[Lapack] Doubtful programing style leads to numerical inconsistency
From: Sven Hammarling
Date: Thu, 16 Nov 2006 14:11:02 +0000
In the example you quote I believe that the code is safe and is correct.
The body of the IF statement can safely be skipped if Y(JY) is exactly zero.
Is it possible to introduce changes into current LAPACK sources to fix
these issues? It does not deem to be a hreat number of them. If yes,
we can prepare proposals on how to do it and perform comprehensive
I think that the answer is no, since I doubt that there are issues.
There are already extensive tests for the BLAS and for LAPACK.
(One of the BLAS authors and one of the LAPACK authors.)