OpenBLAS/lapack-netlib/SRC/dlasy2.f

*> \brief \b DLASY2 solves the Sylvester matrix equation where the matrices are of order 1 or 2.
*
*  =========== DOCUMENTATION ===========
*
* Online html documentation available at
*            http://www.netlib.org/lapack/explore-html/
*
*> \htmlonly
*> Download DLASY2 + dependencies
*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlasy2.f">
*> [TGZ]</a>
*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlasy2.f">
*> [ZIP]</a>
*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlasy2.f">
*> [TXT]</a>
*> \endhtmlonly
*
*  Definition:
*  ===========
*
*       SUBROUTINE DLASY2( LTRANL, LTRANR, ISGN, N1, N2, TL, LDTL, TR,
*                          LDTR, B, LDB, SCALE, X, LDX, XNORM, INFO )
*
*       .. Scalar Arguments ..
*       LOGICAL            LTRANL, LTRANR
*       INTEGER            INFO, ISGN, LDB, LDTL, LDTR, LDX, N1, N2
*       DOUBLE PRECISION   SCALE, XNORM
*       ..
*       .. Array Arguments ..
*       DOUBLE PRECISION   B( LDB, * ), TL( LDTL, * ), TR( LDTR, * ),
*      $                   X( LDX, * )
*       ..
*
*
*> \par Purpose:
*  =============
*>
*> \verbatim
*>
*> DLASY2 solves for the N1 by N2 matrix X, 1 <= N1,N2 <= 2, in
*>
*>        op(TL)*X + ISGN*X*op(TR) = SCALE*B,
*>
*> where TL is N1 by N1, TR is N2 by N2, B is N1 by N2, and ISGN = 1 or
*> -1.  op(T) = T or T**T, where T**T denotes the transpose of T.
*> \endverbatim
*
*  Arguments:
*  ==========
*
*> \param[in] LTRANL
*> \verbatim
*>          LTRANL is LOGICAL
*>          On entry, LTRANL specifies the op(TL):
*>             = .FALSE., op(TL) = TL,
*>             = .TRUE., op(TL) = TL**T.
*> \endverbatim
*>
*> \param[in] LTRANR
*> \verbatim
*>          LTRANR is LOGICAL
*>          On entry, LTRANR specifies the op(TR):
*>            = .FALSE., op(TR) = TR,
*>            = .TRUE., op(TR) = TR**T.
*> \endverbatim
*>
*> \param[in] ISGN
*> \verbatim
*>          ISGN is INTEGER
*>          On entry, ISGN specifies the sign of the equation
*>          as described before. ISGN may only be 1 or -1.
*> \endverbatim
*>
*> \param[in] N1
*> \verbatim
*>          N1 is INTEGER
*>          On entry, N1 specifies the order of matrix TL.
*>          N1 may only be 0, 1 or 2.
*> \endverbatim
*>
*> \param[in] N2
*> \verbatim
*>          N2 is INTEGER
*>          On entry, N2 specifies the order of matrix TR.
*>          N2 may only be 0, 1 or 2.
*> \endverbatim
*>
*> \param[in] TL
*> \verbatim
*>          TL is DOUBLE PRECISION array, dimension (LDTL,2)
*>          On entry, TL contains an N1 by N1 matrix.
*> \endverbatim
*>
*> \param[in] LDTL
*> \verbatim
*>          LDTL is INTEGER
*>          The leading dimension of the matrix TL. LDTL >= max(1,N1).
*> \endverbatim
*>
*> \param[in] TR
*> \verbatim
*>          TR is DOUBLE PRECISION array, dimension (LDTR,2)
*>          On entry, TR contains an N2 by N2 matrix.
*> \endverbatim
*>
*> \param[in] LDTR
*> \verbatim
*>          LDTR is INTEGER
*>          The leading dimension of the matrix TR. LDTR >= max(1,N2).
*> \endverbatim
*>
*> \param[in] B
*> \verbatim
*>          B is DOUBLE PRECISION array, dimension (LDB,2)
*>          On entry, the N1 by N2 matrix B contains the right-hand
*>          side of the equation.
*> \endverbatim
*>
*> \param[in] LDB
*> \verbatim
*>          LDB is INTEGER
*>          The leading dimension of the matrix B. LDB >= max(1,N1).
