/[escript]/trunk/finley/src/Assemble_PDE_Single2_C.c
ViewVC logotype

Contents of /trunk/finley/src/Assemble_PDE_Single2_C.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 853 - (show annotations)
Wed Sep 20 05:56:36 2006 UTC (13 years, 1 month ago) by gross
File MIME type: text/plain
File size: 7217 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 a single PDEs into the stiffness matrix S right hand side F */
16
17 /* D u_m and Y */
18
19 /* in a 3D domain. The shape functions for test and solution must be identical */
20 /* and 2* row_NS == row_NN (contact elements) */
21
22 /* Shape of the coefficients: */
23
24 /* D = scalar */
25 /* Y = scalar */
26
27
28 /**************************************************************/
29
30 /* Author: gross@access.edu.au */
31 /* Version: $Id:$ */
32
33 /**************************************************************/
34
35
36 #include "Assemble.h"
37 #include "Util.h"
38 #ifdef _OPENMP
39 #include <omp.h>
40 #endif
41
42
43 /**************************************************************/
44
45 void Finley_Assemble_PDE_Single2_C(Assemble_Parameters p, Finley_ElementFile* elements,
46 Paso_SystemMatrix* Mat, escriptDataC* F, escriptDataC* D, escriptDataC* Y) {
47
48 index_t color;
49 dim_t e;
50 double *EM_S, *EM_F, *Vol, *D_p, *Y_p;
51 index_t *row_index;
52 register dim_t q, s,r;
53 register double rtmp, rtmp_D;
54 bool_t add_EM_F, add_EM_S;
55
56 bool_t extendedD=isExpanded(D);
57 bool_t extendedY=isExpanded(Y);
58 double *F_p=getSampleData(F,0);
59 double *S=p.row_jac->ReferenceElement->S;
60
61
62 #pragma omp parallel private(color,EM_S, EM_F, Vol, D_p, Y_p,row_index,q, s,r,rtmp, rtmp_D,add_EM_F, add_EM_S)
63 {
64 EM_S=THREAD_MEMALLOC(p.row_NN*p.col_NN,double);
65 EM_F=THREAD_MEMALLOC(p.row_NN,double);
66 row_index=THREAD_MEMALLOC(p.row_NN,index_t);
67
68 if (!Finley_checkPtr(EM_S) && !Finley_checkPtr(EM_F) && !Finley_checkPtr(row_index) ) {
69
70 #ifndef PASO_MPI
71 for (color=elements->minColor;color<=elements->maxColor;color++) {
72 /* open loop over all elements: */
73 #pragma omp for private(e) schedule(static)
74 for(e=0;e<elements->numElements;e++){
75 if (elements->Color[e]==color) {
76 #else
77 {
78 for(e=0;e<elements->numElements;e++) {
79 {
80 #endif
81 Vol=&(p.row_jac->volume[INDEX2(0,e,p.numQuad)]);
82 add_EM_F=FALSE;
83 add_EM_S=FALSE;
84 /************************************************************* */
85 /* process D */
86 /**************************************************************/
87 D_p=getSampleData(D,e);
88 if (NULL!=D_p) {
89 add_EM_S=TRUE;
90 if (extendedD) {
91 for (s=0;s<p.row_NS;s++) {
92 for (r=0;r<p.col_NS;r++) {
93 rtmp=0;
94 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)];
95 EM_S[INDEX4(0,0,s ,r ,p.numEqu,p.numComp,p.row_NN)]= rtmp;
96 EM_S[INDEX4(0,0,s ,r+p.col_NS,p.numEqu,p.numComp,p.row_NN)]=-rtmp;
97 EM_S[INDEX4(0,0,s+p.row_NS,r ,p.numEqu,p.numComp,p.row_NN)]=-rtmp;
98 EM_S[INDEX4(0,0,s+p.row_NS,r+p.col_NS,p.numEqu,p.numComp,p.row_NN)]= rtmp;
99 }
100 }
101 } else {
102 for (s=0;s<p.row_NS;s++) {
103 for (r=0;r<p.col_NS;r++) {
104 rtmp=0;
105 for (q=0;q<p.numQuad;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_NS)]*S[INDEX2(r,q,p.row_NS)];
106 rtmp_D=rtmp*D_p[0];
107 EM_S[INDEX4(0,0,s ,r ,p.numEqu,p.numComp,p.row_NN)]= rtmp_D;
108 EM_S[INDEX4(0,0,s ,r+p.col_NS,p.numEqu,p.numComp,p.row_NN)]=-rtmp_D;
109 EM_S[INDEX4(0,0,s+p.row_NS,r ,p.numEqu,p.numComp,p.row_NN)]=-rtmp_D;
110 EM_S[INDEX4(0,0,s+p.row_NS,r+p.col_NS,p.numEqu,p.numComp,p.row_NN)]= rtmp_D;
111 }
112 }
113 }
114 }
115 /**************************************************************/
116 /* process Y: */
117 /**************************************************************/
118 Y_p=getSampleData(Y,e);
119 if (NULL!=Y_p) {
120 add_EM_F=TRUE;
121 if (extendedY) {
122 for (s=0;s<p.row_NS;s++) {
123 rtmp=0;
124 for (q=0;q<p.numQuad;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_NS)]*Y_p[q];
125 EM_F[INDEX2(0,s ,p.numEqu)]=-rtmp;
126 EM_F[INDEX2(0,s+p.row_NS,p.numEqu)]= rtmp;
127 }
128 } else {
129 for (s=0;s<p.row_NS;s++) {
130 rtmp=0;
131 for (q=0;q<p.numQuad;q++) rtmp+=Vol[q]*S[INDEX2(s,q,p.row_NS)];
132 rtmp_D=rtmp*Y_p[0];
133 EM_F[INDEX2(0,s ,p.numEqu)]=-rtmp_D;
134 EM_F[INDEX2(0,s+p.row_NS,p.numEqu)]= rtmp_D;
135 }
136 }
137 }
138 /***********************************************************************************************/
139 /* add the element matrices onto the matrix and right hand side */
140 /***********************************************************************************************/
141 for (q=0;q<p.row_NN;q++) row_index[q]=p.row_DOF[elements->Nodes[INDEX2(p.row_node[q],e,p.row_NN)]];
142
143 if (add_EM_F) Finley_Util_AddScatter(p.row_NN,row_index,p.numEqu,EM_F,F_p, p.row_DOF_UpperBound);
144 if (add_EM_S) Finley_Assemble_addToSystemMatrix(Mat,p.row_NN,row_index,p.numEqu,p.col_NN,row_index,p.numComp,EM_S);
145
146 } /* end color check */
147 } /* end element loop */
148 } /* end color loop */
149
150 THREAD_MEMFREE(EM_S);
151 THREAD_MEMFREE(EM_F);
152 THREAD_MEMFREE(row_index);
153
154 } /* end of pointer check */
155 } /* end parallel region */
156 }
157 /*
158 * $Log$
159 */

  ViewVC Help
Powered by ViewVC 1.1.26