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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 4650 - (hide annotations)
Wed Feb 5 04:16:01 2014 UTC (5 years, 8 months ago) by jfenwick
File size: 88932 byte(s)
Fixed a spelling error and missing virtual (not sure that one matters).
Added _untested_ 3D gaussian smoothed random data


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 sshaw 4622 #include <ripley/DefaultAssembler2D.h>
20     #include <ripley/WaveAssembler2D.h>
21 caltinay 4477 #include <boost/scoped_array.hpp>
22 jfenwick 4521 #include "esysUtils/EsysRandom.h"
23 caltinay 4477
24 caltinay 4013 #ifdef USE_NETCDF
25     #include <netcdfcpp.h>
26     #endif
27    
28 caltinay 3691 #if USE_SILO
29     #include <silo.h>
30     #ifdef ESYS_MPI
31     #include <pmpio.h>
32     #endif
33     #endif
34    
35     #include <iomanip>
36    
37     using namespace std;
38 caltinay 4334 using esysUtils::FileWriter;
39 caltinay 3691
40     namespace ripley {
41    
42 caltinay 3781 Rectangle::Rectangle(int n0, int n1, double x0, double y0, double x1,
43 sshaw 4622 double y1, int d0, int d1,
44     const std::vector<double>& points,
45     const std::vector<int>& tags,
46 jfenwick 4638 const simap_t& tagnamestonums) :
47 caltinay 4334 RipleyDomain(2)
48 caltinay 3691 {
49 caltinay 3943 // ignore subdivision parameters for serial run
50     if (m_mpiInfo->size == 1) {
51     d0=1;
52     d1=1;
53     }
54    
55     bool warn=false;
56     // if number of subdivisions is non-positive, try to subdivide by the same
57     // ratio as the number of elements
58     if (d0<=0 && d1<=0) {
59     warn=true;
60 caltinay 4010 d0=max(1, (int)(sqrt(m_mpiInfo->size*(n0+1)/(float)(n1+1))));
61 caltinay 3943 d1=m_mpiInfo->size/d0;
62     if (d0*d1 != m_mpiInfo->size) {
63     // ratios not the same so subdivide side with more elements only
64     if (n0>n1) {
65     d0=0;
66     d1=1;
67     } else {
68     d0=1;
69     d1=0;
70     }
71     }
72     }
73     if (d0<=0) {
74     // d1 is preset, determine d0 - throw further down if result is no good
75     d0=m_mpiInfo->size/d1;
76     } else if (d1<=0) {
77     // d0 is preset, determine d1 - throw further down if result is no good
78     d1=m_mpiInfo->size/d0;
79     }
80    
81 caltinay 3691 // ensure number of subdivisions is valid and nodes can be distributed
82     // among number of ranks
83 caltinay 4334 if (d0*d1 != m_mpiInfo->size)
84 caltinay 3691 throw RipleyException("Invalid number of spatial subdivisions");
85    
86 caltinay 3943 if (warn) {
87     cout << "Warning: Automatic domain subdivision (d0=" << d0 << ", d1="
88     << d1 << "). This may not be optimal!" << endl;
89     }
90 caltinay 3691
91 caltinay 4334 double l0 = x1-x0;
92     double l1 = y1-y0;
93     m_dx[0] = l0/n0;
94     m_dx[1] = l1/n1;
95    
96     if ((n0+1)%d0 > 0) {
97 caltinay 3943 n0=(int)round((float)(n0+1)/d0+0.5)*d0-1;
98 caltinay 4334 l0=m_dx[0]*n0;
99 caltinay 3943 cout << "Warning: Adjusted number of elements and length. N0="
100 caltinay 4334 << n0 << ", l0=" << l0 << endl;
101 caltinay 3943 }
102 caltinay 4334 if ((n1+1)%d1 > 0) {
103 caltinay 3943 n1=(int)round((float)(n1+1)/d1+0.5)*d1-1;
104 caltinay 4334 l1=m_dx[1]*n1;
105 caltinay 3943 cout << "Warning: Adjusted number of elements and length. N1="
106 caltinay 4334 << n1 << ", l1=" << l1 << endl;
107 caltinay 3943 }
108    
109 caltinay 4334 if ((d0 > 1 && (n0+1)/d0<2) || (d1 > 1 && (n1+1)/d1<2))
110 caltinay 3752 throw RipleyException("Too few elements for the number of ranks");
111    
112 caltinay 4334 m_gNE[0] = n0;
113     m_gNE[1] = n1;
114     m_origin[0] = x0;
115     m_origin[1] = y0;
116     m_length[0] = l0;
117     m_length[1] = l1;
118     m_NX[0] = d0;
119     m_NX[1] = d1;
120    
121 caltinay 3764 // local number of elements (with and without overlap)
122 caltinay 4334 m_NE[0] = m_ownNE[0] = (d0>1 ? (n0+1)/d0 : n0);
123     if (m_mpiInfo->rank%d0>0 && m_mpiInfo->rank%d0<d0-1)
124     m_NE[0]++;
125     else if (d0>1 && m_mpiInfo->rank%d0==d0-1)
126     m_ownNE[0]--;
127 caltinay 3764
128 caltinay 4334 m_NE[1] = m_ownNE[1] = (d1>1 ? (n1+1)/d1 : n1);
129     if (m_mpiInfo->rank/d0>0 && m_mpiInfo->rank/d0<d1-1)
130     m_NE[1]++;
131     else if (d1>1 && m_mpiInfo->rank/d0==d1-1)
132     m_ownNE[1]--;
133 caltinay 3752
134     // local number of nodes
135 caltinay 4334 m_NN[0] = m_NE[0]+1;
136     m_NN[1] = m_NE[1]+1;
137 caltinay 3752
138 caltinay 3691 // bottom-left node is at (offset0,offset1) in global mesh
139 caltinay 4334 m_offset[0] = (n0+1)/d0*(m_mpiInfo->rank%d0);
140     if (m_offset[0] > 0)
141     m_offset[0]--;
142     m_offset[1] = (n1+1)/d1*(m_mpiInfo->rank/d0);
143     if (m_offset[1] > 0)
144     m_offset[1]--;
145 caltinay 3752
146 caltinay 3691 populateSampleIds();
147 caltinay 3756 createPattern();
148 sshaw 4622 assembler = new DefaultAssembler2D(this, m_dx, m_NX, m_NE, m_NN);
149     for (map<string, int>::const_iterator i = tagnamestonums.begin();
150     i != tagnamestonums.end(); i++) {
151     setTagMap(i->first, i->second);
152     }
153     addPoints(tags.size(), &points[0], &tags[0]);
154 caltinay 3691 }
155    
156     Rectangle::~Rectangle()
157     {
158 caltinay 3785 Paso_SystemMatrixPattern_free(m_pattern);
159     Paso_Connector_free(m_connector);
160 sshaw 4622 delete assembler;
161 caltinay 3691 }
162    
163     string Rectangle::getDescription() const
164     {
165     return "ripley::Rectangle";
166     }
167    
168     bool Rectangle::operator==(const AbstractDomain& other) const
169     {
170 caltinay 3744 const Rectangle* o=dynamic_cast<const Rectangle*>(&other);
171     if (o) {
172     return (RipleyDomain::operator==(other) &&
173 caltinay 4334 m_gNE[0]==o->m_gNE[0] && m_gNE[1]==o->m_gNE[1]
174     && m_origin[0]==o->m_origin[0] && m_origin[1]==o->m_origin[1]
175     && m_length[0]==o->m_length[0] && m_length[1]==o->m_length[1]
176     && m_NX[0]==o->m_NX[0] && m_NX[1]==o->m_NX[1]);
177 caltinay 3744 }
178 caltinay 3691
179     return false;
180     }
181    
182 caltinay 4013 void Rectangle::readNcGrid(escript::Data& out, string filename, string varname,
183 caltinay 4618 const ReaderParameters& params) const
184 caltinay 4013 {
185     #ifdef USE_NETCDF
186     // check destination function space
187     int myN0, myN1;
188     if (out.getFunctionSpace().getTypeCode() == Nodes) {
189 caltinay 4334 myN0 = m_NN[0];
190     myN1 = m_NN[1];
191 caltinay 4013 } else if (out.getFunctionSpace().getTypeCode() == Elements ||
192     out.getFunctionSpace().getTypeCode() == ReducedElements) {
193 caltinay 4334 myN0 = m_NE[0];
194     myN1 = m_NE[1];
195 caltinay 4013 } else
196     throw RipleyException("readNcGrid(): invalid function space for output data object");
197    
198 caltinay 4615 if (params.first.size() != 2)
199 caltinay 4013 throw RipleyException("readNcGrid(): argument 'first' must have 2 entries");
200    
201 caltinay 4615 if (params.numValues.size() != 2)
202 caltinay 4013 throw RipleyException("readNcGrid(): argument 'numValues' must have 2 entries");
203    
204 caltinay 4615 if (params.multiplier.size() != 2)
205 caltinay 4277 throw RipleyException("readNcGrid(): argument 'multiplier' must have 2 entries");
206 caltinay 4615 for (size_t i=0; i<params.multiplier.size(); i++)
207     if (params.multiplier[i]<1)
208 caltinay 4277 throw RipleyException("readNcGrid(): all multipliers must be positive");
209 caltinay 4618 if (params.reverse.size() != 2)
210     throw RipleyException("readNcGrid(): argument 'reverse' must have 2 entries");
211 caltinay 4277
212 caltinay 4013 // check file existence and size
213     NcFile f(filename.c_str(), NcFile::ReadOnly);
214     if (!f.is_valid())
215     throw RipleyException("readNcGrid(): cannot open file");
216    
217     NcVar* var = f.get_var(varname.c_str());
218     if (!var)
219     throw RipleyException("readNcGrid(): invalid variable");
220    
221     // TODO: rank>0 data support
222     const int numComp = out.getDataPointSize();
223     if (numComp > 1)
224     throw RipleyException("readNcGrid(): only scalar data supported");
225    
226     const int dims = var->num_dims();
227 caltinay 4477 boost::scoped_array<long> edges(var->edges());
228 caltinay 4013
229     // is this a slice of the data object (dims!=2)?
