Actual source code: greedy.c
1: #include <petsc/private/matimpl.h>
2: #include <../src/mat/impls/aij/seq/aij.h>
3: #include <../src/mat/impls/aij/mpi/mpiaij.h>
4: #include <petscsf.h>
6: typedef struct {
7: PetscBool symmetric;
8: } MC_Greedy;
10: static PetscErrorCode MatColoringDestroy_Greedy(MatColoring mc)
11: {
15: PetscFree(mc->data);
16: return(0);
17: }
19: static PetscErrorCode GreedyColoringLocalDistanceOne_Private(MatColoring mc,PetscReal *wts,PetscInt *lperm,ISColoringValue *colors)
20: {
21: PetscInt i,j,k,s,e,n,no,nd,nd_global,n_global,idx,ncols,maxcolors,masksize,ccol,*mask;
22: PetscErrorCode ierr;
23: Mat m=mc->mat;
24: Mat_MPIAIJ *aij = (Mat_MPIAIJ*)m->data;
25: Mat md=NULL,mo=NULL;
26: const PetscInt *md_i,*mo_i,*md_j,*mo_j;
27: PetscBool isMPIAIJ,isSEQAIJ;
28: ISColoringValue pcol;
29: const PetscInt *cidx;
30: PetscInt *lcolors,*ocolors;
31: PetscReal *owts=NULL;
32: PetscSF sf;
33: PetscLayout layout;
36: MatGetSize(m,&n_global,NULL);
37: MatGetOwnershipRange(m,&s,&e);
38: n=e-s;
39: masksize=20;
40: nd_global = 0;
41: /* get the matrix communication structures */
42: PetscObjectTypeCompare((PetscObject)m, MATMPIAIJ, &isMPIAIJ);
43: PetscObjectBaseTypeCompare((PetscObject)m, MATSEQAIJ, &isSEQAIJ);
44: if (isMPIAIJ) {
45: /* get the CSR data for on and off diagonal portions of m */
46: Mat_SeqAIJ *dseq;
47: Mat_SeqAIJ *oseq;
48: md=aij->A;
49: dseq = (Mat_SeqAIJ*)md->data;
50: mo=aij->B;
51: oseq = (Mat_SeqAIJ*)mo->data;
52: md_i = dseq->i;
53: md_j = dseq->j;
54: mo_i = oseq->i;
55: mo_j = oseq->j;
56: } else if (isSEQAIJ) {
57: /* get the CSR data for m */
58: Mat_SeqAIJ *dseq;
59: /* no off-processor nodes */
60: md=m;
61: dseq = (Mat_SeqAIJ*)md->data;
62: mo=NULL;
63: no=0;
64: md_i = dseq->i;
65: md_j = dseq->j;
66: mo_i = NULL;
67: mo_j = NULL;
68: } else SETERRQ(PetscObjectComm((PetscObject)mc),PETSC_ERR_ARG_WRONG,"Matrix must be AIJ for greedy coloring");
69: MatColoringGetMaxColors(mc,&maxcolors);
70: if (mo) {
71: VecGetSize(aij->lvec,&no);
72: PetscMalloc2(no,&ocolors,no,&owts);
73: for (i=0;i<no;i++) {
74: ocolors[i]=maxcolors;
75: }
76: }
78: PetscMalloc1(masksize,&mask);
79: PetscMalloc1(n,&lcolors);
80: for (i=0;i<n;i++) {
81: lcolors[i]=maxcolors;
82: }
83: for (i=0;i<masksize;i++) {
84: mask[i]=-1;
85: }
86: if (mo) {
87: /* transfer neighbor weights */
88: PetscSFCreate(PetscObjectComm((PetscObject)m),&sf);
89: MatGetLayouts(m,&layout,NULL);
90: PetscSFSetGraphLayout(sf,layout,no,NULL,PETSC_COPY_VALUES,aij->garray);
91: PetscSFBcastBegin(sf,MPIU_REAL,wts,owts,MPI_REPLACE);
92: PetscSFBcastEnd(sf,MPIU_REAL,wts,owts,MPI_REPLACE);
93: }
94: while (nd_global < n_global) {
95: nd=n;
