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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 3757 - (hide annotations)
Fri Jan 6 04:48:27 2012 UTC (7 years, 10 months ago) by caltinay
Original Path: branches/ripleygmg_from_3668/ripley/src/Rectangle.cpp
File size: 104957 byte(s)
Implemented ownSample() for DOF and Elements.

1 caltinay 3691
2     /*******************************************************
3     *
4     * Copyright (c) 2003-2011 by University of Queensland
5     * 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     // local number of elements (including overlap)
53     m_NE0 = (m_NX>1 ? (n0+1)/m_NX : n0);
54     if (m_mpiInfo->rank%m_NX>0 && m_mpiInfo->rank%m_NX<m_NX-1)
55     m_NE0++;
56     m_NE1 = (m_NY>1 ? (n1+1)/m_NY : n1);
57     if (m_mpiInfo->rank/m_NX>0 && m_mpiInfo->rank/m_NX<m_NY-1)
58     m_NE1++;
59    
60     // local number of nodes
61 caltinay 3691 m_N0 = m_NE0+1;
62     m_N1 = m_NE1+1;
63 caltinay 3752
64 caltinay 3691 // bottom-left node is at (offset0,offset1) in global mesh
65 caltinay 3752 m_offset0 = (n0+1)/m_NX*(m_mpiInfo->rank%m_NX);
66 caltinay 3753 if (m_offset0 > 0)
67 caltinay 3752 m_offset0--;
68     m_offset1 = (n1+1)/m_NY*(m_mpiInfo->rank/m_NX);
69 caltinay 3753 if (m_offset1 > 0)
70 caltinay 3752 m_offset1--;
71    
72 caltinay 3691 populateSampleIds();
73 caltinay 3756 createPattern();
74 caltinay 3691 }
75    
76     Rectangle::~Rectangle()
77     {
78     }
79    
80     string Rectangle::getDescription() const
81     {
82     return "ripley::Rectangle";
83     }
84    
85     bool Rectangle::operator==(const AbstractDomain& other) const
86     {
87 caltinay 3744 const Rectangle* o=dynamic_cast<const Rectangle*>(&other);
88     if (o) {
89     return (RipleyDomain::operator==(other) &&
90     m_gNE0==o->m_gNE0 && m_gNE1==o->m_gNE1
91     && m_l0==o->m_l0 && m_l1==o->m_l1
92     && m_NX==o->m_NX && m_NY==o->m_NY);
93     }
94 caltinay 3691
95     return false;
96     }
97    
98     void Rectangle::dump(const string& fileName) const
99     {
100     #if USE_SILO
101     string fn(fileName);
102     if (fileName.length() < 6 || fileName.compare(fileName.length()-5, 5, ".silo") != 0) {
103     fn+=".silo";
104     }
105    
106     const int NUM_SILO_FILES = 1;
107     const char* blockDirFmt = "/block%04d";
108     int driver=DB_HDF5;
109     string siloPath;
110     DBfile* dbfile = NULL;
111    
112     #ifdef ESYS_MPI
113     PMPIO_baton_t* baton = NULL;
114     #endif
115    
116     if (m_mpiInfo->size > 1) {
117     #ifdef ESYS_MPI
118     baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
119     0x1337, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
120     PMPIO_DefaultClose, (void*)&driver);
121     // try the fallback driver in case of error
122     if (!baton && driver != DB_PDB) {
123     driver = DB_PDB;
124     baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
125     0x1338, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
126     PMPIO_DefaultClose, (void*)&driver);
127     }
128     if (baton) {
129     char str[64];
130     snprintf(str, 64, blockDirFmt, PMPIO_RankInGroup(baton, m_mpiInfo->rank));
131     siloPath = str;
132     dbfile = (DBfile*) PMPIO_WaitForBaton(baton, fn.c_str(), siloPath.c_str());
133     }
134     #endif
135     } else {
136     dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
137     getDescription().c_str(), driver);
138     // try the fallback driver in case of error
139     if (!dbfile && driver != DB_PDB) {
140     driver = DB_PDB;
141     dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
142     getDescription().c_str(), driver);
143     }
144     }
145    
146     if (!dbfile)
147     throw RipleyException("dump: Could not create Silo file");
148    
149     /*
150     if (driver==DB_HDF5) {
151     // gzip level 1 already provides good compression with minimal
152     // performance penalty. Some tests showed that gzip levels >3 performed
153     // rather badly on escript data both in terms of time and space
154     DBSetCompression("ERRMODE=FALLBACK METHOD=GZIP LEVEL=1");
155     }
156     */
157    
158     boost::scoped_ptr<double> x(new double[m_N0]);
159     boost::scoped_ptr<double> y(new double[m_N1]);
160     double* coords[2] = { x.get(), y.get() };
161 caltinay 3697 pair<double,double> xdx = getFirstCoordAndSpacing(0);
162     pair<double,double> ydy = getFirstCoordAndSpacing(1);
163 caltinay 3691 #pragma omp parallel
164     {
165 caltinay 3722 #pragma omp for nowait
166 caltinay 3691 for (dim_t i0 = 0; i0 < m_N0; i0++) {
167 caltinay 3697 coords[0][i0]=xdx.first+i0*xdx.second;
168 caltinay 3691 }
169 caltinay 3722 #pragma omp for nowait
170 caltinay 3691 for (dim_t i1 = 0; i1 < m_N1; i1++) {
171 caltinay 3697 coords[1][i1]=ydy.first+i1*ydy.second;
172 caltinay 3691 }
173     }
174 caltinay 3697 IndexVector dims = getNumNodesPerDim();
175    
176     // write mesh
177     DBPutQuadmesh(dbfile, "mesh", NULL, coords, &dims[0], 2, DB_DOUBLE,
178 caltinay 3691 DB_COLLINEAR, NULL);
179    
180 caltinay 3697 // write node ids
181     DBPutQuadvar1(dbfile, "nodeId", "mesh", (void*)&m_nodeId[0], &dims[0], 2,
182     NULL, 0, DB_INT, DB_NODECENT, NULL);
183    
184     // write element ids
185     dims = getNumElementsPerDim();
186     DBPutQuadvar1(dbfile, "elementId", "mesh", (void*)&m_elementId[0],
187     &dims[0], 2, NULL, 0, DB_INT, DB_ZONECENT, NULL);
188    
189     // rank 0 writes multimesh and multivar
190 caltinay 3691 if (m_mpiInfo->rank == 0) {
191     vector<string> tempstrings;
192 caltinay 3697 vector<char*> names;
193 caltinay 3691 for (dim_t i=0; i<m_mpiInfo->size; i++) {
194     stringstream path;
195     path << "/block" << setw(4) << setfill('0') << right << i << "/mesh";
196     tempstrings.push_back(path.str());
197 caltinay 3697 names.push_back((char*)tempstrings.back().c_str());
198 caltinay 3691 }
199 caltinay 3697 vector<int> types(m_mpiInfo->size, DB_QUAD_RECT);
200 caltinay 3691 DBSetDir(dbfile, "/");
201 caltinay 3697 DBPutMultimesh(dbfile, "multimesh", m_mpiInfo->size, &names[0],
202     &types[0], NULL);
203     tempstrings.clear();
204     names.clear();
205     for (dim_t i=0; i<m_mpiInfo->size; i++) {
206     stringstream path;
207     path << "/block" << setw(4) << setfill('0') << right << i << "/nodeId";
208     tempstrings.push_back(path.str());
209     names.push_back((char*)tempstrings.back().c_str());
210     }
211     types.assign(m_mpiInfo->size, DB_QUADVAR);
212     DBPutMultivar(dbfile, "nodeId", 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 << "/elementId";
219     tempstrings.push_back(path.str());
220     names.push_back((char*)tempstrings.back().c_str());
221     }
222     DBPutMultivar(dbfile, "elementId", m_mpiInfo->size, &names[0],
223     &types[0], NULL);
224 caltinay 3691 }
225    
226     if (m_mpiInfo->size > 1) {
227     #ifdef ESYS_MPI
228     PMPIO_HandOffBaton(baton, dbfile);
229     PMPIO_Finish(baton);
230     #endif
231     } else {
232     DBClose(dbfile);
233     }
234    
235     #else // USE_SILO
236     throw RipleyException("dump(): no Silo support");
237     #endif
238     }
239    
240 caltinay 3697 const int* Rectangle::borrowSampleReferenceIDs(int fsType) const
241 caltinay 3691 {
242 caltinay 3697 switch (fsType) {
243 caltinay 3691 case Nodes:
244 caltinay 3748 case ReducedNodes: //FIXME: reduced
245 caltinay 3691 return &m_nodeId[0];
246 caltinay 3750 case DegreesOfFreedom:
247     case ReducedDegreesOfFreedom: //FIXME: reduced
248     return &m_dofId[0];
249 caltinay 3691 case Elements:
250 caltinay 3733 case ReducedElements:
251 caltinay 3691 return &m_elementId[0];
252     case FaceElements:
253 caltinay 3733 case ReducedFaceElements:
254 caltinay 3691 return &m_faceId[0];
255     default:
256     break;
257     }
258    
259     stringstream msg;
260     msg << "borrowSampleReferenceIDs() not implemented for function space type "
261 caltinay 3701 << functionSpaceTypeAsString(fsType);
262 caltinay 3691 throw RipleyException(msg.str());
263     }
264    
265 caltinay 3757 bool Rectangle::ownSample(int fsType, index_t id) const
266 caltinay 3691 {
267     #ifdef ESYS_MPI
268 caltinay 3757 switch (fsType) {
269     case Nodes:
270     case ReducedNodes: //FIXME: reduced
271     return (m_dofMap[id] < getNumDOF());
272     case DegreesOfFreedom:
273     case ReducedDegreesOfFreedom:
274     return true;
275     case Elements:
276     case ReducedElements:
277     // check ownership of element's bottom left node
278     return (m_dofMap[id%m_NE0+m_N0*(id/m_NE0)] < getNumDOF());
279     case FaceElements:
280     case ReducedFaceElements:
281     default:
282     break;
283 caltinay 3702 }
284 caltinay 3757
285     stringstream msg;
286     msg << "ownSample() not implemented for "
287     << functionSpaceTypeAsString(fsType);
288     throw RipleyException(msg.str());
289 caltinay 3691 #else
290     return true;
291     #endif
292     }
293    
294 caltinay 3702 void Rectangle::setToGradient(escript::Data& out, const escript::Data& cIn) const
295 caltinay 3691 {
296 caltinay 3702 escript::Data& in = *const_cast<escript::Data*>(&cIn);
297 caltinay 3701 const dim_t numComp = in.getDataPointSize();
298 caltinay 3702 const double h0 = m_l0/m_gNE0;
299     const double h1 = m_l1/m_gNE1;
300 caltinay 3724 const double cx0 = -1./h0;
301     const double cx1 = -.78867513459481288225/h0;
302     const double cx2 = -.5/h0;
303     const double cx3 = -.21132486540518711775/h0;
304     const double cx4 = .21132486540518711775/h0;
305     const double cx5 = .5/h0;
306     const double cx6 = .78867513459481288225/h0;
307     const double cx7 = 1./h0;
308     const double cy0 = -1./h1;
309     const double cy1 = -.78867513459481288225/h1;
310     const double cy2 = -.5/h1;
311     const double cy3 = -.21132486540518711775/h1;
312     const double cy4 = .21132486540518711775/h1;
313     const double cy5 = .5/h1;
314     const double cy6 = .78867513459481288225/h1;
315     const double cy7 = 1./h1;
316    
317 caltinay 3702 if (out.getFunctionSpace().getTypeCode() == Elements) {
318 caltinay 3724 /*** GENERATOR SNIP_GRAD_ELEMENTS TOP */
319 caltinay 3697 #pragma omp parallel for
320 caltinay 3707 for (index_t k1=0; k1 < m_NE1; ++k1) {
321     for (index_t k0=0; k0 < m_NE0; ++k0) {
322 caltinay 3702 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
323     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
324     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
325     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
326     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
327     for (index_t i=0; i < numComp; ++i) {
328 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;
329     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
330     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
331     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
332     o[INDEX3(i,0,2,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
333     o[INDEX3(i,1,2,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
334     o[INDEX3(i,0,3,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
335     o[INDEX3(i,1,3,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
336 caltinay 3702 } /* end of component loop i */
337     } /* end of k0 loop */
338     } /* end of k1 loop */
339     /* GENERATOR SNIP_GRAD_ELEMENTS BOTTOM */
340 caltinay 3711 } else if (out.getFunctionSpace().getTypeCode() == ReducedElements) {
341 caltinay 3724 /*** GENERATOR SNIP_GRAD_REDUCED_ELEMENTS TOP */
342 caltinay 3711 #pragma omp parallel for
343     for (index_t k1=0; k1 < m_NE1; ++k1) {
344     for (index_t k0=0; k0 < m_NE0; ++k0) {
345     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
