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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 4353 - (hide annotations)
Fri Apr 5 00:14:35 2013 UTC (6 years, 6 months ago) by caltinay
File size: 243040 byte(s)
For some reason std::isnan<float> is not available on NCI so changed this
to ::isnan. Added options file for NCI vayu.

1 caltinay 3691
2 jfenwick 3981 /*****************************************************************************
3 caltinay 3691 *
4 jfenwick 4154 * Copyright (c) 2003-2013 by University of Queensland
5 jfenwick 3981 * http://www.uq.edu.au
6 caltinay 3691 *
7     * Primary Business: Queensland, Australia
8     * Licensed under the Open Software License version 3.0
9     * http://www.opensource.org/licenses/osl-3.0.php
10     *
11 jfenwick 3981 * Development until 2012 by Earth Systems Science Computational Center (ESSCC)
12     * Development since 2012 by School of Earth Sciences
13     *
14     *****************************************************************************/
15 caltinay 3691
16     #include <ripley/Rectangle.h>
17 caltinay 3791 #include <paso/SystemMatrix.h>
18 caltinay 4334 #include <esysUtils/esysFileWriter.h>
19 caltinay 3691
20 caltinay 4013 #ifdef USE_NETCDF
21     #include <netcdfcpp.h>
22     #endif
23    
24 caltinay 3691 #if USE_SILO
25     #include <silo.h>
26     #ifdef ESYS_MPI
27     #include <pmpio.h>
28     #endif
29     #endif
30    
31     #include <iomanip>
32    
33     using namespace std;
34 caltinay 4334 using esysUtils::FileWriter;
35 caltinay 3691
36     namespace ripley {
37    
38 caltinay 3781 Rectangle::Rectangle(int n0, int n1, double x0, double y0, double x1,
39     double y1, int d0, int d1) :
40 caltinay 4334 RipleyDomain(2)
41 caltinay 3691 {
42 caltinay 3943 // ignore subdivision parameters for serial run
43     if (m_mpiInfo->size == 1) {
44     d0=1;
45     d1=1;
46     }
47    
48     bool warn=false;
49     // if number of subdivisions is non-positive, try to subdivide by the same
50     // ratio as the number of elements
51     if (d0<=0 && d1<=0) {
52     warn=true;
53 caltinay 4010 d0=max(1, (int)(sqrt(m_mpiInfo->size*(n0+1)/(float)(n1+1))));
54 caltinay 3943 d1=m_mpiInfo->size/d0;
55     if (d0*d1 != m_mpiInfo->size) {
56     // ratios not the same so subdivide side with more elements only
57     if (n0>n1) {
58     d0=0;
59     d1=1;
60     } else {
61     d0=1;
62     d1=0;
63     }
64     }
65     }
66     if (d0<=0) {
67     // d1 is preset, determine d0 - throw further down if result is no good
68     d0=m_mpiInfo->size/d1;
69     } else if (d1<=0) {
70     // d0 is preset, determine d1 - throw further down if result is no good
71     d1=m_mpiInfo->size/d0;
72     }
73    
74 caltinay 3691 // ensure number of subdivisions is valid and nodes can be distributed
75     // among number of ranks
76 caltinay 4334 if (d0*d1 != m_mpiInfo->size)
77 caltinay 3691 throw RipleyException("Invalid number of spatial subdivisions");
78    
79 caltinay 3943 if (warn) {
80     cout << "Warning: Automatic domain subdivision (d0=" << d0 << ", d1="
81     << d1 << "). This may not be optimal!" << endl;
82     }
83 caltinay 3691
84 caltinay 4334 double l0 = x1-x0;
85     double l1 = y1-y0;
86     m_dx[0] = l0/n0;
87     m_dx[1] = l1/n1;
88    
89     if ((n0+1)%d0 > 0) {
90 caltinay 3943 n0=(int)round((float)(n0+1)/d0+0.5)*d0-1;
91 caltinay 4334 l0=m_dx[0]*n0;
92 caltinay 3943 cout << "Warning: Adjusted number of elements and length. N0="
93 caltinay 4334 << n0 << ", l0=" << l0 << endl;
94 caltinay 3943 }
95 caltinay 4334 if ((n1+1)%d1 > 0) {
96 caltinay 3943 n1=(int)round((float)(n1+1)/d1+0.5)*d1-1;
97 caltinay 4334 l1=m_dx[1]*n1;
98 caltinay 3943 cout << "Warning: Adjusted number of elements and length. N1="
99 caltinay 4334 << n1 << ", l1=" << l1 << endl;
100 caltinay 3943 }
101    
102 caltinay 4334 if ((d0 > 1 && (n0+1)/d0<2) || (d1 > 1 && (n1+1)/d1<2))
103 caltinay 3752 throw RipleyException("Too few elements for the number of ranks");
104    
105 caltinay 4334 m_gNE[0] = n0;
106     m_gNE[1] = n1;
107     m_origin[0] = x0;
108     m_origin[1] = y0;
109     m_length[0] = l0;
110     m_length[1] = l1;
111     m_NX[0] = d0;
112     m_NX[1] = d1;
113    
114 caltinay 3764 // local number of elements (with and without overlap)
115 caltinay 4334 m_NE[0] = m_ownNE[0] = (d0>1 ? (n0+1)/d0 : n0);
116     if (m_mpiInfo->rank%d0>0 && m_mpiInfo->rank%d0<d0-1)
117     m_NE[0]++;
118     else if (d0>1 && m_mpiInfo->rank%d0==d0-1)
119     m_ownNE[0]--;
120 caltinay 3764
121 caltinay 4334 m_NE[1] = m_ownNE[1] = (d1>1 ? (n1+1)/d1 : n1);
122     if (m_mpiInfo->rank/d0>0 && m_mpiInfo->rank/d0<d1-1)
123     m_NE[1]++;
124     else if (d1>1 && m_mpiInfo->rank/d0==d1-1)
125     m_ownNE[1]--;
126 caltinay 3752
127     // local number of nodes
128 caltinay 4334 m_NN[0] = m_NE[0]+1;
129     m_NN[1] = m_NE[1]+1;
130 caltinay 3752
131 caltinay 3691 // bottom-left node is at (offset0,offset1) in global mesh
132 caltinay 4334 m_offset[0] = (n0+1)/d0*(m_mpiInfo->rank%d0);
133     if (m_offset[0] > 0)
134     m_offset[0]--;
135     m_offset[1] = (n1+1)/d1*(m_mpiInfo->rank/d0);
136     if (m_offset[1] > 0)
137     m_offset[1]--;
138 caltinay 3752
139 caltinay 3691 populateSampleIds();
140 caltinay 3756 createPattern();
141 caltinay 3691 }
142    
143     Rectangle::~Rectangle()
144     {
145 caltinay 3785 Paso_SystemMatrixPattern_free(m_pattern);
146     Paso_Connector_free(m_connector);
147 caltinay 3691 }
148    
149     string Rectangle::getDescription() const
150     {
151     return "ripley::Rectangle";
152     }
153    
154     bool Rectangle::operator==(const AbstractDomain& other) const
155     {
156 caltinay 3744 const Rectangle* o=dynamic_cast<const Rectangle*>(&other);
157     if (o) {
158     return (RipleyDomain::operator==(other) &&
159 caltinay 4334 m_gNE[0]==o->m_gNE[0] && m_gNE[1]==o->m_gNE[1]
160     && m_origin[0]==o->m_origin[0] && m_origin[1]==o->m_origin[1]
161     && m_length[0]==o->m_length[0] && m_length[1]==o->m_length[1]
162     && m_NX[0]==o->m_NX[0] && m_NX[1]==o->m_NX[1]);
163 caltinay 3744 }
164 caltinay 3691
165     return false;
166     }
167    
168 caltinay 4013 void Rectangle::readNcGrid(escript::Data& out, string filename, string varname,
169 caltinay 4277 const vector<int>& first, const vector<int>& numValues,
170     const vector<int>& multiplier) const
171 caltinay 4013 {
172     #ifdef USE_NETCDF
173     // check destination function space
174     int myN0, myN1;
175     if (out.getFunctionSpace().getTypeCode() == Nodes) {
176 caltinay 4334 myN0 = m_NN[0];
177     myN1 = m_NN[1];
178 caltinay 4013 } else if (out.getFunctionSpace().getTypeCode() == Elements ||
179     out.getFunctionSpace().getTypeCode() == ReducedElements) {
180 caltinay 4334 myN0 = m_NE[0];
181     myN1 = m_NE[1];
182 caltinay 4013 } else
183     throw RipleyException("readNcGrid(): invalid function space for output data object");
184    
185     if (first.size() != 2)
186     throw RipleyException("readNcGrid(): argument 'first' must have 2 entries");
187    
188     if (numValues.size() != 2)
189     throw RipleyException("readNcGrid(): argument 'numValues' must have 2 entries");
190    
191 caltinay 4277 if (multiplier.size() != 2)
192     throw RipleyException("readNcGrid(): argument 'multiplier' must have 2 entries");
193     for (size_t i=0; i<multiplier.size(); i++)
194     if (multiplier[i]<1)
195     throw RipleyException("readNcGrid(): all multipliers must be positive");
196    
197 caltinay 4013 // check file existence and size
198     NcFile f(filename.c_str(), NcFile::ReadOnly);
199     if (!f.is_valid())
200     throw RipleyException("readNcGrid(): cannot open file");
201    
202     NcVar* var = f.get_var(varname.c_str());
203     if (!var)
204     throw RipleyException("readNcGrid(): invalid variable");
205    
206     // TODO: rank>0 data support
207     const int numComp = out.getDataPointSize();
208     if (numComp > 1)
209     throw RipleyException("readNcGrid(): only scalar data supported");
210    
211     const int dims = var->num_dims();
212     const long *edges = var->edges();
213    
214     // is this a slice of the data object (dims!=2)?
