Actual source code: dense.c
petsc-dev 2012-05-24
2: /*
3: Defines the basic matrix operations for sequential dense.
4: */
6: #include <../src/mat/impls/dense/seq/dense.h>
7: #include <petscblaslapack.h>
9: #include <../src/mat/impls/aij/seq/aij.h>
10: EXTERN_C_BEGIN
13: PetscErrorCode MatConvert_SeqDense_SeqAIJ(Mat A, MatType newtype,MatReuse reuse,Mat *newmat)
14: {
15: Mat B;
16: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
18: PetscInt i;
19: PetscInt *rows;
20: MatScalar *aa = a->v;
23: MatCreate(((PetscObject)A)->comm,&B);
24: MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N);
25: MatSetType(B,MATSEQAIJ);
26: MatSeqAIJSetPreallocation(B,A->cmap->n,PETSC_NULL);
28: PetscMalloc(A->rmap->n*sizeof(PetscInt),&rows);
29: for (i=0; i<A->rmap->n; i++) rows[i] = i;
31: for (i=0; i<A->cmap->n; i++) {
32: MatSetValues(B,A->rmap->n,rows,1,&i,aa,INSERT_VALUES);
33: aa += a->lda;
34: }
35: PetscFree(rows);
36: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
37: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
38:
39: if (reuse == MAT_REUSE_MATRIX) {
40: MatHeaderReplace(A,B);
41: } else {
42: *newmat = B;
43: }
44: return(0);
45: }
46: EXTERN_C_END
50: PetscErrorCode MatAXPY_SeqDense(Mat Y,PetscScalar alpha,Mat X,MatStructure str)
51: {
52: Mat_SeqDense *x = (Mat_SeqDense*)X->data,*y = (Mat_SeqDense*)Y->data;
53: PetscScalar oalpha = alpha;
54: PetscInt j;
55: PetscBLASInt N,m,ldax,lday,one = 1;
59: N = PetscBLASIntCast(X->rmap->n*X->cmap->n);
60: m = PetscBLASIntCast(X->rmap->n);
61: ldax = PetscBLASIntCast(x->lda);
62: lday = PetscBLASIntCast(y->lda);
63: if (ldax>m || lday>m) {
64: for (j=0; j<X->cmap->n; j++) {
65: BLASaxpy_(&m,&oalpha,x->v+j*ldax,&one,y->v+j*lday,&one);
66: }
67: } else {
68: BLASaxpy_(&N,&oalpha,x->v,&one,y->v,&one);
69: }
70: PetscLogFlops(PetscMax(2*N-1,0));
71: return(0);
72: }
76: PetscErrorCode MatGetInfo_SeqDense(Mat A,MatInfoType flag,MatInfo *info)
77: {
78: PetscInt N = A->rmap->n*A->cmap->n;
81: info->block_size = 1.0;
82: info->nz_allocated = (double)N;
83: info->nz_used = (double)N;
84: info->nz_unneeded = (double)0;
85: info->assemblies = (double)A->num_ass;
86: info->mallocs = 0;
87: info->memory = ((PetscObject)A)->mem;
88: info->fill_ratio_given = 0;
89: info->fill_ratio_needed = 0;
90: info->factor_mallocs = 0;
91: return(0);
92: }
96: PetscErrorCode MatScale_SeqDense(Mat A,PetscScalar alpha)
97: {
98: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
99: PetscScalar oalpha = alpha;
101: PetscBLASInt one = 1,j,nz,lda = PetscBLASIntCast(a->lda);
104: if (lda>A->rmap->n) {
105: nz = PetscBLASIntCast(A->rmap->n);
106: for (j=0; j<A->cmap->n; j++) {
107: BLASscal_(&nz,&oalpha,a->v+j*lda,&one);
108: }
109: } else {
110: nz = PetscBLASIntCast(A->rmap->n*A->cmap->n);
111: BLASscal_(&nz,&oalpha,a->v,&one);
112: }
113: PetscLogFlops(nz);
114: return(0);
115: }
119: PetscErrorCode MatIsHermitian_SeqDense(Mat A,PetscReal rtol,PetscBool *fl)
120: {
121: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
122: PetscInt i,j,m = A->rmap->n,N;
123: PetscScalar *v = a->v;
126: *fl = PETSC_FALSE;
127: if (A->rmap->n != A->cmap->n) return(0);
128: N = a->lda;
130: for (i=0; i<m; i++) {
131: for (j=i+1; j<m; j++) {
132: if (PetscAbsScalar(v[i+j*N] - PetscConj(v[j+i*N])) > rtol) return(0);
133: }
134: }
135: *fl = PETSC_TRUE;
136: return(0);
137: }
138:
141: PetscErrorCode MatDuplicateNoCreate_SeqDense(Mat newi,Mat A,MatDuplicateOption cpvalues)
142: {
143: Mat_SeqDense *mat = (Mat_SeqDense*)A->data,*l;
145: PetscInt lda = (PetscInt)mat->lda,j,m;
148: PetscLayoutReference(A->rmap,&newi->rmap);
149: PetscLayoutReference(A->cmap,&newi->cmap);
150: MatSeqDenseSetPreallocation(newi,PETSC_NULL);
151: if (cpvalues == MAT_COPY_VALUES) {
152: l = (Mat_SeqDense*)newi->data;
153: if (lda>A->rmap->n) {
154: m = A->rmap->n;
155: for (j=0; j<A->cmap->n; j++) {
156: PetscMemcpy(l->v+j*m,mat->v+j*lda,m*sizeof(PetscScalar));
157: }
158: } else {
159: PetscMemcpy(l->v,mat->v,A->rmap->n*A->cmap->n*sizeof(PetscScalar));
160: }
161: }
162: newi->assembled = PETSC_TRUE;
163: return(0);
164: }
168: PetscErrorCode MatDuplicate_SeqDense(Mat A,MatDuplicateOption cpvalues,Mat *newmat)
169: {
173: MatCreate(((PetscObject)A)->comm,newmat);
174: MatSetSizes(*newmat,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);
175: MatSetType(*newmat,((PetscObject)A)->type_name);
176: MatDuplicateNoCreate_SeqDense(*newmat,A,cpvalues);
177: return(0);
178: }
181: extern PetscErrorCode MatLUFactor_SeqDense(Mat,IS,IS,const MatFactorInfo*);
185: PetscErrorCode MatLUFactorNumeric_SeqDense(Mat fact,Mat A,const MatFactorInfo *info_dummy)
186: {
187: MatFactorInfo info;
191: MatDuplicateNoCreate_SeqDense(fact,A,MAT_COPY_VALUES);
192: MatLUFactor_SeqDense(fact,0,0,&info);
193: return(0);
194: }
198: PetscErrorCode MatSolve_SeqDense(Mat A,Vec xx,Vec yy)
199: {
200: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
202: PetscScalar *x,*y;
203: PetscBLASInt one = 1,info,m = PetscBLASIntCast(A->rmap->n);
204:
206: VecGetArray(xx,&x);
207: VecGetArray(yy,&y);
208: PetscMemcpy(y,x,A->rmap->n*sizeof(PetscScalar));
209: if (A->factortype == MAT_FACTOR_LU) {
210: #if defined(PETSC_MISSING_LAPACK_GETRS)
211: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"GETRS - Lapack routine is unavailable.");
212: #else
213: LAPACKgetrs_("N",&m,&one,mat->v,&mat->lda,mat->pivots,y,&m,&info);
214: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"GETRS - Bad solve");
215: #endif
216: } else if (A->factortype == MAT_FACTOR_CHOLESKY){
217: #if defined(PETSC_MISSING_LAPACK_POTRS)
218: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"POTRS - Lapack routine is unavailable.");
219: #else
220: LAPACKpotrs_("L",&m,&one,mat->v,&mat->lda,y,&m,&info);
221: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"POTRS Bad solve");
222: #endif
223: }
224: else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix must be factored to solve");
225: VecRestoreArray(xx,&x);
226: VecRestoreArray(yy,&y);
227: PetscLogFlops(2.0*A->cmap->n*A->cmap->n - A->cmap->n);
228: return(0);
229: }
233: PetscErrorCode MatMatSolve_SeqDense(Mat A,Mat B,Mat X)
234: {
235: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
237: PetscScalar *b,*x;
238: PetscInt n;
239: PetscBLASInt nrhs,info,m=PetscBLASIntCast(A->rmap->n);
240: PetscBool flg;
243: PetscObjectTypeCompareAny((PetscObject)B,&flg,MATSEQDENSE,MATMPIDENSE,PETSC_NULL);
244: if (!flg) SETERRQ(((PetscObject)A)->comm,PETSC_ERR_ARG_WRONG,"Matrix B must be MATDENSE matrix");
245: PetscObjectTypeCompareAny((PetscObject)X,&flg,MATSEQDENSE,MATMPIDENSE,PETSC_NULL);
246: if (!flg) SETERRQ(((PetscObject)A)->comm,PETSC_ERR_ARG_WRONG,"Matrix X must be MATDENSE matrix");
248: MatGetSize(B,PETSC_NULL,&n);
249: nrhs = PetscBLASIntCast(n);
250: MatGetArray(B,&b);
251: MatGetArray(X,&x);
253: PetscMemcpy(x,b,m*nrhs*sizeof(PetscScalar));
255: if (A->factortype == MAT_FACTOR_LU) {
256: #if defined(PETSC_MISSING_LAPACK_GETRS)
257: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"GETRS - Lapack routine is unavailable.");
258: #else
259: LAPACKgetrs_("N",&m,&nrhs,mat->v,&mat->lda,mat->pivots,x,&m,&info);
260: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"GETRS - Bad solve");
261: #endif
262: } else if (A->factortype == MAT_FACTOR_CHOLESKY){
263: #if defined(PETSC_MISSING_LAPACK_POTRS)
264: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"POTRS - Lapack routine is unavailable.");
265: #else
266: LAPACKpotrs_("L",&m,&nrhs,mat->v,&mat->lda,x,&m,&info);
267: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"POTRS Bad solve");
268: #endif
269: }
270: else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix must be factored to solve");
272: MatRestoreArray(B,&b);
273: MatRestoreArray(X,&x);
274: PetscLogFlops(nrhs*(2.0*m*m - m));
275: return(0);
276: }
