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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 3756 - (show annotations)
Fri Jan 6 02:35:19 2012 UTC (7 years, 10 months ago) by caltinay
Original Path: branches/ripleygmg_from_3668/ripley/src/Rectangle.cpp
File size: 104546 byte(s)
Fixed interpolation from DOF to nodes and moved common code to the base class.

1
2 /*******************************************************
3 *
4 * Copyright (c) 2003-2011 by University of Queensland
5 * Earth Systems Science Computational Center (ESSCC)
6 * http://www.uq.edu.au/esscc
7 *
8 * Primary Business: Queensland, Australia
9 * Licensed under the Open Software License version 3.0
10 * http://www.opensource.org/licenses/osl-3.0.php
11 *
12 *******************************************************/
13
14 #include <ripley/Rectangle.h>
15 extern "C" {
16 #include "paso/SystemMatrix.h"
17 }
18
19 #if USE_SILO
20 #include <silo.h>
21 #ifdef ESYS_MPI
22 #include <pmpio.h>
23 #endif
24 #endif
25
26 #include <iomanip>
27
28 using namespace std;
29
30 namespace ripley {
31
32 Rectangle::Rectangle(int n0, int n1, double l0, double l1, int d0, int d1) :
33 RipleyDomain(2),
34 m_gNE0(n0),
35 m_gNE1(n1),
36 m_l0(l0),
37 m_l1(l1),
38 m_NX(d0),
39 m_NY(d1)
40 {
41 // ensure number of subdivisions is valid and nodes can be distributed
42 // among number of ranks
43 if (m_NX*m_NY != m_mpiInfo->size)
44 throw RipleyException("Invalid number of spatial subdivisions");
45
46 if ((n0+1)%m_NX > 0 || (n1+1)%m_NY > 0)
47 throw RipleyException("Number of elements+1 must be separable into number of ranks in each dimension");
48
49 if ((m_NX > 1 && (n0+1)/m_NX<2) || (m_NY > 1 && (n1+1)/m_NY<2))
50 throw RipleyException("Too few elements for the number of ranks");
51
52 // local number of elements (including overlap)
53 m_NE0 = (m_NX>1 ? (n0+1)/m_NX : n0);
54 if (m_mpiInfo->rank%m_NX>0 && m_mpiInfo->rank%m_NX<m_NX-1)
55 m_NE0++;
56 m_NE1 = (m_NY>1 ? (n1+1)/m_NY : n1);
57 if (m_mpiInfo->rank/m_NX>0 && m_mpiInfo->rank/m_NX<m_NY-1)
58 m_NE1++;
59
60 // local number of nodes
61 m_N0 = m_NE0+1;
62 m_N1 = m_NE1+1;
63
64 // bottom-left node is at (offset0,offset1) in global mesh
65 m_offset0 = (n0+1)/m_NX*(m_mpiInfo->rank%m_NX);
66 if (m_offset0 > 0)
67 m_offset0--;
68 m_offset1 = (n1+1)/m_NY*(m_mpiInfo->rank/m_NX);
69 if (m_offset1 > 0)
70 m_offset1--;
71
72 populateSampleIds();
73 createPattern();
74 }
75
76 Rectangle::~Rectangle()
77 {
78 }
79
80 string Rectangle::getDescription() const
81 {
82 return "ripley::Rectangle";
83 }
84
85 bool Rectangle::operator==(const AbstractDomain& other) const
86 {
87 const Rectangle* o=dynamic_cast<const Rectangle*>(&other);
88 if (o) {
89 return (RipleyDomain::operator==(other) &&
90 m_gNE0==o->m_gNE0 && m_gNE1==o->m_gNE1
91 && m_l0==o->m_l0 && m_l1==o->m_l1
92 && m_NX==o->m_NX && m_NY==o->m_NY);
93 }
94
95 return false;
96 }
97
98 void Rectangle::dump(const string& fileName) const
99 {
100 #if USE_SILO
101 string fn(fileName);
102 if (fileName.length() < 6 || fileName.compare(fileName.length()-5, 5, ".silo") != 0) {
103 fn+=".silo";
104 }
105
106 const int NUM_SILO_FILES = 1;
107 const char* blockDirFmt = "/block%04d";
108 int driver=DB_HDF5;
109 string siloPath;
110 DBfile* dbfile = NULL;
111
112 #ifdef ESYS_MPI
113 PMPIO_baton_t* baton = NULL;
114 #endif
115
116 if (m_mpiInfo->size > 1) {
117 #ifdef ESYS_MPI
118 baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
119 0x1337, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
120 PMPIO_DefaultClose, (void*)&driver);
121 // try the fallback driver in case of error
122 if (!baton && driver != DB_PDB) {
123 driver = DB_PDB;
124 baton = PMPIO_Init(NUM_SILO_FILES, PMPIO_WRITE, m_mpiInfo->comm,
125 0x1338, PMPIO_DefaultCreate, PMPIO_DefaultOpen,
126 PMPIO_DefaultClose, (void*)&driver);
127 }
128 if (baton) {
129 char str[64];
130 snprintf(str, 64, blockDirFmt, PMPIO_RankInGroup(baton, m_mpiInfo->rank));
131 siloPath = str;
132 dbfile = (DBfile*) PMPIO_WaitForBaton(baton, fn.c_str(), siloPath.c_str());
133 }
134 #endif
135 } else {
136 dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
137 getDescription().c_str(), driver);
138 // try the fallback driver in case of error
139 if (!dbfile && driver != DB_PDB) {
140 driver = DB_PDB;
141 dbfile = DBCreate(fn.c_str(), DB_CLOBBER, DB_LOCAL,
142 getDescription().c_str(), driver);
143 }
144 }
145
146 if (!dbfile)
147 throw RipleyException("dump: Could not create Silo file");
148
149 /*
150 if (driver==DB_HDF5) {
151 // gzip level 1 already provides good compression with minimal
152 // performance penalty. Some tests showed that gzip levels >3 performed
153 // rather badly on escript data both in terms of time and space
154 DBSetCompression("ERRMODE=FALLBACK METHOD=GZIP LEVEL=1");
155 }
156 */
157
158 boost::scoped_ptr<double> x(new double[m_N0]);
159 boost::scoped_ptr<double> y(new double[m_N1]);
160 double* coords[2] = { x.get(), y.get() };
161 pair<double,double> xdx = getFirstCoordAndSpacing(0);
162 pair<double,double> ydy = getFirstCoordAndSpacing(1);
163 #pragma omp parallel
164 {
165 #pragma omp for nowait
166 for (dim_t i0 = 0; i0 < m_N0; i0++) {
167 coords[0][i0]=xdx.first+i0*xdx.second;
168 }
169 #pragma omp for nowait
170 for (dim_t i1 = 0; i1 < m_N1; i1++) {
171 coords[1][i1]=ydy.first+i1*ydy.second;
172 }
173 }
174 IndexVector dims = getNumNodesPerDim();
175
176 // write mesh
177 DBPutQuadmesh(dbfile, "mesh", NULL, coords, &dims[0], 2, DB_DOUBLE,
178 DB_COLLINEAR, NULL);
179
180 // write node ids
181 DBPutQuadvar1(dbfile, "nodeId", "mesh", (void*)&m_nodeId[0], &dims[0], 2,
182 NULL, 0, DB_INT, DB_NODECENT, NULL);
183
184 // write element ids
185 dims = getNumElementsPerDim();
186 DBPutQuadvar1(dbfile, "elementId", "mesh", (void*)&m_elementId[0],
187 &dims[0], 2, NULL, 0, DB_INT, DB_ZONECENT, NULL);
188
189 // rank 0 writes multimesh and multivar
190 if (m_mpiInfo->rank == 0) {
191 vector<string> tempstrings;
192 vector<char*> names;
193 for (dim_t i=0; i<m_mpiInfo->size; i++) {
194 stringstream path;
195 path << "/block" << setw(4) << setfill('0') << right << i << "/mesh";
196 tempstrings.push_back(path.str());
197 names.push_back((char*)tempstrings.back().c_str());
198 }
199 vector<int> types(m_mpiInfo->size, DB_QUAD_RECT);
200 DBSetDir(dbfile, "/");
201 DBPutMultimesh(dbfile, "multimesh", m_mpiInfo->size, &names[0],
202 &types[0], NULL);
203 tempstrings.clear();
204 names.clear();
205 for (dim_t i=0; i<m_mpiInfo->size; i++) {
206 stringstream path;
207 path << "/block" << setw(4) << setfill('0') << right << i << "/nodeId";
208 tempstrings.push_back(path.str());
209 names.push_back((char*)tempstrings.back().c_str());
210 }
211 types.assign(m_mpiInfo->size, DB_QUADVAR);
212 DBPutMultivar(dbfile, "nodeId", m_mpiInfo->size, &names[0],
213 &types[0], NULL);
214 tempstrings.clear();
215 names.clear();
216 for (dim_t i=0; i<m_mpiInfo->size; i++) {
217 stringstream path;
218 path << "/block" << setw(4) << setfill('0') << right << i << "/elementId";
219 tempstrings.push_back(path.str());
220 names.push_back((char*)tempstrings.back().c_str());
221 }
222 DBPutMultivar(dbfile, "elementId", m_mpiInfo->size, &names[0],
223 &types[0], NULL);
224 }
225
226 if (m_mpiInfo->size > 1) {
227 #ifdef ESYS_MPI
228 PMPIO_HandOffBaton(baton, dbfile);
229 PMPIO_Finish(baton);
230 #endif
231 } else {
232 DBClose(dbfile);
233 }
234
235 #else // USE_SILO
236 throw RipleyException("dump(): no Silo support");
237 #endif
238 }
239
240 const int* Rectangle::borrowSampleReferenceIDs(int fsType) const
241 {
242 switch (fsType) {
243 case Nodes:
244 case ReducedNodes: //FIXME: reduced
245 return &m_nodeId[0];
246 case DegreesOfFreedom:
247 case ReducedDegreesOfFreedom: //FIXME: reduced
248 return &m_dofId[0];
249 case Elements:
250 case ReducedElements:
251 return &m_elementId[0];
252 case FaceElements:
253 case ReducedFaceElements:
254 return &m_faceId[0];
255 default:
256 break;
257 }
258
259 stringstream msg;
260 msg << "borrowSampleReferenceIDs() not implemented for function space type "
261 << functionSpaceTypeAsString(fsType);
262 throw RipleyException(msg.str());
263 }
264
265 bool Rectangle::ownSample(int fsCode, index_t id) const
266 {
267 #ifdef ESYS_MPI
268 if (fsCode == Nodes) {
269 return (m_dofMap[id] < getNumDOF());
270 } else {
271 stringstream msg;
272 msg << "ownSample() not implemented for "
273 << functionSpaceTypeAsString(fsCode);
274 throw RipleyException(msg.str());
275 }
276 #else
277 return true;
278 #endif
279 }
280
281 void Rectangle::setToGradient(escript::Data& out, const escript::Data& cIn) const
282 {
283 escript::Data& in = *const_cast<escript::Data*>(&cIn);
284 const dim_t numComp = in.getDataPointSize();
285 const double h0 = m_l0/m_gNE0;
286 const double h1 = m_l1/m_gNE1;
287 const double cx0 = -1./h0;
288 const double cx1 = -.78867513459481288225/h0;
289 const double cx2 = -.5/h0;
290 const double cx3 = -.21132486540518711775/h0;
291 const double cx4 = .21132486540518711775/h0;
292 const double cx5 = .5/h0;
293 const double cx6 = .78867513459481288225/h0;
294 const double cx7 = 1./h0;
295 const double cy0 = -1./h1;
296 const double cy1 = -.78867513459481288225/h1;
297 const double cy2 = -.5/h1;
298 const double cy3 = -.21132486540518711775/h1;
299 const double cy4 = .21132486540518711775/h1;
300 const double cy5 = .5/h1;
301 const double cy6 = .78867513459481288225/h1;
302 const double cy7 = 1./h1;
303
304 if (out.getFunctionSpace().getTypeCode() == Elements) {
305 /*** GENERATOR SNIP_GRAD_ELEMENTS TOP */
306 #pragma omp parallel for
307 for (index_t k1=0; k1 < m_NE1; ++k1) {
308 for (index_t k0=0; k0 < m_NE0; ++k0) {
309 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
310 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
311 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
312 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
313 double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
314 for (index_t i=0; i < numComp; ++i) {
315 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
316 o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
317 o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
318 o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
319 o[INDEX3(i,0,2,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
320 o[INDEX3(i,1,2,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
321 o[INDEX3(i,0,3,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
322 o[INDEX3(i,1,3,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
323 } /* end of component loop i */
324 } /* end of k0 loop */
325 } /* end of k1 loop */
326 /* GENERATOR SNIP_GRAD_ELEMENTS BOTTOM */
327 } else if (out.getFunctionSpace().getTypeCode() == ReducedElements) {
328 /*** GENERATOR SNIP_GRAD_REDUCED_ELEMENTS TOP */
329 #pragma omp parallel for
330 for (index_t k1=0; k1 < m_NE1; ++k1) {
331 for (index_t k0=0; k0 < m_NE0; ++k0) {
332 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
333 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
334 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
335 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
336 double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
337 for (index_t i=0; i < numComp; ++i) {
338 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
339 o[INDEX3(i,1,0,numComp,2)] = cy2*(f_00[i] + f_10[i]) + cy5*(f_01[i] + f_11[i]);
340 } /* end of component loop i */
341 } /* end of k0 loop */