215     // note the expected ordering of edges (as in numpy: y,x)
216     if ( (dims==2 && (numValues[1] > edges[0] || numValues[0] > edges[1]))
217     || (dims==1 && numValues[1]>1) ) {
218     throw RipleyException("readNcGrid(): not enough data in file");
219     }
220    
221     // check if this rank contributes anything
222 caltinay 4334 if (first[0] >= m_offset[0]+myN0 || first[0]+numValues[0]*multiplier[0] <= m_offset[0] ||
223     first[1] >= m_offset[1]+myN1 || first[1]+numValues[1]*multiplier[1] <= m_offset[1])
224 caltinay 4013 return;
225    
226     // now determine how much this rank has to write
227    
228     // first coordinates in data object to write to
229 caltinay 4334 const int first0 = max(0, first[0]-m_offset[0]);
230     const int first1 = max(0, first[1]-m_offset[1]);
231 caltinay 4013 // indices to first value in file
232 caltinay 4334 const int idx0 = max(0, m_offset[0]-first[0]);
233     const int idx1 = max(0, m_offset[1]-first[1]);
234 caltinay 4277 // number of values to read
235 caltinay 4013 const int num0 = min(numValues[0]-idx0, myN0-first0);
236     const int num1 = min(numValues[1]-idx1, myN1-first1);
237    
238     vector<double> values(num0*num1);
239     if (dims==2) {
240     var->set_cur(idx1, idx0);
241     var->get(&values[0], num1, num0);
242     } else {
243     var->set_cur(idx0);
244     var->get(&values[0], num0);
245     }
246    
247     const int dpp = out.getNumDataPointsPerSample();
248     out.requireWrite();
249    
250     for (index_t y=0; y<num1; y++) {
251     #pragma omp parallel for
252     for (index_t x=0; x<num0; x++) {
253 caltinay 4277 const int baseIndex = first0+x*multiplier[0]
254     +(first1+y*multiplier[1])*myN0;
255     const int srcIndex = y*num0+x;
256 caltinay 4174 if (!isnan(values[srcIndex])) {
257 caltinay 4277 for (index_t m1=0; m1<multiplier[1]; m1++) {
258     for (index_t m0=0; m0<multiplier[0]; m0++) {
259     const int dataIndex = baseIndex+m0+m1*myN0;
260     double* dest = out.getSampleDataRW(dataIndex);
261     for (index_t q=0; q<dpp; q++) {
262     *dest++ = values[srcIndex];
263     }
264     }
265 caltinay 4174 }
266 caltinay 4013 }
267     }
268     }
269     #else
270     throw RipleyException("readNcGrid(): not compiled with netCDF support");
271     #endif
272     }
273    
274 caltinay 3971 void Rectangle::readBinaryGrid(escript::Data& out, string filename,
275 caltinay 4277 const vector<int>& first,
276     const vector<int>& numValues,
277     const vector<int>& multiplier) const
278 caltinay 3971 {
279     // check destination function space
280     int myN0, myN1;
281     if (out.getFunctionSpace().getTypeCode() == Nodes) {
282 caltinay 4334 myN0 = m_NN[0];
283     myN1 = m_NN[1];
284 caltinay 3971 } else if (out.getFunctionSpace().getTypeCode() == Elements ||
285     out.getFunctionSpace().getTypeCode() == ReducedElements) {
286 caltinay 4334 myN0 = m_NE[0];
287     myN1 = m_NE[1];
288 caltinay 3971 } else
289     throw RipleyException("readBinaryGrid(): invalid function space for output data object");
290    
291     // check file existence and size
292     ifstream f(filename.c_str(), ifstream::binary);
293     if (f.fail()) {
294     throw RipleyException("readBinaryGrid(): cannot open file");
295     }
296     f.seekg(0, ios::end);
297     const int numComp = out.getDataPointSize();
298     const int filesize = f.tellg();
299     const int reqsize = numValues[0]*numValues[1]*numComp*sizeof(float);
300     if (filesize < reqsize) {
301     f.close();
302     throw RipleyException("readBinaryGrid(): not enough data in file");
303     }
304    
305     // check if this rank contributes anything
306 caltinay 4334 if (first[0] >= m_offset[0]+myN0 || first[0]+numValues[0] <= m_offset[0] ||
307     first[1] >= m_offset[1]+myN1 || first[1]+numValues[1] <= m_offset[1]) {
308 caltinay 3971 f.close();
309     return;
310     }
311    
312     // now determine how much this rank has to write
313    
314     // first coordinates in data object to write to
315 caltinay 4334 const int first0 = max(0, first[0]-m_offset[0]);
316     const int first1 = max(0, first[1]-m_offset[1]);
317 caltinay 3971 // indices to first value in file
318 caltinay 4334 const int idx0 = max(0, m_offset[0]-first[0]);
319     const int idx1 = max(0, m_offset[1]-first[1]);
320 caltinay 4277 // number of values to read
321 caltinay 3971 const int num0 = min(numValues[0]-idx0, myN0-first0);
322     const int num1 = min(numValues[1]-idx1, myN1-first1);
323    
324     out.requireWrite();
325     vector<float> values(num0*numComp);
326     const int dpp = out.getNumDataPointsPerSample();
327    
328     for (index_t y=0; y<num1; y++) {
329     const int fileofs = numComp*(idx0+(idx1+y)*numValues[0]);
330     f.seekg(fileofs*sizeof(float));
331     f.read((char*)&values[0], num0*numComp*sizeof(float));
332     for (index_t x=0; x<num0; x++) {
333 caltinay 4277 const int baseIndex = first0+x*multiplier[0]
334     +(first1+y*multiplier[1])*myN0;
335     for (index_t m1=0; m1<multiplier[1]; m1++) {
336     for (index_t m0=0; m0<multiplier[0]; m0++) {
337     const int dataIndex = baseIndex+m0+m1*myN0;
338     double* dest = out.getSampleDataRW(dataIndex);
339     for (index_t c=0; c<numComp; c++) {
340 caltinay 4353 if (!::isnan(values[x*numComp+c])) {
341 caltinay 4277 for (index_t q=0; q<dpp; q++) {
342     *dest++ = static_cast<double>(values[x*numComp+c]);
343     }
344     }
345 caltinay 4174 }
346 caltinay 3971 }
347     }
348     }
349     }
350    
351     f.close();
352     }
353    
354 caltinay 4334 void Rectangle::writeBinaryGrid(const escript::Data& in, string filename, int byteOrder) const
355     {
356     // check function space and determine number of points
357     int myN0, myN1;
358     int totalN0, totalN1;
359     if (in.getFunctionSpace().getTypeCode() == Nodes) {
360     myN0 = m_NN[0];
361     myN1 = m_NN[1];
362     totalN0 = m_gNE[0]+1;
363     totalN1 = m_gNE[1]+1;
364     } else if (in.getFunctionSpace().getTypeCode() == Elements ||
365     in.getFunctionSpace().getTypeCode() == ReducedElements) {
366     myN0 = m_NE[0];
367     myN1 = m_NE[1];
368     totalN0 = m_gNE[0];
369     totalN1 = m_gNE[1];
370     } else
371     throw RipleyException("writeBinaryGrid(): invalid function space of data object");
372    
373     const int numComp = in.getDataPointSize();
374     const int dpp = in.getNumDataPointsPerSample();
375     const int fileSize = sizeof(float)*numComp*dpp*totalN0*totalN1;
376    
377     if (numComp > 1 || dpp > 1)
378     throw RipleyException("writeBinaryGrid(): only scalar, single-value data supported");
379    
380     escript::Data* _in = const_cast<escript::Data*>(&in);
381    
382     // from here on we know that each sample consists of one value
383     FileWriter* fw = new FileWriter();
384     fw->openFile(filename, fileSize);
385     MPIBarrier();
386    
387     for (index_t y=0; y<myN1; y++) {
388     const int fileofs = (m_offset[0]+(m_offset[1]+y)*totalN0)*sizeof(float);
389     ostringstream oss;
390    
391     for (index_t x=0; x<myN0; x++) {
392     const double* sample = _in->getSampleDataRO(y*myN0+x);
393     float fvalue = (float)(*sample);
394     if (byteOrder == RIPLEY_BYTE_ORDER) {
395     oss.write((char*)&fvalue, sizeof(fvalue));
396     } else {
397     char* value = reinterpret_cast<char*>(&fvalue);
398     oss.write(RIPLEY_BYTE_SWAP32(value), sizeof(fvalue));
399     }
400     }
401     fw->writeAt(oss, fileofs);
402     }
403     fw->close();
404     }
405    
406 caltinay 3691 void Rectangle::dump(const string& fileName) const
407     {
408     #if USE_SILO
409     string fn(fileName);
410     if (fileName.length() < 6 || fileName.compare(fileName.length()-5, 5, ".silo") != 0) {
411     fn+=".silo";
412     }
413    
414     int driver=DB_HDF5;
415     DBfile* dbfile = NULL;
416 gross 3793 const char* blockDirFmt = "/block%04d";
417 caltinay 3691
418     #ifdef ESYS_MPI
419     PMPIO_baton_t* baton = NULL;
420 gross 3793 const int NUM_SILO_FILES = 1;
421 caltinay 3691 #endif
422    
423     if (m_mpiInfo->size > 1) {
424     #ifdef ESYS_MPI
425     baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
426     0x1337, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
427     PMPIO_DefaultClose, (void*)&driver);
428     // try the fallback driver in case of error
429     if (!baton && driver != DB_PDB) {
430     driver = DB_PDB;
431     baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
432     0x1338, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
433     PMPIO_DefaultClose, (void*)&driver);
434     }
435     if (baton) {
436 caltinay 3766 char siloPath[64];
437     snprintf(siloPath, 64, blockDirFmt, PMPIO_RankInGroup(baton, m_mpiInfo->rank));
438     dbfile = (DBfile*) PMPIO_WaitForBaton(baton, fn.c_str(), siloPath);
439 caltinay 3691 }
440     #endif
441     } else {
442     dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
443     getDescription().c_str(), driver);
444     // try the fallback driver in case of error
445     if (!dbfile && driver != DB_PDB) {
446     driver = DB_PDB;
447     dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
448     getDescription().c_str(), driver);
449     }
450 caltinay 3766 char siloPath[64];
451     snprintf(siloPath, 64, blockDirFmt, 0);
452     DBMkDir(dbfile, siloPath);
453     DBSetDir(dbfile, siloPath);
454 caltinay 3691 }
455    
456     if (!dbfile)
457     throw RipleyException("dump: Could not create Silo file");
458    
459     /*
460     if (driver==DB_HDF5) {
461     // gzip level 1 already provides good compression with minimal
462     // performance penalty. Some tests showed that gzip levels >3 performed
463     // rather badly on escript data both in terms of time and space
464     DBSetCompression("ERRMODE=FALLBACK METHOD=GZIP LEVEL=1");
465     }
466     */
467    
468 caltinay 4334 boost::scoped_ptr<double> x(new double[m_NN[0]]);
469     boost::scoped_ptr<double> y(new double[m_NN[1]]);
470 caltinay 3691 double* coords[2] = { x.get(), y.get() };
471     #pragma omp parallel
472     {
473 caltinay 3722 #pragma omp for nowait
474 caltinay 4334 for (dim_t i0 = 0; i0 < m_NN[0]; i0++) {
475     coords[0][i0]=getLocalCoordinate(i0, 0);
476 caltinay 3691 }
477 caltinay 3722 #pragma omp for nowait
478 caltinay 4334 for (dim_t i1 = 0; i1 < m_NN[1]; i1++) {
479     coords[1][i1]=getLocalCoordinate(i1, 1);
480 caltinay 3691 }
481     }
482 caltinay 4334 int* dims = const_cast<int*>(getNumNodesPerDim());
483 caltinay 3697
484     // write mesh
485 caltinay 4334 DBPutQuadmesh(dbfile, "mesh", NULL, coords, dims, 2, DB_DOUBLE,
