Actual source code: dsnhepts.c
slepc-main 2024-05-17
1: /*
2: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
3: SLEPc - Scalable Library for Eigenvalue Problem Computations
4: Copyright (c) 2002-, Universitat Politecnica de Valencia, Spain
6: This file is part of SLEPc.
7: SLEPc is distributed under a 2-clause BSD license (see LICENSE).
8: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
9: */
11: #include <slepc/private/dsimpl.h>
12: #include <slepcblaslapack.h>
14: typedef struct {
15: PetscScalar *wr,*wi; /* eigenvalues of B */
16: } DS_NHEPTS;
18: static PetscErrorCode DSAllocate_NHEPTS(DS ds,PetscInt ld)
19: {
20: DS_NHEPTS *ctx = (DS_NHEPTS*)ds->data;
22: PetscFunctionBegin;
23: PetscCall(DSAllocateMat_Private(ds,DS_MAT_A));
24: PetscCall(DSAllocateMat_Private(ds,DS_MAT_B));
25: PetscCall(DSAllocateMat_Private(ds,DS_MAT_Q));
26: PetscCall(DSAllocateMat_Private(ds,DS_MAT_Z));
27: PetscCall(PetscFree(ds->perm));
28: PetscCall(PetscMalloc1(ld,&ds->perm));
29: PetscCall(PetscMalloc1(ld,&ctx->wr));
30: #if !defined(PETSC_USE_COMPLEX)
31: PetscCall(PetscMalloc1(ld,&ctx->wi));
32: #endif
33: PetscFunctionReturn(PETSC_SUCCESS);
34: }
36: static PetscErrorCode DSView_NHEPTS(DS ds,PetscViewer viewer)
37: {
38: PetscViewerFormat format;
40: PetscFunctionBegin;
41: PetscCall(PetscViewerGetFormat(viewer,&format));
42: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) PetscFunctionReturn(PETSC_SUCCESS);
43: PetscCall(DSViewMat(ds,viewer,DS_MAT_A));
44: PetscCall(DSViewMat(ds,viewer,DS_MAT_B));
45: if (ds->state>DS_STATE_INTERMEDIATE) {
46: PetscCall(DSViewMat(ds,viewer,DS_MAT_Q));
47: PetscCall(DSViewMat(ds,viewer,DS_MAT_Z));
48: }
49: if (ds->omat[DS_MAT_X]) PetscCall(DSViewMat(ds,viewer,DS_MAT_X));
50: if (ds->omat[DS_MAT_Y]) PetscCall(DSViewMat(ds,viewer,DS_MAT_Y));
51: PetscFunctionReturn(PETSC_SUCCESS);
52: }
54: static PetscErrorCode DSVectors_NHEPTS_Eigen_Some(DS ds,PetscInt *k,PetscReal *rnorm,PetscBool left)
55: {
56: PetscInt i;
57: PetscBLASInt mm=1,mout,info,ld,n,*select,inc=1,cols=1,zero=0;
58: PetscScalar sone=1.0,szero=0.0;
59: PetscReal norm,done=1.0;
60: PetscBool iscomplex = PETSC_FALSE;
61: PetscScalar *X,*Y;
62: const PetscScalar *A,*Q;
64: PetscFunctionBegin;
65: PetscCall(PetscBLASIntCast(ds->n,&n));
66: PetscCall(PetscBLASIntCast(ds->ld,&ld));
67: PetscCall(DSAllocateWork_Private(ds,0,0,ld));
68: select = ds->iwork;
69: for (i=0;i<n;i++) select[i] = (PetscBLASInt)PETSC_FALSE;
71: /* compute k-th eigenvector Y of A */
72: PetscCall(MatDenseGetArrayRead(ds->omat[left?DS_MAT_B:DS_MAT_A],&A));