96: /* assign lowest possible color to each local vertex */
97: PetscLogEventBegin(MATCOLORING_Local,mc,0,0,0);
98: for (i=0;i<n;i++) {
99: idx=lperm[i];
100: if (lcolors[idx] == maxcolors) {
101: ncols = md_i[idx+1]-md_i[idx];
102: cidx = &(md_j[md_i[idx]]);
103: for (j=0;j<ncols;j++) {
104: if (lcolors[cidx[j]] != maxcolors) {
105: ccol=lcolors[cidx[j]];
106: if (ccol>=masksize) {
107: PetscInt *newmask;
108: PetscMalloc1(masksize*2,&newmask);
109: for (k=0;k<2*masksize;k++) {
110: newmask[k]=-1;
111: }
112: for (k=0;k<masksize;k++) {
113: newmask[k]=mask[k];
114: }
115: PetscFree(mask);
116: mask=newmask;
117: masksize*=2;
118: }
119: mask[ccol]=idx;
120: }
121: }
122: if (mo) {
123: ncols = mo_i[idx+1]-mo_i[idx];
124: cidx = &(mo_j[mo_i[idx]]);
125: for (j=0;j<ncols;j++) {
126: if (ocolors[cidx[j]] != maxcolors) {
127: ccol=ocolors[cidx[j]];
128: if (ccol>=masksize) {
129: PetscInt *newmask;
130: PetscMalloc1(masksize*2,&newmask);
131: for (k=0;k<2*masksize;k++) {
132: newmask[k]=-1;
133: }
134: for (k=0;k<masksize;k++) {
135: newmask[k]=mask[k];
136: }
137: PetscFree(mask);
138: mask=newmask;
139: masksize*=2;
140: }
141: mask[ccol]=idx;
142: }
143: }
144: }
145: for (j=0;j<masksize;j++) {
146: if (mask[j]!=idx) {
147: break;
148: }
149: }
150: pcol=j;
151: if (pcol>maxcolors)pcol=maxcolors;
152: lcolors[idx]=pcol;
153: }
154: }
155: PetscLogEventEnd(MATCOLORING_Local,mc,0,0,0);
156: if (mo) {
157: /* transfer neighbor colors */
158: PetscLogEventBegin(MATCOLORING_Comm,mc,0,0,0);
159: PetscSFBcastBegin(sf,MPIU_INT,lcolors,ocolors,MPI_REPLACE);
160: PetscSFBcastEnd(sf,MPIU_INT,lcolors,ocolors,MPI_REPLACE);
161: /* check for conflicts -- this is merely checking if any adjacent off-processor rows have the same color and marking the ones that are lower weight locally for changing */
162: for (i=0;i<n;i++) {
163: ncols = mo_i[i+1]-mo_i[i];
164: cidx = &(mo_j[mo_i[i]]);
165: for (j=0;j<ncols;j++) {
166: /* in the case of conflicts, the highest weight one stays and the others go */
167: if ((ocolors[cidx[j]] == lcolors[i]) && (owts[cidx[j]] > wts[i]) && lcolors[i] < maxcolors) {
168: lcolors[i]=maxcolors;
169: nd--;
170: }
171: }
172: }
173: nd_global=0;
174: }
175: MPIU_Allreduce(&nd,&nd_global,1,MPIU_INT,MPI_SUM,PetscObjectComm((PetscObject)mc));
176: }
177: for (i=0;i<n;i++) {
178: colors[i] = (ISColoringValue)lcolors[i];
179: }
180: PetscFree(mask);
181: PetscFree(lcolors);
182: if (mo) {
183: PetscFree2(ocolors,owts);
184: PetscSFDestroy(&sf);
185: }
186: return(0);
187: }
189: static PetscErrorCode GreedyColoringLocalDistanceTwo_Private(MatColoring mc,PetscReal *wts,PetscInt *lperm,ISColoringValue *colors)
190: {
191: MC_Greedy *gr = (MC_Greedy *) mc->data;