486 caltinay 3691 DB_COLLINEAR, NULL);
487    
488 caltinay 3697 // write node ids
489 caltinay 4334 DBPutQuadvar1(dbfile, "nodeId", "mesh", (void*)&m_nodeId[0], dims, 2,
490 caltinay 3697 NULL, 0, DB_INT, DB_NODECENT, NULL);
491    
492     // write element ids
493 caltinay 4334 dims = const_cast<int*>(getNumElementsPerDim());
494 caltinay 3697 DBPutQuadvar1(dbfile, "elementId", "mesh", (void*)&m_elementId[0],
495 caltinay 4334 dims, 2, NULL, 0, DB_INT, DB_ZONECENT, NULL);
496 caltinay 3697
497     // rank 0 writes multimesh and multivar
498 caltinay 3691 if (m_mpiInfo->rank == 0) {
499     vector<string> tempstrings;
500 caltinay 3697 vector<char*> names;
501 caltinay 3691 for (dim_t i=0; i<m_mpiInfo->size; i++) {
502     stringstream path;
503     path << "/block" << setw(4) << setfill('0') << right << i << "/mesh";
504     tempstrings.push_back(path.str());
505 caltinay 3697 names.push_back((char*)tempstrings.back().c_str());
506 caltinay 3691 }
507 caltinay 3697 vector<int> types(m_mpiInfo->size, DB_QUAD_RECT);
508 caltinay 3691 DBSetDir(dbfile, "/");
509 caltinay 3697 DBPutMultimesh(dbfile, "multimesh", m_mpiInfo->size, &names[0],
510     &types[0], NULL);
511     tempstrings.clear();
512     names.clear();
513     for (dim_t i=0; i<m_mpiInfo->size; i++) {
514     stringstream path;
515     path << "/block" << setw(4) << setfill('0') << right << i << "/nodeId";
516     tempstrings.push_back(path.str());
517     names.push_back((char*)tempstrings.back().c_str());
518     }
519     types.assign(m_mpiInfo->size, DB_QUADVAR);
520     DBPutMultivar(dbfile, "nodeId", m_mpiInfo->size, &names[0],
521     &types[0], NULL);
522     tempstrings.clear();
523     names.clear();
524     for (dim_t i=0; i<m_mpiInfo->size; i++) {
525     stringstream path;
526     path << "/block" << setw(4) << setfill('0') << right << i << "/elementId";
527     tempstrings.push_back(path.str());
528     names.push_back((char*)tempstrings.back().c_str());
529     }
530     DBPutMultivar(dbfile, "elementId", m_mpiInfo->size, &names[0],
531     &types[0], NULL);
532 caltinay 3691 }
533    
534     if (m_mpiInfo->size > 1) {
535     #ifdef ESYS_MPI
536     PMPIO_HandOffBaton(baton, dbfile);
537     PMPIO_Finish(baton);
538     #endif
539     } else {
540     DBClose(dbfile);
541     }
542    
543     #else // USE_SILO
544 caltinay 3791 throw RipleyException("dump: no Silo support");
545 caltinay 3691 #endif
546     }
547    
548 caltinay 3697 const int* Rectangle::borrowSampleReferenceIDs(int fsType) const
549 caltinay 3691 {
550 caltinay 3697 switch (fsType) {
551 caltinay 3691 case Nodes:
552 caltinay 3769 case ReducedNodes: // FIXME: reduced
553 caltinay 3691 return &m_nodeId[0];
554 caltinay 3750 case DegreesOfFreedom:
555 caltinay 3769 case ReducedDegreesOfFreedom: // FIXME: reduced
556 caltinay 3750 return &m_dofId[0];
557 caltinay 3691 case Elements:
558 caltinay 3733 case ReducedElements:
559 caltinay 3691 return &m_elementId[0];
560     case FaceElements:
561 caltinay 3733 case ReducedFaceElements:
562 caltinay 3691 return &m_faceId[0];
563     default:
564     break;
565     }
566    
567     stringstream msg;
568 caltinay 3791 msg << "borrowSampleReferenceIDs: invalid function space type " << fsType;
569 caltinay 3691 throw RipleyException(msg.str());
570     }
571    
572 caltinay 3757 bool Rectangle::ownSample(int fsType, index_t id) const
573 caltinay 3691 {
574 caltinay 3759 if (getMPISize()==1)
575     return true;
576    
577 caltinay 3757 switch (fsType) {
578     case Nodes:
579 caltinay 3769 case ReducedNodes: // FIXME: reduced
580 caltinay 3757 return (m_dofMap[id] < getNumDOF());
581     case DegreesOfFreedom:
582     case ReducedDegreesOfFreedom:
583     return true;
584     case Elements:
585     case ReducedElements:
586     // check ownership of element's bottom left node
587 caltinay 4334 return (m_dofMap[id%m_NE[0]+m_NN[0]*(id/m_NE[0])] < getNumDOF());
588 caltinay 3757 case FaceElements:
589     case ReducedFaceElements:
590 caltinay 3759 {
591 caltinay 3764 // determine which face the sample belongs to before
592 caltinay 3768 // checking ownership of corresponding element's first node
593 caltinay 3759 dim_t n=0;
594 caltinay 4334 for (size_t i=0; i<4; i++) {
595     n+=m_faceCount[i];
596 caltinay 3759 if (id<n) {
597     index_t k;
598     if (i==1)
599 caltinay 4334 k=m_NN[0]-2;
600 caltinay 3759 else if (i==3)
601 caltinay 4334 k=m_NN[0]*(m_NN[1]-2);
602 caltinay 3759 else
603     k=0;
604     // determine whether to move right or up
605 caltinay 4334 const index_t delta=(i/2==0 ? m_NN[0] : 1);
606     return (m_dofMap[k+(id-n+m_faceCount[i])*delta] < getNumDOF());
607 caltinay 3759 }
608     }
609     return false;
610     }
611 caltinay 3757 default:
612     break;
613 caltinay 3702 }
614 caltinay 3757
615     stringstream msg;
616 caltinay 3791 msg << "ownSample: invalid function space type " << fsType;
617 caltinay 3757 throw RipleyException(msg.str());
618 caltinay 3691 }
619    
620 caltinay 3764 void Rectangle::setToNormal(escript::Data& out) const
621 caltinay 3691 {
622 caltinay 3764 if (out.getFunctionSpace().getTypeCode() == FaceElements) {
623     out.requireWrite();
624     #pragma omp parallel
625     {
626     if (m_faceOffset[0] > -1) {
627     #pragma omp for nowait
628 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
629 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
630     // set vector at two quadrature points
631     *o++ = -1.;
632     *o++ = 0.;
633     *o++ = -1.;
634     *o = 0.;
635     }
636     }
637    
638     if (m_faceOffset[1] > -1) {
639     #pragma omp for nowait
640 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
641 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
642     // set vector at two quadrature points
643     *o++ = 1.;
644     *o++ = 0.;
645     *o++ = 1.;
646     *o = 0.;
647     }
648     }
649    
650     if (m_faceOffset[2] > -1) {
651     #pragma omp for nowait
652 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
653 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
654     // set vector at two quadrature points
655     *o++ = 0.;
656     *o++ = -1.;
657     *o++ = 0.;
658     *o = -1.;
659     }
660     }
661    
662     if (m_faceOffset[3] > -1) {
663     #pragma omp for nowait
664 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
665 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
666     // set vector at two quadrature points
667     *o++ = 0.;
668     *o++ = 1.;
669     *o++ = 0.;
670     *o = 1.;
671     }
672     }
673     } // end of parallel section
674     } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
675     out.requireWrite();
676     #pragma omp parallel
677     {
678     if (m_faceOffset[0] > -1) {
679     #pragma omp for nowait
680 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
681 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
682     *o++ = -1.;
683     *o = 0.;
684     }
685     }
686    
687     if (m_faceOffset[1] > -1) {
688     #pragma omp for nowait
689 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
690 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
691     *o++ = 1.;
692     *o = 0.;
693     }
694     }
695    
696     if (m_faceOffset[2] > -1) {
697     #pragma omp for nowait
698 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
699 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
700     *o++ = 0.;
701     *o = -1.;
702     }
703     }
704    
705     if (m_faceOffset[3] > -1) {
706     #pragma omp for nowait
707 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
708 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
709     *o++ = 0.;
710     *o = 1.;
711     }
712     }
713     } // end of parallel section
714    
715     } else {
716     stringstream msg;
717 caltinay 3791 msg << "setToNormal: invalid function space type "
718     << out.getFunctionSpace().getTypeCode();
719 caltinay 3764 throw RipleyException(msg.str());
720     }
721     }
722    
723     void Rectangle::setToSize(escript::Data& out) const
724     {
725     if (out.getFunctionSpace().getTypeCode() == Elements
726     || out.getFunctionSpace().getTypeCode() == ReducedElements) {
727     out.requireWrite();
728     const dim_t numQuad=out.getNumDataPointsPerSample();
729 caltinay 4334 const double size=sqrt(m_dx[0]*m_dx[0]+m_dx[1]*m_dx[1]);
730 caltinay 3764 #pragma omp parallel for
731     for (index_t k = 0; k < getNumElements(); ++k) {
732     double* o = out.getSampleDataRW(k);
733     fill(o, o+numQuad, size);
734     }
735     } else if (out.getFunctionSpace().getTypeCode() == FaceElements
736     || out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
737     out.requireWrite();
738     const dim_t numQuad=out.getNumDataPointsPerSample();
739     #pragma omp parallel
740     {
741     if (m_faceOffset[0] > -1) {
742     #pragma omp for nowait
743 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
744 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
745 caltinay 4334 fill(o, o+numQuad, m_dx[1]);
746 caltinay 3764 }
747     }
748    
749     if (m_faceOffset[1] > -1) {
750     #pragma omp for nowait
751 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
752 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
753 caltinay 4334 fill(o, o+numQuad, m_dx[1]);
754 caltinay 3764 }
755     }
756    
757     if (m_faceOffset[2] > -1) {
758     #pragma omp for nowait
759 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
760 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
761 caltinay 4334 fill(o, o+numQuad, m_dx[0]);
762 caltinay 3764 }
763     }
764    
765     if (m_faceOffset[3] > -1) {
766     #pragma omp for nowait
767 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
768 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
769 caltinay 4334 fill(o, o+numQuad, m_dx[0]);
770 caltinay 3764 }
771     }
772     } // end of parallel section
773    
774     } else {
775     stringstream msg;
776 caltinay 3791 msg << "setToSize: invalid function space type "
777     << out.getFunctionSpace().getTypeCode();
778 caltinay 3764 throw RipleyException(msg.str());
779     }
780     }
781    
782     void Rectangle::Print_Mesh_Info(const bool full) const
783     {
784     RipleyDomain::Print_Mesh_Info(full);
785     if (full) {
786     cout << " Id Coordinates" << endl;
787     cout.precision(15);
788     cout.setf(ios::scientific, ios::floatfield);
789     for (index_t i=0; i < getNumNodes(); i++) {
790     cout << " " << setw(5) << m_nodeId[i]
791 caltinay 4334 << " " << getLocalCoordinate(i%m_NN[0], 0)
792     << " " << getLocalCoordinate(i/m_NN[0], 1) << endl;
