Lapack
Lapack
Lapack
Lapack

LAPACK Users Webpage

LAPack 2007

Bugs track

Support

More soon

Logs of installation

Contributions to LAPACK/SCALAPACK

Hello Contributor of LAPACK,

here is your place to check work form other, and how the work is splitted, you should have access to CVS soon, how-to for lots of tools to give you life easier, auto built and testing for your routines everynight (if you want), intructions to have code eventually integrated in LAPACK, ...

Our goal is to provide the community a nice framework to work altogether to produce a new high quality release of LAPACK. All you support is appreciated!

Each contributions has several sections :
  • date of last change
  • implementers
  • references
  • will be done (10%)
  • will be included (10%)
  • progress
  • expected gain (if modif of existing)
  • change of interface (if modif of existing)
  • special features
    • - needs XBLAS, needs BLAS 2.5,
    • - needs recursive Fortran, need allocate Fortran
This is a work document!!!!! Things can be completely wrong, use it, improve it, correct it, but if something looks wrong, let us know, if your proposed contribution is completely forgotten, this is highly possible, let us know as well.

(Prospective) contributions to LAPACK:
Improvements of already existing routines

  • DHGEQZ: multishift QZ algorithm with aggressive early deflation techniques (+block)

    Bo Kågström (Umea) and Daniel Kressner (Zagreb) Bo Kågström wrote: You are already informed about the multishift QZ with aggressive early deflation techniques (joint work with Daniel Kressner). Presently, we are doing extensive testings including comparisons with other implementations (LAPACK, Dackland-Kagstrom's blocked variant, both with/without agressive early deflation).

  • ?????: condition estimator for triangular matrices

    Nick Higham (Manchester) + students (+ Sven Hammarling)

  • DGEHQR: reduction to Hessenberg form

    • Implementation: Stanimire Tomov (UTK)
    • Will be done: 99%
    • Progress: (06/02/2005) double prec. routine is working, ok with LAPACK test, timing ok, two versions implemented
    • Expected gain: 20% on some machines in term of time, accuracy should be the same
    • Related work:
      • from the two papers of Robert van de Geijin (put ref here)
      • error analysis of the Hessenberg reduction is ongoing work from: Nela Bosner (Zagreb) + Daniel Kressner (Zagreb)

  • DGGBAL: balancing GEP

  • DTRSEN + DTGESEN: eigenvalue reordering -- block --

  • DHSEQR:

    • Edward Smyth (NAG) -- Ralph Byers code for QR algorithm
    • paper K. Braman, R. Byers, and R. Mathias, The multishift QR algorithm. Part II: Aggressive early deflation, SIAM J. Matrix Anal. Appl., 23-4 (2002), 948-973.

  • Jacobi SVD of Drmac/Veselic

  • MRRR algorithm for symmetric eigenproblem

    • Christof, Jim, Bresford, Osni

  • MRRR algorithm for SVD

    • Christof, Lang * reduction to tridiagonal form
    • Bishof and Lang

  • Reduction to bidiagonal form

    • Howell/Fulton
    • needs BLAS 2.5
(Prospective) contributions to LAPACK: New Stuff

  • Updating and Downdating

    • Implementation: Pete Stewart (Maryland) Dianne O Leary (Maryland) Julien Langou (UTK)
    • Will be done: 99%
    • Progress: discussions with Pete
    • Related work:
      • block Gram-Schmidt (Jalby and Philippe) Stathopoulos
      • Downdating Jesse Barlow + ref
      • thesis of Craig Lucas (see Nick Higham and Sven Hammarling)

  • Cholesky with pivoting

    • Craig Lucas and Nick Higham

  • GSVD, CS decomposition, product SVD of two matrices

    • Bai (UC Davis) + students

  • MRRR

    • Christof, Osni, Jim
    • Will be done: 99%
    • Progress: almost ok, need to fix parameters accuracy-speed
    • related work LAWN???

  • Iterative refinement with extra precision

    • Sherry Li, Jason, Yozo, Jim, W. Kahan, ...
    • Will be done: 99%
    • Progress: almost done
    • related work LAWN???
    • needs XBLAS

  • Proposed by Bo:

    (see below) need to decide if included
(Prospective) contributions to LAPACK: more LAPACK in ScaLAPACK
  • see after meeting with Jack at 1:00pm Monday 6 June 2005
  • Luc and Serge and Marc:
    • PACKED FORMAT + HYBRID + ... : evaluation of this in Cholesky
    • Up dating of QR facto
(Prospective) contributions to LAPACK: new general feature
in evaluation:
  • support of quads
  • support of infinite precision
  • Babel
  • C-wrapper
FROM BO : Dear Jim and Jack, I think we have some interesting contributions for the new LAPACK and ScaLAPACK. We are developing two libraries for solving different types of matrix equations (RECSY and SCASY). RECSY (joint work with Isak Jonsson) provides recursive blocked solvers for 42 sign and transpose variants of matrix equations in reduced (triangular) form. SCASY (joint work with Robert Granat) provides ScaLAPACK-style solvers for the same 42 sign and transpose variants of undreduced matrix equations. Links to the two libraries: RECSY is completed and all SCASY routines are planned to be completed during the Fall Semester 2005. An important part of SCASY is our ScaLAPACK-style QZ algorithm (jointly with Bjorn Adlerborn and Daniel Kressner), for which we have had some good progress on recently. Some more testings are needed for verification and evaluation.
Last update: Tuesday, May 31 2005
.