/[escript]/trunk/finley/src/Assemble_LumpedSystem.c
ViewVC logotype

Contents of /trunk/finley/src/Assemble_LumpedSystem.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 2988 - (show annotations)
Tue Mar 16 06:14:30 2010 UTC (9 years, 1 month ago) by gross
File MIME type: text/plain
File size: 14831 byte(s)
C-level lumping switched on
1
2 /*******************************************************
3 *
4 * Copyright (c) 2003-2010 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
15 /**************************************************************/
16
17 /* assembles the mass matrix in lumped form */
18
19 /* The coefficient D has to be defined on the integration points or not present. */
20
21 /* lumpedMat has to be initialized before the routine is called. */
22
23 /**************************************************************/
24
25 #include "Assemble.h"
26 #include "Util.h"
27 #ifdef _OPENMP
28 #include <omp.h>
29 #endif
30
31
32 /* Disabled until the tests pass */
33 #define NEW_LUMPING
34
35 /**************************************************************/
36
37 void Finley_Assemble_LumpedSystem(Finley_NodeFile* nodes,Finley_ElementFile* elements, escriptDataC* lumpedMat, escriptDataC* D)
38 {
39
40 bool_t reducedIntegrationOrder=FALSE, expandedD;
41 char error_msg[LenErrorMsg_MAX];
42 Assemble_Parameters p;
43 dim_t dimensions[ESCRIPT_MAX_DATA_RANK], k, e, len_EM_lumpedMat, q, s, isub;
44 type_t funcspace;
45 index_t color,*row_index=NULL;
46 __const double *D_p=NULL;
47 double *S=NULL, *EM_lumpedMat=NULL, *Vol=NULL, *lumpedMat_p=NULL;
48 register double rtmp;
49 size_t len_EM_lumpedMat_size;
50 #ifdef NEW_LUMPING
51 register double m_t=0., diagS=0.;
52 #endif
53
54 Finley_resetError();
55
56 if (nodes==NULL || elements==NULL) return;
57 if (isEmpty(lumpedMat) || isEmpty(D)) return;
58 if (isEmpty(lumpedMat) && !isEmpty(D)) {
59 Finley_setError(TYPE_ERROR,"Assemble_LumpedSystem: coefficients are non-zero but no lumped matrix is given.");
60 return;
61 }
62 funcspace=getFunctionSpaceType(D);
63 /* check if all function spaces are the same */
64 if (funcspace==FINLEY_ELEMENTS) {
65 reducedIntegrationOrder=FALSE;
66 } else if (funcspace==FINLEY_FACE_ELEMENTS) {
67 reducedIntegrationOrder=FALSE;
68 } else if (funcspace==FINLEY_REDUCED_ELEMENTS) {
69 reducedIntegrationOrder=TRUE;
70 } else if (funcspace==FINLEY_REDUCED_FACE_ELEMENTS) {
71 reducedIntegrationOrder=TRUE;
72 } else {
73 Finley_setError(TYPE_ERROR,"Assemble_LumpedSystem: assemblage failed because of illegal function space.");
74 }
75 if (! Finley_noError()) return;
76
77 /* set all parameters in p*/
78 Assemble_getAssembleParameters(nodes,elements,NULL,lumpedMat, reducedIntegrationOrder, &p);
79 if (! Finley_noError()) return;
80
81 /* check if all function spaces are the same */
82
83 if (! numSamplesEqual(D,p.numQuadSub,elements->numElements) ) {