793 caltinay 3764 }
794     }
795     }
796    
797    
798     //protected
799     void Rectangle::assembleCoordinates(escript::Data& arg) const
800     {
801     escriptDataC x = arg.getDataC();
802     int numDim = m_numDim;
803     if (!isDataPointShapeEqual(&x, 1, &numDim))
804     throw RipleyException("setToX: Invalid Data object shape");
805     if (!numSamplesEqual(&x, 1, getNumNodes()))
806     throw RipleyException("setToX: Illegal number of samples in Data object");
807    
808     arg.requireWrite();
809     #pragma omp parallel for
810 caltinay 4334 for (dim_t i1 = 0; i1 < m_NN[1]; i1++) {
811     for (dim_t i0 = 0; i0 < m_NN[0]; i0++) {
812     double* point = arg.getSampleDataRW(i0+m_NN[0]*i1);
813     point[0] = getLocalCoordinate(i0, 0);
814     point[1] = getLocalCoordinate(i1, 1);
815 caltinay 3764 }
816     }
817     }
818    
819     //protected
820     void Rectangle::assembleGradient(escript::Data& out, escript::Data& in) const
821     {
822 caltinay 3701 const dim_t numComp = in.getDataPointSize();
823 caltinay 4334 const double cx0 = -1./m_dx[0];
824     const double cx1 = -.78867513459481288225/m_dx[0];
825     const double cx2 = -.5/m_dx[0];
826     const double cx3 = -.21132486540518711775/m_dx[0];
827     const double cx4 = .21132486540518711775/m_dx[0];
828     const double cx5 = .5/m_dx[0];
829     const double cx6 = .78867513459481288225/m_dx[0];
830     const double cx7 = 1./m_dx[0];
831     const double cy0 = -1./m_dx[1];
832     const double cy1 = -.78867513459481288225/m_dx[1];
833     const double cy2 = -.5/m_dx[1];
834     const double cy3 = -.21132486540518711775/m_dx[1];
835     const double cy4 = .21132486540518711775/m_dx[1];
836     const double cy5 = .5/m_dx[1];
837     const double cy6 = .78867513459481288225/m_dx[1];
838     const double cy7 = 1./m_dx[1];
839 caltinay 3724
840 caltinay 3702 if (out.getFunctionSpace().getTypeCode() == Elements) {
841 caltinay 3760 out.requireWrite();
842 caltinay 3913 #pragma omp parallel
843     {
844     vector<double> f_00(numComp);
845     vector<double> f_01(numComp);
846     vector<double> f_10(numComp);
847     vector<double> f_11(numComp);
848     #pragma omp for
849 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
850     for (index_t k0=0; k0 < m_NE[0]; ++k0) {
851     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
852     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
853     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
854     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
855     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
856 caltinay 3913 for (index_t i=0; i < numComp; ++i) {
857     o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
858     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
859     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
860     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
861     o[INDEX3(i,0,2,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
862     o[INDEX3(i,1,2,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
863     o[INDEX3(i,0,3,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
864     o[INDEX3(i,1,3,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
865     } // end of component loop i
866     } // end of k0 loop
867     } // end of k1 loop
868     } // end of parallel section
869 caltinay 3711 } else if (out.getFunctionSpace().getTypeCode() == ReducedElements) {
870 caltinay 3760 out.requireWrite();
871 caltinay 3913 #pragma omp parallel
872     {
873     vector<double> f_00(numComp);
874     vector<double> f_01(numComp);
875     vector<double> f_10(numComp);
876     vector<double> f_11(numComp);
877     #pragma omp for
878 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
879     for (index_t k0=0; k0 < m_NE[0]; ++k0) {
880     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
881     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
882     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
883     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
884     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
885 caltinay 3913 for (index_t i=0; i < numComp; ++i) {
886     o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
887     o[INDEX3(i,1,0,numComp,2)] = cy2*(f_00[i] + f_10[i]) + cy5*(f_01[i] + f_11[i]);
888     } // end of component loop i
889     } // end of k0 loop
890     } // end of k1 loop
891     } // end of parallel section
892 caltinay 3707 } else if (out.getFunctionSpace().getTypeCode() == FaceElements) {
893 caltinay 3760 out.requireWrite();
894 caltinay 3722 #pragma omp parallel
895     {
896 caltinay 3913 vector<double> f_00(numComp);
897     vector<double> f_01(numComp);
898     vector<double> f_10(numComp);
899     vector<double> f_11(numComp);
900 caltinay 3722 if (m_faceOffset[0] > -1) {
901     #pragma omp for nowait
902 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
903     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
904     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
905     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(1,k1, m_NN[0])), numComp*sizeof(double));
906     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(1,k1+1, m_NN[0])), numComp*sizeof(double));
907 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
908     for (index_t i=0; i < numComp; ++i) {
909 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;
910     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
911     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
912     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
913 caltinay 3800 } // end of component loop i
914     } // end of k1 loop
915     } // end of face 0
916 caltinay 3722 if (m_faceOffset[1] > -1) {
917     #pragma omp for nowait
918 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
919     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1, m_NN[0])), numComp*sizeof(double));
920     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1+1, m_NN[0])), numComp*sizeof(double));
921     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
922     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
923 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
924     for (index_t i=0; i < numComp; ++i) {
925 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;
926     o[INDEX3(i,1,0,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
927     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
928     o[INDEX3(i,1,1,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
929 caltinay 3800 } // end of component loop i
930     } // end of k1 loop
931     } // end of face 1
932 caltinay 3722 if (m_faceOffset[2] > -1) {
933     #pragma omp for nowait
934 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
935     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
936     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,1, m_NN[0])), numComp*sizeof(double));
937     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
938     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,1, m_NN[0])), numComp*sizeof(double));
939 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
940     for (index_t i=0; i < numComp; ++i) {
941 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
942     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
943     o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
944     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
945 caltinay 3800 } // end of component loop i
946     } // end of k0 loop
947     } // end of face 2
948 caltinay 3722 if (m_faceOffset[3] > -1) {
949     #pragma omp for nowait
950 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
951     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
952     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
953     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
954     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
955 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
956     for (index_t i=0; i < numComp; ++i) {
957 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
958     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
959     o[INDEX3(i,0,1,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
960     o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
961 caltinay 3800 } // end of component loop i
962     } // end of k0 loop
963     } // end of face 3
964 caltinay 3722 } // end of parallel section
965 caltinay 3800
966 caltinay 3711 } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
967 caltinay 3760 out.requireWrite();
968 caltinay 3722 #pragma omp parallel
969     {
970 caltinay 3913 vector<double> f_00(numComp);
971     vector<double> f_01(numComp);
972     vector<double> f_10(numComp);
973     vector<double> f_11(numComp);
974 caltinay 3722 if (m_faceOffset[0] > -1) {
975     #pragma omp for nowait
976 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
977     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
978     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
979     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(1,k1, m_NN[0])), numComp*sizeof(double));
980     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(1,k1+1, m_NN[0])), numComp*sizeof(double));
981 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
982     for (index_t i=0; i < numComp; ++i) {
983 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
984     o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
985 caltinay 3800 } // end of component loop i
986     } // end of k1 loop
987     } // end of face 0
988 caltinay 3722 if (m_faceOffset[1] > -1) {
989     #pragma omp for nowait
990 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
991     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1, m_NN[0])), numComp*sizeof(double));
992     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1+1, m_NN[0])), numComp*sizeof(double));
993     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
994     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
995 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
996     for (index_t i=0; i < numComp; ++i) {
997 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
998     o[INDEX3(i,1,0,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
999 caltinay 3800 } // end of component loop i
1000     } // end of k1 loop
1001     } // end of face 1
1002 caltinay 3722 if (m_faceOffset[2] > -1) {
1003     #pragma omp for nowait
1004 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1005     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