230     // note the expected ordering of edges (as in numpy: y,x)
231 caltinay 4615 if ( (dims==2 && (params.numValues[1] > edges[0] || params.numValues[0] > edges[1]))
232     || (dims==1 && params.numValues[1]>1) ) {
233 caltinay 4013 throw RipleyException("readNcGrid(): not enough data in file");
234     }
235    
236     // check if this rank contributes anything
237 caltinay 4615 if (params.first[0] >= m_offset[0]+myN0 ||
238     params.first[0]+params.numValues[0]*params.multiplier[0] <= m_offset[0] ||
239     params.first[1] >= m_offset[1]+myN1 ||
240     params.first[1]+params.numValues[1]*params.multiplier[1] <= m_offset[1])
241 caltinay 4013 return;
242    
243     // now determine how much this rank has to write
244    
245     // first coordinates in data object to write to
246 caltinay 4615 const int first0 = max(0, params.first[0]-m_offset[0]);
247     const int first1 = max(0, params.first[1]-m_offset[1]);
248 caltinay 4618 // indices to first value in file (not accounting for reverse yet)
249     int idx0 = max(0, m_offset[0]-params.first[0]);
250     int idx1 = max(0, m_offset[1]-params.first[1]);
251 caltinay 4277 // number of values to read
252 caltinay 4615 const int num0 = min(params.numValues[0]-idx0, myN0-first0);
253     const int num1 = min(params.numValues[1]-idx1, myN1-first1);
254 caltinay 4013
255 caltinay 4618 // make sure we read the right block if going backwards through file
256     if (params.reverse[0])
257     idx0 = edges[dims-1]-num0-idx0;
258     if (dims>1 && params.reverse[1])
259     idx1 = edges[dims-2]-num1-idx1;
260    
261 caltinay 4013 vector<double> values(num0*num1);
262     if (dims==2) {
263     var->set_cur(idx1, idx0);
264     var->get(&values[0], num1, num0);
265     } else {
266     var->set_cur(idx0);
267     var->get(&values[0], num0);
268     }
269    
270     const int dpp = out.getNumDataPointsPerSample();
271     out.requireWrite();
272    
273 caltinay 4618 // helpers for reversing
274     const int x0 = (params.reverse[0] ? num0-1 : 0);
275     const int x_mult = (params.reverse[0] ? -1 : 1);
276     const int y0 = (params.reverse[1] ? num1-1 : 0);
277     const int y_mult = (params.reverse[1] ? -1 : 1);
278    
279 caltinay 4013 for (index_t y=0; y<num1; y++) {
280     #pragma omp parallel for
281     for (index_t x=0; x<num0; x++) {
282 caltinay 4615 const int baseIndex = first0+x*params.multiplier[0]
283     +(first1+y*params.multiplier[1])*myN0;
284 caltinay 4618 const int srcIndex = (y0+y_mult*y)*num0+(x0+x_mult*x);
285 caltinay 4174 if (!isnan(values[srcIndex])) {
286 caltinay 4615 for (index_t m1=0; m1<params.multiplier[1]; m1++) {
287     for (index_t m0=0; m0<params.multiplier[0]; m0++) {
288 caltinay 4277 const int dataIndex = baseIndex+m0+m1*myN0;
289     double* dest = out.getSampleDataRW(dataIndex);
290     for (index_t q=0; q<dpp; q++) {
291     *dest++ = values[srcIndex];
292     }
293     }
294 caltinay 4174 }
295 caltinay 4013 }
296     }
297     }
298     #else
299     throw RipleyException("readNcGrid(): not compiled with netCDF support");
300     #endif
301     }
302    
303 caltinay 3971 void Rectangle::readBinaryGrid(escript::Data& out, string filename,
304 caltinay 4618 const ReaderParameters& params) const
305 caltinay 3971 {
306 caltinay 4495 // the mapping is not universally correct but should work on our
307     // supported platforms
308 caltinay 4615 switch (params.dataType) {
309 caltinay 4495 case DATATYPE_INT32:
310 caltinay 4615 readBinaryGridImpl<int>(out, filename, params);
311 caltinay 4495 break;
312     case DATATYPE_FLOAT32:
313 caltinay 4615 readBinaryGridImpl<float>(out, filename, params);
314 caltinay 4495 break;
315     case DATATYPE_FLOAT64:
316 caltinay 4615 readBinaryGridImpl<double>(out, filename, params);
317 caltinay 4495 break;
318     default:
319     throw RipleyException("readBinaryGrid(): invalid or unsupported datatype");
320     }
321     }
322    
323     template<typename ValueType>
324     void Rectangle::readBinaryGridImpl(escript::Data& out, const string& filename,
325 caltinay 4618 const ReaderParameters& params) const
326 caltinay 4495 {
327 caltinay 3971 // check destination function space
328     int myN0, myN1;
329     if (out.getFunctionSpace().getTypeCode() == Nodes) {
330 caltinay 4334 myN0 = m_NN[0];
331     myN1 = m_NN[1];
332 caltinay 3971 } else if (out.getFunctionSpace().getTypeCode() == Elements ||
333     out.getFunctionSpace().getTypeCode() == ReducedElements) {
334 caltinay 4334 myN0 = m_NE[0];
335     myN1 = m_NE[1];
336 caltinay 3971 } else
337     throw RipleyException("readBinaryGrid(): invalid function space for output data object");
338    
339     // check file existence and size
340     ifstream f(filename.c_str(), ifstream::binary);
341     if (f.fail()) {
342     throw RipleyException("readBinaryGrid(): cannot open file");
343     }
344     f.seekg(0, ios::end);
345     const int numComp = out.getDataPointSize();
346     const int filesize = f.tellg();
347 caltinay 4615 const int reqsize = params.numValues[0]*params.numValues[1]*numComp*sizeof(ValueType);
348 caltinay 3971 if (filesize < reqsize) {
349     f.close();
350     throw RipleyException("readBinaryGrid(): not enough data in file");
351     }
352    
353     // check if this rank contributes anything
354 caltinay 4615 if (params.first[0] >= m_offset[0]+myN0 ||
355     params.first[0]+params.numValues[0] <= m_offset[0] ||
356     params.first[1] >= m_offset[1]+myN1 ||
357     params.first[1]+params.numValues[1] <= m_offset[1]) {
358 caltinay 3971 f.close();
359     return;
360     }
361    
362     // now determine how much this rank has to write
363    
364     // first coordinates in data object to write to
365 caltinay 4615 const int first0 = max(0, params.first[0]-m_offset[0]);
366     const int first1 = max(0, params.first[1]-m_offset[1]);
367 caltinay 3971 // indices to first value in file
368 caltinay 4615 const int idx0 = max(0, m_offset[0]-params.first[0]);
369     const int idx1 = max(0, m_offset[1]-params.first[1]);
370 caltinay 4277 // number of values to read
371 caltinay 4615 const int num0 = min(params.numValues[0]-idx0, myN0-first0);
372     const int num1 = min(params.numValues[1]-idx1, myN1-first1);
373 caltinay 3971
374     out.requireWrite();
375 caltinay 4495 vector<ValueType> values(num0*numComp);
376 caltinay 3971 const int dpp = out.getNumDataPointsPerSample();
377    
378 caltinay 4529 for (int y=0; y<num1; y++) {
379 caltinay 4615 const int fileofs = numComp*(idx0+(idx1+y)*params.numValues[0]);
380 caltinay 4495 f.seekg(fileofs*sizeof(ValueType));
381     f.read((char*)&values[0], num0*numComp*sizeof(ValueType));
382     for (int x=0; x<num0; x++) {
383 caltinay 4615 const int baseIndex = first0+x*params.multiplier[0]
384     +(first1+y*params.multiplier[1])*myN0;
385     for (int m1=0; m1<params.multiplier[1]; m1++) {
386     for (int m0=0; m0<params.multiplier[0]; m0++) {
387 caltinay 4277 const int dataIndex = baseIndex+m0+m1*myN0;
388     double* dest = out.getSampleDataRW(dataIndex);
389 caltinay 4495 for (int c=0; c<numComp; c++) {
390 caltinay 4529 ValueType val = values[x*numComp+c];
391    
392 caltinay 4615 if (params.byteOrder != BYTEORDER_NATIVE) {
393 caltinay 4529 char* cval = reinterpret_cast<char*>(&val);
394     // this will alter val!!
