Hi, Code performance is very important of course. I haven't done any benchmarks against LAPACK yet, since I want to make sure the algorithm is stable first. It should be running in O(N^3) time right now since thats what Golub & Van Loan say and I implemented their algorithm pretty much as they outlined it. I will optimize what I can once its working perfectly.
Patrick Alken On Fri, May 19, 2006 at 12:51:04PM +0000, C J Kenneth Tan -- OptimaNumerics wrote: > Patrick, > > On 2006-05-18 10:40 -0600 [EMAIL PROTECTED] wrote: > > On Thu, May 18, 2006 at 11:21:33AM -0500, Linas Vepstas wrote: > > > On Wed, May 17, 2006 at 02:45:10PM -0600, [EMAIL PROTECTED] wrote: > > > > > > > > I have recently needed to compute eigenvalues of unsymmetric > > > > matrices, > > > > > > When I hit ths problem, someone recommended LAPACK, which > > > I then used happily (routine DGEEV). This mailing list has > > > seen occasional discusions about how to wrap/merge/whatever > > > LAPACK routines with gsl (a technical problem is that LAPACK > > > uses the fortran calling convention, making it ugly in C.) > > > > LAPACK I believe uses the same QR algorithm which I have been > > implementing, so if I can make it as robust as LAPACK's > > implementation it should be a far better alternative than > > linking to fortran code. > > Are you taking into consideration performance of the code? How do you > currently compare against NETLIB LAPACK? > > > Kenneth Tan > ----------------------------------------------------------------------- > C J Kenneth Tan, PhD > OptimaNumerics Ltd Telephone: +44 798 941 7838 > E-mail: [EMAIL PROTECTED] Telephone: +44 207 099 4428 > Web: http://www.OptimaNumerics.com Facsimile: +44 207 100 4572 > -----------------------------------------------------------------------