192: PetscInt i,j,k,l,s,e,n,nd,nd_global,n_global,idx,ncols,maxcolors,mcol,mcol_global,nd1cols,*mask,masksize,*d1cols,*bad,*badnext,nbad,badsize,ccol,no,cbad;
193: Mat m = mc->mat, mt;
194: Mat_MPIAIJ *aij = (Mat_MPIAIJ*)m->data;
195: Mat md=NULL,mo=NULL;
196: const PetscInt *md_i,*mo_i,*md_j,*mo_j;
197: const PetscInt *rmd_i,*rmo_i,*rmd_j,*rmo_j;
198: PetscBool isMPIAIJ,isSEQAIJ;
199: PetscInt pcol,*dcolors,*ocolors;
200: ISColoringValue *badidx;
201: const PetscInt *cidx;
202: PetscReal *owts,*colorweights;
203: PetscInt *oconf,*conf;
204: PetscSF sf;
205: PetscLayout layout;
206: PetscErrorCode ierr;
209: MatGetSize(m,&n_global,NULL);
210: MatGetOwnershipRange(m,&s,&e);
211: n=e-s;
212: nd_global = 0;
213: /* get the matrix communication structures */
214: PetscObjectBaseTypeCompare((PetscObject)m, MATMPIAIJ, &isMPIAIJ);
215: PetscObjectBaseTypeCompare((PetscObject)m, MATSEQAIJ, &isSEQAIJ);
216: if (isMPIAIJ) {
217: Mat_SeqAIJ *dseq;
218: Mat_SeqAIJ *oseq;
219: md=aij->A;
220: dseq = (Mat_SeqAIJ*)md->data;
221: mo=aij->B;
222: oseq = (Mat_SeqAIJ*)mo->data;
223: md_i = dseq->i;
224: md_j = dseq->j;
225: mo_i = oseq->i;
226: mo_j = oseq->j;
227: rmd_i = dseq->i;
228: rmd_j = dseq->j;
229: rmo_i = oseq->i;
230: rmo_j = oseq->j;
231: } else if (isSEQAIJ) {
232: Mat_SeqAIJ *dseq;
233: /* no off-processor nodes */
234: md=m;
235: dseq = (Mat_SeqAIJ*)md->data;
236: md_i = dseq->i;
237: md_j = dseq->j;
238: mo_i = NULL;
239: mo_j = NULL;
240: rmd_i = dseq->i;
241: rmd_j = dseq->j;
242: rmo_i = NULL;
243: rmo_j = NULL;
244: } else SETERRQ(PetscObjectComm((PetscObject)mc),PETSC_ERR_ARG_WRONG,"Matrix must be AIJ for greedy coloring");
245: if (!gr->symmetric) {
246: MatTranspose(m, MAT_INITIAL_MATRIX, &mt);
247: if (isSEQAIJ) {
248: Mat_SeqAIJ *dseq = (Mat_SeqAIJ*) mt->data;
249: rmd_i = dseq->i;
250: rmd_j = dseq->j;
251: rmo_i = NULL;
252: rmo_j = NULL;
253: } else SETERRQ(PetscObjectComm((PetscObject) mc), PETSC_ERR_SUP, "Nonsymmetric greedy coloring only works in serial");
254: }
255: /* create the vectors and communication structures if necessary */
256: no=0;
257: if (mo) {
258: VecGetLocalSize(aij->lvec,&no);
259: PetscSFCreate(PetscObjectComm((PetscObject)m),&sf);
260: MatGetLayouts(m,&layout,NULL);
261: PetscSFSetGraphLayout(sf,layout,no,NULL,PETSC_COPY_VALUES,aij->garray);
262: }
263: MatColoringGetMaxColors(mc,&maxcolors);
264: masksize=n;
265: nbad=0;
266: badsize=n;
267: PetscMalloc1(masksize,&mask);
268: PetscMalloc4(n,&d1cols,n,&dcolors,n,&conf,n,&bad);
269: PetscMalloc2(badsize,&badidx,badsize,&badnext);
270: for (i=0;i<masksize;i++) {
271: mask[i]=-1;
272: }
273: for (i=0;i<n;i++) {
274: dcolors[i]=maxcolors;
