1
       2
       3
       4
       5
       6
       7
       8
       9
      10
      11
      12
      13
      14
      15
      16
      17
      18
      19
      20
      21
      22
      23
      24
      25
      26
      27
      28
      29
      30
      31
      32
      33
      34
      35
      36
      37
      38
      39
      40
      41
      42
      43
      44
      45
      46
      47
      48
      49
      50
      51
      52
      53
      54
      55
      56
      57
      58
      59
      60
      61
      62
      63
      64
      65
      66
      67
      68
      69
      70
      71
      72
      73
      74
      75
      76
      77
      78
      79
      80
      81
      82
      83
      84
      85
      86
      87
      88
      89
      90
      91
      92
      93
      94
      95
      96
      97
      98
      99
     100
     101
     102
     103
     104
     105
     106
     107
     108
     109
     110
     111
     112
     113
     114
     115
     116
     117
     118
     119
     120
     121
     122
     123
     124
     125
     126
     127
     128
     129
     130
     131
     132
     133
     134
     135
     136
     137
     138
     139
     140
     141
     142
     143
     144
     145
     146
     147
     148
     149
     150
     151
     152
     153
     154
     155
     156
     157
     158
     159
     160
     161
     162
     163
     164
     165
     166
     167
     168
     169
     170
     171
     172
     173
     174
     175
     176
     177
     178
     179
     180
     181
     182
     183
     184
     185
     186
     187
     188
     189
     190
     191
     192
     193
     194
     195
     196
     197
     198
     199
     200
     201
     202
     203
     204
     205
     206
     207
     208
     209
     210
     211
     212
     213
     214
     215
     216
     217
     218
     219
     220
     221
     222
     223
     224
     225
     226
     227
     228
     229
     230
     231
     232
     233
     234
     235
     236
     237
     238
     239
     240
     241
     242
     243
     244
     245
     246
     247
     248
     249
     250
     251
     252
     253
     254
     255
     256
     257
     258
     259
     260
     261
     262
     263
     264
     265
     266
     267
     268
     269
     270
     271
     272
     273
     274
     275
     276
     277
     278
     279
     280
     281
     282
     283
     284
     285
     286
     287
     288
     289
     290
     291
     292
     293
     294
     295
     296
     297
     298
     299
     300
     301
     302
     303
     304
     305
     306
     307
     308
     309
     310
     311
     312
     313
     314
     315
     316
     317
     318
     319
     320
     321
     322
     323
     324
     325
     326
     327
     328
     329
     330
     331
     332
     333
     334
     335
     336
     337
     338
     339
     340
     341
     342
     343
     344
     345
     346
     347
     348
     349
     350
     351
     352
     353
     354
     355
     356
     357
     358
     359
     360
     361
     362
     363
     364
     365
     366
     367
     368
     369
     370
     371
     372
     373
     374
     375
     376
     377
     378
     379
     380
     381
     382
     383
     384
     385
     386
     387
     388
     389
     390
     391
     392
     393
     394
     395
     396
     397
     398
     399
     400
     401
     402
     403
     404
     405
     406
     407
     408
     409
     410
     411
     412
     413
     414
     415
     416
     417
     418
     419
     420
     421
     422
     423
     424
     425
     426
     427
     428
     429
     430
     431
     432
     433
     434
     435
      SUBROUTINE SLALSD( UPLO, SMLSIZ, N, NRHS, D, E, B, LDB, RCOND,
     $                   RANK, WORK, IWORK, INFO )
*
*  -- LAPACK routine (version 3.2.2) --
*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
*     June 2010
*
*     .. Scalar Arguments ..
      CHARACTER          UPLO
      INTEGER            INFO, LDB, N, NRHS, RANK, SMLSIZ
      REAL               RCOND
*     ..
*     .. Array Arguments ..
      INTEGER            IWORK( * )
      REAL               B( LDB, * ), D( * ), E( * ), WORK( * )
*     ..
*
*  Purpose
*  =======
*
*  SLALSD uses the singular value decomposition of A to solve the least
*  squares problem of finding X to minimize the Euclidean norm of each
*  column of A*X-B, where A is N-by-N upper bidiagonal, and X and B
*  are N-by-NRHS. The solution X overwrites B.
*
*  The singular values of A smaller than RCOND times the largest
*  singular value are treated as zero in solving the least squares
*  problem; in this case a minimum norm solution is returned.
