From: jesse AT lenny DOT dseg DOT ti DOT com (Jesse Bennett)
Newsgroups: comp.os.msdos.djgpp
Subject: Netlib code [was Re: flops...]
Date: 8 Feb 1997 05:55:30 GMT
Organization: Texas Instruments
Lines: 40
Message-ID: <5dh4gi$ek9$1@superb.csc.ti.com>
References: <Pine DOT LNX DOT 3 DOT 91 DOT 970208044630 DOT 1946B-100000 AT aditya DOT unigoa DOT ernet DOT in>
Reply-To: jbennett AT ti DOT com (Jesse Bennett)
NNTP-Posting-Host: lenny.dseg.ti.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp

In article <Pine DOT LNX DOT 3 DOT 91 DOT 970208044630 DOT 1946B-100000 AT aditya DOT unigoa DOT ernet DOT in>,
	"Gurunandan R. Bhat" <grbhat AT unigoa DOT ernet DOT in> writes:
> On Fri, 7 Feb 1997, A. Sinan Unur wrote:
> 
>> in general, NetLib is a great place to look at numerical stuff.
>>    -- Sinan
> 
> yes, but most of it is in fortran :(
> there is an urgent need to translate some important netlib packages to c, 
> and asm to get something better than what f2c gives. 
> 
> any volunteers? i would be happy to chip in.

I would **LOVE** to see a native C version of LAPACK/BLAS.  This is
probably the most widely used piece of code from netlib.  I am
currently using CLAPACK (the f2c'd version of LAPACK) extensively in
my work.  The performance of this code is ... hmmm ... let's see if I
can put this nicely ... absolutely pathetic.  This is no fault of
those who did the translation, indeed I would like to personally thank
those individuals who were involved in this effort.  The availability
of this code has been a tremendous benefit to me, performance issues
notwithstanding.  The code works flawlessly.

To be perfectly honest, to do this right would be much more than a
simple translation of the Fortran code, especially for the BLAS.  I'm
sure this is the reason it hasn't been done already.  Much of the code
is optimized for the "Fortran" way of storing data, i.e. matrices
stored in column-major order vs. row-major order for C.  To say the
least, this makes integrating these routines with existing code a
major pain in the !&%#.  I have written wrappers for the routines I
use which basically do a lot of data reordering/copying to get the
matrices in the "Fortran" structure for the LAPACK code and then back
to "C" structure.

In summary, I would be happy to contribute to any serious effort to
convert this code.  I think there are many others who would benefit
from such a project.

Best Regards,
Jesse