AVX Extensions - Ongoing development?

Message boards : Number crunching : AVX Extensions - Ongoing development?
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · Next

AuthorMessage
B-Man
Volunteer tester

Send message
Joined: 11 Feb 01
Posts: 253
Credit: 147,366
RAC: 0
United States
Message 1086211 - Posted: 11 Mar 2011, 23:32:54 UTC

How is the dev going?
ID: 1086211 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1086374 - Posted: 12 Mar 2011, 10:36:12 UTC - in response to Message 1086211.  
Last modified: 12 Mar 2011, 10:39:53 UTC

How is the dev going?


FYI: On the Lunatics side of things, focus has temporarily shifted to getting Raistmer's OpenCL apps into the installer, getting the recently optimised Cuda code into the X Branch, and perhaps most importantly assisting Berkeley with SaH_v7 stock refinement.

Fortunately there is quite a lot of crossover between stock SaH_v7 refinement and the additions to AK_v8b required to support AVX hand optimisation (as opposed to compiler rebuilds). That means a steady development through stock leading into a major AK_v8b update to support V7 & AVX at the same time, seems the logical path at the moment, though it's likely to continue to be slow going since where AVX would likely benefit the most is some tough code.

Jason
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1086374 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1086482 - Posted: 12 Mar 2011, 17:55:57 UTC - in response to Message 1086374.  

How is the dev going?


FYI: On the Lunatics side of things, focus has temporarily shifted to getting Raistmer's OpenCL apps into the installer, getting the recently optimised Cuda code into the X Branch, and perhaps most importantly assisting Berkeley with SaH_v7 stock refinement.

Fortunately there is quite a lot of crossover between stock SaH_v7 refinement and the additions to AK_v8b required to support AVX hand optimisation (as opposed to compiler rebuilds). That means a steady development through stock leading into a major AK_v8b update to support V7 & AVX at the same time, seems the logical path at the moment, though it's likely to continue to be slow going since where AVX would likely benefit the most is some tough code.

Jason


Forget AVX and OpenCL, FERMI FERMI FERMI FERMI FERMI......

Eck....shake it off....shake it off....good job guys can't wait to see what's added in the next version! Take your time. ;)
Traveling through space at ~67,000mph!
ID: 1086482 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1086511 - Posted: 12 Mar 2011, 19:58:29 UTC - in response to Message 1086482.  

there is work on a fermi OpenCL app IIRC at Lunatics


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1086511 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1086529 - Posted: 12 Mar 2011, 21:04:59 UTC - in response to Message 1086511.  

there is work on a fermi OpenCL app IIRC at Lunatics


LoL. Sadly despite that some OpenCL apps can, & will, run even some ATi SDK (as well as nVidia) builds on nVidia cards, doesn't make it as fast on Cuda Cores as even relatively unoptimised Cuda code yet. The tools and libraries have quite a way to catch up (though getting there slowly), so I wouldn't hold your breath for Cuda to be superseded on these cards.

Forget AVX and OpenCL, FERMI FERMI FERMI FERMI FERMI......

Eck....shake it off....shake it off....good job guys can't wait to see what's added in the next version! Take your time. ;)


:D. A few other people are getting anxious for an improved Fermi app too. I understand that completely. Some description of how I work might be in order just to fill in the picture a bit, bearing in mind I don't work for the public service, don't have time to keep a blog to give constant updates to the curious & the well wishers, and do have other responsibilities.

I work in highly productive 'bursts', with thorough investigation, repeated analysis & proving/disproving different methods & techniques (sometimes seemingly unconnected) that tend to converge & combine on a final outcome. Much of that time is spent reading & experimenting behind the scenes. Some of the in-between is sometimes also spent recovering from those unstoppable bursts, like right now taking a break to answer questions here :)

The main publicly visible artefacts of that process are either the 'final product' or viable looking intermediate stages toward the goal. In the case of the first phase optimisations to the Cuda codebase, the experimentation & refinement of certain heuristics & optimisations occurred very publicly, though that this particular strain of development applies to a near 60% total rewrite of the cuda application, applicable to all cards (inc Fermi), isn't immediately recognisable to casual observer.

