5 * SUBROUTINE SLAMSWLQ( SIDE, TRANS, M, N, K, MB, NB, A, LDA, T,
6 * $ LDT, C, LDC, WORK, LWORK, INFO )
9 * .. Scalar Arguments ..
10 * CHARACTER SIDE, TRANS
11 * INTEGER INFO, LDA, M, N, K, MB, NB, LDT, LWORK, LDC
13 * .. Array Arguments ..
14 * DOUBLE A( LDA, * ), WORK( * ), C(LDC, * ),
21 *> SLAMSWLQ overwrites the general real M-by-N matrix C with
24 *> SIDE = 'L' SIDE = 'R'
25 *> TRANS = 'N': Q * C C * Q
26 *> TRANS = 'T': Q**T * C C * Q**T
27 *> where Q is a real orthogonal matrix defined as the product of blocked
28 *> elementary reflectors computed by short wide LQ
29 *> factorization (SLASWLQ)
37 *> SIDE is CHARACTER*1
38 *> = 'L': apply Q or Q**T from the Left;
39 *> = 'R': apply Q or Q**T from the Right.
44 *> TRANS is CHARACTER*1
45 *> = 'N': No transpose, apply Q;
46 *> = 'T': Transpose, apply Q**T.
52 *> The number of rows of the matrix C. M >=0.
58 *> The number of columns of the matrix C. N >= M.
64 *> The number of elementary reflectors whose product defines
72 *> The row block size to be used in the blocked QR.
79 *> The column block size to be used in the blocked QR.
86 *> The block size to be used in the blocked QR.
93 *> A is REAL array, dimension
94 *> (LDA,M) if SIDE = 'L',
95 *> (LDA,N) if SIDE = 'R'
96 *> The i-th row must contain the vector which defines the blocked
97 *> elementary reflector H(i), for i = 1,2,...,k, as returned by
98 *> SLASWLQ in the first k rows of its array argument A.
104 *> The leading dimension of the array A.
105 *> If SIDE = 'L', LDA >= max(1,M);
106 *> if SIDE = 'R', LDA >= max(1,N).
111 *> T is REAL array, dimension
112 *> ( M * Number of blocks(CEIL(N-K/NB-K)),
113 *> The blocked upper triangular block reflectors stored in compact form
114 *> as a sequence of upper triangular blocks. See below
115 *> for further details.
121 *> The leading dimension of the array T. LDT >= MB.
126 *> C is REAL array, dimension (LDC,N)
127 *> On entry, the M-by-N matrix C.
128 *> On exit, C is overwritten by Q*C or Q**T*C or C*Q**T or C*Q.
134 *> The leading dimension of the array C. LDC >= max(1,M).
139 *> (workspace) REAL array, dimension (MAX(1,LWORK))
145 *> The dimension of the array WORK.
146 *> If SIDE = 'L', LWORK >= max(1,NB) * MB;
147 *> if SIDE = 'R', LWORK >= max(1,M) * MB.
148 *> If LWORK = -1, then a workspace query is assumed; the routine
149 *> only calculates the optimal size of the WORK array, returns
150 *> this value as the first entry of the WORK array, and no error
151 *> message related to LWORK is issued by XERBLA.
157 *> = 0: successful exit
158 *> < 0: if INFO = -i, the i-th argument had an illegal value
164 *> \author Univ. of Tennessee
165 *> \author Univ. of California Berkeley
166 *> \author Univ. of Colorado Denver
169 *> \par Further Details:
170 * =====================
173 *> Short-Wide LQ (SWLQ) performs LQ by a sequence of orthogonal transformations,
174 *> representing Q as a product of other orthogonal matrices
175 *> Q = Q(1) * Q(2) * . . . * Q(k)
176 *> where each Q(i) zeros out upper diagonal entries of a block of NB rows of A:
177 *> Q(1) zeros out the upper diagonal entries of rows 1:NB of A
178 *> Q(2) zeros out the bottom MB-N rows of rows [1:M,NB+1:2*NB-M] of A
179 *> Q(3) zeros out the bottom MB-N rows of rows [1:M,2*NB-M+1:3*NB-2*M] of A
182 *> Q(1) is computed by GELQT, which represents Q(1) by Householder vectors
183 *> stored under the diagonal of rows 1:MB of A, and by upper triangular
184 *> block reflectors, stored in array T(1:LDT,1:N).