*> \endverbatim
*>
*> \param[out] SCALE
*> \verbatim
*>          SCALE is DOUBLE PRECISION
*>          On exit, SCALE contains the scale factor. SCALE is chosen
*>          less than or equal to 1 to prevent the solution overflowing.
*> \endverbatim
*>
*> \param[out] X
*> \verbatim
*>          X is DOUBLE PRECISION array, dimension (LDX,2)
*>          On exit, X contains the N1 by N2 solution.
*> \endverbatim
*>
*> \param[in] LDX
*> \verbatim
*>          LDX is INTEGER
*>          The leading dimension of the matrix X. LDX >= max(1,N1).
*> \endverbatim
*>
*> \param[out] XNORM
*> \verbatim
*>          XNORM is DOUBLE PRECISION
*>          On exit, XNORM is the infinity-norm of the solution.
*> \endverbatim
*>
*> \param[out] INFO
*> \verbatim
*>          INFO is INTEGER
*>          On exit, INFO is set to
*>             0: successful exit.
*>             1: TL and TR have too close eigenvalues, so TL or
*>                TR is perturbed to get a nonsingular equation.
*>          NOTE: In the interests of speed, this routine does not
*>                check the inputs for errors.
*> \endverbatim
*
*  Authors:
*  ========
*
*> \author Univ. of Tennessee
*> \author Univ. of California Berkeley
*> \author Univ. of Colorado Denver
*> \author NAG Ltd.
*
*> \ingroup doubleSYauxiliary
*
*  =====================================================================
      SUBROUTINE DLASY2( LTRANL, LTRANR, ISGN, N1, N2, TL, LDTL, TR,
     $                   LDTR, B, LDB, SCALE, X, LDX, XNORM, INFO )
*
*  -- LAPACK auxiliary routine --
*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
*
*     .. Scalar Arguments ..
      LOGICAL            LTRANL, LTRANR
      INTEGER            INFO, ISGN, LDB, LDTL, LDTR, LDX, N1, N2
      DOUBLE PRECISION   SCALE, XNORM
*     ..
*     .. Array Arguments ..
      DOUBLE PRECISION   B( LDB, * ), TL( LDTL, * ), TR( LDTR, * ),
     $                   X( LDX, * )
*     ..
*
* =====================================================================
*
*     .. Parameters ..
      DOUBLE PRECISION   ZERO, ONE
      PARAMETER          ( ZERO = 0.0D+0, ONE = 1.0D+0 )
      DOUBLE PRECISION   TWO, HALF, EIGHT
      PARAMETER          ( TWO = 2.0D+0, HALF = 0.5D+0, EIGHT = 8.0D+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            BSWAP, XSWAP
      INTEGER            I, IP, IPIV, IPSV, J, JP, JPSV, K
      DOUBLE PRECISION   BET, EPS, GAM, L21, SGN, SMIN, SMLNUM, TAU1,
     $                   TEMP, U11, U12, U22, XMAX
*     ..
*     .. Local Arrays ..
      LOGICAL            BSWPIV( 4 ), XSWPIV( 4 )
      INTEGER            JPIV( 4 ), LOCL21( 4 ), LOCU12( 4 ),
     $                   LOCU22( 4 )
      DOUBLE PRECISION   BTMP( 4 ), T16( 4, 4 ), TMP( 4 ), X2( 2 )
*     ..
*     .. External Functions ..
      INTEGER            IDAMAX
      DOUBLE PRECISION   DLAMCH
      EXTERNAL           IDAMAX, DLAMCH
*     ..
*     .. External Subroutines ..
      EXTERNAL           DCOPY, DSWAP
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          ABS, MAX
*     ..
*     .. Data statements ..
      DATA               LOCU12 / 3, 4, 1, 2 / , LOCL21 / 2, 1, 4, 3 / ,
     $                   LOCU22 / 4, 3, 2, 1 /
      DATA               XSWPIV / .FALSE., .FALSE., .TRUE., .TRUE. /
      DATA               BSWPIV / .FALSE., .TRUE., .FALSE., .TRUE. /
*     ..
*     .. Executable Statements ..
