Status of the ATI GPU Version ?

Moderator: BarsMonster

Post Reply
DarkPrince
Posts: 148
Joined: Wed Oct 15, 2008 5:50 pm

Re: Status of the ATI GPU Version ?

Post by DarkPrince » Fri Dec 26, 2008 2:43 am

BarsMonster wrote:Development notes: Boost is so great :-) Looks like I had too little time to improve my C++ skills over last 8 years or so. It just have almost everything I needed to make BarsWF crossplatform :-) Shame on me I was not looking at boost for so long time... :-)

Shame on you as a new recruit

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Fri Dec 26, 2008 2:52 am

DarkPrince wrote:Shame on you as a new recruit
At least I have an excuse :-)
During my life I never had commercial C++ projects, only Java (with it's long list of related technologies), C#, Delphi, PHP and even Perl ones.
That is kinda crazy - I was working only with technologies which I HATE :crazy:

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Fri Dec 26, 2008 3:17 am

BarsMonster wrote:At least I have an excuse :-)
Hmm... probably that is not an excuse.

DarkPrince
Posts: 148
Joined: Wed Oct 15, 2008 5:50 pm

Re: Status of the ATI GPU Version ?

Post by DarkPrince » Fri Dec 26, 2008 3:58 am

BarsMonster wrote:
BarsMonster wrote:At least I have an excuse :-)
Hmm... probably that is not an excuse.
Erything is an exuce.

The hard part is finding a good one

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Fri Dec 26, 2008 5:39 am

Boost developers are crazy... I am shocked...
Downloaded 17Mb sources archive, compiled for 32-bit and 64-bit MSVC compilers, and it took 6Gb :str: just to store all the result libs... And I still need to compile it for Intel compiler... Also, compilation took around 1 hour on my C2D@4Ghz for both 32-bit & 64-bit, provided that everything was in cache(and with unlimited write-caching) with ~70% average CPU load. 1 hour of cpp compiler & linker work :mad: :str: :mad: That is insane, that is biggest piece of CPP code I've ever seen...

That is 100% crazy. :crazy: :crazy: :crazy:

DarkPrince
Posts: 148
Joined: Wed Oct 15, 2008 5:50 pm

Re: Status of the ATI GPU Version ?

Post by DarkPrince » Fri Dec 26, 2008 5:53 am

BarsMonster wrote:Boost developers are crazy... I am shocked...
Downloaded 17Mb sources archive, compiled for 32-bit and 64-bit MSVC compilers, and it took 6Gb :str: just to store all the result libs... And I still need to compile it for Intel compiler... Also, compilation took around 1 hour on my C2D@4Ghz for both 32-bit & 64-bit, provided that everything was in cache(and with unlimited write-caching) with ~70% average CPU load. 1 hour of cpp compiler & linker work :mad: :str: :mad: That is insane, that is biggest piece of CPP code I've ever seen...

That is 100% crazy. :crazy: :crazy: :crazy:
Crazy like a fox

Corni

Re: Status of the ATI GPU Version ?

Post by Corni » Fri Dec 26, 2008 8:07 am

yeah, and it's even mroe crazier that the biggest part of boost is stuff you include, not stuff you link against 8except a few libraries)!

neinbrucke
Posts: 82
Joined: Sun Nov 02, 2008 8:53 pm

Re: Status of the ATI GPU Version ?

Post by neinbrucke » Fri Dec 26, 2008 11:49 am

is boost that much easier or more powerful then just using posix threads? :)

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Fri Dec 26, 2008 5:37 pm

neinbrucke wrote:is boost that much easier or more powerful then just using posix threads? :)
Do we have posix threads on windows? Or you mean use native function undes *nix and some proxy library under windows?
That might we worth it.

Corni

Re: Status of the ATI GPU Version ?

Post by Corni » Fri Dec 26, 2008 8:13 pm

There's a pthreads port to win32, somewhere at SF ;)

neinbrucke
Posts: 82
Joined: Sun Nov 02, 2008 8:53 pm

Re: Status of the ATI GPU Version ?

Post by neinbrucke » Fri Dec 26, 2008 9:55 pm

yeah, i use it for a multithreaded version of rcracki. i first used windows threads, then thought it would be nice to have multiplatform support... it seemed to me that boost is more 'high level' then pthreads (and windows threads). but i could be wrong, i just choose pthreads as it seemed like a faster switch coming from the windows thread model :P (you could also take a look at OpenMP, although i somehow doubt it will make things faster for you... you seem to me like someone that wants to do it all by himself, low level, tweaking every bit :crazy: )

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Fri Dec 26, 2008 10:38 pm

neinbrucke wrote:you seem to me like someone that wants to do it all by himself, low level, tweaking every bit :crazy: )
Yes, I want to make all by myself :-) But unfortunately I have so little time each day (just 24 hours) :crazy:, so I have to use 3rd party things :-)

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Sat Dec 27, 2008 1:18 am

Mind dump: our peak performance target is 80TFlop - that's what shows top super computer in exUSSR, we have to beat him :-)
Hmm... Is that possible to implement linpack for a distributed supercomputer?

