octave-maintainers
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

LAPACK revisions


From: Ross A. Lippert
Subject: LAPACK revisions
Date: Wed, 23 Jun 1999 16:17:52 -0600

(Actually, let me send you and the rest of the maintainers the
whole LAPACK 3.0 revisions file so that other people can see if
they want this or that included in the LAPACK-3.0-octave.)
LAPACK, version 3.0

DATE:  April 30, 1999

LAPACK is a library of numerical linear algebra subroutines designed for
high performance on workstations, vector computers, and shared memory 
multiprocessors.  Release 3.0 of LAPACK introduces new routines, as well 
as extending the functionality of existing routines.  The most significant 
new routines and functions are:

 1) singular value decomposition (SVD) computed
    by the divide-and-conquer method, (xGESDD)
 2) SVD based divide-and-conquer least squares (xGELSD)
 3) new simple and expert drivers for the generalized nonsymmetric
    eigenproblem (xGGES,xGGEV,xGGESX,xGGEVX)
 4) new generalized symmetric eigenproblem drivers (xHBGVD/xSBGVD,
    xHBGVX/xSBGVX, xHEGVD/xSYGVD, xHEGVX/xSYGVX, xHPGVD/xSPGVD,
    xHPGVX/xSPGVX) 
 5) symmetric eigenproblem drivers using fast but robust eigenvector
    computations (xSTEGR, xSYEVR/xHEEVR, SSTEVR)
 6) a faster QR decomposition with column pivoting (xGEQP3)
 7) a faster solver for the rank-deficient least squares (xGELSY)
 8) a blocked version of xTZRQF (xTZRZF), and associated xORMRZ/xUNMRZ 
 9) solver for generalized Sylvester equation (xTGSYL)
10) computational routines (xTGEXC, xTGSEN, xTGSNA)

Please note that the structure of the lapack directory on netlib is such
that the SRC files are NOT contained in the top-level lapack directory
but in its subdirectories single, double, complex, and complex16.
The indexes for these subdirectories list each of the different precisions
of a subroutines individually.  (This feature has eased the development
of html search capabilities.)  For brevity, only driver and computational
routines are listed on the indexes.  Auxiliary routines are contained in
the respective subdirectories but NOT listed on the index.  These routines
may be downloaded via anonymous ftp or the "with dependencies" option
via the World Wide Web.
   http://www.netlib.org/lapack/single/
   http://www.netlib.org/lapack/double/
   http://www.netlib.org/lapack/complex/
   http://www.netlib.org/lapack/complex16/

Alternative language interfaces to LAPACK (or translations/conversions
of LAPACK) are available in Fortran~90, C, C++, and Java.
   http://www.netlib.org/lapack90/
   http://www.netlib.org/clapack/
   http://www.netlib.org/lapack++/
   http://www.cs.utk.edu/f2j/download.html

Bug fixes, code modifications, and/or commenting changes have
also occurred.  Some of these modifications are briefly listed below.
For complete information, please refer to the release_notes file
in the lapack directory on netlib.
   http://www.netlib.org/lapack/release_notes

Bug Fixes and/or Code Modifications:

     Tzu-Yi Chen's modified xGEBAL, TESTING/_bal.in and
           TESTING/EIG/_chkbl.f

     Clint's performance improvement to xLASWP

     Nick and Francoise's modification to CLACON and ICMAX1

     Francoise's reduced-workspace xSYEVD, xSBEVD, xSPEVD, xSTEVD

     John Lewis's modified SSYTF2/CSYTF2/CHETF2 and propogated to
                                    SSPTRF/CSPTRF/CHPTRF.

     Linda's modified xSBTRD/xHBTRD to include BLAS 2 calls;
              xSBGST, Linda's modification to include BLAS 2 calls;

     David Day's revised SLAHQR.  And Bai's 2x2 bug fix to CLAHQR.