73: PetscCall(MatDenseGetArray(ds->omat[left?DS_MAT_Y:DS_MAT_X],&X));
74: Y = X+(*k)*ld;
75: select[*k] = (PetscBLASInt)PETSC_TRUE;
76: #if !defined(PETSC_USE_COMPLEX)
77: if ((*k)<n-1 && A[(*k)+1+(*k)*ld]!=0.0) iscomplex = PETSC_TRUE;
78: mm = iscomplex? 2: 1;
79: if (iscomplex) select[(*k)+1] = (PetscBLASInt)PETSC_TRUE;
80: PetscCall(DSAllocateWork_Private(ds,3*ld,0,0));
81: PetscCallBLAS("LAPACKtrevc",LAPACKtrevc_("R","S",select,&n,(PetscScalar*)A,&ld,Y,&ld,Y,&ld,&mm,&mout,ds->work,&info));
82: #else
83: PetscCall(DSAllocateWork_Private(ds,2*ld,ld,0));
84: PetscCallBLAS("LAPACKtrevc",LAPACKtrevc_("R","S",select,&n,(PetscScalar*)A,&ld,Y,&ld,Y,&ld,&mm,&mout,ds->work,ds->rwork,&info));
85: #endif
86: SlepcCheckLapackInfo("trevc",info);
87: PetscCheck(mout==mm,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Inconsistent arguments");
88: PetscCall(MatDenseRestoreArrayRead(ds->omat[left?DS_MAT_B:DS_MAT_A],&A));
90: /* accumulate and normalize eigenvectors */
91: if (ds->state>=DS_STATE_CONDENSED) {
92: PetscCall(MatDenseGetArrayRead(ds->omat[left?DS_MAT_Z:DS_MAT_Q],&Q));
93: PetscCall(PetscArraycpy(ds->work,Y,mout*ld));
94: PetscCallBLAS("BLASgemv",BLASgemv_("N",&n,&n,&sone,Q,&ld,ds->work,&inc,&szero,Y,&inc));
95: #if !defined(PETSC_USE_COMPLEX)
96: if (iscomplex) PetscCallBLAS("BLASgemv",BLASgemv_("N",&n,&n,&sone,Q,&ld,ds->work+ld,&inc,&szero,Y+ld,&inc));
97: #endif
98: PetscCall(MatDenseRestoreArrayRead(ds->omat[left?DS_MAT_Z:DS_MAT_Q],&Q));
99: cols = 1;
100: norm = BLASnrm2_(&n,Y,&inc);
101: #if !defined(PETSC_USE_COMPLEX)
102: if (iscomplex) {
103: norm = SlepcAbsEigenvalue(norm,BLASnrm2_(&n,Y+ld,&inc));
104: cols = 2;
105: }
106: #endif
107: PetscCallBLAS("LAPACKlascl",LAPACKlascl_("G",&zero,&zero,&norm,&done,&n,&cols,Y,&ld,&info));
108: SlepcCheckLapackInfo("lascl",info);
109: }
111: /* set output arguments */
112: if (iscomplex) (*k)++;
113: if (rnorm) {
114: if (iscomplex) *rnorm = SlepcAbsEigenvalue(Y[n-1],Y[n-1+ld]);
115: else *rnorm = PetscAbsScalar(Y[n-1]);
116: }
117: PetscCall(MatDenseRestoreArray(ds->omat[left?DS_MAT_Y:DS_MAT_X],&X));
118: PetscFunctionReturn(PETSC_SUCCESS);
119: }
121: static PetscErrorCode DSVectors_NHEPTS_Eigen_All(DS ds,PetscBool left)
122: {
123: PetscInt i;
124: PetscBLASInt n,ld,mout,info,inc=1,cols,zero=0;
125: PetscBool iscomplex;
126: PetscScalar *X;
127: const PetscScalar *A;
128: PetscReal norm,done=1.0;
129: const char *back;
131: PetscFunctionBegin;
132: PetscCall(PetscBLASIntCast(ds->n,&n));
133: PetscCall(PetscBLASIntCast(ds->ld,&ld));