185 *> For more information see Further Details in GELQT.
187 *> Q(i) for i>1 is computed by TPLQT, which represents Q(i) by Householder vectors
188 *> stored in columns [(i-1)*(NB-M)+M+1:i*(NB-M)+M] of A, and by upper triangular
189 *> block reflectors, stored in array T(1:LDT,(i-1)*M+1:i*M).
190 *> The last Q(k) may use fewer rows.
191 *> For more information see Further Details in TPQRT.
193 *> For more details of the overall algorithm, see the description of
194 *> Sequential TSQR in Section 2.2 of [1].
196 *> [1] “Communication-Optimal Parallel and Sequential QR and LU Factorizations,”
197 *> J. Demmel, L. Grigori, M. Hoemmen, J. Langou,
198 *> SIAM J. Sci. Comput, vol. 34, no. 1, 2012
201 * =====================================================================
202 SUBROUTINE SLAMSWLQ( SIDE, TRANS, M, N, K, MB, NB, A, LDA, T,
203 $ LDT, C, LDC, WORK, LWORK, INFO )
205 * -- LAPACK computational routine (version 3.7.0) --
206 * -- LAPACK is a software package provided by Univ. of Tennessee, --
207 * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
210 * .. Scalar Arguments ..
211 CHARACTER SIDE, TRANS
212 INTEGER INFO, LDA, M, N, K, MB, NB, LDT, LWORK, LDC
214 * .. Array Arguments ..
215 REAL A( LDA, * ), WORK( * ), C(LDC, * ),
219 * =====================================================================
222 * .. Local Scalars ..
223 LOGICAL LEFT, RIGHT, TRAN, NOTRAN, LQUERY
224 INTEGER I, II, KK, LW, CTR
226 * .. External Functions ..
229 * .. External Subroutines ..
230 EXTERNAL STPMLQT, SGEMLQT, XERBLA
232 * .. Executable Statements ..
234 * Test the input arguments
237 NOTRAN = LSAME( TRANS, 'N' )
238 TRAN = LSAME( TRANS, 'T' )
239 LEFT = LSAME( SIDE, 'L' )
240 RIGHT = LSAME( SIDE, 'R' )
248 IF( .NOT.LEFT .AND. .NOT.RIGHT ) THEN
250 ELSE IF( .NOT.TRAN .AND. .NOT.NOTRAN ) THEN
252 ELSE IF( M.LT.0 ) THEN
254 ELSE IF( N.LT.0 ) THEN
256 ELSE IF( K.LT.0 ) THEN
258 ELSE IF( LDA.LT.MAX( 1, K ) ) THEN
260 ELSE IF( LDT.LT.MAX( 1, MB) ) THEN
262 ELSE IF( LDC.LT.MAX( 1, M ) ) THEN
264 ELSE IF(( LWORK.LT.MAX(1,LW)).AND.(.NOT.LQUERY)) THEN
269 CALL XERBLA( 'SLAMSWLQ', -INFO )
272 ELSE IF (LQUERY) THEN
277 * Quick return if possible
279 IF( MIN(M,N,K).EQ.0 ) THEN
283 IF((NB.LE.K).OR.(NB.GE.MAX(M,N,K))) THEN
284 CALL DGEMLQT( SIDE, TRANS, M, N, K, MB, A, LDA,
285 $ T, LDT, C, LDC, WORK, INFO)
