/[escript]/trunk/paso/src/Smoother.c
ViewVC logotype

Annotation of /trunk/paso/src/Smoother.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 3158 - (hide annotations)
Mon Sep 6 06:09:11 2010 UTC (8 years, 7 months ago) by gross
File MIME type: text/plain
File size: 14710 byte(s)
GS and Jacobi are now used through the same interface.
1 jfenwick 1851
2     /*******************************************************
3     *
4 jfenwick 2881 * Copyright (c) 2003-2010 by University of Queensland
5 jfenwick 1851 * 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 gross 3097 /* Paso: Gauss-Seidel */
18 jfenwick 1851
19     /**************************************************************/
20    
21     /* Author: artak@uq.edu.au */
22    
23     /**************************************************************/
24    
25     #include "Paso.h"
26 gross 3120 #include "Preconditioner.h"
27 jfenwick 1851 #include "PasoUtil.h"
28 gross 3097 #include "BlockOps.h"
29 jfenwick 1851
30     #include <stdio.h>
31    
32 gross 3094
33 jfenwick 1851 /**************************************************************/
34    
35 gross 3158 /* free all memory used by Smoother */
36 jfenwick 1851
37 gross 3158 void Paso_Preconditioner_Smoother_free(Paso_Preconditioner_Smoother * in) {
38 jfenwick 1851 if (in!=NULL) {
39 gross 3158 Paso_Preconditioner_LocalSmoother_free(in->localSmoother);
40 jfenwick 1851 MEMFREE(in);
41     }
42     }
43 gross 3158 void Paso_Preconditioner_LocalSmoother_free(Paso_Preconditioner_LocalSmoother * in) {
44 gross 3094 if (in!=NULL) {
45     MEMFREE(in->diag);
46     MEMFREE(in->pivot);
47 gross 3097 MEMFREE(in->buffer);
48 gross 3094 MEMFREE(in);
49     }
50     }
51 jfenwick 1851 /**************************************************************/
52    
53 gross 3094 /* constructs the symmetric Gauss-Seidel preconditioner
54 jfenwick 1851
55     */
56 gross 3158 Paso_Preconditioner_Smoother* Paso_Preconditioner_Smoother_alloc(Paso_SystemMatrix * A_p, const bool_t jacobi, const bool_t is_local, const bool_t verbose)
57 gross 3094 {
58    
59 jfenwick 1851 /* allocations: */
60 gross 3158 Paso_Preconditioner_Smoother* out=MEMALLOC(1,Paso_Preconditioner_Smoother);
61 gross 3094 if (! Paso_checkPtr(out)) {
62 gross 3158 out->localSmoother=Paso_Preconditioner_LocalSmoother_alloc(A_p->mainBlock,jacobi,verbose);
63 gross 3094 out->is_local=is_local;
64 jfenwick 1851 }
65 gross 3094 if (Paso_MPIInfo_noError(A_p->mpi_info)) {
66 jfenwick 1851 return out;
67 gross 3094 } else {
68 gross 3158 Paso_Preconditioner_Smoother_free(out);
69 jfenwick 1851 return NULL;
70     }
71     }
72 gross 3158 Paso_Preconditioner_LocalSmoother* Paso_Preconditioner_LocalSmoother_alloc(Paso_SparseMatrix * A_p, const bool_t jacobi, bool_t verbose)
73     {
74 gross 3094
75     dim_t n=A_p->numRows;
76     dim_t n_block=A_p->row_block_size;
77     dim_t block_size=A_p->block_size;
78    
79     double time0=Paso_timer();
80     /* allocations: */
