/[escript]/trunk/finley/src/Assemble_PDE_Single2_2D.c
ViewVC logotype

Contents of /trunk/finley/src/Assemble_PDE_Single2_2D.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 853 - (show annotations)
Wed Sep 20 05:56:36 2006 UTC (12 years, 9 months ago) by gross
File MIME type: text/plain
File size: 14738 byte(s)
some performance problems wit openmp fixed.
1 /*
2 ************************************************************
3 * Copyright 2006 by ACcESS MNRF *
4 * *
5 * http://www.access.edu.au *
6 * Primary Business: Queensland, Australia *
7 * Licensed under the Open Software License version 3.0 *
8 * http://www.opensource.org/licenses/osl-3.0.php *
9 * *
10 ************************************************************
11 */
12
13 /**************************************************************/
14
15 /* assembles the system of numEq PDEs into the stiffness matrix S right hand side F */
16 /* the shape functions for test and solution must be identical */
17
18
19 /* -(A_{i,j} u_,j)_i-(B_{i} u)_i+C_{j} u_,j-D u_m and -(X_,i)_i + Y */
20
21 /* in a 2D domain. The shape functions for test and solution must be identical */
22 /* and row_NS == row_NN */
23
24 /* Shape of the coefficients: */
25
26 /* A = 2 x 2 */
27 /* B = 2 */
28 /* C = 2 */
29 /* D = scalar */
30 /* X = 2 */
31 /* Y = scalar */
32
33
34 /**************************************************************/
35
36 /* Author: gross@access.edu.au */
37 /* Version: $Id:$ */
38
39 /**************************************************************/
40
41
42 #include "Assemble.h"
43 #include "Util.h"
44 #ifdef _OPENMP
45 #include <omp.h>
46 #endif
47
48
49 /**************************************************************/
50
51 void Finley_Assemble_PDE_Single2_2D(Assemble_Parameters p, Finley_ElementFile* elements,
52 Paso_SystemMatrix* Mat, escriptDataC* F,
53 escriptDataC* A, escriptDataC* B, escriptDataC* C, escriptDataC* D, escriptDataC* X, escriptDataC* Y) {
54
55 #define DIM 2
56 index_t color;
57 dim_t e;
58 double *EM_S, *EM_F, *Vol, *DSDX, *A_p, *B_p, *C_p, *D_p, *X_p, *Y_p;
59 index_t *row_index;
60 register dim_t q, s,r;
61 register double rtmp00, rtmp01, rtmp10, rtmp11, rtmp, rtmp0, rtmp1;
62 bool_t add_EM_F, add_EM_S;
63
64 bool_t extendedA=isExpanded(A);
65 bool_t extendedB=isExpanded(B);
66 bool_t extendedC=isExpanded(C);
67 bool_t extendedD=isExpanded(D);
68 bool_t extendedX=isExpanded(X);
69 bool_t extendedY=isExpanded(Y);
70 double *F_p=getSampleData(F,0);
71 double *S=p.row_jac->ReferenceElement->S;
72 dim_t len_EM_S=p.row_NN*p.col_NN;
73 dim_t len_EM_F=p.row_NN;
74
75
76 #pragma omp parallel private(color,EM_S, EM_F, Vol, DSDX, A_p, B_p, C_p, D_p, X_p, Y_p,row_index,q, s,r,rtmp00, rtmp01, rtmp10, rtmp11, rtmp, rtmp0, rtmp1,add_EM_F, add_EM_S)
77 {
78 EM_S=THREAD_MEMALLOC(len_EM_S,double);
79 EM_F=THREAD_MEMALLOC(len_EM_F,double);
80 row_index=THREAD_MEMALLOC(p.row_NN,index_t);
81
82 if (!Finley_checkPtr(EM_S) && !Finley_checkPtr(EM_F) && !Finley_checkPtr(row_index) ) {
83
84 #ifndef PASO_MPI
85 for (color=elements->minColor;color<=elements->maxColor;color++) {
86 /* open loop over all elements: */
87 #pragma omp for private(e) schedule(static)
88 for(e=0;e<elements->numElements;e++){
89 if (elements->Color[e]==color) {
90 #else
91 {
92 for(e=0;e<elements->numElements;e++) {
93 {
94 #endif
95 Vol=&(p.row_jac->volume[INDEX2(0,e,p.numQuad)]);
