1: SUBROUTINE ZGESDD( JOBZ, M, N, A, LDA, S, U, LDU, VT, LDVT, WORK,
2: $ LWORK, RWORK, IWORK, INFO )
3: *
4: * -- LAPACK driver routine (version 3.2.2) --
5: * -- LAPACK is a software package provided by Univ. of Tennessee, --
6: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
7: * June 2010
8: * 8-15-00: Improve consistency of WS calculations (eca)
9: *
10: * .. Scalar Arguments ..
11: CHARACTER JOBZ
12: INTEGER INFO, LDA, LDU, LDVT, LWORK, M, N
13: * ..
14: * .. Array Arguments ..
15: INTEGER IWORK( * )
16: DOUBLE PRECISION RWORK( * ), S( * )
17: COMPLEX*16 A( LDA, * ), U( LDU, * ), VT( LDVT, * ),
18: $ WORK( * )
19: * ..
20: *
21: * Purpose
22: * =======
23: *
24: * ZGESDD computes the singular value decomposition (SVD) of a complex
25: * M-by-N matrix A, optionally computing the left and/or right singular
26: * vectors, by using divide-and-conquer method. The SVD is written
27: *
28: * A = U * SIGMA * conjugate-transpose(V)
29: *
30: * where SIGMA is an M-by-N matrix which is zero except for its
31: * min(m,n) diagonal elements, U is an M-by-M unitary matrix, and
32: * V is an N-by-N unitary matrix. The diagonal elements of SIGMA
33: * are the singular values of A; they are real and non-negative, and
34: * are returned in descending order. The first min(m,n) columns of
35: * U and V are the left and right singular vectors of A.
36: *
37: * Note that the routine returns VT = V**H, not V.
38: *
39: * The divide and conquer algorithm makes very mild assumptions about
40: * floating point arithmetic. It will work on machines with a guard
41: * digit in add/subtract, or on those binary machines without guard
42: * digits which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or
43: * Cray-2. It could conceivably fail on hexadecimal or decimal machines
44: * without guard digits, but we know of none.
45: *
46: * Arguments
47: * =========
48: *
49: * JOBZ (input) CHARACTER*1
50: * Specifies options for computing all or part of the matrix U:
51: * = 'A': all M columns of U and all N rows of V**H are
52: * returned in the arrays U and VT;
53: * = 'S': the first min(M,N) columns of U and the first
54: * min(M,N) rows of V**H are returned in the arrays U
55: * and VT;
56: * = 'O': If M >= N, the first N columns of U are overwritten
57: * in the array A and all rows of V**H are returned in
58: * the array VT;
59: * otherwise, all columns of U are returned in the
60: * array U and the first M rows of V**H are overwritten
61: * in the array A;
62: * = 'N': no columns of U or rows of V**H are computed.
63: *
64: * M (input) INTEGER
65: * The number of rows of the input matrix A. M >= 0.
66: *
67: * N (input) INTEGER
68: * The number of columns of the input matrix A. N >= 0.
69: *
70: * A (input/output) COMPLEX*16 array, dimension (LDA,N)
71: * On entry, the M-by-N matrix A.
72: * On exit,
73: * if JOBZ = 'O', A is overwritten with the first N columns
74: * of U (the left singular vectors, stored
75: * columnwise) if M >= N;
76: * A is overwritten with the first M rows
77: * of V**H (the right singular vectors, stored
78: * rowwise) otherwise.
79: * if JOBZ .ne. 'O', the contents of A are destroyed.
80: *
81: * LDA (input) INTEGER
82: * The leading dimension of the array A. LDA >= max(1,M).
83: *
84: * S (output) DOUBLE PRECISION array, dimension (min(M,N))
85: * The singular values of A, sorted so that S(i) >= S(i+1).
86: *
87: * U (output) COMPLEX*16 array, dimension (LDU,UCOL)
88: * UCOL = M if JOBZ = 'A' or JOBZ = 'O' and M < N;
89: * UCOL = min(M,N) if JOBZ = 'S'.
90: * If JOBZ = 'A' or JOBZ = 'O' and M < N, U contains the M-by-M
91: * unitary matrix U;
92: * if JOBZ = 'S', U contains the first min(M,N) columns of U
93: * (the left singular vectors, stored columnwise);
94: * if JOBZ = 'O' and M >= N, or JOBZ = 'N', U is not referenced.
95: *
96: * LDU (input) INTEGER
97: * The leading dimension of the array U. LDU >= 1; if
98: * JOBZ = 'S' or 'A' or JOBZ = 'O' and M < N, LDU >= M.
99: *
100: * VT (output) COMPLEX*16 array, dimension (LDVT,N)
101: * If JOBZ = 'A' or JOBZ = 'O' and M >= N, VT contains the
102: * N-by-N unitary matrix V**H;
103: * if JOBZ = 'S', VT contains the first min(M,N) rows of
104: * V**H (the right singular vectors, stored rowwise);
105: * if JOBZ = 'O' and M < N, or JOBZ = 'N', VT is not referenced.
106: *
107: * LDVT (input) INTEGER
108: * The leading dimension of the array VT. LDVT >= 1; if
109: * JOBZ = 'A' or JOBZ = 'O' and M >= N, LDVT >= N;
110: * if JOBZ = 'S', LDVT >= min(M,N).
111: *
112: * WORK (workspace/output) COMPLEX*16 array, dimension (MAX(1,LWORK))
113: * On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
114: *
115: * LWORK (input) INTEGER
116: * The dimension of the array WORK. LWORK >= 1.
117: * if JOBZ = 'N', LWORK >= 2*min(M,N)+max(M,N).
118: * if JOBZ = 'O',
119: * LWORK >= 2*min(M,N)*min(M,N)+2*min(M,N)+max(M,N).
120: * if JOBZ = 'S' or 'A',
121: * LWORK >= min(M,N)*min(M,N)+2*min(M,N)+max(M,N).
122: * For good performance, LWORK should generally be larger.
123: *
124: * If LWORK = -1, a workspace query is assumed. The optimal
125: * size for the WORK array is calculated and stored in WORK(1),
126: * and no other work except argument checking is performed.
127: *
128: * RWORK (workspace) DOUBLE PRECISION array, dimension (MAX(1,LRWORK))
129: * If JOBZ = 'N', LRWORK >= 5*min(M,N).
130: * Otherwise,
131: * LRWORK >= min(M,N)*max(5*min(M,N)+7,2*max(M,N)+2*min(M,N)+1)
132: *
133: * IWORK (workspace) INTEGER array, dimension (8*min(M,N))
134: *
135: * INFO (output) INTEGER
136: * = 0: successful exit.
137: * < 0: if INFO = -i, the i-th argument had an illegal value.
138: * > 0: The updating process of DBDSDC did not converge.
139: *
140: * Further Details
141: * ===============
142: *
143: * Based on contributions by
144: * Ming Gu and Huan Ren, Computer Science Division, University of
145: * California at Berkeley, USA
146: *
147: * =====================================================================
148: *
149: * .. Parameters ..
150: INTEGER LQUERV
151: PARAMETER ( LQUERV = -1 )
152: COMPLEX*16 CZERO, CONE
153: PARAMETER ( CZERO = ( 0.0D+0, 0.0D+0 ),
154: $ CONE = ( 1.0D+0, 0.0D+0 ) )
155: DOUBLE PRECISION ZERO, ONE
156: PARAMETER ( ZERO = 0.0D+0, ONE = 1.0D+0 )
157: * ..
158: * .. Local Scalars ..
159: LOGICAL WNTQA, WNTQAS, WNTQN, WNTQO, WNTQS
160: INTEGER BLK, CHUNK, I, IE, IERR, IL, IR, IRU, IRVT,
161: $ ISCL, ITAU, ITAUP, ITAUQ, IU, IVT, LDWKVT,
162: $ LDWRKL, LDWRKR, LDWRKU, MAXWRK, MINMN, MINWRK,
163: $ MNTHR1, MNTHR2, NRWORK, NWORK, WRKBL
164: DOUBLE PRECISION ANRM, BIGNUM, EPS, SMLNUM
165: * ..
166: * .. Local Arrays ..
167: INTEGER IDUM( 1 )
168: DOUBLE PRECISION DUM( 1 )
169: * ..
170: * .. External Subroutines ..
171: EXTERNAL DBDSDC, DLASCL, XERBLA, ZGEBRD, ZGELQF, ZGEMM,
172: $ ZGEQRF, ZLACP2, ZLACPY, ZLACRM, ZLARCM, ZLASCL,
173: $ ZLASET, ZUNGBR, ZUNGLQ, ZUNGQR, ZUNMBR
174: * ..
175: * .. External Functions ..
176: LOGICAL LSAME
177: INTEGER ILAENV
178: DOUBLE PRECISION DLAMCH, ZLANGE
179: EXTERNAL LSAME, ILAENV, DLAMCH, ZLANGE
180: * ..
181: * .. Intrinsic Functions ..
182: INTRINSIC INT, MAX, MIN, SQRT
183: * ..
184: * .. Executable Statements ..
185: *
186: * Test the input arguments
187: *
188: INFO = 0
189: MINMN = MIN( M, N )
190: MNTHR1 = INT( MINMN*17.0D0 / 9.0D0 )
191: MNTHR2 = INT( MINMN*5.0D0 / 3.0D0 )
192: WNTQA = LSAME( JOBZ, 'A' )
193: WNTQS = LSAME( JOBZ, 'S' )
194: WNTQAS = WNTQA .OR. WNTQS
195: WNTQO = LSAME( JOBZ, 'O' )
196: WNTQN = LSAME( JOBZ, 'N' )
197: MINWRK = 1
198: MAXWRK = 1
199: *
200: IF( .NOT.( WNTQA .OR. WNTQS .OR. WNTQO .OR. WNTQN ) ) THEN
201: INFO = -1
202: ELSE IF( M.LT.0 ) THEN
203: INFO = -2
204: ELSE IF( N.LT.0 ) THEN
205: INFO = -3
206: ELSE IF( LDA.LT.MAX( 1, M ) ) THEN
207: INFO = -5
208: ELSE IF( LDU.LT.1 .OR. ( WNTQAS .AND. LDU.LT.M ) .OR.
209: $ ( WNTQO .AND. M.LT.N .AND. LDU.LT.M ) ) THEN
210: INFO = -8
211: ELSE IF( LDVT.LT.1 .OR. ( WNTQA .AND. LDVT.LT.N ) .OR.
212: $ ( WNTQS .AND. LDVT.LT.MINMN ) .OR.
213: $ ( WNTQO .AND. M.GE.N .AND. LDVT.LT.N ) ) THEN
214: INFO = -10
215: END IF
216: *
217: * Compute workspace
218: * (Note: Comments in the code beginning "Workspace:" describe the
219: * minimal amount of workspace needed at that point in the code,
220: * as well as the preferred amount for good performance.
221: * CWorkspace refers to complex workspace, and RWorkspace to
222: * real workspace. NB refers to the optimal block size for the
223: * immediately following subroutine, as returned by ILAENV.)
224: *
225: IF( INFO.EQ.0 .AND. M.GT.0 .AND. N.GT.0 ) THEN
226: IF( M.GE.N ) THEN
227: *
228: * There is no complex work space needed for bidiagonal SVD
229: * The real work space needed for bidiagonal SVD is BDSPAC
230: * for computing singular values and singular vectors; BDSPAN
231: * for computing singular values only.
