673     for (index_t i=0; i < numComp; ++i) {
674 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
675     o[INDEX3(i,1,0,numComp,2)] = cy2*(f_00[i] + f_10[i]) + cy5*(f_01[i] + f_11[i]);
676 caltinay 3711 } /* end of component loop i */
677     } /* end of k0 loop */
678     } /* end of k1 loop */
679 caltinay 3707 } else if (out.getFunctionSpace().getTypeCode() == FaceElements) {
680 caltinay 3760 out.requireWrite();
681 caltinay 3722 #pragma omp parallel
682     {
683     if (m_faceOffset[0] > -1) {
684     #pragma omp for nowait
685     for (index_t k1=0; k1 < m_NE1; ++k1) {
686     const register double* f_10 = in.getSampleDataRO(INDEX2(1,k1, m_N0));
687     const register double* f_11 = in.getSampleDataRO(INDEX2(1,k1+1, m_N0));
688     const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
689     const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
690     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
691     for (index_t i=0; i < numComp; ++i) {
692 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
693     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
694     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
695     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
696 caltinay 3722 } /* end of component loop i */
697     } /* end of k1 loop */
698     } /* end of face 0 */
699     if (m_faceOffset[1] > -1) {
700     #pragma omp for nowait
701     for (index_t k1=0; k1 < m_NE1; ++k1) {
702     const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
703     const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
704     const register double* f_01 = in.getSampleDataRO(INDEX2(m_N0-2,k1+1, m_N0));
705     const register double* f_00 = in.getSampleDataRO(INDEX2(m_N0-2,k1, m_N0));
706     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
707     for (index_t i=0; i < numComp; ++i) {
708 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
709     o[INDEX3(i,1,0,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
710     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
711     o[INDEX3(i,1,1,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
712 caltinay 3722 } /* end of component loop i */
713     } /* end of k1 loop */
714     } /* end of face 1 */
715     if (m_faceOffset[2] > -1) {
716     #pragma omp for nowait
717     for (index_t k0=0; k0 < m_NE0; ++k0) {
718     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
719     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
720     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,1, m_N0));
721     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,1, m_N0));
722     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
723     for (index_t i=0; i < numComp; ++i) {
724 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
725     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
726     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
727     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
728 caltinay 3722 } /* end of component loop i */
729     } /* end of k0 loop */
730     } /* end of face 2 */
731     if (m_faceOffset[3] > -1) {
732     #pragma omp for nowait
733     for (index_t k0=0; k0 < m_NE0; ++k0) {
734     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
735     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
736     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,m_N1-2, m_N0));
737     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,m_N1-2, m_N0));
738     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
739     for (index_t i=0; i < numComp; ++i) {
740 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
741     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
742     o[INDEX3(i,0,1,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
743     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
744 caltinay 3722 } /* end of component loop i */
745     } /* end of k0 loop */
746     } /* end of face 3 */
747     } // end of parallel section
748 caltinay 3711 } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
749 caltinay 3760 out.requireWrite();
750 caltinay 3722 #pragma omp parallel
751     {
752     if (m_faceOffset[0] > -1) {
753     #pragma omp for nowait
754     for (index_t k1=0; k1 < m_NE1; ++k1) {
755     const register double* f_10 = in.getSampleDataRO(INDEX2(1,k1, m_N0));
756     const register double* f_11 = in.getSampleDataRO(INDEX2(1,k1+1, m_N0));
757     const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
758     const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
759     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
760     for (index_t i=0; i < numComp; ++i) {
761 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
762     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
763 caltinay 3722 } /* end of component loop i */
764     } /* end of k1 loop */
765     } /* end of face 0 */
766     if (m_faceOffset[1] > -1) {
767     #pragma omp for nowait
768     for (index_t k1=0; k1 < m_NE1; ++k1) {
769     const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
770     const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
771     const register double* f_01 = in.getSampleDataRO(INDEX2(m_N0-2,k1+1, m_N0));
772     const register double* f_00 = in.getSampleDataRO(INDEX2(m_N0-2,k1, m_N0));
773     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
774     for (index_t i=0; i < numComp; ++i) {
775 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
776     o[INDEX3(i,1,0,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
777 caltinay 3722 } /* end of component loop i */
778     } /* end of k1 loop */
779     } /* end of face 1 */
780     if (m_faceOffset[2] > -1) {
781     #pragma omp for nowait
782     for (index_t k0=0; k0 < m_NE0; ++k0) {
783     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
784     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
785     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,1, m_N0));
786     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,1, m_N0));
787     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
788     for (index_t i=0; i < numComp; ++i) {
789 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
790     o[INDEX3(i,1,0,numComp,2)] = cy2*(f_00[i] + f_10[i]) + cy5*(f_01[i] + f_11[i]);
791 caltinay 3722 } /* end of component loop i */
792     } /* end of k0 loop */
793     } /* end of face 2 */
794     if (m_faceOffset[3] > -1) {
795     #pragma omp for nowait
796     for (index_t k0=0; k0 < m_NE0; ++k0) {
797     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
798     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
799     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,m_N1-2, m_N0));
800     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,m_N1-2, m_N0));
801     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
802     for (index_t i=0; i < numComp; ++i) {
803 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
804     o[INDEX3(i,1,0,numComp,2)] = cy5*(f_01[i] + f_11[i]) + cy2*(f_00[i] + f_10[i]);
805 caltinay 3722 } /* end of component loop i */
806     } /* end of k0 loop */
807     } /* end of face 3 */
808     } // end of parallel section
809 caltinay 3702 }
810 caltinay 3701 }
811 caltinay 3697
812 caltinay 3764 //protected
813     void Rectangle::assembleIntegrate(vector<double>& integrals, escript::Data& arg) const
814 caltinay 3713 {
815 caltinay 3764 const dim_t numComp = arg.getDataPointSize();
816 caltinay 3713 const double h0 = m_l0/m_gNE0;
817     const double h1 = m_l1/m_gNE1;
818 caltinay 3764 const index_t left = (m_offset0==0 ? 0 : 1);
819     const index_t bottom = (m_offset1==0 ? 0 : 1);
820 caltinay 3713 if (arg.getFunctionSpace().getTypeCode() == Elements) {
821 caltinay 3764 const double w = h0*h1/4.;
822 caltinay 3713 #pragma omp parallel
823     {
824     vector<double> int_local(numComp, 0);
825 caltinay 3722 #pragma omp for nowait
826 caltinay 3764 for (index_t k1 = bottom; k1 < bottom+m_ownNE1; ++k1) {
827     for (index_t k0 = left; k0 < left+m_ownNE0; ++k0) {
828     const double* f = arg.getSampleDataRO(INDEX2(k0, k1, m_NE0));
829 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
830 caltinay 3764 const register double f0 = f[INDEX2(i,0,numComp)];
831     const register double f1 = f[INDEX2(i,1,numComp)];
