1: #include <petsc/private/matimpl.h>

  3: /*@
  4:   MatGetColumnVector - Gets the values from a given column of a matrix.

  6:   Not Collective

  8:   Input Parameters:
  9: + A   - the matrix
 10: . yy  - the vector
 11: - col - the column requested (in global numbering)

 13:   Level: advanced

 15:   Notes:
 16:   If a `MatType` does not implement the operation, each processor for which this is called
 17:   gets the values for its rows using `MatGetRow()`.

 19:   The vector must have the same parallel row layout as the matrix.

 21:   Contributed by: Denis Vanderstraeten

 23: .seealso: [](ch_matrices), `Mat`, `MatGetRow()`, `MatGetDiagonal()`, `MatMult()`
 24: @*/
 25: PetscErrorCode MatGetColumnVector(Mat A, Vec yy, PetscInt col)
 26: {
 27:   PetscScalar       *y;
 28:   const PetscScalar *v;
 29:   PetscInt           i, j, nz, N, Rs, Re, rs, re;
 30:   const PetscInt    *idx;

 32:   PetscFunctionBegin;
 36:   PetscCheck(col >= 0, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_OUTOFRANGE, "Requested negative column: %" PetscInt_FMT, col);
 37:   PetscCall(MatGetSize(A, NULL, &N));
 38:   PetscCheck(col < N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_OUTOFRANGE, "Requested column %" PetscInt_FMT " larger than number columns in matrix %" PetscInt_FMT, col, N);
 39:   PetscCall(MatGetOwnershipRange(A, &Rs, &Re));
 40:   PetscCall(VecGetOwnershipRange(yy, &rs, &re));
 41:   PetscCheck(Rs == rs && Re == re, PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "Matrix %" PetscInt_FMT " %" PetscInt_FMT " does not have same ownership range (size) as vector %" PetscInt_FMT " %" PetscInt_FMT, Rs, Re, rs, re);

 43:   if (A->ops->getcolumnvector) PetscUseTypeMethod(A, getcolumnvector, yy, col);
 44:   else {
 45:     PetscCall(VecSet(yy, 0.0));
 46:     PetscCall(VecGetArray(yy, &y));
 47:     /* TODO for general matrices */
 48:     for (i = Rs; i < Re; i++) {
 49:       PetscCall(MatGetRow(A, i, &nz, &idx, &v));
 50:       if (nz && idx[0] <= col) {
 51:         /*
 52:           Should use faster search here
 53:         */
 54:         for (j = 0; j < nz; j++) {
 55:           if (idx[j] >= col) {
 56:             if (idx[j] == col) y[i - rs] = v[j];
 57:             break;
 58:           }
 59:         }
 60:       }
 61:       PetscCall(MatRestoreRow(A, i, &nz, &idx, &v));
 62:     }
 63:     PetscCall(VecRestoreArray(yy, &y));
 64:   }
 65:   PetscFunctionReturn(PETSC_SUCCESS);
 66: }

 68: /*@
 69:   MatGetColumnNorms - Gets the norms of each column of a sparse or dense matrix.

 71:   Input Parameters:
 72: + A    - the matrix
 73: - type - `NORM_2`, `NORM_1` or `NORM_INFINITY`

 75:   Output Parameter:
 76: . norms - an array as large as the TOTAL number of columns in the matrix

 78:   Level: intermediate

 80:   Note:
 81:   Each process has ALL the column norms after the call. Because of the way this is computed each process gets all the values,
 82:   if each process wants only some of the values it should extract the ones it wants from the array.

 84: .seealso: [](ch_matrices), `Mat`, `NormType`, `MatNorm()`
 85: @*/
 86: PetscErrorCode MatGetColumnNorms(Mat A, NormType type, PetscReal norms[])
 87: {
 88:   /* NOTE: MatGetColumnNorms() could simply be a macro that calls MatGetColumnReductions().
 89:    * I've kept this as a function because it allows slightly more in the way of error checking,
 90:    * erroring out if MatGetColumnNorms() is not called with a valid NormType. */

 92:   PetscFunctionBegin;
 93:   if (type == NORM_2 || type == NORM_1 || type == NORM_FROBENIUS || type == NORM_INFINITY || type == NORM_1_AND_2) {
 94:     PetscCall(MatGetColumnReductions(A, type, norms));
 95:   } else SETERRQ(PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Unknown NormType");
 96:   PetscFunctionReturn(PETSC_SUCCESS);
 97: }

 99: /*@
100:   MatGetColumnSumsRealPart - Gets the sums of the real part of each column of a sparse or dense matrix.