289 IF(LEFT.AND.TRAN) THEN
291 * Multiply Q to the last block of C
293 KK = MOD((M-K),(NB-K))
298 CALL STPMLQT('L','T',KK , N, K, 0, MB, A(1,II), LDA,
299 $ T(1,CTR*K+1), LDT, C(1,1), LDC,
300 $ C(II,1), LDC, WORK, INFO )
305 DO I=II-(NB-K),NB+1,-(NB-K)
307 * Multiply Q to the current block of C (1:M,I:I+NB)
310 CALL STPMLQT('L','T',NB-K , N, K, 0,MB, A(1,I), LDA,
311 $ T(1,CTR*K+1),LDT, C(1,1), LDC,
312 $ C(I,1), LDC, WORK, INFO )
315 * Multiply Q to the first block of C (1:M,1:NB)
317 CALL SGEMLQT('L','T',NB , N, K, MB, A(1,1), LDA, T
318 $ ,LDT ,C(1,1), LDC, WORK, INFO )
320 ELSE IF (LEFT.AND.NOTRAN) THEN
322 * Multiply Q to the first block of C
324 KK = MOD((M-K),(NB-K))
327 CALL SGEMLQT('L','N',NB , N, K, MB, A(1,1), LDA, T
328 $ ,LDT ,C(1,1), LDC, WORK, INFO )
330 DO I=NB+1,II-NB+K,(NB-K)
332 * Multiply Q to the current block of C (I:I+NB,1:N)
334 CALL STPMLQT('L','N',NB-K , N, K, 0,MB, A(1,I), LDA,
335 $ T(1,CTR * K+1), LDT, C(1,1), LDC,
336 $ C(I,1), LDC, WORK, INFO )
342 * Multiply Q to the last block of C
344 CALL STPMLQT('L','N',KK , N, K, 0, MB, A(1,II), LDA,
345 $ T(1,CTR*K+1), LDT, C(1,1), LDC,
346 $ C(II,1), LDC, WORK, INFO )
350 ELSE IF(RIGHT.AND.NOTRAN) THEN
352 * Multiply Q to the last block of C
354 KK = MOD((N-K),(NB-K))
358 CALL STPMLQT('R','N',M , KK, K, 0, MB, A(1, II), LDA,
359 $ T(1,CTR*K+1), LDT, C(1,1), LDC,
360 $ C(1,II), LDC, WORK, INFO )
365 DO I=II-(NB-K),NB+1,-(NB-K)
367 * Multiply Q to the current block of C (1:M,I:I+MB)
370 CALL STPMLQT('R','N', M, NB-K, K, 0, MB, A(1, I), LDA,
371 $ T(1,CTR*K+1), LDT, C(1,1), LDC,
372 $ C(1,I), LDC, WORK, INFO )
376 * Multiply Q to the first block of C (1:M,1:MB)
378 CALL SGEMLQT('R','N',M , NB, K, MB, A(1,1), LDA, T
379 $ ,LDT ,C(1,1), LDC, WORK, INFO )
381 ELSE IF (RIGHT.AND.TRAN) THEN
383 * Multiply Q to the first block of C
385 KK = MOD((N-K),(NB-K))
388 CALL SGEMLQT('R','T',M , NB, K, MB, A(1,1), LDA, T
389 $ ,LDT ,C(1,1), LDC, WORK, INFO )
391 DO I=NB+1,II-NB+K,(NB-K)
393 * Multiply Q to the current block of C (1:M,I:I+MB)
395 CALL STPMLQT('R','T',M , NB-K, K, 0,MB, A(1,I), LDA,
396 $ T(1, CTR*K+1), LDT, C(1,1), LDC,
397 $ C(1,I), LDC, WORK, INFO )
403 * Multiply Q to the last block of C
405 CALL STPMLQT('R','T',M , KK, K, 0,MB, A(1,II), LDA,
406 $ T(1,CTR*K+1),LDT, C(1,1), LDC,
407 $ C(1,II), LDC, WORK, INFO )