342 } /* end of k1 loop */
343 /* GENERATOR SNIP_GRAD_REDUCED_ELEMENTS BOTTOM */
344 } else if (out.getFunctionSpace().getTypeCode() == FaceElements) {
345 #pragma omp parallel
346 {
347 /*** GENERATOR SNIP_GRAD_FACES TOP */
348 if (m_faceOffset[0] > -1) {
349 #pragma omp for nowait
350 for (index_t k1=0; k1 < m_NE1; ++k1) {
351 const register double* f_10 = in.getSampleDataRO(INDEX2(1,k1, m_N0));
352 const register double* f_11 = in.getSampleDataRO(INDEX2(1,k1+1, m_N0));
353 const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
354 const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
355 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
356 for (index_t i=0; i < numComp; ++i) {
357 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
358 o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
359 o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
360 o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
361 } /* end of component loop i */
362 } /* end of k1 loop */
363 } /* end of face 0 */
364 if (m_faceOffset[1] > -1) {
365 #pragma omp for nowait
366 for (index_t k1=0; k1 < m_NE1; ++k1) {
367 const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
368 const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
369 const register double* f_01 = in.getSampleDataRO(INDEX2(m_N0-2,k1+1, m_N0));
370 const register double* f_00 = in.getSampleDataRO(INDEX2(m_N0-2,k1, m_N0));
371 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
372 for (index_t i=0; i < numComp; ++i) {
373 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx1 + f_01[i]*cx3 + f_10[i]*cx6 + f_11[i]*cx4;
374 o[INDEX3(i,1,0,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
375 o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx3 + f_01[i]*cx1 + f_10[i]*cx4 + f_11[i]*cx6;
376 o[INDEX3(i,1,1,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
377 } /* end of component loop i */
378 } /* end of k1 loop */
379 } /* end of face 1 */
380 if (m_faceOffset[2] > -1) {
381 #pragma omp for nowait
382 for (index_t k0=0; k0 < m_NE0; ++k0) {
383 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
384 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
385 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,1, m_N0));
386 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,1, m_N0));
387 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
388 for (index_t i=0; i < numComp; ++i) {
389 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
390 o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
391 o[INDEX3(i,0,1,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
392 o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
393 } /* end of component loop i */
394 } /* end of k0 loop */
395 } /* end of face 2 */
396 if (m_faceOffset[3] > -1) {
397 #pragma omp for nowait
398 for (index_t k0=0; k0 < m_NE0; ++k0) {
399 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
400 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
401 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,m_N1-2, m_N0));
402 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,m_N1-2, m_N0));
403 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
404 for (index_t i=0; i < numComp; ++i) {
405 o[INDEX3(i,0,0,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
406 o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy1 + f_01[i]*cy6 + f_10[i]*cy3 + f_11[i]*cy4;
407 o[INDEX3(i,0,1,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
408 o[INDEX3(i,1,1,numComp,2)] = f_00[i]*cy3 + f_01[i]*cy4 + f_10[i]*cy1 + f_11[i]*cy6;
409 } /* end of component loop i */
410 } /* end of k0 loop */
411 } /* end of face 3 */
412 /* GENERATOR SNIP_GRAD_FACES BOTTOM */
413 } // end of parallel section
414 } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
415 #pragma omp parallel
416 {
417 /*** GENERATOR SNIP_GRAD_REDUCED_FACES TOP */
418 if (m_faceOffset[0] > -1) {
419 #pragma omp for nowait
420 for (index_t k1=0; k1 < m_NE1; ++k1) {
421 const register double* f_10 = in.getSampleDataRO(INDEX2(1,k1, m_N0));
422 const register double* f_11 = in.getSampleDataRO(INDEX2(1,k1+1, m_N0));
423 const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
424 const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
425 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
426 for (index_t i=0; i < numComp; ++i) {
427 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
428 o[INDEX3(i,1,0,numComp,2)] = f_00[i]*cy0 + f_01[i]*cy7;
429 } /* end of component loop i */
430 } /* end of k1 loop */
431 } /* end of face 0 */
432 if (m_faceOffset[1] > -1) {
433 #pragma omp for nowait
434 for (index_t k1=0; k1 < m_NE1; ++k1) {
435 const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
436 const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
437 const register double* f_01 = in.getSampleDataRO(INDEX2(m_N0-2,k1+1, m_N0));
438 const register double* f_00 = in.getSampleDataRO(INDEX2(m_N0-2,k1, m_N0));
439 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
440 for (index_t i=0; i < numComp; ++i) {
441 o[INDEX3(i,0,0,numComp,2)] = cx5*(f_10[i] + f_11[i]) + cx2*(f_00[i] + f_01[i]);
442 o[INDEX3(i,1,0,numComp,2)] = f_10[i]*cy0 + f_11[i]*cy7;
443 } /* end of component loop i */
444 } /* end of k1 loop */
445 } /* end of face 1 */
446 if (m_faceOffset[2] > -1) {
447 #pragma omp for nowait
448 for (index_t k0=0; k0 < m_NE0; ++k0) {
449 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
450 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
451 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,1, m_N0));
452 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,1, m_N0));
453 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
454 for (index_t i=0; i < numComp; ++i) {
455 o[INDEX3(i,0,0,numComp,2)] = f_00[i]*cx0 + f_10[i]*cx7;
456 o[INDEX3(i,1,0,numComp,2)] = cy2*(f_00[i] + f_10[i]) + cy5*(f_01[i] + f_11[i]);
457 } /* end of component loop i */
458 } /* end of k0 loop */
459 } /* end of face 2 */
460 if (m_faceOffset[3] > -1) {
461 #pragma omp for nowait
462 for (index_t k0=0; k0 < m_NE0; ++k0) {
463 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
464 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
465 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,m_N1-2, m_N0));
466 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,m_N1-2, m_N0));
467 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
468 for (index_t i=0; i < numComp; ++i) {
469 o[INDEX3(i,0,0,numComp,2)] = f_01[i]*cx0 + f_11[i]*cx7;
470 o[INDEX3(i,1,0,numComp,2)] = cy5*(f_01[i] + f_11[i]) + cy2*(f_00[i] + f_10[i]);
471 } /* end of component loop i */
472 } /* end of k0 loop */
473 } /* end of face 3 */
474 /* GENERATOR SNIP_GRAD_REDUCED_FACES BOTTOM */
475 } // end of parallel section
476 } else {
477 stringstream msg;
478 msg << "setToGradient() not implemented for "
479 << functionSpaceTypeAsString(out.getFunctionSpace().getTypeCode());
480 throw RipleyException(msg.str());
481 }
482 }
483
484 void Rectangle::setToIntegrals(vector<double>& integrals, const escript::Data& arg) const
485 {
486 escript::Data& in = *const_cast<escript::Data*>(&arg);
487 const dim_t numComp = in.getDataPointSize();
488 const double h0 = m_l0/m_gNE0;
489 const double h1 = m_l1/m_gNE1;
490 if (arg.getFunctionSpace().getTypeCode() == Elements) {
491 const double w_0 = h0*h1/4.;
492 #pragma omp parallel
493 {
494 vector<double> int_local(numComp, 0);
495 #pragma omp for nowait
496 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
497 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
498 const double* f = in.getSampleDataRO(INDEX2(k0, k1, m_NE0));
499 for (index_t i=0; i < numComp; ++i) {
500 const register double f_0 = f[INDEX2(i,0,numComp)];
501 const register double f_1 = f[INDEX2(i,1,numComp)];
502 const register double f_2 = f[INDEX2(i,2,numComp)];
503 const register double f_3 = f[INDEX2(i,3,numComp)];
504 int_local[i]+=(f_0+f_1+f_2+f_3)*w_0;
505 } /* end of component loop i */
506 } /* end of k0 loop */
507 } /* end of k1 loop */
508
509 #pragma omp critical
510 for (index_t i=0; i<numComp; i++)
511 integrals[i]+=int_local[i];
512 } // end of parallel section
513 } else if (arg.getFunctionSpace().getTypeCode() == ReducedElements) {
514 const double w_0 = h0*h1;
515 #pragma omp parallel
516 {
517 vector<double> int_local(numComp, 0);
518 #pragma omp for nowait
519 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
520 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
521 const double* f = in.getSampleDataRO(INDEX2(k0, k1, m_NE0));
522 for (index_t i=0; i < numComp; ++i) {
523 int_local[i]+=f[i]*w_0;
524 } /* end of component loop i */
525 } /* end of k0 loop */
526 } /* end of k1 loop */
527
528 #pragma omp critical
529 for (index_t i=0; i<numComp; i++)
530 integrals[i]+=int_local[i];
531 } // end of parallel section
532 } else if (arg.getFunctionSpace().getTypeCode() == FaceElements) {
533 const double w_0 = h0/2.;
534 const double w_1 = h1/2.;
535 #pragma omp parallel
536 {
537 vector<double> int_local(numComp, 0);
538 if (m_faceOffset[0] > -1) {
539 #pragma omp for nowait
540 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
541 const double* f = in.getSampleDataRO(m_faceOffset[0]+k1);
542 for (index_t i=0; i < numComp; ++i) {
543 const register double f_0 = f[INDEX2(i,0,numComp)];
544 const register double f_1 = f[INDEX2(i,1,numComp)];
545 int_local[i]+=(f_0+f_1)*w_1;
546 } /* end of component loop i */
547 } /* end of k1 loop */
548 }
549
550 if (m_faceOffset[1] > -1) {
551 #pragma omp for nowait
552 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
553 const double* f = in.getSampleDataRO(m_faceOffset[1]+k1);
554 for (index_t i=0; i < numComp; ++i) {
555 const register double f_0 = f[INDEX2(i,0,numComp)];
556 const register double f_1 = f[INDEX2(i,1,numComp)];
557 int_local[i]+=(f_0+f_1)*w_1;
558 } /* end of component loop i */
559 } /* end of k1 loop */
560 }
561
562 if (m_faceOffset[2] > -1) {
563 #pragma omp for nowait
564 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
565 const double* f = in.getSampleDataRO(m_faceOffset[2]+k0);
566 for (index_t i=0; i < numComp; ++i) {
567 const register double f_0 = f[INDEX2(i,0,numComp)];
568 const register double f_1 = f[INDEX2(i,1,numComp)];
569 int_local[i]+=(f_0+f_1)*w_0;
570 } /* end of component loop i */
571 } /* end of k0 loop */
572 }
573
574 if (m_faceOffset[3] > -1) {
575 #pragma omp for nowait
576 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
577 const double* f = in.getSampleDataRO(m_faceOffset[3]+k0);
578 for (index_t i=0; i < numComp; ++i) {
579 const register double f_0 = f[INDEX2(i,0,numComp)];
580 const register double f_1 = f[INDEX2(i,1,numComp)];
581 int_local[i]+=(f_0+f_1)*w_0;
582 } /* end of component loop i */
583 } /* end of k0 loop */
584 }
585
586 #pragma omp critical
587 for (index_t i=0; i<numComp; i++)
588 integrals[i]+=int_local[i];
589 } // end of parallel section
590 } else if (arg.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
591 #pragma omp parallel
592 {
593 vector<double> int_local(numComp, 0);
594 if (m_faceOffset[0] > -1) {
595 #pragma omp for nowait
596 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
597 const double* f = in.getSampleDataRO(m_faceOffset[0]+k1);
598 for (index_t i=0; i < numComp; ++i) {
599 int_local[i]+=f[i]*h1;
600 } /* end of component loop i */
601 } /* end of k1 loop */
602 }
603
604 if (m_faceOffset[1] > -1) {
605 #pragma omp for nowait