280: PetscErrorCode MatSolveTranspose_SeqDense(Mat A,Vec xx,Vec yy)
281: {
282: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
284: PetscScalar *x,*y;
285: PetscBLASInt one = 1,info,m = PetscBLASIntCast(A->rmap->n);
286:
288: VecGetArray(xx,&x);
289: VecGetArray(yy,&y);
290: PetscMemcpy(y,x,A->rmap->n*sizeof(PetscScalar));
291: /* assume if pivots exist then use LU; else Cholesky */
292: if (mat->pivots) {
293: #if defined(PETSC_MISSING_LAPACK_GETRS)
294: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"GETRS - Lapack routine is unavailable.");
295: #else
296: LAPACKgetrs_("T",&m,&one,mat->v,&mat->lda,mat->pivots,y,&m,&info);
297: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"POTRS - Bad solve");
298: #endif
299: } else {
300: #if defined(PETSC_MISSING_LAPACK_POTRS)
301: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"POTRS - Lapack routine is unavailable.");
302: #else
303: LAPACKpotrs_("L",&m,&one,mat->v,&mat->lda,y,&m,&info);
304: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"POTRS - Bad solve");
305: #endif
306: }
307: VecRestoreArray(xx,&x);
308: VecRestoreArray(yy,&y);
309: PetscLogFlops(2.0*A->cmap->n*A->cmap->n - A->cmap->n);
310: return(0);
311: }
315: PetscErrorCode MatSolveAdd_SeqDense(Mat A,Vec xx,Vec zz,Vec yy)
316: {
317: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
319: PetscScalar *x,*y,sone = 1.0;
320: Vec tmp = 0;
321: PetscBLASInt one = 1,info,m = PetscBLASIntCast(A->rmap->n);
322:
324: VecGetArray(xx,&x);
325: VecGetArray(yy,&y);
326: if (!A->rmap->n || !A->cmap->n) return(0);
327: if (yy == zz) {
328: VecDuplicate(yy,&tmp);
329: PetscLogObjectParent(A,tmp);
330: VecCopy(yy,tmp);
331: }
332: PetscMemcpy(y,x,A->rmap->n*sizeof(PetscScalar));
333: /* assume if pivots exist then use LU; else Cholesky */
334: if (mat->pivots) {
335: #if defined(PETSC_MISSING_LAPACK_GETRS)
336: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"GETRS - Lapack routine is unavailable.");
337: #else
338: LAPACKgetrs_("N",&m,&one,mat->v,&mat->lda,mat->pivots,y,&m,&info);
339: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Bad solve");
340: #endif
341: } else {
342: #if defined(PETSC_MISSING_LAPACK_POTRS)
343: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"POTRS - Lapack routine is unavailable.");
344: #else
345: LAPACKpotrs_("L",&m,&one,mat->v,&mat->lda,y,&m,&info);
346: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Bad solve");
347: #endif
348: }
349: if (tmp) {
350: VecAXPY(yy,sone,tmp);
351: VecDestroy(&tmp);
352: } else {
353: VecAXPY(yy,sone,zz);
354: }
355: VecRestoreArray(xx,&x);
356: VecRestoreArray(yy,&y);
357: PetscLogFlops(2.0*A->cmap->n*A->cmap->n);
358: return(0);
359: }
363: PetscErrorCode MatSolveTransposeAdd_SeqDense(Mat A,Vec xx,Vec zz,Vec yy)
364: {
365: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
367: PetscScalar *x,*y,sone = 1.0;
368: Vec tmp;
369: PetscBLASInt one = 1,info,m = PetscBLASIntCast(A->rmap->n);
370:
372: if (!A->rmap->n || !A->cmap->n) return(0);
373: VecGetArray(xx,&x);
374: VecGetArray(yy,&y);
375: if (yy == zz) {
376: VecDuplicate(yy,&tmp);
377: PetscLogObjectParent(A,tmp);
378: VecCopy(yy,tmp);
379: }
380: PetscMemcpy(y,x,A->rmap->n*sizeof(PetscScalar));
381: /* assume if pivots exist then use LU; else Cholesky */
382: if (mat->pivots) {
383: #if defined(PETSC_MISSING_LAPACK_GETRS)
384: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"GETRS - Lapack routine is unavailable.");
385: #else
386: LAPACKgetrs_("T",&m,&one,mat->v,&mat->lda,mat->pivots,y,&m,&info);
387: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Bad solve");
388: #endif
389: } else {
390: #if defined(PETSC_MISSING_LAPACK_POTRS)
391: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"POTRS - Lapack routine is unavailable.");
392: #else
393: LAPACKpotrs_("L",&m,&one,mat->v,&mat->lda,y,&m,&info);
394: if (info) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Bad solve");
395: #endif
396: }
397: if (tmp) {
398: VecAXPY(yy,sone,tmp);
399: VecDestroy(&tmp);
400: } else {
401: VecAXPY(yy,sone,zz);
402: }
403: VecRestoreArray(xx,&x);
404: VecRestoreArray(yy,&y);
405: PetscLogFlops(2.0*A->cmap->n*A->cmap->n);
406: return(0);
407: }
409: /* ---------------------------------------------------------------*/
410: /* COMMENT: I have chosen to hide row permutation in the pivots,
411: rather than put it in the Mat->row slot.*/
414: PetscErrorCode MatLUFactor_SeqDense(Mat A,IS row,IS col,const MatFactorInfo *minfo)
415: {
416: #if defined(PETSC_MISSING_LAPACK_GETRF)
418: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"GETRF - Lapack routine is unavailable.");
419: #else
420: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
422: PetscBLASInt n,m,info;
425: n = PetscBLASIntCast(A->cmap->n);
426: m = PetscBLASIntCast(A->rmap->n);
427: if (!mat->pivots) {
428: PetscMalloc((A->rmap->n+1)*sizeof(PetscBLASInt),&mat->pivots);
429: PetscLogObjectMemory(A,A->rmap->n*sizeof(PetscBLASInt));
430: }
431: if (!A->rmap->n || !A->cmap->n) return(0);
432: PetscFPTrapPush(PETSC_FP_TRAP_OFF);
433: LAPACKgetrf_(&m,&n,mat->v,&mat->lda,mat->pivots,&info);
434: PetscFPTrapPop();
436: if (info<0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Bad argument to LU factorization");
437: if (info>0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Bad LU factorization");
438: A->ops->solve = MatSolve_SeqDense;
439: A->ops->solvetranspose = MatSolveTranspose_SeqDense;
440: A->ops->solveadd = MatSolveAdd_SeqDense;
441: A->ops->solvetransposeadd = MatSolveTransposeAdd_SeqDense;
442: A->factortype = MAT_FACTOR_LU;
444: PetscLogFlops((2.0*A->cmap->n*A->cmap->n*A->cmap->n)/3);
445: #endif
446: return(0);
447: }
451: PetscErrorCode MatCholeskyFactor_SeqDense(Mat A,IS perm,const MatFactorInfo *factinfo)
452: {
453: #if defined(PETSC_MISSING_LAPACK_POTRF)
455: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"POTRF - Lapack routine is unavailable.");
456: #else
457: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
459: PetscBLASInt info,n = PetscBLASIntCast(A->cmap->n);
460:
462: PetscFree(mat->pivots);
464: if (!A->rmap->n || !A->cmap->n) return(0);
465: LAPACKpotrf_("L",&n,mat->v,&mat->lda,&info);
466: if (info) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_CH_ZRPVT,"Bad factorization: zero pivot in row %D",(PetscInt)info-1);
467: A->ops->solve = MatSolve_SeqDense;
468: A->ops->solvetranspose = MatSolveTranspose_SeqDense;
469: A->ops->solveadd = MatSolveAdd_SeqDense;
470: A->ops->solvetransposeadd = MatSolveTransposeAdd_SeqDense;
471: A->factortype = MAT_FACTOR_CHOLESKY;
472: PetscLogFlops((A->cmap->n*A->cmap->n*A->cmap->n)/3.0);
473: #endif
474: return(0);
475: }
480: PetscErrorCode MatCholeskyFactorNumeric_SeqDense(Mat fact,Mat A,const MatFactorInfo *info_dummy)
481: {
483: MatFactorInfo info;
486: info.fill = 1.0;
487: MatDuplicateNoCreate_SeqDense(fact,A,MAT_COPY_VALUES);
488: MatCholeskyFactor_SeqDense(fact,0,&info);
489: return(0);
490: }
494: PetscErrorCode MatCholeskyFactorSymbolic_SeqDense(Mat fact,Mat A,IS row,const MatFactorInfo *info)
495: {
497: fact->assembled = PETSC_TRUE;
498: fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqDense;
499: return(0);
500: }
504: PetscErrorCode MatLUFactorSymbolic_SeqDense(Mat fact,Mat A,IS row,IS col,const MatFactorInfo *info)
505: {
507: fact->preallocated = PETSC_TRUE;
508: fact->assembled = PETSC_TRUE;
509: fact->ops->lufactornumeric = MatLUFactorNumeric_SeqDense;
510: return(0);
511: }
513: EXTERN_C_BEGIN
516: PetscErrorCode MatGetFactor_seqdense_petsc(Mat A,MatFactorType ftype,Mat *fact)
517: {
521: MatCreate(((PetscObject)A)->comm,fact);
522: MatSetSizes(*fact,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);
523: MatSetType(*fact,((PetscObject)A)->type_name);
524: if (ftype == MAT_FACTOR_LU){
525: (*fact)->ops->lufactorsymbolic = MatLUFactorSymbolic_SeqDense;
526: } else {