102:   Input Parameter:
103: . A - the matrix

105:   Output Parameter:
106: . sums - an array as large as the TOTAL number of columns in the matrix

108:   Level: intermediate

110:   Note:
111:   Each process has ALL the column sums after the call. Because of the way this is computed each process gets all the values,
112:   if each process wants only some of the values it should extract the ones it wants from the array.

114: .seealso: [](ch_matrices), `Mat`, `MatGetColumnSumsImaginaryPart()`, `VecSum()`, `MatGetColumnMeans()`, `MatGetColumnNorms()`, `MatGetColumnReductions()`
115: @*/
116: PetscErrorCode MatGetColumnSumsRealPart(Mat A, PetscReal sums[])
117: {
118:   PetscFunctionBegin;
119:   PetscCall(MatGetColumnReductions(A, REDUCTION_SUM_REALPART, sums));
120:   PetscFunctionReturn(PETSC_SUCCESS);
121: }

123: /*@
124:   MatGetColumnSumsImaginaryPart - Gets the sums of the imaginary part of each column of a sparse or dense matrix.

126:   Input Parameter:
127: . A - the matrix

129:   Output Parameter:
130: . sums - an array as large as the TOTAL number of columns in the matrix

132:   Level: intermediate

134:   Note:
135:   Each process has ALL the column sums after the call. Because of the way this is computed each process gets all the values,
136:   if each process wants only some of the values it should extract the ones it wants from the array.

138: .seealso: [](ch_matrices), `Mat`, `MatGetColumnSumsRealPart()`, `VecSum()`, `MatGetColumnMeans()`, `MatGetColumnNorms()`, `MatGetColumnReductions()`
139: @*/
140: PetscErrorCode MatGetColumnSumsImaginaryPart(Mat A, PetscReal sums[])
141: {
142:   PetscFunctionBegin;
143:   PetscCall(MatGetColumnReductions(A, REDUCTION_SUM_IMAGINARYPART, sums));
144:   PetscFunctionReturn(PETSC_SUCCESS);
145: }

147: /*@
148:   MatGetColumnSums - Gets the sums of each column of a sparse or dense matrix.

150:   Input Parameter:
151: . A - the matrix

153:   Output Parameter:
154: . sums - an array as large as the TOTAL number of columns in the matrix

156:   Level: intermediate

158:   Note:
159:   Each process has ALL the column sums after the call. Because of the way this is computed each process gets all the values,
160:   if each process wants only some of the values it should extract the ones it wants from the array.

162: .seealso: [](ch_matrices), `Mat`, `VecSum()`, `MatGetColumnMeans()`, `MatGetColumnNorms()`, `MatGetColumnReductions()`
163: @*/
164: PetscErrorCode MatGetColumnSums(Mat A, PetscScalar sums[])
165: {
166: #if defined(PETSC_USE_COMPLEX)
167:   PetscInt   i, n;
168:   PetscReal *work;
169: #endif

171:   PetscFunctionBegin;

173: #if !defined(PETSC_USE_COMPLEX)
174:   PetscCall(MatGetColumnSumsRealPart(A, sums));
175: #else
176:   PetscCall(MatGetSize(A, NULL, &n));
177:   PetscCall(PetscArrayzero(sums, n));
178:   PetscCall(PetscCalloc1(n, &work));
179:   PetscCall(MatGetColumnSumsRealPart(A, work));
180:   for (i = 0; i < n; i++) sums[i] = work[i];
181:   PetscCall(MatGetColumnSumsImaginaryPart(A, work));
182:   for (i = 0; i < n; i++) sums[i] += work[i] * PETSC_i;
183:   PetscCall(PetscFree(work));
184: #endif
185:   PetscFunctionReturn(PETSC_SUCCESS);
186: }

188: /*@
189:   MatGetColumnMeansRealPart - Gets the arithmetic means of the real part of each column of a sparse or dense matrix.

191:   Input Parameter:
192: . A - the matrix

194:   Output Parameter:
195: . means - an array as large as the TOTAL number of columns in the matrix

197:   Level: intermediate

199:   Note:
200:   Each process has ALL the column means after the call. Because of the way this is computed each process gets all the values,
201:   if each process wants only some of the values it should extract the ones it wants from the array.

