/[escript]/branches/intelc_win32/finley/src/Util.c
ViewVC logotype

Contents of /branches/intelc_win32/finley/src/Util.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 753 - (show annotations)
Mon Jun 26 02:51:25 2006 UTC (14 years, 5 months ago) by woo409
File MIME type: text/plain
File size: 16175 byte(s)
+ Made qsortG usage conditional on USE_QSORTG being defined

Note both altix and win32 fail the test_normal_onFunctionContactOne/Zero tests when using QSORTG.

Perhaps QSORTG isn't stable or altix isn't or I've introduced some other bug...I'll try altix without qsortG and see what happens.
This is becoming painful.
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 /* Some utility routines: */
16
17 /**************************************************************/
18
19 /* author: gross@access.edu.au */
20 /* Version: $Id$ */
21
22 /**************************************************************/
23
24 #include "Finley.h"
25 #include "Util.h"
26 #ifdef USE_QSORTG
27 #include "qsortG.h"
28 #endif
29
30 #ifdef _OPENMP
31 #include <omp.h>
32 #endif
33
34 /**************************************************************/
35
36 /* returns true if any of the values in the short array values is not equalt to Zero */
37
38 bool_t Finley_Util_anyNonZeroDouble(dim_t N, double* values) {
39 dim_t q;
40 for (q=0;q<N;++q) if (ABS(values[q])>0) return TRUE;
41 return FALSE;
42 }
43 /**************************************************************/
44
45 /* gathers double values out from in by index: */
46
47 /* out(1:numData,1:len)=in(1:numData,index(1:len)) */
48
49 void Finley_Util_Gather_double(dim_t len,index_t* index,dim_t numData,double* in, double * out){
50 dim_t s,i;
51 for (s=0;s<len;s++) {
52 for (i=0;i<numData;i++) {
53 out[INDEX2(i,s,numData)]=in[INDEX2(i,index[s],numData)];
54 }
55 }
56 }
57
58 /**************************************************************/
59
60
61 /* gathers maybelong values out from in by index: */
62
63 /* out(1:numData,1:len)=in(1:numData,index(1:len)) */
64
65 void Finley_Util_Gather_int(dim_t len,index_t* index,dim_t numData, index_t* in, index_t * out){
66 dim_t s,i;
67 for (s=0;s<len;s++) {
68 for (i=0;i<numData;i++) {
69 out[INDEX2(i,s,numData)]=in[INDEX2(i,index[s],numData)];
70 }
71 }
72 }
73
74 /**************************************************************/
75
76 /* adds a vector in into out using and index. */
77
78 /* out(1:numData,index(1:len))+=in(1:numData,1:len) */
79
80 void Finley_Util_AddScatter(dim_t len,index_t* index,dim_t numData,double* in,double * out){
81 dim_t i,s;
82 for (s=0;s<len;s++) {
83 for(i=0;i<numData;i++) {
84 #pragma omp atomic
85 out[INDEX2(i,index[s],numData)]+=in[INDEX2(i,s,numData)];
86 }
87 }
88 }
89
90 /* multiplies two matrices */
91
92 /* A(1:A1,1:A2)=B(1:A1,1:B2)*C(1:B2,1:A2) */
93
94 void Finley_Util_SmallMatMult(dim_t A1,dim_t A2, double* A, dim_t B2, double*B, double* C) {
95 dim_t i,j,s;
96 for (i=0;i<A1*A2;i++) A[i]=0;
97 for (i=0;i<A1;i++) {
98 for (j=0;j<A2;j++) {
