LAPACK 3.2.1 dlarfp

[meconlen]

I have code linked to CLAPACK 3.2.1. I'm running in to a problem with dlarfp. I've seen

viewtopic.php?f=2&t=1986&p=5701&hilit=dlarfp#p5701

and I'm hoping someone can confirm the problem I'm seeing is related to the bug (or another bug documented somewhere else) so I can determine if I should port to something newer (lapacke or magma) or try to understand how to avoid the situation if it's something I'll run in to anyway. (Obviously I *should* upgrade at some point, but I also hope to graduate this May...)

The issue is that I'm doing a least squares regression using dgels. On occasion the code does not return but computes for hours. When I attach to the process with GDB I'm able to 'next' down a few stack frames until I get to dlarfp_(), but the code never returns from dlarfp. as shown below (sorry, I lost the rest of the info above the scrollback).

If there's info I could get to help I can try to dig it out, but the algorithm is a genetic algorithm and so the B array is randomly generated.

In short, is the bug I mentioned, or another one I'm not aware of, one which would cause computation to carry on without stoping?

Thanks for the help.

[rodney]

Unfortunately, there are no newer versions of CLAPACK beyond the 3.2.1 release you are using, and the dlarfp problem with the QR was fixed in LAPACK after version 3.2.1. The more recent versions of LAPACK use more FORTRAN-90 features that the f2c translator cannot handle, although it is possible to modify the LAPACK source to get f2c to work.

I think your best options would be to either use an earlier release of CLAPACK from http://www.netlib.org/clapack/ such as http://www.netlib.org/clapack/clapack-3.1.1.1.tgz, or use a different least squares routine such as the SVD based dgelss that does not depend on the QR in version 3.2.1.

Rodney