[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: you might be interested


>I thought I did subscribe, no?  No wonder I haven't received anything
>:-)!  What is the list server again?

Send mail to majordomo@cs.utk.edu, and in the body of the message
(not the subject), say:
subscribe atlas-comm

If it works, you should recieve a reply back from majordomo (automated
process), saying you have been subscribed, and giving directions on
how to unsubscribe yourself . . .

>The below (snip) is very interesting.  Please let me know where
>everything stands.  I.e.  what jobs are left uncovered, what beta
>Atlas should be worked on, etc.  Looks like sgemm has a taker.

You and Goto are the only guys who have actually sent in code.  Doug
(the emmerald guy) is presently scoping SSE sgemm; there is a Danish
student who is looking at using AMD's 3DNow in some fashion (not clear
yet what he will tackle first).  Those are the only things I'm at
all sure about right now (still, it's not too bad considering I only sent
the original mail to roughly 5 guys, and haven't officially released
the user-modifiable ATLAS yet).

So, there is certainly plenty of things to be worked on.  No one is tackling
complex codes at all, that I know of (I still think a SSE/3DNOW-enabled
CGEMV might well match or even exceed DGEMM performance-wise), no one is
looking into whether SSE/3DNOW prefetch could help double precision, etc . . .

I'm kind of a bottleneck at the moment.  I've been working on my software
maintainence tool, extract, for the last week.  As soon as I finish that,
I hope to update the tarfile on the developer page with the bugs Doug 
found fixed, and your and Goto's contributions included . . .

After I get that bare minimum finished, I want to do some work on the
developer page, so we can have a signup kind of deal where who's working
on what could be explained.  I also want to find a way to start getting
all atlas-comm mail archived and web-browsable so that old mail can be
a resource for people hoping to contribute to ATLAS . . .