395     byte_swap32(cval);
396     }
397     if (!std::isnan(val)) {
398 caltinay 4495 for (int q=0; q<dpp; q++) {
399 caltinay 4529 *dest++ = static_cast<double>(val);
400 caltinay 4277 }
401     }
402 caltinay 4174 }
403 caltinay 3971 }
404     }
405     }
406     }
407    
408     f.close();
409     }
410    
411 caltinay 4357 void Rectangle::writeBinaryGrid(const escript::Data& in, string filename,
412     int byteOrder, int dataType) const
413 caltinay 4334 {
414 caltinay 4357 // the mapping is not universally correct but should work on our
415     // supported platforms
416     switch (dataType) {
417     case DATATYPE_INT32:
418     writeBinaryGridImpl<int>(in, filename, byteOrder);
419     break;
420     case DATATYPE_FLOAT32:
421     writeBinaryGridImpl<float>(in, filename, byteOrder);
422     break;
423     case DATATYPE_FLOAT64:
424     writeBinaryGridImpl<double>(in, filename, byteOrder);
425     break;
426     default:
427     throw RipleyException("writeBinaryGrid(): invalid or unsupported datatype");
428     }
429     }
430    
431     template<typename ValueType>
432     void Rectangle::writeBinaryGridImpl(const escript::Data& in,
433     const string& filename, int byteOrder) const
434     {
435 caltinay 4334 // check function space and determine number of points
436     int myN0, myN1;
437     int totalN0, totalN1;
438     if (in.getFunctionSpace().getTypeCode() == Nodes) {
439     myN0 = m_NN[0];
440     myN1 = m_NN[1];
441     totalN0 = m_gNE[0]+1;
442     totalN1 = m_gNE[1]+1;
443     } else if (in.getFunctionSpace().getTypeCode() == Elements ||
444     in.getFunctionSpace().getTypeCode() == ReducedElements) {
445     myN0 = m_NE[0];
446     myN1 = m_NE[1];
447     totalN0 = m_gNE[0];
448     totalN1 = m_gNE[1];
449     } else
450     throw RipleyException("writeBinaryGrid(): invalid function space of data object");
451    
452     const int numComp = in.getDataPointSize();
453     const int dpp = in.getNumDataPointsPerSample();
454    
455     if (numComp > 1 || dpp > 1)
456     throw RipleyException("writeBinaryGrid(): only scalar, single-value data supported");
457    
458 caltinay 4357 const int fileSize = sizeof(ValueType)*numComp*dpp*totalN0*totalN1;
459 caltinay 4334
460     // from here on we know that each sample consists of one value
461 caltinay 4482 FileWriter fw;
462     fw.openFile(filename, fileSize);
463 caltinay 4334 MPIBarrier();
464    
465     for (index_t y=0; y<myN1; y++) {
466 caltinay 4357 const int fileofs = (m_offset[0]+(m_offset[1]+y)*totalN0)*sizeof(ValueType);
467 caltinay 4334 ostringstream oss;
468    
469     for (index_t x=0; x<myN0; x++) {
470 caltinay 4626 const double* sample = in.getSampleDataRO(y*myN0+x);
471 caltinay 4357 ValueType fvalue = static_cast<ValueType>(*sample);
472     if (byteOrder == BYTEORDER_NATIVE) {
473 caltinay 4334 oss.write((char*)&fvalue, sizeof(fvalue));
474     } else {
475     char* value = reinterpret_cast<char*>(&fvalue);
476 caltinay 4357 oss.write(byte_swap32(value), sizeof(fvalue));
477 caltinay 4334 }
478     }
479 caltinay 4482 fw.writeAt(oss, fileofs);
480 caltinay 4334 }
481 caltinay 4482 fw.close();
482 caltinay 4334 }
483    
484 caltinay 3691 void Rectangle::dump(const string& fileName) const
485     {
486     #if USE_SILO
487     string fn(fileName);
488     if (fileName.length() < 6 || fileName.compare(fileName.length()-5, 5, ".silo") != 0) {
489     fn+=".silo";
490     }
491    
492     int driver=DB_HDF5;
493     DBfile* dbfile = NULL;
494 gross 3793 const char* blockDirFmt = "/block%04d";
495 caltinay 3691
496     #ifdef ESYS_MPI
497     PMPIO_baton_t* baton = NULL;
498 gross 3793 const int NUM_SILO_FILES = 1;
499 caltinay 3691 #endif
500    
501     if (m_mpiInfo->size > 1) {
502     #ifdef ESYS_MPI
503     baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
504     0x1337, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
505     PMPIO_DefaultClose, (void*)&driver);
506     // try the fallback driver in case of error
507     if (!baton && driver != DB_PDB) {
508     driver = DB_PDB;
509     baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
510     0x1338, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
511     PMPIO_DefaultClose, (void*)&driver);
512     }
513     if (baton) {
514 caltinay 3766 char siloPath[64];
515     snprintf(siloPath, 64, blockDirFmt, PMPIO_RankInGroup(baton, m_mpiInfo->rank));
516     dbfile = (DBfile*) PMPIO_WaitForBaton(baton, fn.c_str(), siloPath);
517 caltinay 3691 }
518     #endif
519     } else {
520     dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
521     getDescription().c_str(), driver);
522     // try the fallback driver in case of error
523     if (!dbfile && driver != DB_PDB) {
524     driver = DB_PDB;
525     dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
526     getDescription().c_str(), driver);
527     }
528 caltinay 3766 char siloPath[64];
529     snprintf(siloPath, 64, blockDirFmt, 0);
530     DBMkDir(dbfile, siloPath);
531     DBSetDir(dbfile, siloPath);
532 caltinay 3691 }
533    
534     if (!dbfile)
535     throw RipleyException("dump: Could not create Silo file");
536    
537     /*
538     if (driver==DB_HDF5) {
539     // gzip level 1 already provides good compression with minimal
540     // performance penalty. Some tests showed that gzip levels >3 performed
541     // rather badly on escript data both in terms of time and space
542     DBSetCompression("ERRMODE=FALLBACK METHOD=GZIP LEVEL=1");
543     }
544     */
545    
546 caltinay 4334 boost::scoped_ptr<double> x(new double[m_NN[0]]);
547     boost::scoped_ptr<double> y(new double[m_NN[1]]);
548 caltinay 3691 double* coords[2] = { x.get(), y.get() };
549     #pragma omp parallel
550     {
551 caltinay 3722 #pragma omp for nowait
552 caltinay 4334 for (dim_t i0 = 0; i0 < m_NN[0]; i0++) {
553     coords[0][i0]=getLocalCoordinate(i0, 0);
554 caltinay 3691 }
555 caltinay 3722 #pragma omp for nowait
556 caltinay 4334 for (dim_t i1 = 0; i1 < m_NN[1]; i1++) {
557     coords[1][i1]=getLocalCoordinate(i1, 1);
558 caltinay 3691 }
559     }
560 caltinay 4334 int* dims = const_cast<int*>(getNumNodesPerDim());
561 caltinay 3697
562     // write mesh
563 caltinay 4334 DBPutQuadmesh(dbfile, "mesh", NULL, coords, dims, 2, DB_DOUBLE,
564 caltinay 3691 DB_COLLINEAR, NULL);
565    
566 caltinay 3697 // write node ids
567 caltinay 4334 DBPutQuadvar1(dbfile, "nodeId", "mesh", (void*)&m_nodeId[0], dims, 2,
568 caltinay 3697 NULL, 0, DB_INT, DB_NODECENT, NULL);
569    
570     // write element ids
571 caltinay 4334 dims = const_cast<int*>(getNumElementsPerDim());
572 caltinay 3697 DBPutQuadvar1(dbfile, "elementId", "mesh", (void*)&m_elementId[0],
573 caltinay 4334 dims, 2, NULL, 0, DB_INT, DB_ZONECENT, NULL);
574 caltinay 3697
575     // rank 0 writes multimesh and multivar
576 caltinay 3691 if (m_mpiInfo->rank == 0) {
577     vector<string> tempstrings;
578 caltinay 3697 vector<char*> names;
579 caltinay 3691 for (dim_t i=0; i<m_mpiInfo->size; i++) {
580     stringstream path;
581     path << "/block" << setw(4) << setfill('0') << right << i << "/mesh";
582     tempstrings.push_back(path.str());
583 caltinay 3697 names.push_back((char*)tempstrings.back().c_str());
584 caltinay 3691 }
585 caltinay 3697 vector<int> types(m_mpiInfo->size, DB_QUAD_RECT);
586 caltinay 3691 DBSetDir(dbfile, "/");
587 caltinay 3697 DBPutMultimesh(dbfile, "multimesh", m_mpiInfo->size, &names[0],
588     &types[0], NULL);
589     tempstrings.clear();
590     names.clear();
591     for (dim_t i=0; i<m_mpiInfo->size; i++) {
592     stringstream path;
593     path << "/block" << setw(4) << setfill('0') << right << i << "/nodeId";
594     tempstrings.push_back(path.str());
595     names.push_back((char*)tempstrings.back().c_str());
596     }
597     types.assign(m_mpiInfo->size, DB_QUADVAR);
598     DBPutMultivar(dbfile, "nodeId", m_mpiInfo->size, &names[0],
599     &types[0], NULL);
600     tempstrings.clear();
601     names.clear();
602     for (dim_t i=0; i<m_mpiInfo->size; i++) {
603     stringstream path;
604     path << "/block" << setw(4) << setfill('0') << right << i << "/elementId";
605     tempstrings.push_back(path.str());
606     names.push_back((char*)tempstrings.back().c_str());
607     }
608     DBPutMultivar(dbfile, "elementId", m_mpiInfo->size, &names[0],
609     &types[0], NULL);
610 caltinay 3691 }
611    
612     if (m_mpiInfo->size > 1) {
613     #ifdef ESYS_MPI
614     PMPIO_HandOffBaton(baton, dbfile);
615     PMPIO_Finish(baton);
616     #endif
617     } else {
618     DBClose(dbfile);
619     }
620    
621     #else // USE_SILO
622 caltinay 3791 throw RipleyException("dump: no Silo support");
623 caltinay 3691 #endif
624     }
625    
626 caltinay 3697 const int* Rectangle::borrowSampleReferenceIDs(int fsType) const
627 caltinay 3691 {
628 caltinay 3697 switch (fsType) {
629 caltinay 3691 case Nodes:
630 caltinay 3769 case ReducedNodes: // FIXME: reduced
631 caltinay 3691 return &m_nodeId[0];
632 caltinay 3750 case DegreesOfFreedom:
633 caltinay 3769 case ReducedDegreesOfFreedom: // FIXME: reduced
634 caltinay 3750 return &m_dofId[0];
635 caltinay 3691 case Elements:
636 caltinay 3733 case ReducedElements:
637 caltinay 3691 return &m_elementId[0];
638     case FaceElements:
639 caltinay 3733 case ReducedFaceElements:
640 caltinay 3691 return &m_faceId[0];
641 sshaw 4629 case Points:
642     return &m_diracPointNodeIDs[0];
643 caltinay 3691 default:
644     break;
645     }
646    
647     stringstream msg;
648 caltinay 3791 msg << "borrowSampleReferenceIDs: invalid function space type " << fsType;
649 caltinay 3691 throw RipleyException(msg.str());
650     }
651    
652 caltinay 3757 bool Rectangle::ownSample(int fsType, index_t id) const
653 caltinay 3691 {
654 caltinay 3759 if (getMPISize()==1)
655     return true;
656    
657 caltinay 3757 switch (fsType) {
658     case Nodes:
659 caltinay 3769 case ReducedNodes: // FIXME: reduced
660 caltinay 3757 return (m_dofMap[id] < getNumDOF());
661     case DegreesOfFreedom:
662     case ReducedDegreesOfFreedom:
663     return true;
664     case Elements:
665     case ReducedElements:
666     // check ownership of element's bottom left node
667 caltinay 4334 return (m_dofMap[id%m_NE[0]+m_NN[0]*(id/m_NE[0])] < getNumDOF());
668 caltinay 3757 case FaceElements:
669     case ReducedFaceElements:
670 caltinay 3759 {
671 caltinay 3764 // determine which face the sample belongs to before
672 caltinay 3768 // checking ownership of corresponding element's first node
673 caltinay 3759 dim_t n=0;
674 caltinay 4334 for (size_t i=0; i<4; i++) {
675     n+=m_faceCount[i];
676 caltinay 3759 if (id<n) {
677     index_t k;
678     if (i==1)
679 caltinay 4334 k=m_NN[0]-2;
680 caltinay 3759 else if (i==3)
681 caltinay 4334 k=m_NN[0]*(m_NN[1]-2);
682 caltinay 3759 else
683     k=0;
684     // determine whether to move right or up
685 caltinay 4334 const index_t delta=(i/2==0 ? m_NN[0] : 1);
686     return (m_dofMap[k+(id-n+m_faceCount[i])*delta] < getNumDOF());
687 caltinay 3759 }
688     }
689     return false;
690     }
691 caltinay 3757 default:
692     break;
693 caltinay 3702 }
694 caltinay 3757