346     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
347     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
348     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
349     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
350     for (index_t i=0; i < numComp; ++i) {
351 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
352     o[INDEX3(i,1,0,numComp,2)] = cy2*(f_00[i] + f_10[i]) + cy5*(f_01[i] + f_11[i]);
353 caltinay 3711 } /* end of component loop i */
354     } /* end of k0 loop */
355     } /* end of k1 loop */
356     /* GENERATOR SNIP_GRAD_REDUCED_ELEMENTS BOTTOM */
357 caltinay 3707 } else if (out.getFunctionSpace().getTypeCode() == FaceElements) {
358 caltinay 3722 #pragma omp parallel
359     {
360 caltinay 3724 /*** GENERATOR SNIP_GRAD_FACES TOP */
361 caltinay 3722 if (m_faceOffset[0] > -1) {
362     #pragma omp for nowait
363     for (index_t k1=0; k1 < m_NE1; ++k1) {
364     const register double* f_10 = in.getSampleDataRO(INDEX2(1,k1, m_N0));
365     const register double* f_11 = in.getSampleDataRO(INDEX2(1,k1+1, m_N0));
366     const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
367     const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
368     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
369     for (index_t i=0; i < numComp; ++i) {
370 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;
371     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
372     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
373     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
374 caltinay 3722 } /* end of component loop i */
375     } /* end of k1 loop */
376     } /* end of face 0 */
377     if (m_faceOffset[1] > -1) {
378     #pragma omp for nowait
379     for (index_t k1=0; k1 < m_NE1; ++k1) {
380     const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
381     const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
382     const register double* f_01 = in.getSampleDataRO(INDEX2(m_N0-2,k1+1, m_N0));
383     const register double* f_00 = in.getSampleDataRO(INDEX2(m_N0-2,k1, m_N0));
384     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
385     for (index_t i=0; i < numComp; ++i) {
386 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;
387     o[INDEX3(i,1,0,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
388     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
389     o[INDEX3(i,1,1,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
390 caltinay 3722 } /* end of component loop i */
391     } /* end of k1 loop */
392     } /* end of face 1 */
393     if (m_faceOffset[2] > -1) {
394     #pragma omp for nowait
395     for (index_t k0=0; k0 < m_NE0; ++k0) {
396     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
397     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
398     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,1, m_N0));
399     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,1, m_N0));
400     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
401     for (index_t i=0; i < numComp; ++i) {
402 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
403     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
404     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
405     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
406 caltinay 3722 } /* end of component loop i */
407     } /* end of k0 loop */
408     } /* end of face 2 */
409     if (m_faceOffset[3] > -1) {
410     #pragma omp for nowait
411     for (index_t k0=0; k0 < m_NE0; ++k0) {
412     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
413     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
414     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,m_N1-2, m_N0));
415     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,m_N1-2, m_N0));
416     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
417     for (index_t i=0; i < numComp; ++i) {
418 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
419     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
420     o[INDEX3(i,0,1,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
421     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
422 caltinay 3722 } /* end of component loop i */
423     } /* end of k0 loop */
424     } /* end of face 3 */
425     /* GENERATOR SNIP_GRAD_FACES BOTTOM */
426     } // end of parallel section
427 caltinay 3711 } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
428 caltinay 3722 #pragma omp parallel
429     {
430 caltinay 3724 /*** GENERATOR SNIP_GRAD_REDUCED_FACES TOP */
431 caltinay 3722 if (m_faceOffset[0] > -1) {
432     #pragma omp for nowait
433     for (index_t k1=0; k1 < m_NE1; ++k1) {
434     const register double* f_10 = in.getSampleDataRO(INDEX2(1,k1, m_N0));
435     const register double* f_11 = in.getSampleDataRO(INDEX2(1,k1+1, m_N0));
436     const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
437     const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
438     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
439     for (index_t i=0; i < numComp; ++i) {
440 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
441     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
442 caltinay 3722 } /* end of component loop i */
443     } /* end of k1 loop */
444     } /* end of face 0 */
445     if (m_faceOffset[1] > -1) {
446     #pragma omp for nowait
447     for (index_t k1=0; k1 < m_NE1; ++k1) {
448     const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
449     const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
450     const register double* f_01 = in.getSampleDataRO(INDEX2(m_N0-2,k1+1, m_N0));
451     const register double* f_00 = in.getSampleDataRO(INDEX2(m_N0-2,k1, m_N0));
452     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
453     for (index_t i=0; i < numComp; ++i) {
454 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
455     o[INDEX3(i,1,0,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
456 caltinay 3722 } /* end of component loop i */
457     } /* end of k1 loop */
458     } /* end of face 1 */
459     if (m_faceOffset[2] > -1) {
460     #pragma omp for nowait
461     for (index_t k0=0; k0 < m_NE0; ++k0) {
462     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
463     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
464     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,1, m_N0));
465     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,1, m_N0));
466     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
467     for (index_t i=0; i < numComp; ++i) {
468 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
469     o[INDEX3(i,1,0,numComp,2)] = cy2*(f_00[i] + f_10[i]) + cy5*(f_01[i] + f_11[i]);
470 caltinay 3722 } /* end of component loop i */
471     } /* end of k0 loop */
472     } /* end of face 2 */
473     if (m_faceOffset[3] > -1) {
474     #pragma omp for nowait
475     for (index_t k0=0; k0 < m_NE0; ++k0) {
476     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
477     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
478     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,m_N1-2, m_N0));
479     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,m_N1-2, m_N0));
480     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
481     for (index_t i=0; i < numComp; ++i) {
482 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
483     o[INDEX3(i,1,0,numComp,2)] = cy5*(f_01[i] + f_11[i]) + cy2*(f_00[i] + f_10[i]);
484 caltinay 3722 } /* end of component loop i */
485     } /* end of k0 loop */
486     } /* end of face 3 */
487     /* GENERATOR SNIP_GRAD_REDUCED_FACES BOTTOM */
488     } // end of parallel section
489 caltinay 3702 } else {
490 caltinay 3707 stringstream msg;
491     msg << "setToGradient() not implemented for "
492     << functionSpaceTypeAsString(out.getFunctionSpace().getTypeCode());
493     throw RipleyException(msg.str());
494 caltinay 3702 }
495 caltinay 3701 }
496 caltinay 3697
497 caltinay 3713 void Rectangle::setToIntegrals(vector<double>& integrals, const escript::Data& arg) const
498     {
499     escript::Data& in = *const_cast<escript::Data*>(&arg);
500     const dim_t numComp = in.getDataPointSize();
501     const double h0 = m_l0/m_gNE0;
502     const double h1 = m_l1/m_gNE1;
503     if (arg.getFunctionSpace().getTypeCode() == Elements) {
504     const double w_0 = h0*h1/4.;
505     #pragma omp parallel
506     {
507     vector<double> int_local(numComp, 0);
508 caltinay 3722 #pragma omp for nowait
509 caltinay 3713 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
510     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
511     const double* f = in.getSampleDataRO(INDEX2(k0, k1, m_NE0));
512     for (index_t i=0; i < numComp; ++i) {
513     const register double f_0 = f[INDEX2(i,0,numComp)];
514     const register double f_1 = f[INDEX2(i,1,numComp)];
515     const register double f_2 = f[INDEX2(i,2,numComp)];
516     const register double f_3 = f[INDEX2(i,3,numComp)];
517     int_local[i]+=(f_0+f_1+f_2+f_3)*w_0;
518     } /* end of component loop i */
519     } /* end of k0 loop */
520     } /* end of k1 loop */
521    
522     #pragma omp critical
523     for (index_t i=0; i<numComp; i++)
524     integrals[i]+=int_local[i];
525 caltinay 3722 } // end of parallel section
526 caltinay 3713 } else if (arg.getFunctionSpace().getTypeCode() == ReducedElements) {
527     const double w_0 = h0*h1;
528     #pragma omp parallel
529     {
530     vector<double> int_local(numComp, 0);
531 caltinay 3722 #pragma omp for nowait
532 caltinay 3713 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
533     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
534     const double* f = in.getSampleDataRO(INDEX2(k0, k1, m_NE0));
535     for (index_t i=0; i < numComp; ++i) {
536     int_local[i]+=f[i]*w_0;
537     } /* end of component loop i */
538     } /* end of k0 loop */
539     } /* end of k1 loop */
540    
541     #pragma omp critical
542     for (index_t i=0; i<numComp; i++)
543     integrals[i]+=int_local[i];
544 caltinay 3722 } // end of parallel section
545 caltinay 3713 } else if (arg.getFunctionSpace().getTypeCode() == FaceElements) {
546     const double w_0 = h0/2.;
547     const double w_1 = h1/2.;
548     #pragma omp parallel
549     {
550     vector<double> int_local(numComp, 0);
551     if (m_faceOffset[0] > -1) {
552 caltinay 3722 #pragma omp for nowait
553 caltinay 3713 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
554     const double* f = in.getSampleDataRO(m_faceOffset[0]+k1);
555     for (index_t i=0; i < numComp; ++i) {
556     const register double f_0 = f[INDEX2(i,0,numComp)];
557     const register double f_1 = f[INDEX2(i,1,numComp)];
558     int_local[i]+=(f_0+f_1)*w_1;
559     } /* end of component loop i */
560     } /* end of k1 loop */
561     }
562    
563     if (m_faceOffset[1] > -1) {
564 caltinay 3722 #pragma omp for nowait
565 caltinay 3713 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
566     const double* f = in.getSampleDataRO(m_faceOffset[1]+k1);
567     for (index_t i=0; i < numComp; ++i) {
568     const register double f_0 = f[INDEX2(i,0,numComp)];
569     const register double f_1 = f[INDEX2(i,1,numComp)];
570     int_local[i]+=(f_0+f_1)*w_1;
571     } /* end of component loop i */
572     } /* end of k1 loop */
573     }
574    
575     if (m_faceOffset[2] > -1) {
576 caltinay 3722 #pragma omp for nowait
577 caltinay 3713 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
578     const double* f = in.getSampleDataRO(m_faceOffset[2]+k0);
579     for (index_t i=0; i < numComp; ++i) {
580     const register double f_0 = f[INDEX2(i,0,numComp)];