99 for (s=0;s<B2;s++) {
100 A[INDEX2(i,j,A1)]+=B[INDEX2(i,s,A1)]*C[INDEX2(s,j,B2)];
101 }
102 }
103 }
104 }
105
106 /* multiplies a two sets of matries: */
107
108 /* A(1:A1,1:A2,i)=B(1:A1,1:B2,i)*C(1:B2,1:A2,i) i=1,len */
109
110 void Finley_Util_SmallMatSetMult(dim_t len,dim_t A1,dim_t A2, double* A, dim_t B2, double*B, double* C) {
111 dim_t q,i,j,s;
112 for (i=0;i<A1*A2*len;i++) A[i]=0;
113 for (q=0;q<len;q++) {
114 for (i=0;i<A1;i++) {
115 for (j=0;j<A2;j++) {
116 for (s=0;s<B2;s++) {
117 A[INDEX3(i,j,q,A1,A2)]+=B[INDEX3(i,s,q,A1,B2)]*C[INDEX3(s,j,q,B2,A2)];
118 }
119 }
120 }
121 }
122 }
123 /* inverts the set of dim x dim matrices A(:,:,1:len) with dim=1,2,3 */
124 /* the determinante is returned. */
125
126 void Finley_Util_InvertSmallMat(dim_t len,dim_t dim,double* A,double *invA, double* det){
127 dim_t q;
128 register double D,A11,A12,A13,A21,A22,A23,A31,A32,A33;
129
130 switch(dim) {
131 case 1:
132 for (q=0;q<len;q++) {
133 D=A[q];
134 if (ABS(D) > 0 ){
135 det[q]=D;
136 D=1./D;
137 invA[q]=D;
138 } else {
139 Finley_setError(ZERO_DIVISION_ERROR,"__FILE__: Non-regular matrix");
140 return;
141 }
142 }
143 break;
144
145 case 2:
146 for (q=0;q<len;q++) {
147 A11=A[INDEX3(0,0,q,2,2)];
148 A12=A[INDEX3(0,1,q,2,2)];
149 A21=A[INDEX3(1,0,q,2,2)];
150 A22=A[INDEX3(1,1,q,2,2)];
151
152 D = A11*A22-A12*A21;
153 if (ABS(D) > 0 ){
154 det[q]=D;
155 D=1./D;
156 invA[INDEX3(0,0,q,2,2)]= A22*D;
157 invA[INDEX3(1,0,q,2,2)]=-A21*D;
158 invA[INDEX3(0,1,q,2,2)]=-A12*D;
159 invA[INDEX3(1,1,q,2,2)]= A11*D;
160 } else {
161 Finley_setError(ZERO_DIVISION_ERROR,"__FILE__: Non-regular matrix");
162 return;
163 }
164 }
165 break;
166
167 case 3:
168 for (q=0;q<len;q++) {
169 A11=A[INDEX3(0,0,q,3,3)];
170 A21=A[INDEX3(1,0,q,3,3)];
171 A31=A[INDEX3(2,0,q,3,3)];
172 A12=A[INDEX3(0,1,q,3,3)];
173 A22=A[INDEX3(1,1,q,3,3)];
174 A32=A[INDEX3(2,1,q,3,3)];
175 A13=A[INDEX3(0,2,q,3,3)];
176 A23=A[INDEX3(1,2,q,3,3)];
177 A33=A[INDEX3(2,2,q,3,3)];
178
179 D = A11*(A22*A33-A23*A32)+ A12*(A31*A23-A21*A33)+A13*(A21*A32-A31*A22);
180 if (ABS(D) > 0 ){
181 det[q] =D;
182 D=1./D;
183 invA[INDEX3(0,0,q,3,3)]=(A22*A33-A23*A32)*D;
184 invA[INDEX3(1,0,q,3,3)]=(A31*A23-A21*A33)*D;
185 invA[INDEX3(2,0,q,3,3)]=(A21*A32-A31*A22)*D;
186 invA[INDEX3(0,1,q,3,3)]=(A13*A32-A12*A33)*D;
187 invA[INDEX3(1,1,q,3,3)]=(A11*A33-A31*A13)*D;
188 invA[INDEX3(2,1,q,3,3)]=(A12*A31-A11*A32)*D;
189 invA[INDEX3(0,2,q,3,3)]=(A12*A23-A13*A22)*D;
190 invA[INDEX3(1,2,q,3,3)]=(A13*A21-A11*A23)*D;
191 invA[INDEX3(2,2,q,3,3)]=(A11*A22-A12*A21)*D;
192 } else {
193 Finley_setError(ZERO_DIVISION_ERROR,"__FILE__: Non-regular matrix");
194 return;
195 }
196 }
197 break;
198
199 }
200 return;
201 }
202
203 /* sets the derterminate of a set of dim x dim matrices A(:,:,1:len) with dim=1,2,3 */
204
205 void Finley_Util_DetOfSmallMat(dim_t len,dim_t dim,double* A, double* det){
206 dim_t q;
207 register double A11,A12,A13,A21,A22,A23,A31,A32,A33;
208
209 switch(dim) {
210 case 1:
211 for (q=0;q<len;q++) {
212 det[q]=A[q];