*
*     Do not check the input parameters for errors
*
      INFO = 0
*
*     Quick return if possible
*
      IF( N1.EQ.0 .OR. N2.EQ.0 )
     $   RETURN
*
*     Set constants to control overflow
*
      EPS = DLAMCH( 'P' )
      SMLNUM = DLAMCH( 'S' ) / EPS
      SGN = ISGN
*
      K = N1 + N1 + N2 - 2
      GO TO ( 10, 20, 30, 50 )K
*
*     1 by 1: TL11*X + SGN*X*TR11 = B11
*
   10 CONTINUE
      TAU1 = TL( 1, 1 ) + SGN*TR( 1, 1 )
      BET = ABS( TAU1 )
      IF( BET.LE.SMLNUM ) THEN
         TAU1 = SMLNUM
         BET = SMLNUM
         INFO = 1
      END IF
*
      SCALE = ONE
      GAM = ABS( B( 1, 1 ) )
      IF( SMLNUM*GAM.GT.BET )
     $   SCALE = ONE / GAM
*
      X( 1, 1 ) = ( B( 1, 1 )*SCALE ) / TAU1
      XNORM = ABS( X( 1, 1 ) )
      RETURN
*
*     1 by 2:
*     TL11*[X11 X12] + ISGN*[X11 X12]*op[TR11 TR12]  = [B11 B12]
*                                       [TR21 TR22]
*
   20 CONTINUE
*
      SMIN = MAX( EPS*MAX( ABS( TL( 1, 1 ) ), ABS( TR( 1, 1 ) ),
     $       ABS( TR( 1, 2 ) ), ABS( TR( 2, 1 ) ), ABS( TR( 2, 2 ) ) ),
     $       SMLNUM )
      TMP( 1 ) = TL( 1, 1 ) + SGN*TR( 1, 1 )
      TMP( 4 ) = TL( 1, 1 ) + SGN*TR( 2, 2 )
      IF( LTRANR ) THEN
         TMP( 2 ) = SGN*TR( 2, 1 )
         TMP( 3 ) = SGN*TR( 1, 2 )
      ELSE
         TMP( 2 ) = SGN*TR( 1, 2 )
         TMP( 3 ) = SGN*TR( 2, 1 )
      END IF
      BTMP( 1 ) = B( 1, 1 )
      BTMP( 2 ) = B( 1, 2 )
      GO TO 40
*
*     2 by 1:
*          op[TL11 TL12]*[X11] + ISGN* [X11]*TR11  = [B11]
*            [TL21 TL22] [X21]         [X21]         [B21]
*
   30 CONTINUE
      SMIN = MAX( EPS*MAX( ABS( TR( 1, 1 ) ), ABS( TL( 1, 1 ) ),
     $       ABS( TL( 1, 2 ) ), ABS( TL( 2, 1 ) ), ABS( TL( 2, 2 ) ) ),
     $       SMLNUM )
      TMP( 1 ) = TL( 1, 1 ) + SGN*TR( 1, 1 )
      TMP( 4 ) = TL( 2, 2 ) + SGN*TR( 1, 1 )
      IF( LTRANL ) THEN
         TMP( 2 ) = TL( 1, 2 )
         TMP( 3 ) = TL( 2, 1 )
      ELSE
         TMP( 2 ) = TL( 2, 1 )
         TMP( 3 ) = TL( 1, 2 )
      END IF
      BTMP( 1 ) = B( 1, 1 )
      BTMP( 2 ) = B( 2, 1 )
   40 CONTINUE
*
*     Solve 2 by 2 system using complete pivoting.
*     Set pivots less than SMIN to SMIN.
*
      IPIV = IDAMAX( 4, TMP, 1 )
      U11 = TMP( IPIV )
      IF( ABS( U11 ).LE.SMIN ) THEN
         INFO = 1
         U11 = SMIN
      END IF
      U12 = TMP( LOCU12( IPIV ) )
      L21 = TMP( LOCL21( IPIV ) ) / U11
      U22 = TMP( LOCU22( IPIV ) ) - U12*L21
      XSWAP = XSWPIV( IPIV )
      BSWAP = BSWPIV( IPIV )
      IF( ABS( U22 ).LE.SMIN ) THEN
         INFO = 1
         U22 = SMIN
      END IF
      IF( BSWAP ) THEN
         TEMP = BTMP( 2 )
         BTMP( 2 ) = BTMP( 1 ) - L21*TEMP
         BTMP( 1 ) = TEMP
      ELSE
         BTMP( 2 ) = BTMP( 2 ) - L21*BTMP( 1 )
      END IF
      SCALE = ONE
      IF( ( TWO*SMLNUM )*ABS( BTMP( 2 ) ).GT.ABS( U22 ) .OR.