832     const register double f2 = f[INDEX2(i,2,numComp)];
833     const register double f3 = f[INDEX2(i,3,numComp)];
834     int_local[i]+=(f0+f1+f2+f3)*w;
835 caltinay 3713 } /* end of component loop i */
836     } /* end of k0 loop */
837     } /* end of k1 loop */
838    
839     #pragma omp critical
840     for (index_t i=0; i<numComp; i++)
841     integrals[i]+=int_local[i];
842 caltinay 3722 } // end of parallel section
843 caltinay 3713 } else if (arg.getFunctionSpace().getTypeCode() == ReducedElements) {
844 caltinay 3764 const double w = h0*h1;
845 caltinay 3713 #pragma omp parallel
846     {
847     vector<double> int_local(numComp, 0);
848 caltinay 3722 #pragma omp for nowait
849 caltinay 3764 for (index_t k1 = bottom; k1 < bottom+m_ownNE1; ++k1) {
850     for (index_t k0 = left; k0 < left+m_ownNE0; ++k0) {
851     const double* f = arg.getSampleDataRO(INDEX2(k0, k1, m_NE0));
852 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
853 caltinay 3764 int_local[i]+=f[i]*w;
854 caltinay 3713 } /* end of component loop i */
855     } /* end of k0 loop */
856     } /* end of k1 loop */
857    
858     #pragma omp critical
859     for (index_t i=0; i<numComp; i++)
860     integrals[i]+=int_local[i];
861 caltinay 3722 } // end of parallel section
862 caltinay 3713 } else if (arg.getFunctionSpace().getTypeCode() == FaceElements) {
863 caltinay 3764 const double w0 = h0/2.;
864     const double w1 = h1/2.;
865 caltinay 3713 #pragma omp parallel
866     {
867     vector<double> int_local(numComp, 0);
868     if (m_faceOffset[0] > -1) {
869 caltinay 3722 #pragma omp for nowait
870 caltinay 3764 for (index_t k1 = bottom; k1 < bottom+m_ownNE1; ++k1) {
871     const double* f = arg.getSampleDataRO(m_faceOffset[0]+k1);
872 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
873 caltinay 3764 const register double f0 = f[INDEX2(i,0,numComp)];
874     const register double f1 = f[INDEX2(i,1,numComp)];
875     int_local[i]+=(f0+f1)*w1;
876 caltinay 3713 } /* end of component loop i */
877     } /* end of k1 loop */
878     }
879    
880     if (m_faceOffset[1] > -1) {
881 caltinay 3722 #pragma omp for nowait
882 caltinay 3764 for (index_t k1 = bottom; k1 < bottom+m_ownNE1; ++k1) {
883     const double* f = arg.getSampleDataRO(m_faceOffset[1]+k1);
884 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
885 caltinay 3764 const register double f0 = f[INDEX2(i,0,numComp)];
886     const register double f1 = f[INDEX2(i,1,numComp)];
887     int_local[i]+=(f0+f1)*w1;
888 caltinay 3713 } /* end of component loop i */
889     } /* end of k1 loop */
890     }
891    
892     if (m_faceOffset[2] > -1) {
893 caltinay 3722 #pragma omp for nowait
894 caltinay 3764 for (index_t k0 = left; k0 < left+m_ownNE0; ++k0) {
895     const double* f = arg.getSampleDataRO(m_faceOffset[2]+k0);
896 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
897 caltinay 3764 const register double f0 = f[INDEX2(i,0,numComp)];
898     const register double f1 = f[INDEX2(i,1,numComp)];
899     int_local[i]+=(f0+f1)*w0;
900 caltinay 3713 } /* end of component loop i */
901     } /* end of k0 loop */
902     }
903    
904     if (m_faceOffset[3] > -1) {
905 caltinay 3722 #pragma omp for nowait
906 caltinay 3764 for (index_t k0 = left; k0 < left+m_ownNE0; ++k0) {
907     const double* f = arg.getSampleDataRO(m_faceOffset[3]+k0);
908 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
909 caltinay 3764 const register double f0 = f[INDEX2(i,0,numComp)];
910     const register double f1 = f[INDEX2(i,1,numComp)];
911     int_local[i]+=(f0+f1)*w0;
912 caltinay 3713 } /* end of component loop i */
913     } /* end of k0 loop */
914     }
915    
916     #pragma omp critical
917     for (index_t i=0; i<numComp; i++)
918     integrals[i]+=int_local[i];
919 caltinay 3722 } // end of parallel section
920 caltinay 3713 } else if (arg.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
921     #pragma omp parallel
922     {
923     vector<double> int_local(numComp, 0);
924     if (m_faceOffset[0] > -1) {
925 caltinay 3722 #pragma omp for nowait
926 caltinay 3764 for (index_t k1 = bottom; k1 < bottom+m_ownNE1; ++k1) {
927     const double* f = arg.getSampleDataRO(m_faceOffset[0]+k1);
928 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
929     int_local[i]+=f[i]*h1;
930     } /* end of component loop i */
931     } /* end of k1 loop */
932     }
933    
934     if (m_faceOffset[1] > -1) {
935 caltinay 3722 #pragma omp for nowait
936 caltinay 3764 for (index_t k1 = bottom; k1 < bottom+m_ownNE1; ++k1) {
937     const double* f = arg.getSampleDataRO(m_faceOffset[1]+k1);
938 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
939     int_local[i]+=f[i]*h1;
940     } /* end of component loop i */
941     } /* end of k1 loop */
942     }
943    
944     if (m_faceOffset[2] > -1) {
945 caltinay 3722 #pragma omp for nowait
946 caltinay 3764 for (index_t k0 = left; k0 < left+m_ownNE0; ++k0) {
947     const double* f = arg.getSampleDataRO(m_faceOffset[2]+k0);
948 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
949     int_local[i]+=f[i]*h0;
950     } /* end of component loop i */
951     } /* end of k0 loop */
952     }
953    
954     if (m_faceOffset[3] > -1) {
955 caltinay 3722 #pragma omp for nowait
956 caltinay 3764 for (index_t k0 = left; k0 < left+m_ownNE0; ++k0) {
957     const double* f = arg.getSampleDataRO(m_faceOffset[3]+k0);
958 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
959     int_local[i]+=f[i]*h0;
960     } /* end of component loop i */
961     } /* end of k0 loop */
962     }
963    
964     #pragma omp critical
965     for (index_t i=0; i<numComp; i++)
966     integrals[i]+=int_local[i];
967 caltinay 3722 } // end of parallel section
968 caltinay 3713 }
969     }
970    
971 caltinay 3691 //protected
972 caltinay 3756 dim_t Rectangle::insertNeighbourNodes(IndexVector& index, index_t node) const
973     {
974     const dim_t nDOF0 = (m_gNE0+1)/m_NX;
975     const dim_t nDOF1 = (m_gNE1+1)/m_NY;
976     const int x=node%nDOF0;
977     const int y=node/nDOF0;
978     dim_t num=0;
979     // loop through potential neighbours and add to index if positions are
980     // within bounds
981     for (int i1=-1; i1<2; i1++) {
982     for (int i0=-1; i0<2; i0++) {
983     // skip node itself
984     if (i0==0 && i1==0)
985     continue;
986     // location of neighbour node
987     const int nx=x+i0;
988     const int ny=y+i1;
989     if (nx>=0 && ny>=0 && nx<nDOF0 && ny<nDOF1) {
990     index.push_back(ny*nDOF0+nx);
991     num++;
992     }
993     }
994     }
995    
996     return num;
997     }
998    
999     //protected
1000     void Rectangle::nodesToDOF(escript::Data& out, escript::Data& in) const
1001     {
1002     const dim_t numComp = in.getDataPointSize();
1003     out.requireWrite();
1004    
1005     const index_t left = (m_offset0==0 ? 0 : 1);
1006     const index_t bottom = (m_offset1==0 ? 0 : 1);
1007     const dim_t nDOF0 = (m_gNE0+1)/m_NX;
1008     const dim_t nDOF1 = (m_gNE1+1)/m_NY;
1009     #pragma omp parallel for
1010     for (index_t i=0; i<nDOF1; i++) {
1011     for (index_t j=0; j<nDOF0; j++) {
1012     const index_t n=j+left+(i+bottom)*m_N0;
1013     const double* src=in.getSampleDataRO(n);
1014     copy(src, src+numComp, out.getSampleDataRW(j+i*nDOF0));
1015     }
1016     }
1017     }
1018    
1019     //protected
1020     void Rectangle::dofToNodes(escript::Data& out, escript::Data& in) const
1021     {
1022     const dim_t numComp = in.getDataPointSize();
1023     Paso_Coupler* coupler = Paso_Coupler_alloc(m_connector, numComp);
1024     in.requireWrite();
1025     Paso_Coupler_startCollect(coupler, in.getSampleDataRW(0));
1026    
1027     const dim_t numDOF = getNumDOF();
1028     out.requireWrite();
1029     const double* buffer = Paso_Coupler_finishCollect(coupler);
1030    
1031     #pragma omp parallel for
1032     for (index_t i=0; i<getNumNodes(); i++) {
1033     const double* src=(m_dofMap[i]<numDOF ?
1034     in.getSampleDataRO(m_dofMap[i])
1035     : &buffer[(m_dofMap[i]-numDOF)*numComp]);
1036     copy(src, src+numComp, out.getSampleDataRW(i));
1037     }
1038     }
1039    
1040 caltinay 3691 //private
1041     void Rectangle::populateSampleIds()
1042     {
1043 caltinay 3752 // identifiers are ordered from left to right, bottom to top globablly.
1044 caltinay 3697
1045     // build node distribution vector first.