606 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
607 const double* f = in.getSampleDataRO(m_faceOffset[1]+k1);
608 for (index_t i=0; i < numComp; ++i) {
609 int_local[i]+=f[i]*h1;
610 } /* end of component loop i */
611 } /* end of k1 loop */
612 }
613
614 if (m_faceOffset[2] > -1) {
615 #pragma omp for nowait
616 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
617 const double* f = in.getSampleDataRO(m_faceOffset[2]+k0);
618 for (index_t i=0; i < numComp; ++i) {
619 int_local[i]+=f[i]*h0;
620 } /* end of component loop i */
621 } /* end of k0 loop */
622 }
623
624 if (m_faceOffset[3] > -1) {
625 #pragma omp for nowait
626 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
627 const double* f = in.getSampleDataRO(m_faceOffset[3]+k0);
628 for (index_t i=0; i < numComp; ++i) {
629 int_local[i]+=f[i]*h0;
630 } /* end of component loop i */
631 } /* end of k0 loop */
632 }
633
634 #pragma omp critical
635 for (index_t i=0; i<numComp; i++)
636 integrals[i]+=int_local[i];
637 } // end of parallel section
638 } else {
639 stringstream msg;
640 msg << "setToIntegrals() not implemented for "
641 << functionSpaceTypeAsString(arg.getFunctionSpace().getTypeCode());
642 throw RipleyException(msg.str());
643 }
644 }
645
646 void Rectangle::setToNormal(escript::Data& out) const
647 {
648 if (out.getFunctionSpace().getTypeCode() == FaceElements) {
649 #pragma omp parallel
650 {
651 if (m_faceOffset[0] > -1) {
652 #pragma omp for nowait
653 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
654 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
655 // set vector at two quadrature points
656 *o++ = -1.;
657 *o++ = 0.;
658 *o++ = -1.;
659 *o = 0.;
660 }
661 }
662
663 if (m_faceOffset[1] > -1) {
664 #pragma omp for nowait
665 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
666 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
667 // set vector at two quadrature points
668 *o++ = 1.;
669 *o++ = 0.;
670 *o++ = 1.;
671 *o = 0.;
672 }
673 }
674
675 if (m_faceOffset[2] > -1) {
676 #pragma omp for nowait
677 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
678 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
679 // set vector at two quadrature points
680 *o++ = 0.;
681 *o++ = -1.;
682 *o++ = 0.;
683 *o = -1.;
684 }
685 }
686
687 if (m_faceOffset[3] > -1) {
688 #pragma omp for nowait
689 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
690 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
691 // set vector at two quadrature points
692 *o++ = 0.;
693 *o++ = 1.;
694 *o++ = 0.;
695 *o = 1.;
696 }
697 }
698 } // end of parallel section
699 } else if (out.getFunctionSpace().getTypeCode() == ReducedFaceElements) {
700 #pragma omp parallel
701 {
702 if (m_faceOffset[0] > -1) {
703 #pragma omp for nowait
704 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
705 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
706 *o++ = -1.;
707 *o = 0.;
708 }
709 }
710
711 if (m_faceOffset[1] > -1) {
712 #pragma omp for nowait
713 for (index_t k1 = 0; k1 < m_NE1; ++k1) {
714 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
715 *o++ = 1.;
716 *o = 0.;
717 }
718 }
719
720 if (m_faceOffset[2] > -1) {
721 #pragma omp for nowait
722 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
723 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
724 *o++ = 0.;
725 *o = -1.;
726 }
727 }
728
729 if (m_faceOffset[3] > -1) {
730 #pragma omp for nowait
731 for (index_t k0 = 0; k0 < m_NE0; ++k0) {
732 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
733 *o++ = 0.;
734 *o = 1.;
735 }
736 }
737 } // end of parallel section
738
739 } else {
740 stringstream msg;
741 msg << "setToNormal() not implemented for "
742 << functionSpaceTypeAsString(out.getFunctionSpace().getTypeCode());
743 throw RipleyException(msg.str());
744 }
745 }
746
747 Paso_SystemMatrixPattern* Rectangle::getPattern(bool reducedRowOrder,
748 bool reducedColOrder) const
749 {
750 if (reducedRowOrder || reducedColOrder)
751 throw RipleyException("getPattern() not implemented for reduced order");
752
753 return m_pattern;
754 }
755
756 void Rectangle::Print_Mesh_Info(const bool full) const
757 {
758 RipleyDomain::Print_Mesh_Info(full);
759 if (full) {
760 cout << " Id Coordinates" << endl;
761 cout.precision(15);
762 cout.setf(ios::scientific, ios::floatfield);
763 pair<double,double> xdx = getFirstCoordAndSpacing(0);
764 pair<double,double> ydy = getFirstCoordAndSpacing(1);
765 for (index_t i=0; i < getNumNodes(); i++) {
766 cout << " " << setw(5) << m_nodeId[i]
767 << " " << xdx.first+(i%m_N0)*xdx.second
768 << " " << ydy.first+(i/m_N0)*ydy.second << endl;
769 }
770 }
771 }
772
773 IndexVector Rectangle::getNumNodesPerDim() const
774 {
775 IndexVector ret;
776 ret.push_back(m_N0);
777 ret.push_back(m_N1);
778 return ret;
779 }
780
781 IndexVector Rectangle::getNumElementsPerDim() const
782 {
783 IndexVector ret;
784 ret.push_back(m_NE0);
785 ret.push_back(m_NE1);
786 return ret;
787 }
788
789 IndexVector Rectangle::getNumFacesPerBoundary() const
790 {
791 IndexVector ret(4, 0);
792 //left
793 if (m_offset0==0)
794 ret[0]=m_NE1;
795 //right
796 if (m_mpiInfo->rank%m_NX==m_NX-1)
797 ret[1]=m_NE1;
798 //bottom
799 if (m_offset1==0)
800 ret[2]=m_NE0;
801 //top
802 if (m_mpiInfo->rank/m_NX==m_NY-1)
803 ret[3]=m_NE0;
804 return ret;
805 }
806
807 pair<double,double> Rectangle::getFirstCoordAndSpacing(dim_t dim) const
808 {
809 if (dim==0) {
810 return pair<double,double>((m_l0*m_offset0)/m_gNE0, m_l0/m_gNE0);
811 } else if (dim==1) {
812 return pair<double,double>((m_l1*m_offset1)/m_gNE1, m_l1/m_gNE1);
813 }
814 throw RipleyException("getFirstCoordAndSpacing(): invalid argument");
815 }
816
817 //protected
818 dim_t Rectangle::getNumDOF() const
819 {
820 return (m_gNE0+1)/m_NX*(m_gNE1+1)/m_NY;
821 }
822
823 //protected
824 dim_t Rectangle::getNumFaceElements() const
825 {
826 const IndexVector faces = getNumFacesPerBoundary();
827 dim_t n=0;
828 for (size_t i=0; i<faces.size(); i++)
829 n+=faces[i];
830 return n;
831 }
832
833 //protected
834 void Rectangle::assembleCoordinates(escript::Data& arg) const
835 {
836 escriptDataC x = arg.getDataC();
837 int numDim = m_numDim;
838 if (!isDataPointShapeEqual(&x, 1, &numDim))
839 throw RipleyException("setToX: Invalid Data object shape");
840 if (!numSamplesEqual(&x, 1, getNumNodes()))
841 throw RipleyException("setToX: Illegal number of samples in Data object");
842
843 pair<double,double> xdx = getFirstCoordAndSpacing(0);
844 pair<double,double> ydy = getFirstCoordAndSpacing(1);
845 arg.requireWrite();
846 #pragma omp parallel for
847 for (dim_t i1 = 0; i1 < m_N1; i1++) {
848 for (dim_t i0 = 0; i0 < m_N0; i0++) {
849 double* point = arg.getSampleDataRW(i0+m_N0*i1);
850 point[0] = xdx.first+i0*xdx.second;
851 point[1] = ydy.first+i1*ydy.second;
852 }
853 }
854 }
855
856 //protected
857 dim_t Rectangle::insertNeighbourNodes(IndexVector& index, index_t node) const
858 {
859 const dim_t nDOF0 = (m_gNE0+1)/m_NX;
860 const dim_t nDOF1 = (m_gNE1+1)/m_NY;
861 const int x=node%nDOF0;
862 const int y=node/nDOF0;
863 dim_t num=0;
864 // loop through potential neighbours and add to index if positions are
865 // within bounds
866 for (int i1=-1; i1<2; i1++) {
867 for (int i0=-1; i0<2; i0++) {
868 // skip node itself
869 if (i0==0 && i1==0)
870 continue;
871 // location of neighbour node
872 const int nx=x+i0;
873 const int ny=y+i1;
874 if (nx>=0 && ny>=0 && nx<nDOF0 && ny<nDOF1) {
875 index.push_back(ny*nDOF0+nx);
876 num++;
877 }
878 }
879 }
880
881 return num;
882 }
883
884 //protected
885 void Rectangle::nodesToDOF(escript::Data& out, escript::Data& in) const
886 {
887 const dim_t numComp = in.getDataPointSize();
888 out.requireWrite();
889
890 const index_t left = (m_offset0==0 ? 0 : 1);
891 const index_t bottom = (m_offset1==0 ? 0 : 1);
892 const dim_t nDOF0 = (m_gNE0+1)/m_NX;
893 const dim_t nDOF1 = (m_gNE1+1)/m_NY;
894 #pragma omp parallel for
895 for (index_t i=0; i<nDOF1; i++) {
896 for (index_t j=0; j<nDOF0; j++) {
897 const index_t n=j+left+(i+bottom)*m_N0;
898 const double* src=in.getSampleDataRO(n);
899 copy(src, src+numComp, out.getSampleDataRW(j+i*nDOF0));
900 }
901 }
902 }
903
904 //protected
905 void Rectangle::dofToNodes(escript::Data& out, escript::Data& in) const
906 {
907 const dim_t numComp = in.getDataPointSize();
908 Paso_Coupler* coupler = Paso_Coupler_alloc(m_connector, numComp);
909 in.requireWrite();
910 Paso_Coupler_startCollect(coupler, in.getSampleDataRW(0));
911
912 const dim_t numDOF = getNumDOF();
913 out.requireWrite();
914 const double* buffer = Paso_Coupler_finishCollect(coupler);
915
916 #pragma omp parallel for
917 for (index_t i=0; i<getNumNodes(); i++) {
918 const double* src=(m_dofMap[i]<numDOF ?
919 in.getSampleDataRO(m_dofMap[i])
920 : &buffer[(m_dofMap[i]-numDOF)*numComp]);
921 copy(src, src+numComp, out.getSampleDataRW(i));
922 }
923 }
924
925 //private
926 void Rectangle::populateSampleIds()
927 {
928 // identifiers are ordered from left to right, bottom to top globablly.
929
930 // build node distribution vector first.
931 // rank i owns m_nodeDistribution[i+1]-nodeDistribution[i] nodes
932 m_nodeDistribution.assign(m_mpiInfo->size+1, 0);
933 const dim_t numDOF=getNumDOF();
934 for (dim_t k=1; k<m_mpiInfo->size; k++) {
935 m_nodeDistribution[k]=k*numDOF;
936 }
937 m_nodeDistribution[m_mpiInfo->size]=getNumDataPointsGlobal();
938 m_nodeId.resize(getNumNodes());
939 m_dofId.resize(numDOF);
940 m_elementId.resize(getNumElements());
941 m_faceId.resize(getNumFaceElements());
942
943 #pragma omp parallel
944 {
945 // nodes
946 #pragma omp for nowait
947 for (dim_t i1=0; i1<m_N1; i1++) {
948 for (dim_t i0=0; i0<m_N0; i0++) {
949 m_nodeId[i0+i1*m_N0] = (m_offset1+i1)*(m_gNE0+1)+m_offset0+i0;
950 }
951 }
952
953 // degrees of freedom
954 #pragma omp for nowait
955 for (dim_t k=0; k<numDOF; k++)
956 m_dofId[k] = m_nodeDistribution[m_mpiInfo->rank]+k;
957
958 // elements
959 #pragma omp for nowait
960 for (dim_t i1=0; i1<m_NE1; i1++) {
961 for (dim_t i0=0; i0<m_NE0; i0++) {
962 m_elementId[i0+i1*m_NE0]=(m_offset1+i1)*m_gNE0+m_offset0+i0;
963 }
964 }
965
966 // face elements
967 #pragma omp for
968 for (dim_t k=0; k<getNumFaceElements(); k++)
969 m_faceId[k]=k;
970 } // end parallel section
971
972 m_nodeTags.assign(getNumNodes(), 0);
973 updateTagsInUse(Nodes);
974
975 m_elementTags.assign(getNumElements(), 0);
976 updateTagsInUse(Elements);
977
978 // generate face offset vector and set face tags
979 const IndexVector facesPerEdge = getNumFacesPerBoundary();
980 const index_t LEFT=1, RIGHT=2, BOTTOM=10, TOP=20;
981 const index_t faceTag[] = { LEFT, RIGHT, BOTTOM, TOP };
982 m_faceOffset.assign(facesPerEdge.size(), -1);
983 m_faceTags.clear();
984 index_t offset=0;
985 for (size_t i=0; i<facesPerEdge.size(); i++) {
986 if (facesPerEdge[i]>0) {
987 m_faceOffset[i]=offset;
988 offset+=facesPerEdge[i];
989 m_faceTags.insert(m_faceTags.end(), facesPerEdge[i], faceTag[i]);
990 }
991 }
992 setTagMap("left", LEFT);
993 setTagMap("right", RIGHT);
994 setTagMap("bottom", BOTTOM);
995 setTagMap("top", TOP);
996 updateTagsInUse(FaceElements);
997 }
998
999 //private
1000 void Rectangle::createPattern()
1001 {
1002 const dim_t nDOF0 = (m_gNE0+1)/m_NX;
1003 const dim_t nDOF1 = (m_gNE1+1)/m_NY;
1004 const index_t left = (m_offset0==0 ? 0 : 1);
1005 const index_t bottom = (m_offset1==0 ? 0 : 1);
1006
1007 // populate node->DOF mapping with own degrees of freedom.
