2: #include <petsc/private/pcimpl.h> /*I "petscpc.h" I*/
3: #include <../src/mat/impls/aij/seq/aij.h>
5: /*
6: Private context (data structure) for the CP preconditioner.
7: */
8: typedef struct {
9: PetscInt n,m;
10: Vec work;
11: PetscScalar *d; /* sum of squares of each column */
12: PetscScalar *a; /* non-zeros by column */
13: PetscInt *i,*j; /* offsets of nonzeros by column, non-zero indices by column */
14: } PC_CP;
19: static PetscErrorCode PCSetUp_CP(PC pc) 20: {
21: PC_CP *cp = (PC_CP*)pc->data;
22: PetscInt i,j,*colcnt;
24: PetscBool flg;
25: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)pc->pmat->data;
28: PetscObjectTypeCompare((PetscObject)pc->pmat,MATSEQAIJ,&flg);
29: if (!flg) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_SUP,"Currently only handles SeqAIJ matrices");
31: MatGetLocalSize(pc->pmat,&cp->m,&cp->n);
32: if (cp->m != cp->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Currently only for square matrices");
34: if (!cp->work) {MatCreateVecs(pc->pmat,&cp->work,NULL);}
35: if (!cp->d) {PetscMalloc1(cp->n,&cp->d);}
36: if (cp->a && pc->flag != SAME_NONZERO_PATTERN) {
37: PetscFree3(cp->a,cp->i,cp->j);
38: cp->a = 0;
39: }
41: /* convert to column format */
42: if (!cp->a) {
43: PetscMalloc3(aij->nz,&cp->a,cp->n+1,&cp->i,aij->nz,&cp->j);
44: }
45: PetscCalloc1(cp->n,&colcnt);
47: for (i=0; i<aij->nz; i++) colcnt[aij->j[i]]++;
48: cp->i[0] = 0;
49: for (i=0; i<cp->n; i++) cp->i[i+1] = cp->i[i] + colcnt[i];
50: PetscMemzero(colcnt,cp->n*sizeof(PetscInt));
51: for (i=0; i<cp->m; i++) { /* over rows */
52: for (j=aij->i[i]; j<aij->i[i+1]; j++) { /* over columns in row */
53: cp->j[cp->i[aij->j[j]]+colcnt[aij->j[j]]] = i;
54: cp->a[cp->i[aij->j[j]]+colcnt[aij->j[j]]++] = aij->a[j];
55: }
56: }
57: PetscFree(colcnt);
59: /* compute sum of squares of each column d[] */
60: for (i=0; i<cp->n; i++) { /* over columns */
61: cp->d[i] = 0.;
62: for (j=cp->i[i]; j<cp->i[i+1]; j++) cp->d[i] += cp->a[j]*cp->a[j]; /* over rows in column */
63: cp->d[i] = 1.0/cp->d[i];
64: }
65: return(0);
66: }
67: /* -------------------------------------------------------------------------- */
70: static PetscErrorCode PCApply_CP(PC pc,Vec bb,Vec xx) 71: {
72: PC_CP *cp = (PC_CP*)pc->data;
74: PetscScalar *b,*x,xt;
75: PetscInt i,j;
78: VecCopy(bb,cp->work);
79: VecGetArray(cp->work,&b);
80: VecGetArray(xx,&x);
82: for (i=0; i<cp->n; i++) { /* over columns */
83: xt = 0.;
84: for (j=cp->i[i]; j<cp->i[i+1]; j++) xt += cp->a[j]*b[cp->j[j]]; /* over rows in column */
85: xt *= cp->d[i];
86: x[i] = xt;
87: for (j=cp->i[i]; j<cp->i[i+1]; j++) b[cp->j[j]] -= xt*cp->a[j]; /* over rows in column updating b*/
88: }
89: for (i=cp->n-1; i>-1; i--) { /* over columns */
90: xt = 0.;
91: for (j=cp->i[i]; j<cp->i[i+1]; j++) xt += cp->a[j]*b[cp->j[j]]; /* over rows in column */
92: xt *= cp->d[i];
93: x[i] = xt;
94: for (j=cp->i[i]; j<cp->i[i+1]; j++) b[cp->j[j]] -= xt*cp->a[j]; /* over rows in column updating b*/
95: }
97: VecRestoreArray(cp->work,&b);
98: VecRestoreArray(xx,&x);
99: return(0);
100: }
101: /* -------------------------------------------------------------------------- */
104: static PetscErrorCode PCReset_CP(PC pc)105: {
106: PC_CP *cp = (PC_CP*)pc->data;
110: PetscFree(cp->d);
111: VecDestroy(&cp->work);
112: PetscFree3(cp->a,cp->i,cp->j);
113: return(0);
114: }
118: static PetscErrorCode PCDestroy_CP(PC pc)119: {
120: PC_CP *cp = (PC_CP*)pc->data;
124: PCReset_CP(pc);
125: PetscFree(cp->d);
126: PetscFree3(cp->a,cp->i,cp->j);
127: PetscFree(pc->data);
128: return(0);
129: }
133: static PetscErrorCode PCSetFromOptions_CP(PetscOptionItems *PetscOptionsObject,PC pc)134: {
136: return(0);
137: }
140: /*MC
141: PCCP - a "column-projection" preconditioner
143: This is a terrible preconditioner and is not recommended, ever!
145: Loops over the entries of x computing dx_i to
146: $
147: $ min || b - A(x + dx_i e_i ||_2
148: $ dx_i
149: $
150: $ That is, it changes a single entry of x to minimize the new residual.
151: $ Let A_i represent the ith column of A, then the minimization can be written as
152: $
153: $ min || r - (dx_i) A e_i ||_2
154: $ dx_i
155: $ or min || r - (dx_i) A_i ||_2
156: $ dx_i
157: $
158: $ take the derivative with respect to dx_i to obtain
159: $ dx_i = (A_i^T A_i)^(-1) A_i^T r
160: $
161: $ This algorithm can be thought of as Gauss-Seidel on the normal equations
163: Notes: This proceedure can also be done with block columns or any groups of columns
164: but this is not coded.
166: These "projections" can be done simultaneously for all columns (similar to Jacobi)
167: or sequentially (similar to Gauss-Seidel/SOR). This is only coded for SOR type.
169: This is related to, but not the same as "row projection" methods.
171: This is currently coded only for SeqAIJ matrices in sequential (SOR) form.
173: Level: intermediate
175: .seealso: PCCreate(), PCSetType(), PCType (for list of available types), PCJACOBI, PCSOR177: M*/
181: PETSC_EXTERN PetscErrorCode PCCreate_CP(PC pc)182: {
183: PC_CP *cp;
187: PetscNewLog(pc,&cp);
188: pc->data = (void*)cp;
190: pc->ops->apply = PCApply_CP;
191: pc->ops->applytranspose = PCApply_CP;
192: pc->ops->setup = PCSetUp_CP;
193: pc->ops->reset = PCReset_CP;
194: pc->ops->destroy = PCDestroy_CP;
195: pc->ops->setfromoptions = PCSetFromOptions_CP;
196: pc->ops->view = 0;
197: pc->ops->applyrichardson = 0;
198: return(0);
199: }