Of course, integrating a 60% rewrite of what is after all prototype code that uses higher level engineering practices not well documented in the gpGPU field, and that the approach itself tends to expose limitations/bugs in preceding work, complicates things a lot. It does so by having to first analyse to make sure the new work is 'right', make decisions about whether the old work should be improved, or accept the 'original' as reference, etc.., then come up with fixes for the 'other stuff' if that's thought appropriate.

Make no mistake about it, I am far enough along toward the skillset needed to complete the first total rewrite, that I've drawn a line under the opt1 60% rewrite and said 'good enough for phase one'. How that goes over the coming weeks will likely determine the priorities for 'Opt2', and how both are to mesh with coming SaH_v7.

I hope that clarifies a bit for the curious, as I see a lot of questions floating about. One thing I can say is that someone that claims they can churn out faster Cuda code than nVidia Engineers, that doesn't take the time to do all the reading & experiment ... is either a genius or a liar.

Jason

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1086529 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13797
Credit: 208,696,464
RAC: 304
Australia
Message 1086534 - Posted: 12 Mar 2011, 21:11:15 UTC - in response to Message 1086529.  

One thing I can say is that someone that claims they can churn out faster Cuda code than nVidia Engineers, that doesn't take the time to do all the reading & experiment ... is either a genius or a liar.

Or just a deluded fool.
I've met enough of them over the years.
*rolls eyes*

Grant
Darwin NT
ID: 1086534 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1086557 - Posted: 12 Mar 2011, 23:00:38 UTC - in response to Message 1086529.  

:D. A few other people are getting anxious for an improved Fermi app too. I understand that completely. Some description of how I work might be in order just to fill in the picture a bit, bearing in mind I don't work for the public service, don't have time to keep a blog to give constant updates to the curious & the well wishers, and do have other responsibilities.


Thanks for that Jason. We know everything is coming along for better apps for us, no doubt about that. It's just gets a bit aggravating knowing my equipment could be possibly doing a lot more with the proper optimizations. With that being said, I understand it takes time, and making every step you are taking may not be advantageous in helping keep the curious at bay as it would probably only bring more questions in what you are doing. However it would be nice to kind of see a public check list of what you are working on, is that something that's available of could be made available. IE a road map on the work you are doing etc. Would be interesting and kind of keep people in the right lane on what's going on with development on new apps etc. Thanks again!
Traveling through space at ~67,000mph!
ID: 1086557 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1086656 - Posted: 13 Mar 2011, 4:25:17 UTC - in response to Message 1086557.  

... IE a road map on the work you are doing etc. Would be interesting and kind of keep people in the right lane on what's going on with development on new apps etc.


That's not a bad idea at all. I'll see what I can do.
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1086656 · Report as offensive
Profile perryjay
Volunteer tester
Avatar

Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 20,676,751
RAC: 0
United States
Message 1086725 - Posted: 13 Mar 2011, 13:36:29 UTC - in response to Message 1086656.  

Hmmm, so that means it will be a few more days then? Should we send more coffee? :-)


PROUD MEMBER OF Team Starfire World BOINC
ID: 1086725 · Report as offensive
Profile Helli_retiered
Volunteer tester
Avatar

Send message
Joined: 15 Dec 99
Posts: 707
Credit: 108,785,585
RAC: 0
Germany
Message 1086727 - Posted: 13 Mar 2011, 13:45:42 UTC - in response to Message 1086725.  

Hmmm, so that means it will be a few more days then :-)


Phew...thankfully that i did not say that... ;-)


Helli
A loooong time ago: First Credits after SETI@home Restart
ID: 1086727 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1086732 - Posted: 13 Mar 2011, 14:33:30 UTC - in response to Message 1086727.  
Last modified: 13 Mar 2011, 14:38:17 UTC

Hmmm, so that means it will be a few more days then :-)


Phew...thankfully that i did not say that... ;-)


Helli



Hmmm, so that means it will be a few more days then? Should we send more coffee? :-)


Right now a cloning device would be more appreciated. (i.e. 1am on Monday morning :) )
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1086732 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1086733 - Posted: 13 Mar 2011, 14:39:59 UTC - in response to Message 1086732.  