581     const register double f_1 = f[INDEX2(i,1,numComp)];
582     int_local[i]+=(f_0+f_1)*w_0;
583     } /* end of component loop i */
584     } /* end of k0 loop */
585     }
586    
587     if (m_faceOffset[3] > -1) {
588 caltinay 3722 #pragma omp for nowait
589 caltinay 3713 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
590     const double* f = in.getSampleDataRO(m_faceOffset[3]+k0);
591     for (index_t i=0; i < numComp; ++i) {
592     const register double f_0 = f[INDEX2(i,0,numComp)];
593     const register double f_1 = f[INDEX2(i,1,numComp)];
594     int_local[i]+=(f_0+f_1)*w_0;
595     } /* end of component loop i */
596     } /* end of k0 loop */
597     }
598    
599     #pragma omp critical
600     for (index_t i=0; i<numComp; i++)
601     integrals[i]+=int_local[i];
602 caltinay 3722 } // end of parallel section
603 caltinay 3713 } else if (arg.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
604     #pragma omp parallel
605     {
606     vector<double> int_local(numComp, 0);
607     if (m_faceOffset[0] > -1) {
608 caltinay 3722 #pragma omp for nowait
609 caltinay 3713 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
610     const double* f = in.getSampleDataRO(m_faceOffset[0]+k1);
611     for (index_t i=0; i < numComp; ++i) {
612     int_local[i]+=f[i]*h1;
613     } /* end of component loop i */
614     } /* end of k1 loop */
615     }
616    
617     if (m_faceOffset[1] > -1) {
618 caltinay 3722 #pragma omp for nowait
619 caltinay 3713 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
620     const double* f = in.getSampleDataRO(m_faceOffset[1]+k1);
621     for (index_t i=0; i < numComp; ++i) {
622     int_local[i]+=f[i]*h1;
623     } /* end of component loop i */
624     } /* end of k1 loop */
625     }
626    
627     if (m_faceOffset[2] > -1) {
628 caltinay 3722 #pragma omp for nowait
629 caltinay 3713 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
630     const double* f = in.getSampleDataRO(m_faceOffset[2]+k0);
631     for (index_t i=0; i < numComp; ++i) {
632     int_local[i]+=f[i]*h0;
633     } /* end of component loop i */
634     } /* end of k0 loop */
635     }
636    
637     if (m_faceOffset[3] > -1) {
638 caltinay 3722 #pragma omp for nowait
639 caltinay 3713 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
640     const double* f = in.getSampleDataRO(m_faceOffset[3]+k0);
641     for (index_t i=0; i < numComp; ++i) {
642     int_local[i]+=f[i]*h0;
643     } /* end of component loop i */
644     } /* end of k0 loop */
645     }
646    
647     #pragma omp critical
648     for (index_t i=0; i<numComp; i++)
649     integrals[i]+=int_local[i];
650 caltinay 3722 } // end of parallel section
651 caltinay 3713 } else {
652     stringstream msg;
653     msg << "setToIntegrals() not implemented for "
654     << functionSpaceTypeAsString(arg.getFunctionSpace().getTypeCode());
655     throw RipleyException(msg.str());
656     }
657     }
658    
659 caltinay 3722 void Rectangle::setToNormal(escript::Data& out) const
660     {
661     if (out.getFunctionSpace().getTypeCode() == FaceElements) {
662     #pragma omp parallel
663     {
664     if (m_faceOffset[0] > -1) {
665     #pragma omp for nowait
666     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
667     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
668     // set vector at two quadrature points
669     *o++ = -1.;
670     *o++ = 0.;
671     *o++ = -1.;
672     *o = 0.;
673     }
674     }
675    
676     if (m_faceOffset[1] > -1) {
677     #pragma omp for nowait
678     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
679     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
680     // set vector at two quadrature points
681     *o++ = 1.;
682     *o++ = 0.;
683     *o++ = 1.;
684     *o = 0.;
685     }
686     }
687    
688     if (m_faceOffset[2] > -1) {
689     #pragma omp for nowait
690     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
691     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
692     // set vector at two quadrature points
693     *o++ = 0.;
694     *o++ = -1.;
695     *o++ = 0.;
696     *o = -1.;
697     }
698     }
699    
700     if (m_faceOffset[3] > -1) {
701     #pragma omp for nowait
702     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
703     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
704     // set vector at two quadrature points
705     *o++ = 0.;
706     *o++ = 1.;
707     *o++ = 0.;
708     *o = 1.;
709     }
710     }
711     } // end of parallel section
712     } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
713     #pragma omp parallel
714     {
715     if (m_faceOffset[0] > -1) {
716     #pragma omp for nowait
717     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
718     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
719     *o++ = -1.;
720     *o = 0.;
721     }
722     }
723    
724     if (m_faceOffset[1] > -1) {
725     #pragma omp for nowait
726     for (index_t k1 = 0; k1 < m_NE1; ++k1) {
727     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
728     *o++ = 1.;
729     *o = 0.;
730     }
731     }
732    
733     if (m_faceOffset[2] > -1) {
734     #pragma omp for nowait
735     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
736     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
737     *o++ = 0.;
738     *o = -1.;
739     }
740     }
741    
742     if (m_faceOffset[3] > -1) {
743     #pragma omp for nowait
744     for (index_t k0 = 0; k0 < m_NE0; ++k0) {
745     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
746     *o++ = 0.;
747     *o = 1.;
748     }
749     }
750     } // end of parallel section
751    
752     } else {
753     stringstream msg;
754     msg << "setToNormal() not implemented for "
755     << functionSpaceTypeAsString(out.getFunctionSpace().getTypeCode());
756     throw RipleyException(msg.str());
757     }
758     }
759    
760 caltinay 3691 Paso_SystemMatrixPattern* Rectangle::getPattern(bool reducedRowOrder,
761     bool reducedColOrder) const
762     {
763     if (reducedRowOrder || reducedColOrder)
764     throw RipleyException("getPattern() not implemented for reduced order");
765    
766 caltinay 3756 return m_pattern;
767 caltinay 3691 }
768    
769     void Rectangle::Print_Mesh_Info(const bool full) const
770     {
771     RipleyDomain::Print_Mesh_Info(full);
772     if (full) {
773     cout << " Id Coordinates" << endl;
774     cout.precision(15);
775     cout.setf(ios::scientific, ios::floatfield);
776 caltinay 3697 pair<double,double> xdx = getFirstCoordAndSpacing(0);
777     pair<double,double> ydy = getFirstCoordAndSpacing(1);
778 caltinay 3691 for (index_t i=0; i < getNumNodes(); i++) {
779     cout << " " << setw(5) << m_nodeId[i]
780 caltinay 3697 << " " << xdx.first+(i%m_N0)*xdx.second
781     << " " << ydy.first+(i/m_N0)*ydy.second << endl;
782 caltinay 3691 }
783     }
784     }
785    
786 caltinay 3697 IndexVector Rectangle::getNumNodesPerDim() const
787     {
788     IndexVector ret;
789     ret.push_back(m_N0);
790     ret.push_back(m_N1);
791     return ret;
792     }
793    
794     IndexVector Rectangle::getNumElementsPerDim() const
795     {
796     IndexVector ret;
797     ret.push_back(m_NE0);
798     ret.push_back(m_NE1);
799     return ret;
800     }
801    
802     IndexVector Rectangle::getNumFacesPerBoundary() const
803     {
804     IndexVector ret(4, 0);
805     //left
806     if (m_offset0==0)
807     ret[0]=m_NE1;
808     //right
809     if (m_mpiInfo->rank%m_NX==m_NX-1)
810     ret[1]=m_NE1;
811     //bottom
812     if (m_offset1==0)
813     ret[2]=m_NE0;
814     //top
815     if (m_mpiInfo->rank/m_NX==m_NY-1)
816     ret[3]=m_NE0;
817     return ret;
818     }
819    
820     pair<double,double> Rectangle::getFirstCoordAndSpacing(dim_t dim) const
821     {
822     if (dim==0) {
823     return pair<double,double>((m_l0*m_offset0)/m_gNE0, m_l0/m_gNE0);
824     } else if (dim==1) {
825     return pair<double,double>((m_l1*m_offset1)/m_gNE1, m_l1/m_gNE1);
826     }
827     throw RipleyException("getFirstCoordAndSpacing(): invalid argument");
828     }
829    
830 caltinay 3691 //protected
831 caltinay 3750 dim_t Rectangle::getNumDOF() const
832     {
833 caltinay 3752 return (m_gNE0+1)/m_NX*(m_gNE1+1)/m_NY;
834 caltinay 3750 }
835    
836     //protected
837 caltinay 3691 dim_t Rectangle::getNumFaceElements() const
838     {
839 caltinay 3699 const IndexVector faces = getNumFacesPerBoundary();
840 caltinay 3691 dim_t n=0;
841 caltinay 3699 for (size_t i=0; i<faces.size(); i++)
842     n+=faces[i];
843 caltinay 3691 return n;
844     }
845    
846     //protected
847     void Rectangle::assembleCoordinates(escript::Data& arg) const
848     {
849     escriptDataC x = arg.getDataC();
850     int numDim = m_numDim;
851     if (!isDataPointShapeEqual(&x, 1, &numDim))
852     throw RipleyException("setToX: Invalid Data object shape");
853     if (!numSamplesEqual(&x, 1, getNumNodes()))
854     throw RipleyException("setToX: Illegal number of samples in Data object");
855    
856 caltinay 3697 pair<double,double> xdx = getFirstCoordAndSpacing(0);
857     pair<double,double> ydy = getFirstCoordAndSpacing(1);
858 caltinay 3691 arg.requireWrite();
859     #pragma omp parallel for
860     for (dim_t i1 = 0; i1 < m_N1; i1++) {
861     for (dim_t i0 = 0; i0 < m_N0; i0++) {
862     double* point = arg.getSampleDataRW(i0+m_N0*i1);
863 caltinay 3697 point[0] = xdx.first+i0*xdx.second;
864     point[1] = ydy.first+i1*ydy.second;
865 caltinay 3691 }
866     }
867     }
868    
869 caltinay 3756 //protected
870     dim_t Rectangle::insertNeighbourNodes(IndexVector& index, index_t node) const
871     {
872     const dim_t nDOF0 = (m_gNE0+1)/m_NX;
873     const dim_t nDOF1 = (m_gNE1+1)/m_NY;
874     const int x=node%nDOF0;
875     const int y=node/nDOF0;
876     dim_t num=0;
877     // loop through potential neighbours and add to index if positions are
878     // within bounds
879     for (int i1=-1; i1<2; i1++) {
880     for (int i0=-1; i0<2; i0++) {
881     // skip node itself
882     if (i0==0 && i1==0)
883     continue;
884     // location of neighbour node
885     const int nx=x+i0;
886     const int ny=y+i1;
887     if (nx>=0 && ny>=0 && nx<nDOF0 && ny<nDOF1) {
888     index.push_back(ny*nDOF0+nx);
889     num++;
890     }
891     }
892     }
893    
894     return num;
895     }
896    
897     //protected
898     void Rectangle::nodesToDOF(escript::Data& out, escript::Data& in) const
899     {
900     const dim_t numComp = in.getDataPointSize();
901     out.requireWrite();
902    
903     const index_t left = (m_offset0==0 ? 0 : 1);
904     const index_t bottom = (m_offset1==0 ? 0 : 1);
905     const dim_t nDOF0 = (m_gNE0+1)/m_NX;
906     const dim_t nDOF1 = (m_gNE1+1)/m_NY;
907     #pragma omp parallel for
908     for (index_t i=0; i<nDOF1; i++) {
909     for (index_t j=0; j<nDOF0; j++) {
910     const index_t n=j+left+(i+bottom)*m_N0;
911     const double* src=in.getSampleDataRO(n);
912     copy(src, src+numComp, out.getSampleDataRW(j+i*nDOF0));
913     }
914     }
915     }
916    
917     //protected
918     void Rectangle::dofToNodes(escript::Data& out, escript::Data& in) const
919     {
920     const dim_t numComp = in.getDataPointSize();
921     Paso_Coupler* coupler = Paso_Coupler_alloc(m_connector, numComp);
922     in.requireWrite();
923     Paso_Coupler_startCollect(coupler, in.getSampleDataRW(0));
924    
925     const dim_t numDOF = getNumDOF();
926     out.requireWrite();
927     const double* buffer = Paso_Coupler_finishCollect(coupler);
928    
929     #pragma omp parallel for
930     for (index_t i=0; i<getNumNodes(); i++) {
931     const double* src=(m_dofMap[i]<numDOF ?