1006     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,1, m_NN[0])), numComp*sizeof(double));
1007     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
1008     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,1, m_NN[0])), numComp*sizeof(double));
1009 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1010     for (index_t i=0; i < numComp; ++i) {
1011 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
1012     o[INDEX3(i,1,0,numComp,2)] = cy2*(f_00[i] + f_10[i]) + cy5*(f_01[i] + f_11[i]);
1013 caltinay 3800 } // end of component loop i
1014     } // end of k0 loop
1015     } // end of face 2
1016 caltinay 3722 if (m_faceOffset[3] > -1) {
1017     #pragma omp for nowait
1018 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1019     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
1020     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1021     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
1022     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1023 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1024     for (index_t i=0; i < numComp; ++i) {
1025 caltinay 3724 o[INDEX3(i,0,0,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
1026     o[INDEX3(i,1,0,numComp,2)] = cy5*(f_01[i] + f_11[i]) + cy2*(f_00[i] + f_10[i]);
1027 caltinay 3800 } // end of component loop i
1028     } // end of k0 loop
1029     } // end of face 3
1030 caltinay 3722 } // end of parallel section
1031 caltinay 3702 }
1032 caltinay 3701 }
1033 caltinay 3697
1034 caltinay 3764 //protected
1035     void Rectangle::assembleIntegrate(vector<double>& integrals, escript::Data& arg) const
1036 caltinay 3713 {
1037 caltinay 3764 const dim_t numComp = arg.getDataPointSize();
1038 caltinay 4334 const index_t left = (m_offset[0]==0 ? 0 : 1);
1039     const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1040 caltinay 3800 const int fs=arg.getFunctionSpace().getTypeCode();
1041     if (fs == Elements && arg.actsExpanded()) {
1042 caltinay 3713 #pragma omp parallel
1043     {
1044     vector<double> int_local(numComp, 0);
1045 caltinay 4334 const double w = m_dx[0]*m_dx[1]/4.;
1046 caltinay 3722 #pragma omp for nowait
1047 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1048     for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1049     const double* f = arg.getSampleDataRO(INDEX2(k0, k1, m_NE[0]));
1050 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1051 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1052     const double f1 = f[INDEX2(i,1,numComp)];
1053     const double f2 = f[INDEX2(i,2,numComp)];
1054     const double f3 = f[INDEX2(i,3,numComp)];
1055 caltinay 3764 int_local[i]+=(f0+f1+f2+f3)*w;
1056 caltinay 3800 } // end of component loop i
1057     } // end of k0 loop
1058     } // end of k1 loop
1059 caltinay 3713 #pragma omp critical
1060     for (index_t i=0; i<numComp; i++)
1061     integrals[i]+=int_local[i];
1062 caltinay 3722 } // end of parallel section
1063 caltinay 3800
1064     } else if (fs==ReducedElements || (fs==Elements && !arg.actsExpanded())) {
1065 caltinay 4334 const double w = m_dx[0]*m_dx[1];
1066 caltinay 3713 #pragma omp parallel
1067     {
1068     vector<double> int_local(numComp, 0);
1069 caltinay 3722 #pragma omp for nowait
1070 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1071     for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1072     const double* f = arg.getSampleDataRO(INDEX2(k0, k1, m_NE[0]));
1073 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1074 caltinay 3764 int_local[i]+=f[i]*w;
1075 caltinay 3800 }
1076     }
1077     }
1078 caltinay 3713 #pragma omp critical
1079     for (index_t i=0; i<numComp; i++)
1080     integrals[i]+=int_local[i];
1081 caltinay 3722 } // end of parallel section
1082 caltinay 3800
1083     } else if (fs == FaceElements && arg.actsExpanded()) {
1084 caltinay 3713 #pragma omp parallel
1085     {
1086     vector<double> int_local(numComp, 0);
1087 caltinay 4334 const double w0 = m_dx[0]/2.;
1088     const double w1 = m_dx[1]/2.;
1089 caltinay 3713 if (m_faceOffset[0] > -1) {
1090 caltinay 3722 #pragma omp for nowait
1091 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1092 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[0]+k1);
1093 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1094 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1095     const double f1 = f[INDEX2(i,1,numComp)];
1096 caltinay 3764 int_local[i]+=(f0+f1)*w1;
1097 caltinay 3800 } // end of component loop i
1098     } // end of k1 loop
1099 caltinay 3713 }
1100    
1101     if (m_faceOffset[1] > -1) {
1102 caltinay 3722 #pragma omp for nowait
1103 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1104 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[1]+k1);
1105 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1106 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1107     const double f1 = f[INDEX2(i,1,numComp)];
1108 caltinay 3764 int_local[i]+=(f0+f1)*w1;
1109 caltinay 3800 } // end of component loop i
1110     } // end of k1 loop
1111 caltinay 3713 }
1112    
1113     if (m_faceOffset[2] > -1) {
1114 caltinay 3722 #pragma omp for nowait
1115 caltinay 4334 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1116 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[2]+k0);
1117 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1118 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1119     const double f1 = f[INDEX2(i,1,numComp)];
1120 caltinay 3764 int_local[i]+=(f0+f1)*w0;
1121 caltinay 3800 } // end of component loop i
1122     } // end of k0 loop
1123 caltinay 3713 }
1124    
1125     if (m_faceOffset[3] > -1) {
1126 caltinay 3722 #pragma omp for nowait
1127 caltinay 4334 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1128 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[3]+k0);
1129 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1130 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1131     const double f1 = f[INDEX2(i,1,numComp)];
1132 caltinay 3764 int_local[i]+=(f0+f1)*w0;
1133 caltinay 3800 } // end of component loop i
1134     } // end of k0 loop
1135 caltinay 3713 }
1136     #pragma omp critical
1137     for (index_t i=0; i<numComp; i++)
1138     integrals[i]+=int_local[i];
1139 caltinay 3722 } // end of parallel section
1140 caltinay 3800
1141     } else if (fs==ReducedFaceElements || (fs==FaceElements && !arg.actsExpanded())) {
1142 caltinay 3713 #pragma omp parallel
1143     {
1144     vector<double> int_local(numComp, 0);
1145     if (m_faceOffset[0] > -1) {
1146 caltinay 3722 #pragma omp for nowait
1147 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1148 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[0]+k1);
1149 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1150 caltinay 4334 int_local[i]+=f[i]*m_dx[1];
1151 caltinay 3800 }
1152     }
1153 caltinay 3713 }
1154    
1155     if (m_faceOffset[1] > -1) {
1156 caltinay 3722 #pragma omp for nowait
1157 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1158 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[1]+k1);
1159 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1160 caltinay 4334 int_local[i]+=f[i]*m_dx[1];
1161 caltinay 3800 }
1162     }
1163 caltinay 3713 }
1164    
1165     if (m_faceOffset[2] > -1) {
1166 caltinay 3722 #pragma omp for nowait
1167 caltinay 4334 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1168 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[2]+k0);
1169 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1170 caltinay 4334 int_local[i]+=f[i]*m_dx[0];
1171 caltinay 3800 }
1172     }
1173 caltinay 3713 }
1174    
1175     if (m_faceOffset[3] > -1) {
1176 caltinay 3722 #pragma omp for nowait
1177 caltinay 4334 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1178 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[3]+k0);
1179 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1180 caltinay 4334 int_local[i]+=f[i]*m_dx[0];
1181 caltinay 3800 }
1182     }
1183 caltinay 3713 }
1184    
1185     #pragma omp critical
1186     for (index_t i=0; i<numComp; i++)
1187     integrals[i]+=int_local[i];
1188 caltinay 3722 } // end of parallel section
1189 caltinay 3800 } // function space selector
1190 caltinay 3713 }
1191    
1192 caltinay 3691 //protected
1193 caltinay 3756 dim_t Rectangle::insertNeighbourNodes(IndexVector& index, index_t node) const
1194     {
1195 caltinay 4334 const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1196     const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1197 caltinay 3756 const int x=node%nDOF0;
1198     const int y=node/nDOF0;
1199     dim_t num=0;
1200     // loop through potential neighbours and add to index if positions are
1201     // within bounds
1202     for (int i1=-1; i1<2; i1++) {
1203     for (int i0=-1; i0<2; i0++) {
1204     // skip node itself
1205     if (i0==0 && i1==0)
1206     continue;
1207     // location of neighbour node
1208     const int nx=x+i0;
1209     const int ny=y+i1;
1210     if (nx>=0 && ny>=0 && nx<nDOF0 && ny<nDOF1) {
1211     index.push_back(ny*nDOF0+nx);
1212     num++;
1213     }
1214     }
1215     }
1216    
1217     return num;
1218     }
1219    
1220     //protected
1221     void Rectangle::nodesToDOF(escript::Data& out, escript::Data& in) const
1222     {
1223     const dim_t numComp = in.getDataPointSize();
1224     out.requireWrite();
1225    
1226 caltinay 4334 const index_t left = (m_offset[0]==0 ? 0 : 1);
1227     const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1228     const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1229     const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1230 caltinay 3756 #pragma omp parallel for
1231     for (index_t i=0; i<nDOF1; i++) {
1232     for (index_t j=0; j<nDOF0; j++) {
1233 caltinay 4334 const index_t n=j+left+(i+bottom)*m_NN[0];
1234 caltinay 3756 const double* src=in.getSampleDataRO(n);
1235     copy(src, src+numComp, out.getSampleDataRW(j+i*nDOF0));
1236     }
1237     }
1238     }
1239    
1240     //protected
1241     void Rectangle::dofToNodes(escript::Data& out, escript::Data& in) const
1242     {
1243     const dim_t numComp = in.getDataPointSize();
1244     Paso_Coupler* coupler = Paso_Coupler_alloc(m_connector, numComp);
1245     in.requireWrite();
1246     Paso_Coupler_startCollect(coupler, in.getSampleDataRW(0));
1247    
1248     const dim_t numDOF = getNumDOF();
1249     out.requireWrite();
1250     const double* buffer = Paso_Coupler_finishCollect(coupler);
1251    
1252     #pragma omp parallel for
1253     for (index_t i=0; i<getNumNodes(); i++) {
1254     const double* src=(m_dofMap[i]<numDOF ?
