LinuxQuestions.org
Visit Jeremy's Blog.
Go Back   LinuxQuestions.org > Forums > Linux Forums > Linux - Distributions > Slackware
User Name
Password
Slackware This Forum is for the discussion of Slackware Linux.

Notices

Reply
 
Search this Thread
Old 10-07-2012, 11:53 AM   #76
H_TeXMeX_H
Guru
 
Registered: Oct 2005
Location: $RANDOM
Distribution: slackware64
Posts: 12,928
Blog Entries: 2

Rep: Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269

It might be fun to benchmark against Gentoo. Of course, no self respecting Gentoo user would ever believe them if they came out not showing Gentoo to be far superior.
 
Old 10-07-2012, 12:09 PM   #77
bosth
Member
 
Registered: Apr 2011
Posts: 229

Rep: Reputation: 68
Quote:
Originally Posted by H_TeXMeX_H View Post
It might be fun to benchmark against Gentoo. Of course, no self respecting Gentoo user would ever believe them if they came out not showing Gentoo to be far superior.
No one seems to believe results unless they show their distro being the best... just look at this thread!
 
Old 10-07-2012, 12:40 PM   #78
H_TeXMeX_H
Guru
 
Registered: Oct 2005
Location: $RANDOM
Distribution: slackware64
Posts: 12,928
Blog Entries: 2

Rep: Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269
Quote:
Originally Posted by bosth View Post
No one seems to believe results unless they show their distro being the best... just look at this thread!
I don't expect Slackware to be the fastest, in fact I'd bet on Gentoo, but certainly I don't believe that Slackware is much slower than the other distros like the biased phoronix benchmark shows. I don't expect huge differences, just small ones. My own benchmark, although not in-depth shows that it's not possible for Slackware to be much slower.
 
1 members found this post helpful.
Old 10-07-2012, 01:07 PM   #79
konsolebox
Senior Member
 
Registered: Oct 2005
Distribution: Gentoo, Slackware, LFS
Posts: 2,245
Blog Entries: 15

Rep: Reputation: 233Reputation: 233Reputation: 233
Quote:
Originally Posted by H_TeXMeX_H View Post
It might be fun to benchmark against Gentoo. Of course, no self respecting Gentoo user would ever believe them if they came out not showing Gentoo to be far superior.
With proper settings I expect Gentoo to be superior. But how would you test it anyway? Package-specific tests do not really reflect the overall performance of the system. One could have Slackware and just rebuild a specific package with native settings and it would be almost the same as Gentoo already i.e. if the package does really call much libraries routines. Tests would also vary depending on the instructions that could be included during build and other flags like -floop-block and -fgraphite-identity from graphite. -O3 seemed to yield faster builds (before) but I'm not sure about newer machines now. -unroll-loops as well is not essentially helpful most of the time and just cause too much memory usage. Using graphite is much better to arrange loops to make them work optimized within the native processor's cache.
 
Old 10-07-2012, 01:08 PM   #80
H_TeXMeX_H
Guru
 
Registered: Oct 2005
Location: $RANDOM
Distribution: slackware64
Posts: 12,928
Blog Entries: 2

Rep: Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269Reputation: 1269
I would never use '-O3' unless I wanted an unstable system, but I suppose that's normal for Gentoo.
 
Old 10-07-2012, 01:10 PM   #81
konsolebox
Senior Member
 
Registered: Oct 2005
Distribution: Gentoo, Slackware, LFS
Posts: 2,245
Blog Entries: 15

Rep: Reputation: 233Reputation: 233Reputation: 233
Quote:
Originally Posted by H_TeXMeX_H View Post
I would never use '-O3' unless I wanted an unstable system, but I suppose that's normal for Gentoo.
I used -O3 before with stronger options and my system was stable.
 
Old 10-07-2012, 01:36 PM   #82
konsolebox
Senior Member
 
Registered: Oct 2005
Distribution: Gentoo, Slackware, LFS
Posts: 2,245
Blog Entries: 15

Rep: Reputation: 233Reputation: 233Reputation: 233
Quote:
Originally Posted by jtsn View Post
Wrong conclusion.

Most softwares (like Microsoft Windows) don't include machine-specific instructions, CPU vendors know this and optimize their hardware for it (and not for Gentoo). A modern x86 CPU has an instruction decoder to feed RISC-like execution units with micro-instructions for the optimal execution speed. Of course exotic instructions get a penalty and will be emulated in micro-code.

So as a compiler you are aiming for using only the most generic instructions, because the others are slow.
I'm quite skeptic. Are you sure that's always true?
Quote:
Performance-wise the best thing you can do on modern hardware is to implement multi-threading
Parallel-processing? Not always true in -most- operations, afaik.
Quote:
and make use of specialised units like GPGPU and AESNI. Both you can't do by compiler flags, you have to change the algorithm. Case closed.
I'm not sure if that's the best, but do you think that could even be done with normal packages as well? Or do you mean the packages should be hacked first to have them optimized?

