Posted by Paul J. Nikolai on February 07, 1999 at 14:44:04:
Ladies and Gentlemen:
The undersigned is currently engaged in a project to rouse from the TOMS archives a F66 eigenvalue code and to give it new life in F90/95. Testing is well along using the subject symmetric eigenvalue solver and SVD solver from F90_LAPACK required by the code as auxiliary procedures. Results have been excellent, but the resulting executable is grossly inflated in field length by the procedures present as a result of the calls to these F90_LAPACK drivers. A colleague who is running the tests writes:
"I have one comment - must you call Lapack? It
increases the size of the program enormously.
You have 188 dependent routines. If I comment out
the two calls, the executable reduces in size
from 3442004 to 130052 bytes."
What I seek are F90/95 codes which are self-
contained, compact and yield results comparable
to the subject codes for full matrices of
"moderate size", say, up to 300-square. Recall
F90/95 includes F77. It would help if the codes
were in the public domain - like the EISPACK
equivalents - only more modern - like LAPACK.
Thank you for your interest.
Paul J. Nikolai
p.j.nikolai@worldnet.att.net