1046     // rank i owns m_nodeDistribution[i+1]-nodeDistribution[i] nodes
1047     m_nodeDistribution.assign(m_mpiInfo->size+1, 0);
1048 caltinay 3752 const dim_t numDOF=getNumDOF();
1049     for (dim_t k=1; k<m_mpiInfo->size; k++) {
1050     m_nodeDistribution[k]=k*numDOF;
1051 caltinay 3697 }
1052     m_nodeDistribution[m_mpiInfo->size]=getNumDataPointsGlobal();
1053 caltinay 3691 m_nodeId.resize(getNumNodes());
1054 caltinay 3753 m_dofId.resize(numDOF);
1055     m_elementId.resize(getNumElements());
1056     m_faceId.resize(getNumFaceElements());
1057 caltinay 3697
1058 caltinay 3753 #pragma omp parallel
1059     {
1060     // nodes
1061     #pragma omp for nowait
1062     for (dim_t i1=0; i1<m_N1; i1++) {
1063     for (dim_t i0=0; i0<m_N0; i0++) {
1064     m_nodeId[i0+i1*m_N0] = (m_offset1+i1)*(m_gNE0+1)+m_offset0+i0;
1065     }
1066 caltinay 3697 }
1067    
1068 caltinay 3753 // degrees of freedom
1069     #pragma omp for nowait
1070     for (dim_t k=0; k<numDOF; k++)
1071     m_dofId[k] = m_nodeDistribution[m_mpiInfo->rank]+k;
1072 caltinay 3752
1073 caltinay 3753 // elements
1074     #pragma omp for nowait
1075 caltinay 3755 for (dim_t i1=0; i1<m_NE1; i1++) {
1076     for (dim_t i0=0; i0<m_NE0; i0++) {
1077     m_elementId[i0+i1*m_NE0]=(m_offset1+i1)*m_gNE0+m_offset0+i0;
1078     }
1079     }
1080 caltinay 3753
1081     // face elements
1082     #pragma omp for
1083     for (dim_t k=0; k<getNumFaceElements(); k++)
1084     m_faceId[k]=k;
1085     } // end parallel section
1086    
1087 caltinay 3735 m_nodeTags.assign(getNumNodes(), 0);
1088     updateTagsInUse(Nodes);
1089 caltinay 3697
1090 caltinay 3735 m_elementTags.assign(getNumElements(), 0);
1091     updateTagsInUse(Elements);
1092 caltinay 3697
1093 caltinay 3722 // generate face offset vector and set face tags
1094 caltinay 3704 const IndexVector facesPerEdge = getNumFacesPerBoundary();
1095 caltinay 3722 const index_t LEFT=1, RIGHT=2, BOTTOM=10, TOP=20;
1096     const index_t faceTag[] = { LEFT, RIGHT, BOTTOM, TOP };
1097 caltinay 3704 m_faceOffset.assign(facesPerEdge.size(), -1);
1098 caltinay 3722 m_faceTags.clear();
1099 caltinay 3704 index_t offset=0;
1100     for (size_t i=0; i<facesPerEdge.size(); i++) {
1101     if (facesPerEdge[i]>0) {
1102     m_faceOffset[i]=offset;
1103     offset+=facesPerEdge[i];
1104 caltinay 3722 m_faceTags.insert(m_faceTags.end(), facesPerEdge[i], faceTag[i]);
1105 caltinay 3704 }
1106     }
1107 caltinay 3722 setTagMap("left", LEFT);
1108     setTagMap("right", RIGHT);
1109     setTagMap("bottom", BOTTOM);
1110     setTagMap("top", TOP);
1111     updateTagsInUse(FaceElements);
1112 caltinay 3691 }
1113    
1114 caltinay 3699 //private
1115 caltinay 3756 void Rectangle::createPattern()
1116 caltinay 3699 {
1117 caltinay 3756 const dim_t nDOF0 = (m_gNE0+1)/m_NX;
1118     const dim_t nDOF1 = (m_gNE1+1)/m_NY;
1119     const index_t left = (m_offset0==0 ? 0 : 1);
1120     const index_t bottom = (m_offset1==0 ? 0 : 1);
1121    
1122     // populate node->DOF mapping with own degrees of freedom.
1123     // The rest is assigned in the loop further down
1124     m_dofMap.assign(getNumNodes(), 0);
1125     #pragma omp parallel for
1126 caltinay 3766 for (index_t i=bottom; i<bottom+nDOF1; i++) {
1127     for (index_t j=left; j<left+nDOF0; j++) {
1128 caltinay 3756 m_dofMap[i*m_N0+j]=(i-bottom)*nDOF0+j-left;
1129     }
1130     }
1131    
1132     // build list of shared components and neighbours by looping through
1133     // all potential neighbouring ranks and checking if positions are
1134 caltinay 3754 // within bounds
1135 caltinay 3756 const dim_t numDOF=nDOF0*nDOF1;
1136     vector<IndexVector> colIndices(numDOF); // for the couple blocks
1137     RankVector neighbour;
1138     IndexVector offsetInShared(1,0);
1139     IndexVector sendShared, recvShared;
1140     int numShared=0;
1141     const int x=m_mpiInfo->rank%m_NX;
1142     const int y=m_mpiInfo->rank/m_NX;
1143 caltinay 3754 for (int i1=-1; i1<2; i1++) {
1144     for (int i0=-1; i0<2; i0++) {
1145 caltinay 3756 // skip this rank
1146 caltinay 3754 if (i0==0 && i1==0)
1147     continue;
1148 caltinay 3756 // location of neighbour rank
1149 caltinay 3754 const int nx=x+i0;
1150     const int ny=y+i1;
1151 caltinay 3756 if (nx>=0 && ny>=0 && nx<m_NX && ny<m_NY) {
1152     neighbour.push_back(ny*m_NX+nx);
1153     if (i0==0) {
1154     // sharing top or bottom edge
1155     const int firstDOF=(i1==-1 ? 0 : numDOF-nDOF0);
1156     const int firstNode=(i1==-1 ? left : m_N0*(m_N1-1)+left);
1157     offsetInShared.push_back(offsetInShared.back()+nDOF0);
1158     for (dim_t i=0; i<nDOF0; i++, numShared++) {
1159     sendShared.push_back(firstDOF+i);
1160     recvShared.push_back(numDOF+numShared);
1161     if (i>0)
1162     colIndices[firstDOF+i-1].push_back(numShared);
1163     colIndices[firstDOF+i].push_back(numShared);
1164     if (i<nDOF0-1)
1165     colIndices[firstDOF+i+1].push_back(numShared);
1166     m_dofMap[firstNode+i]=numDOF+numShared;
1167     }
1168     } else if (i1==0) {
1169     // sharing left or right edge
1170     const int firstDOF=(i0==-1 ? 0 : nDOF0-1);
1171     const int firstNode=(i0==-1 ? bottom*m_N0 : (bottom+1)*m_N0-1);
1172     offsetInShared.push_back(offsetInShared.back()+nDOF1);
1173     for (dim_t i=0; i<nDOF1; i++, numShared++) {
1174     sendShared.push_back(firstDOF+i*nDOF0);
1175     recvShared.push_back(numDOF+numShared);
1176     if (i>0)
1177     colIndices[firstDOF+(i-1)*nDOF0].push_back(numShared);
1178     colIndices[firstDOF+i*nDOF0].push_back(numShared);
1179     if (i<nDOF1-1)
1180     colIndices[firstDOF+(i+1)*nDOF0].push_back(numShared);
1181     m_dofMap[firstNode+i*m_N0]=numDOF+numShared;
1182     }
1183     } else {
1184     // sharing a node
1185     const int dof=(i0+1)/2*(nDOF0-1)+(i1+1)/2*(numDOF-nDOF0);
1186     const int node=(i0+1)/2*(m_N0-1)+(i1+1)/2*m_N0*(m_N1-1);
1187     offsetInShared.push_back(offsetInShared.back()+1);
1188     sendShared.push_back(dof);
1189     recvShared.push_back(numDOF+numShared);
1190     colIndices[dof].push_back(numShared);
1191     m_dofMap[node]=numDOF+numShared;
1192     ++numShared;
1193     }
1194 caltinay 3754 }
1195 caltinay 3699 }
1196     }
1197 caltinay 3754
1198 caltinay 3756 // create connector
1199     Paso_SharedComponents *snd_shcomp = Paso_SharedComponents_alloc(
1200     numDOF, neighbour.size(), &neighbour[0], &sendShared[0],
1201     &offsetInShared[0], 1, 0, m_mpiInfo);
1202     Paso_SharedComponents *rcv_shcomp = Paso_SharedComponents_alloc(
1203     numDOF, neighbour.size(), &neighbour[0], &recvShared[0],
1204     &offsetInShared[0], 1, 0, m_mpiInfo);
1205     m_connector = Paso_Connector_alloc(snd_shcomp, rcv_shcomp);
