LAPACK 3.3.1
Linear Algebra PACKage

dlaqps.f

Go to the documentation of this file.
00001       SUBROUTINE DLAQPS( M, N, OFFSET, NB, KB, A, LDA, JPVT, TAU, VN1,
00002      $                   VN2, AUXV, F, LDF )
00003 *
00004 *  -- LAPACK auxiliary routine (version 3.3.1) --
00005 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
00006 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
00007 *  -- April 2011                                                      --
00008 *
00009 *     .. Scalar Arguments ..
00010       INTEGER            KB, LDA, LDF, M, N, NB, OFFSET
00011 *     ..
00012 *     .. Array Arguments ..
00013       INTEGER            JPVT( * )
00014       DOUBLE PRECISION   A( LDA, * ), AUXV( * ), F( LDF, * ), TAU( * ),
00015      $                   VN1( * ), VN2( * )
00016 *     ..
00017 *
00018 *  Purpose
00019 *  =======
00020 *
00021 *  DLAQPS computes a step of QR factorization with column pivoting
00022 *  of a real M-by-N matrix A by using Blas-3.  It tries to factorize
00023 *  NB columns from A starting from the row OFFSET+1, and updates all
00024 *  of the matrix with Blas-3 xGEMM.
00025 *
00026 *  In some cases, due to catastrophic cancellations, it cannot
00027 *  factorize NB columns.  Hence, the actual number of factorized
00028 *  columns is returned in KB.
00029 *
00030 *  Block A(1:OFFSET,1:N) is accordingly pivoted, but not factorized.
00031 *
00032 *  Arguments
00033 *  =========
00034 *
00035 *  M       (input) INTEGER
00036 *          The number of rows of the matrix A. M >= 0.
00037 *
00038 *  N       (input) INTEGER
00039 *          The number of columns of the matrix A. N >= 0
00040 *
00041 *  OFFSET  (input) INTEGER
00042 *          The number of rows of A that have been factorized in
00043 *          previous steps.
00044 *
00045 *  NB      (input) INTEGER
00046 *          The number of columns to factorize.
00047 *
00048 *  KB      (output) INTEGER
00049 *          The number of columns actually factorized.
00050 *
00051 *  A       (input/output) DOUBLE PRECISION array, dimension (LDA,N)
00052 *          On entry, the M-by-N matrix A.
00053 *          On exit, block A(OFFSET+1:M,1:KB) is the triangular
00054 *          factor obtained and block A(1:OFFSET,1:N) has been
00055 *          accordingly pivoted, but no factorized.
00056 *          The rest of the matrix, block A(OFFSET+1:M,KB+1:N) has
00057 *          been updated.
00058 *
00059 *  LDA     (input) INTEGER
00060 *          The leading dimension of the array A. LDA >= max(1,M).
00061 *
00062 *  JPVT    (input/output) INTEGER array, dimension (N)
00063 *          JPVT(I) = K <==> Column K of the full matrix A has been
00064 *          permuted into position I in AP.
00065 *
00066 *  TAU     (output) DOUBLE PRECISION array, dimension (KB)
00067 *          The scalar factors of the elementary reflectors.
00068 *
00069 *  VN1     (input/output) DOUBLE PRECISION array, dimension (N)
00070 *          The vector with the partial column norms.
00071 *
00072 *  VN2     (input/output) DOUBLE PRECISION array, dimension (N)
00073 *          The vector with the exact column norms.
00074 *
00075 *  AUXV    (input/output) DOUBLE PRECISION array, dimension (NB)
00076 *          Auxiliar vector.
00077 *
00078 *  F       (input/output) DOUBLE PRECISION array, dimension (LDF,NB)
00079 *          Matrix F**T = L*Y**T*A.
00080 *
00081 *  LDF     (input) INTEGER
00082 *          The leading dimension of the array F. LDF >= max(1,N).
00083 *
00084 *  Further Details
00085 *  ===============
00086 *
00087 *  Based on contributions by
00088 *    G. Quintana-Orti, Depto. de Informatica, Universidad Jaime I, Spain
00089 *    X. Sun, Computer Science Dept., Duke University, USA
00090 *
00091 *  Partial column norm updating strategy modified by
00092 *    Z. Drmac and Z. Bujanovic, Dept. of Mathematics,
00093 *    University of Zagreb, Croatia.
