64-bit cleanlinest of gfortran

[Steve Kargl]

On Mon, Dec 13, 2004 at 07:32:01PM -0500, Andrew Pinski wrote:
> 
> On Dec 13, 2004, at 7:29 PM, Steve Kargl wrote:
> 
> >Has anyone looked at the gfortran code for its cleanliness
> >with respect to a 64-bit OS?  I reported very good
> >success with building and running netlib's LAPACK
> >code.  That success was on i386-*-FreeBSD6.0.  I am
> >having spectacular failure on amd64-*-FreeBSD6.0.  If
> >the LAPACK tests do not die with a BUS ERROR, then the
> >single and double complex tests produce hundreds of
> >failures.  These failures are producing conditions
> >numbers on the order of 1e16.  Can someone, using
> >a different 64-bit OS, build and run the LAPACK test
> >program?
> 
> The complex problem is most likely a target bug only on x86-64,
> see PR 17603.
> 

Thanks for the PR number.  However, the BUS ERROR problem is more
of a concern to me.  This smells of an assumption of the sizeof
a type (i.e., sizeof(int) == sizeof(long)) and some not so safe
pointer mathematics.  I'll hunt around and see if I can find anything.

-- 
Steve