Date: Tue, 15 Sep 2009 10:49:58 -0600
From: Shivkumar Chandrasekaran
To: Julien Langou
Cc: James Demmel, lapackers, lapackers, mastewart, Peng Du 
Subject: Re: [Lapackers] [Fwd: CLAPACK/slamch]

I noticed that if I use CLAPACK3.2.1 zgesvd on a random (say 40 x 30) 
matrix it almost always fails, whereas the *exact* same matrix with 
LAPACK zgesvd works fine. The CLAPACK zgesvd fails claiming (in the 40 x 
30) case that 29 super-diagonal elements did not converge (hard to 
believe). However, I am not sure exactly where in zgesvd.c this problem 
arises.

Thanks,

--shiv--