*  The actual singular values are returned in D in ascending order.
*
*  This code makes very mild assumptions about floating point
*  arithmetic. It will work on machines with a guard digit in
*  add/subtract, or on those binary machines without guard digits
*  which subtract like the Cray XMP, Cray YMP, Cray C 90, or Cray 2.
*  It could conceivably fail on hexadecimal or decimal machines
*  without guard digits, but we know of none.
*
*  Arguments
*  =========
*
*  UPLO   (input) CHARACTER*1
*         = 'U': D and E define an upper bidiagonal matrix.
*         = 'L': D and E define a  lower bidiagonal matrix.
*
*  SMLSIZ (input) INTEGER
*         The maximum size of the subproblems at the bottom of the
*         computation tree.
*
*  N      (input) INTEGER
*         The dimension of the  bidiagonal matrix.  N >= 0.
*
*  NRHS   (input) INTEGER
*         The number of columns of B. NRHS must be at least 1.
*
*  D      (input/output) REAL array, dimension (N)
*         On entry D contains the main diagonal of the bidiagonal
*         matrix. On exit, if INFO = 0, D contains its singular values.
*
*  E      (input/output) REAL array, dimension (N-1)
*         Contains the super-diagonal entries of the bidiagonal matrix.
*         On exit, E has been destroyed.
*
*  B      (input/output) REAL array, dimension (LDB,NRHS)
*         On input, B contains the right hand sides of the least
*         squares problem. On output, B contains the solution X.
*
*  LDB    (input) INTEGER
*         The leading dimension of B in the calling subprogram.
*         LDB must be at least max(1,N).
*
*  RCOND  (input) REAL
*         The singular values of A less than or equal to RCOND times
*         the largest singular value are treated as zero in solving
*         the least squares problem. If RCOND is negative,
*         machine precision is used instead.
*         For example, if diag(S)*X=B were the least squares problem,
*         where diag(S) is a diagonal matrix of singular values, the
*         solution would be X(i) = B(i) / S(i) if S(i) is greater than
*         RCOND*max(S), and X(i) = 0 if S(i) is less than or equal to
*         RCOND*max(S).
*
*  RANK   (output) INTEGER
*         The number of singular values of A greater than RCOND times
*         the largest singular value.
*
*  WORK   (workspace) REAL array, dimension at least
*         (9*N + 2*N*SMLSIZ + 8*N*NLVL + N*NRHS + (SMLSIZ+1)**2),
*         where NLVL = max(0, INT(log_2 (N/(SMLSIZ+1))) + 1).
*
*  IWORK  (workspace) INTEGER array, dimension at least
*         (3*N*NLVL + 11*N)
*
*  INFO   (output) INTEGER
*         = 0:  successful exit.
*         < 0:  if INFO = -i, the i-th argument had an illegal value.
*         > 0:  The algorithm failed to compute a singular value while
*               working on the submatrix lying in rows and columns
*               INFO/(N+1) through MOD(INFO,N+1).
*
*  Further Details
*  ===============
*
*  Based on contributions by
*     Ming Gu and Ren-Cang Li, Computer Science Division, University of
*       California at Berkeley, USA
*     Osni Marques, LBNL/NERSC, USA
*
*  =====================================================================
*
*     .. Parameters ..
      REAL               ZERO, ONE, TWO
      PARAMETER          ( ZERO = 0.0E0, ONE = 1.0E0, TWO = 2.0E0 )
*     ..
*     .. Local Scalars ..
      INTEGER            BX, BXST, C, DIFL, DIFR, GIVCOL, GIVNUM,
     $                   GIVPTR, I, ICMPQ1, ICMPQ2, IWK, J, K, NLVL,
     $                   NM1, NSIZE, NSUB, NWORK, PERM, POLES, S, SIZEI,
     $                   SMLSZP, SQRE, ST, ST1, U, VT, Z
      REAL               CS, EPS, ORGNRM, R, RCND, SN, TOL
*     ..
*     .. External Functions ..
      INTEGER            ISAMAX
      REAL               SLAMCH, SLANST
      EXTERNAL           ISAMAX, SLAMCH, SLANST
*     ..
*     .. External Subroutines ..
      EXTERNAL           SCOPY, SGEMM, SLACPY, SLALSA, SLARTG, SLASCL,
     $                   SLASDA, SLASDQ, SLASET, SLASRT, SROT, XERBLA
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          ABSINTLOG, REAL, SIGN
*     ..