00094 *  -- April 2011                                                      --
00095 *  For more details see LAPACK Working Note 176.
00096 *  =====================================================================
00097 *
00098 *     .. Parameters ..
00099       DOUBLE PRECISION   ZERO, ONE
00100       PARAMETER          ( ZERO = 0.0D+0, ONE = 1.0D+0 )
00101 *     ..
00102 *     .. Local Scalars ..
00103       INTEGER            ITEMP, J, K, LASTRK, LSTICC, PVT, RK
00104       DOUBLE PRECISION   AKK, TEMP, TEMP2, TOL3Z
00105 *     ..
00106 *     .. External Subroutines ..
00107       EXTERNAL           DGEMM, DGEMV, DLARFG, DSWAP
00108 *     ..
00109 *     .. Intrinsic Functions ..
00110       INTRINSIC          ABS, DBLE, MAX, MIN, NINT, SQRT
00111 *     ..
00112 *     .. External Functions ..
00113       INTEGER            IDAMAX
00114       DOUBLE PRECISION   DLAMCH, DNRM2
00115       EXTERNAL           IDAMAX, DLAMCH, DNRM2
00116 *     ..
00117 *     .. Executable Statements ..
00118 *
00119       LASTRK = MIN( M, N+OFFSET )
00120       LSTICC = 0
00121       K = 0
00122       TOL3Z = SQRT(DLAMCH('Epsilon'))
00123 *
00124 *     Beginning of while loop.
00125 *
00126    10 CONTINUE
00127       IF( ( K.LT.NB ) .AND. ( LSTICC.EQ.0 ) ) THEN
00128          K = K + 1
00129          RK = OFFSET + K
00130 *
00131 *        Determine ith pivot column and swap if necessary
00132 *
00133          PVT = ( K-1 ) + IDAMAX( N-K+1, VN1( K ), 1 )
00134          IF( PVT.NE.K ) THEN
00135             CALL DSWAP( M, A( 1, PVT ), 1, A( 1, K ), 1 )
00136             CALL DSWAP( K-1, F( PVT, 1 ), LDF, F( K, 1 ), LDF )
00137             ITEMP = JPVT( PVT )
00138             JPVT( PVT ) = JPVT( K )
00139             JPVT( K ) = ITEMP
00140             VN1( PVT ) = VN1( K )
00141             VN2( PVT ) = VN2( K )
00142          END IF
00143 *
00144 *        Apply previous Householder reflectors to column K:
00145 *        A(RK:M,K) := A(RK:M,K) - A(RK:M,1:K-1)*F(K,1:K-1)**T.
00146 *
00147          IF( K.GT.1 ) THEN
00148             CALL DGEMV( 'No transpose', M-RK+1, K-1, -ONE, A( RK, 1 ),
00149      $                  LDA, F( K, 1 ), LDF, ONE, A( RK, K ), 1 )
00150          END IF
00151 *
00152 *        Generate elementary reflector H(k).
00153 *
00154          IF( RK.LT.M ) THEN
00155             CALL DLARFG( M-RK+1, A( RK, K ), A( RK+1, K ), 1, TAU( K ) )
00156          ELSE
00157             CALL DLARFG( 1, A( RK, K ), A( RK, K ), 1, TAU( K ) )
00158          END IF
00159 *
00160          AKK = A( RK, K )
00161          A( RK, K ) = ONE
00162 *
00163 *        Compute Kth column of F:
00164 *
00165 *        Compute  F(K+1:N,K) := tau(K)*A(RK:M,K+1:N)**T*A(RK:M,K).
00166 *
00167          IF( K.LT.N ) THEN
00168             CALL DGEMV( 'Transpose', M-RK+1, N-K, TAU( K ),
00169      $                  A( RK, K+1 ), LDA, A( RK, K ), 1, ZERO,
00170      $                  F( K+1, K ), 1 )
00171          END IF
00172 *
00173 *        Padding F(1:K,K) with zeros.