84 sprintf(error_msg,"Assemble_LumpedSystem: sample points of coefficient D don't match (%d,%d)",p.numQuadSub,elements->numElements);
85 Finley_setError(TYPE_ERROR,error_msg);
86 }
87
88 /* check the dimensions: */
89
90 if (p.numEqu==1) {
91 if (!isEmpty(D)) {
92 if (!isDataPointShapeEqual(D,0,dimensions)) {
93 Finley_setError(TYPE_ERROR,"Assemble_LumpedSystem: coefficient D, rank 0 expected.");
94 }
95
96 }
97 } else {
98 if (!isEmpty(D)) {
99 dimensions[0]=p.numEqu;
100 if (!isDataPointShapeEqual(D,1,dimensions)) {
101 sprintf(error_msg,"Assemble_LumpedSystem: coefficient D, expected shape (%d,)",dimensions[0]);
102 Finley_setError(TYPE_ERROR,error_msg);
103 }
104 }
105 }
106
107 if (Finley_noError()) {
108 requireWrite(lumpedMat);
109 lumpedMat_p=getSampleDataRW(lumpedMat,0);
110 len_EM_lumpedMat=p.row_numShapesTotal*p.numEqu;
111 len_EM_lumpedMat_size=len_EM_lumpedMat*sizeof(double);
112
113 expandedD=isExpanded(D);
114 S=p.row_jac->BasisFunctions->S;
115
116 #ifdef NEW_LUMPING
117 #pragma omp parallel private(color, EM_lumpedMat, row_index, Vol, D_p, s, q, k, rtmp, diagS, m_t)
118 #else
119 #pragma omp parallel private(color, EM_lumpedMat, row_index, Vol, D_p, s, q, k, rtmp)
120 #endif
121 {
122 EM_lumpedMat=THREAD_MEMALLOC(len_EM_lumpedMat,double);
123 row_index=THREAD_MEMALLOC(p.row_numShapesTotal,index_t);
124 if ( !Finley_checkPtr(EM_lumpedMat) && !Finley_checkPtr(row_index) ) {
125 if (p.numEqu == 1) { /* single equation */
126 if (expandedD) { /* with expanded D */
127 for (color=elements->minColor;color<=elements->maxColor;color++) {
128 /* open loop over all elements: */
129 #pragma omp for private(e) schedule(static)
130 for(e=0;e<elements->numElements;e++){
131 if (elements->Color[e]==color) {
132 for (isub=0; isub<p.numSub; isub++) {
133 Vol=&(p.row_jac->volume[INDEX3(0,isub,e, p.numQuadSub,p.numSub)]);;
134
135 D_p=getSampleDataRO(D,e);
136 #ifdef NEW_LUMPING /* HRZ lumping */
137
138 m_t=0; /* mass of the element: m_t */
139 for (q=0;q<p.numQuadSub;q++) m_t+=Vol[q]*D_p[INDEX2(q, isub,p.numQuadSub) ];
140
141 diagS=0; /* diagonal sum: S */
142 for (s=0;s<p.row_numShapes;s++) {
143 rtmp=0;
144 for (q=0;q<p.numQuadSub;q++) rtmp+=Vol[q]*D_p[INDEX2(q, isub,p.numQuadSub)]*S[INDEX2(s,q,p.row_numShapes)]*S[INDEX2(s,q,p.row_numShapes)];
145 EM_lumpedMat[INDEX2(0,s,p.numEqu)]=rtmp;
146 diagS+=rtmp;
147 }
148 /* rescale diagonals by m_t/diagS to ensure consistent mass over element */
149 rtmp=m_t/diagS;
150 for (s=0;s<p.row_numShapes;s++) EM_lumpedMat[INDEX2(0,s,p.numEqu)]*=rtmp;
151
152 #else /* row-sum lumping */
153 for (s=0;s<p.row_numShapes;s++) {
154 rtmp=0;
155 for (q=0;q<p.numQuadSub;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_numShapes)]*D_p[INDEX2(q, isub,p.numQuadSub)];
156 EM_lumpedMat[INDEX2(0,s,p.numEqu)]+=rtmp;
157 }
158 #endif
159 for (q=0;q<p.row_numShapesTotal;q++) row_index[q]=p.row_DOF[elements->Nodes[INDEX2(p.row_node[INDEX2(q,isub,p.row_numShapesTotal)],e,p.NN)]];
