spstrf.f

*> \brief \b SPSTRF computes the Cholesky factorization with complete pivoting of a real symmetric positive semidefinite matrix.
*
*  =========== DOCUMENTATION ===========
*
* Online html documentation available at
*            http://www.netlib.org/lapack/explore-html/
*
*> \htmlonly
*> Download SPSTRF + dependencies
*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/spstrf.f">
*> [TGZ]</a>
*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/spstrf.f">
*> [ZIP]</a>
*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/spstrf.f">
*> [TXT]</a>
*> \endhtmlonly
*
*  Definition:
*  ===========
*
*       SUBROUTINE SPSTRF( UPLO, N, A, LDA, PIV, RANK, TOL, WORK, INFO )
*
*       .. Scalar Arguments ..
*       REAL               TOL
*       INTEGER            INFO, LDA, N, RANK
*       CHARACTER          UPLO
*       ..
*       .. Array Arguments ..
*       REAL               A( LDA, * ), WORK( 2*N )
*       INTEGER            PIV( N )
*       ..
*
*
*> \par Purpose:
*  =============
*>
*> \verbatim
*>
*> SPSTRF computes the Cholesky factorization with complete
*> pivoting of a real symmetric positive semidefinite matrix A.
*>
*> The factorization has the form
*>    P**T * A * P = U**T * U ,  if UPLO = 'U',
*>    P**T * A * P = L  * L**T,  if UPLO = 'L',
*> where U is an upper triangular matrix and L is lower triangular, and
*> P is stored as vector PIV.
*>
*> This algorithm does not attempt to check that A is positive
*> semidefinite. This version of the algorithm calls level 3 BLAS.
*> \endverbatim
*
*  Arguments:
*  ==========
*
*> \param[in] UPLO
*> \verbatim
*>          UPLO is CHARACTER*1
*>          Specifies whether the upper or lower triangular part of the
*>          symmetric matrix A is stored.
*>          = 'U':  Upper triangular
*>          = 'L':  Lower triangular
*> \endverbatim
*>
*> \param[in] N
*> \verbatim
*>          N is INTEGER
*>          The order of the matrix A.  N >= 0.
*> \endverbatim
*>
*> \param[in,out] A
*> \verbatim
*>          A is REAL array, dimension (LDA,N)
*>          On entry, the symmetric matrix A.  If UPLO = 'U', the leading
*>          n by n upper triangular part of A contains the upper
*>          triangular part of the matrix A, and the strictly lower
*>          triangular part of A is not referenced.  If UPLO = 'L', the
*>          leading n by n lower triangular part of A contains the lower
*>          triangular part of the matrix A, and the strictly upper
*>          triangular part of A is not referenced.
*>
*>          On exit, if INFO = 0, the factor U or L from the Cholesky
*>          factorization as above.
*> \endverbatim
*>
*> \param[in] LDA
*> \verbatim
*>          LDA is INTEGER
*>          The leading dimension of the array A.  LDA >= max(1,N).
*> \endverbatim
*>
*> \param[out] PIV
*> \verbatim
*>          PIV is INTEGER array, dimension (N)
*>          PIV is such that the nonzero entries are P( PIV(K), K ) = 1.
*> \endverbatim
*>
*> \param[out] RANK
*> \verbatim
*>          RANK is INTEGER
*>          The rank of A given by the number of steps the algorithm
*>          completed.
*> \endverbatim
*>
*> \param[in] TOL
*> \verbatim
*>          TOL is REAL
*>          User defined tolerance. If TOL < 0, then N*U*MAX( A(K,K) )
*>          will be used. The algorithm terminates at the (K-1)st step
*>          if the pivot <= TOL.
*> \endverbatim
*>
*> \param[out] WORK
*> \verbatim
*>          WORK is REAL array, dimension (2*N)
*>          Work space.
*> \endverbatim
*>
*> \param[out] INFO
*> \verbatim
*>          INFO is INTEGER
*>          < 0: If INFO = -K, the K-th argument had an illegal value,
*>          = 0: algorithm completed successfully, and
*>          > 0: the matrix A is either rank deficient with computed rank
*>               as returned in RANK, or is not positive semidefinite. See
*>               Section 7 of LAPACK Working Note #161 for further
*>               information.
*> \endverbatim
*
*  Authors:
*  ========
*
*> \author Univ. of Tennessee
*> \author Univ. of California Berkeley
*> \author Univ. of Colorado Denver
*> \author NAG Ltd.
