sgetrf.f

Go to the documentation of this file.

C> \brief \b SGETRF VARIANT: iterative version of Sivan Toledo's recursive LU algorithm
*
*  =========== DOCUMENTATION ===========
*
* Online html documentation available at
*            http://www.netlib.org/lapack/explore-html/
*
*  Definition:
*  ===========
*
*       SUBROUTINE SGETRF( M, N, A, LDA, IPIV, INFO )
*
*       .. Scalar Arguments ..
*       INTEGER            INFO, LDA, M, N
*       ..
*       .. Array Arguments ..
*       INTEGER            IPIV( * )
*       REAL               A( LDA, * )
*       ..
*
*  Purpose
*  =======
*
C>\details \b Purpose:
C>\verbatim
C>
C> SGETRF computes an LU factorization of a general M-by-N matrix A
C> using partial pivoting with row interchanges.
C>
C> The factorization has the form
C>    A = P * L * U
C> where P is a permutation matrix, L is lower triangular with unit
C> diagonal elements (lower trapezoidal if m > n), and U is upper
C> triangular (upper trapezoidal if m < n).
C>
C> This code implements an iterative version of Sivan Toledo's recursive
C> LU algorithm[1].  For square matrices, this iterative versions should
C> be within a factor of two of the optimum number of memory transfers.
C>
C> The pattern is as follows, with the large blocks of U being updated
C> in one call to STRSM, and the dotted lines denoting sections that
C> have had all pending permutations applied:
C>
C>  1 2 3 4 5 6 7 8
C> +-+-+---+-------+------
C> | |1|   |       |
C> |.+-+ 2 |       |
C> | | |   |       |
C> |.|.+-+-+   4   |
C> | | | |1|       |
C> | | |.+-+       |
C> | | | | |       |
C> |.|.|.|.+-+-+---+  8
C> | | | | | |1|   |
C> | | | | |.+-+ 2 |
C> | | | | | | |   |
C> | | | | |.|.+-+-+
C> | | | | | | | |1|
C> | | | | | | |.+-+
C> | | | | | | | | |
C> |.|.|.|.|.|.|.|.+-----
C> | | | | | | | | |
C>
C> The 1-2-1-4-1-2-1-8-... pattern is the position of the last 1 bit in
C> the binary expansion of the current column.  Each Schur update is
C> applied as soon as the necessary portion of U is available.
C>
C> [1] Toledo, S. 1997. Locality of Reference in LU Decomposition with
C> Partial Pivoting. SIAM J. Matrix Anal. Appl. 18, 4 (Oct. 1997),
C> 1065-1081. http://dx.doi.org/10.1137/S0895479896297744
C>
C>\endverbatim
*
*  Arguments:
*  ==========
*
C> \param[in] M
C> \verbatim
C>          M is INTEGER
C>          The number of rows of the matrix A.  M >= 0.
C> \endverbatim
C>
C> \param[in] N
C> \verbatim
C>          N is INTEGER
C>          The number of columns of the matrix A.  N >= 0.
C> \endverbatim
C>
C> \param[in,out] A
C> \verbatim
C>          A is REAL array, dimension (LDA,N)
C>          On entry, the M-by-N matrix to be factored.
C>          On exit, the factors L and U from the factorization
C>          A = P*L*U; the unit diagonal elements of L are not stored.
C> \endverbatim
C>
C> \param[in] LDA
C> \verbatim
C>          LDA is INTEGER
C>          The leading dimension of the array A.  LDA >= max(1,M).
C> \endverbatim
C>
C> \param[out] IPIV
C> \verbatim
C>          IPIV is INTEGER array, dimension (min(M,N))
C>          The pivot indices; for 1 <= i <= min(M,N), row i of the
C>          matrix was interchanged with row IPIV(i).
C> \endverbatim
C>
C> \param[out] INFO
C> \verbatim
C>          INFO is INTEGER
C>          = 0:  successful exit
C>          < 0:  if INFO = -i, the i-th argument had an illegal value
C>          > 0:  if INFO = i, U(i,i) is exactly zero. The factorization
C>                has been completed, but the factor U is exactly
C>                singular, and division by zero will occur if it is used
C>                to solve a system of equations.
C> \endverbatim
C>
*
*  Authors:
*  ========
*
C> \author Univ. of Tennessee
C> \author Univ. of California Berkeley
C> \author Univ. of Colorado Denver
C> \author NAG Ltd.
*
C> \date December 2016
*
C> \ingroup variantsGEcomputational
*
*  =====================================================================
      SUBROUTINE sgetrf( M, N, A, LDA, IPIV, INFO )
*
*  -- LAPACK computational routine (version 3.X) --
*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
*
*     .. Scalar Arguments ..
      INTEGER            INFO, LDA, M, N
*     ..
*     .. Array Arguments ..
      INTEGER            IPIV( * )
      REAL               A( LDA, * )
*     ..
*
*  =====================================================================
*
*     .. Parameters ..
      REAL               ONE, ZERO, NEGONE
      parameter( one = 1.0e+0, zero = 0.0e+0 )
      parameter( negone = -1.0e+0 )
*     ..
*     .. Local Scalars ..
      REAL               SFMIN, TMP
      INTEGER            I, J, JP, NSTEP, NTOPIV, NPIVED, KAHEAD
      INTEGER            KSTART, IPIVSTART, JPIVSTART, KCOLS
*     ..
*     .. External Functions ..
      REAL               SLAMCH
      INTEGER            ISAMAX
      LOGICAL            SISNAN
      EXTERNAL           slamch, isamax, sisnan
*     ..
*     .. External Subroutines ..
      EXTERNAL           strsm, sscal, xerbla, slaswp
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          max, min, iand
*     ..
*     .. Executable Statements ..
*
*     Test the input parameters.
*
      info = 0
      IF( m.LT.0 ) THEN
         info = -1
      ELSE IF( n.LT.0 ) THEN
         info = -2
      ELSE IF( lda.LT.max( 1, m ) ) THEN
         info = -4
      END IF
      IF( info.NE.0 ) THEN
         CALL xerbla( 'SGETRF', -info )
         RETURN
      END IF
*
*     Quick return if possible
*
      IF( m.EQ.0 .OR. n.EQ.0 )
     $   RETURN
*
*     Compute machine safe minimum
*
      sfmin = slamch( 'S' )
*
      nstep = min( m, n )
      DO j = 1, nstep
         kahead = iand( j, -j )
         kstart = j + 1 - kahead
         kcols = min( kahead, m-j )
*
*        Find pivot.
*
         jp = j - 1 + isamax( m-j+1, a( j, j ), 1 )
         ipiv( j ) = jp
 
