xref: /petsc/src/mat/utils/axpy.c (revision f3fe499b4cc4d64bf04aa4f5e4963dcc4eb56541)
1 #define PETSCMAT_DLL
2 
3 #include "private/matimpl.h"  /*I   "petscmat.h"  I*/
4 
5 #undef __FUNCT__
6 #define __FUNCT__ "MatAXPY"
7 /*@
8    MatAXPY - Computes Y = a*X + Y.
9 
10    Logically  Collective on Mat
11 
12    Input Parameters:
13 +  a - the scalar multiplier
14 .  X - the first matrix
15 .  Y - the second matrix
16 -  str - either SAME_NONZERO_PATTERN, DIFFERENT_NONZERO_PATTERN
17          or SUBSET_NONZERO_PATTERN (nonzeros of X is a subset of Y's)
18 
19    Notes:
20      Will only be efficient if one has the SAME_NONZERO_PATTERN or SUBSET_NONZERO_PATTERN
21 
22    Level: intermediate
23 
24 .keywords: matrix, add
25 
26 .seealso: MatAYPX()
27  @*/
28 PetscErrorCode PETSCMAT_DLLEXPORT MatAXPY(Mat Y,PetscScalar a,Mat X,MatStructure str)
29 {
30   PetscErrorCode ierr;
31   PetscInt       m1,m2,n1,n2;
32 
33   PetscFunctionBegin;
34   PetscValidHeaderSpecific(X,MAT_CLASSID,3);
35   PetscValidHeaderSpecific(Y,MAT_CLASSID,1);
36   PetscValidLogicalCollectiveScalar(Y,a,2);
37   ierr = MatGetSize(X,&m1,&n1);CHKERRQ(ierr);
38   ierr = MatGetSize(Y,&m2,&n2);CHKERRQ(ierr);
39   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);
40 
41   ierr = PetscLogEventBegin(MAT_AXPY,Y,0,0,0);CHKERRQ(ierr);
42   if (Y->ops->axpy) {
43     ierr = (*Y->ops->axpy)(Y,a,X,str);CHKERRQ(ierr);
44   } else {
45     ierr = MatAXPY_Basic(Y,a,X,str);CHKERRQ(ierr);
46   }
47   ierr = PetscLogEventEnd(MAT_AXPY,Y,0,0,0);CHKERRQ(ierr);
48   PetscFunctionReturn(0);
49 }
50 
51 
52 #undef __FUNCT__
53 #define __FUNCT__ "MatAXPY_Basic"
54 PetscErrorCode MatAXPY_Basic(Mat Y,PetscScalar a,Mat X,MatStructure str)
55 {
56   PetscInt          i,start,end,j,ncols,m,n;
57   PetscErrorCode    ierr;
58   const PetscInt    *row;
59   PetscScalar       *val;
60   const PetscScalar *vals;
61 
62   PetscFunctionBegin;
63   ierr = MatGetSize(X,&m,&n);CHKERRQ(ierr);
64   ierr = MatGetOwnershipRange(X,&start,&end);CHKERRQ(ierr);
65   if (a == 1.0) {
66     for (i = start; i < end; i++) {
67       ierr = MatGetRow(X,i,&ncols,&row,&vals);CHKERRQ(ierr);
68       ierr = MatSetValues(Y,1,&i,ncols,row,vals,ADD_VALUES);CHKERRQ(ierr);
69       ierr = MatRestoreRow(X,i,&ncols,&row,&vals);CHKERRQ(ierr);
70     }
71   } else {
72     ierr = PetscMalloc((n+1)*sizeof(PetscScalar),&val);CHKERRQ(ierr);
73     for (i=start; i<end; i++) {
74       ierr = MatGetRow(X,i,&ncols,&row,&vals);CHKERRQ(ierr);
75       for (j=0; j<ncols; j++) {
76 	val[j] = a*vals[j];
77       }
78       ierr = MatSetValues(Y,1,&i,ncols,row,val,ADD_VALUES);CHKERRQ(ierr);
79       ierr = MatRestoreRow(X,i,&ncols,&row,&vals);CHKERRQ(ierr);
80     }
81     ierr = PetscFree(val);CHKERRQ(ierr);
82   }
83   ierr = MatAssemblyBegin(Y,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
84   ierr = MatAssemblyEnd(Y,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
85   PetscFunctionReturn(0);
86 }
87 
88 #undef __FUNCT__
89 #define __FUNCT__ "MatShift"