*     .. Executable Statements ..
*
*     Test the input parameters.
*
      INFO = 0
*
      IF( N.LT.0 ) THEN
         INFO = -3
      ELSE IF( NRHS.LT.1 ) THEN
         INFO = -4
      ELSE IF( ( LDB.LT.1 ) .OR. ( LDB.LT.N ) ) THEN
         INFO = -8
      END IF
      IF( INFO.NE.0 ) THEN
         CALL XERBLA( 'SLALSD'-INFO )
         RETURN
      END IF
*
      EPS = SLAMCH( 'Epsilon' )
*
*     Set up the tolerance.
*
      IF( ( RCOND.LE.ZERO ) .OR. ( RCOND.GE.ONE ) ) THEN
         RCND = EPS
      ELSE
         RCND = RCOND
      END IF
*
      RANK = 0
*
*     Quick return if possible.
*
      IF( N.EQ.0 ) THEN
         RETURN
      ELSE IF( N.EQ.1 ) THEN
         IF( D( 1 ).EQ.ZERO ) THEN
            CALL SLASET( 'A'1, NRHS, ZERO, ZERO, B, LDB )
         ELSE
            RANK = 1
            CALL SLASCL( 'G'00, D( 1 ), ONE, 1, NRHS, B, LDB, INFO )
            D( 1 ) = ABS( D( 1 ) )
         END IF
         RETURN
      END IF
*
*     Rotate the matrix if it is lower bidiagonal.
*
      IF( UPLO.EQ.'L' ) THEN
         DO 10 I = 1, N - 1
            CALL SLARTG( D( I ), E( I ), CS, SN, R )
            D( I ) = R
            E( I ) = SN*D( I+1 )
            D( I+1 ) = CS*D( I+1 )
            IF( NRHS.EQ.1 ) THEN
               CALL SROT( 1, B( I, 1 ), 1, B( I+11 ), 1, CS, SN )
            ELSE
               WORK( I*2-1 ) = CS
               WORK( I*2 ) = SN
            END IF
   10    CONTINUE
         IF( NRHS.GT.1 ) THEN
            DO 30 I = 1, NRHS
               DO 20 J = 1, N - 1
                  CS = WORK( J*2-1 )
                  SN = WORK( J*2 )
                  CALL SROT( 1, B( J, I ), 1, B( J+1, I ), 1, CS, SN )
   20          CONTINUE
   30       CONTINUE
         END IF
      END IF
*
*     Scale.
*
      NM1 = N - 1
      ORGNRM = SLANST( 'M', N, D, E )
      IF( ORGNRM.EQ.ZERO ) THEN
         CALL SLASET( 'A', N, NRHS, ZERO, ZERO, B, LDB )
         RETURN
      END IF
*
      CALL SLASCL( 'G'00, ORGNRM, ONE, N, 1, D, N, INFO )
      CALL SLASCL( 'G'00, ORGNRM, ONE, NM1, 1, E, NM1, INFO )
*
*     If N is smaller than the minimum divide size SMLSIZ, then solve
*     the problem with another solver.
*
      IF( N.LE.SMLSIZ ) THEN
         NWORK = 1 + N*N
         CALL SLASET( 'A', N, N, ZERO, ONE, WORK, N )
         CALL SLASDQ( 'U'0, N, N, 0, NRHS, D, E, WORK, N, WORK, N, B,
     $                LDB, WORK( NWORK ), INFO )
         IF( INFO.NE.0 ) THEN
            RETURN
         END IF
         TOL = RCND*ABS( D( ISAMAX( N, D, 1 ) ) )
         DO 40 I = 1, N
            IF( D( I ).LE.TOL ) THEN
               CALL SLASET( 'A'1, NRHS, ZERO, ZERO, B( I, 1 ), LDB )
            ELSE
               CALL SLASCL( 'G'00, D( I ), ONE, 1, NRHS, B( I, 1 ),
     $                      LDB, INFO )
               RANK = RANK + 1
            END IF
   40    CONTINUE
         CALL SGEMM( 'T''N', N, NRHS, N, ONE, WORK, N, B, LDB, ZERO,
     $               WORK( NWORK ), N )
         CALL SLACPY( 'A', N, NRHS, WORK( NWORK ), N, B, LDB )
*
*        Unscale.