1206     Paso_SharedComponents_free(snd_shcomp);
1207     Paso_SharedComponents_free(rcv_shcomp);
1208 caltinay 3754
1209 caltinay 3756 // create main and couple blocks
1210     Paso_Pattern *mainPattern = createMainPattern();
1211     Paso_Pattern *colPattern, *rowPattern;
1212     createCouplePatterns(colIndices, numShared, &colPattern, &rowPattern);
1213 caltinay 3754
1214 caltinay 3756 // allocate paso distribution
1215     Paso_Distribution* distribution = Paso_Distribution_alloc(m_mpiInfo,
1216     const_cast<index_t*>(&m_nodeDistribution[0]), 1, 0);
1217 caltinay 3755
1218 caltinay 3756 // finally create the system matrix
1219     m_pattern = Paso_SystemMatrixPattern_alloc(MATRIX_FORMAT_DEFAULT,
1220     distribution, distribution, mainPattern, colPattern, rowPattern,
1221     m_connector, m_connector);
1222 caltinay 3755
1223 caltinay 3756 Paso_Distribution_free(distribution);
1224 caltinay 3755
1225 caltinay 3756 // useful debug output
1226     /*
1227     cout << "--- rcv_shcomp ---" << endl;
1228     cout << "numDOF=" << numDOF << ", numNeighbors=" << neighbour.size() << endl;
1229     for (size_t i=0; i<neighbour.size(); i++) {
1230     cout << "neighbor[" << i << "]=" << neighbour[i]
1231     << " offsetInShared[" << i+1 << "]=" << offsetInShared[i+1] << endl;
1232 caltinay 3699 }
1233 caltinay 3756 for (size_t i=0; i<recvShared.size(); i++) {
1234     cout << "shared[" << i << "]=" << recvShared[i] << endl;
1235     }
1236     cout << "--- snd_shcomp ---" << endl;
1237     for (size_t i=0; i<sendShared.size(); i++) {
1238     cout << "shared[" << i << "]=" << sendShared[i] << endl;
1239     }
1240     cout << "--- dofMap ---" << endl;
1241     for (size_t i=0; i<m_dofMap.size(); i++) {
1242     cout << "m_dofMap[" << i << "]=" << m_dofMap[i] << endl;
1243     }
1244     cout << "--- colIndices ---" << endl;
1245     for (size_t i=0; i<colIndices.size(); i++) {
1246     cout << "colIndices[" << i << "].size()=" << colIndices[i].size() << endl;
1247     }
1248     */
1249 caltinay 3754
1250 caltinay 3756 /*
1251     cout << "--- main_pattern ---" << endl;
1252     cout << "M=" << mainPattern->numOutput << ", N=" << mainPattern->numInput << endl;
1253     for (size_t i=0; i<mainPattern->numOutput+1; i++) {
1254     cout << "ptr[" << i << "]=" << mainPattern->ptr[i] << endl;
1255     }
1256     for (size_t i=0; i<mainPattern->ptr[mainPattern->numOutput]; i++) {
1257     cout << "index[" << i << "]=" << mainPattern->index[i] << endl;
1258     }
1259     */
1260 caltinay 3754
1261 caltinay 3756 /*
1262     cout << "--- colCouple_pattern ---" << endl;
1263     cout << "M=" << colPattern->numOutput << ", N=" << colPattern->numInput << endl;
1264     for (size_t i=0; i<colPattern->numOutput+1; i++) {
1265     cout << "ptr[" << i << "]=" << colPattern->ptr[i] << endl;
1266     }
1267     for (size_t i=0; i<colPattern->ptr[colPattern->numOutput]; i++) {
1268     cout << "index[" << i << "]=" << colPattern->index[i] << endl;
1269     }
1270     */
1271 caltinay 3754
1272 caltinay 3756 /*
1273     cout << "--- rowCouple_pattern ---" << endl;
1274     cout << "M=" << rowPattern->numOutput << ", N=" << rowPattern->numInput << endl;
1275     for (size_t i=0; i<rowPattern->numOutput+1; i++) {
1276     cout << "ptr[" << i << "]=" << rowPattern->ptr[i] << endl;
1277 caltinay 3699 }
1278 caltinay 3756 for (size_t i=0; i<rowPattern->ptr[rowPattern->numOutput]; i++) {
1279     cout << "index[" << i << "]=" << rowPattern->index[i] << endl;
1280     }
1281     */
1282    
1283     Paso_Pattern_free(mainPattern);
1284     Paso_Pattern_free(colPattern);
1285     Paso_Pattern_free(rowPattern);
1286 caltinay 3699 }
1287    
1288 caltinay 3702 //protected
1289 caltinay 3711 void Rectangle::interpolateNodesOnElements(escript::Data& out,
1290     escript::Data& in, bool reduced) const
1291 caltinay 3702 {
1292     const dim_t numComp = in.getDataPointSize();
1293 caltinay 3711 if (reduced) {
1294 caltinay 3760 out.requireWrite();
1295 caltinay 3724 const double c0 = .25;
1296 caltinay 3702 #pragma omp parallel for
1297 caltinay 3711 for (index_t k1=0; k1 < m_NE1; ++k1) {
1298     for (index_t k0=0; k0 < m_NE0; ++k0) {
1299     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
1300     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
1301     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
1302     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
1303     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
1304     for (index_t i=0; i < numComp; ++i) {
1305 caltinay 3724 o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i] + f_10[i] + f_11[i]);
1306 caltinay 3711 } /* end of component loop i */
1307     } /* end of k0 loop */
1308     } /* end of k1 loop */
1309     } else {
1310 caltinay 3760 out.requireWrite();
1311 caltinay 3724 const double c0 = .16666666666666666667;
1312     const double c1 = .044658198738520451079;
1313     const double c2 = .62200846792814621559;
1314 caltinay 3711 #pragma omp parallel for
1315     for (index_t k1=0; k1 < m_NE1; ++k1) {
1316     for (index_t k0=0; k0 < m_NE0; ++k0) {
1317     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
1318     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
1319     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
1320     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
1321     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
1322     for (index_t i=0; i < numComp; ++i) {
1323 caltinay 3724 o[INDEX2(i,numComp,0)] = f_00[i]*c2 + f_11[i]*c1 + c0*(f_01[i] + f_10[i]);
1324     o[INDEX2(i,numComp,1)] = f_01[i]*c1 + f_10[i]*c2 + c0*(f_00[i] + f_11[i]);
1325     o[INDEX2(i,numComp,2)] = f_01[i]*c2 + f_10[i]*c1 + c0*(f_00[i] + f_11[i]);
1326     o[INDEX2(i,numComp,3)] = f_00[i]*c1 + f_11[i]*c2 + c0*(f_01[i] + f_10[i]);
1327 caltinay 3711 } /* end of component loop i */
1328     } /* end of k0 loop */
1329     } /* end of k1 loop */
1330     }
1331 caltinay 3702 }
1332    
1333     //protected
1334 caltinay 3711 void Rectangle::interpolateNodesOnFaces(escript::Data& out, escript::Data& in,
1335     bool reduced) const
1336 caltinay 3702 {
1337 caltinay 3704 const dim_t numComp = in.getDataPointSize();
1338 caltinay 3711 if (reduced) {
1339 caltinay 3760 out.requireWrite();
1340 caltinay 3724 const double c0 = .5;
1341     #pragma omp parallel
1342     {
1343     if (m_faceOffset[0] > -1) {
1344     #pragma omp for nowait
1345     for (index_t k1=0; k1 < m_NE1; ++k1) {
1346     const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
1347     const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
1348     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1349     for (index_t i=0; i < numComp; ++i) {