932     in.getSampleDataRO(m_dofMap[i])
933     : &buffer[(m_dofMap[i]-numDOF)*numComp]);
934     copy(src, src+numComp, out.getSampleDataRW(i));
935     }
936     }
937    
938 caltinay 3691 //private
939     void Rectangle::populateSampleIds()
940     {
941 caltinay 3752 // identifiers are ordered from left to right, bottom to top globablly.
942 caltinay 3697
943     // build node distribution vector first.
944     // rank i owns m_nodeDistribution[i+1]-nodeDistribution[i] nodes
945     m_nodeDistribution.assign(m_mpiInfo->size+1, 0);
946 caltinay 3752 const dim_t numDOF=getNumDOF();
947     for (dim_t k=1; k<m_mpiInfo->size; k++) {
948     m_nodeDistribution[k]=k*numDOF;
949 caltinay 3697 }
950     m_nodeDistribution[m_mpiInfo->size]=getNumDataPointsGlobal();
951 caltinay 3691 m_nodeId.resize(getNumNodes());
952 caltinay 3753 m_dofId.resize(numDOF);
953     m_elementId.resize(getNumElements());
954     m_faceId.resize(getNumFaceElements());
955 caltinay 3697
956 caltinay 3753 #pragma omp parallel
957     {
958     // nodes
959     #pragma omp for nowait
960     for (dim_t i1=0; i1<m_N1; i1++) {
961     for (dim_t i0=0; i0<m_N0; i0++) {
962     m_nodeId[i0+i1*m_N0] = (m_offset1+i1)*(m_gNE0+1)+m_offset0+i0;
963     }
964 caltinay 3697 }
965    
966 caltinay 3753 // degrees of freedom
967     #pragma omp for nowait
968     for (dim_t k=0; k<numDOF; k++)
969     m_dofId[k] = m_nodeDistribution[m_mpiInfo->rank]+k;
970 caltinay 3752
971 caltinay 3753 // elements
972     #pragma omp for nowait
973 caltinay 3755 for (dim_t i1=0; i1<m_NE1; i1++) {
974     for (dim_t i0=0; i0<m_NE0; i0++) {
975     m_elementId[i0+i1*m_NE0]=(m_offset1+i1)*m_gNE0+m_offset0+i0;
976     }
977     }
978 caltinay 3753
979     // face elements
980     #pragma omp for
981     for (dim_t k=0; k<getNumFaceElements(); k++)
982     m_faceId[k]=k;
983     } // end parallel section
984    
985 caltinay 3735 m_nodeTags.assign(getNumNodes(), 0);
986     updateTagsInUse(Nodes);
987 caltinay 3697
988 caltinay 3735 m_elementTags.assign(getNumElements(), 0);
989     updateTagsInUse(Elements);
990 caltinay 3697
991 caltinay 3722 // generate face offset vector and set face tags
992 caltinay 3704 const IndexVector facesPerEdge = getNumFacesPerBoundary();
993 caltinay 3722 const index_t LEFT=1, RIGHT=2, BOTTOM=10, TOP=20;
994     const index_t faceTag[] = { LEFT, RIGHT, BOTTOM, TOP };
995 caltinay 3704 m_faceOffset.assign(facesPerEdge.size(), -1);
996 caltinay 3722 m_faceTags.clear();
997 caltinay 3704 index_t offset=0;
998     for (size_t i=0; i<facesPerEdge.size(); i++) {
999     if (facesPerEdge[i]>0) {
1000     m_faceOffset[i]=offset;
1001     offset+=facesPerEdge[i];
1002 caltinay 3722 m_faceTags.insert(m_faceTags.end(), facesPerEdge[i], faceTag[i]);
1003 caltinay 3704 }
1004     }
1005 caltinay 3722 setTagMap("left", LEFT);
1006     setTagMap("right", RIGHT);
1007     setTagMap("bottom", BOTTOM);
1008     setTagMap("top", TOP);
1009     updateTagsInUse(FaceElements);
1010 caltinay 3691 }
1011    
1012 caltinay 3699 //private
1013 caltinay 3756 void Rectangle::createPattern()
1014 caltinay 3699 {
1015 caltinay 3756 const dim_t nDOF0 = (m_gNE0+1)/m_NX;
1016     const dim_t nDOF1 = (m_gNE1+1)/m_NY;
1017     const index_t left = (m_offset0==0 ? 0 : 1);
1018     const index_t bottom = (m_offset1==0 ? 0 : 1);
1019    
1020     // populate node->DOF mapping with own degrees of freedom.
1021     // The rest is assigned in the loop further down
1022     m_dofMap.assign(getNumNodes(), 0);
1023     #pragma omp parallel for
1024     for (index_t i=bottom; i<m_N1; i++) {
1025     for (index_t j=left; j<m_N0; j++) {
1026     m_dofMap[i*m_N0+j]=(i-bottom)*nDOF0+j-left;
1027     }
1028     }
1029    
1030     // build list of shared components and neighbours by looping through
1031     // all potential neighbouring ranks and checking if positions are
1032 caltinay 3754 // within bounds
1033 caltinay 3756 const dim_t numDOF=nDOF0*nDOF1;
1034     vector<IndexVector> colIndices(numDOF); // for the couple blocks
1035     RankVector neighbour;
1036     IndexVector offsetInShared(1,0);
1037     IndexVector sendShared, recvShared;
1038     int numShared=0;
1039     const int x=m_mpiInfo->rank%m_NX;
1040     const int y=m_mpiInfo->rank/m_NX;
1041 caltinay 3754 for (int i1=-1; i1<2; i1++) {
1042     for (int i0=-1; i0<2; i0++) {
1043 caltinay 3756 // skip this rank
1044 caltinay 3754 if (i0==0 && i1==0)
1045     continue;
1046 caltinay 3756 // location of neighbour rank
1047 caltinay 3754 const int nx=x+i0;
1048     const int ny=y+i1;
1049 caltinay 3756 if (nx>=0 && ny>=0 && nx<m_NX && ny<m_NY) {
1050     neighbour.push_back(ny*m_NX+nx);
1051     if (i0==0) {
1052     // sharing top or bottom edge
1053     const int firstDOF=(i1==-1 ? 0 : numDOF-nDOF0);
1054     const int firstNode=(i1==-1 ? left : m_N0*(m_N1-1)+left);
1055     offsetInShared.push_back(offsetInShared.back()+nDOF0);
1056     for (dim_t i=0; i<nDOF0; i++, numShared++) {
1057     sendShared.push_back(firstDOF+i);
1058     recvShared.push_back(numDOF+numShared);
1059     if (i>0)
1060     colIndices[firstDOF+i-1].push_back(numShared);
1061     colIndices[firstDOF+i].push_back(numShared);
1062     if (i<nDOF0-1)
1063     colIndices[firstDOF+i+1].push_back(numShared);
1064     m_dofMap[firstNode+i]=numDOF+numShared;
1065     }
1066     } else if (i1==0) {
1067     // sharing left or right edge
1068     const int firstDOF=(i0==-1 ? 0 : nDOF0-1);
1069     const int firstNode=(i0==-1 ? bottom*m_N0 : (bottom+1)*m_N0-1);
1070     offsetInShared.push_back(offsetInShared.back()+nDOF1);
1071     for (dim_t i=0; i<nDOF1; i++, numShared++) {
1072     sendShared.push_back(firstDOF+i*nDOF0);
1073     recvShared.push_back(numDOF+numShared);
1074     if (i>0)
1075     colIndices[firstDOF+(i-1)*nDOF0].push_back(numShared);
1076     colIndices[firstDOF+i*nDOF0].push_back(numShared);
1077     if (i<nDOF1-1)
1078     colIndices[firstDOF+(i+1)*nDOF0].push_back(numShared);
1079     m_dofMap[firstNode+i*m_N0]=numDOF+numShared;
1080     }
1081     } else {
1082     // sharing a node
1083     const int dof=(i0+1)/2*(nDOF0-1)+(i1+1)/2*(numDOF-nDOF0);
1084     const int node=(i0+1)/2*(m_N0-1)+(i1+1)/2*m_N0*(m_N1-1);
1085     offsetInShared.push_back(offsetInShared.back()+1);
1086     sendShared.push_back(dof);
1087     recvShared.push_back(numDOF+numShared);
1088     colIndices[dof].push_back(numShared);
1089     m_dofMap[node]=numDOF+numShared;
1090     ++numShared;
1091     }
1092 caltinay 3754 }
1093 caltinay 3699 }
1094     }
1095 caltinay 3754
1096 caltinay 3756 // create connector
1097     Paso_SharedComponents *snd_shcomp = Paso_SharedComponents_alloc(
1098     numDOF, neighbour.size(), &neighbour[0], &sendShared[0],
1099     &offsetInShared[0], 1, 0, m_mpiInfo);
1100     Paso_SharedComponents *rcv_shcomp = Paso_SharedComponents_alloc(
1101     numDOF, neighbour.size(), &neighbour[0], &recvShared[0],
1102     &offsetInShared[0], 1, 0, m_mpiInfo);
1103     m_connector = Paso_Connector_alloc(snd_shcomp, rcv_shcomp);
1104     Paso_SharedComponents_free(snd_shcomp);
1105     Paso_SharedComponents_free(rcv_shcomp);
1106 caltinay 3754
1107 caltinay 3756 // create main and couple blocks
1108     Paso_Pattern *mainPattern = createMainPattern();
1109     Paso_Pattern *colPattern, *rowPattern;
1110     createCouplePatterns(colIndices, numShared, &colPattern, &rowPattern);
1111 caltinay 3754
1112 caltinay 3756 // allocate paso distribution
1113     Paso_Distribution* distribution = Paso_Distribution_alloc(m_mpiInfo,
1114     const_cast<index_t*>(&m_nodeDistribution[0]), 1, 0);
1115 caltinay 3755
1116 caltinay 3756 // finally create the system matrix
1117     m_pattern = Paso_SystemMatrixPattern_alloc(MATRIX_FORMAT_DEFAULT,
1118     distribution, distribution, mainPattern, colPattern, rowPattern,
1119     m_connector, m_connector);
1120 caltinay 3755
1121 caltinay 3756 Paso_Distribution_free(distribution);
1122 caltinay 3755
1123 caltinay 3756 // useful debug output
1124     /*
1125     cout << "--- rcv_shcomp ---" << endl;
1126     cout << "numDOF=" << numDOF << ", numNeighbors=" << neighbour.size() << endl;
1127     for (size_t i=0; i<neighbour.size(); i++) {
1128     cout << "neighbor[" << i << "]=" << neighbour[i]
1129     << " offsetInShared[" << i+1 << "]=" << offsetInShared[i+1] << endl;
1130 caltinay 3699 }
1131 caltinay 3756 for (size_t i=0; i<recvShared.size(); i++) {
1132     cout << "shared[" << i << "]=" << recvShared[i] << endl;
1133     }
1134     cout << "--- snd_shcomp ---" << endl;
1135     for (size_t i=0; i<sendShared.size(); i++) {
1136     cout << "shared[" << i << "]=" << sendShared[i] << endl;
1137     }
1138     cout << "--- dofMap ---" << endl;
1139     for (size_t i=0; i<m_dofMap.size(); i++) {
1140     cout << "m_dofMap[" << i << "]=" << m_dofMap[i] << endl;
1141     }
1142     cout << "--- colIndices ---" << endl;
1143     for (size_t i=0; i<colIndices.size(); i++) {
1144     cout << "colIndices[" << i << "].size()=" << colIndices[i].size() << endl;
1145     }
1146     */
1147 caltinay 3754
1148 caltinay 3756 /*