*
*> \ingroup pstrf
*
*  =====================================================================
      SUBROUTINE spstrf( UPLO, N, A, LDA, PIV, RANK, TOL, WORK, INFO )
*
*  -- LAPACK computational routine --
*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
*
*     .. Scalar Arguments ..
      REAL               TOL
      INTEGER            INFO, LDA, N, RANK
      CHARACTER          UPLO
*     ..
*     .. Array Arguments ..
      REAL               A( lda, * ), WORK( 2*n )
      INTEGER            PIV( n )
*     ..
*
*  =====================================================================
*
*     .. Parameters ..
      REAL               ONE, ZERO
      parameter( one = 1.0e+0, zero = 0.0e+0 )
*     ..
*     .. Local Scalars ..
      REAL               AJJ, SSTOP, STEMP
      INTEGER            I, ITEMP, J, JB, K, NB, PVT
      LOGICAL            UPPER
*     ..
*     .. External Functions ..
      REAL               SLAMCH
      INTEGER            ILAENV
      LOGICAL            LSAME, SISNAN
      EXTERNAL           slamch, ilaenv, lsame, sisnan
*     ..
*     .. External Subroutines ..
      EXTERNAL           sgemv, spstf2, sscal, sswap, ssyrk, xerbla
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          max, min, sqrt, maxloc
*     ..
*     .. Executable Statements ..
*
*     Test the input parameters.
*
      info = 0
      upper = lsame( uplo, 'U' )
      IF( .NOT.upper .AND. .NOT.lsame( uplo, 'L' ) ) THEN
         info = -1
      ELSE IF( n.LT.0 ) THEN
         info = -2
      ELSE IF( lda.LT.max( 1, n ) ) THEN
         info = -4
      END IF
      IF( info.NE.0 ) THEN
         CALL xerbla( 'SPSTRF', -info )
         RETURN
      END IF
*
*     Quick return if possible
*
      IF( n.EQ.0 )
     $   RETURN
*
*     Get block size
*
      nb = ilaenv( 1, 'SPOTRF', uplo, n, -1, -1, -1 )
      IF( nb.LE.1 .OR. nb.GE.n ) THEN
*
*        Use unblocked code
*
         CALL spstf2( uplo, n, a( 1, 1 ), lda, piv, rank, tol, work,
     $                info )
         GO TO 200
*
      ELSE
*
*     Initialize PIV
*
         DO 100 i = 1, n
            piv( i ) = i
  100    CONTINUE
*
*     Compute stopping value
*
         pvt = 1
         ajj = a( pvt, pvt )
         DO i = 2, n
            IF( a( i, i ).GT.ajj ) THEN
               pvt = i
               ajj = a( pvt, pvt )
            END IF
         END DO
         IF( ajj.LE.zero.OR.sisnan( ajj ) ) THEN
            rank = 0
            info = 1
            GO TO 200
         END IF
*
*     Compute stopping value if not supplied
*
         IF( tol.LT.zero ) THEN
            sstop = n * slamch( 'Epsilon' ) * ajj
         ELSE
            sstop = tol
         END IF
*
*
         IF( upper ) THEN
*
*           Compute the Cholesky factorization P**T * A * P = U**T * U
*
            DO 140 k = 1, n, nb
*
*              Account for last block not being NB wide
*
               jb = min( nb, n-k+1 )
*
*              Set relevant part of first half of WORK to zero,
*              holds dot products
*
               DO 110 i = k, n
                  work( i ) = 0
  110          CONTINUE
*
               DO 130 j = k, k + jb - 1
*
*              Find pivot, test for exit, else swap rows and columns
*              Update dot products, compute possible pivots which are
*              stored in the second half of WORK
*
                  DO 120 i = j, n
*
                     IF( j.GT.k ) THEN
                        work( i ) = work( i ) + a( j-1, i )**2
                     END IF
                     work( n+i ) = a( i, i ) - work( i )
*
  120             CONTINUE
*
                  IF( j.GT.1 ) THEN
                     itemp = maxloc( work( (n+j):(2*n) ), 1 )
                     pvt = itemp + j - 1
                     ajj = work( n+pvt )
                     IF( ajj.LE.sstop.OR.sisnan( ajj ) ) THEN
                        a( j, j ) = ajj
                        GO TO 190
                     END IF
                  END IF
*
                  IF( j.NE.pvt ) THEN
*
*                    Pivot OK, so can now swap pivot rows and columns
*
                     a( pvt, pvt ) = a( j, j )
                     CALL sswap( j-1, a( 1, j ), 1, a( 1, pvt ), 1 )
                     IF( pvt.LT.n )
     $                  CALL sswap( n-pvt, a( j, pvt+1 ), lda,
     $                              a( pvt, pvt+1 ), lda )
                     CALL sswap( pvt-j-1, a( j, j+1 ), lda,
     $                           a( j+1, pvt ), 1 )
*
*                    Swap dot products and PIV
*
                     stemp = work( j )
                     work( j ) = work( pvt )
                     work( pvt ) = stemp
                     itemp = piv( pvt )
                     piv( pvt ) = piv( j )
                     piv( j ) = itemp
                  END IF
*
                  ajj = sqrt( ajj )
                  a( j, j ) = ajj
*
*                 Compute elements J+1:N of row J.