232: * BDSPAC = 5*N*N + 7*N
233: * BDSPAN = MAX(7*N+4, 3*N+2+SMLSIZ*(SMLSIZ+8))
234: *
235: IF( M.GE.MNTHR1 ) THEN
236: IF( WNTQN ) THEN
237: *
238: * Path 1 (M much larger than N, JOBZ='N')
239: *
240: MAXWRK = N + N*ILAENV( 1, 'ZGEQRF', ' ', M, N, -1,
241: $ -1 )
242: MAXWRK = MAX( MAXWRK, 2*N+2*N*
243: $ ILAENV( 1, 'ZGEBRD', ' ', N, N, -1, -1 ) )
244: MINWRK = 3*N
245: ELSE IF( WNTQO ) THEN
246: *
247: * Path 2 (M much larger than N, JOBZ='O')
248: *
249: WRKBL = N + N*ILAENV( 1, 'ZGEQRF', ' ', M, N, -1, -1 )
250: WRKBL = MAX( WRKBL, N+N*ILAENV( 1, 'ZUNGQR', ' ', M,
251: $ N, N, -1 ) )
252: WRKBL = MAX( WRKBL, 2*N+2*N*
253: $ ILAENV( 1, 'ZGEBRD', ' ', N, N, -1, -1 ) )
254: WRKBL = MAX( WRKBL, 2*N+N*
255: $ ILAENV( 1, 'ZUNMBR', 'QLN', N, N, N, -1 ) )
256: WRKBL = MAX( WRKBL, 2*N+N*
257: $ ILAENV( 1, 'ZUNMBR', 'PRC', N, N, N, -1 ) )
258: MAXWRK = M*N + N*N + WRKBL
259: MINWRK = 2*N*N + 3*N
260: ELSE IF( WNTQS ) THEN
261: *
262: * Path 3 (M much larger than N, JOBZ='S')
263: *
264: WRKBL = N + N*ILAENV( 1, 'ZGEQRF', ' ', M, N, -1, -1 )
265: WRKBL = MAX( WRKBL, N+N*ILAENV( 1, 'ZUNGQR', ' ', M,
266: $ N, N, -1 ) )
267: WRKBL = MAX( WRKBL, 2*N+2*N*
268: $ ILAENV( 1, 'ZGEBRD', ' ', N, N, -1, -1 ) )
269: WRKBL = MAX( WRKBL, 2*N+N*
270: $ ILAENV( 1, 'ZUNMBR', 'QLN', N, N, N, -1 ) )
271: WRKBL = MAX( WRKBL, 2*N+N*
272: $ ILAENV( 1, 'ZUNMBR', 'PRC', N, N, N, -1 ) )
273: MAXWRK = N*N + WRKBL
274: MINWRK = N*N + 3*N
275: ELSE IF( WNTQA ) THEN
276: *
277: * Path 4 (M much larger than N, JOBZ='A')
278: *
279: WRKBL = N + N*ILAENV( 1, 'ZGEQRF', ' ', M, N, -1, -1 )
280: WRKBL = MAX( WRKBL, N+M*ILAENV( 1, 'ZUNGQR', ' ', M,
281: $ M, N, -1 ) )
282: WRKBL = MAX( WRKBL, 2*N+2*N*
283: $ ILAENV( 1, 'ZGEBRD', ' ', N, N, -1, -1 ) )
284: WRKBL = MAX( WRKBL, 2*N+N*
285: $ ILAENV( 1, 'ZUNMBR', 'QLN', N, N, N, -1 ) )
286: WRKBL = MAX( WRKBL, 2*N+N*
287: $ ILAENV( 1, 'ZUNMBR', 'PRC', N, N, N, -1 ) )
288: MAXWRK = N*N + WRKBL
289: MINWRK = N*N + 2*N + M
290: END IF
291: ELSE IF( M.GE.MNTHR2 ) THEN
292: *
293: * Path 5 (M much larger than N, but not as much as MNTHR1)
294: *
295: MAXWRK = 2*N + ( M+N )*ILAENV( 1, 'ZGEBRD', ' ', M, N,
296: $ -1, -1 )
297: MINWRK = 2*N + M
298: IF( WNTQO ) THEN
299: MAXWRK = MAX( MAXWRK, 2*N+N*
300: $ ILAENV( 1, 'ZUNGBR', 'P', N, N, N, -1 ) )
301: MAXWRK = MAX( MAXWRK, 2*N+N*
302: $ ILAENV( 1, 'ZUNGBR', 'Q', M, N, N, -1 ) )
303: MAXWRK = MAXWRK + M*N
304: MINWRK = MINWRK + N*N
305: ELSE IF( WNTQS ) THEN
306: MAXWRK = MAX( MAXWRK, 2*N+N*
307: $ ILAENV( 1, 'ZUNGBR', 'P', N, N, N, -1 ) )
308: MAXWRK = MAX( MAXWRK, 2*N+N*
309: $ ILAENV( 1, 'ZUNGBR', 'Q', M, N, N, -1 ) )
310: ELSE IF( WNTQA ) THEN
311: MAXWRK = MAX( MAXWRK, 2*N+N*
312: $ ILAENV( 1, 'ZUNGBR', 'P', N, N, N, -1 ) )
313: MAXWRK = MAX( MAXWRK, 2*N+M*
314: $ ILAENV( 1, 'ZUNGBR', 'Q', M, M, N, -1 ) )
315: END IF
316: ELSE
317: *
318: * Path 6 (M at least N, but not much larger)
319: *
320: MAXWRK = 2*N + ( M+N )*ILAENV( 1, 'ZGEBRD', ' ', M, N,
321: $ -1, -1 )
322: MINWRK = 2*N + M
323: IF( WNTQO ) THEN
324: MAXWRK = MAX( MAXWRK, 2*N+N*
325: $ ILAENV( 1, 'ZUNMBR', 'PRC', N, N, N, -1 ) )
326: MAXWRK = MAX( MAXWRK, 2*N+N*
327: $ ILAENV( 1, 'ZUNMBR', 'QLN', M, N, N, -1 ) )
328: MAXWRK = MAXWRK + M*N
329: MINWRK = MINWRK + N*N
330: ELSE IF( WNTQS ) THEN
331: MAXWRK = MAX( MAXWRK, 2*N+N*
332: $ ILAENV( 1, 'ZUNMBR', 'PRC', N, N, N, -1 ) )
333: MAXWRK = MAX( MAXWRK, 2*N+N*
334: $ ILAENV( 1, 'ZUNMBR', 'QLN', M, N, N, -1 ) )
335: ELSE IF( WNTQA ) THEN
336: MAXWRK = MAX( MAXWRK, 2*N+N*
337: $ ILAENV( 1, 'ZUNGBR', 'PRC', N, N, N, -1 ) )
338: MAXWRK = MAX( MAXWRK, 2*N+M*
339: $ ILAENV( 1, 'ZUNGBR', 'QLN', M, M, N, -1 ) )
340: END IF
341: END IF
342: ELSE
343: *
344: * There is no complex work space needed for bidiagonal SVD
345: * The real work space needed for bidiagonal SVD is BDSPAC
346: * for computing singular values and singular vectors; BDSPAN
347: * for computing singular values only.
348: * BDSPAC = 5*M*M + 7*M
349: * BDSPAN = MAX(7*M+4, 3*M+2+SMLSIZ*(SMLSIZ+8))
350: *
351: IF( N.GE.MNTHR1 ) THEN
352: IF( WNTQN ) THEN
353: *
354: * Path 1t (N much larger than M, JOBZ='N')
355: *
356: MAXWRK = M + M*ILAENV( 1, 'ZGELQF', ' ', M, N, -1,
357: $ -1 )
358: MAXWRK = MAX( MAXWRK, 2*M+2*M*
359: $ ILAENV( 1, 'ZGEBRD', ' ', M, M, -1, -1 ) )
360: MINWRK = 3*M
361: ELSE IF( WNTQO ) THEN
362: *
363: * Path 2t (N much larger than M, JOBZ='O')
364: *
365: WRKBL = M + M*ILAENV( 1, 'ZGELQF', ' ', M, N, -1, -1 )
366: WRKBL = MAX( WRKBL, M+M*ILAENV( 1, 'ZUNGLQ', ' ', M,
367: $ N, M, -1 ) )
368: WRKBL = MAX( WRKBL, 2*M+2*M*
369: $ ILAENV( 1, 'ZGEBRD', ' ', M, M, -1, -1 ) )
370: WRKBL = MAX( WRKBL, 2*M+M*
371: $ ILAENV( 1, 'ZUNMBR', 'PRC', M, M, M, -1 ) )
372: WRKBL = MAX( WRKBL, 2*M+M*
373: $ ILAENV( 1, 'ZUNMBR', 'QLN', M, M, M, -1 ) )
374: MAXWRK = M*N + M*M + WRKBL
375: MINWRK = 2*M*M + 3*M
376: ELSE IF( WNTQS ) THEN
377: *
378: * Path 3t (N much larger than M, JOBZ='S')
379: *
380: WRKBL = M + M*ILAENV( 1, 'ZGELQF', ' ', M, N, -1, -1 )
381: WRKBL = MAX( WRKBL, M+M*ILAENV( 1, 'ZUNGLQ', ' ', M,
382: $ N, M, -1 ) )
383: WRKBL = MAX( WRKBL, 2*M+2*M*
384: $ ILAENV( 1, 'ZGEBRD', ' ', M, M, -1, -1 ) )
385: WRKBL = MAX( WRKBL, 2*M+M*
386: $ ILAENV( 1, 'ZUNMBR', 'PRC', M, M, M, -1 ) )
387: WRKBL = MAX( WRKBL, 2*M+M*
388: $ ILAENV( 1, 'ZUNMBR', 'QLN', M, M, M, -1 ) )
389: MAXWRK = M*M + WRKBL
390: MINWRK = M*M + 3*M
391: ELSE IF( WNTQA ) THEN
392: *
393: * Path 4t (N much larger than M, JOBZ='A')
394: *
395: WRKBL = M + M*ILAENV( 1, 'ZGELQF', ' ', M, N, -1, -1 )
396: WRKBL = MAX( WRKBL, M+N*ILAENV( 1, 'ZUNGLQ', ' ', N,
397: $ N, M, -1 ) )
398: WRKBL = MAX( WRKBL, 2*M+2*M*
399: $ ILAENV( 1, 'ZGEBRD', ' ', M, M, -1, -1 ) )
400: WRKBL = MAX( WRKBL, 2*M+M*
401: $ ILAENV( 1, 'ZUNMBR', 'PRC', M, M, M, -1 ) )
402: WRKBL = MAX( WRKBL, 2*M+M*
403: $ ILAENV( 1, 'ZUNMBR', 'QLN', M, M, M, -1 ) )
404: MAXWRK = M*M + WRKBL
405: MINWRK = M*M + 2*M + N
406: END IF
407: ELSE IF( N.GE.MNTHR2 ) THEN
408: *
409: * Path 5t (N much larger than M, but not as much as MNTHR1)
410: *
411: MAXWRK = 2*M + ( M+N )*ILAENV( 1, 'ZGEBRD', ' ', M, N,
412: $ -1, -1 )
413: MINWRK = 2*M + N
414: IF( WNTQO ) THEN
415: MAXWRK = MAX( MAXWRK, 2*M+M*
416: $ ILAENV( 1, 'ZUNGBR', 'P', M, N, M, -1 ) )
417: MAXWRK = MAX( MAXWRK, 2*M+M*
418: $ ILAENV( 1, 'ZUNGBR', 'Q', M, M, N, -1 ) )
419: MAXWRK = MAXWRK + M*N
420: MINWRK = MINWRK + M*M
421: ELSE IF( WNTQS ) THEN
422: MAXWRK = MAX( MAXWRK, 2*M+M*
423: $ ILAENV( 1, 'ZUNGBR', 'P', M, N, M, -1 ) )
424: MAXWRK = MAX( MAXWRK, 2*M+M*
425: $ ILAENV( 1, 'ZUNGBR', 'Q', M, M, N, -1 ) )
426: ELSE IF( WNTQA ) THEN
427: MAXWRK = MAX( MAXWRK, 2*M+N*
428: $ ILAENV( 1, 'ZUNGBR', 'P', N, N, M, -1 ) )
429: MAXWRK = MAX( MAXWRK, 2*M+M*
430: $ ILAENV( 1, 'ZUNGBR', 'Q', M, M, N, -1 ) )
431: END IF
432: ELSE
433: *
434: * Path 6t (N greater than M, but not much larger)
435: *
436: MAXWRK = 2*M + ( M+N )*ILAENV( 1, 'ZGEBRD', ' ', M, N,
437: $ -1, -1 )
438: MINWRK = 2*M + N
439: IF( WNTQO ) THEN
440: MAXWRK = MAX( MAXWRK, 2*M+M*
441: $ ILAENV( 1, 'ZUNMBR', 'PRC', M, N, M, -1 ) )
442: MAXWRK = MAX( MAXWRK, 2*M+M*
443: $ ILAENV( 1, 'ZUNMBR', 'QLN', M, M, N, -1 ) )
444: MAXWRK = MAXWRK + M*N
445: MINWRK = MINWRK + M*M
446: ELSE IF( WNTQS ) THEN
447: MAXWRK = MAX( MAXWRK, 2*M+M*
448: $ ILAENV( 1, 'ZUNGBR', 'PRC', M, N, M, -1 ) )
449: MAXWRK = MAX( MAXWRK, 2*M+M*
450: $ ILAENV( 1, 'ZUNGBR', 'QLN', M, M, N, -1 ) )
451: ELSE IF( WNTQA ) THEN
452: MAXWRK = MAX( MAXWRK, 2*M+N*
453: $ ILAENV( 1, 'ZUNGBR', 'PRC', N, N, M, -1 ) )
454: MAXWRK = MAX( MAXWRK, 2*M+M*
455: $ ILAENV( 1, 'ZUNGBR', 'QLN', M, M, N, -1 ) )
456: END IF
457: END IF
458: END IF
459: MAXWRK = MAX( MAXWRK, MINWRK )
460: END IF
461: IF( INFO.EQ.0 ) THEN
462: WORK( 1 ) = MAXWRK
463: IF( LWORK.LT.MINWRK .AND. LWORK.NE.LQUERV )
464: $ INFO = -13
465: END IF
466: *
467: * Quick returns
468: *
469: IF( INFO.NE.0 ) THEN
470: CALL XERBLA( 'ZGESDD', -INFO )
471: RETURN
472: END IF
473: IF( LWORK.EQ.LQUERV )
474: $ RETURN
475: IF( M.EQ.0 .OR. N.EQ.0 ) THEN
476: RETURN