1008 // The rest is assigned in the loop further down
1009 m_dofMap.assign(getNumNodes(), 0);
1010 #pragma omp parallel for
1011 for (index_t i=bottom; i<m_N1; i++) {
1012 for (index_t j=left; j<m_N0; j++) {
1013 m_dofMap[i*m_N0+j]=(i-bottom)*nDOF0+j-left;
1014 }
1015 }
1016
1017 // build list of shared components and neighbours by looping through
1018 // all potential neighbouring ranks and checking if positions are
1019 // within bounds
1020 const dim_t numDOF=nDOF0*nDOF1;
1021 vector<IndexVector> colIndices(numDOF); // for the couple blocks
1022 RankVector neighbour;
1023 IndexVector offsetInShared(1,0);
1024 IndexVector sendShared, recvShared;
1025 int numShared=0;
1026 const int x=m_mpiInfo->rank%m_NX;
1027 const int y=m_mpiInfo->rank/m_NX;
1028 for (int i1=-1; i1<2; i1++) {
1029 for (int i0=-1; i0<2; i0++) {
1030 // skip this rank
1031 if (i0==0 && i1==0)
1032 continue;
1033 // location of neighbour rank
1034 const int nx=x+i0;
1035 const int ny=y+i1;
1036 if (nx>=0 && ny>=0 && nx<m_NX && ny<m_NY) {
1037 neighbour.push_back(ny*m_NX+nx);
1038 if (i0==0) {
1039 // sharing top or bottom edge
1040 const int firstDOF=(i1==-1 ? 0 : numDOF-nDOF0);
1041 const int firstNode=(i1==-1 ? left : m_N0*(m_N1-1)+left);
1042 offsetInShared.push_back(offsetInShared.back()+nDOF0);
1043 for (dim_t i=0; i<nDOF0; i++, numShared++) {
1044 sendShared.push_back(firstDOF+i);
1045 recvShared.push_back(numDOF+numShared);
1046 if (i>0)
1047 colIndices[firstDOF+i-1].push_back(numShared);
1048 colIndices[firstDOF+i].push_back(numShared);
1049 if (i<nDOF0-1)
1050 colIndices[firstDOF+i+1].push_back(numShared);
1051 m_dofMap[firstNode+i]=numDOF+numShared;
1052 }
1053 } else if (i1==0) {
1054 // sharing left or right edge
1055 const int firstDOF=(i0==-1 ? 0 : nDOF0-1);
1056 const int firstNode=(i0==-1 ? bottom*m_N0 : (bottom+1)*m_N0-1);
1057 offsetInShared.push_back(offsetInShared.back()+nDOF1);
1058 for (dim_t i=0; i<nDOF1; i++, numShared++) {
1059 sendShared.push_back(firstDOF+i*nDOF0);
1060 recvShared.push_back(numDOF+numShared);
1061 if (i>0)
1062 colIndices[firstDOF+(i-1)*nDOF0].push_back(numShared);
1063 colIndices[firstDOF+i*nDOF0].push_back(numShared);
1064 if (i<nDOF1-1)
1065 colIndices[firstDOF+(i+1)*nDOF0].push_back(numShared);
1066 m_dofMap[firstNode+i*m_N0]=numDOF+numShared;
1067 }
1068 } else {
1069 // sharing a node
1070 const int dof=(i0+1)/2*(nDOF0-1)+(i1+1)/2*(numDOF-nDOF0);
1071 const int node=(i0+1)/2*(m_N0-1)+(i1+1)/2*m_N0*(m_N1-1);
1072 offsetInShared.push_back(offsetInShared.back()+1);
1073 sendShared.push_back(dof);
1074 recvShared.push_back(numDOF+numShared);
1075 colIndices[dof].push_back(numShared);
1076 m_dofMap[node]=numDOF+numShared;
1077 ++numShared;
1078 }
1079 }
1080 }
1081 }
1082
1083 // create connector
1084 Paso_SharedComponents *snd_shcomp = Paso_SharedComponents_alloc(
1085 numDOF, neighbour.size(), &neighbour[0], &sendShared[0],
1086 &offsetInShared[0], 1, 0, m_mpiInfo);
1087 Paso_SharedComponents *rcv_shcomp = Paso_SharedComponents_alloc(
1088 numDOF, neighbour.size(), &neighbour[0], &recvShared[0],
1089 &offsetInShared[0], 1, 0, m_mpiInfo);
1090 m_connector = Paso_Connector_alloc(snd_shcomp, rcv_shcomp);
1091 Paso_SharedComponents_free(snd_shcomp);
1092 Paso_SharedComponents_free(rcv_shcomp);
1093
1094 // create main and couple blocks
1095 Paso_Pattern *mainPattern = createMainPattern();
1096 Paso_Pattern *colPattern, *rowPattern;
1097 createCouplePatterns(colIndices, numShared, &colPattern, &rowPattern);
1098
1099 // allocate paso distribution
1100 Paso_Distribution* distribution = Paso_Distribution_alloc(m_mpiInfo,
1101 const_cast<index_t*>(&m_nodeDistribution[0]), 1, 0);
1102
1103 // finally create the system matrix
1104 m_pattern = Paso_SystemMatrixPattern_alloc(MATRIX_FORMAT_DEFAULT,
1105 distribution, distribution, mainPattern, colPattern, rowPattern,
1106 m_connector, m_connector);
1107
1108 Paso_Distribution_free(distribution);
1109
1110 // useful debug output
1111 /*
1112 cout << "--- rcv_shcomp ---" << endl;
1113 cout << "numDOF=" << numDOF << ", numNeighbors=" << neighbour.size() << endl;
1114 for (size_t i=0; i<neighbour.size(); i++) {
1115 cout << "neighbor[" << i << "]=" << neighbour[i]
1116 << " offsetInShared[" << i+1 << "]=" << offsetInShared[i+1] << endl;
1117 }
1118 for (size_t i=0; i<recvShared.size(); i++) {
1119 cout << "shared[" << i << "]=" << recvShared[i] << endl;
1120 }
1121 cout << "--- snd_shcomp ---" << endl;
1122 for (size_t i=0; i<sendShared.size(); i++) {
1123 cout << "shared[" << i << "]=" << sendShared[i] << endl;
1124 }
1125 cout << "--- dofMap ---" << endl;
1126 for (size_t i=0; i<m_dofMap.size(); i++) {
1127 cout << "m_dofMap[" << i << "]=" << m_dofMap[i] << endl;
1128 }
1129 cout << "--- colIndices ---" << endl;
1130 for (size_t i=0; i<colIndices.size(); i++) {
1131 cout << "colIndices[" << i << "].size()=" << colIndices[i].size() << endl;
1132 }
1133 */
1134
1135 /*
1136 cout << "--- main_pattern ---" << endl;
1137 cout << "M=" << mainPattern->numOutput << ", N=" << mainPattern->numInput << endl;
1138 for (size_t i=0; i<mainPattern->numOutput+1; i++) {
1139 cout << "ptr[" << i << "]=" << mainPattern->ptr[i] << endl;
1140 }
1141 for (size_t i=0; i<mainPattern->ptr[mainPattern->numOutput]; i++) {
1142 cout << "index[" << i << "]=" << mainPattern->index[i] << endl;
1143 }
1144 */
1145
1146 /*
1147 cout << "--- colCouple_pattern ---" << endl;
1148 cout << "M=" << colPattern->numOutput << ", N=" << colPattern->numInput << endl;
1149 for (size_t i=0; i<colPattern->numOutput+1; i++) {
1150 cout << "ptr[" << i << "]=" << colPattern->ptr[i] << endl;
1151 }
1152 for (size_t i=0; i<colPattern->ptr[colPattern->numOutput]; i++) {
1153 cout << "index[" << i << "]=" << colPattern->index[i] << endl;
1154 }
1155 */
1156
1157 /*
1158 cout << "--- rowCouple_pattern ---" << endl;
1159 cout << "M=" << rowPattern->numOutput << ", N=" << rowPattern->numInput << endl;
1160 for (size_t i=0; i<rowPattern->numOutput+1; i++) {
1161 cout << "ptr[" << i << "]=" << rowPattern->ptr[i] << endl;
1162 }
1163 for (size_t i=0; i<rowPattern->ptr[rowPattern->numOutput]; i++) {
1164 cout << "index[" << i << "]=" << rowPattern->index[i] << endl;
1165 }
1166 */
1167
1168 Paso_Pattern_free(mainPattern);
1169 Paso_Pattern_free(colPattern);
1170 Paso_Pattern_free(rowPattern);
1171 }
1172
1173 //protected
1174 void Rectangle::interpolateNodesOnElements(escript::Data& out,
1175 escript::Data& in, bool reduced) const
1176 {
1177 const dim_t numComp = in.getDataPointSize();
1178 if (reduced) {
1179 /*** GENERATOR SNIP_INTERPOLATE_REDUCED_ELEMENTS TOP */
1180 const double c0 = .25;
1181 #pragma omp parallel for
1182 for (index_t k1=0; k1 < m_NE1; ++k1) {
1183 for (index_t k0=0; k0 < m_NE0; ++k0) {
1184 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
1185 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
1186 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
1187 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
1188 double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
1189 for (index_t i=0; i < numComp; ++i) {
1190 o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i] + f_10[i] + f_11[i]);
1191 } /* end of component loop i */
1192 } /* end of k0 loop */
1193 } /* end of k1 loop */
1194 /* GENERATOR SNIP_INTERPOLATE_REDUCED_ELEMENTS BOTTOM */
1195 } else {
1196 /*** GENERATOR SNIP_INTERPOLATE_ELEMENTS TOP */
1197 const double c0 = .16666666666666666667;
1198 const double c1 = .044658198738520451079;
1199 const double c2 = .62200846792814621559;
1200 #pragma omp parallel for
1201 for (index_t k1=0; k1 < m_NE1; ++k1) {
1202 for (index_t k0=0; k0 < m_NE0; ++k0) {
1203 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,k1, m_N0));
1204 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,k1+1, m_N0));
1205 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,k1+1, m_N0));
1206 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,k1, m_N0));
1207 double* o = out.getSampleDataRW(INDEX2(k0,k1,m_NE0));
1208 for (index_t i=0; i < numComp; ++i) {
1209 o[INDEX2(i,numComp,0)] = f_00[i]*c2 + f_11[i]*c1 + c0*(f_01[i] + f_10[i]);
1210 o[INDEX2(i,numComp,1)] = f_01[i]*c1 + f_10[i]*c2 + c0*(f_00[i] + f_11[i]);
1211 o[INDEX2(i,numComp,2)] = f_01[i]*c2 + f_10[i]*c1 + c0*(f_00[i] + f_11[i]);
1212 o[INDEX2(i,numComp,3)] = f_00[i]*c1 + f_11[i]*c2 + c0*(f_01[i] + f_10[i]);
1213 } /* end of component loop i */
1214 } /* end of k0 loop */
1215 } /* end of k1 loop */
1216 /* GENERATOR SNIP_INTERPOLATE_ELEMENTS BOTTOM */
1217 }
1218 }
1219
1220 //protected
1221 void Rectangle::interpolateNodesOnFaces(escript::Data& out, escript::Data& in,
1222 bool reduced) const
1223 {
1224 const dim_t numComp = in.getDataPointSize();
1225 if (reduced) {
1226 const double c0 = .5;
1227 #pragma omp parallel
1228 {
1229 /*** GENERATOR SNIP_INTERPOLATE_REDUCED_FACES TOP */
1230 if (m_faceOffset[0] > -1) {
1231 #pragma omp for nowait
1232 for (index_t k1=0; k1 < m_NE1; ++k1) {
1233 const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
1234 const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
1235 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1236 for (index_t i=0; i < numComp; ++i) {
1237 o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_01[i]);
1238 } /* end of component loop i */
1239 } /* end of k1 loop */
1240 } /* end of face 0 */
1241 if (m_faceOffset[1] > -1) {
1242 #pragma omp for nowait
1243 for (index_t k1=0; k1 < m_NE1; ++k1) {
1244 const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
1245 const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
1246 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1247 for (index_t i=0; i < numComp; ++i) {
1248 o[INDEX2(i,numComp,0)] = c0*(f_10[i] + f_11[i]);
1249 } /* end of component loop i */
1250 } /* end of k1 loop */
1251 } /* end of face 1 */