*
                  IF( j.LT.n ) THEN
                     CALL sgemv( 'Trans', j-k, n-j, -one, a( k, j+1 ),
     $                           lda, a( k, j ), 1, one, a( j, j+1 ),
     $                           lda )
                     CALL sscal( n-j, one / ajj, a( j, j+1 ), lda )
                  END IF
*
  130          CONTINUE
*
*              Update trailing matrix, J already incremented
*
               IF( k+jb.LE.n ) THEN
                  CALL ssyrk( 'Upper', 'Trans', n-j+1, jb, -one,
     $                        a( k, j ), lda, one, a( j, j ), lda )
               END IF
*
  140       CONTINUE
*
         ELSE
*
*        Compute the Cholesky factorization P**T * A * P = L * L**T
*
            DO 180 k = 1, n, nb
*
*              Account for last block not being NB wide
*
               jb = min( nb, n-k+1 )
*
*              Set relevant part of first half of WORK to zero,
*              holds dot products
*
               DO 150 i = k, n
                  work( i ) = 0
  150          CONTINUE
*
               DO 170 j = k, k + jb - 1
*
*              Find pivot, test for exit, else swap rows and columns
*              Update dot products, compute possible pivots which are
*              stored in the second half of WORK
*
                  DO 160 i = j, n
*
                     IF( j.GT.k ) THEN
                        work( i ) = work( i ) + a( i, j-1 )**2
                     END IF
                     work( n+i ) = a( i, i ) - work( i )
*
  160             CONTINUE
*
                  IF( j.GT.1 ) THEN
                     itemp = maxloc( work( (n+j):(2*n) ), 1 )
                     pvt = itemp + j - 1
                     ajj = work( n+pvt )
                     IF( ajj.LE.sstop.OR.sisnan( ajj ) ) THEN
                        a( j, j ) = ajj
                        GO TO 190
                     END IF
                  END IF
*
                  IF( j.NE.pvt ) THEN
*
*                    Pivot OK, so can now swap pivot rows and columns
*
                     a( pvt, pvt ) = a( j, j )
                     CALL sswap( j-1, a( j, 1 ), lda, a( pvt, 1 ), lda )
                     IF( pvt.LT.n )
     $                  CALL sswap( n-pvt, a( pvt+1, j ), 1,
     $                              a( pvt+1, pvt ), 1 )
                     CALL sswap( pvt-j-1, a( j+1, j ), 1, a( pvt, j+1 ),
     $                           lda )
*
*                    Swap dot products and PIV
*
                     stemp = work( j )
                     work( j ) = work( pvt )
                     work( pvt ) = stemp
                     itemp = piv( pvt )
                     piv( pvt ) = piv( j )
                     piv( j ) = itemp
                  END IF
*
                  ajj = sqrt( ajj )
                  a( j, j ) = ajj
*
*                 Compute elements J+1:N of column J.
*
                  IF( j.LT.n ) THEN
                     CALL sgemv( 'No Trans', n-j, j-k, -one,
     $                           a( j+1, k ), lda, a( j, k ), lda, one,
     $                           a( j+1, j ), 1 )
                     CALL sscal( n-j, one / ajj, a( j+1, j ), 1 )
                  END IF
*
  170          CONTINUE
*
*              Update trailing matrix, J already incremented
*
               IF( k+jb.LE.n ) THEN
                  CALL ssyrk( 'Lower', 'No Trans', n-j+1, jb, -one,
     $                        a( j, k ), lda, one, a( j, j ), lda )
               END IF
*
  180       CONTINUE
*
         END IF
      END IF
*
*     Ran to completion, A has full rank
*
      rank = n
*
      GO TO 200
  190 CONTINUE
*
*     Rank is the number of steps completed.  Set INFO = 1 to signal
*     that the factorization cannot be used to solve a system.
*
      rank = j - 1
      info = 1
*
  200 CONTINUE
      RETURN
*
*     End of SPSTRF
*
      END