1149     cout << "--- main_pattern ---" << endl;
1150     cout << "M=" << mainPattern->numOutput << ", N=" << mainPattern->numInput << endl;
1151     for (size_t i=0; i<mainPattern->numOutput+1; i++) {
1152     cout << "ptr[" << i << "]=" << mainPattern->ptr[i] << endl;
1153     }
1154     for (size_t i=0; i<mainPattern->ptr[mainPattern->numOutput]; i++) {
1155     cout << "index[" << i << "]=" << mainPattern->index[i] << endl;
1156     }
1157     */
1158 caltinay 3754
1159 caltinay 3756 /*
1160     cout << "--- colCouple_pattern ---" << endl;
1161     cout << "M=" << colPattern->numOutput << ", N=" << colPattern->numInput << endl;
1162     for (size_t i=0; i<colPattern->numOutput+1; i++) {
1163     cout << "ptr[" << i << "]=" << colPattern->ptr[i] << endl;
1164     }
1165     for (size_t i=0; i<colPattern->ptr[colPattern->numOutput]; i++) {
1166     cout << "index[" << i << "]=" << colPattern->index[i] << endl;
1167     }
1168     */
1169 caltinay 3754
1170 caltinay 3756 /*
1171     cout << "--- rowCouple_pattern ---" << endl;
1172     cout << "M=" << rowPattern->numOutput << ", N=" << rowPattern->numInput << endl;
1173     for (size_t i=0; i<rowPattern->numOutput+1; i++) {
1174     cout << "ptr[" << i << "]=" << rowPattern->ptr[i] << endl;
1175 caltinay 3699 }
1176 caltinay 3756 for (size_t i=0; i<rowPattern->ptr[rowPattern->numOutput]; i++) {
1177     cout << "index[" << i << "]=" << rowPattern->index[i] << endl;
1178     }
1179     */
1180    
1181     Paso_Pattern_free(mainPattern);
1182     Paso_Pattern_free(colPattern);
1183     Paso_Pattern_free(rowPattern);
1184 caltinay 3699 }
1185    
1186 caltinay 3702 //protected
1187 caltinay 3711 void Rectangle::interpolateNodesOnElements(escript::Data& out,
1188     escript::Data& in, bool reduced) const
1189 caltinay 3702 {
1190     const dim_t numComp = in.getDataPointSize();
1191 caltinay 3711 if (reduced) {
1192 caltinay 3724 /*** GENERATOR SNIP_INTERPOLATE_REDUCED_ELEMENTS TOP */
1193     const double c0 = .25;
1194 caltinay 3702 #pragma omp parallel for
1195 caltinay 3711 for (index_t k1=0; k1 < m_NE1; ++k1) {
1196     for (index_t k0=0; k0 < m_NE0; ++k0) {
1197     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
1198     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
1199     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
1200     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
1201     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
1202     for (index_t i=0; i < numComp; ++i) {
1203 caltinay 3724 o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i] + f_10[i] + f_11[i]);
1204 caltinay 3711 } /* end of component loop i */
1205     } /* end of k0 loop */
1206     } /* end of k1 loop */
1207     /* GENERATOR SNIP_INTERPOLATE_REDUCED_ELEMENTS BOTTOM */
1208     } else {
1209 caltinay 3724 /*** GENERATOR SNIP_INTERPOLATE_ELEMENTS TOP */
1210     const double c0 = .16666666666666666667;
1211     const double c1 = .044658198738520451079;
1212     const double c2 = .62200846792814621559;
1213 caltinay 3711 #pragma omp parallel for
1214     for (index_t k1=0; k1 < m_NE1; ++k1) {
1215     for (index_t k0=0; k0 < m_NE0; ++k0) {
1216     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
1217     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
1218     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
1219     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
1220     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
1221     for (index_t i=0; i < numComp; ++i) {
1222 caltinay 3724 o[INDEX2(i,numComp,0)] = f_00[i]*c2 + f_11[i]*c1 + c0*(f_01[i] + f_10[i]);
1223     o[INDEX2(i,numComp,1)] = f_01[i]*c1 + f_10[i]*c2 + c0*(f_00[i] + f_11[i]);
1224     o[INDEX2(i,numComp,2)] = f_01[i]*c2 + f_10[i]*c1 + c0*(f_00[i] + f_11[i]);
1225     o[INDEX2(i,numComp,3)] = f_00[i]*c1 + f_11[i]*c2 + c0*(f_01[i] + f_10[i]);
1226 caltinay 3711 } /* end of component loop i */
1227     } /* end of k0 loop */
1228     } /* end of k1 loop */
1229     /* GENERATOR SNIP_INTERPOLATE_ELEMENTS BOTTOM */
1230     }
1231 caltinay 3702 }
1232    
1233     //protected
1234 caltinay 3711 void Rectangle::interpolateNodesOnFaces(escript::Data& out, escript::Data& in,
1235     bool reduced) const
1236 caltinay 3702 {
1237 caltinay 3704 const dim_t numComp = in.getDataPointSize();
1238 caltinay 3711 if (reduced) {
1239 caltinay 3724 const double c0 = .5;
1240     #pragma omp parallel
1241     {
1242     /*** GENERATOR SNIP_INTERPOLATE_REDUCED_FACES TOP */
1243     if (m_faceOffset[0] > -1) {
1244     #pragma omp for nowait
1245     for (index_t k1=0; k1 < m_NE1; ++k1) {
1246     const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
1247     const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
1248     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1249     for (index_t i=0; i < numComp; ++i) {
1250     o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i]);
1251     } /* end of component loop i */
1252     } /* end of k1 loop */
1253     } /* end of face 0 */
1254     if (m_faceOffset[1] > -1) {
1255     #pragma omp for nowait
1256     for (index_t k1=0; k1 < m_NE1; ++k1) {
1257     const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
1258     const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
1259     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1260     for (index_t i=0; i < numComp; ++i) {
1261     o[INDEX2(i,numComp,0)] = c0*(f_10[i] + f_11[i]);
1262     } /* end of component loop i */
1263     } /* end of k1 loop */
1264     } /* end of face 1 */
1265     if (m_faceOffset[2] > -1) {
1266     #pragma omp for nowait
1267     for (index_t k0=0; k0 < m_NE0; ++k0) {
1268     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
1269     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
1270     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1271     for (index_t i=0; i < numComp; ++i) {
1272     o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_10[i]);
1273     } /* end of component loop i */
1274     } /* end of k0 loop */
1275     } /* end of face 2 */
1276     if (m_faceOffset[3] > -1) {
1277     #pragma omp for nowait
1278     for (index_t k0=0; k0 < m_NE0; ++k0) {
1279     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
1280     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
1281     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1282     for (index_t i=0; i < numComp; ++i) {
1283     o[INDEX2(i,numComp,0)] = c0*(f_01[i] + f_11[i]);
1284     } /* end of component loop i */
1285     } /* end of k0 loop */
1286     } /* end of face 3 */
1287     /* GENERATOR SNIP_INTERPOLATE_REDUCED_FACES BOTTOM */
1288     } // end of parallel section
1289 caltinay 3711 } else {
1290 caltinay 3724 const double c0 = 0.21132486540518711775;
1291     const double c1 = 0.78867513459481288225;
1292     #pragma omp parallel
1293     {
1294     /*** GENERATOR SNIP_INTERPOLATE_FACES TOP */
1295     if (m_faceOffset[0] > -1) {
1296     #pragma omp for nowait
1297     for (index_t k1=0; k1 < m_NE1; ++k1) {
1298     const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
1299     const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
1300     double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1301     for (index_t i=0; i < numComp; ++i) {
1302     o[INDEX2(i,numComp,0)] = f_00[i]*c1 + f_01[i]*c0;
1303     o[INDEX2(i,numComp,1)] = f_00[i]*c0 + f_01[i]*c1;
1304     } /* end of component loop i */
1305     } /* end of k1 loop */
1306     } /* end of face 0 */
1307     if (m_faceOffset[1] > -1) {
1308     #pragma omp for nowait
1309     for (index_t k1=0; k1 < m_NE1; ++k1) {
1310     const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
1311     const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
1312     double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1313     for (index_t i=0; i < numComp; ++i) {
1314     o[INDEX2(i,numComp,0)] = f_10[i]*c1 + f_11[i]*c0;
1315     o[INDEX2(i,numComp,1)] = f_10[i]*c0 + f_11[i]*c1;
1316     } /* end of component loop i */
1317     } /* end of k1 loop */
1318     } /* end of face 1 */
1319     if (m_faceOffset[2] > -1) {
1320     #pragma omp for nowait
1321     for (index_t k0=0; k0 < m_NE0; ++k0) {
1322     const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
1323     const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
1324     double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1325     for (index_t i=0; i < numComp; ++i) {
1326     o[INDEX2(i,numComp,0)] = f_00[i]*c1 + f_10[i]*c0;
1327     o[INDEX2(i,numComp,1)] = f_00[i]*c0 + f_10[i]*c1;
1328     } /* end of component loop i */
1329     } /* end of k0 loop */
1330     } /* end of face 2 */
1331     if (m_faceOffset[3] > -1) {
1332     #pragma omp for nowait
1333     for (index_t k0=0; k0 < m_NE0; ++k0) {
1334     const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
1335     const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
1336     double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1337     for (index_t i=0; i < numComp; ++i) {
1338     o[INDEX2(i,numComp,0)] = f_01[i]*c1 + f_11[i]*c0;