1350     o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i]);
1351     } /* end of component loop i */
1352     } /* end of k1 loop */
1353     } /* end of face 0 */
1354     if (m_faceOffset[1] > -1) {
1355     #pragma omp for nowait
1356     for (index_t k1=0; k1 < m_NE1; ++k1) {
1357     const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
1358     const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
1359     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1360     for (index_t i=0; i < numComp; ++i) {
1361     o[INDEX2(i,numComp,0)] = c0*(f_10[i] + f_11[i]);
1362     } /* end of component loop i */
1363     } /* end of k1 loop */
1364     } /* end of face 1 */
1365     if (m_faceOffset[2] > -1) {
1366     #pragma omp for nowait
1367     for (index_t k0=0; k0 < m_NE0; ++k0) {
1368     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
1369     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
1370     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1371     for (index_t i=0; i < numComp; ++i) {
1372     o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_10[i]);
1373     } /* end of component loop i */
1374     } /* end of k0 loop */
1375     } /* end of face 2 */
1376     if (m_faceOffset[3] > -1) {
1377     #pragma omp for nowait
1378     for (index_t k0=0; k0 < m_NE0; ++k0) {
1379     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
1380     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
1381     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1382     for (index_t i=0; i < numComp; ++i) {
1383     o[INDEX2(i,numComp,0)] = c0*(f_01[i] + f_11[i]);
1384     } /* end of component loop i */
1385     } /* end of k0 loop */
1386     } /* end of face 3 */
1387     } // end of parallel section
1388 caltinay 3711 } else {
1389 caltinay 3760 out.requireWrite();
1390 caltinay 3724 const double c0 = 0.21132486540518711775;
1391     const double c1 = 0.78867513459481288225;
1392     #pragma omp parallel
1393     {
1394     if (m_faceOffset[0] > -1) {
1395     #pragma omp for nowait
1396     for (index_t k1=0; k1 < m_NE1; ++k1) {
1397     const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
1398     const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
1399     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1400     for (index_t i=0; i < numComp; ++i) {
1401     o[INDEX2(i,numComp,0)] = f_00[i]*c1 + f_01[i]*c0;
1402     o[INDEX2(i,numComp,1)] = f_00[i]*c0 + f_01[i]*c1;
1403     } /* end of component loop i */
1404     } /* end of k1 loop */
1405     } /* end of face 0 */
1406     if (m_faceOffset[1] > -1) {
1407     #pragma omp for nowait
1408     for (index_t k1=0; k1 < m_NE1; ++k1) {
1409     const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
1410     const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
1411     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1412     for (index_t i=0; i < numComp; ++i) {
1413     o[INDEX2(i,numComp,0)] = f_10[i]*c1 + f_11[i]*c0;
1414     o[INDEX2(i,numComp,1)] = f_10[i]*c0 + f_11[i]*c1;
1415     } /* end of component loop i */
1416     } /* end of k1 loop */
1417     } /* end of face 1 */
1418     if (m_faceOffset[2] > -1) {
1419     #pragma omp for nowait
1420     for (index_t k0=0; k0 < m_NE0; ++k0) {
1421     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
1422     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
1423     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1424     for (index_t i=0; i < numComp; ++i) {
1425     o[INDEX2(i,numComp,0)] = f_00[i]*c1 + f_10[i]*c0;
1426     o[INDEX2(i,numComp,1)] = f_00[i]*c0 + f_10[i]*c1;
1427     } /* end of component loop i */
1428     } /* end of k0 loop */
1429     } /* end of face 2 */
1430     if (m_faceOffset[3] > -1) {
1431     #pragma omp for nowait
1432     for (index_t k0=0; k0 < m_NE0; ++k0) {
1433     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
1434     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
1435     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1436     for (index_t i=0; i < numComp; ++i) {
1437     o[INDEX2(i,numComp,0)] = f_01[i]*c1 + f_11[i]*c0;
1438     o[INDEX2(i,numComp,1)] = f_01[i]*c0 + f_11[i]*c1;
1439     } /* end of component loop i */
1440     } /* end of k0 loop */
1441     } /* end of face 3 */
1442     } // end of parallel section
1443 caltinay 3711 }
1444 caltinay 3702 }
1445    
1446 caltinay 3748 //protected
1447     void Rectangle::assemblePDESingle(Paso_SystemMatrix* mat,
1448     escript::Data& rhs, const escript::Data& A, const escript::Data& B,
1449     const escript::Data& C, const escript::Data& D,
1450     const escript::Data& X, const escript::Data& Y,
1451     const escript::Data& d, const escript::Data& y) const
1452     {
1453     const double h0 = m_l0/m_gNE0;
1454     const double h1 = m_l1/m_gNE1;
1455     const double w0 = -0.1555021169820365539*h1/h0;
1456     const double w1 = 0.041666666666666666667;
1457 caltinay 3764 const double w2 = -0.15550211698203655390;
1458     const double w3 = 0.041666666666666666667*h0/h1;
1459     const double w4 = 0.15550211698203655390;
1460     const double w5 = -0.041666666666666666667;
1461     const double w6 = -0.01116454968463011277*h1/h0;
1462     const double w7 = 0.011164549684630112770;
1463     const double w8 = -0.011164549684630112770;
1464     const double w9 = -0.041666666666666666667*h1/h0;
1465 caltinay 3748 const double w10 = -0.041666666666666666667*h0/h1;
1466     const double w11 = 0.1555021169820365539*h1/h0;
1467     const double w12 = 0.1555021169820365539*h0/h1;
1468     const double w13 = 0.01116454968463011277*h0/h1;
1469     const double w14 = 0.01116454968463011277*h1/h0;
1470     const double w15 = 0.041666666666666666667*h1/h0;
1471     const double w16 = -0.01116454968463011277*h0/h1;
1472     const double w17 = -0.1555021169820365539*h0/h1;
1473     const double w18 = -0.33333333333333333333*h1/h0;
1474     const double w19 = 0.25000000000000000000;
1475     const double w20 = -0.25000000000000000000;
1476     const double w21 = 0.16666666666666666667*h0/h1;
1477     const double w22 = -0.16666666666666666667*h1/h0;
1478     const double w23 = -0.16666666666666666667*h0/h1;
1479     const double w24 = 0.33333333333333333333*h1/h0;
1480     const double w25 = 0.33333333333333333333*h0/h1;
1481     const double w26 = 0.16666666666666666667*h1/h0;
1482     const double w27 = -0.33333333333333333333*h0/h1;
1483     const double w28 = -0.032861463941450536761*h1;
1484     const double w29 = -0.032861463941450536761*h0;
1485     const double w30 = -0.12264065304058601714*h1;
1486     const double w31 = -0.0023593469594139828636*h1;
1487     const double w32 = -0.008805202725216129906*h0;