1255     in.getSampleDataRO(m_dofMap[i])
1256     : &buffer[(m_dofMap[i]-numDOF)*numComp]);
1257     copy(src, src+numComp, out.getSampleDataRW(i));
1258     }
1259 caltinay 4002 Paso_Coupler_free(coupler);
1260 caltinay 3756 }
1261    
1262 caltinay 3691 //private
1263     void Rectangle::populateSampleIds()
1264     {
1265 caltinay 4334 // degrees of freedom are numbered from left to right, bottom to top in
1266     // each rank, continuing on the next rank (ranks also go left-right,
1267     // bottom-top).
1268     // This means rank 0 has id 0...n0-1, rank 1 has id n0...n1-1 etc. which
1269     // helps when writing out data rank after rank.
1270 caltinay 3697
1271     // build node distribution vector first.
1272 caltinay 4334 // rank i owns m_nodeDistribution[i+1]-nodeDistribution[i] nodes which is
1273     // constant for all ranks in this implementation
1274 caltinay 3697 m_nodeDistribution.assign(m_mpiInfo->size+1, 0);
1275 caltinay 3752 const dim_t numDOF=getNumDOF();
1276     for (dim_t k=1; k<m_mpiInfo->size; k++) {
1277     m_nodeDistribution[k]=k*numDOF;
1278 caltinay 3697 }
1279     m_nodeDistribution[m_mpiInfo->size]=getNumDataPointsGlobal();
1280 caltinay 3691 m_nodeId.resize(getNumNodes());
1281 caltinay 3753 m_dofId.resize(numDOF);
1282     m_elementId.resize(getNumElements());
1283 caltinay 4334
1284     // populate face element counts
1285     //left
1286     if (m_offset[0]==0)
1287     m_faceCount[0]=m_NE[1];
1288     else
1289     m_faceCount[0]=0;
1290     //right
1291     if (m_mpiInfo->rank%m_NX[0]==m_NX[0]-1)
1292     m_faceCount[1]=m_NE[1];
1293     else
1294     m_faceCount[1]=0;
1295     //bottom
1296     if (m_offset[1]==0)
1297     m_faceCount[2]=m_NE[0];
1298     else
1299     m_faceCount[2]=0;
1300     //top
1301     if (m_mpiInfo->rank/m_NX[0]==m_NX[1]-1)
1302     m_faceCount[3]=m_NE[0];
1303     else
1304     m_faceCount[3]=0;
1305    
1306 caltinay 3753 m_faceId.resize(getNumFaceElements());
1307 caltinay 3697
1308 caltinay 4334 const index_t left = (m_offset[0]==0 ? 0 : 1);
1309     const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1310     const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1311     const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1312    
1313     #define globalNodeId(x,y) \
1314     ((m_offset[0]+x)/nDOF0)*nDOF0*nDOF1+(m_offset[0]+x)%nDOF0 \
1315     + ((m_offset[1]+y)/nDOF1)*nDOF0*nDOF1*m_NX[0]+((m_offset[1]+y)%nDOF1)*nDOF0
1316    
1317     // set corner id's outside the parallel region
1318     m_nodeId[0] = globalNodeId(0, 0);
1319     m_nodeId[m_NN[0]-1] = globalNodeId(m_NN[0]-1, 0);
1320     m_nodeId[m_NN[0]*(m_NN[1]-1)] = globalNodeId(0, m_NN[1]-1);
1321     m_nodeId[m_NN[0]*m_NN[1]-1] = globalNodeId(m_NN[0]-1,m_NN[1]-1);
1322     #undef globalNodeId
1323    
1324 caltinay 3753 #pragma omp parallel
1325     {
1326 caltinay 4334 // populate degrees of freedom and own nodes (identical id)
1327 caltinay 3753 #pragma omp for nowait
1328 caltinay 4334 for (dim_t i=0; i<nDOF1; i++) {
1329     for (dim_t j=0; j<nDOF0; j++) {
1330     const index_t nodeIdx=j+left+(i+bottom)*m_NN[0];
1331     const index_t dofIdx=j+i*nDOF0;
1332     m_dofId[dofIdx] = m_nodeId[nodeIdx]
1333     = m_nodeDistribution[m_mpiInfo->rank]+dofIdx;
1334 caltinay 3753 }
1335 caltinay 3697 }
1336    
1337 caltinay 4334 // populate the rest of the nodes (shared with other ranks)
1338     if (m_faceCount[0]==0) { // left column
1339 caltinay 3753 #pragma omp for nowait
1340 caltinay 4334 for (dim_t i=0; i<nDOF1; i++) {
1341     const index_t nodeIdx=(i+bottom)*m_NN[0];
1342     const index_t dofId=(i+1)*nDOF0-1;
1343     m_nodeId[nodeIdx]
1344     = m_nodeDistribution[m_mpiInfo->rank-1]+dofId;
1345     }
1346     }
1347     if (m_faceCount[1]==0) { // right column
1348     #pragma omp for nowait
1349     for (dim_t i=0; i<nDOF1; i++) {
1350     const index_t nodeIdx=(i+bottom+1)*m_NN[0]-1;
1351     const index_t dofId=i*nDOF0;
1352     m_nodeId[nodeIdx]
1353     = m_nodeDistribution[m_mpiInfo->rank+1]+dofId;
1354     }
1355     }
1356     if (m_faceCount[2]==0) { // bottom row
1357     #pragma omp for nowait
1358     for (dim_t i=0; i<nDOF0; i++) {
1359     const index_t nodeIdx=i+left;
1360     const index_t dofId=nDOF0*(nDOF1-1)+i;
1361     m_nodeId[nodeIdx]
1362     = m_nodeDistribution[m_mpiInfo->rank-m_NX[0]]+dofId;
1363     }
1364     }
1365     if (m_faceCount[3]==0) { // top row
1366     #pragma omp for nowait
1367     for (dim_t i=0; i<nDOF0; i++) {
1368     const index_t nodeIdx=m_NN[0]*(m_NN[1]-1)+i+left;
1369     const index_t dofId=i;
1370     m_nodeId[nodeIdx]
1371     = m_nodeDistribution[m_mpiInfo->rank+m_NX[0]]+dofId;
1372     }
1373     }
1374 caltinay 3752
1375 caltinay 4334 // populate element id's
1376 caltinay 3753 #pragma omp for nowait
1377 caltinay 4334 for (dim_t i1=0; i1<m_NE[1]; i1++) {
1378     for (dim_t i0=0; i0<m_NE[0]; i0++) {
1379     m_elementId[i0+i1*m_NE[0]]=(m_offset[1]+i1)*m_gNE[0]+m_offset[0]+i0;
1380 caltinay 3755 }
1381     }
1382 caltinay 3753
1383     // face elements
1384     #pragma omp for
1385     for (dim_t k=0; k<getNumFaceElements(); k++)
1386     m_faceId[k]=k;
1387     } // end parallel section
1388    
1389 caltinay 3735 m_nodeTags.assign(getNumNodes(), 0);
1390     updateTagsInUse(Nodes);
1391 caltinay 3697
1392 caltinay 3735 m_elementTags.assign(getNumElements(), 0);
1393     updateTagsInUse(Elements);
1394 caltinay 3697
1395 caltinay 3722 // generate face offset vector and set face tags
1396     const index_t LEFT=1, RIGHT=2, BOTTOM=10, TOP=20;
1397     const index_t faceTag[] = { LEFT, RIGHT, BOTTOM, TOP };
1398 caltinay 4334 m_faceOffset.assign(4, -1);
1399 caltinay 3722 m_faceTags.clear();
1400 caltinay 3704 index_t offset=0;
1401 caltinay 4334 for (size_t i=0; i<4; i++) {
1402     if (m_faceCount[i]>0) {
1403 caltinay 3704 m_faceOffset[i]=offset;
1404 caltinay 4334 offset+=m_faceCount[i];
1405     m_faceTags.insert(m_faceTags.end(), m_faceCount[i], faceTag[i]);
1406 caltinay 3704 }
1407     }
1408 caltinay 3722 setTagMap("left", LEFT);
1409     setTagMap("right", RIGHT);
1410     setTagMap("bottom", BOTTOM);
1411     setTagMap("top", TOP);
1412     updateTagsInUse(FaceElements);
1413 caltinay 3691 }
1414    
1415 caltinay 3699 //private
1416 caltinay 3756 void Rectangle::createPattern()
1417 caltinay 3699 {
1418 caltinay 4334 const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1419     const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1420     const index_t left = (m_offset[0]==0 ? 0 : 1);
1421     const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1422 caltinay 3756
1423     // populate node->DOF mapping with own degrees of freedom.