1339     o[INDEX2(i,numComp,1)] = f_01[i]*c0 + f_11[i]*c1;
1340     } /* end of component loop i */
1341     } /* end of k0 loop */
1342     } /* end of face 3 */
1343     /* GENERATOR SNIP_INTERPOLATE_FACES BOTTOM */
1344     } // end of parallel section
1345 caltinay 3711 }
1346 caltinay 3702 }
1347    
1348 caltinay 3748 //protected
1349     void Rectangle::assemblePDESingle(Paso_SystemMatrix* mat,
1350     escript::Data& rhs, const escript::Data& A, const escript::Data& B,
1351     const escript::Data& C, const escript::Data& D,
1352     const escript::Data& X, const escript::Data& Y,
1353     const escript::Data& d, const escript::Data& y) const
1354     {
1355     const double h0 = m_l0/m_gNE0;
1356     const double h1 = m_l1/m_gNE1;
1357 caltinay 3754 /*** GENERATOR SNIP_PDE_SINGLE_PRE TOP */
1358 caltinay 3748 const double w0 = -0.1555021169820365539*h1/h0;
1359     const double w1 = 0.041666666666666666667;
1360     const double w10 = -0.041666666666666666667*h0/h1;
1361     const double w11 = 0.1555021169820365539*h1/h0;
1362     const double w12 = 0.1555021169820365539*h0/h1;
1363     const double w13 = 0.01116454968463011277*h0/h1;
1364     const double w14 = 0.01116454968463011277*h1/h0;
1365     const double w15 = 0.041666666666666666667*h1/h0;
1366     const double w16 = -0.01116454968463011277*h0/h1;
1367     const double w17 = -0.1555021169820365539*h0/h1;
1368     const double w18 = -0.33333333333333333333*h1/h0;
1369     const double w19 = 0.25000000000000000000;
1370     const double w2 = -0.15550211698203655390;
1371     const double w20 = -0.25000000000000000000;
1372     const double w21 = 0.16666666666666666667*h0/h1;
1373     const double w22 = -0.16666666666666666667*h1/h0;
1374     const double w23 = -0.16666666666666666667*h0/h1;
1375     const double w24 = 0.33333333333333333333*h1/h0;
1376     const double w25 = 0.33333333333333333333*h0/h1;
1377     const double w26 = 0.16666666666666666667*h1/h0;
1378     const double w27 = -0.33333333333333333333*h0/h1;
1379     const double w28 = -0.032861463941450536761*h1;
1380     const double w29 = -0.032861463941450536761*h0;
1381     const double w3 = 0.041666666666666666667*h0/h1;
1382     const double w30 = -0.12264065304058601714*h1;
1383     const double w31 = -0.0023593469594139828636*h1;
1384     const double w32 = -0.008805202725216129906*h0;
1385     const double w33 = -0.008805202725216129906*h1;
1386     const double w34 = 0.032861463941450536761*h1;
1387     const double w35 = 0.008805202725216129906*h1;
1388     const double w36 = 0.008805202725216129906*h0;
1389     const double w37 = 0.0023593469594139828636*h1;
1390     const double w38 = 0.12264065304058601714*h1;
1391     const double w39 = 0.032861463941450536761*h0;
1392     const double w4 = 0.15550211698203655390;
1393     const double w40 = -0.12264065304058601714*h0;
1394     const double w41 = -0.0023593469594139828636*h0;
1395     const double w42 = 0.0023593469594139828636*h0;
1396     const double w43 = 0.12264065304058601714*h0;
1397     const double w44 = -0.16666666666666666667*h1;
1398     const double w45 = -0.083333333333333333333*h0;
1399     const double w46 = 0.083333333333333333333*h1;
1400     const double w47 = 0.16666666666666666667*h1;
1401     const double w48 = 0.083333333333333333333*h0;
1402     const double w49 = -0.16666666666666666667*h0;
1403     const double w5 = -0.041666666666666666667;
1404     const double w50 = 0.16666666666666666667*h0;
1405     const double w51 = -0.083333333333333333333*h1;
1406     const double w52 = 0.025917019497006092316*h0*h1;
1407     const double w53 = 0.0018607582807716854616*h0*h1;
1408     const double w54 = 0.0069444444444444444444*h0*h1;
1409     const double w55 = 0.09672363354357992482*h0*h1;
1410     const double w56 = 0.00049858867864229740201*h0*h1;
1411     const double w57 = 0.055555555555555555556*h0*h1;
1412     const double w58 = 0.027777777777777777778*h0*h1;
1413     const double w59 = 0.11111111111111111111*h0*h1;
1414     const double w6 = -0.01116454968463011277*h1/h0;
1415     const double w60 = -0.19716878364870322056*h1;
1416     const double w61 = -0.19716878364870322056*h0;
1417     const double w62 = -0.052831216351296779436*h0;
1418     const double w63 = -0.052831216351296779436*h1;
1419     const double w64 = 0.19716878364870322056*h1;
1420     const double w65 = 0.052831216351296779436*h1;
1421     const double w66 = 0.19716878364870322056*h0;
1422     const double w67 = 0.052831216351296779436*h0;
1423     const double w68 = -0.5*h1;
1424     const double w69 = -0.5*h0;
1425     const double w7 = 0.011164549684630112770;
1426     const double w70 = 0.5*h1;
1427     const double w71 = 0.5*h0;
1428     const double w72 = 0.1555021169820365539*h0*h1;
1429     const double w73 = 0.041666666666666666667*h0*h1;
1430     const double w74 = 0.01116454968463011277*h0*h1;
1431     const double w75 = 0.25*h0*h1;
1432     const double w8 = -0.011164549684630112770;
1433     const double w9 = -0.041666666666666666667*h1/h0;
1434     /* GENERATOR SNIP_PDE_SINGLE_PRE BOTTOM */
1435    
1436     rhs.requireWrite();
1437     #pragma omp parallel
1438     {
1439 caltinay 3754 for (index_t k1_0=0; k1_0<2; k1_0++) { // colouring
1440 caltinay 3748 #pragma omp for
1441     for (index_t k1=k1_0; k1<m_NE1; k1+=2) {
1442     for (index_t k0=0; k0<m_NE0; ++k0) {
1443     bool add_EM_S=false;
1444     bool add_EM_F=false;
1445     vector<double> EM_S(4*4, 0);
1446     vector<double> EM_F(4, 0);
1447     const index_t e = k0 + m_NE0*k1;
1448 caltinay 3754 /*** GENERATOR SNIP_PDE_SINGLE TOP */
1449 caltinay 3748 ///////////////
1450     // process A //
1451     ///////////////
1452     if (!A.isEmpty()) {
1453     add_EM_S=true;
1454     const double* A_p=const_cast<escript::Data*>(&A)->getSampleDataRO(e);
1455     if (A.actsExpanded()) {
1456     const register double A_00_0 = A_p[INDEX3(0,0,0,2,2)];
1457     const register double A_01_0 = A_p[INDEX3(0,1,0,2,2)];
1458     const register double A_10_0 = A_p[INDEX3(1,0,0,2,2)];
1459     const register double A_11_0 = A_p[INDEX3(1,1,0,2,2)];
1460     const register double A_00_1 = A_p[INDEX3(0,0,1,2,2)];
1461     const register double A_01_1 = A_p[INDEX3(0,1,1,2,2)];
1462     const register double A_10_1 = A_p[INDEX3(1,0,1,2,2)];
1463     const register double A_11_1 = A_p[INDEX3(1,1,1,2,2)];
1464     const register double A_00_2 = A_p[INDEX3(0,0,2,2,2)];
1465     const register double A_01_2 = A_p[INDEX3(0,1,2,2,2)];
1466     const register double A_10_2 = A_p[INDEX3(1,0,2,2,2)];
1467     const register double A_11_2 = A_p[INDEX3(1,1,2,2,2)];
1468     const register double A_00_3 = A_p[INDEX3(0,0,3,2,2)];
1469     const register double A_01_3 = A_p[INDEX3(0,1,3,2,2)];
1470     const register double A_10_3 = A_p[INDEX3(1,0,3,2,2)];
1471     const register double A_11_3 = A_p[INDEX3(1,1,3,2,2)];
1472     const register double tmp4_0 = A_10_1 + A_10_2;
1473     const register double tmp12_0 = A_11_0 + A_11_2;
1474     const register double tmp2_0 = A_11_0 + A_11_1 + A_11_2 + A_11_3;
1475     const register double tmp10_0 = A_01_3 + A_10_3;
1476     const register double tmp14_0 = A_01_0 + A_01_3 + A_10_0 + A_10_3;
1477     const register double tmp0_0 = A_01_0 + A_01_3;
1478     const register double tmp13_0 = A_01_2 + A_10_1;
1479     const register double tmp3_0 = A_00_2 + A_00_3;
1480     const register double tmp11_0 = A_11_1 + A_11_3;
1481     const register double tmp18_0 = A_01_1 + A_10_1;
1482     const register double tmp1_0 = A_00_0 + A_00_1;
1483     const register double tmp15_0 = A_01_1 + A_10_2;
1484     const register double tmp5_0 = A_00_0 + A_00_1 + A_00_2 + A_00_3;
1485     const register double tmp16_0 = A_10_0 + A_10_3;
1486     const register double tmp6_0 = A_01_3 + A_10_0;
1487     const register double tmp17_0 = A_01_1 + A_01_2;
1488     const register double tmp9_0 = A_01_0 + A_10_0;
1489     const register double tmp7_0 = A_01_0 + A_10_3;
1490     const register double tmp8_0 = A_01_1 + A_01_2 + A_10_1 + A_10_2;
1491     const register double tmp19_0 = A_01_2 + A_10_2;
1492     const register double tmp14_1 = A_10_0*w8;
1493     const register double tmp23_1 = tmp3_0*w14;
1494     const register double tmp35_1 = A_01_0*w8;
1495     const register double tmp54_1 = tmp13_0*w8;
1496     const register double tmp20_1 = tmp9_0*w4;
1497     const register double tmp25_1 = tmp12_0*w12;
1498     const register double tmp2_1 = A_01_1*w4;
1499     const register double tmp44_1 = tmp7_0*w7;
1500     const register double tmp26_1 = tmp10_0*w4;
1501     const register double tmp52_1 = tmp18_0*w8;
1502     const register double tmp48_1 = A_10_1*w7;
1503     const register double tmp46_1 = A_01_3*w8;
1504     const register double tmp50_1 = A_01_0*w2;
1505     const register double tmp8_1 = tmp4_0*w5;
1506     const register double tmp56_1 = tmp19_0*w8;
1507     const register double tmp9_1 = tmp2_0*w10;
1508     const register double tmp19_1 = A_10_3*w2;
1509     const register double tmp47_1 = A_10_2*w4;
1510     const register double tmp16_1 = tmp3_0*w0;