134: PetscCall(MatDenseGetArrayRead(ds->omat[left?DS_MAT_B:DS_MAT_A],&A));
135: PetscCall(MatDenseGetArrayWrite(ds->omat[left?DS_MAT_Y:DS_MAT_X],&X));
136: if (ds->state>=DS_STATE_CONDENSED) {
137: /* DSSolve() has been called, backtransform with matrix Q */
138: back = "B";
139: PetscCall(MatCopy(ds->omat[left?DS_MAT_Z:DS_MAT_Q],ds->omat[left?DS_MAT_Y:DS_MAT_X],SAME_NONZERO_PATTERN));
140: } else back = "A";
141: #if !defined(PETSC_USE_COMPLEX)
142: PetscCall(DSAllocateWork_Private(ds,3*ld,0,0));
143: PetscCallBLAS("LAPACKtrevc",LAPACKtrevc_("R",back,NULL,&n,(PetscScalar*)A,&ld,X,&ld,X,&ld,&n,&mout,ds->work,&info));
144: #else
145: PetscCall(DSAllocateWork_Private(ds,2*ld,ld,0));
146: PetscCallBLAS("LAPACKtrevc",LAPACKtrevc_("R",back,NULL,&n,(PetscScalar*)A,&ld,X,&ld,X,&ld,&n,&mout,ds->work,ds->rwork,&info));
147: #endif
148: SlepcCheckLapackInfo("trevc",info);
150: /* normalize eigenvectors */
151: for (i=0;i<n;i++) {
152: iscomplex = (i<n-1 && A[i+1+i*ld]!=0.0)? PETSC_TRUE: PETSC_FALSE;
153: cols = 1;
154: norm = BLASnrm2_(&n,X+i*ld,&inc);
155: #if !defined(PETSC_USE_COMPLEX)
156: if (iscomplex) {
157: norm = SlepcAbsEigenvalue(norm,BLASnrm2_(&n,X+(i+1)*ld,&inc));
158: cols = 2;
159: }
160: #endif
161: PetscCallBLAS("LAPACKlascl",LAPACKlascl_("G",&zero,&zero,&norm,&done,&n,&cols,X+i*ld,&ld,&info));
162: SlepcCheckLapackInfo("lascl",info);
163: if (iscomplex) i++;
164: }
165: PetscCall(MatDenseRestoreArrayRead(ds->omat[left?DS_MAT_B:DS_MAT_A],&A));
166: PetscCall(MatDenseRestoreArrayWrite(ds->omat[left?DS_MAT_Y:DS_MAT_X],&X));
167: PetscFunctionReturn(PETSC_SUCCESS);
168: }
170: static PetscErrorCode DSVectors_NHEPTS(DS ds,DSMatType mat,PetscInt *j,PetscReal *rnorm)
171: {
172: PetscFunctionBegin;
173: switch (mat) {
174: case DS_MAT_X:
175: PetscCheck(!ds->refined,PetscObjectComm((PetscObject)ds),PETSC_ERR_SUP,"Not implemented yet");
176: if (j) PetscCall(DSVectors_NHEPTS_Eigen_Some(ds,j,rnorm,PETSC_FALSE));
177: else PetscCall(DSVectors_NHEPTS_Eigen_All(ds,PETSC_FALSE));
178: break;
179: case DS_MAT_Y:
180: PetscCheck(!ds->refined,PetscObjectComm((PetscObject)ds),PETSC_ERR_SUP,"Not implemented yet");
181: if (j) PetscCall(DSVectors_NHEPTS_Eigen_Some(ds,j,rnorm,PETSC_TRUE));
182: else PetscCall(DSVectors_NHEPTS_Eigen_All(ds,PETSC_TRUE));
183: break;
184: case DS_MAT_U:
185: case DS_MAT_V:
186: SETERRQ(PetscObjectComm((PetscObject)ds),PETSC_ERR_SUP,"Not implemented yet");
187: default:
188: SETERRQ(PetscObjectComm((PetscObject)ds),PETSC_ERR_ARG_OUTOFRANGE,"Invalid mat parameter");
189: }