1424     // The rest is assigned in the loop further down
1425     m_dofMap.assign(getNumNodes(), 0);
1426     #pragma omp parallel for
1427 caltinay 3766 for (index_t i=bottom; i<bottom+nDOF1; i++) {
1428     for (index_t j=left; j<left+nDOF0; j++) {
1429 caltinay 4334 m_dofMap[i*m_NN[0]+j]=(i-bottom)*nDOF0+j-left;
1430 caltinay 3756 }
1431     }
1432    
1433     // build list of shared components and neighbours by looping through
1434     // all potential neighbouring ranks and checking if positions are
1435 caltinay 3754 // within bounds
1436 caltinay 3756 const dim_t numDOF=nDOF0*nDOF1;
1437     vector<IndexVector> colIndices(numDOF); // for the couple blocks
1438     RankVector neighbour;
1439     IndexVector offsetInShared(1,0);
1440     IndexVector sendShared, recvShared;
1441     int numShared=0;
1442 caltinay 4334 const int x=m_mpiInfo->rank%m_NX[0];
1443     const int y=m_mpiInfo->rank/m_NX[0];
1444 caltinay 3754 for (int i1=-1; i1<2; i1++) {
1445     for (int i0=-1; i0<2; i0++) {
1446 caltinay 3756 // skip this rank
1447 caltinay 3754 if (i0==0 && i1==0)
1448     continue;
1449 caltinay 3756 // location of neighbour rank
1450 caltinay 3754 const int nx=x+i0;
1451     const int ny=y+i1;
1452 caltinay 4334 if (nx>=0 && ny>=0 && nx<m_NX[0] && ny<m_NX[1]) {
1453     neighbour.push_back(ny*m_NX[0]+nx);
1454 caltinay 3756 if (i0==0) {
1455     // sharing top or bottom edge
1456     const int firstDOF=(i1==-1 ? 0 : numDOF-nDOF0);
1457 caltinay 4334 const int firstNode=(i1==-1 ? left : m_NN[0]*(m_NN[1]-1)+left);
1458 caltinay 3756 offsetInShared.push_back(offsetInShared.back()+nDOF0);
1459     for (dim_t i=0; i<nDOF0; i++, numShared++) {
1460     sendShared.push_back(firstDOF+i);
1461     recvShared.push_back(numDOF+numShared);
1462     if (i>0)
1463     colIndices[firstDOF+i-1].push_back(numShared);
1464     colIndices[firstDOF+i].push_back(numShared);
1465     if (i<nDOF0-1)
1466     colIndices[firstDOF+i+1].push_back(numShared);
1467     m_dofMap[firstNode+i]=numDOF+numShared;
1468     }
1469     } else if (i1==0) {
1470     // sharing left or right edge
1471     const int firstDOF=(i0==-1 ? 0 : nDOF0-1);
1472 caltinay 4334 const int firstNode=(i0==-1 ? bottom*m_NN[0] : (bottom+1)*m_NN[0]-1);
1473 caltinay 3756 offsetInShared.push_back(offsetInShared.back()+nDOF1);
1474     for (dim_t i=0; i<nDOF1; i++, numShared++) {
1475     sendShared.push_back(firstDOF+i*nDOF0);
1476     recvShared.push_back(numDOF+numShared);
1477     if (i>0)
1478     colIndices[firstDOF+(i-1)*nDOF0].push_back(numShared);
1479     colIndices[firstDOF+i*nDOF0].push_back(numShared);
1480     if (i<nDOF1-1)
1481     colIndices[firstDOF+(i+1)*nDOF0].push_back(numShared);
1482 caltinay 4334 m_dofMap[firstNode+i*m_NN[0]]=numDOF+numShared;
1483 caltinay 3756 }
1484     } else {
1485     // sharing a node
1486     const int dof=(i0+1)/2*(nDOF0-1)+(i1+1)/2*(numDOF-nDOF0);
1487 caltinay 4334 const int node=(i0+1)/2*(m_NN[0]-1)+(i1+1)/2*m_NN[0]*(m_NN[1]-1);
1488 caltinay 3756 offsetInShared.push_back(offsetInShared.back()+1);
1489     sendShared.push_back(dof);
1490     recvShared.push_back(numDOF+numShared);
1491     colIndices[dof].push_back(numShared);
1492     m_dofMap[node]=numDOF+numShared;
1493     ++numShared;
1494     }
1495 caltinay 3754 }
1496 caltinay 3699 }
1497     }
1498 caltinay 3754
1499 caltinay 3756 // create connector
1500     Paso_SharedComponents *snd_shcomp = Paso_SharedComponents_alloc(
1501     numDOF, neighbour.size(), &neighbour[0], &sendShared[0],
1502     &offsetInShared[0], 1, 0, m_mpiInfo);
1503     Paso_SharedComponents *rcv_shcomp = Paso_SharedComponents_alloc(
1504     numDOF, neighbour.size(), &neighbour[0], &recvShared[0],
1505     &offsetInShared[0], 1, 0, m_mpiInfo);
1506     m_connector = Paso_Connector_alloc(snd_shcomp, rcv_shcomp);
1507     Paso_SharedComponents_free(snd_shcomp);
1508     Paso_SharedComponents_free(rcv_shcomp);
1509 caltinay 3754
1510 caltinay 3756 // create main and couple blocks
1511     Paso_Pattern *mainPattern = createMainPattern();
1512     Paso_Pattern *colPattern, *rowPattern;
1513     createCouplePatterns(colIndices, numShared, &colPattern, &rowPattern);
1514 caltinay 3754
1515 caltinay 3756 // allocate paso distribution
1516     Paso_Distribution* distribution = Paso_Distribution_alloc(m_mpiInfo,
1517     const_cast<index_t*>(&m_nodeDistribution[0]), 1, 0);
1518 caltinay 3755
1519 caltinay 3756 // finally create the system matrix
1520     m_pattern = Paso_SystemMatrixPattern_alloc(MATRIX_FORMAT_DEFAULT,
1521     distribution, distribution, mainPattern, colPattern, rowPattern,
1522     m_connector, m_connector);
1523 caltinay 3755
1524 caltinay 3756 Paso_Distribution_free(distribution);
1525 caltinay 3755
1526 caltinay 3756 // useful debug output
1527     /*
1528     cout << "--- rcv_shcomp ---" << endl;
1529     cout << "numDOF=" << numDOF << ", numNeighbors=" << neighbour.size() << endl;
1530     for (size_t i=0; i<neighbour.size(); i++) {
1531     cout << "neighbor[" << i << "]=" << neighbour[i]
1532     << " offsetInShared[" << i+1 << "]=" << offsetInShared[i+1] << endl;
1533 caltinay 3699 }
1534 caltinay 3756 for (size_t i=0; i<recvShared.size(); i++) {
1535     cout << "shared[" << i << "]=" << recvShared[i] << endl;
1536     }
1537     cout << "--- snd_shcomp ---" << endl;
1538     for (size_t i=0; i<sendShared.size(); i++) {
1539     cout << "shared[" << i << "]=" << sendShared[i] << endl;
1540     }
1541     cout << "--- dofMap ---" << endl;
1542     for (size_t i=0; i<m_dofMap.size(); i++) {
1543     cout << "m_dofMap[" << i << "]=" << m_dofMap[i] << endl;
1544     }
1545     cout << "--- colIndices ---" << endl;
1546     for (size_t i=0; i<colIndices.size(); i++) {
1547     cout << "colIndices[" << i << "].size()=" << colIndices[i].size() << endl;
1548     }
1549     */
1550 caltinay 3754
1551 caltinay 3756 /*
1552     cout << "--- main_pattern ---" << endl;
1553     cout << "M=" << mainPattern->numOutput << ", N=" << mainPattern->numInput << endl;
1554     for (size_t i=0; i<mainPattern->numOutput+1; i++) {
1555     cout << "ptr[" << i << "]=" << mainPattern->ptr[i] << endl;
1556     }
1557     for (size_t i=0; i<mainPattern->ptr[mainPattern->numOutput]; i++) {
1558     cout << "index[" << i << "]=" << mainPattern->index[i] << endl;
1559     }
1560     */
1561 caltinay 3754
1562 caltinay 3756 /*
1563     cout << "--- colCouple_pattern ---" << endl;
1564     cout << "M=" << colPattern->numOutput << ", N=" << colPattern->numInput << endl;
1565     for (size_t i=0; i<colPattern->numOutput+1; i++) {
1566     cout << "ptr[" << i << "]=" << colPattern->ptr[i] << endl;
1567     }
1568     for (size_t i=0; i<colPattern->ptr[colPattern->numOutput]; i++) {
1569     cout << "index[" << i << "]=" << colPattern->index[i] << endl;
1570     }
1571     */
1572 caltinay 3754
1573 caltinay 3756 /*
1574     cout << "--- rowCouple_pattern ---" << endl;
1575     cout << "M=" << rowPattern->numOutput << ", N=" << rowPattern->numInput << endl;
1576     for (size_t i=0; i<rowPattern->numOutput+1; i++) {
1577     cout << "ptr[" << i << "]=" << rowPattern->ptr[i] << endl;
1578 caltinay 3699 }
1579 caltinay 3756 for (size_t i=0; i<rowPattern->ptr[rowPattern->numOutput]; i++) {
1580     cout << "index[" << i << "]=" << rowPattern->index[i] << endl;
1581     }
1582     */
1583    
1584     Paso_Pattern_free(mainPattern);
1585     Paso_Pattern_free(colPattern);
1586     Paso_Pattern_free(rowPattern);
1587 caltinay 3699 }
1588    
1589 caltinay 3776 //private
1590     void Rectangle::addToMatrixAndRHS(Paso_SystemMatrix* S, escript::Data& F,
1591     const vector<double>& EM_S, const vector<double>& EM_F, bool addS,
1592     bool addF, index_t firstNode, dim_t nEq, dim_t nComp) const
1593     {
1594     IndexVector rowIndex;
1595     rowIndex.push_back(m_dofMap[firstNode]);
1596     rowIndex.push_back(m_dofMap[firstNode+1]);
1597 caltinay 4334 rowIndex.push_back(m_dofMap[firstNode+m_NN[0]]);
1598     rowIndex.push_back(m_dofMap[firstNode+m_NN[0]+1]);
1599 caltinay 3776 if (addF) {
1600     double *F_p=F.getSampleDataRW(0);
1601     for (index_t i=0; i<rowIndex.size(); i++) {
1602     if (rowIndex[i]<getNumDOF()) {
1603     for (index_t eq=0; eq<nEq; eq++) {
1604     F_p[INDEX2(eq, rowIndex[i], nEq)]+=EM_F[INDEX2(eq,i,nEq)];
1605     }
1606     }
1607     }
1608     }
1609     if (addS) {
1610     addToSystemMatrix(S, rowIndex, nEq, rowIndex, nComp, EM_S);
1611     }
1612     }
1613    
1614 caltinay 3702 //protected
1615 caltinay 3711 void Rectangle::interpolateNodesOnElements(escript::Data& out,
1616     escript::Data& in, bool reduced) const
1617 caltinay 3702 {
1618     const dim_t numComp = in.getDataPointSize();
1619 caltinay 3711 if (reduced) {
1620 caltinay 3760 out.requireWrite();
1621 caltinay 3913 const double c0 = 0.25;
1622     #pragma omp parallel
1623     {
1624     vector<double> f_00(numComp);
1625     vector<double> f_01(numComp);
1626     vector<double> f_10(numComp);
1627     vector<double> f_11(numComp);
1628     #pragma omp for
1629 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1630     for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1631     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
1632     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
1633     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
1634     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
1635     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