527: (*fact)->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SeqDense;
528: }
529: (*fact)->factortype = ftype;
530: return(0);
531: }
532: EXTERN_C_END
534: /* ------------------------------------------------------------------*/
537: PetscErrorCode MatSOR_SeqDense(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal shift,PetscInt its,PetscInt lits,Vec xx)
538: {
539: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
540: PetscScalar *x,*b,*v = mat->v,zero = 0.0,xt;
542: PetscInt m = A->rmap->n,i;
543: PetscBLASInt o = 1,bm = PetscBLASIntCast(m);
546: if (flag & SOR_ZERO_INITIAL_GUESS) {
547: /* this is a hack fix, should have another version without the second BLASdot */
548: VecSet(xx,zero);
549: }
550: VecGetArray(xx,&x);
551: VecGetArray(bb,&b);
552: its = its*lits;
553: if (its <= 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
554: while (its--) {
555: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
556: for (i=0; i<m; i++) {
557: xt = b[i] - BLASdot_(&bm,v+i,&bm,x,&o);
558: x[i] = (1. - omega)*x[i] + omega*(xt+v[i + i*m]*x[i])/(v[i + i*m]+shift);
559: }
560: }
561: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
562: for (i=m-1; i>=0; i--) {
563: xt = b[i] - BLASdot_(&bm,v+i,&bm,x,&o);
564: x[i] = (1. - omega)*x[i] + omega*(xt+v[i + i*m]*x[i])/(v[i + i*m]+shift);
565: }
566: }
567: }
568: VecRestoreArray(bb,&b);
569: VecRestoreArray(xx,&x);
570: return(0);
571: }
573: /* -----------------------------------------------------------------*/
576: PetscErrorCode MatMultTranspose_SeqDense(Mat A,Vec xx,Vec yy)
577: {
578: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
579: PetscScalar *v = mat->v,*x,*y;
581: PetscBLASInt m, n,_One=1;
582: PetscScalar _DOne=1.0,_DZero=0.0;
585: m = PetscBLASIntCast(A->rmap->n);
586: n = PetscBLASIntCast(A->cmap->n);
587: if (!A->rmap->n || !A->cmap->n) return(0);
588: VecGetArray(xx,&x);
589: VecGetArray(yy,&y);
590: BLASgemv_("T",&m,&n,&_DOne,v,&mat->lda,x,&_One,&_DZero,y,&_One);
591: VecRestoreArray(xx,&x);
592: VecRestoreArray(yy,&y);
593: PetscLogFlops(2.0*A->rmap->n*A->cmap->n - A->cmap->n);
594: return(0);
595: }
599: PetscErrorCode MatMult_SeqDense(Mat A,Vec xx,Vec yy)
600: {
601: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
602: PetscScalar *v = mat->v,*x,*y,_DOne=1.0,_DZero=0.0;
604: PetscBLASInt m, n, _One=1;
607: m = PetscBLASIntCast(A->rmap->n);
608: n = PetscBLASIntCast(A->cmap->n);
609: if (!A->rmap->n || !A->cmap->n) return(0);
610: VecGetArray(xx,&x);
611: VecGetArray(yy,&y);
612: BLASgemv_("N",&m,&n,&_DOne,v,&(mat->lda),x,&_One,&_DZero,y,&_One);
613: VecRestoreArray(xx,&x);
614: VecRestoreArray(yy,&y);
615: PetscLogFlops(2.0*A->rmap->n*A->cmap->n - A->rmap->n);
616: return(0);
617: }
621: PetscErrorCode MatMultAdd_SeqDense(Mat A,Vec xx,Vec zz,Vec yy)
622: {
623: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
624: PetscScalar *v = mat->v,*x,*y,_DOne=1.0;
626: PetscBLASInt m, n, _One=1;
629: m = PetscBLASIntCast(A->rmap->n);
630: n = PetscBLASIntCast(A->cmap->n);
631: if (!A->rmap->n || !A->cmap->n) return(0);
632: if (zz != yy) {VecCopy(zz,yy);}
633: VecGetArray(xx,&x);
634: VecGetArray(yy,&y);
635: BLASgemv_("N",&m,&n,&_DOne,v,&(mat->lda),x,&_One,&_DOne,y,&_One);
636: VecRestoreArray(xx,&x);
637: VecRestoreArray(yy,&y);
638: PetscLogFlops(2.0*A->rmap->n*A->cmap->n);
639: return(0);
640: }
644: PetscErrorCode MatMultTransposeAdd_SeqDense(Mat A,Vec xx,Vec zz,Vec yy)
645: {
646: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
647: PetscScalar *v = mat->v,*x,*y;
649: PetscBLASInt m, n, _One=1;
650: PetscScalar _DOne=1.0;
653: m = PetscBLASIntCast(A->rmap->n);
654: n = PetscBLASIntCast(A->cmap->n);
655: if (!A->rmap->n || !A->cmap->n) return(0);
656: if (zz != yy) {VecCopy(zz,yy);}
657: VecGetArray(xx,&x);
658: VecGetArray(yy,&y);
659: BLASgemv_("T",&m,&n,&_DOne,v,&(mat->lda),x,&_One,&_DOne,y,&_One);
660: VecRestoreArray(xx,&x);
661: VecRestoreArray(yy,&y);
662: PetscLogFlops(2.0*A->rmap->n*A->cmap->n);
663: return(0);
664: }
666: /* -----------------------------------------------------------------*/
669: PetscErrorCode MatGetRow_SeqDense(Mat A,PetscInt row,PetscInt *ncols,PetscInt **cols,PetscScalar **vals)
670: {
671: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
672: PetscScalar *v;
674: PetscInt i;
675:
677: *ncols = A->cmap->n;
678: if (cols) {
679: PetscMalloc((A->cmap->n+1)*sizeof(PetscInt),cols);
680: for (i=0; i<A->cmap->n; i++) (*cols)[i] = i;
681: }
682: if (vals) {
683: PetscMalloc((A->cmap->n+1)*sizeof(PetscScalar),vals);
684: v = mat->v + row;
685: for (i=0; i<A->cmap->n; i++) {(*vals)[i] = *v; v += mat->lda;}
686: }
687: return(0);
688: }
692: PetscErrorCode MatRestoreRow_SeqDense(Mat A,PetscInt row,PetscInt *ncols,PetscInt **cols,PetscScalar **vals)
693: {
696: if (cols) {PetscFree(*cols);}
697: if (vals) {PetscFree(*vals); }
698: return(0);
699: }
700: /* ----------------------------------------------------------------*/
703: PetscErrorCode MatSetValues_SeqDense(Mat A,PetscInt m,const PetscInt indexm[],PetscInt n,const PetscInt indexn[],const PetscScalar v[],InsertMode addv)
704: {
705: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
706: PetscInt i,j,idx=0;
707:
710: if (!mat->roworiented) {
711: if (addv == INSERT_VALUES) {
712: for (j=0; j<n; j++) {
713: if (indexn[j] < 0) {idx += m; continue;}
714: #if defined(PETSC_USE_DEBUG)
715: if (indexn[j] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",indexn[j],A->cmap->n-1);
716: #endif
717: for (i=0; i<m; i++) {
718: if (indexm[i] < 0) {idx++; continue;}
719: #if defined(PETSC_USE_DEBUG)
720: if (indexm[i] >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",indexm[i],A->rmap->n-1);
721: #endif
722: mat->v[indexn[j]*mat->lda + indexm[i]] = v[idx++];
723: }
724: }
725: } else {
726: for (j=0; j<n; j++) {
727: if (indexn[j] < 0) {idx += m; continue;}
728: #if defined(PETSC_USE_DEBUG)
729: if (indexn[j] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",indexn[j],A->cmap->n-1);
730: #endif
731: for (i=0; i<m; i++) {
732: if (indexm[i] < 0) {idx++; continue;}
733: #if defined(PETSC_USE_DEBUG)
734: if (indexm[i] >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",indexm[i],A->rmap->n-1);
735: #endif
736: mat->v[indexn[j]*mat->lda + indexm[i]] += v[idx++];
737: }
738: }
739: }
740: } else {
741: if (addv == INSERT_VALUES) {
742: for (i=0; i<m; i++) {
743: if (indexm[i] < 0) { idx += n; continue;}
744: #if defined(PETSC_USE_DEBUG)
745: if (indexm[i] >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",indexm[i],A->rmap->n-1);
746: #endif
747: for (j=0; j<n; j++) {
748: if (indexn[j] < 0) { idx++; continue;}
749: #if defined(PETSC_USE_DEBUG)
750: if (indexn[j] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",indexn[j],A->cmap->n-1);
751: #endif
752: mat->v[indexn[j]*mat->lda + indexm[i]] = v[idx++];
753: }
754: }
755: } else {
756: for (i=0; i<m; i++) {
757: if (indexm[i] < 0) { idx += n; continue;}
758: #if defined(PETSC_USE_DEBUG)
759: if (indexm[i] >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",indexm[i],A->rmap->n-1);
760: #endif
761: for (j=0; j<n; j++) {
762: if (indexn[j] < 0) { idx++; continue;}
763: #if defined(PETSC_USE_DEBUG)
764: if (indexn[j] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",indexn[j],A->cmap->n-1);
765: #endif
766: mat->v[indexn[j]*mat->lda + indexm[i]] += v[idx++];
767: }
768: }
769: }
770: }
771: return(0);
772: }
776: PetscErrorCode MatGetValues_SeqDense(Mat A,PetscInt m,const PetscInt indexm[],PetscInt n,const PetscInt indexn[],PetscScalar v[])
777: {
778: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
779: PetscInt i,j;
782: /* row-oriented output */
783: for (i=0; i<m; i++) {
784: if (indexm[i] < 0) {v += n;continue;}
785: if (indexm[i] >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D requested larger than number rows %D",indexm[i],A->rmap->n);