1488     const double w33 = -0.008805202725216129906*h1;
1489     const double w34 = 0.032861463941450536761*h1;
1490     const double w35 = 0.008805202725216129906*h1;
1491     const double w36 = 0.008805202725216129906*h0;
1492     const double w37 = 0.0023593469594139828636*h1;
1493     const double w38 = 0.12264065304058601714*h1;
1494     const double w39 = 0.032861463941450536761*h0;
1495     const double w40 = -0.12264065304058601714*h0;
1496     const double w41 = -0.0023593469594139828636*h0;
1497     const double w42 = 0.0023593469594139828636*h0;
1498     const double w43 = 0.12264065304058601714*h0;
1499     const double w44 = -0.16666666666666666667*h1;
1500     const double w45 = -0.083333333333333333333*h0;
1501     const double w46 = 0.083333333333333333333*h1;
1502     const double w47 = 0.16666666666666666667*h1;
1503     const double w48 = 0.083333333333333333333*h0;
1504     const double w49 = -0.16666666666666666667*h0;
1505     const double w50 = 0.16666666666666666667*h0;
1506     const double w51 = -0.083333333333333333333*h1;
1507     const double w52 = 0.025917019497006092316*h0*h1;
1508     const double w53 = 0.0018607582807716854616*h0*h1;
1509     const double w54 = 0.0069444444444444444444*h0*h1;
1510     const double w55 = 0.09672363354357992482*h0*h1;
1511     const double w56 = 0.00049858867864229740201*h0*h1;
1512     const double w57 = 0.055555555555555555556*h0*h1;
1513     const double w58 = 0.027777777777777777778*h0*h1;
1514     const double w59 = 0.11111111111111111111*h0*h1;
1515     const double w60 = -0.19716878364870322056*h1;
1516     const double w61 = -0.19716878364870322056*h0;
1517     const double w62 = -0.052831216351296779436*h0;
1518     const double w63 = -0.052831216351296779436*h1;
1519     const double w64 = 0.19716878364870322056*h1;
1520     const double w65 = 0.052831216351296779436*h1;
1521     const double w66 = 0.19716878364870322056*h0;
1522     const double w67 = 0.052831216351296779436*h0;
1523     const double w68 = -0.5*h1;
1524     const double w69 = -0.5*h0;
1525     const double w70 = 0.5*h1;
1526     const double w71 = 0.5*h0;
1527     const double w72 = 0.1555021169820365539*h0*h1;
1528     const double w73 = 0.041666666666666666667*h0*h1;
1529     const double w74 = 0.01116454968463011277*h0*h1;
1530     const double w75 = 0.25*h0*h1;
1531    
1532     rhs.requireWrite();
1533     #pragma omp parallel
1534     {
1535 caltinay 3754 for (index_t k1_0=0; k1_0<2; k1_0++) { // colouring
1536 caltinay 3748 #pragma omp for
1537     for (index_t k1=k1_0; k1<m_NE1; k1+=2) {
1538     for (index_t k0=0; k0<m_NE0; ++k0) {
1539     bool add_EM_S=false;
1540     bool add_EM_F=false;
1541     vector<double> EM_S(4*4, 0);
1542     vector<double> EM_F(4, 0);
1543     const index_t e = k0 + m_NE0*k1;
1544     ///////////////
1545     // process A //
1546     ///////////////
1547     if (!A.isEmpty()) {
1548     add_EM_S=true;
1549     const double* A_p=const_cast<escript::Data*>(&A)->getSampleDataRO(e);
1550     if (A.actsExpanded()) {
1551     const register double A_00_0 = A_p[INDEX3(0,0,0,2,2)];
1552 caltinay 3764 const register double A_10_0 = A_p[INDEX3(1,0,0,2,2)];
1553 caltinay 3748 const register double A_01_0 = A_p[INDEX3(0,1,0,2,2)];
1554     const register double A_11_0 = A_p[INDEX3(1,1,0,2,2)];
1555     const register double A_00_1 = A_p[INDEX3(0,0,1,2,2)];
1556 caltinay 3764 const register double A_10_1 = A_p[INDEX3(1,0,1,2,2)];
1557 caltinay 3748 const register double A_01_1 = A_p[INDEX3(0,1,1,2,2)];
1558     const register double A_11_1 = A_p[INDEX3(1,1,1,2,2)];
1559     const register double A_00_2 = A_p[INDEX3(0,0,2,2,2)];
1560 caltinay 3764 const register double A_10_2 = A_p[INDEX3(1,0,2,2,2)];
1561 caltinay 3748 const register double A_01_2 = A_p[INDEX3(0,1,2,2,2)];
1562     const register double A_11_2 = A_p[INDEX3(1,1,2,2,2)];
1563     const register double A_00_3 = A_p[INDEX3(0,0,3,2,2)];
1564 caltinay 3764 const register double A_10_3 = A_p[INDEX3(1,0,3,2,2)];
1565 caltinay 3748 const register double A_01_3 = A_p[INDEX3(0,1,3,2,2)];
1566     const register double A_11_3 = A_p[INDEX3(1,1,3,2,2)];
1567 caltinay 3764 const register double tmp0_0 = A_01_0 + A_01_3;
1568     const register double tmp1_0 = A_00_0 + A_00_1;
1569     const register double tmp2_0 = A_11_0 + A_11_1 + A_11_2 + A_11_3;
1570     const register double tmp3_0 = A_00_2 + A_00_3;
1571 caltinay 3748 const register double tmp4_0 = A_10_1 + A_10_2;
1572 caltinay 3764 const register double tmp5_0 = A_00_0 + A_00_1 + A_00_2 + A_00_3;
1573     const register double tmp6_0 = A_01_3 + A_10_0;
1574     const register double tmp7_0 = A_01_0 + A_10_3;
1575     const register double tmp8_0 = A_01_1 + A_01_2 + A_10_1 + A_10_2;
1576     const register double tmp9_0 = A_01_0 + A_10_0;
1577 caltinay 3748 const register double tmp12_0 = A_11_0 + A_11_2;
1578     const register double tmp10_0 = A_01_3 + A_10_3;
1579     const register double tmp14_0 = A_01_0 + A_01_3 + A_10_0 + A_10_3;
1580     const register double tmp13_0 = A_01_2 + A_10_1;
1581     const register double tmp11_0 = A_11_1 + A_11_3;
1582     const register double tmp18_0 = A_01_1 + A_10_1;
1583     const register double tmp15_0 = A_01_1 + A_10_2;
1584     const register double tmp16_0 = A_10_0 + A_10_3;
1585     const register double tmp17_0 = A_01_1 + A_01_2;
1586     const register double tmp19_0 = A_01_2 + A_10_2;
1587 caltinay 3764 const register double tmp0_1 = A_10_3*w8;
1588     const register double tmp1_1 = tmp0_0*w1;
1589     const register double tmp2_1 = A_01_1*w4;
1590     const register double tmp3_1 = tmp1_0*w0;
1591     const register double tmp4_1 = A_01_2*w7;
1592     const register double tmp5_1 = tmp2_0*w3;
1593     const register double tmp6_1 = tmp3_0*w6;
1594     const register double tmp7_1 = A_10_0*w2;
1595     const register double tmp8_1 = tmp4_0*w5;
1596     const register double tmp9_1 = tmp2_0*w10;
1597 caltinay 3748 const register double tmp14_1 = A_10_0*w8;
1598     const register double tmp23_1 = tmp3_0*w14;
1599     const register double tmp35_1 = A_01_0*w8;
1600     const register double tmp54_1 = tmp13_0*w8;
1601     const register double tmp20_1 = tmp9_0*w4;
1602     const register double tmp25_1 = tmp12_0*w12;
1603     const register double tmp44_1 = tmp7_0*w7;
1604     const register double tmp26_1 = tmp10_0*w4;
1605     const register double tmp52_1 = tmp18_0*w8;
1606     const register double tmp48_1 = A_10_1*w7;
1607     const register double tmp46_1 = A_01_3*w8;
1608     const register double tmp50_1 = A_01_0*w2;
1609     const register double tmp56_1 = tmp19_0*w8;
1610     const register double tmp19_1 = A_10_3*w2;