477: END IF
478: *
479: * Get machine constants
480: *
481: EPS = DLAMCH( 'P' )
482: SMLNUM = SQRT( DLAMCH( 'S' ) ) / EPS
483: BIGNUM = ONE / SMLNUM
484: *
485: * Scale A if max element outside range [SMLNUM,BIGNUM]
486: *
487: ANRM = ZLANGE( 'M', M, N, A, LDA, DUM )
488: ISCL = 0
489: IF( ANRM.GT.ZERO .AND. ANRM.LT.SMLNUM ) THEN
490: ISCL = 1
491: CALL ZLASCL( 'G', 0, 0, ANRM, SMLNUM, M, N, A, LDA, IERR )
492: ELSE IF( ANRM.GT.BIGNUM ) THEN
493: ISCL = 1
494: CALL ZLASCL( 'G', 0, 0, ANRM, BIGNUM, M, N, A, LDA, IERR )
495: END IF
496: *
497: IF( M.GE.N ) THEN
498: *
499: * A has at least as many rows as columns. If A has sufficiently
500: * more rows than columns, first reduce using the QR
501: * decomposition (if sufficient workspace available)
502: *
503: IF( M.GE.MNTHR1 ) THEN
504: *
505: IF( WNTQN ) THEN
506: *
507: * Path 1 (M much larger than N, JOBZ='N')
508: * No singular vectors to be computed
509: *
510: ITAU = 1
511: NWORK = ITAU + N
512: *
513: * Compute A=Q*R
514: * (CWorkspace: need 2*N, prefer N+N*NB)
515: * (RWorkspace: need 0)
516: *
517: CALL ZGEQRF( M, N, A, LDA, WORK( ITAU ), WORK( NWORK ),
518: $ LWORK-NWORK+1, IERR )
519: *
520: * Zero out below R
521: *
522: CALL ZLASET( 'L', N-1, N-1, CZERO, CZERO, A( 2, 1 ),
523: $ LDA )
524: IE = 1
525: ITAUQ = 1
526: ITAUP = ITAUQ + N
527: NWORK = ITAUP + N
528: *
529: * Bidiagonalize R in A
530: * (CWorkspace: need 3*N, prefer 2*N+2*N*NB)
531: * (RWorkspace: need N)
532: *
533: CALL ZGEBRD( N, N, A, LDA, S, RWORK( IE ), WORK( ITAUQ ),
534: $ WORK( ITAUP ), WORK( NWORK ), LWORK-NWORK+1,
535: $ IERR )
536: NRWORK = IE + N
537: *
538: * Perform bidiagonal SVD, compute singular values only
539: * (CWorkspace: 0)
540: * (RWorkspace: need BDSPAN)
541: *
542: CALL DBDSDC( 'U', 'N', N, S, RWORK( IE ), DUM, 1, DUM, 1,
543: $ DUM, IDUM, RWORK( NRWORK ), IWORK, INFO )
544: *
545: ELSE IF( WNTQO ) THEN
546: *
547: * Path 2 (M much larger than N, JOBZ='O')
548: * N left singular vectors to be overwritten on A and
549: * N right singular vectors to be computed in VT
550: *
551: IU = 1
552: *
553: * WORK(IU) is N by N
554: *
555: LDWRKU = N
556: IR = IU + LDWRKU*N
557: IF( LWORK.GE.M*N+N*N+3*N ) THEN
558: *
559: * WORK(IR) is M by N
560: *
561: LDWRKR = M
562: ELSE
563: LDWRKR = ( LWORK-N*N-3*N ) / N
564: END IF
565: ITAU = IR + LDWRKR*N
566: NWORK = ITAU + N
567: *
568: * Compute A=Q*R
569: * (CWorkspace: need N*N+2*N, prefer M*N+N+N*NB)
570: * (RWorkspace: 0)
571: *
572: CALL ZGEQRF( M, N, A, LDA, WORK( ITAU ), WORK( NWORK ),
573: $ LWORK-NWORK+1, IERR )
574: *
575: * Copy R to WORK( IR ), zeroing out below it
576: *
577: CALL ZLACPY( 'U', N, N, A, LDA, WORK( IR ), LDWRKR )
578: CALL ZLASET( 'L', N-1, N-1, CZERO, CZERO, WORK( IR+1 ),
579: $ LDWRKR )
580: *
581: * Generate Q in A
582: * (CWorkspace: need 2*N, prefer N+N*NB)
583: * (RWorkspace: 0)
584: *
585: CALL ZUNGQR( M, N, N, A, LDA, WORK( ITAU ),
586: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
587: IE = 1
588: ITAUQ = ITAU
589: ITAUP = ITAUQ + N
590: NWORK = ITAUP + N
591: *
592: * Bidiagonalize R in WORK(IR)
593: * (CWorkspace: need N*N+3*N, prefer M*N+2*N+2*N*NB)
594: * (RWorkspace: need N)
595: *
596: CALL ZGEBRD( N, N, WORK( IR ), LDWRKR, S, RWORK( IE ),
597: $ WORK( ITAUQ ), WORK( ITAUP ), WORK( NWORK ),
598: $ LWORK-NWORK+1, IERR )
599: *
600: * Perform bidiagonal SVD, computing left singular vectors
601: * of R in WORK(IRU) and computing right singular vectors
602: * of R in WORK(IRVT)
603: * (CWorkspace: need 0)
604: * (RWorkspace: need BDSPAC)
605: *
606: IRU = IE + N
607: IRVT = IRU + N*N
608: NRWORK = IRVT + N*N
609: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
610: $ N, RWORK( IRVT ), N, DUM, IDUM,
611: $ RWORK( NRWORK ), IWORK, INFO )
612: *
613: * Copy real matrix RWORK(IRU) to complex matrix WORK(IU)
614: * Overwrite WORK(IU) by the left singular vectors of R
615: * (CWorkspace: need 2*N*N+3*N, prefer M*N+N*N+2*N+N*NB)
616: * (RWorkspace: 0)
617: *
618: CALL ZLACP2( 'F', N, N, RWORK( IRU ), N, WORK( IU ),
619: $ LDWRKU )
620: CALL ZUNMBR( 'Q', 'L', 'N', N, N, N, WORK( IR ), LDWRKR,
621: $ WORK( ITAUQ ), WORK( IU ), LDWRKU,
622: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
623: *
624: * Copy real matrix RWORK(IRVT) to complex matrix VT
625: * Overwrite VT by the right singular vectors of R
626: * (CWorkspace: need N*N+3*N, prefer M*N+2*N+N*NB)
627: * (RWorkspace: 0)
628: *
629: CALL ZLACP2( 'F', N, N, RWORK( IRVT ), N, VT, LDVT )
630: CALL ZUNMBR( 'P', 'R', 'C', N, N, N, WORK( IR ), LDWRKR,
631: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
632: $ LWORK-NWORK+1, IERR )
633: *
634: * Multiply Q in A by left singular vectors of R in
635: * WORK(IU), storing result in WORK(IR) and copying to A
636: * (CWorkspace: need 2*N*N, prefer N*N+M*N)
637: * (RWorkspace: 0)
638: *
639: DO 10 I = 1, M, LDWRKR
640: CHUNK = MIN( M-I+1, LDWRKR )
641: CALL ZGEMM( 'N', 'N', CHUNK, N, N, CONE, A( I, 1 ),
642: $ LDA, WORK( IU ), LDWRKU, CZERO,
643: $ WORK( IR ), LDWRKR )
644: CALL ZLACPY( 'F', CHUNK, N, WORK( IR ), LDWRKR,
645: $ A( I, 1 ), LDA )
646: 10 CONTINUE
647: *
648: ELSE IF( WNTQS ) THEN
649: *
650: * Path 3 (M much larger than N, JOBZ='S')
651: * N left singular vectors to be computed in U and
652: * N right singular vectors to be computed in VT
653: *
654: IR = 1
655: *
656: * WORK(IR) is N by N
657: *
658: LDWRKR = N
659: ITAU = IR + LDWRKR*N
660: NWORK = ITAU + N
661: *
662: * Compute A=Q*R
663: * (CWorkspace: need N*N+2*N, prefer N*N+N+N*NB)
664: * (RWorkspace: 0)
665: *
666: CALL ZGEQRF( M, N, A, LDA, WORK( ITAU ), WORK( NWORK ),
667: $ LWORK-NWORK+1, IERR )
668: *
669: * Copy R to WORK(IR), zeroing out below it
670: *
671: CALL ZLACPY( 'U', N, N, A, LDA, WORK( IR ), LDWRKR )
672: CALL ZLASET( 'L', N-1, N-1, CZERO, CZERO, WORK( IR+1 ),
673: $ LDWRKR )
674: *
675: * Generate Q in A
676: * (CWorkspace: need 2*N, prefer N+N*NB)
677: * (RWorkspace: 0)
678: *
679: CALL ZUNGQR( M, N, N, A, LDA, WORK( ITAU ),
680: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
681: IE = 1
682: ITAUQ = ITAU
683: ITAUP = ITAUQ + N
684: NWORK = ITAUP + N
685: *
686: * Bidiagonalize R in WORK(IR)
687: * (CWorkspace: need N*N+3*N, prefer N*N+2*N+2*N*NB)
688: * (RWorkspace: need N)
689: *
690: CALL ZGEBRD( N, N, WORK( IR ), LDWRKR, S, RWORK( IE ),
691: $ WORK( ITAUQ ), WORK( ITAUP ), WORK( NWORK ),
692: $ LWORK-NWORK+1, IERR )
693: *
694: * Perform bidiagonal SVD, computing left singular vectors
695: * of bidiagonal matrix in RWORK(IRU) and computing right
696: * singular vectors of bidiagonal matrix in RWORK(IRVT)
697: * (CWorkspace: need 0)
698: * (RWorkspace: need BDSPAC)
699: *
700: IRU = IE + N
701: IRVT = IRU + N*N
702: NRWORK = IRVT + N*N
703: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
704: $ N, RWORK( IRVT ), N, DUM, IDUM,
705: $ RWORK( NRWORK ), IWORK, INFO )
706: *
707: * Copy real matrix RWORK(IRU) to complex matrix U
708: * Overwrite U by left singular vectors of R
709: * (CWorkspace: need N*N+3*N, prefer N*N+2*N+N*NB)
710: * (RWorkspace: 0)
711: *
712: CALL ZLACP2( 'F', N, N, RWORK( IRU ), N, U, LDU )
713: CALL ZUNMBR( 'Q', 'L', 'N', N, N, N, WORK( IR ), LDWRKR,
714: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
715: $ LWORK-NWORK+1, IERR )
716: *
717: * Copy real matrix RWORK(IRVT) to complex matrix VT
718: * Overwrite VT by right singular vectors of R
719: * (CWorkspace: need N*N+3*N, prefer N*N+2*N+N*NB)
720: * (RWorkspace: 0)
721: *
722: CALL ZLACP2( 'F', N, N, RWORK( IRVT ), N, VT, LDVT )
723: CALL ZUNMBR( 'P', 'R', 'C', N, N, N, WORK( IR ), LDWRKR,
724: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
725: $ LWORK-NWORK+1, IERR )
726: *
727: * Multiply Q in A by left singular vectors of R in
728: * WORK(IR), storing result in U
729: * (CWorkspace: need N*N)
730: * (RWorkspace: 0)
731: *
732: CALL ZLACPY( 'F', N, N, U, LDU, WORK( IR ), LDWRKR )
733: CALL ZGEMM( 'N', 'N', M, N, N, CONE, A, LDA, WORK( IR ),
734: $ LDWRKR, CZERO, U, LDU )
735: *
736: ELSE IF( WNTQA ) THEN