Hmmm, so that means it will be a few more days then :-)


Phew...thankfully that i did not say that... ;-)


Helli



Hmmm, so that means it will be a few more days then? Should we send more coffee? :-)


Right now a cloning device would be more appreciated.


Sorry, the cloning device never got out of alpha testing. Currently it is turning everyone into clowns. There seems to have been a typo in the design specs. :)
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1086733 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1086823 - Posted: 13 Mar 2011, 19:51:08 UTC - in response to Message 1086733.  
Last modified: 13 Mar 2011, 19:51:54 UTC

Sorry, the cloning device never got out of alpha testing. Currently it is turning everyone into clowns. There seems to have been a typo in the design specs. :)


Ahhh, perfect. That would give Raistmer plenty of help with the Ati stuff :P

Rough roadmap presented at Lunatics, trying to place the Cuda work in context of surrounding development, and all the collaboration & overlap that goes on.

http://lunatics.kwsn.net/1-discussion-forum/lunatics-experimental-development-roadmap.msg36185.html#msg36185
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1086823 · Report as offensive
Profile ML1
Volunteer moderator
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 20621
Credit: 7,508,002
RAC: 20
United Kingdom
Message 1087206 - Posted: 15 Mar 2011, 0:07:52 UTC - in response to Message 1086823.  

Rough roadmap presented at Lunatics, trying to place the Cuda work in context of surrounding development, and all the collaboration & overlap that goes on.

http://lunatics.kwsn.net/1-discussion-forum/lunatics-experimental-development-roadmap.msg36185.html#msg36185


OK, very good! And a good summary.

Just one small omission...

You've missed off my mil-spec spiky hair!

:-)


Happy fast crunchin' very soon,
Martin

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 1087206 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1087214 - Posted: 15 Mar 2011, 0:21:50 UTC - in response to Message 1087206.  

You've missed off my mil-spec spiky hair!


LoL. Missed out Richard's beard as well, so don't worry, your do will be on the long term chart.

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1087214 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1087250 - Posted: 15 Mar 2011, 1:58:27 UTC - in response to Message 1087214.  

You've missed off my mil-spec spiky hair!


LoL. Missed out Richard's beard as well, so don't worry, your do will be on the long term chart.

See, spec. changes already! lol

I'm sure you guys will bang out some good stuff as always.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1087250 · Report as offensive
aad

Send message
Joined: 3 Apr 99
Posts: 101
Credit: 204,131,099
RAC: 26
Netherlands
Message 1087461 - Posted: 16 Mar 2011, 18:10:35 UTC

On a serieus side; Don't you mean 'installer 0.38'???
Or was that the older version of the roadmap?
ID: 1087461 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1087468 - Posted: 16 Mar 2011, 18:36:20 UTC - in response to Message 1087461.  
Last modified: 16 Mar 2011, 18:37:04 UTC

On a serieus side; Don't you mean 'installer 0.38'???
Or was that the older version of the roadmap?


Blocks in green are the 'done stuff', yellow 'near term', and orange 'medium term'

Existing, recent, completed work I decided to show, in order to give context to current efforts/direction.

Jason
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1087468 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1087489 - Posted: 16 Mar 2011, 19:42:17 UTC

Very nice Jason, it appear my plan is coming together nicely.....so what are we going to do tomorrow you ask? We are going to......TAKE OVER THE WORLD!!!!!

Sorry got side tracked lol ;)
Traveling through space at ~67,000mph!
ID: 1087489 · Report as offensive
aad

Send message
Joined: 3 Apr 99
Posts: 101
Credit: 204,131,099
RAC: 26
Netherlands
Message 1087597 - Posted: 17 Mar 2011, 0:06:41 UTC - in response to Message 1087468.  

Blocks in green are the 'done stuff'


Argh......feelin' stupid now.......
Thats why you are a developer and I just a dumb-ass-cruncher......
ID: 1087597 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · Next

Message boards : Number crunching : AVX Extensions - Ongoing development?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.