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Sat Dec 27, 2008 1:20 am

BarsMonster wrote:Mind dump: our peak performance target is 80TFlop - that's what shows top super computer in exUSSR, we have to beat him :-)
Hmm... Is that possible to implement linpack for a distributed supercomputer?
Damn
In particular, the operation count for the algorithm must be 2/3 n^3 + O(n^2) double precision floating point operations.

User avatar
the_drag0n
Posts: 217
Joined: Thu Oct 02, 2008 6:48 am

Re: Status of the ATI GPU Version ?

Post by the_drag0n » Sat Dec 27, 2008 9:23 am

80TFLOPS should be no problem. if you started a distributed client you would easily find hundrets willing to try it :P

Irythros
Posts: 20
Joined: Mon Nov 10, 2008 10:00 pm

Re: Status of the ATI GPU Version ?

Post by Irythros » Sat Dec 27, 2008 11:01 am

the_drag0n wrote:80TFLOPS should be no problem. if you started a distributed client you would easily find hundrets willing to try it :P
Many to try it.
Few to contribute power to it.

User avatar
the_drag0n
Posts: 217
Joined: Thu Oct 02, 2008 6:48 am

Re: Status of the ATI GPU Version ?

Post by the_drag0n » Sat Dec 27, 2008 11:16 am

freerainbowtables is now having about 1500 users over boinc from which about 1000 are activ... why should bars wf not be able to do the same ?!
80 TFLOPS means 80 people with a 4870. but if you have 500 people with a 8800gt, that will do it as well :)

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Sat Dec 27, 2008 11:23 am

Compillers will not let me die bored :crazy:
This code(sometimes) produces different results in debug and release mode in MSVC compiler(everything was fine in Intel C compiler, but it is being downloaded again now) :-S

Code: Select all

		int tmp = (unsigned int)floorl(fabsl(sinl(i + 1))*(powl(2.0,32)));
had to replace with

Code: Select all

		int tmp = (int)(boost::int64_t)floorl(fabsl(sinl(i + 1))*(powl(2.0,32)));
Which looks quite silly.

Took a couple of of hours to find the cause in the code which was working quite long ago...
The difference is that release code tends to "saturate" values which are out of int range :-S

Spaztikdude
Posts: 16
Joined: Thu Oct 16, 2008 1:27 am

Re: Status of the ATI GPU Version ?

Post by Spaztikdude » Sat Dec 27, 2008 11:44 am

Less talk, more code!

*cracks whip*

80 teraflops should be easy.

neinbrucke
Posts: 82
Joined: Sun Nov 02, 2008 8:53 pm

Re: Status of the ATI GPU Version ?

Post by neinbrucke » Sat Dec 27, 2008 1:20 pm

no, i like the talk :P

were the problems introduced when you started using boost? or is that just coincidence?
do i understand right, result of floorl is first cast to an int64_t, then to int.... why the step in between ?
(btw, my c(++) isn't that great, so i may be asking stupid questions ;))

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Sat Dec 27, 2008 7:41 pm

neinbrucke wrote:no, i like the talk :P

were the problems introduced when you started using boost? or is that just coincidence?
do i understand right, result of floorl is first cast to an int64_t, then to int.... why the step in between ?
(btw, my c(++) isn't that great, so i may be asking stupid questions ;))
When I introduced boost I started to run tests over the code and found that it is not working in release move in MSVC compiler. Ofcourse I've thought that boost is the cause, but all places where it was used now (not a complete boost "integration") produced correct results.

Well, if I cast directly to int, or to unsigned int, for some reason compiler was not able to deal with values greater than 2^31+1, so that the result value was 0x80000000 no matter what was long double result.
Well, I understand that compiler behavior is not defined when casting from double -> int for values out of int range, but it is still strange that result was different between debug and release mode.

neinbrucke
Posts: 82
Joined: Sun Nov 02, 2008 8:53 pm

Re: Status of the ATI GPU Version ?

Post by neinbrucke » Sat Dec 27, 2008 7:45 pm

strange indeed :)
but this case is not on it's own... VC++ has more weird differences between debug and release... (just can't recall what my problem with it was some months ago :P)
(and tnx for the explanation)

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Mon Dec 29, 2008 1:24 am

budden wrote:BarsMonster

Why don't you wanna release 0.9 version with hash-list ability?

You've spent over 2 month on ATI-version and we're still stuck with 1-hash-only version.
Sorry to disappoint you, but 0.9 will not have multi-hash feature anyway.
Time was spent not just for ATI version, it is total code refactoring to support multiple algorithms, rewriting non-portable code, and only a small fraction of time for ATI code.

I am releasing 0.9 to have some public testing of ATI core before going "production" :-)

Corni

Re: Status of the ATI GPU Version ?

Post by Corni » Mon Dec 29, 2008 4:23 am

will there be any chance for getting this project open source?
especially when you add the distributed stuff this would be very interesting ;)

User avatar
BarsMonster
Site Admin
Posts: 1118
Joined: Wed Oct 01, 2008 7:58 pm

Re: Status of the ATI GPU Version ?

Post by BarsMonster » Mon Dec 29, 2008 5:13 am

Corni wrote:will there be any chance for getting this project open source?
especially when you add the distributed stuff this would be very interesting ;)
This was already widely discussed on the forum earlier.
The short answer is no.
Although, there are ways to contribute to the project.

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest