LAPACK  3.7.0
LAPACK: Linear Algebra PACKage
claswlq.f
Go to the documentation of this file.
1 *
2 * Definition:
3 * ===========
4 *
5 * SUBROUTINE CLASWLQ( M, N, MB, NB, A, LDA, T, LDT, WORK,
6 * LWORK, INFO)
7 *
8 * .. Scalar Arguments ..
9 * INTEGER INFO, LDA, M, N, MB, NB, LDT, LWORK
10 * ..
11 * .. Array Arguments ..
12 * COMPLEX A( LDA, * ), T( LDT, * ), WORK( * )
13 * ..
14 *
15 *
16 *> \par Purpose:
17 * =============
18 *>
19 *> \verbatim
20 *>
21 *> CLASWLQ computes a blocked Short-Wide LQ factorization of a
22 *> M-by-N matrix A, where N >= M:
23 *> A = L * Q
24 *> \endverbatim
25 *
26 * Arguments:
27 * ==========
28 *
29 *> \param[in] M
30 *> \verbatim
31 *> M is INTEGER
32 *> The number of rows of the matrix A. M >= 0.
33 *> \endverbatim
34 *>
35 *> \param[in] N
36 *> \verbatim
37 *> N is INTEGER
38 *> The number of columns of the matrix A. N >= M >= 0.
39 *> \endverbatim
40 *>
41 *> \param[in] MB
42 *> \verbatim
43 *> MB is INTEGER
44 *> The row block size to be used in the blocked QR.
45 *> M >= MB >= 1
46 *> \endverbatim
47 *> \param[in] NB
48 *> \verbatim
49 *> NB is INTEGER
50 *> The column block size to be used in the blocked QR.
51 *> NB > M.
52 *> \endverbatim
53 *>
54 *> \param[in,out] A
55 *> \verbatim
56 *> A is COMPLEX array, dimension (LDA,N)
57 *> On entry, the M-by-N matrix A.
58 *> On exit, the elements on and bleow the diagonal
59 *> of the array contain the N-by-N lower triangular matrix L;
60 *> the elements above the diagonal represent Q by the rows
61 *> of blocked V (see Further Details).
62 *>
63 *> \endverbatim
64 *>
65 *> \param[in] LDA
66 *> \verbatim
67 *> LDA is INTEGER
68 *> The leading dimension of the array A. LDA >= max(1,M).
69 *> \endverbatim
70 *>
71 *> \param[out] T
72 *> \verbatim
73 *> T is COMPLEX array,
74 *> dimension (LDT, N * Number_of_row_blocks)
75 *> where Number_of_row_blocks = CEIL((N-M)/(NB-M))
76 *> The blocked upper triangular block reflectors stored in compact form
77 *> as a sequence of upper triangular blocks.
78 *> See Further Details below.
79 *> \endverbatim
80 *>
81 *> \param[in] LDT
82 *> \verbatim
83 *> LDT is INTEGER
84 *> The leading dimension of the array T. LDT >= MB.
85 *> \endverbatim
86 *>
87 *>
88 *> \param[out] WORK
89 *> \verbatim
90 *> (workspace) COMPLEX array, dimension (MAX(1,LWORK))
91 *>
92 *> \endverbatim
93 *> \param[in] LWORK
94 *> \verbatim
95 *> The dimension of the array WORK. LWORK >= MB*M.
96 *> If LWORK = -1, then a workspace query is assumed; the routine
97 *> only calculates the optimal size of the WORK array, returns
98 *> this value as the first entry of the WORK array, and no error
99 *> message related to LWORK is issued by XERBLA.
100 *>
101 *> \endverbatim
102 *> \param[out] INFO
103 *> \verbatim
104 *> INFO is INTEGER
105 *> = 0: successful exit
106 *> < 0: if INFO = -i, the i-th argument had an illegal value
107 *> \endverbatim
108 *
109 * Authors:
110 * ========
111 *
112 *> \author Univ. of Tennessee
113 *> \author Univ. of California Berkeley
114 *> \author Univ. of Colorado Denver
115 *> \author NAG Ltd.
116 *
117 *> \par Further Details:
118 * =====================
119 *>
120 *> \verbatim
121 *> Short-Wide LQ (SWLQ) performs LQ by a sequence of orthogonal transformations,
122 *> representing Q as a product of other orthogonal matrices
123 *> Q = Q(1) * Q(2) * . . . * Q(k)
124 *> where each Q(i) zeros out upper diagonal entries of a block of NB rows of A:
125 *> Q(1) zeros out the upper diagonal entries of rows 1:NB of A
126 *> Q(2) zeros out the bottom MB-N rows of rows [1:M,NB+1:2*NB-M] of A
127 *> Q(3) zeros out the bottom MB-N rows of rows [1:M,2*NB-M+1:3*NB-2*M] of A
128 *> . . .
129 *>
130 *> Q(1) is computed by GELQT, which represents Q(1) by Householder vectors
131 *> stored under the diagonal of rows 1:MB of A, and by upper triangular
132 *> block reflectors, stored in array T(1:LDT,1:N).
133 *> For more information see Further Details in GELQT.
134 *>
135 *> Q(i) for i>1 is computed by TPLQT, which represents Q(i) by Householder vectors
136 *> stored in columns [(i-1)*(NB-M)+M+1:i*(NB-M)+M] of A, and by upper triangular
137 *> block reflectors, stored in array T(1:LDT,(i-1)*M+1:i*M).