190: PetscFunctionReturn(PETSC_SUCCESS);
191: }
193: static PetscErrorCode DSSort_NHEPTS(DS ds,PetscScalar *wr,PetscScalar *wi,PetscScalar *rr,PetscScalar *ri,PetscInt *k)
194: {
195: DS_NHEPTS *ctx = (DS_NHEPTS*)ds->data;
196: PetscInt i,j,cont,id=0,*p,*idx,*idx2;
197: PetscReal s,t;
198: #if defined(PETSC_USE_COMPLEX)
199: Mat A,U;
200: #endif
202: PetscFunctionBegin;
203: PetscCheck(!rr || wr==rr,PetscObjectComm((PetscObject)ds),PETSC_ERR_SUP,"Not implemented yet");
204: PetscCall(PetscMalloc3(ds->ld,&idx,ds->ld,&idx2,ds->ld,&p));
205: PetscCall(DSSort_NHEP_Total(ds,DS_MAT_A,DS_MAT_Q,wr,wi));
206: #if defined(PETSC_USE_COMPLEX)
207: PetscCall(DSGetMat(ds,DS_MAT_B,&A));
208: PetscCall(MatConjugate(A));
209: PetscCall(DSRestoreMat(ds,DS_MAT_B,&A));
210: PetscCall(DSGetMat(ds,DS_MAT_Z,&U));
211: PetscCall(MatConjugate(U));
212: PetscCall(DSRestoreMat(ds,DS_MAT_Z,&U));
213: for (i=0;i<ds->n;i++) ctx->wr[i] = PetscConj(ctx->wr[i]);
214: #endif
215: PetscCall(DSSort_NHEP_Total(ds,DS_MAT_B,DS_MAT_Z,ctx->wr,ctx->wi));
216: /* check correct eigenvalue correspondence */
217: cont = 0;
218: for (i=0;i<ds->n;i++) {
219: if (SlepcAbsEigenvalue(ctx->wr[i]-wr[i],ctx->wi[i]-wi[i])>PETSC_SQRT_MACHINE_EPSILON) {idx2[cont] = i; idx[cont++] = i;}
220: p[i] = -1;
221: }
222: if (cont) {
223: for (i=0;i<cont;i++) {
224: t = PETSC_MAX_REAL;
225: for (j=0;j<cont;j++) if (idx2[j]!=-1 && (s=SlepcAbsEigenvalue(ctx->wr[idx[j]]-wr[idx[i]],ctx->wi[idx[j]]-wi[idx[i]]))<t) { id = j; t = s; }
226: p[idx[i]] = idx[id];
227: idx2[id] = -1;
228: }
229: for (i=0;i<ds->n;i++) if (p[i]==-1) p[i] = i;
230: PetscCall(DSSortWithPermutation_NHEP_Private(ds,p,DS_MAT_B,DS_MAT_Z,ctx->wr,ctx->wi));
231: }
232: #if defined(PETSC_USE_COMPLEX)
233: PetscCall(DSGetMat(ds,DS_MAT_B,&A));
234: PetscCall(MatConjugate(A));
235: PetscCall(DSRestoreMat(ds,DS_MAT_B,&A));
236: PetscCall(DSGetMat(ds,DS_MAT_Z,&U));
237: PetscCall(MatConjugate(U));
238: PetscCall(DSRestoreMat(ds,DS_MAT_Z,&U));
239: #endif
240: PetscCall(PetscFree3(idx,idx2,p));
241: PetscFunctionReturn(PETSC_SUCCESS);
242: }
244: static PetscErrorCode DSUpdateExtraRow_NHEPTS(DS ds)
245: {
246: PetscInt i;
247: PetscBLASInt n,ld,incx=1;
248: PetscScalar *A,*x,*y,one=1.0,zero=0.0;
249: const PetscScalar *Q;
251: PetscFunctionBegin;
252: PetscCall(PetscBLASIntCast(ds->n,&n));
253: PetscCall(PetscBLASIntCast(ds->ld,&ld));
254: PetscCall(DSAllocateWork_Private(ds,2*ld,0,0));
255: x = ds->work;
256: y = ds->work+ld;
257: PetscCall(MatDenseGetArray(ds->omat[DS_MAT_A],&A));
258: PetscCall(MatDenseGetArrayRead(ds->omat[DS_MAT_Q],&Q));