90 /*@
91    MatShift - Computes Y =  Y + a I, where a is a PetscScalar and I is the identity matrix.
92 
93    Neighbor-wise Collective on Mat
94 
95    Input Parameters:
96 +  Y - the matrices
97 -  a - the PetscScalar
98 
99    Level: intermediate
100 
101 .keywords: matrix, add, shift
102 
103 .seealso: MatDiagonalSet()
104  @*/
105 PetscErrorCode PETSCMAT_DLLEXPORT MatShift(Mat Y,PetscScalar a)
106 {
107   PetscErrorCode ierr;
108   PetscInt       i,start,end;
109 
110   PetscFunctionBegin;
111   PetscValidHeaderSpecific(Y,MAT_CLASSID,1);
112   if (!Y->assembled) SETERRQ(((PetscObject)Y)->comm,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
113   if (Y->factortype) SETERRQ(((PetscObject)Y)->comm,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
114   ierr = MatPreallocated(Y);CHKERRQ(ierr);
115 
116   if (Y->ops->shift) {
117     ierr = (*Y->ops->shift)(Y,a);CHKERRQ(ierr);
118   } else {
119     PetscScalar alpha = a;
120     ierr = MatGetOwnershipRange(Y,&start,&end);CHKERRQ(ierr);
121     for (i=start; i<end; i++) {
122       ierr = MatSetValues(Y,1,&i,1,&i,&alpha,ADD_VALUES);CHKERRQ(ierr);
123     }
124     ierr = MatAssemblyBegin(Y,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
125     ierr = MatAssemblyEnd(Y,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
126   }
127   PetscFunctionReturn(0);
128 }
129 
130 #undef __FUNCT__
131 #define __FUNCT__ "MatDiagonalSet_Default"
132 PetscErrorCode PETSCMAT_DLLEXPORT MatDiagonalSet_Default(Mat Y,Vec D,InsertMode is)
133 {
134   PetscErrorCode ierr;
135   PetscInt       i,start,end,vstart,vend;
136   PetscScalar    *v;
137 
138   PetscFunctionBegin;
139   ierr = VecGetOwnershipRange(D,&vstart,&vend);CHKERRQ(ierr);
140   ierr = MatGetOwnershipRange(Y,&start,&end);CHKERRQ(ierr);
141   if (vstart != start || vend != end) {
142     SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Vector ownership range not compatible with matrix: %D %D vec %D %D mat",vstart,vend,start,end);
143   }
144   ierr = VecGetArray(D,&v);CHKERRQ(ierr);
145   for (i=start; i<end; i++) {
146     ierr = MatSetValues(Y,1,&i,1,&i,v+i-start,is);CHKERRQ(ierr);
147   }
148   ierr = VecRestoreArray(D,&v);CHKERRQ(ierr);
149   ierr = MatAssemblyBegin(Y,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
150   ierr = MatAssemblyEnd(Y,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
151   PetscFunctionReturn(0);
152 }
153 
154 #undef __FUNCT__
155 #define __FUNCT__ "MatDiagonalSet"