203: .seealso: [](ch_matrices), `Mat`, `MatGetColumnMeansImaginaryPart()`, `VecSum()`, `MatGetColumnSums()`, `MatGetColumnNorms()`, `MatGetColumnReductions()`
204: @*/
205: PetscErrorCode MatGetColumnMeansRealPart(Mat A, PetscReal means[])
206: {
207:   PetscFunctionBegin;
208:   PetscCall(MatGetColumnReductions(A, REDUCTION_MEAN_REALPART, means));
209:   PetscFunctionReturn(PETSC_SUCCESS);
210: }

212: /*@
213:   MatGetColumnMeansImaginaryPart - Gets the arithmetic means of the imaginary part of each column of a sparse or dense matrix.

215:   Input Parameter:
216: . A - the matrix

218:   Output Parameter:
219: . means - an array as large as the TOTAL number of columns in the matrix

221:   Level: intermediate

223:   Note:
224:   Each process has ALL the column means after the call. Because of the way this is computed each process gets all the values,
225:   if each process wants only some of the values it should extract the ones it wants from the array.

227: .seealso: [](ch_matrices), `Mat`, `MatGetColumnMeansRealPart()`, `VecSum()`, `MatGetColumnSums()`, `MatGetColumnNorms()`, `MatGetColumnReductions()`
228: @*/
229: PetscErrorCode MatGetColumnMeansImaginaryPart(Mat A, PetscReal means[])
230: {
231:   PetscFunctionBegin;
232:   PetscCall(MatGetColumnReductions(A, REDUCTION_MEAN_IMAGINARYPART, means));
233:   PetscFunctionReturn(PETSC_SUCCESS);
234: }

236: /*@
237:   MatGetColumnMeans - Gets the arithmetic means of each column of a sparse or dense matrix.

239:   Input Parameter:
240: . A - the matrix

242:   Output Parameter:
243: . means - an array as large as the TOTAL number of columns in the matrix

245:   Level: intermediate

247:   Note:
248:   Each process has ALL the column means after the call. Because of the way this is computed each process gets all the values,
249:   if each process wants only some of the values it should extract the ones it wants from the array.

251: .seealso: [](ch_matrices), `Mat`, `VecSum()`, `MatGetColumnSums()`, `MatGetColumnNorms()`, `MatGetColumnReductions()`
252: @*/
253: PetscErrorCode MatGetColumnMeans(Mat A, PetscScalar means[])
254: {
255: #if defined(PETSC_USE_COMPLEX)
256:   PetscInt   i, n;
257:   PetscReal *work;
258: #endif

260:   PetscFunctionBegin;

262: #if !defined(PETSC_USE_COMPLEX)
263:   PetscCall(MatGetColumnMeansRealPart(A, means));
264: #else
265:   PetscCall(MatGetSize(A, NULL, &n));
266:   PetscCall(PetscArrayzero(means, n));
267:   PetscCall(PetscCalloc1(n, &work));
268:   PetscCall(MatGetColumnMeansRealPart(A, work));
269:   for (i = 0; i < n; i++) means[i] = work[i];
270:   PetscCall(MatGetColumnMeansImaginaryPart(A, work));
271:   for (i = 0; i < n; i++) means[i] += work[i] * PETSC_i;
272:   PetscCall(PetscFree(work));
273: #endif
274:   PetscFunctionReturn(PETSC_SUCCESS);
275: }

277: /*@
278:   MatGetColumnReductions - Gets the reductions of each column of a sparse or dense matrix.

280:   Input Parameters:
281: + A    - the matrix
282: - type - A constant defined in `NormType` or `ReductionType`: `NORM_2`, `NORM_1`, `NORM_INFINITY`, `REDUCTION_SUM_REALPART`,
283:           `REDUCTION_SUM_IMAGINARYPART`, `REDUCTION_MEAN_REALPART`, `REDUCTION_MEAN_IMAGINARYPART`

285:   Output Parameter:
286: . reductions - an array as large as the TOTAL number of columns in the matrix

288:   Level: developer

290:   Note:
291:   Each process has ALL the column reductions after the call. Because of the way this is computed each process gets all the values,
292:   if each process wants only some of the values it should extract the ones it wants from the array.

294:   Developer Notes:
295:   This routine is primarily intended as a back-end.
296:   `MatGetColumnNorms()`, `MatGetColumnSums()`, and `MatGetColumnMeans()` are implemented using this routine.

298: .seealso: [](ch_matrices), `Mat`, `ReductionType`, `NormType`, `MatGetColumnNorms()`, `MatGetColumnSums()`, `MatGetColumnMeans()`
299: @*/
300: PetscErrorCode MatGetColumnReductions(Mat A, PetscInt type, PetscReal reductions[])
301: {
302:   PetscFunctionBegin;
304:   PetscUseTypeMethod(A, getcolumnreductions, type, reductions);
305:   PetscFunctionReturn(PETSC_SUCCESS);
306: }