259: for (i=0;i<n;i++) x[i] = PetscConj(A[n+i*ld]);
260: PetscCallBLAS("BLASgemv",BLASgemv_("C",&n,&n,&one,Q,&ld,x,&incx,&zero,y,&incx));
261: for (i=0;i<n;i++) A[n+i*ld] = PetscConj(y[i]);
262: PetscCall(MatDenseRestoreArray(ds->omat[DS_MAT_A],&A));
263: PetscCall(MatDenseRestoreArrayRead(ds->omat[DS_MAT_Q],&Q));
264: PetscCall(MatDenseGetArray(ds->omat[DS_MAT_B],&A));
265: PetscCall(MatDenseGetArrayRead(ds->omat[DS_MAT_Z],&Q));
266: for (i=0;i<n;i++) x[i] = PetscConj(A[n+i*ld]);
267: PetscCallBLAS("BLASgemv",BLASgemv_("C",&n,&n,&one,Q,&ld,x,&incx,&zero,y,&incx));
268: for (i=0;i<n;i++) A[n+i*ld] = PetscConj(y[i]);
269: PetscCall(MatDenseRestoreArray(ds->omat[DS_MAT_B],&A));
270: PetscCall(MatDenseRestoreArrayRead(ds->omat[DS_MAT_Z],&Q));
271: ds->k = n;
272: PetscFunctionReturn(PETSC_SUCCESS);
273: }
275: static PetscErrorCode DSSolve_NHEPTS(DS ds,PetscScalar *wr,PetscScalar *wi)
276: {
277: DS_NHEPTS *ctx = (DS_NHEPTS*)ds->data;
279: PetscFunctionBegin;
280: #if !defined(PETSC_USE_COMPLEX)
281: PetscAssertPointer(wi,3);
282: #endif
283: PetscCall(DSSolve_NHEP_Private(ds,DS_MAT_A,DS_MAT_Q,wr,wi));
284: PetscCall(DSSolve_NHEP_Private(ds,DS_MAT_B,DS_MAT_Z,ctx->wr,ctx->wi));
285: PetscFunctionReturn(PETSC_SUCCESS);
286: }
288: #if !defined(PETSC_HAVE_MPIUNI)
289: static PetscErrorCode DSSynchronize_NHEPTS(DS ds,PetscScalar eigr[],PetscScalar eigi[])
290: {
291: PetscInt ld=ds->ld,l=ds->l,k;
292: PetscMPIInt n,rank,off=0,size,ldn;
293: DS_NHEPTS *ctx = (DS_NHEPTS*)ds->data;
294: PetscScalar *A,*B,*Q,*Z;
296: PetscFunctionBegin;
297: k = 2*(ds->n-l)*ld;
298: if (ds->state>DS_STATE_RAW) k += 2*(ds->n-l)*ld;
299: if (eigr) k += ds->n-l;
300: if (eigi) k += ds->n-l;
301: if (ctx->wr) k += ds->n-l;
302: if (ctx->wi) k += ds->n-l;
303: PetscCall(DSAllocateWork_Private(ds,k,0,0));
304: PetscCall(PetscMPIIntCast(k*sizeof(PetscScalar),&size));
305: PetscCall(PetscMPIIntCast(ds->n-l,&n));
306: PetscCall(PetscMPIIntCast(ld*(ds->n-l),&ldn));
307: PetscCall(MatDenseGetArray(ds->omat[DS_MAT_A],&A));
308: PetscCall(MatDenseGetArray(ds->omat[DS_MAT_B],&B));
309: if (ds->state>DS_STATE_RAW) {
310: PetscCall(MatDenseGetArray(ds->omat[DS_MAT_Q],&Q));
311: PetscCall(MatDenseGetArray(ds->omat[DS_MAT_Z],&Z));
312: }
313: PetscCallMPI(MPI_Comm_rank(PetscObjectComm((PetscObject)ds),&rank));
314: if (!rank) {
315: PetscCallMPI(MPI_Pack(A+l*ld,ldn,MPIU_SCALAR,ds->work,size,&off,PetscObjectComm((PetscObject)ds)));
316: PetscCallMPI(MPI_Pack(B+l*ld,ldn,MPIU_SCALAR,ds->work,size,&off,PetscObjectComm((PetscObject)ds)));