156 /*@
157    MatDiagonalSet - Computes Y = Y + D, where D is a diagonal matrix
158    that is represented as a vector. Or Y[i,i] = D[i] if InsertMode is
159    INSERT_VALUES.
160 
161    Input Parameters:
162 +  Y - the input matrix
163 .  D - the diagonal matrix, represented as a vector
164 -  i - INSERT_VALUES or ADD_VALUES
165 
166    Neighbor-wise Collective on Mat and Vec
167 
168    Level: intermediate
169 
170 .keywords: matrix, add, shift, diagonal
171 
172 .seealso: MatShift()
173 @*/
174 PetscErrorCode PETSCMAT_DLLEXPORT MatDiagonalSet(Mat Y,Vec D,InsertMode is)
175 {
176   PetscErrorCode ierr;
177 
178   PetscFunctionBegin;
179   PetscValidHeaderSpecific(Y,MAT_CLASSID,1);
180   PetscValidHeaderSpecific(D,VEC_CLASSID,2);
181   if (Y->ops->diagonalset) {
182     ierr = (*Y->ops->diagonalset)(Y,D,is);CHKERRQ(ierr);
183   } else {
184     ierr = MatDiagonalSet_Default(Y,D,is);CHKERRQ(ierr);
185   }
186   PetscFunctionReturn(0);
187 }
188 
189 #undef __FUNCT__
190 #define __FUNCT__ "MatAYPX"
191 /*@
192    MatAYPX - Computes Y = a*Y + X.
193 
194    Logically on Mat
195 
196    Input Parameters:
197 +  a - the PetscScalar multiplier
198 .  Y - the first matrix
199 .  X - the second matrix
200 -  str - either SAME_NONZERO_PATTERN, DIFFERENT_NONZERO_PATTERN or SUBSET_NONZERO_PATTERN
201 
202    Notes:
203      Will only be efficient if one has the SAME_NONZERO_PATTERN or SUBSET_NONZERO_PATTERN
204 
205    Level: intermediate
206 
207 .keywords: matrix, add
208 
209 .seealso: MatAXPY()
210  @*/
211 PetscErrorCode PETSCMAT_DLLEXPORT MatAYPX(Mat Y,PetscScalar a,Mat X,MatStructure str)
212 {
213   PetscScalar    one = 1.0;
214   PetscErrorCode ierr;
215   PetscInt       mX,mY,nX,nY;
216 
217   PetscFunctionBegin;
218   PetscValidHeaderSpecific(X,MAT_CLASSID,3);
219   PetscValidHeaderSpecific(Y,MAT_CLASSID,1);
220   PetscValidLogicalCollectiveScalar(Y,a,2);
221   ierr = MatGetSize(X,&mX,&nX);CHKERRQ(ierr);
222   ierr = MatGetSize(X,&mY,&nY);CHKERRQ(ierr);
223   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);
224 
225   ierr = MatScale(Y,a);CHKERRQ(ierr);
226   ierr = MatAXPY(Y,one,X,str);CHKERRQ(ierr);
227   PetscFunctionReturn(0);
228 }
229 
230 #undef __FUNCT__
231 #define __FUNCT__ "MatComputeExplicitOperator"
232 /*@
233     MatComputeExplicitOperator - Computes the explicit matrix
234 
235     Collective on Mat
236 
237     Input Parameter:
238 .   inmat - the matrix
239 
240     Output Parameter:
241 .   mat - the explict preconditioned operator
242 
243     Notes:
244     This computation is done by applying the operators to columns of the
245     identity matrix.
246 
247     Currently, this routine uses a dense matrix format when 1 processor
248     is used and a sparse format otherwise.  This routine is costly in general,
249     and is recommended for use only with relatively small systems.
