Actual source code: axpy.c
petsc-3.6.4 2016-04-12
2: #include <petsc/private/matimpl.h> /*I "petscmat.h" I*/
6: /*@
7: MatAXPY - Computes Y = a*X + Y.
9: Logically Collective on Mat
11: Input Parameters:
12: + a - the scalar multiplier
13: . X - the first matrix
14: . Y - the second matrix
15: - str - either SAME_NONZERO_PATTERN, DIFFERENT_NONZERO_PATTERN
16: or SUBSET_NONZERO_PATTERN (nonzeros of X is a subset of Y's)
18: Level: intermediate
20: .keywords: matrix, add
22: .seealso: MatAYPX()
23: @*/
24: PetscErrorCode MatAXPY(Mat Y,PetscScalar a,Mat X,MatStructure str)
25: {
27: PetscInt m1,m2,n1,n2;
33: MatGetSize(X,&m1,&n1);
34: MatGetSize(Y,&m2,&n2);
35: if (m1 != m2 || n1 != n2) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Non conforming matrix add: %D %D %D %D",m1,m2,n1,n2);
37: PetscLogEventBegin(MAT_AXPY,Y,0,0,0);
38: if (Y->ops->axpy) {
39: (*Y->ops->axpy)(Y,a,X,str);
40: } else {
41: MatAXPY_Basic(Y,a,X,str);
42: }
43: PetscLogEventEnd(MAT_AXPY,Y,0,0,0);
44: #if defined(PETSC_HAVE_CUSP)
45: if (Y->valid_GPU_matrix != PETSC_CUSP_UNALLOCATED) {
46: Y->valid_GPU_matrix = PETSC_CUSP_CPU;
47: }
48: #endif
49: #if defined(PETSC_HAVE_VIENNACL)
50: if (Y->valid_GPU_matrix != PETSC_VIENNACL_UNALLOCATED) {
51: Y->valid_GPU_matrix = PETSC_VIENNACL_CPU;
52: }
53: #endif
54: return(0);
55: }
59: PetscErrorCode MatAXPY_Basic(Mat Y,PetscScalar a,Mat X,MatStructure str)
60: {
61: PetscInt i,start,end,j,ncols,m,n;
62: PetscErrorCode ierr;
63: const PetscInt *row;
64: PetscScalar *val;
65: const PetscScalar *vals;
68: MatGetSize(X,&m,&n);
69: MatGetOwnershipRange(X,&start,&end);
70: if (a == 1.0) {
71: for (i = start; i < end; i++) {
72: MatGetRow(X,i,&ncols,&row,&vals);
73: MatSetValues(Y,1,&i,ncols,row,vals,ADD_VALUES);
74: MatRestoreRow(X,i,&ncols,&row,&vals);
75: }
76: } else {
77: PetscMalloc1(n+1,&val);
78: for (i=start; i<end; i++) {
79: MatGetRow(X,i,&ncols,&row,&vals);
80: for (j=0; j<ncols; j++) {
81: val[j] = a*vals[j];
82: }
83: MatSetValues(Y,1,&i,ncols,row,val,ADD_VALUES);
84: MatRestoreRow(X,i,&ncols,&row,&vals);
85: }
86: PetscFree(val);
87: }
88: MatAssemblyBegin(Y,MAT_FINAL_ASSEMBLY);
89: MatAssemblyEnd(Y,MAT_FINAL_ASSEMBLY);
90: return(0);
91: }
95: PetscErrorCode MatAXPY_BasicWithPreallocation(Mat B,Mat Y,PetscScalar a,Mat X,MatStructure str)
96: {
97: PetscInt i,start,end,j,ncols,m,n;
98: PetscErrorCode ierr;
99: const PetscInt *row;
100: PetscScalar *val;
101: const PetscScalar *vals;
104: MatGetSize(X,&m,&n);
105: MatGetOwnershipRange(X,&start,&end);
106: if (a == 1.0) {
107: for (i = start; i < end; i++) {
108: MatGetRow(Y,i,&ncols,&row,&vals);
109: MatSetValues(B,1,&i,ncols,row,vals,ADD_VALUES);
110: MatRestoreRow(Y,i,&ncols,&row,&vals);
112: MatGetRow(X,i,&ncols,&row,&vals);
113: MatSetValues(B,1,&i,ncols,row,vals,ADD_VALUES);
114: MatRestoreRow(X,i,&ncols,&row,&vals);
115: }
116: } else {
117: PetscMalloc1(n+1,&val);
118: for (i=start; i<end; i++) {
119: MatGetRow(Y,i,&ncols,&row,&vals);
120: MatSetValues(B,1,&i,ncols,row,vals,ADD_VALUES);
121: MatRestoreRow(Y,i,&ncols,&row,&vals);
123: MatGetRow(X,i,&ncols,&row,&vals);
124: for (j=0; j<ncols; j++) {
125: val[j] = a*vals[j];
126: }
127: MatSetValues(B,1,&i,ncols,row,val,ADD_VALUES);
128: MatRestoreRow(X,i,&ncols,&row,&vals);
129: }
130: PetscFree(val);
131: }
132: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
133: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
134: return(0);
135: }
139: /*@
140: MatShift - Computes Y = Y + a I, where a is a PetscScalar and I is the identity matrix.
