1       SUBROUTINE DLAHR2( N, K, NB, A, LDA, TAU, T, LDT, Y, LDY )
  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 2009                                                      --
  7 *
  8 *     .. Scalar Arguments ..
  9       INTEGER            K, LDA, LDT, LDY, N, NB
 10 *     ..
 11 *     .. Array Arguments ..
 12       DOUBLE PRECISION  A( LDA, * ), T( LDT, NB ), TAU( NB ),
 13      $                   Y( LDY, NB )
 14 *     ..
 15 *
 16 *  Purpose
 17 *  =======
 18 *
 19 *  DLAHR2 reduces the first NB columns of A real general n-BY-(n-k+1)
 20 *  matrix A so that elements below the k-th subdiagonal are zero. The
 21 *  reduction is performed by an orthogonal similarity transformation
 22 *  Q**T * A * Q. The routine returns the matrices V and T which determine
 23 *  Q as a block reflector I - V*T*V**T, and also the matrix Y = A * V * T.
 24 *
 25 *  This is an auxiliary routine called by DGEHRD.
 26 *
 27 *  Arguments
 28 *  =========
 29 *
 30 *  N       (input) INTEGER
 31 *          The order of the matrix A.
 32 *
 33 *  K       (input) INTEGER
 34 *          The offset for the reduction. Elements below the k-th
 35 *          subdiagonal in the first NB columns are reduced to zero.
 36 *          K < N.
 37 *
 38 *  NB      (input) INTEGER
 39 *          The number of columns to be reduced.
 40 *
 41 *  A       (input/output) DOUBLE PRECISION array, dimension (LDA,N-K+1)
 42 *          On entry, the n-by-(n-k+1) general matrix A.
 43 *          On exit, the elements on and above the k-th subdiagonal in
 44 *          the first NB columns are overwritten with the corresponding
 45 *          elements of the reduced matrix; the elements below the k-th
 46 *          subdiagonal, with the array TAU, represent the matrix Q as a
 47 *          product of elementary reflectors. The other columns of A are
 48 *          unchanged. See Further Details.
 49 *
 50 *  LDA     (input) INTEGER
 51 *          The leading dimension of the array A.  LDA >= max(1,N).
 52 *
 53 *  TAU     (output) DOUBLE PRECISION array, dimension (NB)
 54 *          The scalar factors of the elementary reflectors. See Further
 55 *          Details.
 56 *
 57 *  T       (output) DOUBLE PRECISION array, dimension (LDT,NB)
 58 *          The upper triangular matrix T.
 59 *
 60 *  LDT     (input) INTEGER
 61 *          The leading dimension of the array T.  LDT >= NB.
 62 *
 63 *  Y       (output) DOUBLE PRECISION array, dimension (LDY,NB)
 64 *          The n-by-nb matrix Y.
 65 *
 66 *  LDY     (input) INTEGER
 67 *          The leading dimension of the array Y. LDY >= N.
 68 *
 69 *  Further Details
 70 *  ===============
 71 *
 72 *  The matrix Q is represented as a product of nb elementary reflectors
 73 *
 74 *     Q = H(1) H(2) . . . H(nb).
 75 *
 76 *  Each H(i) has the form
 77 *
 78 *     H(i) = I - tau * v * v**T
 79 *
 80 *  where tau is a real scalar, and v is a real vector with
 81 *  v(1:i+k-1) = 0, v(i+k) = 1; v(i+k+1:n) is stored on exit in
 82 *  A(i+k+1:n,i), and tau in TAU(i).
 83 *
 84 *  The elements of the vectors v together form the (n-k+1)-by-nb matrix
 85 *  V which is needed, with T and Y, to apply the transformation to the
 86 *  unreduced part of the matrix, using an update of the form:
 87 *  A := (I - V*T*V**T) * (A - Y*V**T).
 88 *
 89 *  The contents of A on exit are illustrated by the following example
 90 *  with n = 7, k = 3 and nb = 2:
 91 *
 92 *     ( a   a   a   a   a )
 93 *     ( a   a   a   a   a )
 94 *     ( a   a   a   a   a )
 95 *     ( h   h   a   a   a )
 96 *     ( v1  h   a   a   a )
 97 *     ( v1  v2  a   a   a )
 98 *     ( v1  v2  a   a   a )
 99 *
100 *  where a denotes an element of the original matrix A, h denotes a
101 *  modified element of the upper Hessenberg matrix H, and vi denotes an
102 *  element of the vector defining H(i).
103 *
104 *  This subroutine is a slight modification of LAPACK-3.0's DLAHRD
105 *  incorporating improvements proposed by Quintana-Orti and Van de
106 *  Gejin. Note that the entries of A(1:K,2:NB) differ from those
107 *  returned by the original LAPACK-3.0's DLAHRD routine. (This
108 *  subroutine is not backward compatible with LAPACK-3.0's DLAHRD.)
109 *
110 *  References
111 *  ==========
112 *
113 *  Gregorio Quintana-Orti and Robert van de Geijn, "Improving the
114 *  performance of reduction to Hessenberg form," ACM Transactions on
115 *  Mathematical Software, 32(2):180-194, June 2006.
116 *
117 *  =====================================================================
118 *
119 *     .. Parameters ..
120       DOUBLE PRECISION  ZERO, ONE
121       PARAMETER          ( ZERO = 0.0D+0
122      $                     ONE = 1.0D+0 )
123 *     ..
124 *     .. Local Scalars ..