737: *
738: * Path 4 (M much larger than N, JOBZ='A')
739: * M left singular vectors to be computed in U and
740: * N right singular vectors to be computed in VT
741: *
742: IU = 1
743: *
744: * WORK(IU) is N by N
745: *
746: LDWRKU = N
747: ITAU = IU + LDWRKU*N
748: NWORK = ITAU + N
749: *
750: * Compute A=Q*R, copying result to U
751: * (CWorkspace: need 2*N, prefer N+N*NB)
752: * (RWorkspace: 0)
753: *
754: CALL ZGEQRF( M, N, A, LDA, WORK( ITAU ), WORK( NWORK ),
755: $ LWORK-NWORK+1, IERR )
756: CALL ZLACPY( 'L', M, N, A, LDA, U, LDU )
757: *
758: * Generate Q in U
759: * (CWorkspace: need N+M, prefer N+M*NB)
760: * (RWorkspace: 0)
761: *
762: CALL ZUNGQR( M, M, N, U, LDU, WORK( ITAU ),
763: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
764: *
765: * Produce R in A, zeroing out below it
766: *
767: CALL ZLASET( 'L', N-1, N-1, CZERO, CZERO, A( 2, 1 ),
768: $ LDA )
769: IE = 1
770: ITAUQ = ITAU
771: ITAUP = ITAUQ + N
772: NWORK = ITAUP + N
773: *
774: * Bidiagonalize R in A
775: * (CWorkspace: need 3*N, prefer 2*N+2*N*NB)
776: * (RWorkspace: need N)
777: *
778: CALL ZGEBRD( N, N, A, LDA, S, RWORK( IE ), WORK( ITAUQ ),
779: $ WORK( ITAUP ), WORK( NWORK ), LWORK-NWORK+1,
780: $ IERR )
781: IRU = IE + N
782: IRVT = IRU + N*N
783: NRWORK = IRVT + N*N
784: *
785: * Perform bidiagonal SVD, computing left singular vectors
786: * of bidiagonal matrix in RWORK(IRU) and computing right
787: * singular vectors of bidiagonal matrix in RWORK(IRVT)
788: * (CWorkspace: need 0)
789: * (RWorkspace: need BDSPAC)
790: *
791: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
792: $ N, RWORK( IRVT ), N, DUM, IDUM,
793: $ RWORK( NRWORK ), IWORK, INFO )
794: *
795: * Copy real matrix RWORK(IRU) to complex matrix WORK(IU)
796: * Overwrite WORK(IU) by left singular vectors of R
797: * (CWorkspace: need N*N+3*N, prefer N*N+2*N+N*NB)
798: * (RWorkspace: 0)
799: *
800: CALL ZLACP2( 'F', N, N, RWORK( IRU ), N, WORK( IU ),
801: $ LDWRKU )
802: CALL ZUNMBR( 'Q', 'L', 'N', N, N, N, A, LDA,
803: $ WORK( ITAUQ ), WORK( IU ), LDWRKU,
804: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
805: *
806: * Copy real matrix RWORK(IRVT) to complex matrix VT
807: * Overwrite VT by right singular vectors of R
808: * (CWorkspace: need 3*N, prefer 2*N+N*NB)
809: * (RWorkspace: 0)
810: *
811: CALL ZLACP2( 'F', N, N, RWORK( IRVT ), N, VT, LDVT )
812: CALL ZUNMBR( 'P', 'R', 'C', N, N, N, A, LDA,
813: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
814: $ LWORK-NWORK+1, IERR )
815: *
816: * Multiply Q in U by left singular vectors of R in
817: * WORK(IU), storing result in A
818: * (CWorkspace: need N*N)
819: * (RWorkspace: 0)
820: *
821: CALL ZGEMM( 'N', 'N', M, N, N, CONE, U, LDU, WORK( IU ),
822: $ LDWRKU, CZERO, A, LDA )
823: *
824: * Copy left singular vectors of A from A to U
825: *
826: CALL ZLACPY( 'F', M, N, A, LDA, U, LDU )
827: *
828: END IF
829: *
830: ELSE IF( M.GE.MNTHR2 ) THEN
831: *
832: * MNTHR2 <= M < MNTHR1
833: *
834: * Path 5 (M much larger than N, but not as much as MNTHR1)
835: * Reduce to bidiagonal form without QR decomposition, use
836: * ZUNGBR and matrix multiplication to compute singular vectors
837: *
838: IE = 1
839: NRWORK = IE + N
840: ITAUQ = 1
841: ITAUP = ITAUQ + N
842: NWORK = ITAUP + N
843: *
844: * Bidiagonalize A
845: * (CWorkspace: need 2*N+M, prefer 2*N+(M+N)*NB)
846: * (RWorkspace: need N)
847: *
848: CALL ZGEBRD( M, N, A, LDA, S, RWORK( IE ), WORK( ITAUQ ),
849: $ WORK( ITAUP ), WORK( NWORK ), LWORK-NWORK+1,
850: $ IERR )
851: IF( WNTQN ) THEN
852: *
853: * Compute singular values only
854: * (Cworkspace: 0)
855: * (Rworkspace: need BDSPAN)
856: *
857: CALL DBDSDC( 'U', 'N', N, S, RWORK( IE ), DUM, 1, DUM, 1,
858: $ DUM, IDUM, RWORK( NRWORK ), IWORK, INFO )
859: ELSE IF( WNTQO ) THEN
860: IU = NWORK
861: IRU = NRWORK
862: IRVT = IRU + N*N
863: NRWORK = IRVT + N*N
864: *
865: * Copy A to VT, generate P**H
866: * (Cworkspace: need 2*N, prefer N+N*NB)
867: * (Rworkspace: 0)
868: *
869: CALL ZLACPY( 'U', N, N, A, LDA, VT, LDVT )
870: CALL ZUNGBR( 'P', N, N, N, VT, LDVT, WORK( ITAUP ),
871: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
872: *
873: * Generate Q in A
874: * (CWorkspace: need 2*N, prefer N+N*NB)
875: * (RWorkspace: 0)
876: *
877: CALL ZUNGBR( 'Q', M, N, N, A, LDA, WORK( ITAUQ ),
878: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
879: *
880: IF( LWORK.GE.M*N+3*N ) THEN
881: *
882: * WORK( IU ) is M by N
883: *
884: LDWRKU = M
885: ELSE
886: *
887: * WORK(IU) is LDWRKU by N
888: *
889: LDWRKU = ( LWORK-3*N ) / N
890: END IF
891: NWORK = IU + LDWRKU*N
892: *
893: * Perform bidiagonal SVD, computing left singular vectors
894: * of bidiagonal matrix in RWORK(IRU) and computing right
895: * singular vectors of bidiagonal matrix in RWORK(IRVT)
896: * (CWorkspace: need 0)
897: * (RWorkspace: need BDSPAC)
898: *
899: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
900: $ N, RWORK( IRVT ), N, DUM, IDUM,
901: $ RWORK( NRWORK ), IWORK, INFO )
902: *
903: * Multiply real matrix RWORK(IRVT) by P**H in VT,
904: * storing the result in WORK(IU), copying to VT
905: * (Cworkspace: need 0)
906: * (Rworkspace: need 3*N*N)
907: *
908: CALL ZLARCM( N, N, RWORK( IRVT ), N, VT, LDVT,
909: $ WORK( IU ), LDWRKU, RWORK( NRWORK ) )
910: CALL ZLACPY( 'F', N, N, WORK( IU ), LDWRKU, VT, LDVT )
911: *
912: * Multiply Q in A by real matrix RWORK(IRU), storing the
913: * result in WORK(IU), copying to A
914: * (CWorkspace: need N*N, prefer M*N)
915: * (Rworkspace: need 3*N*N, prefer N*N+2*M*N)
916: *
917: NRWORK = IRVT
918: DO 20 I = 1, M, LDWRKU
919: CHUNK = MIN( M-I+1, LDWRKU )
920: CALL ZLACRM( CHUNK, N, A( I, 1 ), LDA, RWORK( IRU ),
921: $ N, WORK( IU ), LDWRKU, RWORK( NRWORK ) )
922: CALL ZLACPY( 'F', CHUNK, N, WORK( IU ), LDWRKU,
923: $ A( I, 1 ), LDA )
924: 20 CONTINUE
925: *
926: ELSE IF( WNTQS ) THEN
927: *
928: * Copy A to VT, generate P**H
929: * (Cworkspace: need 2*N, prefer N+N*NB)
930: * (Rworkspace: 0)
931: *
932: CALL ZLACPY( 'U', N, N, A, LDA, VT, LDVT )
933: CALL ZUNGBR( 'P', N, N, N, VT, LDVT, WORK( ITAUP ),
934: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
935: *
936: * Copy A to U, generate Q
937: * (Cworkspace: need 2*N, prefer N+N*NB)
938: * (Rworkspace: 0)
939: *
940: CALL ZLACPY( 'L', M, N, A, LDA, U, LDU )
941: CALL ZUNGBR( 'Q', M, N, N, U, LDU, WORK( ITAUQ ),
942: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
943: *
944: * Perform bidiagonal SVD, computing left singular vectors
945: * of bidiagonal matrix in RWORK(IRU) and computing right
946: * singular vectors of bidiagonal matrix in RWORK(IRVT)
947: * (CWorkspace: need 0)
948: * (RWorkspace: need BDSPAC)
949: *
950: IRU = NRWORK
951: IRVT = IRU + N*N
952: NRWORK = IRVT + N*N
953: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
954: $ N, RWORK( IRVT ), N, DUM, IDUM,
955: $ RWORK( NRWORK ), IWORK, INFO )
956: *
957: * Multiply real matrix RWORK(IRVT) by P**H in VT,
958: * storing the result in A, copying to VT
959: * (Cworkspace: need 0)
960: * (Rworkspace: need 3*N*N)
961: *
962: CALL ZLARCM( N, N, RWORK( IRVT ), N, VT, LDVT, A, LDA,
963: $ RWORK( NRWORK ) )
964: CALL ZLACPY( 'F', N, N, A, LDA, VT, LDVT )
965: *
966: * Multiply Q in U by real matrix RWORK(IRU), storing the
967: * result in A, copying to U
968: * (CWorkspace: need 0)
969: * (Rworkspace: need N*N+2*M*N)
970: *
971: NRWORK = IRVT
972: CALL ZLACRM( M, N, U, LDU, RWORK( IRU ), N, A, LDA,
973: $ RWORK( NRWORK ) )
974: CALL ZLACPY( 'F', M, N, A, LDA, U, LDU )
975: ELSE
976: *
977: * Copy A to VT, generate P**H
978: * (Cworkspace: need 2*N, prefer N+N*NB)
979: * (Rworkspace: 0)
980: *
981: CALL ZLACPY( 'U', N, N, A, LDA, VT, LDVT )
982: CALL ZUNGBR( 'P', N, N, N, VT, LDVT, WORK( ITAUP ),
983: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
984: *
985: * Copy A to U, generate Q
986: * (Cworkspace: need 2*N, prefer N+N*NB)
987: * (Rworkspace: 0)
988: *
989: CALL ZLACPY( 'L', M, N, A, LDA, U, LDU )
990: CALL ZUNGBR( 'Q', M, M, N, U, LDU, WORK( ITAUQ ),