Too bad, but I did consider that. That's old info. I read that page a long time ago. And I don't use unroll-loops. Also, we're not talking about principles here. We're just describing how things really are.
 
Old 10-07-2012, 01:45 PM   #83
konsolebox
Senior Member
 
Registered: Oct 2005
Distribution: Gentoo, Slackware, LFS
Posts: 2,245
Blog Entries: 15

Rep: Reputation: 233Reputation: 233Reputation: 233
Quote:
Originally Posted by jtsn View Post
In modern hyper-threaded superscalar pipelined out-of-order-execution CPUs there is no concept of "cycles". How much cycles an instruction takes depends on the instructions around it, instructions of other threads, which will be executed on the same core or module, the results of the branch prediction and so on...
Basically I just used the term cycles and it's just an example. The concept of gaining efficiency by using encapsulated functions is what I meant.
Quote:
Using "special" CISC instructions from early x86 assembly is a good way to slow your program down.
Well I was actually thinking about some newer SSE2 instructions over SSE or lower, not old ones.
 
Old 10-07-2012, 01:57 PM   #84
ReaperX7
Senior Member
 
Registered: Jul 2011
Distribution: LFS-SVN, FreeBSD 10.0
Posts: 3,413
Blog Entries: 15

Rep: Reputation: 965Reputation: 965Reputation: 965Reputation: 965Reputation: 965Reputation: 965Reputation: 965Reputation: 965
Actually Gentoo's documentation recommends using the -O2 compile flags, not -O3. In fact any and all "from source" distributions all default to using -O2 as their environmental default compile flag.

Being the "fastest" isn't always equal to being the most "stable" in any regards. Real world usage and longevity of a system's stability factors against other software and added dependencies will outweight any synthetic benchmarks by huge volumes.

Here's a suggestion, let you're own usage be your own benchmark and trust distributions you've used that have been stable, offered what you needed, and were the easiest to use, administrate, and maintain over time.

I don't need a benchmark to tell me which system is the best. I just had to figure it out for myself and then use it.
 
Old 10-07-2012, 03:02 PM   #85
konsolebox
Senior Member
 
Registered: Oct 2005
Distribution: Gentoo, Slackware, LFS
Posts: 2,245
Blog Entries: 15

Rep: Reputation: 233Reputation: 233Reputation: 233
@ReaperX7: A vague part of my memory tells that there was actually a time in Gentoo when "-O3 -fomit-frame-pointer -pipe" was recommended, but I'm no longer certain about that. Perhaps if not on the official documentation, it was on a forum post or some updates of list of stable flags somewhere. There might have been a time in Gentoo that even -O3-based compilation was also heavily tested.

With respect to choosing Slackware for stability (and probably, adaptability as well), I won't really question that after all I myself would still prefer Slackware for normal, simple and easier deployments, but with respect to pointing out that Slackware is naturally slower due to the fact that it's more generic and less system-specific, I don't think I could deny myself about that, rationally. It's also based from experience. Anyhow it's not only Slackware that distributes packages generically, I think. There are many distros that are generic as well, but mainly as how I composed my first post I mainly compare it to more optimized distros especially Gentoo or else the thought would be without meaning.

I would like to add something more to the idea: There might be not much difference perhaps on 64 builds as I'm not sure if the compiler makes use of SSE3+ instructions already as I tried to force it but it didn't show on binary dump analysis; maybe the tasks are just rare for the new instructions to be used or perhaps just bad config on my part, or the compiler was patched, not sure, but on 32/i486 there are notable ones. I'm not also sure if Slackware 64 makes use of SSE2 already since it's also available in AMD products I think.
 
Old 10-07-2012, 04:07 PM   #86
TobiSGD
Moderator
 
Registered: Dec 2009
Location: Hanover, Germany
Distribution: Main: Gentoo Others: What fits the task
Posts: 15,592
Blog Entries: 2

Rep: Reputation: 4047Reputation: 4047Reputation: 4047Reputation: 4047Reputation: 4047Reputation: 4047Reputation: 4047Reputation: 4047Reputation: 4047Reputation: 4047Reputation: 4047
Any x86_64 CPU has SSE2 capabilities, so it would be somewhat foolish not to use it.
 
Old 10-07-2012, 04:29 PM   #87
konsolebox
Senior Member
 
Registered: Oct 2005
Distribution: Gentoo, Slackware, LFS
Posts: 2,245
Blog Entries: 15

Rep: Reputation: 233Reputation: 233Reputation: 233
@TobiSGD: I had the idea but I just gave an allowance. Reading wikipedia somehow confirms it.
 