142: Neighbor-wise Collective on Mat
144: Input Parameters:
145: + Y - the matrices
146: - a - the PetscScalar
148: Level: intermediate
150: Notes: If the matrix Y is missing some diagonal entries this routine can be very slow. To make it fast one should initially
151: fill the matrix so that all diagonal entries have a value (with a value of zero for those locations that would not have an
152: entry).
154: Developers Note: If the local "diagonal part" of the matrix Y has no entries then the local diagonal part is
155: preallocated with 1 nonzero per row for the to be added values. This allows for fast shifting of an empty matrix.
157: .keywords: matrix, add, shift
159: .seealso: MatDiagonalSet()
160: @*/
161: PetscErrorCode MatShift(Mat Y,PetscScalar a)
162: {
167: if (!Y->assembled) SETERRQ(PetscObjectComm((PetscObject)Y),PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
168: if (Y->factortype) SETERRQ(PetscObjectComm((PetscObject)Y),PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
169: MatCheckPreallocated(Y,1);
171: (*Y->ops->shift)(Y,a);
173: #if defined(PETSC_HAVE_CUSP)
174: if (Y->valid_GPU_matrix != PETSC_CUSP_UNALLOCATED) {
175: Y->valid_GPU_matrix = PETSC_CUSP_CPU;
176: }
177: #endif
178: #if defined(PETSC_HAVE_VIENNACL)
179: if (Y->valid_GPU_matrix != PETSC_VIENNACL_UNALLOCATED) {
180: Y->valid_GPU_matrix = PETSC_VIENNACL_CPU;
181: }
182: #endif
183: return(0);
184: }
188: PetscErrorCode MatDiagonalSet_Default(Mat Y,Vec D,InsertMode is)
189: {
191: PetscInt i,start,end;
192: PetscScalar *v;
195: MatGetOwnershipRange(Y,&start,&end);
196: VecGetArray(D,&v);
197: for (i=start; i<end; i++) {
198: MatSetValues(Y,1,&i,1,&i,v+i-start,is);
199: }
200: VecRestoreArray(D,&v);
201: MatAssemblyBegin(Y,MAT_FINAL_ASSEMBLY);
202: MatAssemblyEnd(Y,MAT_FINAL_ASSEMBLY);
203: return(0);
204: }
208: /*@
209: MatDiagonalSet - Computes Y = Y + D, where D is a diagonal matrix
210: that is represented as a vector. Or Y[i,i] = D[i] if InsertMode is
211: INSERT_VALUES.
213: Input Parameters:
214: + Y - the input matrix
215: . D - the diagonal matrix, represented as a vector
216: - i - INSERT_VALUES or ADD_VALUES
218: Neighbor-wise Collective on Mat and Vec
220: Notes: If the matrix Y is missing some diagonal entries this routine can be very slow. To make it fast one should initially
221: fill the matrix so that all diagonal entries have a value (with a value of zero for those locations that would not have an
222: entry).
224: Level: intermediate
226: .keywords: matrix, add, shift, diagonal
228: .seealso: MatShift()
229: @*/
230: PetscErrorCode MatDiagonalSet(Mat Y,Vec D,InsertMode is)
231: {
233: PetscInt matlocal,veclocal;
238: MatGetLocalSize(Y,&matlocal,NULL);
239: VecGetLocalSize(D,&veclocal);
240: if (matlocal != veclocal) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Number local rows of matrix %D does not match that of vector for diagonal %D",matlocal,veclocal);
241: if (Y->ops->diagonalset) {
242: (*Y->ops->diagonalset)(Y,D,is);
243: } else {
244: MatDiagonalSet_Default(Y,D,is);
245: }
246: return(0);
247: }
251: /*@
252: MatAYPX - Computes Y = a*Y + X.
254: Logically on Mat
256: Input Parameters:
257: + a - the PetscScalar multiplier
258: . Y - the first matrix
259: . X - the second matrix
260: - str - either SAME_NONZERO_PATTERN, DIFFERENT_NONZERO_PATTERN or SUBSET_NONZERO_PATTERN
262: Level: intermediate
264: .keywords: matrix, add
266: .seealso: MatAXPY()
267: @*/
268: PetscErrorCode MatAYPX(Mat Y,PetscScalar a,Mat X,MatStructure str)
269: {
270: PetscScalar one = 1.0;
272: PetscInt mX,mY,nX,nY;
278: MatGetSize(X,&mX,&nX);
279: MatGetSize(X,&mY,&nY);
280: if (mX != mY || nX != nY) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Non conforming matrices: %D %D first %D %D second",mX,mY,nX,nY);
282: MatScale(Y,a);
283: MatAXPY(Y,one,X,str);
284: return(0);
285: }
289: /*@
290: MatComputeExplicitOperator - Computes the explicit matrix
292: Collective on Mat
294: Input Parameter:
295: . inmat - the matrix
297: Output Parameter:
298: . mat - the explict preconditioned operator
300: Notes:
301: This computation is done by applying the operators to columns of the
302: identity matrix.
