1       SUBROUTINE DLATRD( UPLO, N, NB, A, LDA, E, TAU, W, LDW )
  2 *
  3 *  -- LAPACK auxiliary routine (version 3.3.1) --
  4 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
  5 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
  6 *  -- April 2011                                                      --
  7 *
  8 *     .. Scalar Arguments ..
  9       CHARACTER          UPLO
 10       INTEGER            LDA, LDW, N, NB
 11 *     ..
 12 *     .. Array Arguments ..
 13       DOUBLE PRECISION   A( LDA, * ), E( * ), TAU( * ), W( LDW, * )
 14 *     ..
 15 *
 16 *  Purpose
 17 *  =======
 18 *
 19 *  DLATRD reduces NB rows and columns of a real symmetric matrix A to
 20 *  symmetric tridiagonal form by an orthogonal similarity
 21 *  transformation Q**T * A * Q, and returns the matrices V and W which are
 22 *  needed to apply the transformation to the unreduced part of A.
 23 *
 24 *  If UPLO = 'U', DLATRD reduces the last NB rows and columns of a
 25 *  matrix, of which the upper triangle is supplied;
 26 *  if UPLO = 'L', DLATRD reduces the first NB rows and columns of a
 27 *  matrix, of which the lower triangle is supplied.
 28 *
 29 *  This is an auxiliary routine called by DSYTRD.
 30 *
 31 *  Arguments
 32 *  =========
 33 *
 34 *  UPLO    (input) CHARACTER*1
 35 *          Specifies whether the upper or lower triangular part of the
 36 *          symmetric matrix A is stored:
 37 *          = 'U': Upper triangular
 38 *          = 'L': Lower triangular
 39 *
 40 *  N       (input) INTEGER
 41 *          The order of the matrix A.
 42 *
 43 *  NB      (input) INTEGER
 44 *          The number of rows and columns to be reduced.
 45 *
 46 *  A       (input/output) DOUBLE PRECISION array, dimension (LDA,N)
 47 *          On entry, the symmetric matrix A.  If UPLO = 'U', the leading
 48 *          n-by-n upper triangular part of A contains the upper
 49 *          triangular part of the matrix A, and the strictly lower
 50 *          triangular part of A is not referenced.  If UPLO = 'L', the
 51 *          leading n-by-n lower triangular part of A contains the lower
 52 *          triangular part of the matrix A, and the strictly upper
 53 *          triangular part of A is not referenced.
 54 *          On exit:
 55 *          if UPLO = 'U', the last NB columns have been reduced to
 56 *            tridiagonal form, with the diagonal elements overwriting
 57 *            the diagonal elements of A; the elements above the diagonal
 58 *            with the array TAU, represent the orthogonal matrix Q as a
 59 *            product of elementary reflectors;
 60 *          if UPLO = 'L', the first NB columns have been reduced to
 61 *            tridiagonal form, with the diagonal elements overwriting
 62 *            the diagonal elements of A; the elements below the diagonal
 63 *            with the array TAU, represent the  orthogonal matrix Q as a
 64 *            product of elementary reflectors.
 65 *          See Further Details.
 66 *
 67 *  LDA     (input) INTEGER
 68 *          The leading dimension of the array A.  LDA >= (1,N).
 69 *
 70 *  E       (output) DOUBLE PRECISION array, dimension (N-1)
 71 *          If UPLO = 'U', E(n-nb:n-1) contains the superdiagonal
 72 *          elements of the last NB columns of the reduced matrix;
 73 *          if UPLO = 'L', E(1:nb) contains the subdiagonal elements of
 74 *          the first NB columns of the reduced matrix.
 75 *
 76 *  TAU     (output) DOUBLE PRECISION array, dimension (N-1)
 77 *          The scalar factors of the elementary reflectors, stored in
 78 *          TAU(n-nb:n-1) if UPLO = 'U', and in TAU(1:nb) if UPLO = 'L'.
 79 *          See Further Details.
 80 *
 81 *  W       (output) DOUBLE PRECISION array, dimension (LDW,NB)
 82 *          The n-by-nb matrix W required to update the unreduced part
 83 *          of A.
 84 *
 85 *  LDW     (input) INTEGER
 86 *          The leading dimension of the array W. LDW >= max(1,N).
