[Vxl-users] Integration of portion of LAPACK/CLAPACK routines to vnl

[Jeff S. <jst...@cs...>]

Hi All:

I am not sure if I have fallen victim to your anti-spam effort, I am
re-posting my query. Since my last post, I have obtained and compiled
clapack. At this point I must decide between: using the old VisSDK
library and integrating clapack and an mpeg library, or integrating the
required routines into vxl's vnl library.

Can anyone answer my questions below and suggest a development path?

Thanks in advance for the help.

Regards
--Jeff Strickrott
Multimedia Database Laboratory
Dept. of Computer Science
Florida International University
Miami, FL
 

> -------- Original Message --------
> Subject: Integration of portion of LAPACK/CLAPACK routines to vnl
> Date: Fri, 09 May 2003 15:39:08 -0400
> From: Jeff Strickrott <jst...@cs...>
> To: vxl...@li...,vxl...@li...
> 
> Hello All:
> 
> I am new to the vxl environment and I am trying to port code that was
> written to use the NAG numerical library and/or a subset of LAPACK and
> need to have the following LAPACK/CLAPACK or NAG routines:
> 
> - DSYEV - compute all eigenvalues and, optionally, eigenvectors of a
> real symmetric matrix A.
> - DGETRF - compute an LU factorization of a general M-by-N matrix A
> using partial pivoting with row interchanges.
> - DGETRI - compute the inverse of a matrix using the LU factorization
> computed by DGETRF.
> - DGEEV - compute for an N-by-N real nonsymmetric matrix A, the
> eigenvalues and, optionally, the left and/or right eigenvectors. I
> believe that your vnl_real_eigensystem will do the same thing.
> 
> - F04AEF: calculates the accurate solution of a set of real linear
> equations with multiple right-hand sides using an LU factorization with
> partial pivoting, and iterative refinement. That is given a set of real
> linear equations AX = B , the routine first computes an LU factorization
> of A with partial pivoting, PA =LU , where P is a permutation matrix, L
> is lower triangular and U is unit upper triangular. An approximation to
> X is found by forward and backward substitution. The residual matrix R
> =B - AX is then calculated using additional precision, and a correction
> D to X is found by solving LUD = PR . X is replaced by X + D and this
> iterative refinement of the solution is repeated until full machine
> accuracy has been obtained.
> 
> My expertise is not with computer vision or numerical routines (yet :-)
> ) and I have the following questions:
> 1. Would it be easier to just obtain CLAPACK and use the appropriate
> routines from there? Has anyone compiled both CLAPACK and vnl together?
> 2. Add the FORTRAN routines per the procedures in the appendix?
> 3. What are (if any) the thread safety issues with vxl, and in
> particular the vnl libraries?
> 4. Any idea for a source for the F04AEF routine?
> 5. Any other suggestions?
> 
> Thanks in advance for the help.
> 
> Regards
> --Jeff Strickrott
> Multimedia Database Laboratory
> Dept. of Computer Science
> Florida International University
> Miami, FL