991: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
992: *
993: * Perform bidiagonal SVD, computing left singular vectors
994: * of bidiagonal matrix in RWORK(IRU) and computing right
995: * singular vectors of bidiagonal matrix in RWORK(IRVT)
996: * (CWorkspace: need 0)
997: * (RWorkspace: need BDSPAC)
998: *
999: IRU = NRWORK
1000: IRVT = IRU + N*N
1001: NRWORK = IRVT + N*N
1002: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
1003: $ N, RWORK( IRVT ), N, DUM, IDUM,
1004: $ RWORK( NRWORK ), IWORK, INFO )
1005: *
1006: * Multiply real matrix RWORK(IRVT) by P**H in VT,
1007: * storing the result in A, copying to VT
1008: * (Cworkspace: need 0)
1009: * (Rworkspace: need 3*N*N)
1010: *
1011: CALL ZLARCM( N, N, RWORK( IRVT ), N, VT, LDVT, A, LDA,
1012: $ RWORK( NRWORK ) )
1013: CALL ZLACPY( 'F', N, N, A, LDA, VT, LDVT )
1014: *
1015: * Multiply Q in U by real matrix RWORK(IRU), storing the
1016: * result in A, copying to U
1017: * (CWorkspace: 0)
1018: * (Rworkspace: need 3*N*N)
1019: *
1020: NRWORK = IRVT
1021: CALL ZLACRM( M, N, U, LDU, RWORK( IRU ), N, A, LDA,
1022: $ RWORK( NRWORK ) )
1023: CALL ZLACPY( 'F', M, N, A, LDA, U, LDU )
1024: END IF
1025: *
1026: ELSE
1027: *
1028: * M .LT. MNTHR2
1029: *
1030: * Path 6 (M at least N, but not much larger)
1031: * Reduce to bidiagonal form without QR decomposition
1032: * Use ZUNMBR to compute singular vectors
1033: *
1034: IE = 1
1035: NRWORK = IE + N
1036: ITAUQ = 1
1037: ITAUP = ITAUQ + N
1038: NWORK = ITAUP + N
1039: *
1040: * Bidiagonalize A
1041: * (CWorkspace: need 2*N+M, prefer 2*N+(M+N)*NB)
1042: * (RWorkspace: need N)
1043: *
1044: CALL ZGEBRD( M, N, A, LDA, S, RWORK( IE ), WORK( ITAUQ ),
1045: $ WORK( ITAUP ), WORK( NWORK ), LWORK-NWORK+1,
1046: $ IERR )
1047: IF( WNTQN ) THEN
1048: *
1049: * Compute singular values only
1050: * (Cworkspace: 0)
1051: * (Rworkspace: need BDSPAN)
1052: *
1053: CALL DBDSDC( 'U', 'N', N, S, RWORK( IE ), DUM, 1, DUM, 1,
1054: $ DUM, IDUM, RWORK( NRWORK ), IWORK, INFO )
1055: ELSE IF( WNTQO ) THEN
1056: IU = NWORK
1057: IRU = NRWORK
1058: IRVT = IRU + N*N
1059: NRWORK = IRVT + N*N
1060: IF( LWORK.GE.M*N+3*N ) THEN
1061: *
1062: * WORK( IU ) is M by N
1063: *
1064: LDWRKU = M
1065: ELSE
1066: *
1067: * WORK( IU ) is LDWRKU by N
1068: *
1069: LDWRKU = ( LWORK-3*N ) / N
1070: END IF
1071: NWORK = IU + LDWRKU*N
1072: *
1073: * Perform bidiagonal SVD, computing left singular vectors
1074: * of bidiagonal matrix in RWORK(IRU) and computing right
1075: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1076: * (CWorkspace: need 0)
1077: * (RWorkspace: need BDSPAC)
1078: *
1079: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
1080: $ N, RWORK( IRVT ), N, DUM, IDUM,
1081: $ RWORK( NRWORK ), IWORK, INFO )
1082: *
1083: * Copy real matrix RWORK(IRVT) to complex matrix VT
1084: * Overwrite VT by right singular vectors of A
1085: * (Cworkspace: need 2*N, prefer N+N*NB)
1086: * (Rworkspace: need 0)
1087: *
1088: CALL ZLACP2( 'F', N, N, RWORK( IRVT ), N, VT, LDVT )
1089: CALL ZUNMBR( 'P', 'R', 'C', N, N, N, A, LDA,
1090: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
1091: $ LWORK-NWORK+1, IERR )
1092: *
1093: IF( LWORK.GE.M*N+3*N ) THEN
1094: *
1095: * Copy real matrix RWORK(IRU) to complex matrix WORK(IU)
1096: * Overwrite WORK(IU) by left singular vectors of A, copying
1097: * to A
1098: * (Cworkspace: need M*N+2*N, prefer M*N+N+N*NB)
1099: * (Rworkspace: need 0)
1100: *
1101: CALL ZLASET( 'F', M, N, CZERO, CZERO, WORK( IU ),
1102: $ LDWRKU )
1103: CALL ZLACP2( 'F', N, N, RWORK( IRU ), N, WORK( IU ),
1104: $ LDWRKU )
1105: CALL ZUNMBR( 'Q', 'L', 'N', M, N, N, A, LDA,
1106: $ WORK( ITAUQ ), WORK( IU ), LDWRKU,
1107: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1108: CALL ZLACPY( 'F', M, N, WORK( IU ), LDWRKU, A, LDA )
1109: ELSE
1110: *
1111: * Generate Q in A
1112: * (Cworkspace: need 2*N, prefer N+N*NB)
1113: * (Rworkspace: need 0)
1114: *
1115: CALL ZUNGBR( 'Q', M, N, N, A, LDA, WORK( ITAUQ ),
1116: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1117: *
1118: * Multiply Q in A by real matrix RWORK(IRU), storing the
1119: * result in WORK(IU), copying to A
1120: * (CWorkspace: need N*N, prefer M*N)
1121: * (Rworkspace: need 3*N*N, prefer N*N+2*M*N)
1122: *
1123: NRWORK = IRVT
1124: DO 30 I = 1, M, LDWRKU
1125: CHUNK = MIN( M-I+1, LDWRKU )
1126: CALL ZLACRM( CHUNK, N, A( I, 1 ), LDA,
1127: $ RWORK( IRU ), N, WORK( IU ), LDWRKU,
1128: $ RWORK( NRWORK ) )
1129: CALL ZLACPY( 'F', CHUNK, N, WORK( IU ), LDWRKU,
1130: $ A( I, 1 ), LDA )
1131: 30 CONTINUE
1132: END IF
1133: *
1134: ELSE IF( WNTQS ) THEN
1135: *
1136: * Perform bidiagonal SVD, computing left singular vectors
1137: * of bidiagonal matrix in RWORK(IRU) and computing right
1138: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1139: * (CWorkspace: need 0)
1140: * (RWorkspace: need BDSPAC)
1141: *
1142: IRU = NRWORK
1143: IRVT = IRU + N*N
1144: NRWORK = IRVT + N*N
1145: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
1146: $ N, RWORK( IRVT ), N, DUM, IDUM,
1147: $ RWORK( NRWORK ), IWORK, INFO )
1148: *
1149: * Copy real matrix RWORK(IRU) to complex matrix U
1150: * Overwrite U by left singular vectors of A
1151: * (CWorkspace: need 3*N, prefer 2*N+N*NB)
1152: * (RWorkspace: 0)
1153: *
1154: CALL ZLASET( 'F', M, N, CZERO, CZERO, U, LDU )
1155: CALL ZLACP2( 'F', N, N, RWORK( IRU ), N, U, LDU )
1156: CALL ZUNMBR( 'Q', 'L', 'N', M, N, N, A, LDA,
1157: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
1158: $ LWORK-NWORK+1, IERR )
1159: *
1160: * Copy real matrix RWORK(IRVT) to complex matrix VT
1161: * Overwrite VT by right singular vectors of A
1162: * (CWorkspace: need 3*N, prefer 2*N+N*NB)
1163: * (RWorkspace: 0)
1164: *
1165: CALL ZLACP2( 'F', N, N, RWORK( IRVT ), N, VT, LDVT )
1166: CALL ZUNMBR( 'P', 'R', 'C', N, N, N, A, LDA,
1167: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
1168: $ LWORK-NWORK+1, IERR )
1169: ELSE
1170: *
1171: * Perform bidiagonal SVD, computing left singular vectors
1172: * of bidiagonal matrix in RWORK(IRU) and computing right
1173: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1174: * (CWorkspace: need 0)
1175: * (RWorkspace: need BDSPAC)
1176: *
1177: IRU = NRWORK
1178: IRVT = IRU + N*N
1179: NRWORK = IRVT + N*N
1180: CALL DBDSDC( 'U', 'I', N, S, RWORK( IE ), RWORK( IRU ),
1181: $ N, RWORK( IRVT ), N, DUM, IDUM,
1182: $ RWORK( NRWORK ), IWORK, INFO )
1183: *
1184: * Set the right corner of U to identity matrix
1185: *
1186: CALL ZLASET( 'F', M, M, CZERO, CZERO, U, LDU )
1187: IF( M.GT.N ) THEN
1188: CALL ZLASET( 'F', M-N, M-N, CZERO, CONE,
1189: $ U( N+1, N+1 ), LDU )
1190: END IF
1191: *
1192: * Copy real matrix RWORK(IRU) to complex matrix U
1193: * Overwrite U by left singular vectors of A
1194: * (CWorkspace: need 2*N+M, prefer 2*N+M*NB)
1195: * (RWorkspace: 0)
1196: *
1197: CALL ZLACP2( 'F', N, N, RWORK( IRU ), N, U, LDU )
1198: CALL ZUNMBR( 'Q', 'L', 'N', M, M, N, A, LDA,
1199: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
1200: $ LWORK-NWORK+1, IERR )
1201: *
1202: * Copy real matrix RWORK(IRVT) to complex matrix VT
1203: * Overwrite VT by right singular vectors of A
1204: * (CWorkspace: need 3*N, prefer 2*N+N*NB)
1205: * (RWorkspace: 0)
1206: *
1207: CALL ZLACP2( 'F', N, N, RWORK( IRVT ), N, VT, LDVT )
1208: CALL ZUNMBR( 'P', 'R', 'C', N, N, N, A, LDA,
1209: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
1210: $ LWORK-NWORK+1, IERR )
1211: END IF
1212: *
1213: END IF
1214: *
1215: ELSE
1216: *
1217: * A has more columns than rows. If A has sufficiently more
1218: * columns than rows, first reduce using the LQ decomposition (if
1219: * sufficient workspace available)
1220: *
1221: IF( N.GE.MNTHR1 ) THEN
1222: *
1223: IF( WNTQN ) THEN
1224: *
1225: * Path 1t (N much larger than M, JOBZ='N')
1226: * No singular vectors to be computed
1227: *
1228: ITAU = 1
1229: NWORK = ITAU + M
1230: *
1231: * Compute A=L*Q
1232: * (CWorkspace: need 2*M, prefer M+M*NB)
1233: * (RWorkspace: 0)
1234: *