213 }
214 break;
215
216 case 2:
217 for (q=0;q<len;q++) {
218 A11=A[INDEX3(0,0,q,2,2)];
219 A12=A[INDEX3(0,1,q,2,2)];
220 A21=A[INDEX3(1,0,q,2,2)];
221 A22=A[INDEX3(1,1,q,2,2)];
222
223 det[q] = A11*A22-A12*A21;
224 }
225 break;
226
227 case 3:
228 for (q=0;q<len;q++) {
229 A11=A[INDEX3(0,0,q,3,3)];
230 A21=A[INDEX3(1,0,q,3,3)];
231 A31=A[INDEX3(2,0,q,3,3)];
232 A12=A[INDEX3(0,1,q,3,3)];
233 A22=A[INDEX3(1,1,q,3,3)];
234 A32=A[INDEX3(2,1,q,3,3)];
235 A13=A[INDEX3(0,2,q,3,3)];
236 A23=A[INDEX3(1,2,q,3,3)];
237 A33=A[INDEX3(2,2,q,3,3)];
238
239 det[q] = A11*(A22*A33-A23*A32)+ A12*(A31*A23-A21*A33)+A13*(A21*A32-A31*A22);
240 }
241 break;
242
243 }
244 return;
245 }
246 /* returns the normalized vector Normal[dim,len] orthogonal to A(:,0,q) and A(:,1,q) in the case of dim=3 */
247 /* or the vector A(:,0,q) in the case of dim=2 */
248
249 void Finley_NormalVector(dim_t len, dim_t dim, dim_t dim1, double* A,double* Normal) {
250 dim_t q;
251 register double A11,A12,CO_A13,A21,A22,CO_A23,A31,A32,CO_A33,length,invlength;
252
253 switch(dim) {
254 case 1:
255 for (q=0;q<len;q++) Normal[q] =1;
256 break;
257 case 2:
258 for (q=0;q<len;q++) {
259 A11=A[INDEX3(0,0,q,2,dim1)];
260 A21=A[INDEX3(1,0,q,2,dim1)];
261 length = sqrt(A11*A11+A21*A21);
262 if (! length>0) {
263 Finley_setError(ZERO_DIVISION_ERROR,"__FILE__: area equals zero.");
264 return;
265 } else {
266 invlength=1./length;
267 Normal[INDEX2(0,q,2)]=A21*invlength;
268 Normal[INDEX2(1,q,2)]=-A11*invlength;
269 }
270 }
271 break;
272 case 3:
273 for (q=0;q<len;q++) {
274 A11=A[INDEX3(0,0,q,3,dim1)];
275 A21=A[INDEX3(1,0,q,3,dim1)];
276 A31=A[INDEX3(2,0,q,3,dim1)];
277 A12=A[INDEX3(0,1,q,3,dim1)];
278 A22=A[INDEX3(1,1,q,3,dim1)];
279 A32=A[INDEX3(2,1,q,3,dim1)];
280 CO_A13=A21*A32-A31*A22;
281 CO_A23=A31*A12-A11*A32;
282 CO_A33=A11*A22-A21*A12;
283 length=sqrt(CO_A13*CO_A13+CO_A23*CO_A23+CO_A33*CO_A33);
284 if (! length>0) {
285 Finley_setError(ZERO_DIVISION_ERROR,"__FILE__: area equals zero.");
286 return;
287 } else {
288 invlength=1./length;
289 Normal[INDEX2(0,q,3)]=CO_A13*invlength;
290 Normal[INDEX2(1,q,3)]=CO_A23*invlength;
291 Normal[INDEX2(2,q,3)]=CO_A33*invlength;
292 }
293
294 }
295 break;
296
297 }
298 return;
299 }
300
301 /* return the length of the vector which is orthogonal to the vectors A(:,0,q) and A(:,1,q) in the case of dim=3 */
302 /* or the vector A(:,0,q) in the case of dim=2 */
303
304 void Finley_LengthOfNormalVector(dim_t len, dim_t dim, dim_t dim1, double* A,double* length) {
305 dim_t q;
306 double A11,A12,CO_A13,A21,A22,CO_A23,A31,A32,CO_A33;
307
308 switch(dim) {
309 case 1:
310 for (q=0;q<len;q++) length[q] =1;
311 break;
312 case 2:
313 for (q=0;q<len;q++) {
314 A11=A[INDEX3(0,0,q,2,dim1)];
315 A21=A[INDEX3(1,0,q,2,dim1)];
316 length[q] = sqrt(A11*A11+A21*A21);
317 }
318 break;
319 case 3:
320 for (q=0;q<len;q++) {
321 A11=A[INDEX3(0,0,q,3,dim1)];
322 A21=A[INDEX3(1,0,q,3,dim1)];
323 A31=A[INDEX3(2,0,q,3,dim1)];
324 A12=A[INDEX3(0,1,q,3,dim1)];
325 A22=A[INDEX3(1,1,q,3,dim1)];