00174 *
00175          DO 20 J = 1, K
00176             F( J, K ) = ZERO
00177    20    CONTINUE
00178 *
00179 *        Incremental updating of F:
00180 *        F(1:N,K) := F(1:N,K) - tau(K)*F(1:N,1:K-1)*A(RK:M,1:K-1)**T
00181 *                    *A(RK:M,K).
00182 *
00183          IF( K.GT.1 ) THEN
00184             CALL DGEMV( 'Transpose', M-RK+1, K-1, -TAU( K ), A( RK, 1 ),
00185      $                  LDA, A( RK, K ), 1, ZERO, AUXV( 1 ), 1 )
00186 *
00187             CALL DGEMV( 'No transpose', N, K-1, ONE, F( 1, 1 ), LDF,
00188      $                  AUXV( 1 ), 1, ONE, F( 1, K ), 1 )
00189          END IF
00190 *
00191 *        Update the current row of A:
00192 *        A(RK,K+1:N) := A(RK,K+1:N) - A(RK,1:K)*F(K+1:N,1:K)**T.
00193 *
00194          IF( K.LT.N ) THEN
00195             CALL DGEMV( 'No transpose', N-K, K, -ONE, F( K+1, 1 ), LDF,
00196      $                  A( RK, 1 ), LDA, ONE, A( RK, K+1 ), LDA )
00197          END IF
00198 *
00199 *        Update partial column norms.
00200 *
00201          IF( RK.LT.LASTRK ) THEN
00202             DO 30 J = K + 1, N
00203                IF( VN1( J ).NE.ZERO ) THEN
00204 *
00205 *                 NOTE: The following 4 lines follow from the analysis in
00206 *                 Lapack Working Note 176.
00207 *
00208                   TEMP = ABS( A( RK, J ) ) / VN1( J )
00209                   TEMP = MAX( ZERO, ( ONE+TEMP )*( ONE-TEMP ) )
00210                   TEMP2 = TEMP*( VN1( J ) / VN2( J ) )**2
00211                   IF( TEMP2 .LE. TOL3Z ) THEN
00212                      VN2( J ) = DBLE( LSTICC )
00213                      LSTICC = J
00214                   ELSE
00215                      VN1( J ) = VN1( J )*SQRT( TEMP )
00216                   END IF
00217                END IF
00218    30       CONTINUE
00219          END IF
00220 *
00221          A( RK, K ) = AKK
00222 *
00223 *        End of while loop.
00224 *
00225          GO TO 10
00226       END IF
00227       KB = K
00228       RK = OFFSET + KB
00229 *
00230 *     Apply the block reflector to the rest of the matrix:
00231 *     A(OFFSET+KB+1:M,KB+1:N) := A(OFFSET+KB+1:M,KB+1:N) -
00232 *                         A(OFFSET+KB+1:M,1:KB)*F(KB+1:N,1:KB)**T.
00233 *
00234       IF( KB.LT.MIN( N, M-OFFSET ) ) THEN
00235          CALL DGEMM( 'No transpose', 'Transpose', M-RK, N-KB, KB, -ONE,
00236      $               A( RK+1, 1 ), LDA, F( KB+1, 1 ), LDF, ONE,
00237      $               A( RK+1, KB+1 ), LDA )
00238       END IF
00239 *
00240 *     Recomputation of difficult columns.
00241 *
00242    40 CONTINUE
00243       IF( LSTICC.GT.0 ) THEN
00244          ITEMP = NINT( VN2( LSTICC ) )
00245          VN1( LSTICC ) = DNRM2( M-RK, A( RK+1, LSTICC ), 1 )
00246 *
00247 *        NOTE: The computation of VN1( LSTICC ) relies on the fact that 
00248 *        SNRM2 does not fail on vectors with norm below the value of
00249 *        SQRT(DLAMCH('S')) 
00250 *
00251          VN2( LSTICC ) = VN1( LSTICC )
00252          LSTICC = ITEMP
00253          GO TO 40
00254       END IF
00255 *
00256       RETURN
00257 *
00258 *     End of DLAQPS
00259 *
00260       END
 All Files Functions