250 
251     Level: advanced
252 
253 .keywords: Mat, compute, explicit, operator
254 
255 @*/
256 PetscErrorCode PETSCMAT_DLLEXPORT MatComputeExplicitOperator(Mat inmat,Mat *mat)
257 {
258   Vec            in,out;
259   PetscErrorCode ierr;
260   PetscInt       i,m,n,M,N,*rows,start,end;
261   MPI_Comm       comm;
262   PetscScalar    *array,zero = 0.0,one = 1.0;
263   PetscMPIInt    size;
264 
265   PetscFunctionBegin;
266   PetscValidHeaderSpecific(inmat,MAT_CLASSID,1);
267   PetscValidPointer(mat,2);
268 
269   comm = ((PetscObject)inmat)->comm;
270   ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr);
271 
272   ierr = MatGetLocalSize(inmat,&m,&n);CHKERRQ(ierr);
273   ierr = MatGetSize(inmat,&M,&N);CHKERRQ(ierr);
274   ierr = MatGetVecs(inmat,&in,&out);CHKERRQ(ierr);
275   ierr = VecSetOption(in,VEC_IGNORE_OFF_PROC_ENTRIES,PETSC_TRUE);CHKERRQ(ierr);
276   ierr = VecGetOwnershipRange(out,&start,&end);CHKERRQ(ierr);
277   ierr = PetscMalloc(m*sizeof(PetscInt),&rows);CHKERRQ(ierr);
278   for (i=0; i<m; i++) {rows[i] = start + i;}
279 
280   ierr = MatCreate(comm,mat);CHKERRQ(ierr);
281   ierr = MatSetSizes(*mat,m,n,M,N);CHKERRQ(ierr);
282   if (size == 1) {
283     ierr = MatSetType(*mat,MATSEQDENSE);CHKERRQ(ierr);
284     ierr = MatSeqDenseSetPreallocation(*mat,PETSC_NULL);CHKERRQ(ierr);
285   } else {
286     ierr = MatSetType(*mat,MATMPIAIJ);CHKERRQ(ierr);
287     ierr = MatMPIAIJSetPreallocation(*mat,n,PETSC_NULL,N-n,PETSC_NULL);CHKERRQ(ierr);
288   }
289 
290   for (i=0; i<N; i++) {
291 
292     ierr = VecSet(in,zero);CHKERRQ(ierr);
293     ierr = VecSetValues(in,1,&i,&one,INSERT_VALUES);CHKERRQ(ierr);
294     ierr = VecAssemblyBegin(in);CHKERRQ(ierr);
295     ierr = VecAssemblyEnd(in);CHKERRQ(ierr);
296 
297     ierr = MatMult(inmat,in,out);CHKERRQ(ierr);
298 
299     ierr = VecGetArray(out,&array);CHKERRQ(ierr);
300     ierr = MatSetValues(*mat,m,rows,1,&i,array,INSERT_VALUES);CHKERRQ(ierr);
301     ierr = VecRestoreArray(out,&array);CHKERRQ(ierr);
302 
303   }
304   ierr = PetscFree(rows);CHKERRQ(ierr);
305   ierr = VecDestroy(out);CHKERRQ(ierr);
306   ierr = VecDestroy(in);CHKERRQ(ierr);
307   ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
308   ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
309   PetscFunctionReturn(0);
310 }
311 
312 /* Get the map xtoy which is used by MatAXPY() in the case of SUBSET_NONZERO_PATTERN */
313 #undef __FUNCT__
314 #define __FUNCT__ "MatAXPYGetxtoy_Private"
315 PetscErrorCode MatAXPYGetxtoy_Private(PetscInt m,PetscInt *xi,PetscInt *xj,PetscInt *xgarray, PetscInt *yi,PetscInt *yj,PetscInt *ygarray, PetscInt **xtoy)
316 {
317   PetscErrorCode ierr;
318   PetscInt       row,i,nz,xcol,ycol,jx,jy,*x2y;
319 
320   PetscFunctionBegin;
321   ierr = PetscMalloc(xi[m]*sizeof(PetscInt),&x2y);CHKERRQ(ierr);
322   i = 0;
323   for (row=0; row<m; row++){
324     nz = xi[1] - xi[0];
325     jy = 0;
326     for (jx=0; jx<nz; jx++,jy++){
327       if (xgarray && ygarray){
328         xcol = xgarray[xj[*xi + jx]];
329         ycol = ygarray[yj[*yi + jy]];
330       } else {
331         xcol = xj[*xi + jx];
332         ycol = yj[*yi + jy];  /* col index for y */
333       }
334       while ( ycol < xcol ) {
335         jy++;
336         if (ygarray){
337           ycol = ygarray[yj[*yi + jy]];
338         } else {
339           ycol = yj[*yi + jy];
340         }
341       }
342       if (xcol != ycol) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"X matrix entry (%D,%D) is not in Y matrix",row,ycol);
343       x2y[i++] = *yi + jy;
344     }
345     xi++; yi++;
346   }
347   *xtoy = x2y;
348   PetscFunctionReturn(0);
349 }
350