326 A32=A[INDEX3(2,1,q,3,dim1)];
327 CO_A13=A21*A32-A31*A22;
328 CO_A23=A31*A12-A11*A32;
329 CO_A33=A11*A22-A21*A12;
330 length[q]=sqrt(CO_A13*CO_A13+CO_A23*CO_A23+CO_A33*CO_A33);
331 }
332 break;
333
334 }
335 return;
336 }
337
338 /* inverts the map map of length len */
339 /* there is no range checking! */
340 /* at output Map[invMap[i]]=i for i=0:lenInvMap */
341
342 void Finley_Util_InvertMap(dim_t lenInvMap, index_t* invMap,dim_t lenMap, index_t* Map) {
343 dim_t i;
344 for (i=0;i<lenInvMap;i++) invMap[i]=0;
345 for (i=0;i<lenMap;i++) {
346 if (Map[i]>=0) invMap[Map[i]]=i;
347 }
348 }
349
350 /* orders a Finley_Util_ValueAndIndex array by value */
351 /* it is assumed that n is large */
352
353 int Finley_Util_ValueAndIndex_compar(const void *arg1 , const void *arg2 ) {
354 Finley_Util_ValueAndIndex *e1,*e2;
355 e1=(Finley_Util_ValueAndIndex*) arg1;
356 e2=(Finley_Util_ValueAndIndex*) arg2;
357 if (e1->value < e2->value) return -1;
358 if (e1->value > e2->value) return 1;
359 return 0;
360 }
361
362 void Finley_Util_sortValueAndIndex(dim_t n,Finley_Util_ValueAndIndex* array) {
363 /* OMP : needs parallelization !*/
364 #ifdef USE_QSORTG
365 qsortG(array,n,sizeof(Finley_Util_ValueAndIndex),Finley_Util_ValueAndIndex_compar);
366 #else
367 qsort(array,n,sizeof(Finley_Util_ValueAndIndex),Finley_Util_ValueAndIndex_compar);
368 #endif
369 }
370
371
372 /**************************************************************/
373
374 /* calculates the minimum value from a dim X N integer array */
375
376 index_t Finley_Util_getMinInt(dim_t dim,dim_t N,index_t* values) {
377 dim_t i,j;
378 index_t out,out_local;
379 out=INDEX_T_MAX;
380 if (values!=NULL && dim*N>0 ) {
381 out=values[0];
382 #pragma omp parallel private(out_local)
383 {
384 out_local=out;
385 #pragma omp for private(i,j) schedule(static)
386 for (j=0;j<N;j++) {
387 for (i=0;i<dim;i++) out_local=MIN(out_local,values[INDEX2(i,j,dim)]);
388 }
389 #pragma omp critical
390 out=MIN(out_local,out);
391 }
392 }
393 return out;
394 }
395
396 /* calculates the maximum value from a dim X N integer array */
397
398 index_t Finley_Util_getMaxInt(dim_t dim,dim_t N,index_t* values) {
399 dim_t i,j;
400 index_t out,out_local;
401 out=-INDEX_T_MAX;
402 if (values!=NULL && dim*N>0 ) {
403 out=values[0];
404 #pragma omp parallel private(out_local)
405 {
406 out_local=out;
407 #pragma omp for private(i,j) schedule(static)
408 for (j=0;j<N;j++) {
409 for (i=0;i<dim;i++) out_local=MAX(out_local,values[INDEX2(i,j,dim)]);
410 }
411 #pragma omp critical
412 out=MAX(out_local,out);
413 }
414 }
415 return out;
416 }
417
418 /* set the index of the positive entries in mask. The length of index is returned. */
419
420 dim_t Finley_Util_packMask(dim_t N,index_t* mask,index_t* index) {
421 dim_t out,k;
422 out=0;
423 /*OMP */
424 for (k=0;k<N;k++) {
425 if (mask[k]>=0) {
426 index[out]=k;
427 out++;
428 }
429 }
430 return out;
431 }
432
433 /* returns true if array contains value */