1235: CALL ZGELQF( M, N, A, LDA, WORK( ITAU ), WORK( NWORK ),
1236: $ LWORK-NWORK+1, IERR )
1237: *
1238: * Zero out above L
1239: *
1240: CALL ZLASET( 'U', M-1, M-1, CZERO, CZERO, A( 1, 2 ),
1241: $ LDA )
1242: IE = 1
1243: ITAUQ = 1
1244: ITAUP = ITAUQ + M
1245: NWORK = ITAUP + M
1246: *
1247: * Bidiagonalize L in A
1248: * (CWorkspace: need 3*M, prefer 2*M+2*M*NB)
1249: * (RWorkspace: need M)
1250: *
1251: CALL ZGEBRD( M, M, A, LDA, S, RWORK( IE ), WORK( ITAUQ ),
1252: $ WORK( ITAUP ), WORK( NWORK ), LWORK-NWORK+1,
1253: $ IERR )
1254: NRWORK = IE + M
1255: *
1256: * Perform bidiagonal SVD, compute singular values only
1257: * (CWorkspace: 0)
1258: * (RWorkspace: need BDSPAN)
1259: *
1260: CALL DBDSDC( 'U', 'N', M, S, RWORK( IE ), DUM, 1, DUM, 1,
1261: $ DUM, IDUM, RWORK( NRWORK ), IWORK, INFO )
1262: *
1263: ELSE IF( WNTQO ) THEN
1264: *
1265: * Path 2t (N much larger than M, JOBZ='O')
1266: * M right singular vectors to be overwritten on A and
1267: * M left singular vectors to be computed in U
1268: *
1269: IVT = 1
1270: LDWKVT = M
1271: *
1272: * WORK(IVT) is M by M
1273: *
1274: IL = IVT + LDWKVT*M
1275: IF( LWORK.GE.M*N+M*M+3*M ) THEN
1276: *
1277: * WORK(IL) M by N
1278: *
1279: LDWRKL = M
1280: CHUNK = N
1281: ELSE
1282: *
1283: * WORK(IL) is M by CHUNK
1284: *
1285: LDWRKL = M
1286: CHUNK = ( LWORK-M*M-3*M ) / M
1287: END IF
1288: ITAU = IL + LDWRKL*CHUNK
1289: NWORK = ITAU + M
1290: *
1291: * Compute A=L*Q
1292: * (CWorkspace: need 2*M, prefer M+M*NB)
1293: * (RWorkspace: 0)
1294: *
1295: CALL ZGELQF( M, N, A, LDA, WORK( ITAU ), WORK( NWORK ),
1296: $ LWORK-NWORK+1, IERR )
1297: *
1298: * Copy L to WORK(IL), zeroing about above it
1299: *
1300: CALL ZLACPY( 'L', M, M, A, LDA, WORK( IL ), LDWRKL )
1301: CALL ZLASET( 'U', M-1, M-1, CZERO, CZERO,
1302: $ WORK( IL+LDWRKL ), LDWRKL )
1303: *
1304: * Generate Q in A
1305: * (CWorkspace: need M*M+2*M, prefer M*M+M+M*NB)
1306: * (RWorkspace: 0)
1307: *
1308: CALL ZUNGLQ( M, N, M, A, LDA, WORK( ITAU ),
1309: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1310: IE = 1
1311: ITAUQ = ITAU
1312: ITAUP = ITAUQ + M
1313: NWORK = ITAUP + M
1314: *
1315: * Bidiagonalize L in WORK(IL)
1316: * (CWorkspace: need M*M+3*M, prefer M*M+2*M+2*M*NB)
1317: * (RWorkspace: need M)
1318: *
1319: CALL ZGEBRD( M, M, WORK( IL ), LDWRKL, S, RWORK( IE ),
1320: $ WORK( ITAUQ ), WORK( ITAUP ), WORK( NWORK ),
1321: $ LWORK-NWORK+1, IERR )
1322: *
1323: * Perform bidiagonal SVD, computing left singular vectors
1324: * of bidiagonal matrix in RWORK(IRU) and computing right
1325: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1326: * (CWorkspace: need 0)
1327: * (RWorkspace: need BDSPAC)
1328: *
1329: IRU = IE + M
1330: IRVT = IRU + M*M
1331: NRWORK = IRVT + M*M
1332: CALL DBDSDC( 'U', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1333: $ M, RWORK( IRVT ), M, DUM, IDUM,
1334: $ RWORK( NRWORK ), IWORK, INFO )
1335: *
1336: * Copy real matrix RWORK(IRU) to complex matrix WORK(IU)
1337: * Overwrite WORK(IU) by the left singular vectors of L
1338: * (CWorkspace: need N*N+3*N, prefer M*N+2*N+N*NB)
1339: * (RWorkspace: 0)
1340: *
1341: CALL ZLACP2( 'F', M, M, RWORK( IRU ), M, U, LDU )
1342: CALL ZUNMBR( 'Q', 'L', 'N', M, M, M, WORK( IL ), LDWRKL,
1343: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
1344: $ LWORK-NWORK+1, IERR )
1345: *
1346: * Copy real matrix RWORK(IRVT) to complex matrix WORK(IVT)
1347: * Overwrite WORK(IVT) by the right singular vectors of L
1348: * (CWorkspace: need N*N+3*N, prefer M*N+2*N+N*NB)
1349: * (RWorkspace: 0)
1350: *
1351: CALL ZLACP2( 'F', M, M, RWORK( IRVT ), M, WORK( IVT ),
1352: $ LDWKVT )
1353: CALL ZUNMBR( 'P', 'R', 'C', M, M, M, WORK( IL ), LDWRKL,
1354: $ WORK( ITAUP ), WORK( IVT ), LDWKVT,
1355: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1356: *
1357: * Multiply right singular vectors of L in WORK(IL) by Q
1358: * in A, storing result in WORK(IL) and copying to A
1359: * (CWorkspace: need 2*M*M, prefer M*M+M*N))
1360: * (RWorkspace: 0)
1361: *
1362: DO 40 I = 1, N, CHUNK
1363: BLK = MIN( N-I+1, CHUNK )
1364: CALL ZGEMM( 'N', 'N', M, BLK, M, CONE, WORK( IVT ), M,
1365: $ A( 1, I ), LDA, CZERO, WORK( IL ),
1366: $ LDWRKL )
1367: CALL ZLACPY( 'F', M, BLK, WORK( IL ), LDWRKL,
1368: $ A( 1, I ), LDA )
1369: 40 CONTINUE
1370: *
1371: ELSE IF( WNTQS ) THEN
1372: *
1373: * Path 3t (N much larger than M, JOBZ='S')
1374: * M right singular vectors to be computed in VT and
1375: * M left singular vectors to be computed in U
1376: *
1377: IL = 1
1378: *
1379: * WORK(IL) is M by M
1380: *
1381: LDWRKL = M
1382: ITAU = IL + LDWRKL*M
1383: NWORK = ITAU + M
1384: *
1385: * Compute A=L*Q
1386: * (CWorkspace: need 2*M, prefer M+M*NB)
1387: * (RWorkspace: 0)
1388: *
1389: CALL ZGELQF( M, N, A, LDA, WORK( ITAU ), WORK( NWORK ),
1390: $ LWORK-NWORK+1, IERR )
1391: *
1392: * Copy L to WORK(IL), zeroing out above it
1393: *
1394: CALL ZLACPY( 'L', M, M, A, LDA, WORK( IL ), LDWRKL )
1395: CALL ZLASET( 'U', M-1, M-1, CZERO, CZERO,
1396: $ WORK( IL+LDWRKL ), LDWRKL )
1397: *
1398: * Generate Q in A
1399: * (CWorkspace: need M*M+2*M, prefer M*M+M+M*NB)
1400: * (RWorkspace: 0)
1401: *
1402: CALL ZUNGLQ( M, N, M, A, LDA, WORK( ITAU ),
1403: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1404: IE = 1
1405: ITAUQ = ITAU
1406: ITAUP = ITAUQ + M
1407: NWORK = ITAUP + M
1408: *
1409: * Bidiagonalize L in WORK(IL)
1410: * (CWorkspace: need M*M+3*M, prefer M*M+2*M+2*M*NB)
1411: * (RWorkspace: need M)
1412: *
1413: CALL ZGEBRD( M, M, WORK( IL ), LDWRKL, S, RWORK( IE ),
1414: $ WORK( ITAUQ ), WORK( ITAUP ), WORK( NWORK ),
1415: $ LWORK-NWORK+1, IERR )
1416: *
1417: * Perform bidiagonal SVD, computing left singular vectors
1418: * of bidiagonal matrix in RWORK(IRU) and computing right
1419: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1420: * (CWorkspace: need 0)
1421: * (RWorkspace: need BDSPAC)
1422: *
1423: IRU = IE + M
1424: IRVT = IRU + M*M
1425: NRWORK = IRVT + M*M
1426: CALL DBDSDC( 'U', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1427: $ M, RWORK( IRVT ), M, DUM, IDUM,
1428: $ RWORK( NRWORK ), IWORK, INFO )
1429: *
1430: * Copy real matrix RWORK(IRU) to complex matrix U
1431: * Overwrite U by left singular vectors of L
1432: * (CWorkspace: need M*M+3*M, prefer M*M+2*M+M*NB)
1433: * (RWorkspace: 0)
1434: *
1435: CALL ZLACP2( 'F', M, M, RWORK( IRU ), M, U, LDU )
1436: CALL ZUNMBR( 'Q', 'L', 'N', M, M, M, WORK( IL ), LDWRKL,
1437: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
1438: $ LWORK-NWORK+1, IERR )
1439: *
1440: * Copy real matrix RWORK(IRVT) to complex matrix VT
1441: * Overwrite VT by left singular vectors of L
1442: * (CWorkspace: need M*M+3*M, prefer M*M+2*M+M*NB)
1443: * (RWorkspace: 0)
1444: *
1445: CALL ZLACP2( 'F', M, M, RWORK( IRVT ), M, VT, LDVT )
1446: CALL ZUNMBR( 'P', 'R', 'C', M, M, M, WORK( IL ), LDWRKL,
1447: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
1448: $ LWORK-NWORK+1, IERR )
1449: *
1450: * Copy VT to WORK(IL), multiply right singular vectors of L
1451: * in WORK(IL) by Q in A, storing result in VT
1452: * (CWorkspace: need M*M)
1453: * (RWorkspace: 0)
1454: *
1455: CALL ZLACPY( 'F', M, M, VT, LDVT, WORK( IL ), LDWRKL )
1456: CALL ZGEMM( 'N', 'N', M, N, M, CONE, WORK( IL ), LDWRKL,
1457: $ A, LDA, CZERO, VT, LDVT )
1458: *
1459: ELSE IF( WNTQA ) THEN
1460: *
1461: * Path 9t (N much larger than M, JOBZ='A')
1462: * N right singular vectors to be computed in VT and
1463: * M left singular vectors to be computed in U
1464: *
1465: IVT = 1
1466: *
1467: * WORK(IVT) is M by M
1468: *
1469: LDWKVT = M
1470: ITAU = IVT + LDWKVT*M
1471: NWORK = ITAU + M
1472: *
1473: * Compute A=L*Q, copying result to VT
1474: * (CWorkspace: need 2*M, prefer M+M*NB)
1475: * (RWorkspace: 0)
1476: *
1477: CALL ZGELQF( M, N, A, LDA, WORK( ITAU ), WORK( NWORK ),
1478: $ LWORK-NWORK+1, IERR )
1479: CALL ZLACPY( 'U', M, N, A, LDA, VT, LDVT )
1480: *
1481: * Generate Q in VT
1482: * (CWorkspace: need M+N, prefer M+N*NB)
1483: * (RWorkspace: 0)
1484: *