786: for (j=0; j<n; j++) {
787: if (indexn[j] < 0) {v++; continue;}
788: if (indexn[j] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column %D requested larger than number columns %D",indexn[j],A->cmap->n);
789: *v++ = mat->v[indexn[j]*mat->lda + indexm[i]];
790: }
791: }
792: return(0);
793: }
795: /* -----------------------------------------------------------------*/
799: PetscErrorCode MatLoad_SeqDense(Mat newmat,PetscViewer viewer)
800: {
801: Mat_SeqDense *a;
803: PetscInt *scols,i,j,nz,header[4];
804: int fd;
805: PetscMPIInt size;
806: PetscInt *rowlengths = 0,M,N,*cols,grows,gcols;
807: PetscScalar *vals,*svals,*v,*w;
808: MPI_Comm comm = ((PetscObject)viewer)->comm;
811: MPI_Comm_size(comm,&size);
812: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"view must have one processor");
813: PetscViewerBinaryGetDescriptor(viewer,&fd);
814: PetscBinaryRead(fd,header,4,PETSC_INT);
815: if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Not matrix object");
816: M = header[1]; N = header[2]; nz = header[3];
818: /* set global size if not set already*/
819: if (newmat->rmap->n < 0 && newmat->rmap->N < 0 && newmat->cmap->n < 0 && newmat->cmap->N < 0) {
820: MatSetSizes(newmat,M,N,M,N);
821: } else {
822: /* if sizes and type are already set, check if the vector global sizes are correct */
823: MatGetSize(newmat,&grows,&gcols);
824: if (M != grows || N != gcols) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different length (%d, %d) than the input matrix (%d, %d)",M,N,grows,gcols);
825: }
826: MatSeqDenseSetPreallocation(newmat,PETSC_NULL);
827:
828: if (nz == MATRIX_BINARY_FORMAT_DENSE) { /* matrix in file is dense */
829: a = (Mat_SeqDense*)newmat->data;
830: v = a->v;
831: /* Allocate some temp space to read in the values and then flip them
832: from row major to column major */
833: PetscMalloc((M*N > 0 ? M*N : 1)*sizeof(PetscScalar),&w);
834: /* read in nonzero values */
835: PetscBinaryRead(fd,w,M*N,PETSC_SCALAR);
836: /* now flip the values and store them in the matrix*/
837: for (j=0; j<N; j++) {
838: for (i=0; i<M; i++) {
839: *v++ =w[i*N+j];
840: }
841: }
842: PetscFree(w);
843: } else {
844: /* read row lengths */
845: PetscMalloc((M+1)*sizeof(PetscInt),&rowlengths);
846: PetscBinaryRead(fd,rowlengths,M,PETSC_INT);
848: a = (Mat_SeqDense*)newmat->data;
849: v = a->v;
851: /* read column indices and nonzeros */
852: PetscMalloc((nz+1)*sizeof(PetscInt),&scols);
853: cols = scols;
854: PetscBinaryRead(fd,cols,nz,PETSC_INT);
855: PetscMalloc((nz+1)*sizeof(PetscScalar),&svals);
856: vals = svals;
857: PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);
859: /* insert into matrix */
860: for (i=0; i<M; i++) {
861: for (j=0; j<rowlengths[i]; j++) v[i+M*scols[j]] = svals[j];
862: svals += rowlengths[i]; scols += rowlengths[i];
863: }
864: PetscFree(vals);
865: PetscFree(cols);
866: PetscFree(rowlengths);
867: }
868: MatAssemblyBegin(newmat,MAT_FINAL_ASSEMBLY);
869: MatAssemblyEnd(newmat,MAT_FINAL_ASSEMBLY);
871: return(0);
872: }
876: static PetscErrorCode MatView_SeqDense_ASCII(Mat A,PetscViewer viewer)
877: {
878: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
879: PetscErrorCode ierr;
880: PetscInt i,j;
881: const char *name;
882: PetscScalar *v;
883: PetscViewerFormat format;
884: #if defined(PETSC_USE_COMPLEX)
885: PetscBool allreal = PETSC_TRUE;
886: #endif
889: PetscViewerGetFormat(viewer,&format);
890: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
891: return(0); /* do nothing for now */
892: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
893: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
894: PetscObjectPrintClassNamePrefixType((PetscObject)A,viewer,"Matrix Object");
895: for (i=0; i<A->rmap->n; i++) {
896: v = a->v + i;
897: PetscViewerASCIIPrintf(viewer,"row %D:",i);
898: for (j=0; j<A->cmap->n; j++) {
899: #if defined(PETSC_USE_COMPLEX)
900: if (PetscRealPart(*v) != 0.0 && PetscImaginaryPart(*v) != 0.0) {
901: PetscViewerASCIIPrintf(viewer," (%D, %G + %G i) ",j,PetscRealPart(*v),PetscImaginaryPart(*v));
902: } else if (PetscRealPart(*v)) {
903: PetscViewerASCIIPrintf(viewer," (%D, %G) ",j,PetscRealPart(*v));
904: }
905: #else
906: if (*v) {
907: PetscViewerASCIIPrintf(viewer," (%D, %G) ",j,*v);
908: }
909: #endif
910: v += a->lda;
911: }
912: PetscViewerASCIIPrintf(viewer,"\n");
913: }
914: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
915: } else {
916: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
917: #if defined(PETSC_USE_COMPLEX)
918: /* determine if matrix has all real values */
919: v = a->v;
920: for (i=0; i<A->rmap->n*A->cmap->n; i++) {
921: if (PetscImaginaryPart(v[i])) { allreal = PETSC_FALSE; break ;}
922: }
923: #endif
924: if (format == PETSC_VIEWER_ASCII_MATLAB) {
925: PetscObjectGetName((PetscObject)A,&name);
926: PetscViewerASCIIPrintf(viewer,"%% Size = %D %D \n",A->rmap->n,A->cmap->n);
927: PetscViewerASCIIPrintf(viewer,"%s = zeros(%D,%D);\n",name,A->rmap->n,A->cmap->n);
928: PetscViewerASCIIPrintf(viewer,"%s = [\n",name);
929: } else {
930: PetscObjectPrintClassNamePrefixType((PetscObject)A,viewer,"Matrix Object");
931: }
933: for (i=0; i<A->rmap->n; i++) {
934: v = a->v + i;
935: for (j=0; j<A->cmap->n; j++) {
936: #if defined(PETSC_USE_COMPLEX)
937: if (allreal) {
938: PetscViewerASCIIPrintf(viewer,"%18.16e ",PetscRealPart(*v));
939: } else {
940: PetscViewerASCIIPrintf(viewer,"%18.16e + %18.16ei ",PetscRealPart(*v),PetscImaginaryPart(*v));
941: }
942: #else
943: PetscViewerASCIIPrintf(viewer,"%18.16e ",*v);
944: #endif
945: v += a->lda;
946: }
947: PetscViewerASCIIPrintf(viewer,"\n");
948: }
949: if (format == PETSC_VIEWER_ASCII_MATLAB) {
950: PetscViewerASCIIPrintf(viewer,"];\n");
951: }
952: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
953: }
954: PetscViewerFlush(viewer);
955: return(0);
956: }
960: static PetscErrorCode MatView_SeqDense_Binary(Mat A,PetscViewer viewer)
961: {
962: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
963: PetscErrorCode ierr;
964: int fd;
965: PetscInt ict,j,n = A->cmap->n,m = A->rmap->n,i,*col_lens,nz = m*n;
966: PetscScalar *v,*anonz,*vals;
967: PetscViewerFormat format;
968:
970: PetscViewerBinaryGetDescriptor(viewer,&fd);
972: PetscViewerGetFormat(viewer,&format);
973: if (format == PETSC_VIEWER_NATIVE) {
974: /* store the matrix as a dense matrix */
975: PetscMalloc(4*sizeof(PetscInt),&col_lens);
976: col_lens[0] = MAT_FILE_CLASSID;
977: col_lens[1] = m;
978: col_lens[2] = n;
979: col_lens[3] = MATRIX_BINARY_FORMAT_DENSE;
980: PetscBinaryWrite(fd,col_lens,4,PETSC_INT,PETSC_TRUE);
981: PetscFree(col_lens);
983: /* write out matrix, by rows */
984: PetscMalloc((m*n+1)*sizeof(PetscScalar),&vals);
985: v = a->v;
986: for (j=0; j<n; j++) {
987: for (i=0; i<m; i++) {
988: vals[j + i*n] = *v++;
989: }
990: }
991: PetscBinaryWrite(fd,vals,n*m,PETSC_SCALAR,PETSC_FALSE);
992: PetscFree(vals);
993: } else {
994: PetscMalloc((4+nz)*sizeof(PetscInt),&col_lens);
995: col_lens[0] = MAT_FILE_CLASSID;
996: col_lens[1] = m;
997: col_lens[2] = n;
998: col_lens[3] = nz;
1000: /* store lengths of each row and write (including header) to file */
1001: for (i=0; i<m; i++) col_lens[4+i] = n;
1002: PetscBinaryWrite(fd,col_lens,4+m,PETSC_INT,PETSC_TRUE);
1004: /* Possibly should write in smaller increments, not whole matrix at once? */
1005: /* store column indices (zero start index) */
1006: ict = 0;
1007: for (i=0; i<m; i++) {
1008: for (j=0; j<n; j++) col_lens[ict++] = j;
1009: }
1010: PetscBinaryWrite(fd,col_lens,nz,PETSC_INT,PETSC_FALSE);
1011: PetscFree(col_lens);
1013: /* store nonzero values */
1014: PetscMalloc((nz+1)*sizeof(PetscScalar),&anonz);
1015: ict = 0;
1016: for (i=0; i<m; i++) {
1017: v = a->v + i;
1018: for (j=0; j<n; j++) {
1019: anonz[ict++] = *v; v += a->lda;
1020: }
1021: }
1022: PetscBinaryWrite(fd,anonz,nz,PETSC_SCALAR,PETSC_FALSE);