1252 if (m_faceOffset[2] > -1) {
1253 #pragma omp for nowait
1254 for (index_t k0=0; k0 < m_NE0; ++k0) {
1255 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
1256 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
1257 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1258 for (index_t i=0; i < numComp; ++i) {
1259 o[INDEX2(i,numComp,0)] = c0*(f_00[i] + f_10[i]);
1260 } /* end of component loop i */
1261 } /* end of k0 loop */
1262 } /* end of face 2 */
1263 if (m_faceOffset[3] > -1) {
1264 #pragma omp for nowait
1265 for (index_t k0=0; k0 < m_NE0; ++k0) {
1266 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
1267 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
1268 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1269 for (index_t i=0; i < numComp; ++i) {
1270 o[INDEX2(i,numComp,0)] = c0*(f_01[i] + f_11[i]);
1271 } /* end of component loop i */
1272 } /* end of k0 loop */
1273 } /* end of face 3 */
1274 /* GENERATOR SNIP_INTERPOLATE_REDUCED_FACES BOTTOM */
1275 } // end of parallel section
1276 } else {
1277 const double c0 = 0.21132486540518711775;
1278 const double c1 = 0.78867513459481288225;
1279 #pragma omp parallel
1280 {
1281 /*** GENERATOR SNIP_INTERPOLATE_FACES TOP */
1282 if (m_faceOffset[0] > -1) {
1283 #pragma omp for nowait
1284 for (index_t k1=0; k1 < m_NE1; ++k1) {
1285 const register double* f_01 = in.getSampleDataRO(INDEX2(0,k1+1, m_N0));
1286 const register double* f_00 = in.getSampleDataRO(INDEX2(0,k1, m_N0));
1287 double* o = out.getSampleDataRW(m_faceOffset[0]+k1);
1288 for (index_t i=0; i < numComp; ++i) {
1289 o[INDEX2(i,numComp,0)] = f_00[i]*c1 + f_01[i]*c0;
1290 o[INDEX2(i,numComp,1)] = f_00[i]*c0 + f_01[i]*c1;
1291 } /* end of component loop i */
1292 } /* end of k1 loop */
1293 } /* end of face 0 */
1294 if (m_faceOffset[1] > -1) {
1295 #pragma omp for nowait
1296 for (index_t k1=0; k1 < m_NE1; ++k1) {
1297 const register double* f_10 = in.getSampleDataRO(INDEX2(m_N0-1,k1, m_N0));
1298 const register double* f_11 = in.getSampleDataRO(INDEX2(m_N0-1,k1+1, m_N0));
1299 double* o = out.getSampleDataRW(m_faceOffset[1]+k1);
1300 for (index_t i=0; i < numComp; ++i) {
1301 o[INDEX2(i,numComp,0)] = f_10[i]*c1 + f_11[i]*c0;
1302 o[INDEX2(i,numComp,1)] = f_10[i]*c0 + f_11[i]*c1;
1303 } /* end of component loop i */
1304 } /* end of k1 loop */
1305 } /* end of face 1 */
1306 if (m_faceOffset[2] > -1) {
1307 #pragma omp for nowait
1308 for (index_t k0=0; k0 < m_NE0; ++k0) {
1309 const register double* f_10 = in.getSampleDataRO(INDEX2(k0+1,0, m_N0));
1310 const register double* f_00 = in.getSampleDataRO(INDEX2(k0,0, m_N0));
1311 double* o = out.getSampleDataRW(m_faceOffset[2]+k0);
1312 for (index_t i=0; i < numComp; ++i) {
1313 o[INDEX2(i,numComp,0)] = f_00[i]*c1 + f_10[i]*c0;
1314 o[INDEX2(i,numComp,1)] = f_00[i]*c0 + f_10[i]*c1;
1315 } /* end of component loop i */
1316 } /* end of k0 loop */
1317 } /* end of face 2 */
1318 if (m_faceOffset[3] > -1) {
1319 #pragma omp for nowait
1320 for (index_t k0=0; k0 < m_NE0; ++k0) {
1321 const register double* f_11 = in.getSampleDataRO(INDEX2(k0+1,m_N1-1, m_N0));
1322 const register double* f_01 = in.getSampleDataRO(INDEX2(k0,m_N1-1, m_N0));
1323 double* o = out.getSampleDataRW(m_faceOffset[3]+k0);
1324 for (index_t i=0; i < numComp; ++i) {
1325 o[INDEX2(i,numComp,0)] = f_01[i]*c1 + f_11[i]*c0;
1326 o[INDEX2(i,numComp,1)] = f_01[i]*c0 + f_11[i]*c1;
1327 } /* end of component loop i */
1328 } /* end of k0 loop */
1329 } /* end of face 3 */
1330 /* GENERATOR SNIP_INTERPOLATE_FACES BOTTOM */
1331 } // end of parallel section
1332 }
1333 }
1334
1335 //protected
1336 void Rectangle::assemblePDESingle(Paso_SystemMatrix* mat,
1337 escript::Data& rhs, const escript::Data& A, const escript::Data& B,
1338 const escript::Data& C, const escript::Data& D,
1339 const escript::Data& X, const escript::Data& Y,
1340 const escript::Data& d, const escript::Data& y) const
1341 {
1342 const double h0 = m_l0/m_gNE0;
1343 const double h1 = m_l1/m_gNE1;
1344 /*** GENERATOR SNIP_PDE_SINGLE_PRE TOP */
1345 const double w0 = -0.1555021169820365539*h1/h0;
1346 const double w1 = 0.041666666666666666667;
1347 const double w10 = -0.041666666666666666667*h0/h1;
1348 const double w11 = 0.1555021169820365539*h1/h0;
1349 const double w12 = 0.1555021169820365539*h0/h1;
1350 const double w13 = 0.01116454968463011277*h0/h1;
1351 const double w14 = 0.01116454968463011277*h1/h0;
1352 const double w15 = 0.041666666666666666667*h1/h0;
1353 const double w16 = -0.01116454968463011277*h0/h1;
1354 const double w17 = -0.1555021169820365539*h0/h1;
1355 const double w18 = -0.33333333333333333333*h1/h0;
1356 const double w19 = 0.25000000000000000000;
1357 const double w2 = -0.15550211698203655390;
1358 const double w20 = -0.25000000000000000000;
1359 const double w21 = 0.16666666666666666667*h0/h1;
1360 const double w22 = -0.16666666666666666667*h1/h0;
1361 const double w23 = -0.16666666666666666667*h0/h1;
1362 const double w24 = 0.33333333333333333333*h1/h0;
1363 const double w25 = 0.33333333333333333333*h0/h1;
1364 const double w26 = 0.16666666666666666667*h1/h0;
1365 const double w27 = -0.33333333333333333333*h0/h1;
1366 const double w28 = -0.032861463941450536761*h1;
1367 const double w29 = -0.032861463941450536761*h0;
1368 const double w3 = 0.041666666666666666667*h0/h1;
1369 const double w30 = -0.12264065304058601714*h1;
1370 const double w31 = -0.0023593469594139828636*h1;
1371 const double w32 = -0.008805202725216129906*h0;
1372 const double w33 = -0.008805202725216129906*h1;
1373 const double w34 = 0.032861463941450536761*h1;
1374 const double w35 = 0.008805202725216129906*h1;
1375 const double w36 = 0.008805202725216129906*h0;
1376 const double w37 = 0.0023593469594139828636*h1;
1377 const double w38 = 0.12264065304058601714*h1;
1378 const double w39 = 0.032861463941450536761*h0;
1379 const double w4 = 0.15550211698203655390;
1380 const double w40 = -0.12264065304058601714*h0;
1381 const double w41 = -0.0023593469594139828636*h0;
1382 const double w42 = 0.0023593469594139828636*h0;
1383 const double w43 = 0.12264065304058601714*h0;
1384 const double w44 = -0.16666666666666666667*h1;
1385 const double w45 = -0.083333333333333333333*h0;
1386 const double w46 = 0.083333333333333333333*h1;
1387 const double w47 = 0.16666666666666666667*h1;
1388 const double w48 = 0.083333333333333333333*h0;
1389 const double w49 = -0.16666666666666666667*h0;
1390 const double w5 = -0.041666666666666666667;
1391 const double w50 = 0.16666666666666666667*h0;
1392 const double w51 = -0.083333333333333333333*h1;
1393 const double w52 = 0.025917019497006092316*h0*h1;
1394 const double w53 = 0.0018607582807716854616*h0*h1;
1395 const double w54 = 0.0069444444444444444444*h0*h1;
1396 const double w55 = 0.09672363354357992482*h0*h1;
1397 const double w56 = 0.00049858867864229740201*h0*h1;
1398 const double w57 = 0.055555555555555555556*h0*h1;
1399 const double w58 = 0.027777777777777777778*h0*h1;
1400 const double w59 = 0.11111111111111111111*h0*h1;
1401 const double w6 = -0.01116454968463011277*h1/h0;
1402 const double w60 = -0.19716878364870322056*h1;
1403 const double w61 = -0.19716878364870322056*h0;
1404 const double w62 = -0.052831216351296779436*h0;
1405 const double w63 = -0.052831216351296779436*h1;
1406 const double w64 = 0.19716878364870322056*h1;
1407 const double w65 = 0.052831216351296779436*h1;
1408 const double w66 = 0.19716878364870322056*h0;
1409 const double w67 = 0.052831216351296779436*h0;
1410 const double w68 = -0.5*h1;
1411 const double w69 = -0.5*h0;
1412 const double w7 = 0.011164549684630112770;
1413 const double w70 = 0.5*h1;
1414 const double w71 = 0.5*h0;
1415 const double w72 = 0.1555021169820365539*h0*h1;
1416 const double w73 = 0.041666666666666666667*h0*h1;
1417 const double w74 = 0.01116454968463011277*h0*h1;
1418 const double w75 = 0.25*h0*h1;
1419 const double w8 = -0.011164549684630112770;
1420 const double w9 = -0.041666666666666666667*h1/h0;
1421 /* GENERATOR SNIP_PDE_SINGLE_PRE BOTTOM */
1422
1423 rhs.requireWrite();
1424 #pragma omp parallel
1425 {
1426 for (index_t k1_0=0; k1_0<2; k1_0++) { // colouring
1427 #pragma omp for
1428 for (index_t k1=k1_0; k1<m_NE1; k1+=2) {
1429 for (index_t k0=0; k0<m_NE0; ++k0) {
1430 bool add_EM_S=false;
1431 bool add_EM_F=false;
1432 vector<double> EM_S(4*4, 0);
1433 vector<double> EM_F(4, 0);
1434 const index_t e = k0 + m_NE0*k1;
1435 /*** GENERATOR SNIP_PDE_SINGLE TOP */
1436 ///////////////
1437 // process A //
1438 ///////////////
1439 if (!A.isEmpty()) {
1440 add_EM_S=true;
1441 const double* A_p=const_cast<escript::Data*>(&A)->getSampleDataRO(e);
1442 if (A.actsExpanded()) {
1443 const register double A_00_0 = A_p[INDEX3(0,0,0,2,2)];
1444 const register double A_01_0 = A_p[INDEX3(0,1,0,2,2)];
1445 const register double A_10_0 = A_p[INDEX3(1,0,0,2,2)];
1446 const register double A_11_0 = A_p[INDEX3(1,1,0,2,2)];
1447 const register double A_00_1 = A_p[INDEX3(0,0,1,2,2)];
1448 const register double A_01_1 = A_p[INDEX3(0,1,1,2,2)];
1449 const register double A_10_1 = A_p[INDEX3(1,0,1,2,2)];
1450 const register double A_11_1 = A_p[INDEX3(1,1,1,2,2)];
1451 const register double A_00_2 = A_p[INDEX3(0,0,2,2,2)];
1452 const register double A_01_2 = A_p[INDEX3(0,1,2,2,2)];
1453 const register double A_10_2 = A_p[INDEX3(1,0,2,2,2)];
1454 const register double A_11_2 = A_p[INDEX3(1,1,2,2,2)];
1455 const register double A_00_3 = A_p[INDEX3(0,0,3,2,2)];
1456 const register double A_01_3 = A_p[INDEX3(0,1,3,2,2)];
1457 const register double A_10_3 = A_p[INDEX3(1,0,3,2,2)];
1458 const register double A_11_3 = A_p[INDEX3(1,1,3,2,2)];
1459 const register double tmp4_0 = A_10_1 + A_10_2;
1460 const register double tmp12_0 = A_11_0 + A_11_2;
1461 const register double tmp2_0 = A_11_0 + A_11_1 + A_11_2 + A_11_3;
1462 const register double tmp10_0 = A_01_3 + A_10_3;
