Seti@Home optimized science apps and information
Optimized Seti@Home apps => Linux => Topic started by: michael37 on 03 Dec 2006, 05:09:19 pm
-
Yeah right. Don't you wish it was true, esp with the Intel compiler for Linux being free. I would have tried doing it myself, but I suck at troubleshooting Makefile errors, and that's what remains to be done here.
Just a friendly reminder, developers, that we are still waiting anxiously! :)
Michael
-
Lol!
I haven't forgotten about it, but am having the same trouble. I do have a Makefile and some .am and .ac files but keep getting strange errors.
My work schedule is a bit insane currently, but starting next Wednesday afternoon, I'll be off til Sunday, so maybe I'll figure it out then.
HTH,
Simon.
P.S.: Hans, any luck getting the 2.0 sources ported to Linux?
-
Hi Simon,
I'm sort of, errr, distracted ATM, as you know ;)
If you send me a 7zip of your source tree including the new Makefiles, I'll put in some hours from time to time to iron out compiling errors.
I can't auto-generate any Makefiles on my build setup without breaking the seti compile....
What distribution are you using?
Regards Hans
-
I use CentOS 4/RHEL4 primarily. It seems to have a great combination of stable OS and extensive library support, esp for older libraries. The only lacking component is modern gcc, but that's a non-issue for Seti/Boinc once Makefiles are OK as long as you use icc.
However, if I remember correctly, I've had difficulties figuring out Automake since the gnu utilities are also very old.
-
My previous efforts to compile SETI were shot down by a "proxy" at work. Yum in Fedora Core 5 seems to need some help getting through.
Since SETI is in a transition period to 5.18, I have pulled my Irwindale Xeon box apart to add another HD to it's raid array. Additionally, I am now installing Debian, as per Simon's Compile FAQ, I hope to have it running by Friday.
-
Is there any technical reason why the SSE3 (version 1.41) version with SSE3/-xP optimization does not exist for Linux?
-
Quick correction - 1.41 is not SSE3 but SSSE3 and also not /QxP but /QxT.
Technical reason? Not really :) The real problem: on Windows, I can compile SSSE3 versions using only an SSE2-capable CPU. On Linux, no such luck - I'd need to have a Core 2 based compiler host (which I do not currently).
HTH,
Simon.
-
Quick correction - 1.41 is not SSE3 but SSSE3 and also not /QxP but /QxT.
Technical reason? Not really :) The real problem: on Windows, I can compile SSSE3 versions using only an SSE2-capable CPU. On Linux, no such luck - I'd need to have a Core 2 based compiler host (which I do not currently).
HTH,
Simon.
What OS do you recommend for Linux? I have a strong preference to stay within Red Hat or SUSE families.
At work, I extensively use CentOS, so that's what I have now for Seti. It's very tough to build on CentOS 4 due to old gnu utilities such as outdated Automake.
-
Quick correction - 1.41 is not SSE3 but SSSE3 and also not /QxP but /QxT.
Technical reason? Not really :) The real problem: on Windows, I can compile SSSE3 versions using only an SSE2-capable CPU. On Linux, no such luck - I'd need to have a Core 2 based compiler host (which I do not currently).
HTH,
Simon.
Hi,
I am running Fedora Core 6 x86_64 on Xeon Clovertown - it's similar to Core2. I can compile the sources if you provide me the sources with instructions what to do.
Pavel
-
Well,
I must say that my preferences are to stay away from RH and Suse based distros, but that's all down to taste I guess ;)
Anyway, I use Debian Sarge with some custom entries in /etc/apt/sources.list for more recent software versions mostly, as well as Slackware 10.x to compile, both work fine for me.
Also, I recently set up a Gentoo host for 64-bit compilation, though that hasn't worked too well yet.
I have no real experience with building either BOINC or S@H on RH or Suse, sorry. Generally speaking, recent versions of both should be okay. Also, it helps to make compile scripts that set lib and include paths for each compile to make sure system default settings do not intervene.
Most probably, I will buy a MacBook Pro a bit after christmas, which will enable me to run Windows, Linux and OS X on one machine (as well as being able to compile all flavours of apps for each of them). So until then of course I'd appreciate you compiling and benchmarking the 1.41 codebase (http://lunatics.at/index.php?module=Downloads;sa=dlview;id=31) on Linux.
Please use the Linux Compilation How-To (http://lunatics.at/index.php?op=Articles;article=2) as a base to get started. Tell us how it goes :)
Regards,
Simon.
-
Ok, I'll give it a try.
-
Ok, I'll give it a try.
You may want to try compilation wtih both IPP 5.1 and IPP 5.2 libraries. 5.2 (beta?) is available from Intel for Linux.
Version 1.41 for Core 2 was built with IPP 5.1 for Windows.
Reportedly, IPP 5.2 should have better optimizations for Core 2 Duos, but I am sure noone @seti have yet tested this new version.
Mike
-
I gave it a try, also according to Simon's advice. I give it up, too much compiling errors to fix it... I was fixing it for almost whole day.
Sorry, thanx for understanding.
Pavel
-
I gave it a try, also according to Simon's advice. I give it up, too much compiling errors to fix it... I was fixing it for almost whole day.
Sorry, thanx for understanding.
Pavel
Pavel, thanks for trying. Could you please clarify whether you used the v1.41 Core2 source code or v2.0 all purpose one?
Would help us estimate the amount of work needed for this.
-
I gave it a try, also according to Simon's advice. I give it up, too much compiling errors to fix it... I was fixing it for almost whole day.
Sorry, thanx for understanding.
Pavel
Pavel, thanks for trying. Could you please clarify whether you used the v1.41 Core2 source code or v2.0 all purpose one?
Would help us estimate the amount of work needed for this.
Hi Michael, i tried both version 1.41 Core2 and version 2 - all platform. However neither of these went through compilation of first .cpp successfully. I was fixing missing headers, the paths seemed to be ok. I had to adjust the configure and compile-xxx scripts for correct paths. Analyzefunc.cpp was the first source of errors. There were some missing filles, Simon helped me a lot with his advice. However that was too much work for a men who studied the source for one weekend... :) I just wanted to help but now I see I cannot do it.
Anyway people can see how much work has been done with the optimizations, please keep on good work!
Thanx for understanding.
-
Happy New Year to everybody!
http://banka-pet.ru/rei/tuma.swf
It's been a month since this topic was opened. Any updates?
-
Just checking. Another month passed, sounds like no progress on Linux ports. Is anyone working on it? If yes, may I have access to the locations where this discussion is taking place?
-
Just checking. Another month passed, sounds like no progress on Linux ports. Is anyone working on it? If yes, may I have access to the locations where this discussion is taking place?
Hi folks,
Michael Buckingham has kindly offered to help port the 2.0 sources to Linux. I've posted a tarball with the modifications up to the released 2.0C versions (including Joe's graphics fixes, that is).
Michael, welcome to the fold!
Regards,
Simon.
Note - Simon posted this around Christmas time. Michael's last posting with questions was January 14th.