1611     const register double tmp47_1 = A_10_2*w4;
1612     const register double tmp16_1 = tmp3_0*w0;
1613     const register double tmp18_1 = tmp1_0*w6;
1614     const register double tmp31_1 = tmp11_0*w12;
1615     const register double tmp55_1 = tmp15_0*w2;
1616     const register double tmp39_1 = A_10_2*w7;
1617     const register double tmp11_1 = tmp6_0*w7;
1618     const register double tmp40_1 = tmp11_0*w17;
1619     const register double tmp34_1 = tmp15_0*w8;
1620     const register double tmp33_1 = tmp14_0*w5;
1621     const register double tmp24_1 = tmp11_0*w13;
1622     const register double tmp43_1 = tmp17_0*w5;
1623     const register double tmp15_1 = A_01_2*w4;
1624     const register double tmp53_1 = tmp19_0*w2;
1625     const register double tmp27_1 = tmp3_0*w11;
1626     const register double tmp32_1 = tmp13_0*w2;
1627     const register double tmp10_1 = tmp5_0*w9;
1628     const register double tmp37_1 = A_10_1*w4;
1629     const register double tmp38_1 = tmp5_0*w15;
1630     const register double tmp17_1 = A_01_1*w7;
1631     const register double tmp12_1 = tmp7_0*w4;
1632     const register double tmp22_1 = tmp10_0*w7;
1633     const register double tmp57_1 = tmp18_0*w2;
1634     const register double tmp28_1 = tmp9_0*w7;
1635     const register double tmp29_1 = tmp1_0*w14;
1636     const register double tmp51_1 = tmp11_0*w16;
1637     const register double tmp42_1 = tmp12_0*w16;
1638     const register double tmp49_1 = tmp12_0*w17;
1639     const register double tmp21_1 = tmp1_0*w11;
1640     const register double tmp45_1 = tmp6_0*w4;
1641     const register double tmp13_1 = tmp8_0*w1;
1642     const register double tmp36_1 = tmp16_0*w1;
1643     const register double tmp41_1 = A_01_3*w2;
1644     const register double tmp30_1 = tmp12_0*w13;
1645     EM_S[INDEX2(0,0,4)]+=tmp13_1 + tmp20_1 + tmp21_1 + tmp22_1 + tmp23_1 + tmp24_1 + tmp25_1;
1646 caltinay 3764 EM_S[INDEX2(1,0,4)]+=tmp36_1 + tmp37_1 + tmp39_1 + tmp3_1 + tmp43_1 + tmp46_1 + tmp50_1 + tmp5_1 + tmp6_1;
1647     EM_S[INDEX2(2,0,4)]+=tmp0_1 + tmp15_1 + tmp17_1 + tmp1_1 + tmp38_1 + tmp49_1 + tmp51_1 + tmp7_1 + tmp8_1;
1648 caltinay 3748 EM_S[INDEX2(3,0,4)]+=tmp10_1 + tmp32_1 + tmp33_1 + tmp34_1 + tmp9_1;
1649 caltinay 3764 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1 + tmp4_1 + tmp5_1 + tmp6_1 + tmp7_1 + tmp8_1;
1650     EM_S[INDEX2(1,1,4)]+=tmp21_1 + tmp23_1 + tmp30_1 + tmp31_1 + tmp33_1 + tmp56_1 + tmp57_1;
1651     EM_S[INDEX2(2,1,4)]+=tmp10_1 + tmp13_1 + tmp44_1 + tmp45_1 + tmp9_1;
1652 caltinay 3748 EM_S[INDEX2(3,1,4)]+=tmp35_1 + tmp36_1 + tmp37_1 + tmp38_1 + tmp39_1 + tmp40_1 + tmp41_1 + tmp42_1 + tmp43_1;
1653     EM_S[INDEX2(0,2,4)]+=tmp36_1 + tmp38_1 + tmp43_1 + tmp46_1 + tmp47_1 + tmp48_1 + tmp49_1 + tmp50_1 + tmp51_1;
1654 caltinay 3764 EM_S[INDEX2(1,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1 + tmp9_1;
1655     EM_S[INDEX2(2,2,4)]+=tmp24_1 + tmp25_1 + tmp27_1 + tmp29_1 + tmp33_1 + tmp52_1 + tmp53_1;
1656     EM_S[INDEX2(3,2,4)]+=tmp14_1 + tmp15_1 + tmp16_1 + tmp17_1 + tmp18_1 + tmp19_1 + tmp1_1 + tmp5_1 + tmp8_1;
1657     EM_S[INDEX2(0,3,4)]+=tmp10_1 + tmp33_1 + tmp54_1 + tmp55_1 + tmp9_1;
1658 caltinay 3748 EM_S[INDEX2(1,3,4)]+=tmp14_1 + tmp19_1 + tmp1_1 + tmp2_1 + tmp38_1 + tmp40_1 + tmp42_1 + tmp4_1 + tmp8_1;
1659     EM_S[INDEX2(2,3,4)]+=tmp16_1 + tmp18_1 + tmp35_1 + tmp36_1 + tmp41_1 + tmp43_1 + tmp47_1 + tmp48_1 + tmp5_1;
1660 caltinay 3764 EM_S[INDEX2(3,3,4)]+=tmp13_1 + tmp26_1 + tmp27_1 + tmp28_1 + tmp29_1 + tmp30_1 + tmp31_1;
1661     } else { // constant data
1662 caltinay 3748 const register double A_00 = A_p[INDEX2(0,0,2)];
1663 caltinay 3764 const register double A_10 = A_p[INDEX2(1,0,2)];
1664 caltinay 3748 const register double A_01 = A_p[INDEX2(0,1,2)];
1665     const register double A_11 = A_p[INDEX2(1,1,2)];
1666     const register double tmp0_0 = A_01 + A_10;
1667     const register double tmp0_1 = A_00*w18;
1668 caltinay 3764 const register double tmp1_1 = A_01*w19;
1669     const register double tmp2_1 = A_10*w20;
1670     const register double tmp3_1 = A_11*w21;
1671 caltinay 3748 const register double tmp4_1 = A_00*w22;
1672 caltinay 3764 const register double tmp5_1 = tmp0_0*w19;
1673     const register double tmp6_1 = A_11*w23;
1674     const register double tmp7_1 = A_11*w25;
1675 caltinay 3748 const register double tmp8_1 = A_00*w24;
1676     const register double tmp9_1 = tmp0_0*w20;
1677 caltinay 3764 const register double tmp10_1 = A_01*w20;
1678 caltinay 3748 const register double tmp11_1 = A_11*w27;
1679 caltinay 3764 const register double tmp12_1 = A_00*w26;
1680     const register double tmp13_1 = A_10*w19;
1681 caltinay 3748 EM_S[INDEX2(0,0,4)]+=tmp5_1 + tmp7_1 + tmp8_1;
1682 caltinay 3764 EM_S[INDEX2(1,0,4)]+=tmp0_1 + tmp10_1 + tmp13_1 + tmp3_1;
1683     EM_S[INDEX2(2,0,4)]+=tmp11_1 + tmp12_1 + tmp1_1 + tmp2_1;
1684 caltinay 3748 EM_S[INDEX2(3,0,4)]+=tmp4_1 + tmp6_1 + tmp9_1;
1685 caltinay 3764 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1;
1686     EM_S[INDEX2(1,1,4)]+=tmp7_1 + tmp8_1 + tmp9_1;
1687     EM_S[INDEX2(2,1,4)]+=tmp4_1 + tmp5_1 + tmp6_1;
1688 caltinay 3748 EM_S[INDEX2(3,1,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1;
1689     EM_S[INDEX2(0,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1;
1690 caltinay 3764 EM_S[INDEX2(1,2,4)]+=tmp4_1 + tmp5_1 + tmp6_1;
1691     EM_S[INDEX2(2,2,4)]+=tmp7_1 + tmp8_1 + tmp9_1;
1692     EM_S[INDEX2(3,2,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1;
1693     EM_S[INDEX2(0,3,4)]+=tmp4_1 + tmp6_1 + tmp9_1;
1694 caltinay 3748 EM_S[INDEX2(1,3,4)]+=tmp11_1 + tmp12_1 + tmp1_1 + tmp2_1;
1695     EM_S[INDEX2(2,3,4)]+=tmp0_1 + tmp10_1 + tmp13_1 + tmp3_1;
1696 caltinay 3764 EM_S[INDEX2(3,3,4)]+=tmp5_1 + tmp7_1 + tmp8_1;
1697 caltinay 3748 }
1698     }
1699     ///////////////
1700     // process B //
1701     ///////////////
1702     if (!B.isEmpty()) {
1703     add_EM_S=true;
1704     const double* B_p=const_cast<escript::Data*>(&B)->getSampleDataRO(e);
1705     if (B.actsExpanded()) {
1706     const register double B_0_0 = B_p[INDEX2(0,0,2)];
1707     const register double B_1_0 = B_p[INDEX2(1,0,2)];
1708     const register double B_0_1 = B_p[INDEX2(0,1,2)];
1709     const register double B_1_1 = B_p[INDEX2(1,1,2)];
1710     const register double B_0_2 = B_p[INDEX2(0,2,2)];
1711     const register double B_1_2 = B_p[INDEX2(1,2,2)];
1712     const register double B_0_3 = B_p[INDEX2(0,3,2)];
1713     const register double B_1_3 = B_p[INDEX2(1,3,2)];
1714 caltinay 3764 const register double tmp0_0 = B_1_0 + B_1_1;
1715 caltinay 3748 const register double tmp1_0 = B_1_2 + B_1_3;
1716     const register double tmp2_0 = B_0_1 + B_0_3;
1717 caltinay 3764 const register double tmp3_0 = B_0_0 + B_0_2;