695     stringstream msg;
696 caltinay 3791 msg << "ownSample: invalid function space type " << fsType;
697 caltinay 3757 throw RipleyException(msg.str());
698 caltinay 3691 }
699    
700 caltinay 3764 void Rectangle::setToNormal(escript::Data& out) const
701 caltinay 3691 {
702 caltinay 3764 if (out.getFunctionSpace().getTypeCode() == FaceElements) {
703     out.requireWrite();
704     #pragma omp parallel
705     {
706     if (m_faceOffset[0] > -1) {
707     #pragma omp for nowait
708 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
709 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
710     // set vector at two quadrature points
711     *o++ = -1.;
712     *o++ = 0.;
713     *o++ = -1.;
714     *o = 0.;
715     }
716     }
717    
718     if (m_faceOffset[1] > -1) {
719     #pragma omp for nowait
720 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
721 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
722     // set vector at two quadrature points
723     *o++ = 1.;
724     *o++ = 0.;
725     *o++ = 1.;
726     *o = 0.;
727     }
728     }
729    
730     if (m_faceOffset[2] > -1) {
731     #pragma omp for nowait
732 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
733 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
734     // set vector at two quadrature points
735     *o++ = 0.;
736     *o++ = -1.;
737     *o++ = 0.;
738     *o = -1.;
739     }
740     }
741    
742     if (m_faceOffset[3] > -1) {
743     #pragma omp for nowait
744 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
745 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
746     // set vector at two quadrature points
747     *o++ = 0.;
748     *o++ = 1.;
749     *o++ = 0.;
750     *o = 1.;
751     }
752     }
753     } // end of parallel section
754     } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
755     out.requireWrite();
756     #pragma omp parallel
757     {
758     if (m_faceOffset[0] > -1) {
759     #pragma omp for nowait
760 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
761 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
762     *o++ = -1.;
763     *o = 0.;
764     }
765     }
766    
767     if (m_faceOffset[1] > -1) {
768     #pragma omp for nowait
769 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
770 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
771     *o++ = 1.;
772     *o = 0.;
773     }
774     }
775    
776     if (m_faceOffset[2] > -1) {
777     #pragma omp for nowait
778 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
779 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
780     *o++ = 0.;
781     *o = -1.;
782     }
783     }
784    
785     if (m_faceOffset[3] > -1) {
786     #pragma omp for nowait
787 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
788 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
789     *o++ = 0.;
790     *o = 1.;
791     }
792     }
793     } // end of parallel section
794    
795     } else {
796     stringstream msg;
797 caltinay 3791 msg << "setToNormal: invalid function space type "
798     << out.getFunctionSpace().getTypeCode();
799 caltinay 3764 throw RipleyException(msg.str());
800     }
801     }
802    
803     void Rectangle::setToSize(escript::Data& out) const
804     {
805     if (out.getFunctionSpace().getTypeCode() == Elements
806     || out.getFunctionSpace().getTypeCode() == ReducedElements) {
807     out.requireWrite();
808     const dim_t numQuad=out.getNumDataPointsPerSample();
809 caltinay 4334 const double size=sqrt(m_dx[0]*m_dx[0]+m_dx[1]*m_dx[1]);
810 caltinay 3764 #pragma omp parallel for
811     for (index_t k = 0; k < getNumElements(); ++k) {
812     double* o = out.getSampleDataRW(k);
813     fill(o, o+numQuad, size);
814     }
815     } else if (out.getFunctionSpace().getTypeCode() == FaceElements
816     || out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
817     out.requireWrite();
818     const dim_t numQuad=out.getNumDataPointsPerSample();
819     #pragma omp parallel
820     {
821     if (m_faceOffset[0] > -1) {
822     #pragma omp for nowait
823 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
824 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
825 caltinay 4334 fill(o, o+numQuad, m_dx[1]);
826 caltinay 3764 }
827     }
828    
829     if (m_faceOffset[1] > -1) {
830     #pragma omp for nowait
831 caltinay 4334 for (index_t k1 = 0; k1 < m_NE[1]; ++k1) {
832 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
833 caltinay 4334 fill(o, o+numQuad, m_dx[1]);
834 caltinay 3764 }
835     }
836    
837     if (m_faceOffset[2] > -1) {
838     #pragma omp for nowait
839 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
840 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
841 caltinay 4334 fill(o, o+numQuad, m_dx[0]);
842 caltinay 3764 }
843     }
844    
845     if (m_faceOffset[3] > -1) {
846     #pragma omp for nowait
847 caltinay 4334 for (index_t k0 = 0; k0 < m_NE[0]; ++k0) {
848 caltinay 3764 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
849 caltinay 4334 fill(o, o+numQuad, m_dx[0]);
850 caltinay 3764 }
851     }
852     } // end of parallel section
853    
854     } else {
855     stringstream msg;
856 caltinay 3791 msg << "setToSize: invalid function space type "
857     << out.getFunctionSpace().getTypeCode();
858 caltinay 3764 throw RipleyException(msg.str());
859     }
860     }
861    
862     void Rectangle::Print_Mesh_Info(const bool full) const
863     {
864     RipleyDomain::Print_Mesh_Info(full);
865     if (full) {
866     cout << " Id Coordinates" << endl;
867     cout.precision(15);
868     cout.setf(ios::scientific, ios::floatfield);
869     for (index_t i=0; i < getNumNodes(); i++) {
870     cout << " " << setw(5) << m_nodeId[i]
871 caltinay 4334 << " " << getLocalCoordinate(i%m_NN[0], 0)
872     << " " << getLocalCoordinate(i/m_NN[0], 1) << endl;
873 caltinay 3764 }
874     }
875     }
876    
877    
878     //protected
879     void Rectangle::assembleCoordinates(escript::Data& arg) const
880     {
881     escriptDataC x = arg.getDataC();
882     int numDim = m_numDim;
883     if (!isDataPointShapeEqual(&x, 1, &numDim))
884     throw RipleyException("setToX: Invalid Data object shape");
885     if (!numSamplesEqual(&x, 1, getNumNodes()))
886     throw RipleyException("setToX: Illegal number of samples in Data object");
887    
888     arg.requireWrite();
889     #pragma omp parallel for
890 caltinay 4334 for (dim_t i1 = 0; i1 < m_NN[1]; i1++) {
891     for (dim_t i0 = 0; i0 < m_NN[0]; i0++) {
892     double* point = arg.getSampleDataRW(i0+m_NN[0]*i1);
893     point[0] = getLocalCoordinate(i0, 0);
894     point[1] = getLocalCoordinate(i1, 1);
895 caltinay 3764 }
896     }
897     }
898    
899     //protected
900 caltinay 4626 void Rectangle::assembleGradient(escript::Data& out, const escript::Data& in) const
901 caltinay 3764 {
902 caltinay 3701 const dim_t numComp = in.getDataPointSize();
903 caltinay 4375 const double cx0 = .21132486540518711775/m_dx[0];
904     const double cx1 = .78867513459481288225/m_dx[0];
905     const double cx2 = 1./m_dx[0];
906     const double cy0 = .21132486540518711775/m_dx[1];
907     const double cy1 = .78867513459481288225/m_dx[1];
908     const double cy2 = 1./m_dx[1];
909 caltinay 3724
910 caltinay 3702 if (out.getFunctionSpace().getTypeCode() == Elements) {
911 caltinay 3760 out.requireWrite();
912 caltinay 3913 #pragma omp parallel
913     {
914     vector<double> f_00(numComp);
915     vector<double> f_01(numComp);
916     vector<double> f_10(numComp);
917     vector<double> f_11(numComp);
918     #pragma omp for
919 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
920     for (index_t k0=0; k0 < m_NE[0]; ++k0) {
921     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
922     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
923     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
924     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
925     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
926 caltinay 3913 for (index_t i=0; i < numComp; ++i) {
927 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx1 + (f_11[i]-f_01[i])*cx0;
928     o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy1 + (f_11[i]-f_10[i])*cy0;
929     o[INDEX3(i,0,1,numComp,2)] = (f_10[i]-f_00[i])*cx1 + (f_11[i]-f_01[i])*cx0;
930     o[INDEX3(i,1,1,numComp,2)] = (f_01[i]-f_00[i])*cy0 + (f_11[i]-f_10[i])*cy1;
931     o[INDEX3(i,0,2,numComp,2)] = (f_10[i]-f_00[i])*cx0 + (f_11[i]-f_01[i])*cx1;
932     o[INDEX3(i,1,2,numComp,2)] = (f_01[i]-f_00[i])*cy1 + (f_11[i]-f_10[i])*cy0;
933     o[INDEX3(i,0,3,numComp,2)] = (f_10[i]-f_00[i])*cx0 + (f_11[i]-f_01[i])*cx1;
934     o[INDEX3(i,1,3,numComp,2)] = (f_01[i]-f_00[i])*cy0 + (f_11[i]-f_10[i])*cy1;
935 caltinay 3913 } // end of component loop i
936     } // end of k0 loop
937     } // end of k1 loop
938     } // end of parallel section
939 caltinay 3711 } else if (out.getFunctionSpace().getTypeCode() == ReducedElements) {
940 caltinay 3760 out.requireWrite();
941 caltinay 3913 #pragma omp parallel
942     {
943     vector<double> f_00(numComp);
944     vector<double> f_01(numComp);
945     vector<double> f_10(numComp);
946     vector<double> f_11(numComp);
947     #pragma omp for
948 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
949     for (index_t k0=0; k0 < m_NE[0]; ++k0) {
950     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
951     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
952     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
953     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
954     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
955 caltinay 3913 for (index_t i=0; i < numComp; ++i) {
956 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_10[i] + f_11[i] - f_00[i] - f_01[i])*cx2/2;
957     o[INDEX3(i,1,0,numComp,2)] = (f_01[i] + f_11[i] - f_00[i] - f_10[i])*cy2/2;
958 caltinay 3913 } // end of component loop i
959     } // end of k0 loop
960     } // end of k1 loop
961     } // end of parallel section
962 caltinay 3707 } else if (out.getFunctionSpace().getTypeCode() == FaceElements) {
963 caltinay 3760 out.requireWrite();
964 caltinay 3722 #pragma omp parallel
965     {
966 caltinay 3913 vector<double> f_00(numComp);
967     vector<double> f_01(numComp);
968     vector<double> f_10(numComp);
969     vector<double> f_11(numComp);
970 caltinay 3722 if (m_faceOffset[0] > -1) {
971     #pragma omp for nowait
972 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
973     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
974     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
975     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(1,k1, m_NN[0])), numComp*sizeof(double));
976     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(1,k1+1, m_NN[0])), numComp*sizeof(double));
977 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
978     for (index_t i=0; i < numComp; ++i) {
979 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx1 + (f_11[i]-f_01[i])*cx0;
980     o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy2;
981     o[INDEX3(i,0,1,numComp,2)] = (f_10[i]-f_00[i])*cx0 + (f_11[i]-f_01[i])*cx1;