1023: PetscFree(anonz);
1024: }
1025: return(0);
1026: }
1030: PetscErrorCode MatView_SeqDense_Draw_Zoom(PetscDraw draw,void *Aa)
1031: {
1032: Mat A = (Mat) Aa;
1033: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1034: PetscErrorCode ierr;
1035: PetscInt m = A->rmap->n,n = A->cmap->n,color,i,j;
1036: PetscScalar *v = a->v;
1037: PetscViewer viewer;
1038: PetscDraw popup;
1039: PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r,scale,maxv = 0.0;
1040: PetscViewerFormat format;
1044: PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);
1045: PetscViewerGetFormat(viewer,&format);
1046: PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);
1048: /* Loop over matrix elements drawing boxes */
1049: if (format != PETSC_VIEWER_DRAW_CONTOUR) {
1050: /* Blue for negative and Red for positive */
1051: color = PETSC_DRAW_BLUE;
1052: for(j = 0; j < n; j++) {
1053: x_l = j;
1054: x_r = x_l + 1.0;
1055: for(i = 0; i < m; i++) {
1056: y_l = m - i - 1.0;
1057: y_r = y_l + 1.0;
1058: #if defined(PETSC_USE_COMPLEX)
1059: if (PetscRealPart(v[j*m+i]) > 0.) {
1060: color = PETSC_DRAW_RED;
1061: } else if (PetscRealPart(v[j*m+i]) < 0.) {
1062: color = PETSC_DRAW_BLUE;
1063: } else {
1064: continue;
1065: }
1066: #else
1067: if (v[j*m+i] > 0.) {
1068: color = PETSC_DRAW_RED;
1069: } else if (v[j*m+i] < 0.) {
1070: color = PETSC_DRAW_BLUE;
1071: } else {
1072: continue;
1073: }
1074: #endif
1075: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
1076: }
1077: }
1078: } else {
1079: /* use contour shading to indicate magnitude of values */
1080: /* first determine max of all nonzero values */
1081: for(i = 0; i < m*n; i++) {
1082: if (PetscAbsScalar(v[i]) > maxv) maxv = PetscAbsScalar(v[i]);
1083: }
1084: scale = (245.0 - PETSC_DRAW_BASIC_COLORS)/maxv;
1085: PetscDrawGetPopup(draw,&popup);
1086: if (popup) {PetscDrawScalePopup(popup,0.0,maxv);}
1087: for(j = 0; j < n; j++) {
1088: x_l = j;
1089: x_r = x_l + 1.0;
1090: for(i = 0; i < m; i++) {
1091: y_l = m - i - 1.0;
1092: y_r = y_l + 1.0;
1093: color = PETSC_DRAW_BASIC_COLORS + (int)(scale*PetscAbsScalar(v[j*m+i]));
1094: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
1095: }
1096: }
1097: }
1098: return(0);
1099: }
1103: PetscErrorCode MatView_SeqDense_Draw(Mat A,PetscViewer viewer)
1104: {
1105: PetscDraw draw;
1106: PetscBool isnull;
1107: PetscReal xr,yr,xl,yl,h,w;
1111: PetscViewerDrawGetDraw(viewer,0,&draw);
1112: PetscDrawIsNull(draw,&isnull);
1113: if (isnull) return(0);
1115: PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);
1116: xr = A->cmap->n; yr = A->rmap->n; h = yr/10.0; w = xr/10.0;
1117: xr += w; yr += h; xl = -w; yl = -h;
1118: PetscDrawSetCoordinates(draw,xl,yl,xr,yr);
1119: PetscDrawZoom(draw,MatView_SeqDense_Draw_Zoom,A);
1120: PetscObjectCompose((PetscObject)A,"Zoomviewer",PETSC_NULL);
1121: return(0);
1122: }
1126: PetscErrorCode MatView_SeqDense(Mat A,PetscViewer viewer)
1127: {
1129: PetscBool iascii,isbinary,isdraw;
1132: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
1133: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
1134: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
1136: if (iascii) {
1137: MatView_SeqDense_ASCII(A,viewer);
1138: } else if (isbinary) {
1139: MatView_SeqDense_Binary(A,viewer);
1140: } else if (isdraw) {
1141: MatView_SeqDense_Draw(A,viewer);
1142: } else {
1143: SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"Viewer type %s not supported by dense matrix",((PetscObject)viewer)->type_name);
1144: }
1145: return(0);
1146: }
1150: PetscErrorCode MatDestroy_SeqDense(Mat mat)
1151: {
1152: Mat_SeqDense *l = (Mat_SeqDense*)mat->data;
1156: #if defined(PETSC_USE_LOG)
1157: PetscLogObjectState((PetscObject)mat,"Rows %D Cols %D",mat->rmap->n,mat->cmap->n);
1158: #endif
1159: PetscFree(l->pivots);
1160: if (!l->user_alloc) {PetscFree(l->v);}
1161: PetscFree(mat->data);
1163: PetscObjectChangeTypeName((PetscObject)mat,0);
1164: PetscObjectComposeFunctionDynamic((PetscObject)mat,"MatSeqDenseSetPreallocation_C","",PETSC_NULL);
1165: PetscObjectComposeFunctionDynamic((PetscObject)mat,"MatMatMult_seqaij_seqdense_C","",PETSC_NULL);
1166: PetscObjectComposeFunctionDynamic((PetscObject)mat,"MatMatMultSymbolic_seqaij_seqdense_C","",PETSC_NULL);
1167: PetscObjectComposeFunctionDynamic((PetscObject)mat,"MatMatMultNumeric_seqaij_seqdense_C","",PETSC_NULL);
1168: return(0);
1169: }
1173: PetscErrorCode MatTranspose_SeqDense(Mat A,MatReuse reuse,Mat *matout)
1174: {
1175: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
1177: PetscInt k,j,m,n,M;
1178: PetscScalar *v,tmp;
1181: v = mat->v; m = A->rmap->n; M = mat->lda; n = A->cmap->n;
1182: if (reuse == MAT_REUSE_MATRIX && *matout == A) { /* in place transpose */
1183: if (m != n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Can not transpose non-square matrix in place");
1184: else {
1185: for (j=0; j<m; j++) {
1186: for (k=0; k<j; k++) {
1187: tmp = v[j + k*M];
1188: v[j + k*M] = v[k + j*M];
1189: v[k + j*M] = tmp;
1190: }
1191: }
1192: }
1193: } else { /* out-of-place transpose */
1194: Mat tmat;
1195: Mat_SeqDense *tmatd;
1196: PetscScalar *v2;
1198: if (reuse == MAT_INITIAL_MATRIX) {
1199: MatCreate(((PetscObject)A)->comm,&tmat);
1200: MatSetSizes(tmat,A->cmap->n,A->rmap->n,A->cmap->n,A->rmap->n);
1201: MatSetType(tmat,((PetscObject)A)->type_name);
1202: MatSeqDenseSetPreallocation(tmat,PETSC_NULL);
1203: } else {
1204: tmat = *matout;
1205: }
1206: tmatd = (Mat_SeqDense*)tmat->data;
1207: v = mat->v; v2 = tmatd->v;
1208: for (j=0; j<n; j++) {
1209: for (k=0; k<m; k++) v2[j + k*n] = v[k + j*M];
1210: }
1211: MatAssemblyBegin(tmat,MAT_FINAL_ASSEMBLY);
1212: MatAssemblyEnd(tmat,MAT_FINAL_ASSEMBLY);
1213: *matout = tmat;
1214: }
1215: return(0);
1216: }
1220: PetscErrorCode MatEqual_SeqDense(Mat A1,Mat A2,PetscBool *flg)
1221: {
1222: Mat_SeqDense *mat1 = (Mat_SeqDense*)A1->data;
1223: Mat_SeqDense *mat2 = (Mat_SeqDense*)A2->data;
1224: PetscInt i,j;
1225: PetscScalar *v1 = mat1->v,*v2 = mat2->v;
1228: if (A1->rmap->n != A2->rmap->n) {*flg = PETSC_FALSE; return(0);}
1229: if (A1->cmap->n != A2->cmap->n) {*flg = PETSC_FALSE; return(0);}
1230: for (i=0; i<A1->rmap->n; i++) {
1231: v1 = mat1->v+i; v2 = mat2->v+i;
1232: for (j=0; j<A1->cmap->n; j++) {
1233: if (*v1 != *v2) {*flg = PETSC_FALSE; return(0);}
1234: v1 += mat1->lda; v2 += mat2->lda;
1235: }
1236: }
1237: *flg = PETSC_TRUE;
1238: return(0);
1239: }
1243: PetscErrorCode MatGetDiagonal_SeqDense(Mat A,Vec v)
1244: {
1245: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
1247: PetscInt i,n,len;
1248: PetscScalar *x,zero = 0.0;
1251: VecSet(v,zero);
1252: VecGetSize(v,&n);
1253: VecGetArray(v,&x);
1254: len = PetscMin(A->rmap->n,A->cmap->n);
1255: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming mat and vec");
1256: for (i=0; i<len; i++) {
1257: x[i] = mat->v[i*mat->lda + i];
1258: }
1259: VecRestoreArray(v,&x);
1260: return(0);
1261: }
1265: PetscErrorCode MatDiagonalScale_SeqDense(Mat A,Vec ll,Vec rr)
1266: {
1267: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
1268: PetscScalar *l,*r,x,*v;
1270: PetscInt i,j,m = A->rmap->n,n = A->cmap->n;
1273: if (ll) {
1274: VecGetSize(ll,&m);
1275: VecGetArray(ll,&l);
1276: if (m != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Left scaling vec wrong size");
1277: for (i=0; i<m; i++) {
1278: x = l[i];
1279: v = mat->v + i;
1280: for (j=0; j<n; j++) { (*v) *= x; v+= m;}
1281: }
1282: VecRestoreArray(ll,&l);
1283: PetscLogFlops(n*m);
1284: }
1285: if (rr) {
1286: VecGetSize(rr,&n);
1287: VecGetArray(rr,&r);
1288: if (n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Right scaling vec wrong size");
1289: for (i=0; i<n; i++) {
1290: x = r[i];
1291: v = mat->v + i*m;
1292: for (j=0; j<m; j++) { (*v++) *= x;}
1293: }
1294: VecRestoreArray(rr,&r);
1295: PetscLogFlops(n*m);