125       INTEGER            I
126       DOUBLE PRECISION  EI
127 *     ..
128 *     .. External Subroutines ..
129       EXTERNAL           DAXPY, DCOPY, DGEMM, DGEMV, DLACPY,
130      $                   DLARFG, DSCAL, DTRMM, DTRMV
131 *     ..
132 *     .. Intrinsic Functions ..
133       INTRINSIC          MIN
134 *     ..
135 *     .. Executable Statements ..
136 *
137 *     Quick return if possible
138 *
139       IF( N.LE.1 )
140      $   RETURN
141 *
142       DO 10 I = 1, NB
143          IF( I.GT.1 ) THEN
144 *
145 *           Update A(K+1:N,I)
146 *
147 *           Update I-th column of A - Y * V**T
148 *
149             CALL DGEMV( 'NO TRANSPOSE', N-K, I-1-ONE, Y(K+1,1), LDY,
150      $                  A( K+I-11 ), LDA, ONE, A( K+1, I ), 1 )
151 *
152 *           Apply I - V * T**T * V**T to this column (call it b) from the
153 *           left, using the last column of T as workspace
154 *
155 *           Let  V = ( V1 )   and   b = ( b1 )   (first I-1 rows)
156 *                    ( V2 )             ( b2 )
157 *
158 *           where V1 is unit lower triangular
159 *
160 *           w := V1**T * b1
161 *
162             CALL DCOPY( I-1, A( K+1, I ), 1, T( 1, NB ), 1 )
163             CALL DTRMV( 'Lower''Transpose''UNIT'
164      $                  I-1, A( K+11 ),
165      $                  LDA, T( 1, NB ), 1 )
166 *
167 *           w := w + V2**T * b2
168 *
169             CALL DGEMV( 'Transpose', N-K-I+1, I-1
170      $                  ONE, A( K+I, 1 ),
171      $                  LDA, A( K+I, I ), 1, ONE, T( 1, NB ), 1 )
172 *
173 *           w := T**T * w
174 *
175             CALL DTRMV( 'Upper''Transpose''NON-UNIT'
176      $                  I-1, T, LDT,
177      $                  T( 1, NB ), 1 )
178 *
179 *           b2 := b2 - V2*w
180 *
181             CALL DGEMV( 'NO TRANSPOSE', N-K-I+1, I-1-ONE, 
182      $                  A( K+I, 1 ),
183      $                  LDA, T( 1, NB ), 1, ONE, A( K+I, I ), 1 )
184 *
185 *           b1 := b1 - V1*w
186 *
187             CALL DTRMV( 'Lower''NO TRANSPOSE'
188      $                  'UNIT', I-1,
189      $                  A( K+11 ), LDA, T( 1, NB ), 1 )
190             CALL DAXPY( I-1-ONE, T( 1, NB ), 1, A( K+1, I ), 1 )
191 *
192             A( K+I-1, I-1 ) = EI
193          END IF
194 *
195 *        Generate the elementary reflector H(I) to annihilate
196 *        A(K+I+1:N,I)
197 *
198          CALL DLARFG( N-K-I+1, A( K+I, I ), A( MIN( K+I+1, N ), I ), 1,
199      $                TAU( I ) )
200          EI = A( K+I, I )
201          A( K+I, I ) = ONE
202 *
203 *        Compute  Y(K+1:N,I)
204 *
205          CALL DGEMV( 'NO TRANSPOSE', N-K, N-K-I+1
206      $               ONE, A( K+1, I+1 ),
207      $               LDA, A( K+I, I ), 1, ZERO, Y( K+1, I ), 1 )
208          CALL DGEMV( 'Transpose', N-K-I+1, I-1
209      $               ONE, A( K+I, 1 ), LDA,
210      $               A( K+I, I ), 1, ZERO, T( 1, I ), 1 )
211          CALL DGEMV( 'NO TRANSPOSE', N-K, I-1-ONE, 
212      $               Y( K+11 ), LDY,
213      $               T( 1, I ), 1, ONE, Y( K+1, I ), 1 )
214          CALL DSCAL( N-K, TAU( I ), Y( K+1, I ), 1 )
215 *
216 *        Compute T(1:I,I)
217 *
218          CALL DSCAL( I-1-TAU( I ), T( 1, I ), 1 )
219          CALL DTRMV( 'Upper''No Transpose''NON-UNIT'
220      $               I-1, T, LDT,
221      $               T( 1, I ), 1 )
222          T( I, I ) = TAU( I )
223 *
224    10 CONTINUE
225       A( K+NB, NB ) = EI
226 *
227 *     Compute Y(1:K,1:NB)
228 *
229       CALL DLACPY( 'ALL', K, NB, A( 12 ), LDA, Y, LDY )
230       CALL DTRMM( 'RIGHT''Lower''NO TRANSPOSE'
231      $            'UNIT', K, NB,
232      $            ONE, A( K+11 ), LDA, Y, LDY )
233       IF( N.GT.K+NB )
234      $   CALL DGEMM( 'NO TRANSPOSE''NO TRANSPOSE', K, 
235      $               NB, N-K-NB, ONE,
236      $               A( 12+NB ), LDA, A( K+1+NB, 1 ), LDA, ONE, Y,
237      $               LDY )
238       CALL DTRMM( 'RIGHT''Upper''NO TRANSPOSE'
239      $            'NON-UNIT', K, NB,
240      $            ONE, T, LDT, Y, LDY )
241 *
242       RETURN
243 *
244 *     End of DLAHR2
245 *
246       END