982     o[INDEX3(i,1,1,numComp,2)] = (f_01[i]-f_00[i])*cy2;
983 caltinay 3800 } // end of component loop i
984     } // end of k1 loop
985     } // end of face 0
986 caltinay 3722 if (m_faceOffset[1] > -1) {
987     #pragma omp for nowait
988 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
989     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1, m_NN[0])), numComp*sizeof(double));
990     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1+1, m_NN[0])), numComp*sizeof(double));
991     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
992     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
993 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
994     for (index_t i=0; i < numComp; ++i) {
995 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx1 + (f_11[i]-f_01[i])*cx0;
996     o[INDEX3(i,1,0,numComp,2)] = (f_11[i]-f_10[i])*cy2;
997     o[INDEX3(i,0,1,numComp,2)] = (f_10[i]-f_00[i])*cx0 + (f_11[i]-f_01[i])*cx1;
998     o[INDEX3(i,1,1,numComp,2)] = (f_11[i]-f_10[i])*cy2;
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 4375 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx2;
1012     o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy1 + (f_11[i]-f_10[i])*cy0;
1013     o[INDEX3(i,0,1,numComp,2)] = (f_10[i]-f_00[i])*cx2;
1014     o[INDEX3(i,1,1,numComp,2)] = (f_01[i]-f_00[i])*cy0 + (f_11[i]-f_10[i])*cy1;
1015 caltinay 3800 } // end of component loop i
1016     } // end of k0 loop
1017     } // end of face 2
1018 caltinay 3722 if (m_faceOffset[3] > -1) {
1019     #pragma omp for nowait
1020 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1021     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
1022     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1023     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
1024     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1025 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1026     for (index_t i=0; i < numComp; ++i) {
1027 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_11[i]-f_01[i])*cx2;
1028     o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy1 + (f_11[i]-f_10[i])*cy0;
1029     o[INDEX3(i,0,1,numComp,2)] = (f_11[i]-f_01[i])*cx2;
1030     o[INDEX3(i,1,1,numComp,2)] = (f_01[i]-f_00[i])*cy0 + (f_11[i]-f_10[i])*cy1;
1031 caltinay 3800 } // end of component loop i
1032     } // end of k0 loop
1033     } // end of face 3
1034 caltinay 3722 } // end of parallel section
1035 caltinay 3800
1036 caltinay 3711 } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
1037 caltinay 3760 out.requireWrite();
1038 caltinay 3722 #pragma omp parallel
1039     {
1040 caltinay 3913 vector<double> f_00(numComp);
1041     vector<double> f_01(numComp);
1042     vector<double> f_10(numComp);
1043     vector<double> f_11(numComp);
1044 caltinay 3722 if (m_faceOffset[0] > -1) {
1045     #pragma omp for nowait
1046 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1047     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
1048     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
1049     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(1,k1, m_NN[0])), numComp*sizeof(double));
1050     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(1,k1+1, m_NN[0])), numComp*sizeof(double));
1051 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1052     for (index_t i=0; i < numComp; ++i) {
1053 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_10[i] + f_11[i] - f_00[i] - f_01[i])*cx2/2;
1054     o[INDEX3(i,1,0,numComp,2)] = (f_01[i]-f_00[i])*cy2;
1055 caltinay 3800 } // end of component loop i
1056     } // end of k1 loop
1057     } // end of face 0
1058 caltinay 3722 if (m_faceOffset[1] > -1) {
1059     #pragma omp for nowait
1060 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1061     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1, m_NN[0])), numComp*sizeof(double));
1062     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(m_NN[0]-2,k1+1, m_NN[0])), numComp*sizeof(double));
1063     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
1064     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
1065 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1066     for (index_t i=0; i < numComp; ++i) {
1067 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_10[i] + f_11[i] - f_00[i] - f_01[i])*cx2/2;
1068     o[INDEX3(i,1,0,numComp,2)] = (f_11[i]-f_10[i])*cy2;
1069 caltinay 3800 } // end of component loop i
1070     } // end of k1 loop
1071     } // end of face 1
1072 caltinay 3722 if (m_faceOffset[2] > -1) {
1073     #pragma omp for nowait
1074 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1075     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
1076     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,1, m_NN[0])), numComp*sizeof(double));
1077     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
1078     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,1, m_NN[0])), numComp*sizeof(double));
1079 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1080     for (index_t i=0; i < numComp; ++i) {
1081 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_10[i]-f_00[i])*cx2;
1082     o[INDEX3(i,1,0,numComp,2)] = (f_01[i] + f_11[i] - f_00[i] - f_10[i])*cy2/2;
1083 caltinay 3800 } // end of component loop i
1084     } // end of k0 loop
1085     } // end of face 2
1086 caltinay 3722 if (m_faceOffset[3] > -1) {
1087     #pragma omp for nowait
1088 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1089     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
1090     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1091     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-2, m_NN[0])), numComp*sizeof(double));
1092     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1093 caltinay 3722 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1094     for (index_t i=0; i < numComp; ++i) {
1095 caltinay 4375 o[INDEX3(i,0,0,numComp,2)] = (f_11[i]-f_01[i])*cx2;
1096     o[INDEX3(i,1,0,numComp,2)] = (f_01[i] + f_11[i] - f_00[i] - f_10[i])*cy2/2;
1097 caltinay 3800 } // end of component loop i
1098     } // end of k0 loop
1099     } // end of face 3
1100 caltinay 3722 } // end of parallel section
1101 caltinay 3702 }
1102 caltinay 3701 }
1103 caltinay 3697
1104 caltinay 3764 //protected
1105 caltinay 4626 void Rectangle::assembleIntegrate(vector<double>& integrals,
1106     const escript::Data& arg) const
1107 caltinay 3713 {
1108 caltinay 3764 const dim_t numComp = arg.getDataPointSize();
1109 caltinay 4334 const index_t left = (m_offset[0]==0 ? 0 : 1);
1110     const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1111 caltinay 3800 const int fs=arg.getFunctionSpace().getTypeCode();
1112     if (fs == Elements && arg.actsExpanded()) {
1113 caltinay 3713 #pragma omp parallel
1114     {
1115     vector<double> int_local(numComp, 0);
1116 caltinay 4334 const double w = m_dx[0]*m_dx[1]/4.;
1117 caltinay 3722 #pragma omp for nowait
1118 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1119     for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1120     const double* f = arg.getSampleDataRO(INDEX2(k0, k1, m_NE[0]));
1121 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1122 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1123     const double f1 = f[INDEX2(i,1,numComp)];
1124     const double f2 = f[INDEX2(i,2,numComp)];
1125     const double f3 = f[INDEX2(i,3,numComp)];
1126 caltinay 3764 int_local[i]+=(f0+f1+f2+f3)*w;
1127 caltinay 3800 } // end of component loop i
1128     } // end of k0 loop
1129     } // end of k1 loop
1130 caltinay 3713 #pragma omp critical
1131     for (index_t i=0; i<numComp; i++)
1132     integrals[i]+=int_local[i];
1133 caltinay 3722 } // end of parallel section
1134 caltinay 3800
1135     } else if (fs==ReducedElements || (fs==Elements && !arg.actsExpanded())) {
1136 caltinay 4334 const double w = m_dx[0]*m_dx[1];
1137 caltinay 3713 #pragma omp parallel
1138     {
1139     vector<double> int_local(numComp, 0);
1140 caltinay 3722 #pragma omp for nowait
1141 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1142     for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1143     const double* f = arg.getSampleDataRO(INDEX2(k0, k1, m_NE[0]));
1144 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1145 caltinay 3764 int_local[i]+=f[i]*w;
1146 caltinay 3800 }
1147     }
1148     }
1149 caltinay 3713 #pragma omp critical
1150     for (index_t i=0; i<numComp; i++)
1151     integrals[i]+=int_local[i];
1152 caltinay 3722 } // end of parallel section
1153 caltinay 3800
1154     } else if (fs == FaceElements && arg.actsExpanded()) {
1155 caltinay 3713 #pragma omp parallel
1156     {
1157     vector<double> int_local(numComp, 0);
1158 caltinay 4334 const double w0 = m_dx[0]/2.;
1159     const double w1 = m_dx[1]/2.;
1160 caltinay 3713 if (m_faceOffset[0] > -1) {
1161 caltinay 3722 #pragma omp for nowait
1162 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1163 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[0]+k1);
1164 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1165 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1166     const double f1 = f[INDEX2(i,1,numComp)];
1167 caltinay 3764 int_local[i]+=(f0+f1)*w1;
1168 caltinay 3800 } // end of component loop i
1169     } // end of k1 loop
1170 caltinay 3713 }
1171    
1172     if (m_faceOffset[1] > -1) {
1173 caltinay 3722 #pragma omp for nowait
1174 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1175 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[1]+k1);
1176 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1177 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1178     const double f1 = f[INDEX2(i,1,numComp)];
1179 caltinay 3764 int_local[i]+=(f0+f1)*w1;
1180 caltinay 3800 } // end of component loop i
1181     } // end of k1 loop
1182 caltinay 3713 }
1183    
1184     if (m_faceOffset[2] > -1) {
1185 caltinay 3722 #pragma omp for nowait
1186 caltinay 4334 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1187 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[2]+k0);
1188 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1189 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1190     const double f1 = f[INDEX2(i,1,numComp)];
1191 caltinay 3764 int_local[i]+=(f0+f1)*w0;
1192 caltinay 3800 } // end of component loop i
1193     } // end of k0 loop
1194 caltinay 3713 }
1195    
1196     if (m_faceOffset[3] > -1) {
1197 caltinay 3722 #pragma omp for nowait
1198 caltinay 4334 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1199 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[3]+k0);
1200 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1201 caltinay 3769 const double f0 = f[INDEX2(i,0,numComp)];
1202     const double f1 = f[INDEX2(i,1,numComp)];
1203 caltinay 3764 int_local[i]+=(f0+f1)*w0;
1204 caltinay 3800 } // end of component loop i
1205     } // end of k0 loop
1206 caltinay 3713 }
1207     #pragma omp critical
1208     for (index_t i=0; i<numComp; i++)
1209     integrals[i]+=int_local[i];
1210 caltinay 3722 } // end of parallel section
1211 caltinay 3800