1463 const register double tmp14_0 = A_01_0 + A_01_3 + A_10_0 + A_10_3;
1464 const register double tmp0_0 = A_01_0 + A_01_3;
1465 const register double tmp13_0 = A_01_2 + A_10_1;
1466 const register double tmp3_0 = A_00_2 + A_00_3;
1467 const register double tmp11_0 = A_11_1 + A_11_3;
1468 const register double tmp18_0 = A_01_1 + A_10_1;
1469 const register double tmp1_0 = A_00_0 + A_00_1;
1470 const register double tmp15_0 = A_01_1 + A_10_2;
1471 const register double tmp5_0 = A_00_0 + A_00_1 + A_00_2 + A_00_3;
1472 const register double tmp16_0 = A_10_0 + A_10_3;
1473 const register double tmp6_0 = A_01_3 + A_10_0;
1474 const register double tmp17_0 = A_01_1 + A_01_2;
1475 const register double tmp9_0 = A_01_0 + A_10_0;
1476 const register double tmp7_0 = A_01_0 + A_10_3;
1477 const register double tmp8_0 = A_01_1 + A_01_2 + A_10_1 + A_10_2;
1478 const register double tmp19_0 = A_01_2 + A_10_2;
1479 const register double tmp14_1 = A_10_0*w8;
1480 const register double tmp23_1 = tmp3_0*w14;
1481 const register double tmp35_1 = A_01_0*w8;
1482 const register double tmp54_1 = tmp13_0*w8;
1483 const register double tmp20_1 = tmp9_0*w4;
1484 const register double tmp25_1 = tmp12_0*w12;
1485 const register double tmp2_1 = A_01_1*w4;
1486 const register double tmp44_1 = tmp7_0*w7;
1487 const register double tmp26_1 = tmp10_0*w4;
1488 const register double tmp52_1 = tmp18_0*w8;
1489 const register double tmp48_1 = A_10_1*w7;
1490 const register double tmp46_1 = A_01_3*w8;
1491 const register double tmp50_1 = A_01_0*w2;
1492 const register double tmp8_1 = tmp4_0*w5;
1493 const register double tmp56_1 = tmp19_0*w8;
1494 const register double tmp9_1 = tmp2_0*w10;
1495 const register double tmp19_1 = A_10_3*w2;
1496 const register double tmp47_1 = A_10_2*w4;
1497 const register double tmp16_1 = tmp3_0*w0;
1498 const register double tmp18_1 = tmp1_0*w6;
1499 const register double tmp31_1 = tmp11_0*w12;
1500 const register double tmp55_1 = tmp15_0*w2;
1501 const register double tmp39_1 = A_10_2*w7;
1502 const register double tmp11_1 = tmp6_0*w7;
1503 const register double tmp40_1 = tmp11_0*w17;
1504 const register double tmp34_1 = tmp15_0*w8;
1505 const register double tmp33_1 = tmp14_0*w5;
1506 const register double tmp24_1 = tmp11_0*w13;
1507 const register double tmp3_1 = tmp1_0*w0;
1508 const register double tmp5_1 = tmp2_0*w3;
1509 const register double tmp43_1 = tmp17_0*w5;
1510 const register double tmp15_1 = A_01_2*w4;
1511 const register double tmp53_1 = tmp19_0*w2;
1512 const register double tmp27_1 = tmp3_0*w11;
1513 const register double tmp32_1 = tmp13_0*w2;
1514 const register double tmp10_1 = tmp5_0*w9;
1515 const register double tmp37_1 = A_10_1*w4;
1516 const register double tmp38_1 = tmp5_0*w15;
1517 const register double tmp17_1 = A_01_1*w7;
1518 const register double tmp12_1 = tmp7_0*w4;
1519 const register double tmp22_1 = tmp10_0*w7;
1520 const register double tmp57_1 = tmp18_0*w2;
1521 const register double tmp28_1 = tmp9_0*w7;
1522 const register double tmp29_1 = tmp1_0*w14;
1523 const register double tmp51_1 = tmp11_0*w16;
1524 const register double tmp42_1 = tmp12_0*w16;
1525 const register double tmp49_1 = tmp12_0*w17;
1526 const register double tmp21_1 = tmp1_0*w11;
1527 const register double tmp1_1 = tmp0_0*w1;
1528 const register double tmp45_1 = tmp6_0*w4;
1529 const register double tmp7_1 = A_10_0*w2;
1530 const register double tmp6_1 = tmp3_0*w6;
1531 const register double tmp13_1 = tmp8_0*w1;
1532 const register double tmp36_1 = tmp16_0*w1;
1533 const register double tmp41_1 = A_01_3*w2;
1534 const register double tmp30_1 = tmp12_0*w13;
1535 const register double tmp4_1 = A_01_2*w7;
1536 const register double tmp0_1 = A_10_3*w8;
1537 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1 + tmp4_1 + tmp5_1 + tmp6_1 + tmp7_1 + tmp8_1;
1538 EM_S[INDEX2(1,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1 + tmp9_1;
1539 EM_S[INDEX2(3,2,4)]+=tmp14_1 + tmp15_1 + tmp16_1 + tmp17_1 + tmp18_1 + tmp19_1 + tmp1_1 + tmp5_1 + tmp8_1;
1540 EM_S[INDEX2(0,0,4)]+=tmp13_1 + tmp20_1 + tmp21_1 + tmp22_1 + tmp23_1 + tmp24_1 + tmp25_1;
1541 EM_S[INDEX2(3,3,4)]+=tmp13_1 + tmp26_1 + tmp27_1 + tmp28_1 + tmp29_1 + tmp30_1 + tmp31_1;
1542 EM_S[INDEX2(3,0,4)]+=tmp10_1 + tmp32_1 + tmp33_1 + tmp34_1 + tmp9_1;
1543 EM_S[INDEX2(3,1,4)]+=tmp35_1 + tmp36_1 + tmp37_1 + tmp38_1 + tmp39_1 + tmp40_1 + tmp41_1 + tmp42_1 + tmp43_1;
1544 EM_S[INDEX2(2,1,4)]+=tmp10_1 + tmp13_1 + tmp44_1 + tmp45_1 + tmp9_1;
1545 EM_S[INDEX2(0,2,4)]+=tmp36_1 + tmp38_1 + tmp43_1 + tmp46_1 + tmp47_1 + tmp48_1 + tmp49_1 + tmp50_1 + tmp51_1;
1546 EM_S[INDEX2(2,0,4)]+=tmp0_1 + tmp15_1 + tmp17_1 + tmp1_1 + tmp38_1 + tmp49_1 + tmp51_1 + tmp7_1 + tmp8_1;
1547 EM_S[INDEX2(1,3,4)]+=tmp14_1 + tmp19_1 + tmp1_1 + tmp2_1 + tmp38_1 + tmp40_1 + tmp42_1 + tmp4_1 + tmp8_1;
1548 EM_S[INDEX2(2,3,4)]+=tmp16_1 + tmp18_1 + tmp35_1 + tmp36_1 + tmp41_1 + tmp43_1 + tmp47_1 + tmp48_1 + tmp5_1;
1549 EM_S[INDEX2(2,2,4)]+=tmp24_1 + tmp25_1 + tmp27_1 + tmp29_1 + tmp33_1 + tmp52_1 + tmp53_1;
1550 EM_S[INDEX2(1,0,4)]+=tmp36_1 + tmp37_1 + tmp39_1 + tmp3_1 + tmp43_1 + tmp46_1 + tmp50_1 + tmp5_1 + tmp6_1;
1551 EM_S[INDEX2(0,3,4)]+=tmp10_1 + tmp33_1 + tmp54_1 + tmp55_1 + tmp9_1;
1552 EM_S[INDEX2(1,1,4)]+=tmp21_1 + tmp23_1 + tmp30_1 + tmp31_1 + tmp33_1 + tmp56_1 + tmp57_1;
1553 } else { /* constant data */
1554 const register double A_00 = A_p[INDEX2(0,0,2)];
1555 const register double A_01 = A_p[INDEX2(0,1,2)];
1556 const register double A_10 = A_p[INDEX2(1,0,2)];
1557 const register double A_11 = A_p[INDEX2(1,1,2)];
1558 const register double tmp0_0 = A_01 + A_10;
1559 const register double tmp0_1 = A_00*w18;
1560 const register double tmp10_1 = A_01*w20;
1561 const register double tmp12_1 = A_00*w26;
1562 const register double tmp4_1 = A_00*w22;
1563 const register double tmp8_1 = A_00*w24;
1564 const register double tmp13_1 = A_10*w19;
1565 const register double tmp9_1 = tmp0_0*w20;
1566 const register double tmp3_1 = A_11*w21;
1567 const register double tmp11_1 = A_11*w27;
1568 const register double tmp1_1 = A_01*w19;
1569 const register double tmp6_1 = A_11*w23;
1570 const register double tmp7_1 = A_11*w25;
1571 const register double tmp2_1 = A_10*w20;
1572 const register double tmp5_1 = tmp0_0*w19;
1573 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1;
1574 EM_S[INDEX2(1,2,4)]+=tmp4_1 + tmp5_1 + tmp6_1;
1575 EM_S[INDEX2(3,2,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1;
1576 EM_S[INDEX2(0,0,4)]+=tmp5_1 + tmp7_1 + tmp8_1;
1577 EM_S[INDEX2(3,3,4)]+=tmp5_1 + tmp7_1 + tmp8_1;
1578 EM_S[INDEX2(3,0,4)]+=tmp4_1 + tmp6_1 + tmp9_1;
1579 EM_S[INDEX2(3,1,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1;
1580 EM_S[INDEX2(2,1,4)]+=tmp4_1 + tmp5_1 + tmp6_1;
1581 EM_S[INDEX2(0,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1;
1582 EM_S[INDEX2(2,0,4)]+=tmp11_1 + tmp12_1 + tmp1_1 + tmp2_1;
1583 EM_S[INDEX2(1,3,4)]+=tmp11_1 + tmp12_1 + tmp1_1 + tmp2_1;
1584 EM_S[INDEX2(2,3,4)]+=tmp0_1 + tmp10_1 + tmp13_1 + tmp3_1;
1585 EM_S[INDEX2(2,2,4)]+=tmp7_1 + tmp8_1 + tmp9_1;
1586 EM_S[INDEX2(1,0,4)]+=tmp0_1 + tmp10_1 + tmp13_1 + tmp3_1;
1587 EM_S[INDEX2(0,3,4)]+=tmp4_1 + tmp6_1 + tmp9_1;
1588 EM_S[INDEX2(1,1,4)]+=tmp7_1 + tmp8_1 + tmp9_1;
1589 }
1590 }
1591 ///////////////
1592 // process B //
1593 ///////////////
1594 if (!B.isEmpty()) {
1595 add_EM_S=true;
1596 const double* B_p=const_cast<escript::Data*>(&B)->getSampleDataRO(e);
1597 if (B.actsExpanded()) {
1598 const register double B_0_0 = B_p[INDEX2(0,0,2)];
1599 const register double B_1_0 = B_p[INDEX2(1,0,2)];
1600 const register double B_0_1 = B_p[INDEX2(0,1,2)];
1601 const register double B_1_1 = B_p[INDEX2(1,1,2)];
1602 const register double B_0_2 = B_p[INDEX2(0,2,2)];
1603 const register double B_1_2 = B_p[INDEX2(1,2,2)];
1604 const register double B_0_3 = B_p[INDEX2(0,3,2)];
1605 const register double B_1_3 = B_p[INDEX2(1,3,2)];
1606 const register double tmp3_0 = B_0_0 + B_0_2;
1607 const register double tmp1_0 = B_1_2 + B_1_3;
1608 const register double tmp2_0 = B_0_1 + B_0_3;
1609 const register double tmp0_0 = B_1_0 + B_1_1;
1610 const register double tmp63_1 = B_1_1*w42;
1611 const register double tmp79_1 = B_1_1*w40;
1612 const register double tmp37_1 = tmp3_0*w35;
1613 const register double tmp8_1 = tmp0_0*w32;
1614 const register double tmp71_1 = B_0_1*w34;
1615 const register double tmp19_1 = B_0_3*w31;
1616 const register double tmp15_1 = B_0_3*w34;
1617 const register double tmp9_1 = tmp3_0*w34;
1618 const register double tmp35_1 = B_1_0*w36;
1619 const register double tmp66_1 = B_0_3*w28;
1620 const register double tmp28_1 = B_1_0*w42;
1621 const register double tmp22_1 = B_1_0*w40;
1622 const register double tmp16_1 = B_1_2*w29;
1623 const register double tmp6_1 = tmp2_0*w35;
1624 const register double tmp55_1 = B_1_3*w40;
1625 const register double tmp50_1 = B_1_3*w42;
1626 const register double tmp7_1 = tmp1_0*w29;
1627 const register double tmp1_1 = tmp1_0*w32;
1628 const register double tmp57_1 = B_0_3*w30;
1629 const register double tmp18_1 = B_1_1*w32;
1630 const register double tmp53_1 = B_1_0*w41;
1631 const register double tmp61_1 = B_1_3*w36;
1632 const register double tmp27_1 = B_0_3*w38;
1633 const register double tmp64_1 = B_0_2*w30;
1634 const register double tmp76_1 = B_0_1*w38;
1635 const register double tmp39_1 = tmp2_0*w34;
1636 const register double tmp62_1 = B_0_1*w31;
1637 const register double tmp56_1 = B_0_0*w31;
1638 const register double tmp49_1 = B_1_1*w36;
1639 const register double tmp2_1 = B_0_2*w31;
1640 const register double tmp23_1 = B_0_2*w33;
1641 const register double tmp38_1 = B_1_1*w43;
1642 const register double tmp74_1 = B_1_2*w41;
1643 const register double tmp43_1 = B_1_1*w41;
1644 const register double tmp58_1 = B_0_2*w28;
1645 const register double tmp67_1 = B_0_0*w33;