1296: }
1297: return(0);
1298: }
1302: PetscErrorCode MatNorm_SeqDense(Mat A,NormType type,PetscReal *nrm)
1303: {
1304: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
1305: PetscScalar *v = mat->v;
1306: PetscReal sum = 0.0;
1307: PetscInt lda=mat->lda,m=A->rmap->n,i,j;
1311: if (type == NORM_FROBENIUS) {
1312: if (lda>m) {
1313: for (j=0; j<A->cmap->n; j++) {
1314: v = mat->v+j*lda;
1315: for (i=0; i<m; i++) {
1316: #if defined(PETSC_USE_COMPLEX)
1317: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
1318: #else
1319: sum += (*v)*(*v); v++;
1320: #endif
1321: }
1322: }
1323: } else {
1324: for (i=0; i<A->cmap->n*A->rmap->n; i++) {
1325: #if defined(PETSC_USE_COMPLEX)
1326: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
1327: #else
1328: sum += (*v)*(*v); v++;
1329: #endif
1330: }
1331: }
1332: *nrm = PetscSqrtReal(sum);
1333: PetscLogFlops(2.0*A->cmap->n*A->rmap->n);
1334: } else if (type == NORM_1) {
1335: *nrm = 0.0;
1336: for (j=0; j<A->cmap->n; j++) {
1337: v = mat->v + j*mat->lda;
1338: sum = 0.0;
1339: for (i=0; i<A->rmap->n; i++) {
1340: sum += PetscAbsScalar(*v); v++;
1341: }
1342: if (sum > *nrm) *nrm = sum;
1343: }
1344: PetscLogFlops(A->cmap->n*A->rmap->n);
1345: } else if (type == NORM_INFINITY) {
1346: *nrm = 0.0;
1347: for (j=0; j<A->rmap->n; j++) {
1348: v = mat->v + j;
1349: sum = 0.0;
1350: for (i=0; i<A->cmap->n; i++) {
1351: sum += PetscAbsScalar(*v); v += mat->lda;
1352: }
1353: if (sum > *nrm) *nrm = sum;
1354: }
1355: PetscLogFlops(A->cmap->n*A->rmap->n);
1356: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No two norm");
1357: return(0);
1358: }
1362: PetscErrorCode MatSetOption_SeqDense(Mat A,MatOption op,PetscBool flg)
1363: {
1364: Mat_SeqDense *aij = (Mat_SeqDense*)A->data;
1366:
1368: switch (op) {
1369: case MAT_ROW_ORIENTED:
1370: aij->roworiented = flg;
1371: break;
1372: case MAT_NEW_NONZERO_LOCATIONS:
1373: case MAT_NEW_NONZERO_LOCATION_ERR:
1374: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1375: case MAT_NEW_DIAGONALS:
1376: case MAT_KEEP_NONZERO_PATTERN:
1377: case MAT_IGNORE_OFF_PROC_ENTRIES:
1378: case MAT_USE_HASH_TABLE:
1379: case MAT_SYMMETRIC:
1380: case MAT_STRUCTURALLY_SYMMETRIC:
1381: case MAT_HERMITIAN:
1382: case MAT_SYMMETRY_ETERNAL:
1383: case MAT_IGNORE_LOWER_TRIANGULAR:
1384: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
1385: break;
1386: default:
1387: SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %s",MatOptions[op]);
1388: }
1389: return(0);
1390: }
1394: PetscErrorCode MatZeroEntries_SeqDense(Mat A)
1395: {
1396: Mat_SeqDense *l = (Mat_SeqDense*)A->data;
1398: PetscInt lda=l->lda,m=A->rmap->n,j;
1401: if (lda>m) {
1402: for (j=0; j<A->cmap->n; j++) {
1403: PetscMemzero(l->v+j*lda,m*sizeof(PetscScalar));
1404: }
1405: } else {
1406: PetscMemzero(l->v,A->rmap->n*A->cmap->n*sizeof(PetscScalar));
1407: }
1408: return(0);
1409: }
1413: PetscErrorCode MatZeroRows_SeqDense(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
1414: {
1415: PetscErrorCode ierr;
1416: Mat_SeqDense *l = (Mat_SeqDense*)A->data;
1417: PetscInt m = l->lda, n = A->cmap->n, i,j;
1418: PetscScalar *slot,*bb;
1419: const PetscScalar *xx;
1422: #if defined(PETSC_USE_DEBUG)
1423: for (i=0; i<N; i++) {
1424: if (rows[i] < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row requested to be zeroed");
1425: if (rows[i] >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D requested to be zeroed greater than or equal number of rows %D",rows[i],A->rmap->n);
1426: }
1427: #endif
1429: /* fix right hand side if needed */
1430: if (x && b) {
1431: VecGetArrayRead(x,&xx);
1432: VecGetArray(b,&bb);
1433: for (i=0; i<N; i++) {
1434: bb[rows[i]] = diag*xx[rows[i]];
1435: }
1436: VecRestoreArrayRead(x,&xx);
1437: VecRestoreArray(b,&bb);
1438: }
1440: for (i=0; i<N; i++) {
1441: slot = l->v + rows[i];
1442: for (j=0; j<n; j++) { *slot = 0.0; slot += m;}
1443: }
1444: if (diag != 0.0) {
1445: if (A->rmap->n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only coded for square matrices");
1446: for (i=0; i<N; i++) {
1447: slot = l->v + (m+1)*rows[i];
1448: *slot = diag;
1449: }
1450: }
1451: return(0);
1452: }
1456: PetscErrorCode MatGetArray_SeqDense(Mat A,PetscScalar *array[])
1457: {
1458: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
1461: if (mat->lda != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Cannot get array for Dense matrices with LDA different from number of rows");
1462: *array = mat->v;
1463: return(0);
1464: }
1468: PetscErrorCode MatRestoreArray_SeqDense(Mat A,PetscScalar *array[])
1469: {
1471: *array = 0; /* user cannot accidently use the array later */
1472: return(0);
1473: }
1477: static PetscErrorCode MatGetSubMatrix_SeqDense(Mat A,IS isrow,IS iscol,PetscInt cs,MatReuse scall,Mat *B)
1478: {
1479: Mat_SeqDense *mat = (Mat_SeqDense*)A->data;
1481: PetscInt i,j,nrows,ncols;
1482: const PetscInt *irow,*icol;
1483: PetscScalar *av,*bv,*v = mat->v;
1484: Mat newmat;
1487: ISGetIndices(isrow,&irow);
1488: ISGetIndices(iscol,&icol);
1489: ISGetLocalSize(isrow,&nrows);
1490: ISGetLocalSize(iscol,&ncols);
1491:
1492: /* Check submatrixcall */
1493: if (scall == MAT_REUSE_MATRIX) {
1494: PetscInt n_cols,n_rows;
1495: MatGetSize(*B,&n_rows,&n_cols);
1496: if (n_rows != nrows || n_cols != ncols) {
1497: /* resize the result matrix to match number of requested rows/columns */
1498: MatSetSizes(*B,nrows,ncols,nrows,ncols);
1499: }
1500: newmat = *B;
1501: } else {
1502: /* Create and fill new matrix */
1503: MatCreate(((PetscObject)A)->comm,&newmat);
1504: MatSetSizes(newmat,nrows,ncols,nrows,ncols);
1505: MatSetType(newmat,((PetscObject)A)->type_name);
1506: MatSeqDenseSetPreallocation(newmat,PETSC_NULL);
1507: }
1509: /* Now extract the data pointers and do the copy,column at a time */
1510: bv = ((Mat_SeqDense*)newmat->data)->v;
1511:
1512: for (i=0; i<ncols; i++) {
1513: av = v + mat->lda*icol[i];
1514: for (j=0; j<nrows; j++) {
1515: *bv++ = av[irow[j]];
1516: }
1517: }
1519: /* Assemble the matrices so that the correct flags are set */
1520: MatAssemblyBegin(newmat,MAT_FINAL_ASSEMBLY);
1521: MatAssemblyEnd(newmat,MAT_FINAL_ASSEMBLY);
1523: /* Free work space */
1524: ISRestoreIndices(isrow,&irow);
1525: ISRestoreIndices(iscol,&icol);
1526: *B = newmat;
1527: return(0);
1528: }
1532: PetscErrorCode MatGetSubMatrices_SeqDense(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[])
1533: {
1535: PetscInt i;
1538: if (scall == MAT_INITIAL_MATRIX) {
1539: PetscMalloc((n+1)*sizeof(Mat),B);
1540: }
1542: for (i=0; i<n; i++) {
1543: MatGetSubMatrix_SeqDense(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);
1544: }
1545: return(0);
1546: }
1550: PetscErrorCode MatAssemblyBegin_SeqDense(Mat mat,MatAssemblyType mode)
1551: {
1553: return(0);
1554: }
1558: PetscErrorCode MatAssemblyEnd_SeqDense(Mat mat,MatAssemblyType mode)
1559: {
1561: return(0);
1562: }
1566: PetscErrorCode MatCopy_SeqDense(Mat A,Mat B,MatStructure str)
1567: {
1568: Mat_SeqDense *a = (Mat_SeqDense*)A->data,*b = (Mat_SeqDense *)B->data;
1570: PetscInt lda1=a->lda,lda2=b->lda, m=A->rmap->n,n=A->cmap->n, j;
1573: /* If the two matrices don't have the same copy implementation, they aren't compatible for fast copy. */
1574: if (A->ops->copy != B->ops->copy) {
1575: MatCopy_Basic(A,B,str);
1576: return(0);
1577: }
1578: if (m != B->rmap->n || n != B->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"size(B) != size(A)");
1579: if (lda1>m || lda2>m) {
1580: for (j=0; j<n; j++) {
1581: PetscMemcpy(b->v+j*lda2,a->v+j*lda1,m*sizeof(PetscScalar));
1582: }
1583: } else {
1584: PetscMemcpy(b->v,a->v,A->rmap->n*A->cmap->n*sizeof(PetscScalar));
1585: }
1586: return(0);
1587: }
1591: PetscErrorCode MatSetUp_SeqDense(Mat A)
1592: {
1596: MatSeqDenseSetPreallocation(A,0);
1597: return(0);
1598: }
1602: static PetscErrorCode MatConjugate_SeqDense(Mat A)
