LAPACK Update

[Jerry DeLisle]

On Sun, 2005-01-16 at 21:22 -0800, Steve Kargl wrote:

> The more I look at the standard and the gfortran source, the
> more complicate this sh*t get.  No wonder Paul is cautious 
> with submitted patches.
> 
On i686-pc-linux-gnu I have isolated the CLS driver failures to specific
call to CGELSY.  Pauls fix did not affect these.  The odd thing is the
failures in LAPACK are down to one dang routine(it appears).  You go
look at it and it makes umpteen calls to BLAS.  I have been studying the
code to see what is uniquely being used in this routine.  Can't say I
have figured it out, but it is odd that the failures before and after
the patch are isolated to this one place.  It is using a logical array
which I don't see too much of.  There was a logical statement function
bug I noticed elsewhere so maybe this is a place to look.  I am really
just browsing until I find clues.

I instrumented this thing before and after the call to CGELSY and
compiled it with g77 and gfortran.  There is definitely different output
from the two programs.  Regardless its a pain.  As my luck will go,
somebody will fix the bug before I discover it in LAPACK and the problem
goes away and maybe my effort wasted.  On the other hand this could keep
plaguing the effort until we do discover it.

So my next step is to delve into CGELSY and instrument it as I have
time.

--Jerry