1212     } else if (fs==ReducedFaceElements || (fs==FaceElements && !arg.actsExpanded())) {
1213 caltinay 3713 #pragma omp parallel
1214     {
1215     vector<double> int_local(numComp, 0);
1216     if (m_faceOffset[0] > -1) {
1217 caltinay 3722 #pragma omp for nowait
1218 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1219 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[0]+k1);
1220 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1221 caltinay 4334 int_local[i]+=f[i]*m_dx[1];
1222 caltinay 3800 }
1223     }
1224 caltinay 3713 }
1225    
1226     if (m_faceOffset[1] > -1) {
1227 caltinay 3722 #pragma omp for nowait
1228 caltinay 4334 for (index_t k1 = bottom; k1 < bottom+m_ownNE[1]; ++k1) {
1229 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[1]+k1);
1230 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1231 caltinay 4334 int_local[i]+=f[i]*m_dx[1];
1232 caltinay 3800 }
1233     }
1234 caltinay 3713 }
1235    
1236     if (m_faceOffset[2] > -1) {
1237 caltinay 3722 #pragma omp for nowait
1238 caltinay 4334 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1239 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[2]+k0);
1240 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1241 caltinay 4334 int_local[i]+=f[i]*m_dx[0];
1242 caltinay 3800 }
1243     }
1244 caltinay 3713 }
1245    
1246     if (m_faceOffset[3] > -1) {
1247 caltinay 3722 #pragma omp for nowait
1248 caltinay 4334 for (index_t k0 = left; k0 < left+m_ownNE[0]; ++k0) {
1249 caltinay 3764 const double* f = arg.getSampleDataRO(m_faceOffset[3]+k0);
1250 caltinay 3713 for (index_t i=0; i < numComp; ++i) {
1251 caltinay 4334 int_local[i]+=f[i]*m_dx[0];
1252 caltinay 3800 }
1253     }
1254 caltinay 3713 }
1255    
1256     #pragma omp critical
1257     for (index_t i=0; i<numComp; i++)
1258     integrals[i]+=int_local[i];
1259 caltinay 3722 } // end of parallel section
1260 caltinay 3800 } // function space selector
1261 caltinay 3713 }
1262    
1263 caltinay 3691 //protected
1264 caltinay 3756 dim_t Rectangle::insertNeighbourNodes(IndexVector& index, index_t node) const
1265     {
1266 caltinay 4334 const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1267     const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1268 caltinay 3756 const int x=node%nDOF0;
1269     const int y=node/nDOF0;
1270     dim_t num=0;
1271     // loop through potential neighbours and add to index if positions are
1272     // within bounds
1273     for (int i1=-1; i1<2; i1++) {
1274     for (int i0=-1; i0<2; i0++) {
1275     // skip node itself
1276     if (i0==0 && i1==0)
1277     continue;
1278     // location of neighbour node
1279     const int nx=x+i0;
1280     const int ny=y+i1;
1281     if (nx>=0 && ny>=0 && nx<nDOF0 && ny<nDOF1) {
1282     index.push_back(ny*nDOF0+nx);
1283     num++;
1284     }
1285     }
1286     }
1287    
1288     return num;
1289     }
1290    
1291     //protected
1292 caltinay 4626 void Rectangle::nodesToDOF(escript::Data& out, const escript::Data& in) const
1293 caltinay 3756 {
1294     const dim_t numComp = in.getDataPointSize();
1295     out.requireWrite();
1296    
1297 caltinay 4334 const index_t left = (m_offset[0]==0 ? 0 : 1);
1298     const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1299     const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1300     const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1301 caltinay 3756 #pragma omp parallel for
1302     for (index_t i=0; i<nDOF1; i++) {
1303     for (index_t j=0; j<nDOF0; j++) {
1304 caltinay 4334 const index_t n=j+left+(i+bottom)*m_NN[0];
1305 caltinay 3756 const double* src=in.getSampleDataRO(n);
1306     copy(src, src+numComp, out.getSampleDataRW(j+i*nDOF0));
1307     }
1308     }
1309     }
1310    
1311     //protected
1312 caltinay 4626 void Rectangle::dofToNodes(escript::Data& out, const escript::Data& in) const
1313 caltinay 3756 {
1314     const dim_t numComp = in.getDataPointSize();
1315     Paso_Coupler* coupler = Paso_Coupler_alloc(m_connector, numComp);
1316 caltinay 4633 // expand data object if necessary to be able to grab the whole data
1317     const_cast<escript::Data*>(&in)->expand();
1318 caltinay 4626 Paso_Coupler_startCollect(coupler, in.getSampleDataRO(0));
1319 caltinay 3756
1320     const dim_t numDOF = getNumDOF();
1321     out.requireWrite();
1322     const double* buffer = Paso_Coupler_finishCollect(coupler);
1323    
1324     #pragma omp parallel for
1325     for (index_t i=0; i<getNumNodes(); i++) {
1326     const double* src=(m_dofMap[i]<numDOF ?
1327     in.getSampleDataRO(m_dofMap[i])
1328     : &buffer[(m_dofMap[i]-numDOF)*numComp]);
1329     copy(src, src+numComp, out.getSampleDataRW(i));
1330     }
1331 caltinay 4002 Paso_Coupler_free(coupler);
1332 caltinay 3756 }
1333    
1334 caltinay 3691 //private
1335     void Rectangle::populateSampleIds()
1336     {
1337 caltinay 4334 // degrees of freedom are numbered from left to right, bottom to top in
1338     // each rank, continuing on the next rank (ranks also go left-right,
1339     // bottom-top).
1340     // This means rank 0 has id 0...n0-1, rank 1 has id n0...n1-1 etc. which
1341     // helps when writing out data rank after rank.
1342 caltinay 3697
1343     // build node distribution vector first.
1344 caltinay 4334 // rank i owns m_nodeDistribution[i+1]-nodeDistribution[i] nodes which is
1345     // constant for all ranks in this implementation
1346 caltinay 3697 m_nodeDistribution.assign(m_mpiInfo->size+1, 0);
1347 caltinay 3752 const dim_t numDOF=getNumDOF();
1348     for (dim_t k=1; k<m_mpiInfo->size; k++) {
1349     m_nodeDistribution[k]=k*numDOF;
1350 caltinay 3697 }
1351     m_nodeDistribution[m_mpiInfo->size]=getNumDataPointsGlobal();
1352 caltinay 3691 m_nodeId.resize(getNumNodes());
1353 caltinay 3753 m_dofId.resize(numDOF);
1354     m_elementId.resize(getNumElements());
1355 caltinay 4334
1356     // populate face element counts
1357     //left
1358     if (m_offset[0]==0)
1359     m_faceCount[0]=m_NE[1];
1360     else
1361     m_faceCount[0]=0;
1362     //right
1363     if (m_mpiInfo->rank%m_NX[0]==m_NX[0]-1)
1364     m_faceCount[1]=m_NE[1];
1365     else
1366     m_faceCount[1]=0;
1367     //bottom
1368     if (m_offset[1]==0)
1369     m_faceCount[2]=m_NE[0];
1370     else
1371     m_faceCount[2]=0;
1372     //top
1373     if (m_mpiInfo->rank/m_NX[0]==m_NX[1]-1)
1374     m_faceCount[3]=m_NE[0];
1375     else
1376     m_faceCount[3]=0;
1377    
1378 caltinay 3753 m_faceId.resize(getNumFaceElements());
1379 caltinay 3697
1380 caltinay 4334 const index_t left = (m_offset[0]==0 ? 0 : 1);
1381     const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1382     const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1383     const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1384    
1385     #define globalNodeId(x,y) \
1386     ((m_offset[0]+x)/nDOF0)*nDOF0*nDOF1+(m_offset[0]+x)%nDOF0 \
1387     + ((m_offset[1]+y)/nDOF1)*nDOF0*nDOF1*m_NX[0]+((m_offset[1]+y)%nDOF1)*nDOF0
1388    
1389     // set corner id's outside the parallel region
1390     m_nodeId[0] = globalNodeId(0, 0);
1391     m_nodeId[m_NN[0]-1] = globalNodeId(m_NN[0]-1, 0);
1392     m_nodeId[m_NN[0]*(m_NN[1]-1)] = globalNodeId(0, m_NN[1]-1);
1393     m_nodeId[m_NN[0]*m_NN[1]-1] = globalNodeId(m_NN[0]-1,m_NN[1]-1);
1394     #undef globalNodeId
1395    
1396 caltinay 3753 #pragma omp parallel
1397     {
1398 caltinay 4334 // populate degrees of freedom and own nodes (identical id)
1399 caltinay 3753 #pragma omp for nowait
1400 caltinay 4334 for (dim_t i=0; i<nDOF1; i++) {
1401     for (dim_t j=0; j<nDOF0; j++) {
1402     const index_t nodeIdx=j+left+(i+bottom)*m_NN[0];
1403     const index_t dofIdx=j+i*nDOF0;
1404     m_dofId[dofIdx] = m_nodeId[nodeIdx]
1405     = m_nodeDistribution[m_mpiInfo->rank]+dofIdx;
1406 caltinay 3753 }
1407 caltinay 3697 }
1408    
1409 caltinay 4334 // populate the rest of the nodes (shared with other ranks)
1410     if (m_faceCount[0]==0) { // left column
1411 caltinay 3753 #pragma omp for nowait
1412 caltinay 4334 for (dim_t i=0; i<nDOF1; i++) {
1413     const index_t nodeIdx=(i+bottom)*m_NN[0];
1414     const index_t dofId=(i+1)*nDOF0-1;
1415     m_nodeId[nodeIdx]
1416     = m_nodeDistribution[m_mpiInfo->rank-1]+dofId;
1417     }
1418     }
1419     if (m_faceCount[1]==0) { // right column
1420     #pragma omp for nowait
1421     for (dim_t i=0; i<nDOF1; i++) {
1422     const index_t nodeIdx=(i+bottom+1)*m_NN[0]-1;
1423     const index_t dofId=i*nDOF0;
1424     m_nodeId[nodeIdx]
1425     = m_nodeDistribution[m_mpiInfo->rank+1]+dofId;
1426     }
1427     }
1428     if (m_faceCount[2]==0) { // bottom row
1429     #pragma omp for nowait
1430     for (dim_t i=0; i<nDOF0; i++) {
1431     const index_t nodeIdx=i+left;
1432     const index_t dofId=nDOF0*(nDOF1-1)+i;
1433     m_nodeId[nodeIdx]
1434     = m_nodeDistribution[m_mpiInfo->rank-m_NX[0]]+dofId;
1435     }
1436     }
1437     if (m_faceCount[3]==0) { // top row
1438     #pragma omp for nowait
1439     for (dim_t i=0; i<nDOF0; i++) {
1440     const index_t nodeIdx=m_NN[0]*(m_NN[1]-1)+i+left;
1441     const index_t dofId=i;
1442     m_nodeId[nodeIdx]
1443     = m_nodeDistribution[m_mpiInfo->rank+m_NX[0]]+dofId;
1444     }
1445     }
1446 caltinay 3752
1447 caltinay 4334 // populate element id's
1448 caltinay 3753 #pragma omp for nowait
1449 caltinay 4334 for (dim_t i1=0; i1<m_NE[1]; i1++) {
1450     for (dim_t i0=0; i0<m_NE[0]; i0++) {
1451     m_elementId[i0+i1*m_NE[0]]=(m_offset[1]+i1)*m_gNE[0]+m_offset[0]+i0;
1452 caltinay 3755 }
1453     }
1454 caltinay 3753
1455     // face elements
1456     #pragma omp for
1457     for (dim_t k=0; k<getNumFaceElements(); k++)
1458     m_faceId[k]=k;
1459     } // end parallel section
1460    
1461 caltinay 3735 m_nodeTags.assign(getNumNodes(), 0);
1462     updateTagsInUse(Nodes);
1463 caltinay 3697
1464 caltinay 3735 m_elementTags.assign(getNumElements(), 0);
1465     updateTagsInUse(Elements);
1466 caltinay 3697
1467 caltinay 3722 // generate face offset vector and set face tags
1468     const index_t LEFT=1, RIGHT=2, BOTTOM=10, TOP=20;
1469     const index_t faceTag[] = { LEFT, RIGHT, BOTTOM, TOP };
1470 caltinay 4334 m_faceOffset.assign(4, -1);
1471 caltinay 3722 m_faceTags.clear();
1472 caltinay 3704 index_t offset=0;
1473 caltinay 4334 for (size_t i=0; i<4; i++) {
1474     if (m_faceCount[i]>0) {
1475 caltinay 3704 m_faceOffset[i]=offset;
1476 caltinay 4334 offset+=m_faceCount[i];
1477     m_faceTags.insert(m_faceTags.end(), m_faceCount[i], faceTag[i]);
1478 caltinay 3704 }
1479     }
1480 caltinay 3722 setTagMap("left", LEFT);
1481     setTagMap("right", RIGHT);
1482     setTagMap("bottom", BOTTOM);
1483     setTagMap("top", TOP);
1484     updateTagsInUse(FaceElements);
1485 caltinay 3691 }
1486    
1487 caltinay 3699 //private
1488 caltinay 3756 void Rectangle::createPattern()
1489 caltinay 3699 {
1490 caltinay 4334 const dim_t nDOF0 = (m_gNE[0]+1)/m_NX[0];
1491     const dim_t nDOF1 = (m_gNE[1]+1)/m_NX[1];
1492     const index_t left = (m_offset[0]==0 ? 0 : 1);
1493     const index_t bottom = (m_offset[1]==0 ? 0 : 1);
1494 caltinay 3756
1495     // populate node->DOF mapping with own degrees of freedom.