1511     const register double tmp18_1 = tmp1_0*w6;
1512     const register double tmp31_1 = tmp11_0*w12;
1513     const register double tmp55_1 = tmp15_0*w2;
1514     const register double tmp39_1 = A_10_2*w7;
1515     const register double tmp11_1 = tmp6_0*w7;
1516     const register double tmp40_1 = tmp11_0*w17;
1517     const register double tmp34_1 = tmp15_0*w8;
1518     const register double tmp33_1 = tmp14_0*w5;
1519     const register double tmp24_1 = tmp11_0*w13;
1520     const register double tmp3_1 = tmp1_0*w0;
1521     const register double tmp5_1 = tmp2_0*w3;
1522     const register double tmp43_1 = tmp17_0*w5;
1523     const register double tmp15_1 = A_01_2*w4;
1524     const register double tmp53_1 = tmp19_0*w2;
1525     const register double tmp27_1 = tmp3_0*w11;
1526     const register double tmp32_1 = tmp13_0*w2;
1527     const register double tmp10_1 = tmp5_0*w9;
1528     const register double tmp37_1 = A_10_1*w4;
1529     const register double tmp38_1 = tmp5_0*w15;
1530     const register double tmp17_1 = A_01_1*w7;
1531     const register double tmp12_1 = tmp7_0*w4;
1532     const register double tmp22_1 = tmp10_0*w7;
1533     const register double tmp57_1 = tmp18_0*w2;
1534     const register double tmp28_1 = tmp9_0*w7;
1535     const register double tmp29_1 = tmp1_0*w14;
1536     const register double tmp51_1 = tmp11_0*w16;
1537     const register double tmp42_1 = tmp12_0*w16;
1538     const register double tmp49_1 = tmp12_0*w17;
1539     const register double tmp21_1 = tmp1_0*w11;
1540     const register double tmp1_1 = tmp0_0*w1;
1541     const register double tmp45_1 = tmp6_0*w4;
1542     const register double tmp7_1 = A_10_0*w2;
1543     const register double tmp6_1 = tmp3_0*w6;
1544     const register double tmp13_1 = tmp8_0*w1;
1545     const register double tmp36_1 = tmp16_0*w1;
1546     const register double tmp41_1 = A_01_3*w2;
1547     const register double tmp30_1 = tmp12_0*w13;
1548     const register double tmp4_1 = A_01_2*w7;
1549     const register double tmp0_1 = A_10_3*w8;
1550     EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1 + tmp4_1 + tmp5_1 + tmp6_1 + tmp7_1 + tmp8_1;
1551     EM_S[INDEX2(1,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1 + tmp9_1;
1552     EM_S[INDEX2(3,2,4)]+=tmp14_1 + tmp15_1 + tmp16_1 + tmp17_1 + tmp18_1 + tmp19_1 + tmp1_1 + tmp5_1 + tmp8_1;
1553     EM_S[INDEX2(0,0,4)]+=tmp13_1 + tmp20_1 + tmp21_1 + tmp22_1 + tmp23_1 + tmp24_1 + tmp25_1;
1554     EM_S[INDEX2(3,3,4)]+=tmp13_1 + tmp26_1 + tmp27_1 + tmp28_1 + tmp29_1 + tmp30_1 + tmp31_1;
1555     EM_S[INDEX2(3,0,4)]+=tmp10_1 + tmp32_1 + tmp33_1 + tmp34_1 + tmp9_1;
1556     EM_S[INDEX2(3,1,4)]+=tmp35_1 + tmp36_1 + tmp37_1 + tmp38_1 + tmp39_1 + tmp40_1 + tmp41_1 + tmp42_1 + tmp43_1;
1557     EM_S[INDEX2(2,1,4)]+=tmp10_1 + tmp13_1 + tmp44_1 + tmp45_1 + tmp9_1;
1558     EM_S[INDEX2(0,2,4)]+=tmp36_1 + tmp38_1 + tmp43_1 + tmp46_1 + tmp47_1 + tmp48_1 + tmp49_1 + tmp50_1 + tmp51_1;
1559     EM_S[INDEX2(2,0,4)]+=tmp0_1 + tmp15_1 + tmp17_1 + tmp1_1 + tmp38_1 + tmp49_1 + tmp51_1 + tmp7_1 + tmp8_1;
1560     EM_S[INDEX2(1,3,4)]+=tmp14_1 + tmp19_1 + tmp1_1 + tmp2_1 + tmp38_1 + tmp40_1 + tmp42_1 + tmp4_1 + tmp8_1;
1561     EM_S[INDEX2(2,3,4)]+=tmp16_1 + tmp18_1 + tmp35_1 + tmp36_1 + tmp41_1 + tmp43_1 + tmp47_1 + tmp48_1 + tmp5_1;
1562     EM_S[INDEX2(2,2,4)]+=tmp24_1 + tmp25_1 + tmp27_1 + tmp29_1 + tmp33_1 + tmp52_1 + tmp53_1;
1563     EM_S[INDEX2(1,0,4)]+=tmp36_1 + tmp37_1 + tmp39_1 + tmp3_1 + tmp43_1 + tmp46_1 + tmp50_1 + tmp5_1 + tmp6_1;
1564     EM_S[INDEX2(0,3,4)]+=tmp10_1 + tmp33_1 + tmp54_1 + tmp55_1 + tmp9_1;
1565     EM_S[INDEX2(1,1,4)]+=tmp21_1 + tmp23_1 + tmp30_1 + tmp31_1 + tmp33_1 + tmp56_1 + tmp57_1;
1566     } else { /* constant data */
1567     const register double A_00 = A_p[INDEX2(0,0,2)];
1568     const register double A_01 = A_p[INDEX2(0,1,2)];
1569     const register double A_10 = A_p[INDEX2(1,0,2)];
1570     const register double A_11 = A_p[INDEX2(1,1,2)];
1571     const register double tmp0_0 = A_01 + A_10;
1572     const register double tmp0_1 = A_00*w18;
1573     const register double tmp10_1 = A_01*w20;
1574     const register double tmp12_1 = A_00*w26;
1575     const register double tmp4_1 = A_00*w22;
1576     const register double tmp8_1 = A_00*w24;
1577     const register double tmp13_1 = A_10*w19;
1578     const register double tmp9_1 = tmp0_0*w20;
1579     const register double tmp3_1 = A_11*w21;
1580     const register double tmp11_1 = A_11*w27;
1581     const register double tmp1_1 = A_01*w19;
1582     const register double tmp6_1 = A_11*w23;
1583     const register double tmp7_1 = A_11*w25;
1584     const register double tmp2_1 = A_10*w20;
1585     const register double tmp5_1 = tmp0_0*w19;
1586     EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1;
1587     EM_S[INDEX2(1,2,4)]+=tmp4_1 + tmp5_1 + tmp6_1;
1588     EM_S[INDEX2(3,2,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1;
1589     EM_S[INDEX2(0,0,4)]+=tmp5_1 + tmp7_1 + tmp8_1;
1590     EM_S[INDEX2(3,3,4)]+=tmp5_1 + tmp7_1 + tmp8_1;
1591     EM_S[INDEX2(3,0,4)]+=tmp4_1 + tmp6_1 + tmp9_1;
1592     EM_S[INDEX2(3,1,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1;
1593     EM_S[INDEX2(2,1,4)]+=tmp4_1 + tmp5_1 + tmp6_1;
1594     EM_S[INDEX2(0,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1;
1595     EM_S[INDEX2(2,0,4)]+=tmp11_1 + tmp12_1 + tmp1_1 + tmp2_1;
1596     EM_S[INDEX2(1,3,4)]+=tmp11_1 + tmp12_1 + tmp1_1 + tmp2_1;
1597     EM_S[INDEX2(2,3,4)]+=tmp0_1 + tmp10_1 + tmp13_1 + tmp3_1;
1598     EM_S[INDEX2(2,2,4)]+=tmp7_1 + tmp8_1 + tmp9_1;
1599     EM_S[INDEX2(1,0,4)]+=tmp0_1 + tmp10_1 + tmp13_1 + tmp3_1;
1600     EM_S[INDEX2(0,3,4)]+=tmp4_1 + tmp6_1 + tmp9_1;
1601     EM_S[INDEX2(1,1,4)]+=tmp7_1 + tmp8_1 + tmp9_1;
1602     }
1603     }
1604     ///////////////
1605     // process B //
1606     ///////////////
1607     if (!B.isEmpty()) {
1608     add_EM_S=true;
1609     const double* B_p=const_cast<escript::Data*>(&B)->getSampleDataRO(e);
1610     if (B.actsExpanded()) {
1611     const register double B_0_0 = B_p[INDEX2(0,0,2)];
1612     const register double B_1_0 = B_p[INDEX2(1,0,2)];
1613     const register double B_0_1 = B_p[INDEX2(0,1,2)];
1614     const register double B_1_1 = B_p[INDEX2(1,1,2)];
1615     const register double B_0_2 = B_p[INDEX2(0,2,2)];
1616     const register double B_1_2 = B_p[INDEX2(1,2,2)];
1617     const register double B_0_3 = B_p[INDEX2(0,3,2)];
1618     const register double B_1_3 = B_p[INDEX2(1,3,2)];
1619     const register double tmp3_0 = B_0_0 + B_0_2;
1620     const register double tmp1_0 = B_1_2 + B_1_3;
1621     const register double tmp2_0 = B_0_1 + B_0_3;
1622     const register double tmp0_0 = B_1_0 + B_1_1;
1623     const register double tmp63_1 = B_1_1*w42;
1624     const register double tmp79_1 = B_1_1*w40;
1625     const register double tmp37_1 = tmp3_0*w35;
1626     const register double tmp8_1 = tmp0_0*w32;
1627     const register double tmp71_1 = B_0_1*w34;
1628     const register double tmp19_1 = B_0_3*w31;
1629     const register double tmp15_1 = B_0_3*w34;
1630     const register double tmp9_1 = tmp3_0*w34;
1631     const register double tmp35_1 = B_1_0*w36;
1632     const register double tmp66_1 = B_0_3*w28;
1633     const register double tmp28_1 = B_1_0*w42;
1634     const register double tmp22_1 = B_1_0*w40;
1635     const register double tmp16_1 = B_1_2*w29;
1636     const register double tmp6_1 = tmp2_0*w35;
1637     const register double tmp55_1 = B_1_3*w40;
1638     const register double tmp50_1 = B_1_3*w42;
1639     const register double tmp7_1 = tmp1_0*w29;
1640     const register double tmp1_1 = tmp1_0*w32;
1641     const register double tmp57_1 = B_0_3*w30;
1642     const register double tmp18_1 = B_1_1*w32;
1643     const register double tmp53_1 = B_1_0*w41;
1644     const register double tmp61_1 = B_1_3*w36;
1645     const register double tmp27_1 = B_0_3*w38;
1646     const register double tmp64_1 = B_0_2*w30;
1647     const register double tmp76_1 = B_0_1*w38;
1648     const register double tmp39_1 = tmp2_0*w34;
1649     const register double tmp62_1 = B_0_1*w31;
1650     const register double tmp56_1 = B_0_0*w31;
1651     const register double tmp49_1 = B_1_1*w36;
1652     const register double tmp2_1 = B_0_2*w31;
1653     const register double tmp23_1 = B_0_2*w33;
1654     const register double tmp38_1 = B_1_1*w43;
1655     const register double tmp74_1 = B_1_2*w41;
1656     const register double tmp43_1 = B_1_1*w41;
1657     const register double tmp58_1 = B_0_2*w28;
1658     const register double tmp67_1 = B_0_0*w33;
1659     const register double tmp33_1 = tmp0_0*w39;
1660     const register double tmp4_1 = B_0_0*w28;
1661     const register double tmp20_1 = B_0_0*w30;
1662     const register double tmp13_1 = B_0_2*w38;
1663     const register double tmp65_1 = B_1_2*w43;
1664     const register double tmp0_1 = tmp0_0*w29;
1665     const register double tmp41_1 = tmp3_0*w33;
1666     const register double tmp73_1 = B_0_2*w37;
1667     const register double tmp69_1 = B_0_0*w38;
1668     const register double tmp48_1 = B_1_2*w39;