138 *> The last Q(k) may use fewer rows.
139 *> For more information see Further Details in TPQRT.
140 *>
141 *> For more details of the overall algorithm, see the description of
142 *> Sequential TSQR in Section 2.2 of [1].
143 *>
144 *> [1] “Communication-Optimal Parallel and Sequential QR and LU Factorizations,”
145 *> J. Demmel, L. Grigori, M. Hoemmen, J. Langou,
146 *> SIAM J. Sci. Comput, vol. 34, no. 1, 2012
147 *> \endverbatim
148 *>
149 * =====================================================================
150  SUBROUTINE claswlq( M, N, MB, NB, A, LDA, T, LDT, WORK, LWORK,
151  $ info)
152 *
153 * -- LAPACK computational routine (version 3.7.0) --
154 * -- LAPACK is a software package provided by Univ. of Tennessee, --
155 * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd. --
156 * December 2016
157 *
158 * .. Scalar Arguments ..
159  INTEGER INFO, LDA, M, N, MB, NB, LWORK, LDT
160 * ..
161 * .. Array Arguments ..
162  COMPLEX A( lda, * ), WORK( * ), T( ldt, *)
163 * ..
164 *
165 * =====================================================================
166 *
167 * ..
168 * .. Local Scalars ..
169  LOGICAL LQUERY
170  INTEGER I, II, KK, CTR
171 * ..
172 * .. EXTERNAL FUNCTIONS ..
173  LOGICAL LSAME
174  EXTERNAL lsame
175 * .. EXTERNAL SUBROUTINES ..
176  EXTERNAL cgelqt, ctplqt, xerbla
177 * .. INTRINSIC FUNCTIONS ..
178  INTRINSIC max, min, mod
179 * ..
180 * .. EXTERNAL FUNCTIONS ..
181  INTEGER ILAENV
182  EXTERNAL ilaenv
183 * ..
184 * .. EXECUTABLE STATEMENTS ..
185 *
186 * TEST THE INPUT ARGUMENTS
187 *
188  info = 0
189 *
190  lquery = ( lwork.EQ.-1 )
191 *
192  IF( m.LT.0 ) THEN
193  info = -1
194  ELSE IF( n.LT.0 .OR. n.LT.m ) THEN
195  info = -2
196  ELSE IF( mb.LT.1 .OR. ( mb.GT.m .AND. m.GT.0 )) THEN
197  info = -3
198  ELSE IF( nb.LE.m ) THEN
199  info = -4
200  ELSE IF( lda.LT.max( 1, m ) ) THEN
201  info = -5
202  ELSE IF( ldt.LT.mb ) THEN
203  info = -8
204  ELSE IF( ( lwork.LT.m*mb) .AND. (.NOT.lquery) ) THEN
205  info = -10
206  END IF
207  IF( info.EQ.0) THEN
208  work(1) = mb*m
209  END IF
210 *
211  IF( info.NE.0 ) THEN
212  CALL xerbla( 'CLASWLQ', -info )
213  RETURN
214  ELSE IF (lquery) THEN
215  RETURN
216  END IF
217 *
218 * Quick return if possible
219 *
220  IF( min(m,n).EQ.0 ) THEN
221  RETURN
222  END IF
223 *
224 * The LQ Decomposition
225 *
226  IF((m.GE.n).OR.(nb.LE.m).OR.(nb.GE.n)) THEN
227  CALL cgelqt( m, n, mb, a, lda, t, ldt, work, info)
228  RETURN
229  END IF
230 *
231  kk = mod((n-m),(nb-m))
232  ii=n-kk+1
233 *
234 * Compute the LQ factorization of the first block A(1:M,1:NB)
235 *
236  CALL cgelqt( m, nb, mb, a(1,1), lda, t, ldt, work, info)
237  ctr = 1
238 *
239  DO i = nb+1, ii-nb+m , (nb-m)
240 *
241 * Compute the QR factorization of the current block A(1:M,I:I+NB-M)
242 *
243  CALL ctplqt( m, nb-m, 0, mb, a(1,1), lda, a( 1, i ),
244  $ lda, t(1,ctr*m+1),
245  $ ldt, work, info )
246  ctr = ctr + 1
247  END DO
248 *
249 * Compute the QR factorization of the last block A(1:M,II:N)
250 *
251  IF (ii.LE.n) THEN
252  CALL ctplqt( m, kk, 0, mb, a(1,1), lda, a( 1, ii ),
253  $ lda, t(1,ctr*m+1), ldt,
254  $ work, info )
255  END IF
256 *
257  work( 1 ) = m * mb
258  RETURN
259 *
260 * End of CLASWLQ
261 *
262  END
subroutine ctplqt(M, N, L, MB, A, LDA, B, LDB, T, LDT, WORK, INFO)
Definition: ctplqt.f:174
subroutine cgelqt(M, N, MB, A, LDA, T, LDT, WORK, INFO)
Definition: cgelqt.f:125
subroutine xerbla(SRNAME, INFO)
XERBLA
Definition: xerbla.f:62
subroutine claswlq(M, N, MB, NB, A, LDA, T, LDT, WORK, LWORK, INFO)
Definition: claswlq.f:152