1496     // The rest is assigned in the loop further down
1497     m_dofMap.assign(getNumNodes(), 0);
1498     #pragma omp parallel for
1499 caltinay 3766 for (index_t i=bottom; i<bottom+nDOF1; i++) {
1500     for (index_t j=left; j<left+nDOF0; j++) {
1501 caltinay 4334 m_dofMap[i*m_NN[0]+j]=(i-bottom)*nDOF0+j-left;
1502 caltinay 3756 }
1503     }
1504    
1505     // build list of shared components and neighbours by looping through
1506     // all potential neighbouring ranks and checking if positions are
1507 caltinay 3754 // within bounds
1508 caltinay 3756 const dim_t numDOF=nDOF0*nDOF1;
1509     vector<IndexVector> colIndices(numDOF); // for the couple blocks
1510     RankVector neighbour;
1511     IndexVector offsetInShared(1,0);
1512     IndexVector sendShared, recvShared;
1513     int numShared=0;
1514 caltinay 4334 const int x=m_mpiInfo->rank%m_NX[0];
1515     const int y=m_mpiInfo->rank/m_NX[0];
1516 caltinay 3754 for (int i1=-1; i1<2; i1++) {
1517     for (int i0=-1; i0<2; i0++) {
1518 caltinay 3756 // skip this rank
1519 caltinay 3754 if (i0==0 && i1==0)
1520     continue;
1521 caltinay 3756 // location of neighbour rank
1522 caltinay 3754 const int nx=x+i0;
1523     const int ny=y+i1;
1524 caltinay 4334 if (nx>=0 && ny>=0 && nx<m_NX[0] && ny<m_NX[1]) {
1525     neighbour.push_back(ny*m_NX[0]+nx);
1526 caltinay 3756 if (i0==0) {
1527     // sharing top or bottom edge
1528     const int firstDOF=(i1==-1 ? 0 : numDOF-nDOF0);
1529 caltinay 4334 const int firstNode=(i1==-1 ? left : m_NN[0]*(m_NN[1]-1)+left);
1530 caltinay 3756 offsetInShared.push_back(offsetInShared.back()+nDOF0);
1531     for (dim_t i=0; i<nDOF0; i++, numShared++) {
1532     sendShared.push_back(firstDOF+i);
1533     recvShared.push_back(numDOF+numShared);
1534     if (i>0)
1535     colIndices[firstDOF+i-1].push_back(numShared);
1536     colIndices[firstDOF+i].push_back(numShared);
1537     if (i<nDOF0-1)
1538     colIndices[firstDOF+i+1].push_back(numShared);
1539     m_dofMap[firstNode+i]=numDOF+numShared;
1540     }
1541     } else if (i1==0) {
1542     // sharing left or right edge
1543     const int firstDOF=(i0==-1 ? 0 : nDOF0-1);
1544 caltinay 4334 const int firstNode=(i0==-1 ? bottom*m_NN[0] : (bottom+1)*m_NN[0]-1);
1545 caltinay 3756 offsetInShared.push_back(offsetInShared.back()+nDOF1);
1546     for (dim_t i=0; i<nDOF1; i++, numShared++) {
1547     sendShared.push_back(firstDOF+i*nDOF0);
1548     recvShared.push_back(numDOF+numShared);
1549     if (i>0)
1550     colIndices[firstDOF+(i-1)*nDOF0].push_back(numShared);
1551     colIndices[firstDOF+i*nDOF0].push_back(numShared);
1552     if (i<nDOF1-1)
1553     colIndices[firstDOF+(i+1)*nDOF0].push_back(numShared);
1554 caltinay 4334 m_dofMap[firstNode+i*m_NN[0]]=numDOF+numShared;
1555 caltinay 3756 }
1556     } else {
1557     // sharing a node
1558     const int dof=(i0+1)/2*(nDOF0-1)+(i1+1)/2*(numDOF-nDOF0);
1559 caltinay 4334 const int node=(i0+1)/2*(m_NN[0]-1)+(i1+1)/2*m_NN[0]*(m_NN[1]-1);
1560 caltinay 3756 offsetInShared.push_back(offsetInShared.back()+1);
1561     sendShared.push_back(dof);
1562     recvShared.push_back(numDOF+numShared);
1563     colIndices[dof].push_back(numShared);
1564     m_dofMap[node]=numDOF+numShared;
1565     ++numShared;
1566     }
1567 caltinay 3754 }
1568 caltinay 3699 }
1569     }
1570 caltinay 3754
1571 caltinay 3756 // create connector
1572     Paso_SharedComponents *snd_shcomp = Paso_SharedComponents_alloc(
1573     numDOF, neighbour.size(), &neighbour[0], &sendShared[0],
1574     &offsetInShared[0], 1, 0, m_mpiInfo);
1575     Paso_SharedComponents *rcv_shcomp = Paso_SharedComponents_alloc(
1576     numDOF, neighbour.size(), &neighbour[0], &recvShared[0],
1577     &offsetInShared[0], 1, 0, m_mpiInfo);
1578     m_connector = Paso_Connector_alloc(snd_shcomp, rcv_shcomp);
1579     Paso_SharedComponents_free(snd_shcomp);
1580     Paso_SharedComponents_free(rcv_shcomp);
1581 caltinay 3754
1582 caltinay 3756 // create main and couple blocks
1583     Paso_Pattern *mainPattern = createMainPattern();
1584     Paso_Pattern *colPattern, *rowPattern;
1585     createCouplePatterns(colIndices, numShared, &colPattern, &rowPattern);
1586 caltinay 3754
1587 caltinay 3756 // allocate paso distribution
1588     Paso_Distribution* distribution = Paso_Distribution_alloc(m_mpiInfo,
1589     const_cast<index_t*>(&m_nodeDistribution[0]), 1, 0);
1590 caltinay 3755
1591 caltinay 3756 // finally create the system matrix
1592     m_pattern = Paso_SystemMatrixPattern_alloc(MATRIX_FORMAT_DEFAULT,
1593     distribution, distribution, mainPattern, colPattern, rowPattern,
1594     m_connector, m_connector);
1595 caltinay 3755
1596 caltinay 3756 Paso_Distribution_free(distribution);
1597 caltinay 3755
1598 caltinay 3756 // useful debug output
1599     /*
1600     cout << "--- rcv_shcomp ---" << endl;
1601     cout << "numDOF=" << numDOF << ", numNeighbors=" << neighbour.size() << endl;
1602     for (size_t i=0; i<neighbour.size(); i++) {
1603     cout << "neighbor[" << i << "]=" << neighbour[i]
1604     << " offsetInShared[" << i+1 << "]=" << offsetInShared[i+1] << endl;
1605 caltinay 3699 }
1606 caltinay 3756 for (size_t i=0; i<recvShared.size(); i++) {
1607     cout << "shared[" << i << "]=" << recvShared[i] << endl;
1608     }
1609     cout << "--- snd_shcomp ---" << endl;
1610     for (size_t i=0; i<sendShared.size(); i++) {
1611     cout << "shared[" << i << "]=" << sendShared[i] << endl;
1612     }
1613     cout << "--- dofMap ---" << endl;
1614     for (size_t i=0; i<m_dofMap.size(); i++) {
1615     cout << "m_dofMap[" << i << "]=" << m_dofMap[i] << endl;
1616     }
1617     cout << "--- colIndices ---" << endl;
1618     for (size_t i=0; i<colIndices.size(); i++) {
1619     cout << "colIndices[" << i << "].size()=" << colIndices[i].size() << endl;
1620     }
1621     */
1622 caltinay 3754
1623 caltinay 3756 /*
1624     cout << "--- main_pattern ---" << endl;
1625     cout << "M=" << mainPattern->numOutput << ", N=" << mainPattern->numInput << endl;
1626     for (size_t i=0; i<mainPattern->numOutput+1; i++) {
1627     cout << "ptr[" << i << "]=" << mainPattern->ptr[i] << endl;
1628     }
1629     for (size_t i=0; i<mainPattern->ptr[mainPattern->numOutput]; i++) {
1630     cout << "index[" << i << "]=" << mainPattern->index[i] << endl;
1631     }
1632     */
1633 caltinay 3754
1634 caltinay 3756 /*
1635     cout << "--- colCouple_pattern ---" << endl;
1636     cout << "M=" << colPattern->numOutput << ", N=" << colPattern->numInput << endl;
1637     for (size_t i=0; i<colPattern->numOutput+1; i++) {
1638     cout << "ptr[" << i << "]=" << colPattern->ptr[i] << endl;
1639     }
1640     for (size_t i=0; i<colPattern->ptr[colPattern->numOutput]; i++) {
1641     cout << "index[" << i << "]=" << colPattern->index[i] << endl;
1642     }
1643     */
1644 caltinay 3754
1645 caltinay 3756 /*
1646     cout << "--- rowCouple_pattern ---" << endl;
1647     cout << "M=" << rowPattern->numOutput << ", N=" << rowPattern->numInput << endl;
1648     for (size_t i=0; i<rowPattern->numOutput+1; i++) {
1649     cout << "ptr[" << i << "]=" << rowPattern->ptr[i] << endl;
1650 caltinay 3699 }
1651 caltinay 3756 for (size_t i=0; i<rowPattern->ptr[rowPattern->numOutput]; i++) {
1652     cout << "index[" << i << "]=" << rowPattern->index[i] << endl;
1653     }
1654     */
1655    
1656     Paso_Pattern_free(mainPattern);
1657     Paso_Pattern_free(colPattern);
1658     Paso_Pattern_free(rowPattern);
1659 caltinay 3699 }
1660    
1661 caltinay 3776 //private
1662     void Rectangle::addToMatrixAndRHS(Paso_SystemMatrix* S, escript::Data& F,
1663     const vector<double>& EM_S, const vector<double>& EM_F, bool addS,
1664     bool addF, index_t firstNode, dim_t nEq, dim_t nComp) const
1665     {
1666     IndexVector rowIndex;
1667     rowIndex.push_back(m_dofMap[firstNode]);
1668     rowIndex.push_back(m_dofMap[firstNode+1]);
1669 caltinay 4334 rowIndex.push_back(m_dofMap[firstNode+m_NN[0]]);
1670     rowIndex.push_back(m_dofMap[firstNode+m_NN[0]+1]);
1671 caltinay 3776 if (addF) {
1672     double *F_p=F.getSampleDataRW(0);
1673     for (index_t i=0; i<rowIndex.size(); i++) {