81 gross 3158 Paso_Preconditioner_LocalSmoother* out=MEMALLOC(1,Paso_Preconditioner_LocalSmoother);
82 gross 3094 if (! Paso_checkPtr(out)) {
83    
84     out->diag=MEMALLOC( ((size_t) n) * ((size_t) block_size),double);
85     out->pivot=MEMALLOC( ((size_t) n) * ((size_t) n_block), index_t);
86 gross 3097 out->buffer=MEMALLOC( ((size_t) n) * ((size_t) n_block), double);
87 gross 3158 out->Jacobi=jacobi;
88 gross 3094
89     if ( ! ( Paso_checkPtr(out->diag) || Paso_checkPtr(out->pivot) ) ) {
90     Paso_SparseMatrix_invMain(A_p, out->diag, out->pivot );
91     }
92    
93     }
94     time0=Paso_timer()-time0;
95    
96     if (Paso_noError()) {
97 gross 3158 if (verbose) {
98     if (jacobi) {
99     printf("timing: Jacobi preparation: elemination : %e\n",time0);
100     } else {
101     printf("timing: Gauss-Seidel preparation: elemination : %e\n",time0);
102     }
103     }
104 gross 3094 return out;
105     } else {
106 gross 3158 Paso_Preconditioner_LocalSmoother_free(out);
107 gross 3094 return NULL;
108     }
109     }
110 jfenwick 1851
111 gross 3097 /*
112 jfenwick 1851
113 gross 3097 performs a few sweeps of the from
114 jfenwick 1851
115 gross 3097 S (x_{k} - x_{k-1}) = b - A x_{k-1}
116 jfenwick 1851
117 gross 3097 where x_{0}=0 and S provides some approximatioon of A.
118 jfenwick 1851
119 gross 3097 Under MPI S is build on using A_p->mainBlock only.
120 gross 3158 if Smoother is local the defect b - A x_{k-1} is calculated using A_p->mainBlock only.
121 gross 3094
122 jfenwick 1851 */
123    
124 gross 3158 void Paso_Preconditioner_Smoother_solve(Paso_SystemMatrix* A_p, Paso_Preconditioner_Smoother * smoother, double * x, const double * b, const dim_t sweeps)
125 gross 3094 {
126 gross 3097 const dim_t n= (A_p->mainBlock->numRows) * (A_p->mainBlock->row_block_size);
127    
128 gross 3158 double *b_new = smoother->localSmoother->buffer;
129     dim_t nsweeps=sweeps;
130     if (smoother->is_local) {
131     Paso_Preconditioner_LocalSmoother_solve(A_p->mainBlock,smoother->localSmoother,x,b,sweeps);
132 gross 3097 } else {
133 gross 3158 Paso_Copy(n, x, b);
134 gross 3097
135 gross 3158 Paso_Preconditioner_LocalSmoother_Sweep(A_p->mainBlock,smoother->localSmoother,x);
136 gross 3105
137 gross 3158 while (nsweeps > 1 ) {
138     Paso_Copy(n, b_new, b);
139 gross 3097
140     Paso_SystemMatrix_MatrixVector((-1.), A_p, x, 1., b_new); /* b_new = b - A*x */
141    
142 gross 3158 Paso_Preconditioner_LocalSmoother_Sweep(A_p->mainBlock,smoother->localSmoother,b_new);
143 gross 3097
144 gross 3158 Paso_AXPY(n, x, 1., b_new);
145     nsweeps--;
146 gross 3097 }
147    
148 gross 3094 }
149 gross 3097 }
150 gross 3158 void Paso_Preconditioner_LocalSmoother_solve(Paso_SparseMatrix* A_p, Paso_Preconditioner_LocalSmoother * smoother, double * x, const double * b, const dim_t sweeps)
151 gross 3097 {
152     const dim_t n= (A_p->numRows) * (A_p->row_block_size);
153 gross 3158 double *b_new = smoother->buffer;