96 DSDX=&(p.row_jac->DSDX[INDEX4(0,0,0,e,p.row_NN,DIM,p.numQuad)]);
97 for (q=0;q<len_EM_S;++q) EM_S[q]=0;
98 for (q=0;q<len_EM_F;++q) EM_F[q]=0;
99 add_EM_F=FALSE;
100 add_EM_S=FALSE;
101 /**************************************************************/
102 /* process A: */
103 /**************************************************************/
104 A_p=getSampleData(A,e);
105 if (NULL!=A_p) {
106 add_EM_S=TRUE;
107 if (extendedA) {
108 for (s=0;s<p.row_NS;s++) {
109 for (r=0;r<p.col_NS;r++) {
110 rtmp=0;
111 for (q=0;q<p.numQuad;q++) {
112 rtmp+=Vol[q]*(DSDX[INDEX3(s,0,q,p.row_NN,DIM)]*A_p[INDEX3(0,0,q,DIM,DIM)]*DSDX[INDEX3(r,0,q,p.row_NN,DIM)]
113 + DSDX[INDEX3(s,0,q,p.row_NN,DIM)]*A_p[INDEX3(0,1,q,DIM,DIM)]*DSDX[INDEX3(r,1,q,p.row_NN,DIM)]
114 + DSDX[INDEX3(s,1,q,p.row_NN,DIM)]*A_p[INDEX3(1,0,q,DIM,DIM)]*DSDX[INDEX3(r,0,q,p.row_NN,DIM)]
115 + DSDX[INDEX3(s,1,q,p.row_NN,DIM)]*A_p[INDEX3(1,1,q,DIM,DIM)]*DSDX[INDEX3(r,1,q,p.row_NN,DIM)] );
116 }
117 EM_S[INDEX4(0,0,s,r,p.numEqu,p.numComp,p.row_NN)]+=rtmp;
118 }
119 }
120 } else {
121 for (s=0;s<p.row_NS;s++) {
122 for (r=0;r<p.col_NS;r++) {
123 rtmp00=0;
124 rtmp01=0;
125 rtmp10=0;
126 rtmp11=0;
127 for (q=0;q<p.numQuad;q++) {
128 rtmp0=Vol[q]*DSDX[INDEX3(s,0,q,p.row_NN,DIM)];
129 rtmp1=Vol[q]*DSDX[INDEX3(s,1,q,p.row_NN,DIM)];
130 rtmp00+=rtmp0*DSDX[INDEX3(r,0,q,p.row_NN,DIM)];
131 rtmp01+=rtmp0*DSDX[INDEX3(r,1,q,p.row_NN,DIM)];
132 rtmp10+=rtmp1*DSDX[INDEX3(r,0,q,p.row_NN,DIM)];
133 rtmp11+=rtmp1*DSDX[INDEX3(r,1,q,p.row_NN,DIM)];
134 }
135 EM_S[INDEX4(0,0,s,r,p.numEqu,p.numComp,p.row_NN)]+= rtmp00*A_p[INDEX2(0,0,DIM)]
136 + rtmp01*A_p[INDEX2(0,1,DIM)]
137 + rtmp10*A_p[INDEX2(1,0,DIM)]
138 + rtmp11*A_p[INDEX2(1,1,DIM)];
139 }
140 }
141 }
142 }
143 /**************************************************************/
144 /* process B: */
145 /**************************************************************/
146 B_p=getSampleData(B,e);
147 if (NULL!=B_p) {
148 add_EM_S=TRUE;
149 if (extendedB) {
150 for (s=0;s<p.row_NS;s++) {
151 for (r=0;r<p.col_NS;r++) {
152 rtmp=0.;
153 for (q=0;q<p.numQuad;q++) {
154 rtmp+=Vol[q]*S[INDEX2(r,q,p.row_NS)]*(DSDX[INDEX3(s,0,q,p.row_NN,DIM)]*B_p[INDEX2(0,q,DIM)]
155 + DSDX[INDEX3(s,1,q,p.row_NN,DIM)]*B_p[INDEX2(1,q,DIM)]);
156 }
157 EM_S[INDEX4(0,0,s,r,p.numEqu,p.numComp,p.row_NN)]+=rtmp;
158 }
159 }
160 } else {
161 for (s=0;s<p.row_NS;s++) {
162 for (r=0;r<p.col_NS;r++) {
163 rtmp0=0;
164 rtmp1=0;
165 for (q=0;q<p.numQuad;q++) {
166 rtmp=Vol[q]*S[INDEX2(r,q,p.row_NS)];
167 rtmp0+=rtmp*DSDX[INDEX3(s,0,q,p.row_NN,DIM)];
168 rtmp1+=rtmp*DSDX[INDEX3(s,1,q,p.row_NN,DIM)];
169 }
170 EM_S[INDEX4(0,0,s,r,p.numEqu,p.numComp,p.row_NN)]+=rtmp0*B_p[0]+rtmp1*B_p[1];
171 }
172 }
173 }
174 }
175 /**************************************************************/
176 /* process C: */
177 /**************************************************************/
178 C_p=getSampleData(C,e);
179 if (NULL!=C_p) {
180 add_EM_S=TRUE;
181 if (extendedC) {
182 for (s=0;s<p.row_NS;s++) {
183 for (r=0;r<p.col_NS;r++) {
184 rtmp=0;
185 for (q=0;q<p.numQuad;q++) {
186 rtmp+=Vol[q]*S[INDEX2(s,q,p.row_NS)]*(C_p[INDEX2(0,q,DIM)]*DSDX[INDEX3(r,0,q,p.row_NN,DIM)]
187 + C_p[INDEX2(1,q,DIM)]*DSDX[INDEX3(r,1,q,p.row_NN,DIM)]);
188 }
189 EM_S[INDEX4(0,0,s,r,p.numEqu,p.numComp,p.row_NN)]+=rtmp;
190 }
191 }
192 } else {
193 for (s=0;s<p.row_NS;s++) {
194 for (r=0;r<p.col_NS;r++) {
195 rtmp0=0;
196 rtmp1=0;
197 for (q=0;q<p.numQuad;q++) {