160 Finley_Util_AddScatter(p.row_numShapesTotal,row_index,p.numEqu,EM_lumpedMat,lumpedMat_p, p.row_DOF_UpperBound);
161 } /* end of isub loop */
162 } /* end color check */
163 } /* end element loop */
164 } /* end color loop */
165 } else { /* with constant D */
166 for (color=elements->minColor;color<=elements->maxColor;color++) {
167 /* open loop over all elements: */
168 #pragma omp for private(e) schedule(static)
169 for(e=0;e<elements->numElements;e++){
170 if (elements->Color[e]==color) {
171 for (isub=0; isub<p.numSub; isub++) {
172 Vol=&(p.row_jac->volume[INDEX3(0,isub,e, p.numQuadSub,p.numSub)]);
173 D_p=getSampleDataRO(D,e);
174 #ifdef NEW_LUMPING /* HRZ lumping */
175 m_t=0; /* mass of the element: m_t */
176 for (q=0;q<p.numQuadSub;q++) m_t+=Vol[q];
177
178 diagS=0; /* diagonal sum: S */
179 for (s=0;s<p.row_numShapes;s++) {
180 rtmp=0;
181 for (q=0;q<p.numQuadSub;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_numShapes)]*S[INDEX2(s,q,p.row_numShapes)];
182 EM_lumpedMat[INDEX2(0,s,p.numEqu)]=rtmp;
183 diagS+=rtmp;
184 }
185 /* rescale diagonals by m_t/diagS to ensure consistent mass over element */
186 rtmp=m_t/diagS*D_p[0];
187 for (s=0;s<p.row_numShapes;s++) EM_lumpedMat[INDEX2(0,s,p.numEqu)]*=rtmp;
188
189 #else /* row-sum lumping */
190 for (s=0;s<p.row_numShapes;s++) {
191 rtmp=0;
192 for (q=0;q<p.numQuadSub;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_numShapes)];
193 EM_lumpedMat[INDEX2(0,s,p.numEqu)]=rtmp*D_p[0];
194 }
195 #endif
196 for (q=0;q<p.row_numShapesTotal;q++) row_index[q]=p.row_DOF[elements->Nodes[INDEX2(p.row_node[INDEX2(q,isub,p.row_numShapesTotal)],e,p.NN)]];
197 Finley_Util_AddScatter(p.row_numShapesTotal,row_index,p.numEqu,EM_lumpedMat,lumpedMat_p, p.row_DOF_UpperBound);
198 } /* end of isub loop */
199 } /* end color check */
200 } /* end element loop */
201 } /* end color loop */
202
203 }
204 } else { /* system of equation */
205 if (expandedD) { /* with expanded D */
206 for (color=elements->minColor;color<=elements->maxColor;color++) {
207 /* open loop over all elements: */
208 #pragma omp for private(e) schedule(static)
209 for(e=0;e<elements->numElements;e++){
210 if (elements->Color[e]==color) {
211 for (isub=0; isub<p.numSub; isub++) {
212 Vol=&(p.row_jac->volume[INDEX3(0,isub,e, p.numQuadSub,p.numSub)]);
213 D_p=getSampleDataRO(D,e);
214
215 #ifdef NEW_LUMPING /* HRZ lumping */
216 for (k=0;k<p.numEqu;k++) {
217 m_t=0; /* mass of the element: m_t */
218 for (q=0;q<p.numQuadSub;q++) m_t+=Vol[q]*D_p[INDEX3(k,q,isub,p.numEqu,p.numQuadSub)];
219
220 diagS=0; /* diagonal sum: S */
221 for (s=0;s<p.row_numShapes;s++) {
222 rtmp=0;
223 for (q=0;q<p.numQuadSub;q++) rtmp+=Vol[q]*D_p[INDEX3(k,q,isub,p.numEqu,p.numQuadSub)]*S[INDEX2(s,q,p.row_numShapes)]*S[INDEX2(s,q,p.row_numShapes)];
224 EM_lumpedMat[INDEX2(k,s,p.numEqu)]=rtmp;
225 diagS+=rtmp;
226 }