*
         CALL SLASCL( 'G'00, ONE, ORGNRM, N, 1, D, N, INFO )
         CALL SLASRT( 'D', N, D, INFO )
         CALL SLASCL( 'G'00, ORGNRM, ONE, N, NRHS, B, LDB, INFO )
*
         RETURN
      END IF
*
*     Book-keeping and setting up some constants.
*
      NLVL = INTLOGREAL( N ) / REAL( SMLSIZ+1 ) ) / LOG( TWO ) ) + 1
*
      SMLSZP = SMLSIZ + 1
*
      U = 1
      VT = 1 + SMLSIZ*N
      DIFL = VT + SMLSZP*N
      DIFR = DIFL + NLVL*N
      Z = DIFR + NLVL*N*2
      C = Z + NLVL*N
      S = C + N
      POLES = S + N
      GIVNUM = POLES + 2*NLVL*N
      BX = GIVNUM + 2*NLVL*N
      NWORK = BX + N*NRHS
*
      SIZEI = 1 + N
      K = SIZEI + N
      GIVPTR = K + N
      PERM = GIVPTR + N
      GIVCOL = PERM + NLVL*N
      IWK = GIVCOL + NLVL*N*2
*
      ST = 1
      SQRE = 0
      ICMPQ1 = 1
      ICMPQ2 = 0
      NSUB = 0
*
      DO 50 I = 1, N
         IFABS( D( I ) ).LT.EPS ) THEN
            D( I ) = SIGN( EPS, D( I ) )
         END IF
   50 CONTINUE
*
      DO 60 I = 1, NM1
         IF( ( ABS( E( I ) ).LT.EPS ) .OR. ( I.EQ.NM1 ) ) THEN
            NSUB = NSUB + 1
            IWORK( NSUB ) = ST
*
*           Subproblem found. First determine its size and then
*           apply divide and conquer on it.
*
            IF( I.LT.NM1 ) THEN
*
*              A subproblem with E(I) small for I < NM1.
*
               NSIZE = I - ST + 1
               IWORK( SIZEI+NSUB-1 ) = NSIZE
            ELSE IFABS( E( I ) ).GE.EPS ) THEN
*
*              A subproblem with E(NM1) not too small but I = NM1.
*
               NSIZE = N - ST + 1
               IWORK( SIZEI+NSUB-1 ) = NSIZE
            ELSE
*
*              A subproblem with E(NM1) small. This implies an
*              1-by-1 subproblem at D(N), which is not solved
*              explicitly.
*
               NSIZE = I - ST + 1
               IWORK( SIZEI+NSUB-1 ) = NSIZE
               NSUB = NSUB + 1
               IWORK( NSUB ) = N
               IWORK( SIZEI+NSUB-1 ) = 1
               CALL SCOPY( NRHS, B( N, 1 ), LDB, WORK( BX+NM1 ), N )
            END IF
            ST1 = ST - 1
            IF( NSIZE.EQ.1 ) THEN
*
*              This is a 1-by-1 subproblem and is not solved
*              explicitly.
*
               CALL SCOPY( NRHS, B( ST, 1 ), LDB, WORK( BX+ST1 ), N )
            ELSE IF( NSIZE.LE.SMLSIZ ) THEN
*
*              This is a small subproblem and is solved by SLASDQ.
*
               CALL SLASET( 'A', NSIZE, NSIZE, ZERO, ONE,
     $                      WORK( VT+ST1 ), N )
               CALL SLASDQ( 'U'0, NSIZE, NSIZE, 0, NRHS, D( ST ),
     $                      E( ST ), WORK( VT+ST1 ), N, WORK( NWORK ),
     $                      N, B( ST, 1 ), LDB, WORK( NWORK ), INFO )
               IF( INFO.NE.0 ) THEN
                  RETURN
               END IF
               CALL SLACPY( 'A', NSIZE, NRHS, B( ST, 1 ), LDB,
     $                      WORK( BX+ST1 ), N )
            ELSE
*
*              A large problem. Solve it using divide and conquer.