317: if (ds->state>DS_STATE_RAW) {
318: PetscCallMPI(MPI_Pack(Q+l*ld,ldn,MPIU_SCALAR,ds->work,size,&off,PetscObjectComm((PetscObject)ds)));
319: PetscCallMPI(MPI_Pack(Z+l*ld,ldn,MPIU_SCALAR,ds->work,size,&off,PetscObjectComm((PetscObject)ds)));
320: }
321: if (eigr) PetscCallMPI(MPI_Pack(eigr+l,n,MPIU_SCALAR,ds->work,size,&off,PetscObjectComm((PetscObject)ds)));
322: #if !defined(PETSC_USE_COMPLEX)
323: if (eigi) PetscCallMPI(MPI_Pack(eigi+l,n,MPIU_SCALAR,ds->work,size,&off,PetscObjectComm((PetscObject)ds)));
324: #endif
325: if (ctx->wr) PetscCallMPI(MPI_Pack(ctx->wr+l,n,MPIU_SCALAR,ds->work,size,&off,PetscObjectComm((PetscObject)ds)));
326: if (ctx->wi) PetscCallMPI(MPI_Pack(ctx->wi+l,n,MPIU_SCALAR,ds->work,size,&off,PetscObjectComm((PetscObject)ds)));
327: }
328: PetscCallMPI(MPI_Bcast(ds->work,size,MPI_BYTE,0,PetscObjectComm((PetscObject)ds)));
329: if (rank) {
330: PetscCallMPI(MPI_Unpack(ds->work,size,&off,A+l*ld,ldn,MPIU_SCALAR,PetscObjectComm((PetscObject)ds)));
331: PetscCallMPI(MPI_Unpack(ds->work,size,&off,B+l*ld,ldn,MPIU_SCALAR,PetscObjectComm((PetscObject)ds)));
332: if (ds->state>DS_STATE_RAW) {
333: PetscCallMPI(MPI_Unpack(ds->work,size,&off,Q+l*ld,ldn,MPIU_SCALAR,PetscObjectComm((PetscObject)ds)));
334: PetscCallMPI(MPI_Unpack(ds->work,size,&off,Z+l*ld,ldn,MPIU_SCALAR,PetscObjectComm((PetscObject)ds)));
335: }
336: if (eigr) PetscCallMPI(MPI_Unpack(ds->work,size,&off,eigr+l,n,MPIU_SCALAR,PetscObjectComm((PetscObject)ds)));
337: #if !defined(PETSC_USE_COMPLEX)
338: if (eigi) PetscCallMPI(MPI_Unpack(ds->work,size,&off,eigi+l,n,MPIU_SCALAR,PetscObjectComm((PetscObject)ds)));
339: #endif
340: if (ctx->wr) PetscCallMPI(MPI_Unpack(ds->work,size,&off,ctx->wr+l,n,MPIU_SCALAR,PetscObjectComm((PetscObject)ds)));
341: if (ctx->wi) PetscCallMPI(MPI_Unpack(ds->work,size,&off,ctx->wi+l,n,MPIU_SCALAR,PetscObjectComm((PetscObject)ds)));
342: }
343: PetscCall(MatDenseRestoreArray(ds->omat[DS_MAT_A],&A));
344: PetscCall(MatDenseRestoreArray(ds->omat[DS_MAT_B],&B));
345: if (ds->state>DS_STATE_RAW) {
346: PetscCall(MatDenseRestoreArray(ds->omat[DS_MAT_Q],&Q));
347: PetscCall(MatDenseRestoreArray(ds->omat[DS_MAT_Z],&Z));
348: }
349: PetscFunctionReturn(PETSC_SUCCESS);
350: }
351: #endif
353: static PetscErrorCode DSGetTruncateSize_NHEPTS(DS ds,PetscInt l,PetscInt n,PetscInt *k)
354: {
355: #if !defined(PETSC_USE_COMPLEX)
356: const PetscScalar *A,*B;
357: #endif
359: PetscFunctionBegin;
360: #if !defined(PETSC_USE_COMPLEX)
361: PetscCall(MatDenseGetArrayRead(ds->omat[DS_MAT_A],&A));