     V. Sima's modified SLANV2.

     Ed's proposed code modifications (xGTSV, xGTTRF, xGTTRS,
                                       xPTTRF, xPTTRS, SSTERF,
                                       subset of xGEBAL mods)

     LWORK = -1 query capability, WORK(1) returns optimal LWORK
                                          (except xGEESX, xGGESX)

     Most of the major modifications proposed became new routines since
     it would have meant a change in calling sequence.  (e.g., xGGES, xGGEV,
     xGEQP3, xGELSY)

     xSYTF2/xHETF2 and xSPTRF/xHPTRF, modification of pivoting

     xGEGV, scaling fix (replaced by improved driver xGGEV, as we will
            eventually remove xGEGV and xGEGS from the library)

     Expert linear systems drivers (xyySVX) modifications

     xLARTG and xLARGV, corrections to scaling

     TESTING/    sgd.in/cgd.in, ssg.in/csg.in

     TESTING/LIN/  Modifications to pass NNS and NSVAL (array of NRHS
                   values) into xCHKyy testing routines;
                   xCHKTZ and xERRTZ, added testing of xTZRZF;
                   xDRVLS, modified calling sequence to avoid overflowing
                           workspace for more than 2 RHS;
                   xDRVLS to test xGELSD

     TESTING/EIG/
                 xCHKBD, added testing of xBDSDC and modified calling
                         sequence to include IWORK;
                 xCHKST, added test code for xSTEGR;
                 xDRGES, xDRGEV, xDRGSX, xDRGVX to test drivers (xGGEV,
                     xGGEVX, xGGES, xGGESX)
                 xDRVSG, modified calling sequence for new test code for
                   new gsep drivers, mods to xSGT01.
                 CDRVST, new test code for CHBEV, CHPEV, and CHEEV;
                 xDRVST, added testing of SSTEVR and SSYEVR/CHEEVR;
                 xDRVBD, added testing of xGESDD and modified calling
                         sequence to include IWORK;
                 xERRBD, added testing of xBDSDC and xGESDD;
                 xERRGG, added testing of xGS, xGV, xGX, xXV paths
                 xERRST, added testing of xSTEGR, SSTEVR, SSYEVR/CHEEVR; 


     TIMING/LIN/xTIMLS, new timing code for xGELS, xGELSD, xGELSS,
                        xGELSX, xGELSY
     TIMING/LIN/LINSRC directory created with instrumented routines to
        time individual parts of Least Squares (LS, LSD, LSS, LSX, LSY)

     TIMING/EIG/STIM22, new timing code for SORMTR and SORGTR,
                        and xSTEGR and xSYEVR/xHEEVR
                xTIM26, new timing code for xBDSDC and xGESDD

New Routines:

     xBDSDC, computational routines for divide-and-conquer SVD
     xGELSD, driver routine for divide-and-conquer Least Squares (with
             auxiliaries xLALS0, xLALSA, xLALSD, xLASD4-8, xLASDA, xLASDQ,
             xLASDT)
     xGELSY, driver routine (xGELSX using xGEQP3 instead of xGEQPF,
             xTZRZF instead of xTZRQF, and xORMRZ instead of xLATZM)
     xGEQP3, computational routines (faster QR with column pivoting
             and associated auxiliary routines xLAQP2, xLAQPS)
     xGESDD, driver routine for divide-and-conquer SVD (with auxiliaries
             xLASD0-5, xLASD7-9, xLASDA, xLASDQ, xLASDT)
     xGGES,  simple driver routines (improved xGEGS)
     xGGESX, expert driver routines (expert driver of xGGES)
     xGGEV,  simple driver routines (improved xGEGV)
     xGGEVX, expert driver routines (expert driver for xGGEV)
     xHBGVD/xSBGVD and xHBGVX/xSBGVX, new band GSEP drivers
     xHEGVD/xSYGVD and xHEGVX/xSYGVX, new GSEP drivers
     xHPGVD/xSPGVD and xHPGVX/xSPGVX, new packed storage GSEP drivers
     SSTEVR and xSYEVR/xHEEVR, driver routines
     xSTEGR, computational routines HG (and auxiliaries xLASQ1-6)
     xTGEXC, computational routines (and auxiliaries xTGEX2)
     xTGSEN, computational routines
     xTGSNA, computational routines
     xTGSYL, computational routines (and auxiliaries xTGSY2)
     xTZRZF, computational routines (and associated xORMRZ, xORMR3,
             xLARZ, xLARZB, xLARZT, xLATRZ)

     ===========   The simple drivers xGGES and xGGEV are the improved
     PLEASE NOTE:  versions of the existing simple drivers xGEGS and
     ===========   xGEGV.  xGELSY is the improved version of xGELSX.

Additional Features:

     On-line manpages (troff files) for LAPACK routines, as well as
        most of the BLAS routines.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]