154     dim_t nsweeps=sweeps;
155 gross 3094
156 gross 3158 Paso_Copy(n, x, b);
157     Paso_Preconditioner_LocalSmoother_Sweep(A_p,smoother,x);
158 gross 3105
159 gross 3158 while (nsweeps > 1 ) {
160    
161     Paso_Copy(n, b_new, b);
162 gross 3094
163 gross 3097 Paso_SparseMatrix_MatrixVector_CSC_OFFSET0((-1.), A_p, x, 1., b_new); /* b_new = b - A*x */
164 gross 3158
165     Paso_Preconditioner_LocalSmoother_Sweep(A_p,smoother,b_new);
166    
167     Paso_AXPY(n, x, 1., b_new)
168 gross 3097
169 gross 3158 nsweeps--;
170 gross 3094 }
171     }
172 jfenwick 1851
173 gross 3158 void Paso_Preconditioner_LocalSmoother_Sweep(Paso_SparseMatrix* A, Paso_Preconditioner_LocalSmoother * smoother, double * x)
174 gross 3094 {
175     #ifdef _OPENMP
176     const dim_t nt=omp_get_max_threads();
177     #else
178     const dim_t nt = 1;
179     #endif
180 gross 3158 if (smoother->Jacobi) {
181     Paso_BlockOps_allMV(A->row_block_size,A->numRows,smoother->diag,smoother->pivot,x);
182 gross 3097 } else {
183 gross 3158 if (nt < 2) {
184     Paso_Preconditioner_LocalSmoother_Sweep_sequential(A,smoother,x);
185     } else {
186     Paso_Preconditioner_LocalSmoother_Sweep_colored(A,smoother,x);
187     }
188 gross 3094 }
189 jfenwick 1851 }
190    
191 gross 3097 /* inplace Gauss-Seidel sweep in seqential mode: */
192 gross 3094
193 gross 3158 void Paso_Preconditioner_LocalSmoother_Sweep_sequential(Paso_SparseMatrix* A_p, Paso_Preconditioner_LocalSmoother * smoother, double * x)
194 gross 3097 {
195     const dim_t n=A_p->numRows;
196     const dim_t n_block=A_p->row_block_size;
197 gross 3158 const double *diag = smoother->diag;
198     /* const index_t* pivot = smoother->pivot;
199 gross 3097 const dim_t block_size=A_p->block_size; use for block size >3*/
200 gross 3094
201 gross 3097 register dim_t i,k;
202     register index_t iptr_ik, mm;
203    
204     const index_t* ptr_main = Paso_SparseMatrix_borrowMainDiagonalPointer(A_p);
205     /* forward substitution */
206    
207     if (n_block==1) {
208 gross 3098 Paso_BlockOps_MV_1(&x[0], &diag[0], &x[0]);
209     for (i = 1; i < n; ++i) {
210 gross 3097 mm=ptr_main[i];
211 gross 3105 /* x_i=x_i-a_ik*x_k (with k<i) */
212 gross 3097 for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<mm; ++iptr_ik) {
213 gross 3098 k=A_p->pattern->index[iptr_ik];
214 gross 3097 Paso_BlockOps_SMV_1(&x[i], &A_p->val[iptr_ik], &x[k]);
215     }
216 gross 3098 Paso_BlockOps_MV_1(&x[i], &diag[i], &x[i]);
217 gross 3097 }
218     } else if (n_block==2) {
219 gross 3098 Paso_BlockOps_MV_2(&x[0], &diag[0], &x[0]);
220     for (i = 1; i < n; ++i) {
221 gross 3097 mm=ptr_main[i];
222     for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<mm; ++iptr_ik) {
223     k=A_p->pattern->index[iptr_ik];
224     Paso_BlockOps_SMV_2(&x[2*i], &A_p->val[4*iptr_ik], &x[2*k]);
225     }
226 gross 3098 Paso_BlockOps_MV_2(&x[2*i], &diag[4*i], &x[2*i]);
227 gross 3097 }
228     } else if (n_block==3) {