1718 caltinay 3748 const register double tmp63_1 = B_1_1*w42;
1719     const register double tmp79_1 = B_1_1*w40;
1720     const register double tmp37_1 = tmp3_0*w35;
1721     const register double tmp8_1 = tmp0_0*w32;
1722     const register double tmp71_1 = B_0_1*w34;
1723     const register double tmp19_1 = B_0_3*w31;
1724     const register double tmp15_1 = B_0_3*w34;
1725     const register double tmp9_1 = tmp3_0*w34;
1726     const register double tmp35_1 = B_1_0*w36;
1727     const register double tmp66_1 = B_0_3*w28;
1728     const register double tmp28_1 = B_1_0*w42;
1729     const register double tmp22_1 = B_1_0*w40;
1730     const register double tmp16_1 = B_1_2*w29;
1731     const register double tmp6_1 = tmp2_0*w35;
1732     const register double tmp55_1 = B_1_3*w40;
1733     const register double tmp50_1 = B_1_3*w42;
1734     const register double tmp7_1 = tmp1_0*w29;
1735     const register double tmp1_1 = tmp1_0*w32;
1736     const register double tmp57_1 = B_0_3*w30;
1737     const register double tmp18_1 = B_1_1*w32;
1738     const register double tmp53_1 = B_1_0*w41;
1739     const register double tmp61_1 = B_1_3*w36;
1740     const register double tmp27_1 = B_0_3*w38;
1741     const register double tmp64_1 = B_0_2*w30;
1742     const register double tmp76_1 = B_0_1*w38;
1743     const register double tmp39_1 = tmp2_0*w34;
1744     const register double tmp62_1 = B_0_1*w31;
1745     const register double tmp56_1 = B_0_0*w31;
1746     const register double tmp49_1 = B_1_1*w36;
1747     const register double tmp2_1 = B_0_2*w31;
1748     const register double tmp23_1 = B_0_2*w33;
1749     const register double tmp38_1 = B_1_1*w43;
1750     const register double tmp74_1 = B_1_2*w41;
1751     const register double tmp43_1 = B_1_1*w41;
1752     const register double tmp58_1 = B_0_2*w28;
1753     const register double tmp67_1 = B_0_0*w33;
1754     const register double tmp33_1 = tmp0_0*w39;
1755     const register double tmp4_1 = B_0_0*w28;
1756     const register double tmp20_1 = B_0_0*w30;
1757     const register double tmp13_1 = B_0_2*w38;
1758     const register double tmp65_1 = B_1_2*w43;
1759     const register double tmp0_1 = tmp0_0*w29;
1760     const register double tmp41_1 = tmp3_0*w33;
1761     const register double tmp73_1 = B_0_2*w37;
1762     const register double tmp69_1 = B_0_0*w38;
1763     const register double tmp48_1 = B_1_2*w39;
1764     const register double tmp59_1 = B_0_1*w33;
1765     const register double tmp17_1 = B_1_3*w41;
1766     const register double tmp5_1 = B_0_3*w33;
1767     const register double tmp3_1 = B_0_1*w30;
1768     const register double tmp21_1 = B_0_1*w28;
1769     const register double tmp42_1 = B_1_0*w29;
1770     const register double tmp54_1 = B_1_2*w32;
1771     const register double tmp60_1 = B_1_0*w39;
1772     const register double tmp32_1 = tmp1_0*w36;
1773     const register double tmp10_1 = B_0_1*w37;
1774     const register double tmp14_1 = B_0_0*w35;
1775     const register double tmp29_1 = B_0_1*w35;
1776     const register double tmp26_1 = B_1_2*w36;
1777     const register double tmp30_1 = B_1_3*w43;
1778     const register double tmp70_1 = B_0_2*w35;
1779     const register double tmp34_1 = B_1_3*w39;
1780     const register double tmp51_1 = B_1_0*w43;
1781     const register double tmp31_1 = B_0_2*w34;
1782     const register double tmp45_1 = tmp3_0*w28;
1783     const register double tmp11_1 = tmp1_0*w39;
1784     const register double tmp52_1 = B_1_1*w29;
1785     const register double tmp44_1 = B_1_3*w32;
1786     const register double tmp25_1 = B_1_1*w39;
1787     const register double tmp47_1 = tmp2_0*w33;
1788     const register double tmp72_1 = B_1_3*w29;
1789     const register double tmp40_1 = tmp2_0*w28;
1790     const register double tmp46_1 = B_1_2*w40;
1791     const register double tmp36_1 = B_1_2*w42;
1792     const register double tmp24_1 = B_0_0*w37;
1793     const register double tmp77_1 = B_0_3*w35;
1794     const register double tmp68_1 = B_0_3*w37;
1795     const register double tmp78_1 = B_0_0*w34;
1796     const register double tmp12_1 = tmp0_0*w36;
1797     const register double tmp75_1 = B_1_0*w32;
1798     EM_S[INDEX2(0,0,4)]+=tmp16_1 + tmp17_1 + tmp18_1 + tmp19_1 + tmp20_1 + tmp21_1 + tmp22_1 + tmp23_1;
1799 caltinay 3764 EM_S[INDEX2(1,0,4)]+=tmp0_1 + tmp1_1 + tmp68_1 + tmp69_1 + tmp70_1 + tmp71_1;
1800     EM_S[INDEX2(2,0,4)]+=tmp45_1 + tmp47_1 + tmp48_1 + tmp49_1 + tmp50_1 + tmp51_1;
1801 caltinay 3748 EM_S[INDEX2(3,0,4)]+=tmp32_1 + tmp33_1 + tmp6_1 + tmp9_1;
1802 caltinay 3764 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1 + tmp4_1 + tmp5_1;
1803     EM_S[INDEX2(1,1,4)]+=tmp72_1 + tmp73_1 + tmp74_1 + tmp75_1 + tmp76_1 + tmp77_1 + tmp78_1 + tmp79_1;
1804     EM_S[INDEX2(2,1,4)]+=tmp32_1 + tmp33_1 + tmp40_1 + tmp41_1;
1805 caltinay 3748 EM_S[INDEX2(3,1,4)]+=tmp34_1 + tmp35_1 + tmp36_1 + tmp37_1 + tmp38_1 + tmp39_1;
1806     EM_S[INDEX2(0,2,4)]+=tmp42_1 + tmp43_1 + tmp44_1 + tmp45_1 + tmp46_1 + tmp47_1;
1807 caltinay 3764 EM_S[INDEX2(1,2,4)]+=tmp6_1 + tmp7_1 + tmp8_1 + tmp9_1;
1808     EM_S[INDEX2(2,2,4)]+=tmp60_1 + tmp61_1 + tmp62_1 + tmp63_1 + tmp64_1 + tmp65_1 + tmp66_1 + tmp67_1;
1809     EM_S[INDEX2(3,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1 + tmp14_1 + tmp15_1;
1810     EM_S[INDEX2(0,3,4)]+=tmp40_1 + tmp41_1 + tmp7_1 + tmp8_1;
1811 caltinay 3748 EM_S[INDEX2(1,3,4)]+=tmp37_1 + tmp39_1 + tmp52_1 + tmp53_1 + tmp54_1 + tmp55_1;
1812     EM_S[INDEX2(2,3,4)]+=tmp11_1 + tmp12_1 + tmp56_1 + tmp57_1 + tmp58_1 + tmp59_1;
1813 caltinay 3764 EM_S[INDEX2(3,3,4)]+=tmp24_1 + tmp25_1 + tmp26_1 + tmp27_1 + tmp28_1 + tmp29_1 + tmp30_1 + tmp31_1;
1814     } else { // constant data
1815 caltinay 3748 const register double B_0 = B_p[0];
1816     const register double B_1 = B_p[1];
1817 caltinay 3764 const register double tmp0_1 = B_0*w44;
1818 caltinay 3748 const register double tmp1_1 = B_1*w45;
1819 caltinay 3764 const register double tmp2_1 = B_0*w46;
1820     const register double tmp3_1 = B_0*w47;
1821     const register double tmp4_1 = B_1*w48;
1822 caltinay 3748 const register double tmp5_1 = B_1*w49;
1823 caltinay 3764 const register double tmp6_1 = B_1*w50;
1824 caltinay 3748 const register double tmp7_1 = B_0*w51;
1825     EM_S[INDEX2(0,0,4)]+=tmp0_1 + tmp5_1;
1826 caltinay 3764 EM_S[INDEX2(1,0,4)]+=tmp1_1 + tmp3_1;
1827     EM_S[INDEX2(2,0,4)]+=tmp6_1 + tmp7_1;
1828 caltinay 3748 EM_S[INDEX2(3,0,4)]+=tmp2_1 + tmp4_1;
1829 caltinay 3764 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1;
1830     EM_S[INDEX2(1,1,4)]+=tmp3_1 + tmp5_1;
1831     EM_S[INDEX2(2,1,4)]+=tmp4_1 + tmp7_1;
1832 caltinay 3748 EM_S[INDEX2(3,1,4)]+=tmp2_1 + tmp6_1;
1833     EM_S[INDEX2(0,2<