1646 const register double tmp33_1 = tmp0_0*w39;
1647 const register double tmp4_1 = B_0_0*w28;
1648 const register double tmp20_1 = B_0_0*w30;
1649 const register double tmp13_1 = B_0_2*w38;
1650 const register double tmp65_1 = B_1_2*w43;
1651 const register double tmp0_1 = tmp0_0*w29;
1652 const register double tmp41_1 = tmp3_0*w33;
1653 const register double tmp73_1 = B_0_2*w37;
1654 const register double tmp69_1 = B_0_0*w38;
1655 const register double tmp48_1 = B_1_2*w39;
1656 const register double tmp59_1 = B_0_1*w33;
1657 const register double tmp17_1 = B_1_3*w41;
1658 const register double tmp5_1 = B_0_3*w33;
1659 const register double tmp3_1 = B_0_1*w30;
1660 const register double tmp21_1 = B_0_1*w28;
1661 const register double tmp42_1 = B_1_0*w29;
1662 const register double tmp54_1 = B_1_2*w32;
1663 const register double tmp60_1 = B_1_0*w39;
1664 const register double tmp32_1 = tmp1_0*w36;
1665 const register double tmp10_1 = B_0_1*w37;
1666 const register double tmp14_1 = B_0_0*w35;
1667 const register double tmp29_1 = B_0_1*w35;
1668 const register double tmp26_1 = B_1_2*w36;
1669 const register double tmp30_1 = B_1_3*w43;
1670 const register double tmp70_1 = B_0_2*w35;
1671 const register double tmp34_1 = B_1_3*w39;
1672 const register double tmp51_1 = B_1_0*w43;
1673 const register double tmp31_1 = B_0_2*w34;
1674 const register double tmp45_1 = tmp3_0*w28;
1675 const register double tmp11_1 = tmp1_0*w39;
1676 const register double tmp52_1 = B_1_1*w29;
1677 const register double tmp44_1 = B_1_3*w32;
1678 const register double tmp25_1 = B_1_1*w39;
1679 const register double tmp47_1 = tmp2_0*w33;
1680 const register double tmp72_1 = B_1_3*w29;
1681 const register double tmp40_1 = tmp2_0*w28;
1682 const register double tmp46_1 = B_1_2*w40;
1683 const register double tmp36_1 = B_1_2*w42;
1684 const register double tmp24_1 = B_0_0*w37;
1685 const register double tmp77_1 = B_0_3*w35;
1686 const register double tmp68_1 = B_0_3*w37;
1687 const register double tmp78_1 = B_0_0*w34;
1688 const register double tmp12_1 = tmp0_0*w36;
1689 const register double tmp75_1 = B_1_0*w32;
1690 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1 + tmp4_1 + tmp5_1;
1691 EM_S[INDEX2(1,2,4)]+=tmp6_1 + tmp7_1 + tmp8_1 + tmp9_1;
1692 EM_S[INDEX2(3,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1 + tmp14_1 + tmp15_1;
1693 EM_S[INDEX2(0,0,4)]+=tmp16_1 + tmp17_1 + tmp18_1 + tmp19_1 + tmp20_1 + tmp21_1 + tmp22_1 + tmp23_1;
1694 EM_S[INDEX2(3,3,4)]+=tmp24_1 + tmp25_1 + tmp26_1 + tmp27_1 + tmp28_1 + tmp29_1 + tmp30_1 + tmp31_1;
1695 EM_S[INDEX2(3,0,4)]+=tmp32_1 + tmp33_1 + tmp6_1 + tmp9_1;
1696 EM_S[INDEX2(3,1,4)]+=tmp34_1 + tmp35_1 + tmp36_1 + tmp37_1 + tmp38_1 + tmp39_1;
1697 EM_S[INDEX2(2,1,4)]+=tmp32_1 + tmp33_1 + tmp40_1 + tmp41_1;
1698 EM_S[INDEX2(0,2,4)]+=tmp42_1 + tmp43_1 + tmp44_1 + tmp45_1 + tmp46_1 + tmp47_1;
1699 EM_S[INDEX2(2,0,4)]+=tmp45_1 + tmp47_1 + tmp48_1 + tmp49_1 + tmp50_1 + tmp51_1;
1700 EM_S[INDEX2(1,3,4)]+=tmp37_1 + tmp39_1 + tmp52_1 + tmp53_1 + tmp54_1 + tmp55_1;
1701 EM_S[INDEX2(2,3,4)]+=tmp11_1 + tmp12_1 + tmp56_1 + tmp57_1 + tmp58_1 + tmp59_1;
1702 EM_S[INDEX2(2,2,4)]+=tmp60_1 + tmp61_1 + tmp62_1 + tmp63_1 + tmp64_1 + tmp65_1 + tmp66_1 + tmp67_1;
1703 EM_S[INDEX2(1,0,4)]+=tmp0_1 + tmp1_1 + tmp68_1 + tmp69_1 + tmp70_1 + tmp71_1;
1704 EM_S[INDEX2(0,3,4)]+=tmp40_1 + tmp41_1 + tmp7_1 + tmp8_1;
1705 EM_S[INDEX2(1,1,4)]+=tmp72_1 + tmp73_1 + tmp74_1 + tmp75_1 + tmp76_1 + tmp77_1 + tmp78_1 + tmp79_1;
1706 } else { /* constant data */
1707 const register double B_0 = B_p[0];
1708 const register double B_1 = B_p[1];
1709 const register double tmp6_1 = B_1*w50;
1710 const register double tmp1_1 = B_1*w45;
1711 const register double tmp5_1 = B_1*w49;
1712 const register double tmp4_1 = B_1*w48;
1713 const register double tmp0_1 = B_0*w44;
1714 const register double tmp2_1 = B_0*w46;
1715 const register double tmp7_1 = B_0*w51;
1716 const register double tmp3_1 = B_0*w47;
1717 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1;
1718 EM_S[INDEX2(1,2,4)]+=tmp1_1 + tmp2_1;
1719 EM_S[INDEX2(3,2,4)]+=tmp3_1 + tmp4_1;
1720 EM_S[INDEX2(0,0,4)]+=tmp0_1 + tmp5_1;
1721 EM_S[INDEX2(3,3,4)]+=tmp3_1 + tmp6_1;
1722 EM_S[INDEX2(3,0,4)]+=tmp2_1 + tmp4_1;
1723 EM_S[INDEX2(3,1,4)]+=tmp2_1 + tmp6_1;
1724 EM_S[INDEX2(2,1,4)]+=tmp4_1 + tmp7_1;
1725 EM_S[INDEX2(0,2,4)]+=tmp5_1 + tmp7_1;
1726 EM_S[INDEX2(2,0,4)]+=tmp6_1 + tmp7_1;
1727 EM_S[INDEX2(1,3,4)]+=tmp2_1 + tmp5_1;
1728 EM_S[INDEX2(2,3,4)]+=tmp0_1 + tmp4_1;
1729 EM_S[INDEX2(2,2,4)]+=tmp0_1 + tmp6_1;
1730 EM_S[INDEX2(1,0,4)]+=tmp1_1 + tmp3_1;
1731 EM_S[INDEX2(0,3,4)]+=tmp1_1 + tmp7_1;
1732 EM_S[INDEX2(1,1,4)]+=tmp3_1 + tmp5_1;
1733 }
1734 }
1735 ///////////////
1736 // process C //
1737 ///////////////
1738 if (!C.isEmpty()) {
1739 add_EM_S=true;
1740 const double* C_p=const_cast<escript::Data*>(&C)->getSampleDataRO(e);
1741 if (C.actsExpanded()) {
1742 const register double C_0_0 = C_p[INDEX2(0,0,2)];
1743 const register double C_1_0 = C_p[INDEX2(1,0,2)];
1744 const register double C_0_1 = C_p[INDEX2(0,1,2)];
1745 const register double C_1_1 = C_p[INDEX2(1,1,2)];
1746 const register double C_0_2 = C_p[INDEX2(0,2,2)];
1747 const register double C_1_2 = C_p[INDEX2(1,2,2)];
1748 const register double C_0_3 = C_p[INDEX2(0,3,2)];
1749 const register double C_1_3 = C_p[INDEX2(1,3,2)];
1750 const register double tmp2_0 = C_0_1 + C_0_3;
1751 const register double tmp1_0 = C_1_2 + C_1_3;
1752 const register double tmp3_0 = C_0_0 + C_0_2;
1753 const register double tmp0_0 = C_1_0 + C_1_1;
1754 const register double tmp64_1 = C_0_2*w30;
1755 const register double tmp14_1 = C_0_2*w28;
1756 const register double tmp19_1 = C_0_3*w31;
1757 const register double tmp22_1 = C_1_0*w40;
1758 const register double tmp37_1 = tmp3_0*w35;
1759 const register double tmp29_1 = C_0_1*w35;
1760 const register double tmp73_1 = C_0_2*w37;
1761 const register double tmp74_1 = C_1_2*w41;
1762 const register double tmp52_1 = C_1_3*w39;
1763 const register double tmp25_1 = C_1_1*w39;
1764 const register double tmp62_1 = C_0_1*w31;
1765 const register double tmp79_1 = C_1_1*w40;
1766 const register double tmp43_1 = C_1_1*w36;
1767 const register double tmp27_1 = C_0_3*w38;
1768 const register double tmp28_1 = C_1_0*w42;
1769 const register double tmp63_1 = C_1_1*w42;
1770 const register double tmp59_1 = C_0_3*w34;
1771 const register double tmp72_1 = C_1_3*w29;
1772 const register double tmp40_1 = tmp2_0*w35;
1773 const register double tmp13_1 = C_0_3*w30;
1774 const register double tmp51_1 = C_1_2*w40;
1775 const register double tmp54_1 = C_1_2*w42;
1776 const register double tmp12_1 = C_0_0*w31;
1777 const register double tmp2_1 = tmp1_0*w32;
1778 const register double tmp68_1 = C_0_2*w31;
1779 const register double tmp75_1 = C_1_0*w32;
1780 const register double tmp49_1 = C_1_1*w41;
1781 const register double tmp4_1 = C_0_2*w35;
1782 const register double tmp66_1 = C_0_3*w28;
1783 const register double tmp56_1 = C_0_1*w37;
1784 const register double tmp5_1 = C_0_1*w34;
1785 const register double tmp38_1 = tmp2_0*w34;
1786 const register double tmp76_1 = C_0_1*w38;
1787 const register double tmp21_1 = C_0_1*w28;
1788 const register double tmp69_1 = C_0_1*w30;
1789 const register double tmp53_1 = C_1_0*w36;
1790 const register double tmp42_1 = C_1_2*w39;
1791 const register double tmp32_1 = tmp1_0*w29;
1792 const register double tmp45_1 = C_1_0*w43;
1793 const register double tmp33_1 = tmp0_0*w32;
1794 const register double tmp35_1 = C_1_0*w41;
1795 const register double tmp26_1 = C_1_2*w36;
1796 const register double tmp67_1 = C_0_0*w33;
1797 const register double tmp31_1 = C_0_2*w34;
1798 const register double tmp20_1 = C_0_0*w30;
1799 const register double tmp70_1 = C_0_0*w28;
1800 const register double tmp8_1 = tmp0_0*w39;
1801 const register double tmp30_1 = C_1_3*w43;
1802 const register double tmp0_1 = tmp0_0*w29;
1803 const register double tmp17_1 = C_1_3*w41;
1804 const register double tmp58_1 = C_0_0*w35;
1805 const register double tmp9_1 = tmp3_0*w33;
1806 const register double tmp61_1 = C_1_3*w36;
1807 const register double tmp41_1 = tmp3_0*w34;
1808 const register double tmp50_1 = C_1_3*w32;
1809 const register double tmp18_1 = C_1_1*w32;
1810 const register double tmp6_1 = tmp1_0*w36;
1811 const register double tmp3_1 = C_0_0*w38;
1812 const register double tmp34_1 = C_1_1*w29;
1813 const register double tmp77_1 = C_0_3*w35;
1814 const register double tmp65_1 = C_1_2*w43;
1815 const register double tmp71_1 = C_0_3*w33;
1816 const register double tmp55_1 = C_1_1*w43;
1817 const register double tmp46_1 = tmp3_0*w28;
1818 const register double tmp24_1 = C_0_0*w37;
1819 const register double tmp10_1 = tmp1_0*w39;
1820 const register double tmp48_1 = C_1_0*w29;
1821 const register double tmp15_1 = C_0_1*w33;
1822 const register double tmp36_1 = C_1_2*w32;
1823 const register double tmp60_1 = C_1_0*w39;
1824 const register double tmp47_1 = tmp2_0*w33;
1825 const register double tmp16_1 = C_1_2*w29;
1826 const register double tmp1_1 = C_0_3*w37;
1827 const register double tmp7_1 = tmp2_0*w28;
1828 const register double tmp39_1 = C_1_3*w40;
1829 const register double tmp44_1 = C_1_3*w42;
1830 const register double tmp57_1 = C_0_2*w38;
1831 const register double tmp78_1 = C_0_0*w34;
1832 const register double tmp11_1 = tmp0_0*w36;
1833 const register double tmp23_1 = C_0_2*w33;
1834 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1 + tmp4_1 + tmp5_1;
1835 EM_S[INDEX2(1,2,4)]+=tmp6_1 + tmp7_1 + tmp8_1 + tmp9_1;
1836 EM_S[INDEX2(3,2,4)]+=tmp10_1 + tmp11_1 + tmp12_1 + tmp13_1 + tmp14_1 + tmp15_1;
1837 EM_S[INDEX2(0,0,4)]+=tmp16_1 + tmp17_1 + tmp18_1 + tmp19_1 + tmp20_1 + tmp21_1 + tmp22_1 + tmp23_1;
1838 EM_S[INDEX2(3,3,4)]+=tmp24_1 + tmp25_1 + tmp26_1 + tmp27_1 + tmp28_1 + tmp29_1 + tmp30_1 + tmp31_1;
