LAPACK Archives

[Lapack] your mail

Hello Clint and Jason,

Thanks for the feedback Jason. It's good to know that Clint is not by 
himself complaining about the fact that we removed dopla.f and that 
actually there is other folks complaining! Good, good.

Well, yes, Clint is right, we are really short in resource. The fatal blow 
to dopla.f was really the introduction of the new QR algorithm from Braman 
and Byers in 3.1. We spent all the resource on integrating Ralph's code 
and adapting testings, adapting dopla.f was beyond us ... That would have 
been quite some work. Another problem was that the timing suite (where 
dopla.f lied) was obsolete and not relevant anymore so we ended up 
scratching the whole directory and dopla.f alltogether.

There is still in the air a new writing of the TIMING suite of LAPACK. 
That'll include a dopla like functionnality. Quite frankly, no-one's 
working on it right now. But we listen to feedback.

Best wishes, 

On Tue, 8 Dec 2009, Clint Whaley wrote:


My name is Jason, and I work at Cray in libsci in the same group as 
Kieta Teranishi(who I think you know). Anyway unrelated to that, I came 
across a message of yours on the scalapack users forum about keeping 
good data on flop counts for some of the lapack routines. I noticed 
that dopla.f is gone from netlib as well. This honestly surprises me, 
as I think this kind of information is very important. I was wondering 
if you've used dopla or if you have some other way of figuring out 
accurate flop counts that you recommend.

Yeah, I pled with Julien (the point-man on lapack these days, I think)
to add dopla back to the package and to keep it up to date, since having
a canonical resource for this info is extremely important.  I think Julien
agreed, and was going to look at it, but it is not clear he has been able
to find the time to do it (I CC Julien in case he wants to comment).  In the
meantime, I've just been using the old dopla; as they keep adding/changing
algorithms to LAPACK, this may become increasingly out-of-date, though . . .


** R. Clint Whaley, PhD ** Assist Prof, UTSA ** **

<Prev in Thread] Current Thread [Next in Thread>
  • [Lapack] your mail, Julien Langou <=

For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or