1485: CALL ZUNGLQ( N, N, M, VT, LDVT, WORK( ITAU ),
1486: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1487: *
1488: * Produce L in A, zeroing out above it
1489: *
1490: CALL ZLASET( 'U', M-1, M-1, CZERO, CZERO, A( 1, 2 ),
1491: $ LDA )
1492: IE = 1
1493: ITAUQ = ITAU
1494: ITAUP = ITAUQ + M
1495: NWORK = ITAUP + M
1496: *
1497: * Bidiagonalize L in A
1498: * (CWorkspace: need M*M+3*M, prefer M*M+2*M+2*M*NB)
1499: * (RWorkspace: need M)
1500: *
1501: CALL ZGEBRD( M, M, A, LDA, S, RWORK( IE ), WORK( ITAUQ ),
1502: $ WORK( ITAUP ), WORK( NWORK ), LWORK-NWORK+1,
1503: $ IERR )
1504: *
1505: * Perform bidiagonal SVD, computing left singular vectors
1506: * of bidiagonal matrix in RWORK(IRU) and computing right
1507: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1508: * (CWorkspace: need 0)
1509: * (RWorkspace: need BDSPAC)
1510: *
1511: IRU = IE + M
1512: IRVT = IRU + M*M
1513: NRWORK = IRVT + M*M
1514: CALL DBDSDC( 'U', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1515: $ M, RWORK( IRVT ), M, DUM, IDUM,
1516: $ RWORK( NRWORK ), IWORK, INFO )
1517: *
1518: * Copy real matrix RWORK(IRU) to complex matrix U
1519: * Overwrite U by left singular vectors of L
1520: * (CWorkspace: need 3*M, prefer 2*M+M*NB)
1521: * (RWorkspace: 0)
1522: *
1523: CALL ZLACP2( 'F', M, M, RWORK( IRU ), M, U, LDU )
1524: CALL ZUNMBR( 'Q', 'L', 'N', M, M, M, A, LDA,
1525: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
1526: $ LWORK-NWORK+1, IERR )
1527: *
1528: * Copy real matrix RWORK(IRVT) to complex matrix WORK(IVT)
1529: * Overwrite WORK(IVT) by right singular vectors of L
1530: * (CWorkspace: need M*M+3*M, prefer M*M+2*M+M*NB)
1531: * (RWorkspace: 0)
1532: *
1533: CALL ZLACP2( 'F', M, M, RWORK( IRVT ), M, WORK( IVT ),
1534: $ LDWKVT )
1535: CALL ZUNMBR( 'P', 'R', 'C', M, M, M, A, LDA,
1536: $ WORK( ITAUP ), WORK( IVT ), LDWKVT,
1537: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1538: *
1539: * Multiply right singular vectors of L in WORK(IVT) by
1540: * Q in VT, storing result in A
1541: * (CWorkspace: need M*M)
1542: * (RWorkspace: 0)
1543: *
1544: CALL ZGEMM( 'N', 'N', M, N, M, CONE, WORK( IVT ), LDWKVT,
1545: $ VT, LDVT, CZERO, A, LDA )
1546: *
1547: * Copy right singular vectors of A from A to VT
1548: *
1549: CALL ZLACPY( 'F', M, N, A, LDA, VT, LDVT )
1550: *
1551: END IF
1552: *
1553: ELSE IF( N.GE.MNTHR2 ) THEN
1554: *
1555: * MNTHR2 <= N < MNTHR1
1556: *
1557: * Path 5t (N much larger than M, but not as much as MNTHR1)
1558: * Reduce to bidiagonal form without QR decomposition, use
1559: * ZUNGBR and matrix multiplication to compute singular vectors
1560: *
1561: *
1562: IE = 1
1563: NRWORK = IE + M
1564: ITAUQ = 1
1565: ITAUP = ITAUQ + M
1566: NWORK = ITAUP + M
1567: *
1568: * Bidiagonalize A
1569: * (CWorkspace: need 2*M+N, prefer 2*M+(M+N)*NB)
1570: * (RWorkspace: M)
1571: *
1572: CALL ZGEBRD( M, N, A, LDA, S, RWORK( IE ), WORK( ITAUQ ),
1573: $ WORK( ITAUP ), WORK( NWORK ), LWORK-NWORK+1,
1574: $ IERR )
1575: *
1576: IF( WNTQN ) THEN
1577: *
1578: * Compute singular values only
1579: * (Cworkspace: 0)
1580: * (Rworkspace: need BDSPAN)
1581: *
1582: CALL DBDSDC( 'L', 'N', M, S, RWORK( IE ), DUM, 1, DUM, 1,
1583: $ DUM, IDUM, RWORK( NRWORK ), IWORK, INFO )
1584: ELSE IF( WNTQO ) THEN
1585: IRVT = NRWORK
1586: IRU = IRVT + M*M
1587: NRWORK = IRU + M*M
1588: IVT = NWORK
1589: *
1590: * Copy A to U, generate Q
1591: * (Cworkspace: need 2*M, prefer M+M*NB)
1592: * (Rworkspace: 0)
1593: *
1594: CALL ZLACPY( 'L', M, M, A, LDA, U, LDU )
1595: CALL ZUNGBR( 'Q', M, M, N, U, LDU, WORK( ITAUQ ),
1596: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1597: *
1598: * Generate P**H in A
1599: * (Cworkspace: need 2*M, prefer M+M*NB)
1600: * (Rworkspace: 0)
1601: *
1602: CALL ZUNGBR( 'P', M, N, M, A, LDA, WORK( ITAUP ),
1603: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1604: *
1605: LDWKVT = M
1606: IF( LWORK.GE.M*N+3*M ) THEN
1607: *
1608: * WORK( IVT ) is M by N
1609: *
1610: NWORK = IVT + LDWKVT*N
1611: CHUNK = N
1612: ELSE
1613: *
1614: * WORK( IVT ) is M by CHUNK
1615: *
1616: CHUNK = ( LWORK-3*M ) / M
1617: NWORK = IVT + LDWKVT*CHUNK
1618: END IF
1619: *
1620: * Perform bidiagonal SVD, computing left singular vectors
1621: * of bidiagonal matrix in RWORK(IRU) and computing right
1622: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1623: * (CWorkspace: need 0)
1624: * (RWorkspace: need BDSPAC)
1625: *
1626: CALL DBDSDC( 'L', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1627: $ M, RWORK( IRVT ), M, DUM, IDUM,
1628: $ RWORK( NRWORK ), IWORK, INFO )
1629: *
1630: * Multiply Q in U by real matrix RWORK(IRVT)
1631: * storing the result in WORK(IVT), copying to U
1632: * (Cworkspace: need 0)
1633: * (Rworkspace: need 2*M*M)
1634: *
1635: CALL ZLACRM( M, M, U, LDU, RWORK( IRU ), M, WORK( IVT ),
1636: $ LDWKVT, RWORK( NRWORK ) )
1637: CALL ZLACPY( 'F', M, M, WORK( IVT ), LDWKVT, U, LDU )
1638: *
1639: * Multiply RWORK(IRVT) by P**H in A, storing the
1640: * result in WORK(IVT), copying to A
1641: * (CWorkspace: need M*M, prefer M*N)
1642: * (Rworkspace: need 2*M*M, prefer 2*M*N)
1643: *
1644: NRWORK = IRU
1645: DO 50 I = 1, N, CHUNK
1646: BLK = MIN( N-I+1, CHUNK )
1647: CALL ZLARCM( M, BLK, RWORK( IRVT ), M, A( 1, I ), LDA,
1648: $ WORK( IVT ), LDWKVT, RWORK( NRWORK ) )
1649: CALL ZLACPY( 'F', M, BLK, WORK( IVT ), LDWKVT,
1650: $ A( 1, I ), LDA )
1651: 50 CONTINUE
1652: ELSE IF( WNTQS ) THEN
1653: *
1654: * Copy A to U, generate Q
1655: * (Cworkspace: need 2*M, prefer M+M*NB)
1656: * (Rworkspace: 0)
1657: *
1658: CALL ZLACPY( 'L', M, M, A, LDA, U, LDU )
1659: CALL ZUNGBR( 'Q', M, M, N, U, LDU, WORK( ITAUQ ),
1660: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1661: *
1662: * Copy A to VT, generate P**H
1663: * (Cworkspace: need 2*M, prefer M+M*NB)
1664: * (Rworkspace: 0)
1665: *
1666: CALL ZLACPY( 'U', M, N, A, LDA, VT, LDVT )
1667: CALL ZUNGBR( 'P', M, N, M, VT, LDVT, WORK( ITAUP ),
1668: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1669: *
1670: * Perform bidiagonal SVD, computing left singular vectors
1671: * of bidiagonal matrix in RWORK(IRU) and computing right
1672: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1673: * (CWorkspace: need 0)
1674: * (RWorkspace: need BDSPAC)
1675: *
1676: IRVT = NRWORK
1677: IRU = IRVT + M*M
1678: NRWORK = IRU + M*M
1679: CALL DBDSDC( 'L', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1680: $ M, RWORK( IRVT ), M, DUM, IDUM,
1681: $ RWORK( NRWORK ), IWORK, INFO )
1682: *
1683: * Multiply Q in U by real matrix RWORK(IRU), storing the
1684: * result in A, copying to U
1685: * (CWorkspace: need 0)
1686: * (Rworkspace: need 3*M*M)
1687: *
1688: CALL ZLACRM( M, M, U, LDU, RWORK( IRU ), M, A, LDA,
1689: $ RWORK( NRWORK ) )
1690: CALL ZLACPY( 'F', M, M, A, LDA, U, LDU )
1691: *
1692: * Multiply real matrix RWORK(IRVT) by P**H in VT,
1693: * storing the result in A, copying to VT
1694: * (Cworkspace: need 0)
1695: * (Rworkspace: need M*M+2*M*N)
1696: *
1697: NRWORK = IRU
1698: CALL ZLARCM( M, N, RWORK( IRVT ), M, VT, LDVT, A, LDA,
1699: $ RWORK( NRWORK ) )
1700: CALL ZLACPY( 'F', M, N, A, LDA, VT, LDVT )
1701: ELSE
1702: *
1703: * Copy A to U, generate Q
1704: * (Cworkspace: need 2*M, prefer M+M*NB)
1705: * (Rworkspace: 0)
1706: *
1707: CALL ZLACPY( 'L', M, M, A, LDA, U, LDU )
1708: CALL ZUNGBR( 'Q', M, M, N, U, LDU, WORK( ITAUQ ),
1709: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1710: *
1711: * Copy A to VT, generate P**H
1712: * (Cworkspace: need 2*M, prefer M+M*NB)
1713: * (Rworkspace: 0)
1714: *
1715: CALL ZLACPY( 'U', M, N, A, LDA, VT, LDVT )
1716: CALL ZUNGBR( 'P', N, N, M, VT, LDVT, WORK( ITAUP ),
1717: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1718: *
1719: * Perform bidiagonal SVD, computing left singular vectors
1720: * of bidiagonal matrix in RWORK(IRU) and computing right
1721: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1722: * (CWorkspace: need 0)
1723: * (RWorkspace: need BDSPAC)
1724: *
1725: IRVT = NRWORK