275: bad[i]=-1;
276: }
277: for (i=0;i<badsize;i++) {
278: badnext[i]=-1;
279: }
280: if (mo) {
281: PetscMalloc3(no,&owts,no,&oconf,no,&ocolors);
282: PetscSFBcastBegin(sf,MPIU_REAL,wts,owts,MPI_REPLACE);
283: PetscSFBcastEnd(sf,MPIU_REAL,wts,owts,MPI_REPLACE);
284: for (i=0;i<no;i++) {
285: ocolors[i]=maxcolors;
286: }
287: } else { /* Appease overzealous -Wmaybe-initialized */
288: owts = NULL;
289: oconf = NULL;
290: ocolors = NULL;
291: }
292: mcol=0;
293: while (nd_global < n_global) {
294: nd=n;
295: /* assign lowest possible color to each local vertex */
296: mcol_global=0;
297: PetscLogEventBegin(MATCOLORING_Local,mc,0,0,0);
298: for (i=0;i<n;i++) {
299: idx=lperm[i];
300: if (dcolors[idx] == maxcolors) {
301: /* entries in bad */
302: cbad=bad[idx];
303: while (cbad>=0) {
304: ccol=badidx[cbad];
305: if (ccol>=masksize) {
306: PetscInt *newmask;
307: PetscMalloc1(masksize*2,&newmask);
308: for (k=0;k<2*masksize;k++) {
309: newmask[k]=-1;
310: }
311: for (k=0;k<masksize;k++) {
312: newmask[k]=mask[k];
313: }
314: PetscFree(mask);
315: mask=newmask;
316: masksize*=2;
317: }
318: mask[ccol]=idx;
319: cbad=badnext[cbad];
320: }
321: /* diagonal distance-one rows */
322: nd1cols=0;
323: ncols = rmd_i[idx+1]-rmd_i[idx];
324: cidx = &(rmd_j[rmd_i[idx]]);
325: for (j=0;j<ncols;j++) {
326: d1cols[nd1cols] = cidx[j];
327: nd1cols++;
328: ccol=dcolors[cidx[j]];
329: if (ccol != maxcolors) {
330: if (ccol>=masksize) {
331: PetscInt *newmask;
332: PetscMalloc1(masksize*2,&newmask);
333: for (k=0;k<2*masksize;k++) {
334: newmask[k]=-1;
335: }
336: for (k=0;k<masksize;k++) {
337: newmask[k]=mask[k];
338: }
339: PetscFree(mask);
340: mask=newmask;
341: masksize*=2;
342: }
343: mask[ccol]=idx;
344: }
345: }
346: /* off-diagonal distance-one rows */
347: if (mo) {
348: ncols = rmo_i[idx+1]-rmo_i[idx];
349: cidx = &(rmo_j[rmo_i[idx]]);
350: for (j=0;j<ncols;j++) {
351: ccol=ocolors[cidx[j]];
352: if (ccol != maxcolors) {
353: if (ccol>=masksize) {
354: PetscInt *newmask;
355: PetscMalloc1(masksize*2,&newmask);
356: for (k=0;k<2*masksize;k++) {
357: newmask[k]=-1;
358: }
359: for (k=0;k<masksize;k++) {
360: newmask[k]=mask[k];
361: }
362: PetscFree(mask);
363: mask=newmask;
364: masksize*=2;
365: }
366: mask[ccol]=idx;
367: }
368: }
369: }
370: /* diagonal distance-two rows */
371: for (j=0;j<nd1cols;j++) {
372: ncols = md_i[d1cols[j]+1]-md_i[d1cols[j]];
373: cidx = &(md_j[md_i[d1cols[j]]]);
374: for (l=0;l<ncols;l++) {
375: ccol=dcolors[cidx[l]];
376: if (ccol != maxcolors) {
377: if (ccol>=masksize) {
378: PetscInt *newmask;
379: PetscMalloc1(masksize*2,&newmask);
380: for (k=0;k<2*masksize;k++) {
381: newmask[k]=-1;
382: }
383: for (k=0;k<masksize;k++) {