229 gross 3098 Paso_BlockOps_MV_3(&x[0], &diag[0], &x[0]);
230     for (i = 1; i < n; ++i) {
231 gross 3097 mm=ptr_main[i];
232     /* x_i=x_i-a_ik*x_k */
233     for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<mm; ++iptr_ik) {
234     k=A_p->pattern->index[iptr_ik];
235     Paso_BlockOps_SMV_3(&x[3*i], &A_p->val[9*iptr_ik], &x[3*k]);
236     }
237 gross 3098 Paso_BlockOps_MV_3(&x[3*i], &diag[9*i], &x[3*i]);
238 gross 3097 }
239     } /* add block size >3 */
240 gross 3105
241 gross 3097 /* backward substitution */
242    
243     if (n_block==1) {
244 gross 3105 for (i = n-2; i > -1; --i) {
245 gross 3098 mm=ptr_main[i];
246     Paso_BlockOps_MV_1(&x[i], &A_p->val[mm], &x[i]);
247     for (iptr_ik=mm+1; iptr_ik < A_p->pattern->ptr[i+1]; ++iptr_ik) {
248     k=A_p->pattern->index[iptr_ik];
249     Paso_BlockOps_SMV_1(&x[i], &A_p->val[iptr_ik], &x[k]);
250     }
251     Paso_BlockOps_MV_1(&x[i], &diag[i], &x[i]);
252 gross 3097 }
253 gross 3098
254 gross 3097 } else if (n_block==2) {
255 gross 3105 for (i = n-2; i > -1; --i) {
256 gross 3098 mm=ptr_main[i];
257     Paso_BlockOps_MV_2(&x[2*i], &A_p->val[4*mm], &x[2*i]);
258     for (iptr_ik=mm+1; iptr_ik < A_p->pattern->ptr[i+1]; ++iptr_ik) {
259     k=A_p->pattern->index[iptr_ik];
260     Paso_BlockOps_SMV_2(&x[2*i], &A_p->val[4*iptr_ik], &x[2*k]);
261     }
262     Paso_BlockOps_MV_2(&x[2*i], &diag[i*4], &x[2*i]);
263 gross 3097 }
264     } else if (n_block==3) {
265 gross 3105 for (i = n-2; i > -1; --i) {
266 gross 3098 mm=ptr_main[i];
267     Paso_BlockOps_MV_3(&x[3*i], &A_p->val[9*mm], &x[3*i]);
268 gross 3097
269 gross 3098 for (iptr_ik=mm+1; iptr_ik < A_p->pattern->ptr[i+1]; ++iptr_ik) {
270     k=A_p->pattern->index[iptr_ik];
271     Paso_BlockOps_SMV_3(&x[3*i], &A_p->val[9*iptr_ik], &x[3*k]);
272     }
273     Paso_BlockOps_MV_3(&x[3*i], &diag[i*9], &x[3*i]);
274 gross 3097 }
275    
276     } /* add block size >3 */
277    
278     return;
279     }
280 gross 3094
281 gross 3158 void Paso_Preconditioner_LocalSmoother_Sweep_colored(Paso_SparseMatrix* A_p, Paso_Preconditioner_LocalSmoother * smoother, double * x)
282 gross 3094 {
283     const dim_t n=A_p->numRows;
284     const dim_t n_block=A_p->row_block_size;
285 gross 3158 const double *diag = smoother->diag;
286     index_t* pivot = smoother->pivot;
287 gross 3097 const dim_t block_size=A_p->block_size;
288 gross 3094
289     register dim_t i,k;
290     register index_t color,iptr_ik, mm;
291    
292     const index_t* coloring = Paso_Pattern_borrowColoringPointer(A_p->pattern);
293     const dim_t num_colors = Paso_Pattern_getNumColors(A_p->pattern);
294     const index_t* ptr_main = Paso_SparseMatrix_borrowMainDiagonalPointer(A_p);
295    
296     /* forward substitution */
297 gross 3105
298 gross 3094
299     /* color = 0 */
300     if (n_block==1) {
301 gross 3097 #pragma omp parallel for schedule(static) private(i)
302 gross 3094 for (i = 0; i < n; ++i) {
303 gross 3097 if (coloring[i]== 0 ) Paso_BlockOps_MV_1(&x[i], &diag[i], &x[i]);