Old 10-07-2012, 04:52 PM   #88
kikinovak
Senior Member
 
Registered: Jun 2011
Location: Montpezat (South France)
Distribution: Slackware, Slackware64
Posts: 1,735

Rep: Reputation: 833Reputation: 833Reputation: 833Reputation: 833Reputation: 833Reputation: 833Reputation: 833
Quote:
Originally Posted by H_TeXMeX_H View Post
It might be fun to benchmark against Gentoo. Of course, no self respecting Gentoo user would ever believe them if they came out not showing Gentoo to be far superior.
http://funroll-loops.info/
 
1 members found this post helpful.
Old 10-07-2012, 07:02 PM   #89
jtsn
Member
 
Registered: Sep 2011
Location: Europe
Distribution: Slackware
Posts: 806

Rep: Reputation: 362Reputation: 362Reputation: 362Reputation: 362
Quote:
Originally Posted by konsolebox View Post
I'm quite skeptic. Are you sure that's always true?
Mainstream compilers like Visual C++ only use generic instructions, so CPU vendors are optimizing there to look good in mainstream application benchmarks. They don't go to Phoronix.

Quote:
Parallel-processing? Not always true in -most- operations, afaik.
Even on a dual core Atom CPU, you can gain almost factor 4 by just using openssl -multi 4 with AES128. And of course you can skyrockit it by employing AESNI (not on the Atom though), which a compiler can't do automatically.

Quote:
Originally Posted by konsolebox View Post
Basically I just used the term cycles and it's just an example. The concept of gaining efficiency by using encapsulated functions is what I meant.
What do you mean with "encapsulated functions"?

Quote:
Well I was actually thinking about some newer SSE2 instructions over SSE or lower, not old ones.
SSE is just single-precision floating-point, which isn't very useful outside of games. SSE2 sometimes is faster than i387, but also at the cost of precision. So you can break applications by applying SSE2 via GCC flags.

Instruction set extensions are best to be used by developers of performance-sensitive applications at the corresponding hot-spots in the code.
 
Old 10-07-2012, 08:35 PM   #90
konsolebox
Senior Member
 
Registered: Oct 2005
Distribution: Gentoo, Slackware, LFS
Posts: 2,245
Blog Entries: 15

Rep: Reputation: 233Reputation: 233Reputation: 233
Quote:
Originally Posted by jtsn View Post
Even on a dual core Atom CPU, you can gain almost factor 4 by just using openssl -multi 4 with AES128. And of course you can skyrockit it by employing AESNI (not on the Atom though), which a compiler can't do automatically.
So these instructions do processing silently in multiple internal threads?
Quote:
What do you mean with "encapsulated functions"?
Well just like AESNI. It does one big step easily with one instruction as compared to calling many generic instructions. I refer that one instruction as an encapsulated function.
Quote:
SSE is just single-precision floating-point, which isn't very useful outside of games. SSE2 sometimes is faster than i387, but also at the cost of precision. So you can break applications by applying SSE2 via GCC flags.

Instruction set extensions are best to be used by developers of performance-sensitive applications at the corresponding hot-spots in the code.
I haven't examined instructions much yet but it's not only SSE2 that's around now. What do you think about STTNI of SSE 4.2?

Last edited by konsolebox; 10-07-2012 at 08:46 PM. Reason: Wrong idea.
 
  


Reply

Tags
benchmarking, performance


Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is Off
HTML code is Off


Similar Threads
Thread Thread Starter Forum Replies Last Post
Lightest "jukebox" app/player for Debian? lite compared to rhytmbox/amarok? linus72 Linux - General 4 11-02-2009 08:55 PM
linux distribution supporting "2.6.18-8.1.10.el5" or "2.6.16.21-0.8-smp" mrpc_cambodia Linux - Kernel 3 10-08-2009 02:43 AM
Shouldn't "Slackware64" Become just "Slackware" and 32-bit Become "Slackware32"? foodown Slackware 6 06-23-2009 01:24 PM
Slackware = "most unix-like" distribution ? caustic386 Slackware 41 04-08-2009 08:26 PM
Poor 3d performance with ATI Radeon 7500LE desipte "Direct rendering enabled" tallman Linux - Hardware 5 06-16-2004 11:31 PM


All times are GMT -5. The time now is 06:10 PM.

Main Menu
My LQ
Write for LQ
LinuxQuestions.org is looking for people interested in writing Editorials, Articles, Reviews, and more. If you'd like to contribute content, let us know.
Main Menu
Syndicate
RSS1  Latest Threads
RSS1  LQ News
Twitter: @linuxquestions
identi.ca: @linuxquestions
Facebook: linuxquestions Google+: linuxquestions
Open Source Consulting | Domain Registration