     $    ( TWO*SMLNUM )*ABS( BTMP( 1 ) ).GT.ABS( U11 ) ) THEN
         SCALE = HALF / MAX( ABS( BTMP( 1 ) ), ABS( BTMP( 2 ) ) )
         BTMP( 1 ) = BTMP( 1 )*SCALE
         BTMP( 2 ) = BTMP( 2 )*SCALE
      END IF
      X2( 2 ) = BTMP( 2 ) / U22
      X2( 1 ) = BTMP( 1 ) / U11 - ( U12 / U11 )*X2( 2 )
      IF( XSWAP ) THEN
         TEMP = X2( 2 )
         X2( 2 ) = X2( 1 )
         X2( 1 ) = TEMP
      END IF
      X( 1, 1 ) = X2( 1 )
      IF( N1.EQ.1 ) THEN
         X( 1, 2 ) = X2( 2 )
         XNORM = ABS( X( 1, 1 ) ) + ABS( X( 1, 2 ) )
      ELSE
         X( 2, 1 ) = X2( 2 )
         XNORM = MAX( ABS( X( 1, 1 ) ), ABS( X( 2, 1 ) ) )
      END IF
      RETURN
*
*     2 by 2:
*     op[TL11 TL12]*[X11 X12] +ISGN* [X11 X12]*op[TR11 TR12] = [B11 B12]
*       [TL21 TL22] [X21 X22]        [X21 X22]   [TR21 TR22]   [B21 B22]
*
*     Solve equivalent 4 by 4 system using complete pivoting.
*     Set pivots less than SMIN to SMIN.
*
   50 CONTINUE
      SMIN = MAX( ABS( TR( 1, 1 ) ), ABS( TR( 1, 2 ) ),
     $       ABS( TR( 2, 1 ) ), ABS( TR( 2, 2 ) ) )
      SMIN = MAX( SMIN, ABS( TL( 1, 1 ) ), ABS( TL( 1, 2 ) ),
     $       ABS( TL( 2, 1 ) ), ABS( TL( 2, 2 ) ) )
      SMIN = MAX( EPS*SMIN, SMLNUM )
      BTMP( 1 ) = ZERO
      CALL DCOPY( 16, BTMP, 0, T16, 1 )
      T16( 1, 1 ) = TL( 1, 1 ) + SGN*TR( 1, 1 )
      T16( 2, 2 ) = TL( 2, 2 ) + SGN*TR( 1, 1 )
      T16( 3, 3 ) = TL( 1, 1 ) + SGN*TR( 2, 2 )
      T16( 4, 4 ) = TL( 2, 2 ) + SGN*TR( 2, 2 )
      IF( LTRANL ) THEN
         T16( 1, 2 ) = TL( 2, 1 )
         T16( 2, 1 ) = TL( 1, 2 )
         T16( 3, 4 ) = TL( 2, 1 )
         T16( 4, 3 ) = TL( 1, 2 )
      ELSE
         T16( 1, 2 ) = TL( 1, 2 )
         T16( 2, 1 ) = TL( 2, 1 )
         T16( 3, 4 ) = TL( 1, 2 )
         T16( 4, 3 ) = TL( 2, 1 )
      END IF
      IF( LTRANR ) THEN
         T16( 1, 3 ) = SGN*TR( 1, 2 )
         T16( 2, 4 ) = SGN*TR( 1, 2 )
         T16( 3, 1 ) = SGN*TR( 2, 1 )
         T16( 4, 2 ) = SGN*TR( 2, 1 )
      ELSE
         T16( 1, 3 ) = SGN*TR( 2, 1 )
         T16( 2, 4 ) = SGN*TR( 2, 1 )
         T16( 3, 1 ) = SGN*TR( 1, 2 )
         T16( 4, 2 ) = SGN*TR( 1, 2 )
      END IF
      BTMP( 1 ) = B( 1, 1 )
      BTMP( 2 ) = B( 2, 1 )
      BTMP( 3 ) = B( 1, 2 )
      BTMP( 4 ) = B( 2, 2 )
*
*     Perform elimination
*
      DO 100 I = 1, 3
         XMAX = ZERO
         DO 70 IP = I, 4