304     }
305 gross 3094 } else if (n_block==2) {
306 gross 3097 #pragma omp parallel for schedule(static) private(i)
307 gross 3094 for (i = 0; i < n; ++i) {
308 gross 3097 if (coloring[i]== 0 ) Paso_BlockOps_MV_2(&x[2*i], &diag[i*4], &x[2*i]);
309 gross 3094 }
310 gross 3097 } else if (n_block==3) {
311     #pragma omp parallel for schedule(static) private(i)
312 gross 3094 for (i = 0; i < n; ++i) {
313 gross 3097 if (coloring[i]==0) Paso_BlockOps_MV_3(&x[3*i], &diag[i*9], &x[3*i]);
314 gross 3094 }
315 gross 3097 } else {
316     #pragma omp parallel for schedule(static) private(i)
317     for (i = 0; i < n; ++i) {
318     if (coloring[i]==0) Paso_BlockOps_Solve_N(n_block, &x[n_block*i], &diag[block_size*i], &pivot[n_block*i], &x[n_block*i]);
319     }
320     }
321 gross 3094
322     for (color=1;color<num_colors;++color) {
323 gross 3105
324 gross 3094 if (n_block==1) {
325 gross 3097 #pragma omp parallel for schedule(static) private(i,iptr_ik,k)
326 gross 3094 for (i = 0; i < n; ++i) {
327     if (coloring[i]==color) {
328     /* x_i=x_i-a_ik*x_k */
329     for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<A_p->pattern->ptr[i+1]; ++iptr_ik) {
330 gross 3105 k=A_p->pattern->index[iptr_ik];
331 gross 3097 if (coloring[k]<color) Paso_BlockOps_SMV_1(&x[i], &A_p->val[iptr_ik], &x[k]);
332 gross 3094 }
333 gross 3097 Paso_BlockOps_MV_1(&x[i], &diag[i], &x[i]);
334 gross 3094 }
335     }
336     } else if (n_block==2) {
337 gross 3097 #pragma omp parallel for schedule(static) private(i,iptr_ik,k)
338 gross 3094 for (i = 0; i < n; ++i) {
339     if (coloring[i]==color) {
340     for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<A_p->pattern->ptr[i+1]; ++iptr_ik) {
341     k=A_p->pattern->index[iptr_ik];
342 gross 3097 if (coloring[k]<color) Paso_BlockOps_SMV_2(&x[2*i], &A_p->val[4*iptr_ik], &x[2*k]);
343 gross 3094 }
344 gross 3097 Paso_BlockOps_MV_2(&x[2*i], &diag[4*i], &x[2*i]);
345 gross 3094 }
346     }
347     } else if (n_block==3) {
348 gross 3097 #pragma omp parallel for schedule(static) private(i,iptr_ik,k)
349 gross 3094 for (i = 0; i < n; ++i) {
350     if (coloring[i]==color) {
351     for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<A_p->pattern->ptr[i+1]; ++iptr_ik) {
352     k=A_p->pattern->index[iptr_ik];
353 gross 3097 if (coloring[k]<color) Paso_BlockOps_SMV_3(&x[3*i], &A_p->val[9*iptr_ik], &x[3*k]);
354 gross 3094 }
355 gross 3097 Paso_BlockOps_MV_3(&x[3*i], &diag[9*i], &x[3*i]);
356 gross 3094 }
357     }
358 gross 3097 } else {
359     #pragma omp parallel for schedule(static) private(i,iptr_ik,k)
360     for (i = 0; i < n; ++i) {
361     if (coloring[i] == color) {
362     for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<A_p->pattern->ptr[i+1]; ++iptr_ik) {
363     k=A_p->pattern->index[iptr_ik];
364     if (coloring[k]<color) Paso_BlockOps_SMV_N(n_block, &x[n_block*i], &A_p->val[block_size*iptr_ik], &x[n_block*k]);
365     }