*
               CALL SLASDA( ICMPQ1, SMLSIZ, NSIZE, SQRE, D( ST ),
     $                      E( ST ), WORK( U+ST1 ), N, WORK( VT+ST1 ),
     $                      IWORK( K+ST1 ), WORK( DIFL+ST1 ),
     $                      WORK( DIFR+ST1 ), WORK( Z+ST1 ),
     $                      WORK( POLES+ST1 ), IWORK( GIVPTR+ST1 ),
     $                      IWORK( GIVCOL+ST1 ), N, IWORK( PERM+ST1 ),
     $                      WORK( GIVNUM+ST1 ), WORK( C+ST1 ),
     $                      WORK( S+ST1 ), WORK( NWORK ), IWORK( IWK ),
     $                      INFO )
               IF( INFO.NE.0 ) THEN
                  RETURN
               END IF
               BXST = BX + ST1
               CALL SLALSA( ICMPQ2, SMLSIZ, NSIZE, NRHS, B( ST, 1 ),
     $                      LDB, WORK( BXST ), N, WORK( U+ST1 ), N,
     $                      WORK( VT+ST1 ), IWORK( K+ST1 ),
     $                      WORK( DIFL+ST1 ), WORK( DIFR+ST1 ),
     $                      WORK( Z+ST1 ), WORK( POLES+ST1 ),
     $                      IWORK( GIVPTR+ST1 ), IWORK( GIVCOL+ST1 ), N,
     $                      IWORK( PERM+ST1 ), WORK( GIVNUM+ST1 ),
     $                      WORK( C+ST1 ), WORK( S+ST1 ), WORK( NWORK ),
     $                      IWORK( IWK ), INFO )
               IF( INFO.NE.0 ) THEN
                  RETURN
               END IF
            END IF
            ST = I + 1
         END IF
   60 CONTINUE
*
*     Apply the singular values and treat the tiny ones as zero.
*
      TOL = RCND*ABS( D( ISAMAX( N, D, 1 ) ) )
*
      DO 70 I = 1, N
*
*        Some of the elements in D can be negative because 1-by-1
*        subproblems were not solved explicitly.
*
         IFABS( D( I ) ).LE.TOL ) THEN
            CALL SLASET( 'A'1, NRHS, ZERO, ZERO, WORK( BX+I-1 ), N )
         ELSE
            RANK = RANK + 1
            CALL SLASCL( 'G'00, D( I ), ONE, 1, NRHS,
     $                   WORK( BX+I-1 ), N, INFO )
         END IF
         D( I ) = ABS( D( I ) )
   70 CONTINUE
*
*     Now apply back the right singular vectors.
*
      ICMPQ2 = 1
      DO 80 I = 1, NSUB
         ST = IWORK( I )
         ST1 = ST - 1
         NSIZE = IWORK( SIZEI+I-1 )
         BXST = BX + ST1
         IF( NSIZE.EQ.1 ) THEN
            CALL SCOPY( NRHS, WORK( BXST ), N, B( ST, 1 ), LDB )
         ELSE IF( NSIZE.LE.SMLSIZ ) THEN
            CALL SGEMM( 'T''N', NSIZE, NRHS, NSIZE, ONE,
     $                  WORK( VT+ST1 ), N, WORK( BXST ), N, ZERO,
     $                  B( ST, 1 ), LDB )
         ELSE
            CALL SLALSA( ICMPQ2, SMLSIZ, NSIZE, NRHS, WORK( BXST ), N,
     $                   B( ST, 1 ), LDB, WORK( U+ST1 ), N,
     $                   WORK( VT+ST1 ), IWORK( K+ST1 ),
     $                   WORK( DIFL+ST1 ), WORK( DIFR+ST1 ),
     $                   WORK( Z+ST1 ), WORK( POLES+ST1 ),
     $                   IWORK( GIVPTR+ST1 ), IWORK( GIVCOL+ST1 ), N,
     $                   IWORK( PERM+ST1 ), WORK( GIVNUM+ST1 ),
     $                   WORK( C+ST1 ), WORK( S+ST1 ), WORK( NWORK ),
     $                   IWORK( IWK ), INFO )
            IF( INFO.NE.0 ) THEN
               RETURN
            END IF
         END IF
   80 CONTINUE
*
*     Unscale and sort the singular values.
*
      CALL SLASCL( 'G'00, ONE, ORGNRM, N, 1, D, N, INFO )
      CALL SLASRT( 'D', N, D, INFO )
      CALL SLASCL( 'G'00, ORGNRM, ONE, N, NRHS, B, LDB, INFO )
*
      RETURN
*
*     End of SLALSD
*
      END