1674     if (rowIndex[i]<getNumDOF()) {
1675     for (index_t eq=0; eq<nEq; eq++) {
1676     F_p[INDEX2(eq, rowIndex[i], nEq)]+=EM_F[INDEX2(eq,i,nEq)];
1677     }
1678     }
1679     }
1680     }
1681     if (addS) {
1682     addToSystemMatrix(S, rowIndex, nEq, rowIndex, nComp, EM_S);
1683     }
1684     }
1685    
1686 caltinay 3702 //protected
1687 caltinay 3711 void Rectangle::interpolateNodesOnElements(escript::Data& out,
1688 caltinay 4626 const escript::Data& in,
1689     bool reduced) const
1690 caltinay 3702 {
1691     const dim_t numComp = in.getDataPointSize();
1692 caltinay 3711 if (reduced) {
1693 caltinay 3760 out.requireWrite();
1694 caltinay 3913 const double c0 = 0.25;
1695     #pragma omp parallel
1696     {
1697     vector<double> f_00(numComp);
1698     vector<double> f_01(numComp);
1699     vector<double> f_10(numComp);
1700     vector<double> f_11(numComp);
1701     #pragma omp for
1702 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1703     for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1704     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
1705     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
1706     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
1707     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
1708     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
1709 caltinay 3913 for (index_t i=0; i < numComp; ++i) {
1710     o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i] + f_10[i] + f_11[i]);
1711     } /* end of component loop i */
1712     } /* end of k0 loop */
1713     } /* end of k1 loop */
1714     } /* end of parallel section */
1715 caltinay 3711 } else {
1716 caltinay 3760 out.requireWrite();
1717 caltinay 3913 const double c0 = 0.16666666666666666667;
1718     const double c1 = 0.044658198738520451079;
1719     const double c2 = 0.62200846792814621559;
1720     #pragma omp parallel
1721     {
1722     vector<double> f_00(numComp);
1723     vector<double> f_01(numComp);
1724     vector<double> f_10(numComp);
1725     vector<double> f_11(numComp);
1726     #pragma omp for
1727 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1728     for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1729     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,k1, m_NN[0])), numComp*sizeof(double));
1730     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,k1+1, m_NN[0])), numComp*sizeof(double));
1731     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,k1, m_NN[0])), numComp*sizeof(double));
1732     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,k1+1, m_NN[0])), numComp*sizeof(double));
1733     double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE[0]));
1734 caltinay 3913 for (index_t i=0; i < numComp; ++i) {
1735     o[INDEX2(i,numComp,0)] = c0*(f_01[i] + f_10[i]) + c1*f_11[i] + c2*f_00[i];
1736     o[INDEX2(i,numComp,1)] = c0*(f_00[i] + f_11[i]) + c1*f_01[i] + c2*f_10[i];
1737     o[INDEX2(i,numComp,2)] = c0*(f_00[i] + f_11[i]) + c1*f_10[i] + c2*f_01[i];
1738     o[INDEX2(i,numComp,3)] = c0*(f_01[i] + f_10[i]) + c1*f_00[i] + c2*f_11[i];
1739     } /* end of component loop i */
1740     } /* end of k0 loop */
1741     } /* end of k1 loop */
1742     } /* end of parallel section */
1743 caltinay 3711 }
1744 caltinay 3702 }
1745    
1746     //protected
1747 caltinay 4626 void Rectangle::interpolateNodesOnFaces(escript::Data& out,
1748     const escript::Data& in,
1749 caltinay 3711 bool reduced) const
1750 caltinay 3702 {
1751 caltinay 3704 const dim_t numComp = in.getDataPointSize();
1752 caltinay 3711 if (reduced) {
1753 caltinay 3760 out.requireWrite();
1754 caltinay 3724 #pragma omp parallel
1755     {
1756 caltinay 3913 vector<double> f_00(numComp);
1757     vector<double> f_01(numComp);
1758     vector<double> f_10(numComp);
1759     vector<double> f_11(numComp);
1760 caltinay 3724 if (m_faceOffset[0] > -1) {
1761     #pragma omp for nowait
1762 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1763     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
1764     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
1765 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1766     for (index_t i=0; i < numComp; ++i) {
1767 caltinay 4375 o[INDEX2(i,numComp,0)] = (f_00[i] + f_01[i])/2;
1768 caltinay 3724 } /* end of component loop i */
1769     } /* end of k1 loop */
1770     } /* end of face 0 */
1771     if (m_faceOffset[1] > -1) {
1772     #pragma omp for nowait
1773 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1774     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
1775     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
1776 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1777     for (index_t i=0; i < numComp; ++i) {
1778 caltinay 4375 o[INDEX2(i,numComp,0)] = (f_10[i] + f_11[i])/2;
1779 caltinay 3724 } /* end of component loop i */
1780     } /* end of k1 loop */
1781     } /* end of face 1 */
1782     if (m_faceOffset[2] > -1) {
1783     #pragma omp for nowait
1784 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1785     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(k0,0, m_NN[0])), numComp*sizeof(double));
1786     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(k0+1,0, m_NN[0])), numComp*sizeof(double));
1787 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1788     for (index_t i=0; i < numComp; ++i) {
1789 caltinay 4375 o[INDEX2(i,numComp,0)] = (f_00[i] + f_10[i])/2;
1790 caltinay 3724 } /* end of component loop i */
1791     } /* end of k0 loop */
1792     } /* end of face 2 */
1793     if (m_faceOffset[3] > -1) {
1794     #pragma omp for nowait
1795 caltinay 4334 for (index_t k0=0; k0 < m_NE[0]; ++k0) {
1796     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(k0,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1797     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(k0+1,m_NN[1]-1, m_NN[0])), numComp*sizeof(double));
1798 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1799     for (index_t i=0; i < numComp; ++i) {
1800 caltinay 4375 o[INDEX2(i,numComp,0)] = (f_01[i] + f_11[i])/2;
1801 caltinay 3724 } /* end of component loop i */
1802     } /* end of k0 loop */
1803     } /* end of face 3 */
1804 caltinay 3913 } /* end of parallel section */
1805 caltinay 3711 } else {
1806 caltinay 3760 out.requireWrite();
1807 caltinay 3724 const double c0 = 0.21132486540518711775;
1808     const double c1 = 0.78867513459481288225;
1809     #pragma omp parallel
1810     {
1811 caltinay 3913 vector<double> f_00(numComp);
1812     vector<double> f_01(numComp);
1813     vector<double> f_10(numComp);
1814     vector<double> f_11(numComp);
1815 caltinay 3724 if (m_faceOffset[0] > -1) {
1816     #pragma omp for nowait
1817 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1818     memcpy(&f_00[0], in.getSampleDataRO(INDEX2(0,k1, m_NN[0])), numComp*sizeof(double));
1819     memcpy(&f_01[0], in.getSampleDataRO(INDEX2(0,k1+1, m_NN[0])), numComp*sizeof(double));
1820 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1821     for (index_t i=0; i < numComp; ++i) {
1822 caltinay 3913 o[INDEX2(i,numComp,0)] = c0*f_01[i] + c1*f_00[i];
1823     o[INDEX2(i,numComp,1)] = c0*f_00[i] + c1*f_01[i];
1824 caltinay 3724 } /* end of component loop i */
1825     } /* end of k1 loop */
1826     } /* end of face 0 */
1827     if (m_faceOffset[1] > -1) {
1828     #pragma omp for nowait
1829 caltinay 4334 for (index_t k1=0; k1 < m_NE[1]; ++k1) {
1830     memcpy(&f_10[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1, m_NN[0])), numComp*sizeof(double));
1831     memcpy(&f_11[0], in.getSampleDataRO(INDEX2(m_NN[0]-1,k1+1, m_NN[0])), numComp*sizeof(double));
1832 caltinay 3724 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1833     for (index_t i=0; i < numComp; ++i) {
1834 caltinay 3913 o[INDEX2(i,numComp,0)] = c1*f_10[i] + c0*f_11[i];
1835     o[INDEX2(i,numComp,1)] = c1*f_11[i] + c0*f_10[i];
1836 caltinay 3724 } /* end of component loop i */
1837     } /* end of k1 loop */
1838     } /* end of face 1 */
1839     if (m_faceOffset[2] > -1) {
1840     #pragma omp for nowait
1841 caltinay 4334