1839 EM_S[INDEX2(3,0,4)]+=tmp32_1 + tmp33_1 + tmp7_1 + tmp9_1;
1840 EM_S[INDEX2(3,1,4)]+=tmp34_1 + tmp35_1 + tmp36_1 + tmp37_1 + tmp38_1 + tmp39_1;
1841 EM_S[INDEX2(2,1,4)]+=tmp32_1 + tmp33_1 + tmp40_1 + tmp41_1;
1842 EM_S[INDEX2(0,2,4)]+=tmp42_1 + tmp43_1 + tmp44_1 + tmp45_1 + tmp46_1 + tmp47_1;
1843 EM_S[INDEX2(2,0,4)]+=tmp46_1 + tmp47_1 + tmp48_1 + tmp49_1 + tmp50_1 + tmp51_1;
1844 EM_S[INDEX2(1,3,4)]+=tmp37_1 + tmp38_1 + tmp52_1 + tmp53_1 + tmp54_1 + tmp55_1;
1845 EM_S[INDEX2(2,3,4)]+=tmp10_1 + tmp11_1 + tmp56_1 + tmp57_1 + tmp58_1 + tmp59_1;
1846 EM_S[INDEX2(2,2,4)]+=tmp60_1 + tmp61_1 + tmp62_1 + tmp63_1 + tmp64_1 + tmp65_1 + tmp66_1 + tmp67_1;
1847 EM_S[INDEX2(1,0,4)]+=tmp0_1 + tmp2_1 + tmp68_1 + tmp69_1 + tmp70_1 + tmp71_1;
1848 EM_S[INDEX2(0,3,4)]+=tmp40_1 + tmp41_1 + tmp6_1 + tmp8_1;
1849 EM_S[INDEX2(1,1,4)]+=tmp72_1 + tmp73_1 + tmp74_1 + tmp75_1 + tmp76_1 + tmp77_1 + tmp78_1 + tmp79_1;
1850 } else { /* constant data */
1851 const register double C_0 = C_p[0];
1852 const register double C_1 = C_p[1];
1853 const register double tmp1_1 = C_1*w45;
1854 const register double tmp3_1 = C_0*w51;
1855 const register double tmp4_1 = C_0*w44;
1856 const register double tmp7_1 = C_0*w46;
1857 const register double tmp5_1 = C_1*w49;
1858 const register double tmp2_1 = C_1*w48;
1859 const register double tmp0_1 = C_0*w47;
1860 const register double tmp6_1 = C_1*w50;
1861 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1;
1862 EM_S[INDEX2(1,2,4)]+=tmp2_1 + tmp3_1;
1863 EM_S[INDEX2(3,2,4)]+=tmp2_1 + tmp4_1;
1864 EM_S[INDEX2(0,0,4)]+=tmp4_1 + tmp5_1;
1865 EM_S[INDEX2(3,3,4)]+=tmp0_1 + tmp6_1;
1866 EM_S[INDEX2(3,0,4)]+=tmp1_1 + tmp3_1;
1867 EM_S[INDEX2(3,1,4)]+=tmp5_1 + tmp7_1;
1868 EM_S[INDEX2(2,1,4)]+=tmp1_1 + tmp7_1;
1869 EM_S[INDEX2(0,2,4)]+=tmp3_1 + tmp6_1;
1870 EM_S[INDEX2(2,0,4)]+=tmp3_1 + tmp5_1;
1871 EM_S[INDEX2(1,3,4)]+=tmp6_1 + tmp7_1;
1872 EM_S[INDEX2(2,3,4)]+=tmp0_1 + tmp2_1;
1873 EM_S[INDEX2(2,2,4)]+=tmp4_1 + tmp6_1;
1874 EM_S[INDEX2(1,0,4)]+=tmp1_1 + tmp4_1;
1875 EM_S[INDEX2(0,3,4)]+=tmp2_1 + tmp7_1;
1876 EM_S[INDEX2(1,1,4)]+=tmp0_1 + tmp5_1;
1877 }
1878 }
1879 ///////////////
1880 // process D //
1881 ///////////////
1882 if (!D.isEmpty()) {
1883 add_EM_S=true;
1884 const double* D_p=const_cast<escript::Data*>(&D)->getSampleDataRO(e);
1885 if (D.actsExpanded()) {
1886 const register double D_0 = D_p[0];
1887 const register double D_1 = D_p[1];
1888 const register double D_2 = D_p[2];
1889 const register double D_3 = D_p[3];
1890 const register double tmp4_0 = D_1 + D_3;
1891 const register double tmp2_0 = D_0 + D_1 + D_2 + D_3;
1892 const register double tmp5_0 = D_0 + D_2;
1893 const register double tmp0_0 = D_0 + D_1;
1894 const register double tmp6_0 = D_0 + D_3;
1895 const register double tmp1_0 = D_2 + D_3;
1896 const register double tmp3_0 = D_1 + D_2;
1897 const register double tmp16_1 = D_1*w56;
1898 const register double tmp14_1 = tmp6_0*w54;
1899 const register double tmp8_1 = D_3*w55;
1900 const register double tmp2_1 = tmp2_0*w54;
1901 const register double tmp12_1 = tmp5_0*w52;
1902 const register double tmp4_1 = tmp0_0*w53;
1903 const register double tmp3_1 = tmp1_0*w52;
1904 const register double tmp13_1 = tmp4_0*w53;
1905 const register double tmp10_1 = tmp4_0*w52;
1906 const register double tmp1_1 = tmp1_0*w53;
1907 const register double tmp7_1 = D_3*w56;
1908 const register double tmp0_1 = tmp0_0*w52;
1909 const register double tmp11_1 = tmp5_0*w53;
1910 const register double tmp9_1 = D_0*w56;
1911 const register double tmp5_1 = tmp3_0*w54;
1912 const register double tmp18_1 = D_2*w56;
1913 const register double tmp17_1 = D_1*w55;
1914 const register double tmp6_1 = D_0*w55;
1915 const register double tmp15_1 = D_2*w55;
1916 EM_S[INDEX2(0,1,4)]+=tmp0_1 + tmp1_1;
1917 EM_S[INDEX2(1,2,4)]+=tmp2_1;
1918 EM_S[INDEX2(3,2,4)]+=tmp3_1 + tmp4_1;
1919 EM_S[INDEX2(0,0,4)]+=tmp5_1 + tmp6_1 + tmp7_1;
1920 EM_S[INDEX2(3,3,4)]+=tmp5_1 + tmp8_1 + tmp9_1;
1921 EM_S[INDEX2(3,0,4)]+=tmp2_1;
1922 EM_S[INDEX2(3,1,4)]+=tmp10_1 + tmp11_1;
1923 EM_S[INDEX2(2,1,4)]+=tmp2_1;
1924 EM_S[INDEX2(0,2,4)]+=tmp12_1 + tmp13_1;
1925 EM_S[INDEX2(2,0,4)]+=tmp12_1 + tmp13_1;
1926 EM_S[INDEX2(1,3,4)]+=tmp10_1 + tmp11_1;
1927 EM_S[INDEX2(2,3,4)]+=tmp3_1 + tmp4_1;
1928 EM_S[INDEX2(2,2,4)]+=tmp14_1 + tmp15_1 + tmp16_1;
1929 EM_S[INDEX2(1,0,4)]+=tmp0_1 + tmp1_1;
1930 EM_S[INDEX2(0,3,4)]+=tmp2_1;
1931 EM_S[INDEX2(1,1,4)]+=tmp14_1 + tmp17_1 + tmp18_1;
1932 } else { /* constant data */
1933 const register double D_0 = D_p[0];
1934 const register double tmp2_1 = D_0*w59;
1935 const register double tmp1_1 = D_0*w58;
1936 const register double tmp0_1 = D_0*w57;
1937 EM_S[INDEX2(0,1,4)]+=tmp0_1;
1938 EM_S[INDEX2(1,2,4)]+=tmp1_1;
1939 EM_S[INDEX2(3,2,4)]+=tmp0_1;
1940 EM_S[INDEX2(0,0,4)]+=tmp2_1;
1941 EM_S[INDEX2(3,3,4)]+=tmp2_1;
1942 EM_S[INDEX2(3,0,4)]+=tmp1_1;
1943 EM_S[INDEX2(3,1,4)]+=tmp0_1;
1944 EM_S[INDEX2(2,1,4)]+=tmp1_1;
1945 EM_S[INDEX2(0,2,4)]+=tmp0_1;
1946 EM_S[INDEX2(2,0,4)]+=tmp0_1;
1947 EM_S[INDEX2(1,3,4)]+=tmp0_1;
1948 EM_S[INDEX2(2,3,4)]+=tmp0_1;
1949 EM_S[INDEX2(2,2,4)]+=tmp2_1;
1950 EM_S[INDEX2(1,0,4)]+=tmp0_1;
1951 EM_S[INDEX2(0,3,4)]+=tmp1_1;
1952 EM_S[INDEX2(1,1,4)]+=tmp2_1;
1953 }
1954 }
1955 ///////////////
1956 // process X //
1957 ///////////////
1958 if (!X.isEmpty()) {
1959 add_EM_F=true;
1960 const double* X_p=const_cast<escript::Data*>(&X)->getSampleDataRO(e);
1961 if (X.actsExpanded()) {
1962 const register double X_0_0 = X_p[INDEX2(0,0,2)];
1963 const register double X_1_0 = X_p[INDEX2(1,0,2)];
1964 const register double X_0_1 = X_p[INDEX2(0,1,2)];
1965 const register double X_1_1 = X_p[INDEX2(1,1,2)];
1966 const register double X_0_2 = X_p[INDEX2(0,2,2)];
1967 const register double X_1_2 = X_p[INDEX2(1,2,2)];
1968 const register double X_0_3 = X_p[INDEX2(0,3,2)];
1969 const register double X_1_3 = X_p[INDEX2(1,3,2)];
1970 const register double tmp1_0 = X_1_1 + X_1_3;
1971 const register double tmp3_0 = X_0_0 + X_0_1;
1972 const register double tmp2_0 = X_1_0 + X_1_2;
1973 const register double tmp0_0 = X_0_2 + X_0_3;
1974 const register double tmp8_1 = tmp2_0*w66;
1975 const register double tmp5_1 = tmp3_0*w64;
1976 const register double tmp14_1 = tmp0_0*w64;
1977 const register double tmp3_1 = tmp3_0*w60;
1978 const register double tmp9_1 = tmp3_0*w63;
1979 const register double tmp13_1 = tmp3_0*w65;
1980 const register double tmp12_1 = tmp1_0*w66;
1981 const register double tmp10_1 = tmp0_0*w60;
1982 const register double tmp2_1 = tmp2_0*w61;
1983 const register double tmp6_1 = tmp2_0*w62;
1984 const register double tmp4_1 = tmp0_0*w65;
1985 const register double tmp11_1 = tmp1_0*w67;
1986 const register double tmp1_1 = tmp1_0*w62;
1987 const register double tmp7_1 = tmp1_0*w61;
1988 const register double tmp0_1 = tmp0_0*w63;
1989 const register double tmp15_1 = tmp2_0*w67;
1990 EM_F[0]+=tmp0_1 + tmp1_1 + tmp2_1 + tmp3_1;
1991 EM_F[1]+=tmp4_1 + tmp5_1 + tmp6_1 + tmp7_1;
1992 EM_F[2]+=tmp10_1 + tmp11_1 + tmp8_1 + tmp9_1;
1993 EM_F[3]+=tmp12_1 + tmp13_1 + tmp14_1 + tmp15_1;
1994 } else { /* constant data */
1995 const register double X_0 = X_p[0];
1996 const register double X_1 = X_p[1];
1997 const register double tmp3_1 = X_1*w71;
1998 const register double tmp2_1 = X_0*w70;
1999 const register double tmp1_1 = X_0*w68;
2000 const register double tmp0_1 = X_1*w69;
2001 EM_F[0]+=tmp0_1 + tmp1_1;
2002 EM_F[1]+=tmp0_1 + tmp2_1;
2003 EM_F[2]+=tmp1_1 + tmp3_1;
2004 EM_F[3]+=tmp2_1 + tmp3_1;
2005 }
2006 }
2007 ///////////////
2008 // process Y //
2009 ///////////////
2010 if (!Y.isEmpty()) {
2011 add_EM_F=true;
2012 const double* Y_p=const_cast<escript::Data*>(&Y)->getSampleDataRO(e);
2013 if (Y.actsExpanded()) {
2014 const register double Y_0 = Y_p[0];
2015 const register double Y_1 = Y_p[1];
2016 const register double Y_2 = Y_p[2];
2017 const register double Y_3 = Y_p[3];
2018 const register double tmp0_0 = Y_1 + Y_2;
2019 const register double tmp1_0 = Y_0 + Y_3;
2020 const register double tmp9_1 = Y_0*w74;
2021 const register double tmp4_1 = tmp1_0*w73;
2022 const register double tmp5_1 = Y_2*w74;
2023 const register double tmp7_1 = Y_1*w74;
2024 const register double tmp6_1 = Y_2*w72;
2025 const register double tmp2_1 = Y_3*w74;
2026 const register double tmp8_1 = Y_3*w72;
2027 const register double tmp3_1 = Y_1*w72;
2028 const register double tmp0_1 = Y_0*w72;
2029 const register double tmp1_1 = tmp0_0*w73;
2030 EM_F[0]+=tmp0_1 + tmp1_1 + tmp2_1;
2031 EM_F[1]+=tmp3_1 + tmp4_1 + tmp5_1;
2032 EM_F[2]+=tmp4_1 + tmp6_1 + tmp7_1;
2033 EM_F[3]+=tmp1_1 + tmp8_1 + tmp9_1;
2034 } else { /* constant data */
2035 const register double Y_0 = Y_p[0];
2036 const register double tmp0_1 = Y_0*w75;
2037 EM_F[0]+=tmp0_1;
2038 EM_F[1]+=tmp0_1;
2039 EM_F[2]+=tmp0_1;
2040 EM_F[3]+=tmp0_1;
2041 }
2042 }
2043 /* GENERATOR SNIP_PDE_SINGLE BOTTOM */
2044
2045 // add to matrix (if add_EM_S) and RHS (if add_EM_F)
2046 const index_t firstNode=m_N0*k1+k0;
2047 IndexVector rowIndex;
2048 rowIndex.push_back(m_dofMap[firstNode]);
2049 rowIndex.push_back(m_dofMap[firstNode+1]);
2050 rowIndex.push_back(m_dofMap[firstNode+m_N0]);
2051 rowIndex.push_back(m_dofMap[firstNode+m_N0+1]);
2052 if (add_EM_F) {
2053 //cout << "-- AddtoRHS -- " << endl;
2054 double *F_p=rhs.getSampleDataRW(0);
2055 for (index_t i=0; i<rowIndex.size(); i++) {
2056 if (rowIndex[i]<getNumDOF()) {
2057 F_p[rowIndex[i]]+=EM_F[i];
2058 //cout << "F[" << rowIndex[i] << "]=" << F_p[rowIndex[i]] << endl;
2059 }
2060 }
2061 //cout << "---"<<endl;
2062 }
2063 if (add_EM_S) {
2064 //cout << "-- AddtoSystem -- " << endl;
2065 addToSystemMatrix(mat, rowIndex, 1, rowIndex, 1, EM_S);
2066 }
2067
2068 } // end k0 loop
2069 } // end k1 loop
2070 } // end of coloring
2071 } // end of parallel region
2072 }
2073
2074 } // end of namespace ripley
2075

  ViewVC Help
Powered by ViewVC 1.1.26