366     Paso_BlockOps_Solve_N(n_block, &x[n_block*i], &diag[block_size*i], &pivot[n_block*i], &x[n_block*i]);
367     }
368     }
369     }
370 gross 3094 } /* end of coloring loop */
371    
372     /* backward substitution */
373     for (color=(num_colors)-2 ;color>-1;--color) { /* Note: color=(num_colors)-1 is not required */
374     if (n_block==1) {
375 gross 3097 #pragma omp parallel for schedule(static) private(mm, i,iptr_ik,k)
376 gross 3094 for (i = 0; i < n; ++i) {
377     if (coloring[i]==color) {
378     mm=ptr_main[i];
379 gross 3097 Paso_BlockOps_MV_1(&x[i], &A_p->val[mm], &x[i]);
380 gross 3094 for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<A_p->pattern->ptr[i+1]; ++iptr_ik) {
381     k=A_p->pattern->index[iptr_ik];
382 gross 3097 if (coloring[k]>color) Paso_BlockOps_SMV_1(&x[i], &A_p->val[iptr_ik], &x[k]);
383 gross 3094 }
384 gross 3097 Paso_BlockOps_MV_1(&x[i], &diag[i], &x[i]);
385 gross 3094 }
386     }
387     } else if (n_block==2) {
388 gross 3097 #pragma omp parallel for schedule(static) private(mm, i,iptr_ik,k)
389 gross 3094 for (i = 0; i < n; ++i) {
390     if (coloring[i]==color) {
391     mm=ptr_main[i];
392 gross 3097 Paso_BlockOps_MV_2(&x[2*i], &A_p->val[4*mm], &x[2*i]);
393 gross 3094 for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<A_p->pattern->ptr[i+1]; ++iptr_ik) {
394     k=A_p->pattern->index[iptr_ik];
395 gross 3097 if (coloring[k]>color) Paso_BlockOps_SMV_2(&x[2*i], &A_p->val[4*iptr_ik], &x[2*k]);
396 gross 3094 }
397 gross 3097 Paso_BlockOps_MV_2(&x[2*i], &diag[4*i], &x[2*i]);
398 gross 3094 }
399     }
400     } else if (n_block==3) {
401 gross 3097 #pragma omp parallel for schedule(static) private(mm, i,iptr_ik,k)
402 gross 3094 for (i = 0; i < n; ++i) {
403     if (coloring[i]==color) {
404     mm=ptr_main[i];
405 gross 3097 Paso_BlockOps_MV_3(&x[3*i], &A_p->val[9*mm], &x[3*i]);
406 gross 3094 for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<A_p->pattern->ptr[i+1]; ++iptr_ik) {
407     k=A_p->pattern->index[iptr_ik];
408 gross 3097 if (coloring[k]>color) Paso_BlockOps_SMV_3(&x[3*i], &A_p->val[9*iptr_ik], &x[3*k]);
409 gross 3094 }
410 gross 3097 Paso_BlockOps_MV_3(&x[3*i], &diag[9*i], &x[3*i]);
411 gross 3094 }
412     }
413 gross 3097 } else {
414     #pragma omp parallel for schedule(static) private(mm, i,iptr_ik,k)
415     for (i = 0; i < n; ++i) {
416     if (coloring[i]==color) {
417     mm=ptr_main[i];
418     Paso_BlockOps_MV_N( n_block, &x[n_block*i], &A_p->val[block_size*mm], &x[n_block*i] );
419     for (iptr_ik=A_p->pattern->ptr[i];iptr_ik<A_p->pattern->ptr[i+1]; ++iptr_ik) {
420     k=A_p->pattern->index[iptr_ik];
421     if (coloring[k]>color) Paso_BlockOps_SMV_N(n_block, &x[n_block*i], &A_p->val[block_size*iptr_ik], &x[n_block*k]);
422     }
423     Paso_BlockOps_Solve_N(n_block, &x[n_block*i], &diag[block_size*i], &pivot[n_block*i], &x[n_block*i]);
424     }
425     }
426     }
427 gross 3094 }
428     return;
429 gross 3125 }

  ViewVC Help
Powered by ViewVC 1.1.26