 87 *
 88 *  Further Details
 89 *  ===============
 90 *
 91 *  If UPLO = 'U', the matrix Q is represented as a product of elementary
 92 *  reflectors
 93 *
 94 *     Q = H(n) H(n-1) . . . H(n-nb+1).
 95 *
 96 *  Each H(i) has the form
 97 *
 98 *     H(i) = I - tau * v * v**T
 99 *
100 *  where tau is a real scalar, and v is a real vector with
101 *  v(i:n) = 0 and v(i-1) = 1; v(1:i-1) is stored on exit in A(1:i-1,i),
102 *  and tau in TAU(i-1).
103 *
104 *  If UPLO = 'L', the matrix Q is represented as a product of elementary
105 *  reflectors
106 *
107 *     Q = H(1) H(2) . . . H(nb).
108 *
109 *  Each H(i) has the form
110 *
111 *     H(i) = I - tau * v * v**T
112 *
113 *  where tau is a real scalar, and v is a real vector with
114 *  v(1:i) = 0 and v(i+1) = 1; v(i+1:n) is stored on exit in A(i+1:n,i),
115 *  and tau in TAU(i).
116 *
117 *  The elements of the vectors v together form the n-by-nb matrix V
118 *  which is needed, with W, to apply the transformation to the unreduced
119 *  part of the matrix, using a symmetric rank-2k update of the form:
120 *  A := A - V*W**T - W*V**T.
121 *
122 *  The contents of A on exit are illustrated by the following examples
123 *  with n = 5 and nb = 2:
124 *
125 *  if UPLO = 'U':                       if UPLO = 'L':
126 *
127 *    (  a   a   a   v4  v5 )              (  d                  )
128 *    (      a   a   v4  v5 )              (  1   d              )
129 *    (          a   1   v5 )              (  v1  1   a          )
130 *    (              d   1  )              (  v1  v2  a   a      )
131 *    (                  d  )              (  v1  v2  a   a   a  )
132 *
133 *  where d denotes a diagonal element of the reduced matrix, a denotes
134 *  an element of the original matrix that is unchanged, and vi denotes
135 *  an element of the vector defining H(i).
136 *
137 *  =====================================================================
138 *
139 *     .. Parameters ..
140       DOUBLE PRECISION   ZERO, ONE, HALF
141       PARAMETER          ( ZERO = 0.0D+0, ONE = 1.0D+0, HALF = 0.5D+0 )
142 *     ..
143 *     .. Local Scalars ..
144       INTEGER            I, IW
145       DOUBLE PRECISION   ALPHA
146 *     ..
147 *     .. External Subroutines ..
148       EXTERNAL           DAXPY, DGEMV, DLARFG, DSCAL, DSYMV
149 *     ..
150 *     .. External Functions ..
151       LOGICAL            LSAME
152       DOUBLE PRECISION   DDOT
153       EXTERNAL           LSAME, DDOT
154 *     ..
155 *     .. Intrinsic Functions ..
156       INTRINSIC          MIN
157 *     ..
158 *     .. Executable Statements ..