1603: {
1604: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1605: PetscInt i,nz = A->rmap->n*A->cmap->n;
1606: PetscScalar *aa = a->v;
1609: for (i=0; i<nz; i++) aa[i] = PetscConj(aa[i]);
1610: return(0);
1611: }
1615: static PetscErrorCode MatRealPart_SeqDense(Mat A)
1616: {
1617: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1618: PetscInt i,nz = A->rmap->n*A->cmap->n;
1619: PetscScalar *aa = a->v;
1622: for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]);
1623: return(0);
1624: }
1628: static PetscErrorCode MatImaginaryPart_SeqDense(Mat A)
1629: {
1630: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1631: PetscInt i,nz = A->rmap->n*A->cmap->n;
1632: PetscScalar *aa = a->v;
1635: for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]);
1636: return(0);
1637: }
1639: /* ----------------------------------------------------------------*/
1642: PetscErrorCode MatMatMult_SeqDense_SeqDense(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C)
1643: {
1647: if (scall == MAT_INITIAL_MATRIX){
1648: MatMatMultSymbolic_SeqDense_SeqDense(A,B,fill,C);
1649: }
1650: MatMatMultNumeric_SeqDense_SeqDense(A,B,*C);
1651: return(0);
1652: }
1656: PetscErrorCode MatMatMultSymbolic_SeqDense_SeqDense(Mat A,Mat B,PetscReal fill,Mat *C)
1657: {
1659: PetscInt m=A->rmap->n,n=B->cmap->n;
1660: Mat Cmat;
1663: if (A->cmap->n != B->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"A->cmap->n %d != B->rmap->n %d\n",A->cmap->n,B->rmap->n);
1664: MatCreate(PETSC_COMM_SELF,&Cmat);
1665: MatSetSizes(Cmat,m,n,m,n);
1666: MatSetType(Cmat,MATSEQDENSE);
1667: MatSeqDenseSetPreallocation(Cmat,PETSC_NULL);
1668: Cmat->assembled = PETSC_TRUE;
1669: Cmat->ops->matmult = MatMatMult_SeqDense_SeqDense;
1670: *C = Cmat;
1671: return(0);
1672: }
1676: PetscErrorCode MatMatMultNumeric_SeqDense_SeqDense(Mat A,Mat B,Mat C)
1677: {
1678: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1679: Mat_SeqDense *b = (Mat_SeqDense*)B->data;
1680: Mat_SeqDense *c = (Mat_SeqDense*)C->data;
1681: PetscBLASInt m,n,k;
1682: PetscScalar _DOne=1.0,_DZero=0.0;
1685: m = PetscBLASIntCast(A->rmap->n);
1686: n = PetscBLASIntCast(B->cmap->n);
1687: k = PetscBLASIntCast(A->cmap->n);
1688: BLASgemm_("N","N",&m,&n,&k,&_DOne,a->v,&a->lda,b->v,&b->lda,&_DZero,c->v,&c->lda);
1689: return(0);
1690: }
1694: PetscErrorCode MatTransposeMatMult_SeqDense_SeqDense(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C)
1695: {
1699: if (scall == MAT_INITIAL_MATRIX){
1700: MatTransposeMatMultSymbolic_SeqDense_SeqDense(A,B,fill,C);
1701: }
1702: MatTransposeMatMultNumeric_SeqDense_SeqDense(A,B,*C);
1703: return(0);
1704: }
1708: PetscErrorCode MatTransposeMatMultSymbolic_SeqDense_SeqDense(Mat A,Mat B,PetscReal fill,Mat *C)
1709: {
1711: PetscInt m=A->cmap->n,n=B->cmap->n;
1712: Mat Cmat;
1715: if (A->rmap->n != B->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"A->rmap->n %d != B->rmap->n %d\n",A->rmap->n,B->rmap->n);
1716: MatCreate(PETSC_COMM_SELF,&Cmat);
1717: MatSetSizes(Cmat,m,n,m,n);
1718: MatSetType(Cmat,MATSEQDENSE);
1719: MatSeqDenseSetPreallocation(Cmat,PETSC_NULL);
1720: Cmat->assembled = PETSC_TRUE;
1721: *C = Cmat;
1722: return(0);
1723: }
1727: PetscErrorCode MatTransposeMatMultNumeric_SeqDense_SeqDense(Mat A,Mat B,Mat C)
1728: {
1729: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1730: Mat_SeqDense *b = (Mat_SeqDense*)B->data;
1731: Mat_SeqDense *c = (Mat_SeqDense*)C->data;
1732: PetscBLASInt m,n,k;
1733: PetscScalar _DOne=1.0,_DZero=0.0;
1736: m = PetscBLASIntCast(A->cmap->n);
1737: n = PetscBLASIntCast(B->cmap->n);
1738: k = PetscBLASIntCast(A->rmap->n);
1739: /*
1740: Note the m and n arguments below are the number rows and columns of A', not A!
1741: */
1742: BLASgemm_("T","N",&m,&n,&k,&_DOne,a->v,&a->lda,b->v,&b->lda,&_DZero,c->v,&c->lda);
1743: return(0);
1744: }
1748: PetscErrorCode MatGetRowMax_SeqDense(Mat A,Vec v,PetscInt idx[])
1749: {
1750: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1752: PetscInt i,j,m = A->rmap->n,n = A->cmap->n,p;
1753: PetscScalar *x;
1754: MatScalar *aa = a->v;
1757: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
1759: VecSet(v,0.0);
1760: VecGetArray(v,&x);
1761: VecGetLocalSize(v,&p);
1762: if (p != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
1763: for (i=0; i<m; i++) {
1764: x[i] = aa[i]; if (idx) idx[i] = 0;
1765: for (j=1; j<n; j++){
1766: if (PetscRealPart(x[i]) < PetscRealPart(aa[i+m*j])) {x[i] = aa[i + m*j]; if (idx) idx[i] = j;}
1767: }
1768: }
1769: VecRestoreArray(v,&x);
1770: return(0);
1771: }
1775: PetscErrorCode MatGetRowMaxAbs_SeqDense(Mat A,Vec v,PetscInt idx[])
1776: {
1777: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1779: PetscInt i,j,m = A->rmap->n,n = A->cmap->n,p;
1780: PetscScalar *x;
1781: PetscReal atmp;
1782: MatScalar *aa = a->v;
1785: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
1787: VecSet(v,0.0);
1788: VecGetArray(v,&x);
1789: VecGetLocalSize(v,&p);
1790: if (p != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
1791: for (i=0; i<m; i++) {
1792: x[i] = PetscAbsScalar(aa[i]);
1793: for (j=1; j<n; j++){
1794: atmp = PetscAbsScalar(aa[i+m*j]);
1795: if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = j;}
1796: }
1797: }
1798: VecRestoreArray(v,&x);
1799: return(0);
1800: }
1804: PetscErrorCode MatGetRowMin_SeqDense(Mat A,Vec v,PetscInt idx[])
1805: {
1806: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1808: PetscInt i,j,m = A->rmap->n,n = A->cmap->n,p;
1809: PetscScalar *x;
1810: MatScalar *aa = a->v;
1813: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
1815: VecSet(v,0.0);
1816: VecGetArray(v,&x);
1817: VecGetLocalSize(v,&p);
1818: if (p != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
1819: for (i=0; i<m; i++) {
1820: x[i] = aa[i]; if (idx) idx[i] = 0;
1821: for (j=1; j<n; j++){
1822: if (PetscRealPart(x[i]) > PetscRealPart(aa[i+m*j])) {x[i] = aa[i + m*j]; if (idx) idx[i] = j;}
1823: }
1824: }
1825: VecRestoreArray(v,&x);
1826: return(0);
1827: }
1831: PetscErrorCode MatGetColumnVector_SeqDense(Mat A,Vec v,PetscInt col)
1832: {
1833: Mat_SeqDense *a = (Mat_SeqDense*)A->data;
1835: PetscScalar *x;
1838: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
1840: VecGetArray(v,&x);
1841: PetscMemcpy(x,a->v+col*a->lda,A->rmap->n*sizeof(PetscScalar));
1842: VecRestoreArray(v,&x);
1843: return(0);
1844: }
1849: PetscErrorCode MatGetColumnNorms_SeqDense(Mat A,NormType type,PetscReal *norms)
1850: {
1852: PetscInt i,j,m,n;
1853: PetscScalar *a;
1856: MatGetSize(A,&m,&n);
1857: PetscMemzero(norms,n*sizeof(PetscReal));
1858: MatGetArray(A,&a);
1859: if (type == NORM_2) {
1860: for (i=0; i<n; i++ ){
1861: for (j=0; j<m; j++) {
1862: norms[i] += PetscAbsScalar(a[j]*a[j]);
1863: }
1864: a += m;
1865: }
1866: } else if (type == NORM_1) {
1867: for (i=0; i<n; i++ ){
1868: for (j=0; j<m; j++) {
1869: norms[i] += PetscAbsScalar(a[j]);
1870: }
1871: a += m;
1872: }
1873: } else if (type == NORM_INFINITY) {
1874: for (i=0; i<n; i++ ){
1875: for (j=0; j<m; j++) {
1876: norms[i] = PetscMax(PetscAbsScalar(a[j]),norms[i]);
1877: }
1878: a += m;
1879: }
1880: } else SETERRQ(((PetscObject)A)->comm,PETSC_ERR_ARG_WRONG,"Unknown NormType");
1881: if (type == NORM_2) {
1882: for (i=0; i<n; i++) norms[i] = PetscSqrtReal(norms[i]);
1883: }
1884: return(0);
1885: }
1887: /* -------------------------------------------------------------------*/
1888: static struct _MatOps MatOps_Values = {MatSetValues_SeqDense,
1889: MatGetRow_SeqDense,
1890: MatRestoreRow_SeqDense,
1891: MatMult_SeqDense,
1892: /* 4*/ MatMultAdd_SeqDense,
1893: MatMultTranspose_SeqDense,
1894: MatMultTransposeAdd_SeqDense,
1895: 0,
1896: 0,
1897: 0,
1898: /*10*/ 0,
1899: MatLUFactor_SeqDense,
1900: MatCholeskyFactor_SeqDense,
1901: MatSOR_SeqDense,
1902: MatTranspose_SeqDense,
1903: /*15*/ MatGetInfo_SeqDense,