1726: IRU = IRVT + M*M
1727: NRWORK = IRU + M*M
1728: CALL DBDSDC( 'L', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1729: $ M, RWORK( IRVT ), M, DUM, IDUM,
1730: $ RWORK( NRWORK ), IWORK, INFO )
1731: *
1732: * Multiply Q in U by real matrix RWORK(IRU), storing the
1733: * result in A, copying to U
1734: * (CWorkspace: need 0)
1735: * (Rworkspace: need 3*M*M)
1736: *
1737: CALL ZLACRM( M, M, U, LDU, RWORK( IRU ), M, A, LDA,
1738: $ RWORK( NRWORK ) )
1739: CALL ZLACPY( 'F', M, M, A, LDA, U, LDU )
1740: *
1741: * Multiply real matrix RWORK(IRVT) by P**H in VT,
1742: * storing the result in A, copying to VT
1743: * (Cworkspace: need 0)
1744: * (Rworkspace: need M*M+2*M*N)
1745: *
1746: CALL ZLARCM( M, N, RWORK( IRVT ), M, VT, LDVT, A, LDA,
1747: $ RWORK( NRWORK ) )
1748: CALL ZLACPY( 'F', M, N, A, LDA, VT, LDVT )
1749: END IF
1750: *
1751: ELSE
1752: *
1753: * N .LT. MNTHR2
1754: *
1755: * Path 6t (N greater than M, but not much larger)
1756: * Reduce to bidiagonal form without LQ decomposition
1757: * Use ZUNMBR to compute singular vectors
1758: *
1759: IE = 1
1760: NRWORK = IE + M
1761: ITAUQ = 1
1762: ITAUP = ITAUQ + M
1763: NWORK = ITAUP + M
1764: *
1765: * Bidiagonalize A
1766: * (CWorkspace: need 2*M+N, prefer 2*M+(M+N)*NB)
1767: * (RWorkspace: M)
1768: *
1769: CALL ZGEBRD( M, N, A, LDA, S, RWORK( IE ), WORK( ITAUQ ),
1770: $ WORK( ITAUP ), WORK( NWORK ), LWORK-NWORK+1,
1771: $ IERR )
1772: IF( WNTQN ) THEN
1773: *
1774: * Compute singular values only
1775: * (Cworkspace: 0)
1776: * (Rworkspace: need BDSPAN)
1777: *
1778: CALL DBDSDC( 'L', 'N', M, S, RWORK( IE ), DUM, 1, DUM, 1,
1779: $ DUM, IDUM, RWORK( NRWORK ), IWORK, INFO )
1780: ELSE IF( WNTQO ) THEN
1781: LDWKVT = M
1782: IVT = NWORK
1783: IF( LWORK.GE.M*N+3*M ) THEN
1784: *
1785: * WORK( IVT ) is M by N
1786: *
1787: CALL ZLASET( 'F', M, N, CZERO, CZERO, WORK( IVT ),
1788: $ LDWKVT )
1789: NWORK = IVT + LDWKVT*N
1790: ELSE
1791: *
1792: * WORK( IVT ) is M by CHUNK
1793: *
1794: CHUNK = ( LWORK-3*M ) / M
1795: NWORK = IVT + LDWKVT*CHUNK
1796: END IF
1797: *
1798: * Perform bidiagonal SVD, computing left singular vectors
1799: * of bidiagonal matrix in RWORK(IRU) and computing right
1800: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1801: * (CWorkspace: need 0)
1802: * (RWorkspace: need BDSPAC)
1803: *
1804: IRVT = NRWORK
1805: IRU = IRVT + M*M
1806: NRWORK = IRU + M*M
1807: CALL DBDSDC( 'L', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1808: $ M, RWORK( IRVT ), M, DUM, IDUM,
1809: $ RWORK( NRWORK ), IWORK, INFO )
1810: *
1811: * Copy real matrix RWORK(IRU) to complex matrix U
1812: * Overwrite U by left singular vectors of A
1813: * (Cworkspace: need 2*M, prefer M+M*NB)
1814: * (Rworkspace: need 0)
1815: *
1816: CALL ZLACP2( 'F', M, M, RWORK( IRU ), M, U, LDU )
1817: CALL ZUNMBR( 'Q', 'L', 'N', M, M, N, A, LDA,
1818: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
1819: $ LWORK-NWORK+1, IERR )
1820: *
1821: IF( LWORK.GE.M*N+3*M ) THEN
1822: *
1823: * Copy real matrix RWORK(IRVT) to complex matrix WORK(IVT)
1824: * Overwrite WORK(IVT) by right singular vectors of A,
1825: * copying to A
1826: * (Cworkspace: need M*N+2*M, prefer M*N+M+M*NB)
1827: * (Rworkspace: need 0)
1828: *
1829: CALL ZLACP2( 'F', M, M, RWORK( IRVT ), M, WORK( IVT ),
1830: $ LDWKVT )
1831: CALL ZUNMBR( 'P', 'R', 'C', M, N, M, A, LDA,
1832: $ WORK( ITAUP ), WORK( IVT ), LDWKVT,
1833: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1834: CALL ZLACPY( 'F', M, N, WORK( IVT ), LDWKVT, A, LDA )
1835: ELSE
1836: *
1837: * Generate P**H in A
1838: * (Cworkspace: need 2*M, prefer M+M*NB)
1839: * (Rworkspace: need 0)
1840: *
1841: CALL ZUNGBR( 'P', M, N, M, A, LDA, WORK( ITAUP ),
1842: $ WORK( NWORK ), LWORK-NWORK+1, IERR )
1843: *
1844: * Multiply Q in A by real matrix RWORK(IRU), storing the
1845: * result in WORK(IU), copying to A
1846: * (CWorkspace: need M*M, prefer M*N)
1847: * (Rworkspace: need 3*M*M, prefer M*M+2*M*N)
1848: *
1849: NRWORK = IRU
1850: DO 60 I = 1, N, CHUNK
1851: BLK = MIN( N-I+1, CHUNK )
1852: CALL ZLARCM( M, BLK, RWORK( IRVT ), M, A( 1, I ),
1853: $ LDA, WORK( IVT ), LDWKVT,
1854: $ RWORK( NRWORK ) )
1855: CALL ZLACPY( 'F', M, BLK, WORK( IVT ), LDWKVT,
1856: $ A( 1, I ), LDA )
1857: 60 CONTINUE
1858: END IF
1859: ELSE IF( WNTQS ) THEN
1860: *
1861: * Perform bidiagonal SVD, computing left singular vectors
1862: * of bidiagonal matrix in RWORK(IRU) and computing right
1863: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1864: * (CWorkspace: need 0)
1865: * (RWorkspace: need BDSPAC)
1866: *
1867: IRVT = NRWORK
1868: IRU = IRVT + M*M
1869: NRWORK = IRU + M*M
1870: CALL DBDSDC( 'L', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1871: $ M, RWORK( IRVT ), M, DUM, IDUM,
1872: $ RWORK( NRWORK ), IWORK, INFO )
1873: *
1874: * Copy real matrix RWORK(IRU) to complex matrix U
1875: * Overwrite U by left singular vectors of A
1876: * (CWorkspace: need 3*M, prefer 2*M+M*NB)
1877: * (RWorkspace: M*M)
1878: *
1879: CALL ZLACP2( 'F', M, M, RWORK( IRU ), M, U, LDU )
1880: CALL ZUNMBR( 'Q', 'L', 'N', M, M, N, A, LDA,
1881: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
1882: $ LWORK-NWORK+1, IERR )
1883: *
1884: * Copy real matrix RWORK(IRVT) to complex matrix VT
1885: * Overwrite VT by right singular vectors of A
1886: * (CWorkspace: need 3*M, prefer 2*M+M*NB)
1887: * (RWorkspace: M*M)
1888: *
1889: CALL ZLASET( 'F', M, N, CZERO, CZERO, VT, LDVT )
1890: CALL ZLACP2( 'F', M, M, RWORK( IRVT ), M, VT, LDVT )
1891: CALL ZUNMBR( 'P', 'R', 'C', M, N, M, A, LDA,
1892: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
1893: $ LWORK-NWORK+1, IERR )
1894: ELSE
1895: *
1896: * Perform bidiagonal SVD, computing left singular vectors
1897: * of bidiagonal matrix in RWORK(IRU) and computing right
1898: * singular vectors of bidiagonal matrix in RWORK(IRVT)
1899: * (CWorkspace: need 0)
1900: * (RWorkspace: need BDSPAC)
1901: *
1902: IRVT = NRWORK
1903: IRU = IRVT + M*M
1904: NRWORK = IRU + M*M
1905: *
1906: CALL DBDSDC( 'L', 'I', M, S, RWORK( IE ), RWORK( IRU ),
1907: $ M, RWORK( IRVT ), M, DUM, IDUM,
1908: $ RWORK( NRWORK ), IWORK, INFO )
1909: *
1910: * Copy real matrix RWORK(IRU) to complex matrix U
1911: * Overwrite U by left singular vectors of A
1912: * (CWorkspace: need 3*M, prefer 2*M+M*NB)
1913: * (RWorkspace: M*M)
1914: *
1915: CALL ZLACP2( 'F', M, M, RWORK( IRU ), M, U, LDU )
1916: CALL ZUNMBR( 'Q', 'L', 'N', M, M, N, A, LDA,
1917: $ WORK( ITAUQ ), U, LDU, WORK( NWORK ),
1918: $ LWORK-NWORK+1, IERR )
1919: *
1920: * Set all of VT to identity matrix
1921: *
1922: CALL ZLASET( 'F', N, N, CZERO, CONE, VT, LDVT )
1923: *
1924: * Copy real matrix RWORK(IRVT) to complex matrix VT
1925: * Overwrite VT by right singular vectors of A
1926: * (CWorkspace: need 2*M+N, prefer 2*M+N*NB)
1927: * (RWorkspace: M*M)
1928: *
1929: CALL ZLACP2( 'F', M, M, RWORK( IRVT ), M, VT, LDVT )
1930: CALL ZUNMBR( 'P', 'R', 'C', N, N, M, A, LDA,
1931: $ WORK( ITAUP ), VT, LDVT, WORK( NWORK ),
1932: $ LWORK-NWORK+1, IERR )
1933: END IF
1934: *
1935: END IF
1936: *
1937: END IF
1938: *
1939: * Undo scaling if necessary
1940: *
1941: IF( ISCL.EQ.1 ) THEN
1942: IF( ANRM.GT.BIGNUM )
1943: $ CALL DLASCL( 'G', 0, 0, BIGNUM, ANRM, MINMN, 1, S, MINMN,
1944: $ IERR )
1945: IF( INFO.NE.0 .AND. ANRM.GT.BIGNUM )
1946: $ CALL DLASCL( 'G', 0, 0, BIGNUM, ANRM, MINMN-1, 1,
1947: $ RWORK( IE ), MINMN, IERR )
1948: IF( ANRM.LT.SMLNUM )
1949: $ CALL DLASCL( 'G', 0, 0, SMLNUM, ANRM, MINMN, 1, S, MINMN,
1950: $ IERR )
1951: IF( INFO.NE.0 .AND. ANRM.LT.SMLNUM )
1952: $ CALL DLASCL( 'G', 0, 0, SMLNUM, ANRM, MINMN-1, 1,
1953: $ RWORK( IE ), MINMN, IERR )
1954: END IF
1955: *
1956: * Return optimal workspace in WORK(1)
1957: *
1958: WORK( 1 ) = MAXWRK
1959: *
1960: RETURN
1961: *
1962: * End of ZGESDD
1963: *
1964: END
CVSweb interface <joel.bertrand@systella.fr>
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to zgesdd.f CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [local] / rpl / lapack / lapack