159 *
160 *     Quick return if possible
161 *
162       IF( N.LE.0 )
163      $   RETURN
164 *
165       IF( LSAME( UPLO, 'U' ) ) THEN
166 *
167 *        Reduce last NB columns of upper triangle
168 *
169          DO 10 I = N, N - NB + 1-1
170             IW = I - N + NB
171             IF( I.LT.N ) THEN
172 *
173 *              Update A(1:i,i)
174 *
175                CALL DGEMV( 'No transpose', I, N-I, -ONE, A( 1, I+1 ),
176      $                     LDA, W( I, IW+1 ), LDW, ONE, A( 1, I ), 1 )
177                CALL DGEMV( 'No transpose', I, N-I, -ONE, W( 1, IW+1 ),
178      $                     LDW, A( I, I+1 ), LDA, ONE, A( 1, I ), 1 )
179             END IF
180             IF( I.GT.1 ) THEN
181 *
182 *              Generate elementary reflector H(i) to annihilate
183 *              A(1:i-2,i)
184 *
185                CALL DLARFG( I-1, A( I-1, I ), A( 1, I ), 1, TAU( I-1 ) )
186                E( I-1 ) = A( I-1, I )
187                A( I-1, I ) = ONE
188 *
189 *              Compute W(1:i-1,i)
190 *
191                CALL DSYMV( 'Upper', I-1, ONE, A, LDA, A( 1, I ), 1,
192      $                     ZERO, W( 1, IW ), 1 )
193                IF( I.LT.N ) THEN
194                   CALL DGEMV( 'Transpose', I-1, N-I, ONE, W( 1, IW+1 ),
195      $                        LDW, A( 1, I ), 1, ZERO, W( I+1, IW ), 1 )
196                   CALL DGEMV( 'No transpose', I-1, N-I, -ONE,
197      $                        A( 1, I+1 ), LDA, W( I+1, IW ), 1, ONE,
198      $                        W( 1, IW ), 1 )
199                   CALL DGEMV( 'Transpose', I-1, N-I, ONE, A( 1, I+1 ),
200      $                        LDA, A( 1, I ), 1, ZERO, W( I+1, IW ), 1 )
201                   CALL DGEMV( 'No transpose', I-1, N-I, -ONE,
202      $                        W( 1, IW+1 ), LDW, W( I+1, IW ), 1, ONE,
203      $                        W( 1, IW ), 1 )
204                END IF
205                CALL DSCAL( I-1, TAU( I-1 ), W( 1, IW ), 1 )
206                ALPHA = -HALF*TAU( I-1 )*DDOT( I-1, W( 1, IW ), 1,
207      $                 A( 1, I ), 1 )
208                CALL DAXPY( I-1, ALPHA, A( 1, I ), 1, W( 1, IW ), 1 )
209             END IF
210 *
211    10    CONTINUE
212       ELSE
213 *
214 *        Reduce first NB columns of lower triangle
215 *
216          DO 20 I = 1, NB
217 *
218 *           Update A(i:n,i)
219 *
220             CALL DGEMV( 'No transpose', N-I+1, I-1-ONE, A( I, 1 ),
221      $                  LDA, W( I, 1 ), LDW, ONE, A( I, I ), 1 )
222             CALL DGEMV( 'No transpose', N-I+1, I-1-ONE, W( I, 1 ),
223      $                  LDW, A( I, 1 ), LDA, ONE, A( I, I ), 1 )
224             IF( I.LT.N ) THEN
225 *
226 *              Generate elementary reflector H(i) to annihilate
227 *              A(i+2:n,i)
228 *
229                CALL DLARFG( N-I, A( I+1, I ), A( MIN( I+2, N ), I ), 1,
230      $                      TAU( I ) )
231                E( I ) = A( I+1, I )
232                A( I+1, I ) = ONE
233 *
234 *              Compute W(i+1:n,i)
235 *
236                CALL DSYMV( 'Lower', N-I, ONE, A( I+1, I+1 ), LDA,
237      $                     A( I+1, I ), 1, ZERO, W( I+1, I ), 1 )
238                CALL DGEMV( 'Transpose', N-I, I-1, ONE, W( I+11 ), LDW,
239      $                     A( I+1, I ), 1, ZERO, W( 1, I ), 1 )
240                CALL DGEMV( 'No transpose', N-I, I-1-ONE, A( I+11 ),
241      $                     LDA, W( 1, I ), 1, ONE, W( I+1, I ), 1 )
242                CALL DGEMV( 'Transpose', N-I, I-1, ONE, A( I+11 ), LDA,
243      $                     A( I+1, I ), 1, ZERO, W( 1, I ), 1 )
244                CALL DGEMV( 'No transpose', N-I, I-1-ONE, W( I+11 ),
245      $                     LDW, W( 1, I ), 1, ONE, W( I+1, I ), 1 )
246                CALL DSCAL( N-I, TAU( I ), W( I+1, I ), 1 )
247                ALPHA = -HALF*TAU( I )*DDOT( N-I, W( I+1, I ), 1,
248      $                 A( I+1, I ), 1 )
249                CALL DAXPY( N-I, ALPHA, A( I+1, I ), 1, W( I+1, I ), 1 )
250             END IF
251 *
252    20    CONTINUE
253       END IF
254 *
255       RETURN
256 *
257 *     End of DLATRD
258 *
259       END