!        Permute just this column.
         IF (jp .NE. j) THEN
            tmp = a( j, j )
            a( j, j ) = a( jp, j )
            a( jp, j ) = tmp
         END IF
 
!        Apply pending permutations to L
         ntopiv = 1
         ipivstart = j
         jpivstart = j - ntopiv
         DO WHILE ( ntopiv .LT. kahead )
            CALL slaswp( ntopiv, a( 1, jpivstart ), lda, ipivstart, j,
     $           ipiv, 1 )
            ipivstart = ipivstart - ntopiv;
            ntopiv = ntopiv * 2;
            jpivstart = jpivstart - ntopiv;
         END DO
 
!        Permute U block to match L
         CALL slaswp( kcols, a( 1,j+1 ), lda, kstart, j, ipiv, 1 )
 
!        Factor the current column
         IF( a( j, j ).NE.zero .AND. .NOT.sisnan( a( j, j ) ) ) THEN
               IF( abs(a( j, j )) .GE. sfmin ) THEN
                  CALL sscal( m-j, one / a( j, j ), a( j+1, j ), 1 )
               ELSE
                 DO i = 1, m-j
                    a( j+i, j ) = a( j+i, j ) / a( j, j )
                 END DO
               END IF
         ELSE IF( a( j,j ) .EQ. zero .AND. info .EQ. 0 ) THEN
            info = j
         END IF
 
!        Solve for U block.
         CALL strsm( 'Left', 'Lower', 'No transpose', 'Unit', kahead,
     $        kcols, one, a( kstart, kstart ), lda,
     $        a( kstart, j+1 ), lda )
!        Schur complement.
         CALL sgemm( 'No transpose', 'No transpose', m-j,
     $        kcols, kahead, negone, a( j+1, kstart ), lda,
     $        a( kstart, j+1 ), lda, one, a( j+1, j+1 ), lda )
      END DO
 
!     Handle pivot permutations on the way out of the recursion
      npived = iand( nstep, -nstep )
      j = nstep - npived
      DO WHILE ( j .GT. 0 )
         ntopiv = iand( j, -j )
         CALL slaswp( ntopiv, a( 1, j-ntopiv+1 ), lda, j+1, nstep,
     $        ipiv, 1 )
         j = j - ntopiv
      END DO
 
!     If short and wide, handle the rest of the columns.
      IF ( m .LT. n ) THEN
         CALL slaswp( n-m, a( 1, m+kcols+1 ), lda, 1, m, ipiv, 1 )
         CALL strsm( 'Left', 'Lower', 'No transpose', 'Unit', m,
     $        n-m, one, a, lda, a( 1,m+kcols+1 ), lda )
      END IF
 
      RETURN
*
*     End of SGETRF
*
      END