434 bool_t Finley_Util_isAny(dim_t N,index_t* array,index_t value) {
435 bool_t out=FALSE;
436 dim_t i;
437 #pragma omp parallel for private(i) schedule(static) reduction(||:out)
438 for (i=0;i<N;i++) out = out || (array[i]==value);
439 return out;
440 }
441 /* calculates the cummultative sum in array and returns the total sum */
442 index_t Finley_Util_cumsum(dim_t N,index_t* array) {
443 index_t out=0,tmp;
444 dim_t i;
445 #ifdef _OPENMP
446 index_t partial_sums[omp_get_max_threads()],sum;
447 #pragma omp parallel private(sum,i,tmp)
448 {
449 sum=0;
450 #pragma omp for schedule(static)
451 for (i=0;i<N;++i) sum+=array[i];
452 partial_sums[omp_get_thread_num()]=sum;
453 #pragma omp barrier
454 #pragma omp master
455 {
456 out=0;
457 for (i=0;i<omp_get_max_threads();++i) {
458 tmp=out;
459 out+=partial_sums[i];
460 partial_sums[i]=tmp;
461 }
462 }
463 #pragma omp barrier
464 sum=partial_sums[omp_get_thread_num()];
465 #pragma omp for schedule(static)
466 for (i=0;i<N;++i) {
467 tmp=sum;
468 sum+=array[i];
469 array[i]=tmp;
470 }
471 }
472 #else
473 for (i=0;i<N;++i) {
474 tmp=out;
475 out+=array[i];
476 array[i]=tmp;
477 }
478 #endif
479 return out;
480 }
481
482 void Finley_copyDouble(dim_t n,double* source, double* target) {
483 dim_t i;
484 for (i=0;i<n;i++) target[i]=source[i];
485 }
486
487 /*
488 * Revision 1.8 2005/08/12 01:45:43 jgs
489 * erge of development branch dev-02 back to main trunk on 2005-08-12
490 *
491 * Revision 1.7.2.2 2005/09/07 06:26:22 gross
492 * the solver from finley are put into the standalone package paso now
493 *
494 * Revision 1.7.2.1 2005/08/04 22:41:11 gross
495 * some extra routines for finley that might speed-up RHS assembling in some cases (not actived right now)
496 *
497 * Revision 1.7 2005/07/08 04:07:59 jgs
498 * Merge of development branch back to main trunk on 2005-07-08
499 *
500 * Revision 1.1.1.1.2.4 2005/06/29 02:34:57 gross
501 * some changes towards 64 integers in finley
502 *
503 * Revision 1.1.1.1.2.3 2005/03/02 23:35:06 gross
504 * reimplementation of the ILU in Finley. block size>1 still needs some testing
505 *
506 * Revision 1.1.1.1.2.2 2005/02/18 02:27:31 gross
507 * two function that will be used for a reimplementation of the ILU preconditioner
508 *
509 * Revision 1.1.1.1.2.1 2004/11/12 06:58:19 gross
510 * a lot of changes to get the linearPDE class running: most important change is that there is no matrix format exposed to the user anymore. the format is chosen by the Domain according to the solver and symmetry
511 *
512 * Revision 1.1.1.1 2004/10/26 06:53:57 jgs
513 * initial import of project esys2
514 *
515 * Revision 1.3 2004/08/26 12:03:52 gross
516 * Some other bug in Finley_Assemble_gradient fixed.
517 *
518 * Revision 1.2 2004/07/02 04:21:13 gross
519 * Finley C code has been included
520 *
521 * Revision 1.1.1.1 2004/06/24 04:00:40 johng
522 * Initial version of eys using boost-python.
523 *
524 *
525 */

Properties

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

  ViewVC Help
Powered by ViewVC 1.1.26