1636 caltinay 3913 for (index_t i=0; i < numComp; ++i) {
1637     o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i] + f_10[i] + f_11[i]);
1638     } /* end of component loop i */
1639     } /* end of k0 loop */
1640     } /* end of k1 loop */
1641     } /* end of parallel section */
1642 caltinay 3711 } else {
1643 caltinay 3760 out.requireWrite();
1644 caltinay 3913 const double c0 = 0.16666666666666666667;
1645     const double c1 = 0.044658198738520451079;
1646     const double c2 = 0.62200846792814621559;
1647     #pragma omp parallel
1648     {
1649     vector<double> f_00(numComp);
1650     vector<double> f_01(numComp);
1651     vector<double> f_10(numComp);
1652     vector<double> f_11(numComp);
1653     #pragma omp for
1654 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1655     for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1656     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
1657     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
1658     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
1659     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
1660     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
1661 caltinay 3913 for (index_t i=0; i < numComp; ++i) {
1662     o[INDEX2(i,numComp,0)] = c0*(f_01[i] + f_10[i]) + c1*f_11[i] + c2*f_00[i];
1663     o[INDEX2(i,numComp,1)] = c0*(f_00[i] + f_11[i]) + c1*f_01[i] + c2*f_10[i];
1664     o[INDEX2(i,numComp,2)] = c0*(f_00[i] + f_11[i]) + c1*f_10[i] + c2*f_01[i];
1665     o[INDEX2(i,numComp,3)] = c0*(f_01[i] + f_10[i]) + c1*f_00[i] + c2*f_11[i];
1666     } /* end of component loop i */
1667     } /* end of k0 loop */
1668     } /* end of k1 loop */
1669     } /* end of parallel section */
1670 caltinay 3711 }
1671 caltinay 3702 }
1672    
1673     //protected
1674 caltinay 3711 void Rectangle::interpolateNodesOnFaces(escript::Data& out, escript::Data& in,
1675     bool reduced) const
1676 caltinay 3702 {
1677 caltinay 3704 const dim_t numComp = in.getDataPointSize();
1678 caltinay 3711 if (reduced) {
1679 caltinay 3760 out.requireWrite();
1680 caltinay 3913 const double c0 = 0.5;
1681 caltinay 3724 #pragma omp parallel
1682     {
1683 caltinay 3913 vector<double> f_00(numComp);
1684     vector<double> f_01(numComp);
1685     vector<double> f_10(numComp);
1686     vector<double> f_11(numComp);
1687 caltinay 3724 if (m_faceOffset[0] > -1) {
1688     #pragma omp for nowait
1689 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1690     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
1691     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
1692 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1693     for (index_t i=0; i < numComp; ++i) {
1694     o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i]);
1695     } /* end of component loop i */
1696     } /* end of k1 loop */
1697     } /* end of face 0 */
1698     if (m_faceOffset[1] > -1) {
1699     #pragma omp for nowait
1700 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1701     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
1702     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
1703 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1704     for (index_t i=0; i < numComp; ++i) {
1705     o[INDEX2(i,numComp,0)] = c0*(f_10[i] + f_11[i]);
1706     } /* end of component loop i */
1707     } /* end of k1 loop */
1708     } /* end of face 1 */
1709     if (m_faceOffset[2] > -1) {
1710     #pragma omp for nowait
1711 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1712     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
1713     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
1714 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1715     for (index_t i=0; i < numComp; ++i) {
1716     o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_10[i]);
1717     } /* end of component loop i */
1718     } /* end of k0 loop */
1719     } /* end of face 2 */
1720     if (m_faceOffset[3] > -1) {
1721     #pragma omp for nowait
1722 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1723     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1724     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1725 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1726     for (index_t i=0; i < numComp; ++i) {
1727     o[INDEX2(i,numComp,0)] = c0*(f_01[i] + f_11[i]);
1728     } /* end of component loop i */
1729     } /* end of k0 loop */
1730     } /* end of face 3 */
1731 caltinay 3913 } /* end of parallel section */
1732 caltinay 3711 } else {
1733 caltinay 3760 out.requireWrite();
1734 caltinay 3724 const double c0 = 0.21132486540518711775;
1735     const double c1 = 0.78867513459481288225;
1736     #pragma omp parallel
1737     {
1738 caltinay 3913 vector<double> f_00(numComp);
1739     vector<double> f_01(numComp);
1740     vector<double> f_10(numComp);
1741     vector<double> f_11(numComp);
1742 caltinay 3724 if (m_faceOffset[0] > -1) {
1743     #pragma omp for nowait
1744 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1745     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
1746     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
1747 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1748     for (index_t i=0; i < numComp; ++i) {
1749 caltinay 3913 o[INDEX2(i,numComp,0)] = c0*f_01[i] + c1*f_00[i];
1750     o[INDEX2(i,numComp,1)] = c0*f_00[i] + c1*f_01[i];
1751 caltinay 3724 } /* end of component loop i */
1752     } /* end of k1 loop */
1753     } /* end of face 0 */
1754     if (m_faceOffset[1] > -1) {
1755     #pragma omp for nowait
1756 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1757     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
1758     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
1759 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1760     for (index_t i=0; i < numComp; ++i) {
1761 caltinay 3913 o[INDEX2(i,numComp,0)] = c1*f_10[i] + c0*f_11[i];
1762     o[INDEX2(i,numComp,1)] = c1*f_11[i] + c0*f_10[i];
1763 caltinay 3724 } /* end of component loop i */
1764     } /* end of k1 loop */
1765     } /* end of face 1 */
1766     if (m_faceOffset[2] > -1) {
1767     #pragma omp for nowait
1768 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1769     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
1770     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
1771 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1772     for (index_t i=0; i < numComp; ++i) {
1773 caltinay 3913 o[INDEX2(i,numComp,0)] = c0*f_10[i] + c1*f_00[i];
1774     o[INDEX2(i,numComp,1)] = c0*f_00[i] + c1*f_10[i];
1775 caltinay 3724 } /* end of component loop i */
1776     } /* end of k0 loop */
1777     } /* end of face 2 */
1778     if (m_faceOffset[3] > -1) {
1779     #pragma omp for nowait
1780 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1781     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1782     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1783 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1784     for (index_t i=0; i < numComp; ++i) {
1785 caltinay 3913 o[INDEX2(i,numComp,0)] = c0*f_11[i] + c1*f_01[i];
1786     o[INDEX2(i,numComp,1)] = c0*f_01[i] + c1*f_11[i];
1787 caltinay 3724 } /* end of component loop i */
1788     } /* end of k0 loop */
1789     } /* end of face 3 */
1790 caltinay 3913 } /* end of parallel section */
1791 caltinay 3711 }
1792 caltinay 3702 }
1793    
1794 caltinay 3748 //protected
1795     void Rectangle::assemblePDESingle(Paso_SystemMatrix* mat,
1796     escript::Data& rhs, const escript::Data& A, const escript::Data& B,
1797     const escript::Data& C, const escript::Data& D,
1798 caltinay 3769 const escript::Data& X, const escript::Data& Y) const
1799 caltinay 3748 {
1800 caltinay 4334 const double w0 = -0.1555021169820365539*m_dx[1]/m_dx[0];
1801 caltinay 3748 const double w1 = 0.041666666666666666667;
1802 caltinay 3764 const double w2 = -0.15550211698203655390;
1803 caltinay 4334 const double w3 = 0.041666666666666666667*m_dx[0]/m_dx[1];
1804 caltinay 3764 const double w4 = 0.15550211698203655390;
1805     const double w5 = -0.041666666666666666667;
1806 caltinay 4334 const double w6 = -0.01116454968463011277*m_dx[1]/m_dx[0];
1807 caltinay 3764 const double w7 = 0.011164549684630112770;
1808     const double w8 = -0.011164549684630112770;
1809 caltinay 4334 const double w9 = -0.041666666666666666667*m_dx[1]/m_dx[0];
1810     const double w10 = -0.041666666666666666667*m_dx[0]/m_dx[1];
1811     const double w11 = 0.1555021169820365539*m_dx[1]/m_dx[0];
1812     const double w12 = 0.1555021169820365539*m_dx[0]/m_dx[1];
1813     const double w13 = 0.01116454968463011277*m_dx[0]/m_dx[1];
1814     const double w14 = 0.01116454968463011277*m_dx[1]/m_dx[0];
1815     const double w15 = 0.041666666666666666667*m_dx[1]/m_dx[0];
1816     const double w16 = -0.01116454968463011277*m_dx[0]/m_dx[1];
1817     const double w17 = -0.1555021169820365539*m_dx[0]/m_dx[1];
1818     const double w18 = -0.33333333333333333333*m_dx[1]/m_dx[0];
1819 caltinay 3769 const double w19 = 0.25;
1820     const double w20 = -0.25;
1821 caltinay 4334 const double w21 = 0.16666666666666666667*m_dx[0]/m_dx[1];
1822     const double w22 = -0.16666666666666666667*m_dx[1]/m_dx[0];
1823     const double w23 = -0.16666666666666666667*m_dx[0]/m_dx[1];
1824     const double w24 = 0.33333333333333333333*m_dx[1]/m_dx[0];
1825     const double w25 = 0.33333333333333333333*m_dx[0]/m_dx[1];
1826     const double w26 = 0.16666666666666666667