198 rtmp=Vol[q]*S[INDEX2(s,q,p.row_NS)];
199 rtmp0+=rtmp*DSDX[INDEX3(r,0,q,p.row_NN,DIM)];
200 rtmp1+=rtmp*DSDX[INDEX3(r,1,q,p.row_NN,DIM)];
201 }
202 EM_S[INDEX4(0,0,s,r,p.numEqu,p.numComp,p.row_NN)]+=rtmp0*C_p[0]+rtmp1*C_p[1];
203 }
204 }
205 }
206 }
207 /************************************************************* */
208 /* process D */
209 /**************************************************************/
210 D_p=getSampleData(D,e);
211 if (NULL!=D_p) {
212 add_EM_S=TRUE;
213 if (extendedD) {
214 for (s=0;s<p.row_NS;s++) {
215 for (r=0;r<p.col_NS;r++) {
216 rtmp=0;
217 for (q=0;q<p.numQuad;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_NS)]*D_p[q]*S[INDEX2(r,q,p.row_NS)];
218 EM_S[INDEX4(0,0,s,r,p.numEqu,p.numComp,p.row_NN)]+=rtmp;
219 }
220 }
221 } else {
222 for (s=0;s<p.row_NS;s++) {
223 for (r=0;r<p.col_NS;r++) {
224 rtmp=0;
225 for (q=0;q<p.numQuad;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_NS)]*S[INDEX2(r,q,p.row_NS)];
226 EM_S[INDEX4(0,0,s,r,p.numEqu,p.numComp,p.row_NN)]+=rtmp*D_p[0];
227 }
228 }
229 }
230 }
231 /**************************************************************/
232 /* process X: */
233 /**************************************************************/
234 X_p=getSampleData(X,e);
235 if (NULL!=X_p) {
236 add_EM_F=TRUE;
237 if (extendedX) {
238 for (s=0;s<p.row_NS;s++) {
239 rtmp=0.;
240 for (q=0;q<p.numQuad;q++) {
241 rtmp+=Vol[q]*( DSDX[INDEX3(s,0,q,p.row_NN,DIM)]*X_p[INDEX2(0,q,DIM)]
242 + DSDX[INDEX3(s,1,q,p.row_NN,DIM)]*X_p[INDEX2(1,q,DIM)]);
243 }
244 EM_F[INDEX2(0,s,p.numEqu)]+=rtmp;
245 }
246 } else {
247 for (s=0;s<p.row_NS;s++) {
248 rtmp0=0.;
249 rtmp1=0.;
250 for (q=0;q<p.numQuad;q++) {
251 rtmp0+=Vol[q]*DSDX[INDEX3(s,0,q,p.row_NN,DIM)];
252 rtmp1+=Vol[q]*DSDX[INDEX3(s,1,q,p.row_NN,DIM)];
253 }
254 EM_F[INDEX2(0,s,p.numEqu)]+=rtmp0*X_p[0]+rtmp1*X_p[1];
255 }
256 }
257 }
258 /**************************************************************/
259 /* process Y: */
260 /**************************************************************/
261 Y_p=getSampleData(Y,e);
262 if (NULL!=Y_p) {
263 add_EM_F=TRUE;
264 if (extendedY) {
265 for (s=0;s<p.row_NS;s++) {
266 rtmp=0;
267 for (q=0;q<p.numQuad;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_NS)]*Y_p[q];
268 EM_F[INDEX2(0,s,p.numEqu)]+=rtmp;
269 }
270 } else {
271 for (s=0;s<p.row_NS;s++) {
272 rtmp=0;
273 for (q=0;q<p.numQuad;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_NS)];
274 EM_F[INDEX2(0,s,p.numEqu)]+=rtmp*Y_p[0];
275 }
276 }
277 }
278 /***********************************************************************************************/
279 /* add the element matrices onto the matrix and right hand side */
280 /***********************************************************************************************/
281 for (q=0;q<p.row_NN;q++) row_index[q]=p.row_DOF[elements->Nodes[INDEX2(p.row_node[q],e,p.NN)]];
282 if (add_EM_F) Finley_Util_AddScatter(p.row_NN,row_index,p.numEqu,EM_F,F_p, p.row_DOF_UpperBound);
283 if (add_EM_S) Finley_Assemble_addToSystemMatrix(Mat,p.row_NN,row_index,p.numEqu,p.col_NN,row_index,p.numComp,EM_S);
284
285 } /* end color check */
286 } /* end element loop */
287 } /* end color loop */
288
289 THREAD_MEMFREE(EM_S);
290 THREAD_MEMFREE(EM_F);
291 THREAD_MEMFREE(row_index);
292
293 } /* end of pointer check */
294 } /* end parallel region */
295 }
296 /*
297 * $Log$
298 */

  ViewVC Help
Powered by ViewVC 1.1.26