304: Currently, this routine uses a dense matrix format when 1 processor
305: is used and a sparse format otherwise. This routine is costly in general,
306: and is recommended for use only with relatively small systems.
308: Level: advanced
310: .keywords: Mat, compute, explicit, operator
311: @*/
312: PetscErrorCode MatComputeExplicitOperator(Mat inmat,Mat *mat)
313: {
314: Vec in,out;
316: PetscInt i,m,n,M,N,*rows,start,end;
317: MPI_Comm comm;
318: PetscScalar *array,zero = 0.0,one = 1.0;
319: PetscMPIInt size;
325: PetscObjectGetComm((PetscObject)inmat,&comm);
326: MPI_Comm_size(comm,&size);
328: MatGetLocalSize(inmat,&m,&n);
329: MatGetSize(inmat,&M,&N);
330: MatCreateVecs(inmat,&in,&out);
331: VecSetOption(in,VEC_IGNORE_OFF_PROC_ENTRIES,PETSC_TRUE);
332: VecGetOwnershipRange(out,&start,&end);
333: PetscMalloc1(m,&rows);
334: for (i=0; i<m; i++) rows[i] = start + i;
336: MatCreate(comm,mat);
337: MatSetSizes(*mat,m,n,M,N);
338: if (size == 1) {
339: MatSetType(*mat,MATSEQDENSE);
340: MatSeqDenseSetPreallocation(*mat,NULL);
341: } else {
342: MatSetType(*mat,MATMPIAIJ);
343: MatMPIAIJSetPreallocation(*mat,n,NULL,N-n,NULL);
344: }
346: for (i=0; i<N; i++) {
348: VecSet(in,zero);
349: VecSetValues(in,1,&i,&one,INSERT_VALUES);
350: VecAssemblyBegin(in);
351: VecAssemblyEnd(in);
353: MatMult(inmat,in,out);
355: VecGetArray(out,&array);
356: MatSetValues(*mat,m,rows,1,&i,array,INSERT_VALUES);
357: VecRestoreArray(out,&array);
359: }
360: PetscFree(rows);
361: VecDestroy(&out);
362: VecDestroy(&in);
363: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
364: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
365: return(0);
366: }
370: /*@
371: MatChop - Set all values in the matrix less than the tolerance to zero
373: Input Parameters:
374: + A - The matrix
375: - tol - The zero tolerance
377: Output Parameters:
378: . A - The chopped matrix
380: Level: intermediate
382: .seealso: MatCreate(), MatZeroEntries()
383: @*/
384: PetscErrorCode MatChop(Mat A, PetscReal tol)
385: {
386: PetscScalar *newVals;
387: PetscInt *newCols;
388: PetscInt rStart, rEnd, numRows, maxRows, r, colMax = 0;
392: MatGetOwnershipRange(A, &rStart, &rEnd);
393: for (r = rStart; r < rEnd; ++r) {
394: PetscInt ncols;
396: MatGetRow(A, r, &ncols, NULL, NULL);
397: colMax = PetscMax(colMax, ncols);
398: MatRestoreRow(A, r, &ncols, NULL, NULL);
399: }
400: numRows = rEnd - rStart;
401: MPI_Allreduce(&numRows, &maxRows, 1, MPIU_INT, MPI_MAX, PetscObjectComm((PetscObject)A));
402: PetscMalloc2(colMax,&newCols,colMax,&newVals);
403: for (r = rStart; r < rStart+maxRows; ++r) {
404: const PetscScalar *vals;
405: const PetscInt *cols;
406: PetscInt ncols, newcols, c;
408: if (r < rEnd) {
409: MatGetRow(A, r, &ncols, &cols, &vals);
410: for (c = 0; c < ncols; ++c) {
411: newCols[c] = cols[c];
412: newVals[c] = PetscAbsScalar(vals[c]) < tol ? 0.0 : vals[c];
413: }
414: newcols = ncols;
415: MatRestoreRow(A, r, &ncols, &cols, &vals);
416: MatSetValues(A, 1, &r, newcols, newCols, newVals, INSERT_VALUES);
417: }
418: MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY);
419: MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY);
420: }
421: PetscFree2(newCols,newVals);
422: return(0);
423: }