362: PetscCall(MatDenseGetArrayRead(ds->omat[DS_MAT_B],&B));
363: if (A[l+(*k)+(l+(*k)-1)*ds->ld] != 0.0 || B[l+(*k)+(l+(*k)-1)*ds->ld] != 0.0) {
364: if (l+(*k)<n-1) (*k)++;
365: else (*k)--;
366: }
367: PetscCall(MatDenseRestoreArrayRead(ds->omat[DS_MAT_A],&A));
368: PetscCall(MatDenseRestoreArrayRead(ds->omat[DS_MAT_B],&B));
369: #endif
370: PetscFunctionReturn(PETSC_SUCCESS);
371: }
373: static PetscErrorCode DSTruncate_NHEPTS(DS ds,PetscInt n,PetscBool trim)
374: {
375: PetscInt i,ld=ds->ld,l=ds->l;
376: PetscScalar *A,*B;
378: PetscFunctionBegin;
379: PetscCall(MatDenseGetArray(ds->omat[DS_MAT_A],&A));
380: PetscCall(MatDenseGetArray(ds->omat[DS_MAT_B],&B));
381: #if defined(PETSC_USE_DEBUG)
382: /* make sure diagonal 2x2 block is not broken */
383: PetscCheck(ds->state<DS_STATE_CONDENSED || n==0 || n==ds->n || A[n+(n-1)*ld]==0.0 || B[n+(n-1)*ld]==0.0,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"The given size would break a 2x2 block, call DSGetTruncateSize() first");
384: #endif
385: if (trim) {
386: if (ds->extrarow) { /* clean extra row */
387: for (i=l;i<ds->n;i++) { A[ds->n+i*ld] = 0.0; B[ds->n+i*ld] = 0.0; }
388: }
389: ds->l = 0;
390: ds->k = 0;
391: ds->n = n;
392: ds->t = ds->n; /* truncated length equal to the new dimension */
393: } else {
394: if (ds->extrarow && ds->k==ds->n) {
395: /* copy entries of extra row to the new position, then clean last row */
396: for (i=l;i<n;i++) { A[n+i*ld] = A[ds->n+i*ld]; B[n+i*ld] = B[ds->n+i*ld]; }
397: for (i=l;i<ds->n;i++) { A[ds->n+i*ld] = 0.0; B[ds->n+i*ld] = 0.0; }
398: }
399: ds->k = (ds->extrarow)? n: 0;
400: ds->t = ds->n; /* truncated length equal to previous dimension */
401: ds->n = n;
402: }
403: PetscCall(MatDenseRestoreArray(ds->omat[DS_MAT_A],&A));
404: PetscCall(MatDenseRestoreArray(ds->omat[DS_MAT_B],&B));
405: PetscFunctionReturn(PETSC_SUCCESS);
406: }
408: static PetscErrorCode DSDestroy_NHEPTS(DS ds)
409: {
410: DS_NHEPTS *ctx = (DS_NHEPTS*)ds->data;
412: PetscFunctionBegin;
413: if (ctx->wr) PetscCall(PetscFree(ctx->wr));
414: if (ctx->wi) PetscCall(PetscFree(ctx->wi));
415: PetscCall(PetscFree(ds->data));
416: PetscFunctionReturn(PETSC_SUCCESS);
417: }
419: static PetscErrorCode DSMatGetSize_NHEPTS(DS ds,DSMatType t,PetscInt *rows,PetscInt *cols)
420: {
421: PetscFunctionBegin;
422: *rows = ((t==DS_MAT_A || t==DS_MAT_B) && ds->extrarow)? ds->n+1: ds->n;
423: *cols = ds->n;
424: PetscFunctionReturn(PETSC_SUCCESS);
425: }
427: /*MC
428: DSNHEPTS - Dense Non-Hermitian Eigenvalue Problem (special variant intended
429: for two-sided Krylov solvers).