1669     const register double tmp59_1 = B_0_1*w33;
1670     const register double tmp17_1 = B_1_3*w41;
1671     const register double tmp5_1 = B_0_3*w33;
1672     const register double tmp3_1 = B_0_1*w30;
1673     const register double tmp21_1 = B_0_1*w28;
1674     const register double tmp42_1 = B_1_0*w29;
1675     const register double tmp54_1 = B_1_2*w32;
1676     const register double tmp60_1 = B_1_0*w39;
1677     const register double tmp32_1 = tmp1_0*w36;
1678     const register double tmp10_1 = B_0_1*w37;
1679     const register double tmp14_1 = B_0_0*w35;
1680     const register double tmp29_1 = B_0_1*w35;
1681     const register double tmp26_1 = B_1_2*w36;
1682     const register double tmp30_1 = B_1_3*w43;
1683     const register double tmp70_1 = B_0_2*w35;
1684     const register double tmp34_1 = B_1_3*w39;
1685     const register double tmp51_1 = B_1_0*w43;
1686     const register double tmp31_1 = B_0_2*w34;
1687     const register double tmp45_1 = tmp3_0*w28;
1688     const register double tmp11_1 = tmp1_0*w39;
1689     const register double tmp52_1 = B_1_1*w29;
1690     const register double tmp44_1 = B_1_3*w32;
1691     const register double tmp25_1 = B_1_1*w39;
1692     const register double tmp47_1 = tmp2_0*w33;
1693     const register double tmp72_1 = B_1_3*w29;
1694     const register double tmp40_1 = tmp2_0*w28;
1695     const register double tmp46_1 = B_1_2*w40;
1696     const register double tmp36_1 = B_1_2*w42;
1697     const register double tmp24_1 = B_0_0*w37;
1698     const register double tmp77_1 = B_0_3*w35;
1699     const register double tmp68_1 = B_0_3*w37;
1700     const register double tmp78_1 = B_0_0*w34;
1701     const register double tmp12_1 = tmp0_0*w36;
1702     const register double tmp75_1 = B_1_0*w32;
1703     EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1 + tmp4_1 + tmp5_1;
1704     EM_S[INDEX2(1,2,4)]+=tmp6_1 + tmp7_1 + tmp8_1 + tmp9_1;
1705     EM_S[INDEX2(3,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1 + tmp14_1 + tmp15_1;
1706     EM_S[INDEX2(0,0,4)]+=tmp16_1 + tmp17_1 + tmp18_1 + tmp19_1 + tmp20_1 + tmp21_1 + tmp22_1 + tmp23_1;
1707     EM_S[INDEX2(3,3,4)]+=tmp24_1 + tmp25_1 + tmp26_1 + tmp27_1 + tmp28_1 + tmp29_1 + tmp30_1 + tmp31_1;
1708     EM_S[INDEX2(3,0,4)]+=tmp32_1 + tmp33_1 + tmp6_1 + tmp9_1;
1709     EM_S[INDEX2(3,1,4)]+=tmp34_1 + tmp35_1 + tmp36_1 + tmp37_1 + tmp38_1 + tmp39_1;
1710     EM_S[INDEX2(2,1,4)]+=tmp32_1 + tmp33_1 + tmp40_1 + tmp41_1;
1711     EM_S[INDEX2(0,2,4)]+=tmp42_1 + tmp43_1 + tmp44_1 + tmp45_1 + tmp46_1 + tmp47_1;
1712     EM_S[INDEX2(2,0,4)]+=tmp45_1 + tmp47_1 + tmp48_1 + tmp49_1 + tmp50_1 + tmp51_1;
1713     EM_S[INDEX2(1,3,4)]+=tmp37_1 + tmp39_1 + tmp52_1 + tmp53_1 + tmp54_1 + tmp55_1;
1714     EM_S[INDEX2(2,3,4)]+=tmp11_1 + tmp12_1 + tmp56_1 + tmp57_1 + tmp58_1 + tmp59_1;
1715     EM_S[INDEX2(2,2,4)]+=tmp60_1 + tmp61_1 + tmp62_1 + tmp63_1 + tmp64_1 + tmp65_1 + tmp66_1 + tmp67_1;
1716     EM_S[INDEX2(1,0,4)]+=tmp0_1 + tmp1_1 + tmp68_1 + tmp69_1 + tmp70_1 + tmp71_1;
1717     EM_S[INDEX2(0,3,4)]+=tmp40_1 + tmp41_1 + tmp7_1 + tmp8_1;
1718     EM_S[INDEX2(1,1,4)]+=tmp72_1 + tmp73_1 + tmp74_1 + tmp75_1 + tmp76_1 + tmp77_1 + tmp78_1 + tmp79_1;
1719     } else { /* constant data */
1720     const register double B_0 = B_p[0];
1721     const register double B_1 = B_p[1];
1722     const register double tmp6_1 = B_1*w50;
1723     const register double tmp1_1 = B_1*w45;
1724     const register double tmp5_1 = B_1*w49;
1725     const register double tmp4_1 = B_1*w48;
1726     const register double tmp0_1 = B_0*w44;
1727     const register double tmp2_1 = B_0*w46;
1728     const register double tmp7_1 = B_0*w51;
1729     const register double tmp3_1 = B_0*w47;
1730     EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1;
1731     EM_S[INDEX2(1,2,4)]+=tmp1_1 + tmp2_1;
1732     EM_S[INDEX2(3,2,4)]+=tmp3_1 + tmp4_1;
1733     EM_S[INDEX2(0,0,4)]+=tmp0_1 + tmp5_1;
1734     EM_S[INDEX2(3,3,4)]+=tmp3_1 + tmp6_1;
1735     EM_S[INDEX2(3,0,4)]+=tmp2_1 + tmp4_1;
1736     EM_S[INDEX2(3,1,4)]+=tmp2_1 + tmp6_1;
1737     EM_S[INDEX2(2,1,4)]+=tmp4_1 + tmp7_1;
1738     EM_S[INDEX2(0,2,4)]+=tmp5_1 + tmp7_1;
1739     EM_S[INDEX2(2,0,4)]+=tmp6_1 + tmp7_1;
1740     EM_S[INDEX2(1,3,4)]+=tmp2_1 + tmp5_1;
1741     EM_S[INDEX2(2,3,4)]+=tmp0_1 + tmp4_1;
1742     EM_S[INDEX2(2,2,4)]+=tmp0_1 + tmp6_1;
1743     EM_S[INDEX2(1,0,4)]+=tmp1_1 + tmp3_1;
1744     EM_S[INDEX2(0,3,4)]+=tmp1_1 + tmp7_1;
1745     EM_S[INDEX2(1,1,4)]+=tmp3_1 + tmp5_1;
1746     }
1747     }
1748     ///////////////
1749     // process C //
1750     ///////////////
1751     if (!C.isEmpty()) {
1752     add_EM_S=true;
1753     const double* C_p=const_cast<escript::Data*>(&C)->getSampleDataRO(e);
1754     if (C.actsExpanded()) {
1755     const register double C_0_0 = C_p[INDEX2(0,0,2)];
1756     const register double C_1_0 = C_p[INDEX2(1,0,2)];
1757     const register double C_0_1 = C_p[INDEX2(0,1,2)];
1758     const register double C_1_1 = C_p[INDEX2(1,1,2)];
1759     const register double C_0_2 = C_p[INDEX2(0,2,2)];
1760     const register double C_1_2 = C_p[INDEX2(1,2,2)];
1761     const register double C_0_3 = C_p[INDEX2(0,3,2)];
1762     const register double C_1_3 = C_p[INDEX2(1,3,2)];
1763     const register double tmp2_0 = C_0_1 + C_0_3;
1764     const register double tmp1_0 = C_1_2 + C_1_3;
1765     const register double tmp3_0 = C_0_0 + C_0_2;
1766     const register double tmp0_0 = C_1_0 + C_1_1;
1767     const register double tmp64_1 = C_0_2*w30;
1768     const register double tmp14_1 = C_0_2*w28;
1769     const register double tmp19_1 = C_0_3*w31;
1770     const register double tmp22_1 = C_1_0*w40;
1771     const register double tmp37_1 = tmp3_0*w35;
1772     const register double tmp29_1 = C_0_1*w35;
1773     const register double tmp73_1 = C_0_2*w37;
1774     const register double tmp74_1 = C_1_2*w41;
1775     const register double tmp52_1 = C_1_3*w39;
1776     const register double tmp25_1 = C_1_1*w39;
1777     const register double tmp62_1 = C_0_1*w31;
1778     const register double tmp79_1 = C_1_1*w40;
1779     const register double tmp43_1 = C_1_1*w36;
1780     const register double tmp27_1 = C_0_3*w38;
1781     const register double tmp28_1 = C_1_0*w42;
1782     const register double tmp63_1 = C_1_1*w42;
1783     const register double tmp59_1 = C_0_3*w34;
1784     const register double tmp72_1 = C_1_3*w29;
1785     const register double tmp40_1 = tmp2_0*w35;
1786     const register double tmp13_1 = C_0_3*w30;
1787     const register double tmp51_1 = C_1_2*w40;
1788     const register double tmp54_1 = C_1_2*w42;
1789     const register double tmp12_1 = C_0_0*w31;
1790     const register double tmp2_1 = tmp1_0*w32;
1791     const register double tmp68_1 = C_0_2*w31;
1792     const register double tmp75_1 = C_1_0*w32;
1793     const register double tmp49_1 = C_1_1*w41;
1794     const register double tmp4_1 = C_0_2*w35;
1795     const register double tmp66_1 = C_0_3*w28;
1796     const register double tmp56_1 = C_0_1*w37;
1797     const register double tmp5_1 = C_0_1*w34;
1798     const register double tmp38_1 = tmp2_0*w34;
1799     const register double tmp76_1 = C_0_1*w38;
1800     const register double tmp21_1 = C_0_1*w28;
1801     const register double tmp69_1 = C_0_1*w30;
1802     const register double tmp53_1 = C_1_0*w36;
1803     const register double tmp42_1 = C_1_2*w39;
1804     const register double tmp32_1 = tmp1_0*w29;
1805     const register double tmp45_1 = C_1_0*w43;
1806     const register double tmp33_1 = tmp0_0*w32;
1807     const register double tmp35_1 = C_1_0*w41;
1808     const register double tmp26_1 = C_1_2*w36;
1809     const register double tmp67_1 = C_0_0*w33;
1810     const register double tmp31_1 = C_0_2*w34;
1811     const register double tmp20_1 = C_0_0*w30;
1812     const register double tmp70_1 = C_0_0*w28;
1813     const register double tmp8_1 = tmp0_0*w39;
1814     const register double tmp30_1 = C_1_3*w43;
1815     const register double tmp0_1 = tmp0_0*w29;
1816     const register double tmp17_1 = C_1_3*w41;
1817     const register double tmp58_1 = C_0_0*w35;
1818     const register double tmp9_1 = tmp3_0*w33;
1819     const register double tmp61_1 = C_1_3*w36;
1820     const register double tmp41_1 = tmp3_0*w34;
1821     const register double tmp50_1 = C_1_3*w32;
1822     const register double tmp18_1 = C_1_1*w32;
1823     const register double tmp6_1 = tmp1_0*w36;
1824     const register double tmp3_1 = C_0_0*w38;
1825     const register double tmp34_1 = C_1_1*w29;
1826     const register double tmp77_1 = C_0_3*w35;
1827     const register double tmp65_1 = C_1_2*w43;
1828     const register double tmp71_1 = C_0_3*w33;
1829     const register double tmp55_1 = C_1_1*w43;
1830     const register double tmp46_1 = tmp3_0*w28;
1831     const register double tmp24_1 = C_0_0*w37;
1832     const register double tmp10_1 = tmp1_0*w39;
1833     const register double tmp48_1 = C_1_0*w29;
1834     const register double tmp15_1 = C_0_1*w33;
1835     const register double tmp36_1 = C_1_2*w32;
1836     const register double tmp60_1 = C_1_0*w39;
1837     const