            DO 60 JP = I, 4
               IF( ABS( T16( IP, JP ) ).GE.XMAX ) THEN
                  XMAX = ABS( T16( IP, JP ) )
                  IPSV = IP
                  JPSV = JP
               END IF
   60       CONTINUE
   70    CONTINUE
         IF( IPSV.NE.I ) THEN
            CALL DSWAP( 4, T16( IPSV, 1 ), 4, T16( I, 1 ), 4 )
            TEMP = BTMP( I )
            BTMP( I ) = BTMP( IPSV )
            BTMP( IPSV ) = TEMP
         END IF
         IF( JPSV.NE.I )
     $      CALL DSWAP( 4, T16( 1, JPSV ), 1, T16( 1, I ), 1 )
         JPIV( I ) = JPSV
         IF( ABS( T16( I, I ) ).LT.SMIN ) THEN
            INFO = 1
            T16( I, I ) = SMIN
         END IF
         DO 90 J = I + 1, 4
            T16( J, I ) = T16( J, I ) / T16( I, I )
            BTMP( J ) = BTMP( J ) - T16( J, I )*BTMP( I )
            DO 80 K = I + 1, 4
               T16( J, K ) = T16( J, K ) - T16( J, I )*T16( I, K )
   80       CONTINUE
   90    CONTINUE
  100 CONTINUE
      IF( ABS( T16( 4, 4 ) ).LT.SMIN ) THEN
         INFO = 1
         T16( 4, 4 ) = SMIN
      END IF
      SCALE = ONE
      IF( ( EIGHT*SMLNUM )*ABS( BTMP( 1 ) ).GT.ABS( T16( 1, 1 ) ) .OR.
     $    ( EIGHT*SMLNUM )*ABS( BTMP( 2 ) ).GT.ABS( T16( 2, 2 ) ) .OR.
     $    ( EIGHT*SMLNUM )*ABS( BTMP( 3 ) ).GT.ABS( T16( 3, 3 ) ) .OR.
     $    ( EIGHT*SMLNUM )*ABS( BTMP( 4 ) ).GT.ABS( T16( 4, 4 ) ) ) THEN
         SCALE = ( ONE / EIGHT ) / MAX( ABS( BTMP( 1 ) ),
     $           ABS( BTMP( 2 ) ), ABS( BTMP( 3 ) ), ABS( BTMP( 4 ) ) )
         BTMP( 1 ) = BTMP( 1 )*SCALE
         BTMP( 2 ) = BTMP( 2 )*SCALE
         BTMP( 3 ) = BTMP( 3 )*SCALE
         BTMP( 4 ) = BTMP( 4 )*SCALE
      END IF
      DO 120 I = 1, 4
         K = 5 - I
         TEMP = ONE / T16( K, K )
         TMP( K ) = BTMP( K )*TEMP
         DO 110 J = K + 1, 4
            TMP( K ) = TMP( K ) - ( TEMP*T16( K, J ) )*TMP( J )
  110    CONTINUE
  120 CONTINUE
      DO 130 I = 1, 3
         IF( JPIV( 4-I ).NE.4-I ) THEN
            TEMP = TMP( 4-I )
            TMP( 4-I ) = TMP( JPIV( 4-I ) )
            TMP( JPIV( 4-I ) ) = TEMP
         END IF
  130 CONTINUE
      X( 1, 1 ) = TMP( 1 )
      X( 2, 1 ) = TMP( 2 )
      X( 1, 2 ) = TMP( 3 )
      X( 2, 2 ) = TMP( 4 )
      XNORM = MAX( ABS( TMP( 1 ) )+ABS( TMP( 3 ) ),
     $        ABS( TMP( 2 ) )+ABS( TMP( 4 ) ) )
      RETURN
*
*     End of DLASY2
*
      END