384: newmask[k]=mask[k];
385: }
386: PetscFree(mask);
387: mask=newmask;
388: masksize*=2;
389: }
390: mask[ccol]=idx;
391: }
392: }
393: }
394: /* off-diagonal distance-two rows */
395: if (mo) {
396: for (j=0;j<nd1cols;j++) {
397: ncols = mo_i[d1cols[j]+1]-mo_i[d1cols[j]];
398: cidx = &(mo_j[mo_i[d1cols[j]]]);
399: for (l=0;l<ncols;l++) {
400: ccol=ocolors[cidx[l]];
401: if (ccol != maxcolors) {
402: if (ccol>=masksize) {
403: PetscInt *newmask;
404: PetscMalloc1(masksize*2,&newmask);
405: for (k=0;k<2*masksize;k++) {
406: newmask[k]=-1;
407: }
408: for (k=0;k<masksize;k++) {
409: newmask[k]=mask[k];
410: }
411: PetscFree(mask);
412: mask=newmask;
413: masksize*=2;
414: }
415: mask[ccol]=idx;
416: }
417: }
418: }
419: }
420: /* assign this one the lowest color possible by seeing if there's a gap in the sequence of sorted neighbor colors */
421: for (j=0;j<masksize;j++) {
422: if (mask[j]!=idx) {
423: break;
424: }
425: }
426: pcol=j;
427: if (pcol>maxcolors) pcol=maxcolors;
428: dcolors[idx]=pcol;
429: if (pcol>mcol) mcol=pcol;
430: }
431: }
432: PetscLogEventEnd(MATCOLORING_Local,mc,0,0,0);
433: if (mo) {
434: /* transfer neighbor colors */
435: PetscSFBcastBegin(sf,MPIU_INT,dcolors,ocolors,MPI_REPLACE);
436: PetscSFBcastEnd(sf,MPIU_INT,dcolors,ocolors,MPI_REPLACE);
437: /* find the maximum color assigned locally and allocate a mask */
438: MPIU_Allreduce(&mcol,&mcol_global,1,MPIU_INT,MPI_MAX,PetscObjectComm((PetscObject)mc));
439: PetscMalloc1(mcol_global+1,&colorweights);
440: /* check for conflicts */
441: for (i=0;i<n;i++) {
442: conf[i]=PETSC_FALSE;
443: }
444: for (i=0;i<no;i++) {
445: oconf[i]=PETSC_FALSE;
446: }
447: for (i=0;i<n;i++) {
448: ncols = mo_i[i+1]-mo_i[i];
449: cidx = &(mo_j[mo_i[i]]);
450: if (ncols > 0) {
451: /* fill in the mask */
452: for (j=0;j<mcol_global+1;j++) {
453: colorweights[j]=0;
454: }
455: colorweights[dcolors[i]]=wts[i];
456: /* fill in the off-diagonal part of the mask */
457: for (j=0;j<ncols;j++) {
458: ccol=ocolors[cidx[j]];
459: if (ccol < maxcolors) {
460: if (colorweights[ccol] < owts[cidx[j]]) {
461: colorweights[ccol] = owts[cidx[j]];
462: }
463: }
464: }
465: /* fill in the on-diagonal part of the mask */
466: ncols = md_i[i+1]-md_i[i];
467: cidx = &(md_j[md_i[i]]);
468: for (j=0;j<ncols;j++) {
469: ccol=dcolors[cidx[j]];
470: if (ccol < maxcolors) {
471: if (colorweights[ccol] < wts[cidx[j]]) {
472: colorweights[ccol] = wts[cidx[j]];
473: }
474: }
475: }
476: /* go back through and set up on and off-diagonal conflict vectors */
477: ncols = md_i[i+1]-md_i[i];
478: cidx = &(md_j[md_i[i]]);
479: for (j=0;j<ncols;j++) {
480: ccol=dcolors[cidx[j]];
481: if (ccol < maxcolors) {