431: Level: beginner
433: Notes:
434: Two related problems are solved, A*X = X*Lambda and B*Y = Y*Lambda', where A and
435: B are supposed to come from the Arnoldi factorizations of a certain matrix and its
436: (conjugate) transpose, respectively. Hence, in exact arithmetic the columns of Y
437: are equal to the left eigenvectors of A. Lambda is a diagonal matrix whose diagonal
438: elements are the arguments of DSSolve(). After solve, A is overwritten with the
439: upper quasi-triangular matrix T of the (real) Schur form, A*Q = Q*T, and similarly
440: another (real) Schur relation is computed, B*Z = Z*S, overwriting B.
442: In the intermediate state A and B are reduced to upper Hessenberg form.
444: When left eigenvectors DS_MAT_Y are requested, right eigenvectors of B are returned,
445: while DS_MAT_X contains right eigenvectors of A.
447: Used DS matrices:
448: + DS_MAT_A - first problem matrix obtained from Arnoldi
449: . DS_MAT_B - second problem matrix obtained from Arnoldi on the transpose
450: . DS_MAT_Q - orthogonal/unitary transformation that reduces A to Hessenberg form
451: (intermediate step) or matrix of orthogonal Schur vectors of A
452: - DS_MAT_Z - orthogonal/unitary transformation that reduces B to Hessenberg form
453: (intermediate step) or matrix of orthogonal Schur vectors of B
455: Implemented methods:
456: . 0 - Implicit QR (_hseqr)
458: .seealso: DSCreate(), DSSetType(), DSType
459: M*/
460: SLEPC_EXTERN PetscErrorCode DSCreate_NHEPTS(DS ds)
461: {
462: DS_NHEPTS *ctx;
464: PetscFunctionBegin;
465: PetscCall(PetscNew(&ctx));
466: ds->data = (void*)ctx;
468: ds->ops->allocate = DSAllocate_NHEPTS;
469: ds->ops->view = DSView_NHEPTS;
470: ds->ops->vectors = DSVectors_NHEPTS;
471: ds->ops->solve[0] = DSSolve_NHEPTS;
472: ds->ops->sort = DSSort_NHEPTS;
473: #if !defined(PETSC_HAVE_MPIUNI)
474: ds->ops->synchronize = DSSynchronize_NHEPTS;
475: #endif
476: ds->ops->gettruncatesize = DSGetTruncateSize_NHEPTS;
477: ds->ops->truncate = DSTruncate_NHEPTS;
478: ds->ops->update = DSUpdateExtraRow_NHEPTS;
479: ds->ops->destroy = DSDestroy_NHEPTS;
480: ds->ops->matgetsize = DSMatGetSize_NHEPTS;
481: PetscFunctionReturn(PETSC_SUCCESS);
482: }