1904: MatEqual_SeqDense,
1905: MatGetDiagonal_SeqDense,
1906: MatDiagonalScale_SeqDense,
1907: MatNorm_SeqDense,
1908: /*20*/ MatAssemblyBegin_SeqDense,
1909: MatAssemblyEnd_SeqDense,
1910: MatSetOption_SeqDense,
1911: MatZeroEntries_SeqDense,
1912: /*24*/ MatZeroRows_SeqDense,
1913: 0,
1914: 0,
1915: 0,
1916: 0,
1917: /*29*/ MatSetUp_SeqDense,
1918: 0,
1919: 0,
1920: MatGetArray_SeqDense,
1921: MatRestoreArray_SeqDense,
1922: /*34*/ MatDuplicate_SeqDense,
1923: 0,
1924: 0,
1925: 0,
1926: 0,
1927: /*39*/ MatAXPY_SeqDense,
1928: MatGetSubMatrices_SeqDense,
1929: 0,
1930: MatGetValues_SeqDense,
1931: MatCopy_SeqDense,
1932: /*44*/ MatGetRowMax_SeqDense,
1933: MatScale_SeqDense,
1934: 0,
1935: 0,
1936: 0,
1937: /*49*/ 0,
1938: 0,
1939: 0,
1940: 0,
1941: 0,
1942: /*54*/ 0,
1943: 0,
1944: 0,
1945: 0,
1946: 0,
1947: /*59*/ 0,
1948: MatDestroy_SeqDense,
1949: MatView_SeqDense,
1950: 0,
1951: 0,
1952: /*64*/ 0,
1953: 0,
1954: 0,
1955: 0,
1956: 0,
1957: /*69*/ MatGetRowMaxAbs_SeqDense,
1958: 0,
1959: 0,
1960: 0,
1961: 0,
1962: /*74*/ 0,
1963: 0,
1964: 0,
1965: 0,
1966: 0,
1967: /*79*/ 0,
1968: 0,
1969: 0,
1970: 0,
1971: /*83*/ MatLoad_SeqDense,
1972: 0,
1973: MatIsHermitian_SeqDense,
1974: 0,
1975: 0,
1976: 0,
1977: /*89*/ MatMatMult_SeqDense_SeqDense,
1978: MatMatMultSymbolic_SeqDense_SeqDense,
1979: MatMatMultNumeric_SeqDense_SeqDense,
1980: 0,
1981: 0,
1982: /*94*/ 0,
1983: 0,
1984: 0,
1985: 0,
1986: 0,
1987: /*99*/ 0,
1988: 0,
1989: 0,
1990: MatConjugate_SeqDense,
1991: 0,
1992: /*104*/0,
1993: MatRealPart_SeqDense,
1994: MatImaginaryPart_SeqDense,
1995: 0,
1996: 0,
1997: /*109*/MatMatSolve_SeqDense,
1998: 0,
1999: MatGetRowMin_SeqDense,
2000: MatGetColumnVector_SeqDense,
2001: 0,
2002: /*114*/0,
2003: 0,
2004: 0,
2005: 0,
2006: 0,
2007: /*119*/0,
2008: 0,
2009: 0,
2010: 0,
2011: 0,
2012: /*124*/0,
2013: MatGetColumnNorms_SeqDense,
2014: 0,
2015: 0,
2016: 0,
2017: /*129*/0,
2018: MatTransposeMatMult_SeqDense_SeqDense,
2019: MatTransposeMatMultSymbolic_SeqDense_SeqDense,
2020: MatTransposeMatMultNumeric_SeqDense_SeqDense,
2021: };
2025: /*@C
2026: MatCreateSeqDense - Creates a sequential dense matrix that
2027: is stored in column major order (the usual Fortran 77 manner). Many
2028: of the matrix operations use the BLAS and LAPACK routines.
2030: Collective on MPI_Comm
2032: Input Parameters:
2033: + comm - MPI communicator, set to PETSC_COMM_SELF
2034: . m - number of rows
2035: . n - number of columns
2036: - data - optional location of matrix data in column major order. Set data=PETSC_NULL for PETSc
2037: to control all matrix memory allocation.
2039: Output Parameter:
2040: . A - the matrix
2042: Notes:
2043: The data input variable is intended primarily for Fortran programmers
2044: who wish to allocate their own matrix memory space. Most users should
2045: set data=PETSC_NULL.
2047: Level: intermediate
2049: .keywords: dense, matrix, LAPACK, BLAS
2051: .seealso: MatCreate(), MatCreateDense(), MatSetValues()
2052: @*/
2053: PetscErrorCode MatCreateSeqDense(MPI_Comm comm,PetscInt m,PetscInt n,PetscScalar *data,Mat *A)
2054: {
2058: MatCreate(comm,A);
2059: MatSetSizes(*A,m,n,m,n);
2060: MatSetType(*A,MATSEQDENSE);
2061: MatSeqDenseSetPreallocation(*A,data);
2062: return(0);
2063: }
2067: /*@C
2068: MatSeqDenseSetPreallocation - Sets the array used for storing the matrix elements
2070: Collective on MPI_Comm
2072: Input Parameters:
2073: + A - the matrix
2074: - data - the array (or PETSC_NULL)
2076: Notes:
2077: The data input variable is intended primarily for Fortran programmers
2078: who wish to allocate their own matrix memory space. Most users should
2079: need not call this routine.
2081: Level: intermediate
2083: .keywords: dense, matrix, LAPACK, BLAS
2085: .seealso: MatCreate(), MatCreateDense(), MatSetValues(), MatSeqDenseSetLDA()
2087: @*/
2088: PetscErrorCode MatSeqDenseSetPreallocation(Mat B,PetscScalar data[])
2089: {
2093: PetscTryMethod(B,"MatSeqDenseSetPreallocation_C",(Mat,PetscScalar[]),(B,data));
2094: return(0);
2095: }
2097: EXTERN_C_BEGIN
2100: PetscErrorCode MatSeqDenseSetPreallocation_SeqDense(Mat B,PetscScalar *data)
2101: {
2102: Mat_SeqDense *b;
2106: B->preallocated = PETSC_TRUE;
2108: PetscLayoutSetUp(B->rmap);
2109: PetscLayoutSetUp(B->cmap);
2111: b = (Mat_SeqDense*)B->data;
2112: b->Mmax = B->rmap->n;
2113: b->Nmax = B->cmap->n;
2114: if(b->lda <= 0 || b->changelda) b->lda = B->rmap->n;
2116: if (!data) { /* petsc-allocated storage */
2117: if (!b->user_alloc) { PetscFree(b->v); }
2118: PetscMalloc(b->lda*b->Nmax*sizeof(PetscScalar),&b->v);
2119: PetscMemzero(b->v,b->lda*b->Nmax*sizeof(PetscScalar));
2120: PetscLogObjectMemory(B,b->lda*b->Nmax*sizeof(PetscScalar));
2121: b->user_alloc = PETSC_FALSE;
2122: } else { /* user-allocated storage */
2123: if (!b->user_alloc) { PetscFree(b->v); }
2124: b->v = data;
2125: b->user_alloc = PETSC_TRUE;
2126: }
2127: B->assembled = PETSC_TRUE;
2128: return(0);
2129: }
2130: EXTERN_C_END
2134: /*@C
2135: MatSeqDenseSetLDA - Declare the leading dimension of the user-provided array
2137: Input parameter:
2138: + A - the matrix
2139: - lda - the leading dimension
2141: Notes:
2142: This routine is to be used in conjunction with MatSeqDenseSetPreallocation();
2143: it asserts that the preallocation has a leading dimension (the LDA parameter
2144: of Blas and Lapack fame) larger than M, the first dimension of the matrix.
2146: Level: intermediate
2148: .keywords: dense, matrix, LAPACK, BLAS
2150: .seealso: MatCreate(), MatCreateSeqDense(), MatSeqDenseSetPreallocation(), MatSetMaximumSize()
2152: @*/
2153: PetscErrorCode MatSeqDenseSetLDA(Mat B,PetscInt lda)
2154: {
2155: Mat_SeqDense *b = (Mat_SeqDense*)B->data;
2158: if (lda < B->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"LDA %D must be at least matrix dimension %D",lda,B->rmap->n);
2159: b->lda = lda;
2160: b->changelda = PETSC_FALSE;
2161: b->Mmax = PetscMax(b->Mmax,lda);
2162: return(0);
2163: }
2165: /*MC
2166: MATSEQDENSE - MATSEQDENSE = "seqdense" - A matrix type to be used for sequential dense matrices.
2168: Options Database Keys:
2169: . -mat_type seqdense - sets the matrix type to "seqdense" during a call to MatSetFromOptions()
2171: Level: beginner
2173: .seealso: MatCreateSeqDense()
2175: M*/
2177: EXTERN_C_BEGIN
2180: PetscErrorCode MatCreate_SeqDense(Mat B)
2181: {
2182: Mat_SeqDense *b;
2184: PetscMPIInt size;
2187: MPI_Comm_size(((PetscObject)B)->comm,&size);
2188: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Comm must be of size 1");
2190: PetscNewLog(B,Mat_SeqDense,&b);
2191: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
2192: B->data = (void*)b;
2194: b->pivots = 0;
2195: b->roworiented = PETSC_TRUE;
2196: b->v = 0;
2197: b->changelda = PETSC_FALSE;
2200: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqdense_seqaij_C","MatConvert_SeqDense_SeqAIJ",MatConvert_SeqDense_SeqAIJ);
2201: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetFactor_petsc_C",
2202: "MatGetFactor_seqdense_petsc",
2203: MatGetFactor_seqdense_petsc);
2204: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqDenseSetPreallocation_C",
2205: "MatSeqDenseSetPreallocation_SeqDense",
2206: MatSeqDenseSetPreallocation_SeqDense);
2208: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMatMult_seqaij_seqdense_C",
2209: "MatMatMult_SeqAIJ_SeqDense",
2210: MatMatMult_SeqAIJ_SeqDense);
2213: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMatMultSymbolic_seqaij_seqdense_C",
2214: "MatMatMultSymbolic_SeqAIJ_SeqDense",
2215: MatMatMultSymbolic_SeqAIJ_SeqDense);
2216: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMatMultNumeric_seqaij_seqdense_C",
2217: "MatMatMultNumeric_SeqAIJ_SeqDense",
2218: MatMatMultNumeric_SeqAIJ_SeqDense);
2219: PetscObjectChangeTypeName((PetscObject)B,MATSEQDENSE);
2220: return(0);
2221: }
2222: EXTERN_C_END