227 /* rescale diagonals by m_t/diagS to ensure consistent mass over element */
228 rtmp=m_t/diagS;
229 for (s=0;s<p.row_numShapes;s++) EM_lumpedMat[INDEX2(k,s,p.numEqu)]*=rtmp;
230 }
231 #else /* row-sum lumping */
232 for (s=0;s<p.row_numShapes;s++) {
233 for (k=0;k<p.numEqu;k++) {
234 rtmp=0.;
235 for (q=0;q<p.numQuadSub;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_numShapes)]*D_p[INDEX3(k,q,isub,p.numEqu,p.numQuadSub)];
236 EM_lumpedMat[INDEX2(k,s,p.numEqu)]=rtmp;
237 }
238 }
239 #endif
240 for (q=0;q<p.row_numShapesTotal;q++) row_index[q]=p.row_DOF[elements->Nodes[INDEX2(p.row_node[INDEX2(q,isub,p.row_numShapesTotal)],e,p.NN)]];
241 Finley_Util_AddScatter(p.row_numShapesTotal,row_index,p.numEqu,EM_lumpedMat,lumpedMat_p, p.row_DOF_UpperBound);
242 } /* end of isub loop */
243 } /* end color check */
244 } /* end element loop */
245 } /* end color loop */
246 } else { /* with constant D */
247 for (color=elements->minColor;color<=elements->maxColor;color++) {
248 /* open loop over all elements: */
249 #pragma omp for private(e) schedule(static)
250 for(e=0;e<elements->numElements;e++){
251 if (elements->Color[e]==color) {
252 for (isub=0; isub<p.numSub; isub++) {
253 Vol=&(p.row_jac->volume[INDEX3(0,isub,e, p.numQuadSub,p.numSub)]);
254 memset(EM_lumpedMat,0,len_EM_lumpedMat_size);
255 D_p=getSampleDataRO(D,e);
256
257 #ifdef NEW_LUMPING /* HRZ lumping */
258 m_t=0; /* mass of the element: m_t */
259 for (q=0;q<p.numQuadSub;q++) m_t+=Vol[q];
260 diagS=0; /* diagonal sum: S */
261 for (s=0;s<p.row_numShapes;s++) {
262 rtmp=0;
263 for (q=0;q<p.numQuadSub;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_numShapes)]*S[INDEX2(s,q,p.row_numShapes)];
264 for (k=0;k<p.numEqu;k++) EM_lumpedMat[INDEX2(k,s,p.numEqu)]=rtmp;
265 diagS+=rtmp;
266 }
267 /* rescale diagonals by m_t/diagS to ensure consistent mass over element */
268 rtmp=m_t/diagS;
269 for (s=0;s<p.row_numShapes;s++) {
270 for (k=0;k<p.numEqu;k++) EM_lumpedMat[INDEX2(k,s,p.numEqu)]*=rtmp*D_p[k];
271 }
272 #else /* row-sum lumping */
273 for (s=0;s<p.row_numShapes;s++) {
274 for (k=0;k<p.numEqu;k++) {
275 rtmp=0.;
276 for (q=0;q<p.numQuadSub;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_numShapes)];
277 EM_lumpedMat[INDEX2(k,s,p.numEqu)]+=rtmp*D_p[k];
278 }
279 }
280 #endif
281 for (q=0;q<p.row_numShapesTotal;q++) row_index[q]=p.row_DOF[elements->Nodes[INDEX2(p.row_node[INDEX2(q,isub,p.row_numShapesTotal)],e,p.NN)]];
282 Finley_Util_AddScatter(p.row_numShapesTotal,row_index,p.numEqu,EM_lumpedMat,lumpedMat_p, p.row_DOF_UpperBound);
283 } /* end of isub loop */
284 } /* end color check */
285 } /* end element loop */
286 } /* end color loop */
287 }
288 }
289 } /* end of pointer check */
290 THREAD_MEMFREE(EM_lumpedMat);
291 THREAD_MEMFREE(row_index);
292 } /* end parallel region */
293 }
294 }

Properties

Name Value
svn:eol-style native
svn:keywords Author Date Id Revision

  ViewVC Help
Powered by ViewVC 1.1.26