482: if (colorweights[ccol] > wts[cidx[j]]) {
483: conf[cidx[j]]=PETSC_TRUE;
484: }
485: }
486: }
487: ncols = mo_i[i+1]-mo_i[i];
488: cidx = &(mo_j[mo_i[i]]);
489: for (j=0;j<ncols;j++) {
490: ccol=ocolors[cidx[j]];
491: if (ccol < maxcolors) {
492: if (colorweights[ccol] > owts[cidx[j]]) {
493: oconf[cidx[j]]=PETSC_TRUE;
494: }
495: }
496: }
497: }
498: }
499: nd_global=0;
500: PetscFree(colorweights);
501: PetscLogEventBegin(MATCOLORING_Comm,mc,0,0,0);
502: PetscSFReduceBegin(sf,MPIU_INT,oconf,conf,MPIU_SUM);
503: PetscSFReduceEnd(sf,MPIU_INT,oconf,conf,MPIU_SUM);
504: PetscLogEventEnd(MATCOLORING_Comm,mc,0,0,0);
505: /* go through and unset local colors that have conflicts */
506: for (i=0;i<n;i++) {
507: if (conf[i]>0) {
508: /* push this color onto the bad stack */
509: badidx[nbad]=dcolors[i];
510: badnext[nbad]=bad[i];
511: bad[i]=nbad;
512: nbad++;
513: if (nbad>=badsize) {
514: PetscInt *newbadnext;
515: ISColoringValue *newbadidx;
516: PetscMalloc2(badsize*2,&newbadidx,badsize*2,&newbadnext);
517: for (k=0;k<2*badsize;k++) {
518: newbadnext[k]=-1;
519: }
520: for (k=0;k<badsize;k++) {
521: newbadidx[k]=badidx[k];
522: newbadnext[k]=badnext[k];
523: }
524: PetscFree2(badidx,badnext);
525: badidx=newbadidx;
526: badnext=newbadnext;
527: badsize*=2;
528: }
529: dcolors[i] = maxcolors;
530: nd--;
531: }
532: }
533: }
534: MPIU_Allreduce(&nd,&nd_global,1,MPIU_INT,MPI_SUM,PetscObjectComm((PetscObject)mc));
535: }
536: if (mo) {
537: PetscSFDestroy(&sf);
538: PetscFree3(owts,oconf,ocolors);
539: }
540: for (i=0;i<n;i++) {
541: colors[i]=dcolors[i];
542: }
543: PetscFree(mask);
544: PetscFree4(d1cols,dcolors,conf,bad);
545: PetscFree2(badidx,badnext);
546: if (!gr->symmetric) {MatDestroy(&mt);}
547: return(0);
548: }
550: static PetscErrorCode MatColoringApply_Greedy(MatColoring mc,ISColoring *iscoloring)
551: {
552: PetscErrorCode ierr;
553: PetscInt finalcolor,finalcolor_global;
554: ISColoringValue *colors;
555: PetscInt ncolstotal,ncols;
556: PetscReal *wts;
557: PetscInt i,*lperm;
560: MatGetSize(mc->mat,NULL,&ncolstotal);
561: MatGetLocalSize(mc->mat,NULL,&ncols);
562: if (!mc->user_weights) {
563: MatColoringCreateWeights(mc,&wts,&lperm);
564: } else {
565: wts = mc->user_weights;
566: lperm = mc->user_lperm;
567: }
568: PetscMalloc1(ncols,&colors);
569: if (mc->dist == 1) {
570: GreedyColoringLocalDistanceOne_Private(mc,wts,lperm,colors);
571: } else if (mc->dist == 2) {
572: GreedyColoringLocalDistanceTwo_Private(mc,wts,lperm,colors);
573: } else SETERRQ(PetscObjectComm((PetscObject)mc),PETSC_ERR_ARG_OUTOFRANGE,"Only distance 1 and distance 2 supported by MatColoringGreedy");
574: finalcolor=0;
575: for (i=0;i<ncols;i++) {
576: if (colors[i] > finalcolor) finalcolor=colors[i];
577: }
578: finalcolor_global=0;
579: MPIU_Allreduce(&finalcolor,&finalcolor_global,1,MPIU_INT,MPI_MAX,PetscObjectComm((PetscObject)mc));
580: PetscLogEventBegin(MATCOLORING_ISCreate,mc,0,0,0);
581: ISColoringCreate(PetscObjectComm((PetscObject)mc),finalcolor_global+1,ncols,colors,PETSC_OWN_POINTER,iscoloring);
582: PetscLogEventEnd(MATCOLORING_ISCreate,mc,0,0,0);
583: if (!mc->user_weights) {
584: PetscFree(wts);
585: PetscFree(lperm);
586: }
587: return(0);
588: }
590: static PetscErrorCode MatColoringSetFromOptions_Greedy(PetscOptionItems *PetscOptionsObject, MatColoring mc)
591: {
592: MC_Greedy *gr = (MC_Greedy *) mc->data;
596: PetscOptionsHead(PetscOptionsObject, "Greedy options");
597: PetscOptionsBool("-mat_coloring_greedy_symmetric", "Flag for assuming a symmetric matrix", "", gr->symmetric, &gr->symmetric, NULL);
598: PetscOptionsTail();
599: return(0);
600: }
602: /*MC
603: MATCOLORINGGREEDY - Greedy-with-conflict correction based Matrix Coloring for distance 1 and 2.
605: Level: beginner
607: Notes:
609: These algorithms proceed in two phases -- local coloring and conflict resolution. The local coloring
610: tentatively colors all vertices at the distance required given what's known of the global coloring. Then,
611: the updated colors are transferred to different processors at distance one. In the distance one case, each
612: vertex with nonlocal neighbors is then checked to see if it conforms, with the vertex being
613: marked for recoloring if its lower weight than its same colored neighbor. In the distance two case,
614: each boundary vertex's immediate star is checked for validity of the coloring. Lower-weight conflict
615: vertices are marked, and then the conflicts are gathered back on owning processors. In both cases
616: this is done until each column has received a valid color.
618: Supports both distance one and distance two colorings.
620: References:
621: . 1. - Bozdag et al. "A Parallel Distance 2 Graph Coloring Algorithm for Distributed Memory Computers"
622: HPCC'05 Proceedings of the First international conference on High Performance Computing and Communications
624: .seealso: MatColoringCreate(), MatColoring, MatColoringSetType(), MatColoringType
625: M*/
626: PETSC_EXTERN PetscErrorCode MatColoringCreate_Greedy(MatColoring mc)
627: {
628: MC_Greedy *gr;
632: PetscNewLog(mc,&gr);
633: mc->data = gr;
634: mc->ops->apply = MatColoringApply_Greedy;
635: mc->ops->view = NULL;
636: mc->ops->destroy = MatColoringDestroy_Greedy;
637: mc->ops->setfromoptions = MatColoringSetFromOptions_Greedy;
639: gr->symmetric = PETSC_TRUE;
640: return(0);
641: }