You are not logged in.
I'm wondering why noone is compilng stuff for pentium4 procesors.
I personaly always compile only kernel for pentium4 because I'm using Arch on my laptop and longer compilations are just killers for my fan.
What I (and I guess many other people) would like to see are esential packages compiled for pentium4 procesors. Things like:
standard kernel, glibc, kde, qt, gnome, gcc and maybe other soft stored in one "i886 arch repo" could be interesting for many people.
I wonder what te difference in speed could be.
What do you think about this idea?
Offline
I will put down $300 right now that says if you were to benchmark i686 v pentium4, you would not get over a 1% speed increase (on the same machine).
Offline
I've never heard of i886. Feel free to start a port though, that's what the i586, ppc, and x86_64 people are doing. Nobody would do it for you.
Dusty
Offline
Hmm. So there will be no gain when compiling things with -march higher than 686, you say. Well :oops: - silly me.
Anyhow - if there are any (over)tweakers out there that would like to try (for < 1% maybe or because of gentoo-envy or something ;-) ) ... . It's just greed for speed. :-)
Offline
For those who belives:
int main() {
long double x=2;
for (int i = 0 ; i < 100000; ++i) {
for (int j = 0 ; j < 100000; ++j) {
x *=2;
x +=1;
}
}
}
g++ bla.cc -march=i686 -O2 -s -pipe -ffast-math -o i686 && g++ bla.cc -march=pentium4 -O2 -s -pipe -ffast-math -o i886
(these are repetable (no disc reading etc. - i've run them several times))
$ time ./i886
real 0m6.021s
user 0m5.904s
sys 0m0.001s
$ time ./i686
real 0m8.020s
user 0m7.958s
sys 0m0.001s
And look what will happen when we change long double to int:
$ time ./i686
real 0m8.700s
user 0m8.144s
sys 0m0.002s
$ time ./i886
real 0m6.523s
user 0m6.143s
sys 0m0.002s
In both cases 'i886' files were smaler than 'i686'.
I know this is not real application, but IMO the idea defends a little, don't you think?
Offline
I know this is not real application, but IMO the idea defends a little, don't you think?
Oh no, don't get me wrong - it works great for contrived examples and benchmarks involving simple math, however, most of your apps are going to spend more time waiting for I/O or user input than doing calculations, so the speed increases will be effectively canceled out.
That said, I know many people compile their kernel specifically for their architecture - which makes sense, and may be beneficial to you.
I don't think the gain is enough to warrant an entirely new architecture port, though - the i586 makes sense because i686 won't run on a 586, but the existing packages run just fine on an 886.
I would say build a custom kernel and see where that gets you - if things feel too slow for you, only recompile those applications affecting you directly (as it's going to be alot of work maintaining - this isn't gentoo).
Offline
I would say build a custom kernel and see where that gets you - if things feel too slow for you, only recompile those applications affecting you directly (as it's going to be alot of work maintaining - this isn't gentoo).
As I said I do not wish to recompile everything - only esential apps./libs. I've got laptop - compiling glibc takes 6 - 8h - this is what I wish to avoid. Someone with bigger cooler could do that for more people and this way safe world from global warming.
If only there were glibc, qt and kde compiled for pentium4 somewhere out there ...
Offline
If only there were glibc, qt and kde compiled for pentium4 somewhere out there ...
you want someone else to do your work for you?
Offline
you want someone else to do your work for you?
Always...
And where were all the sportsmen who always pulled you though?
They're all resting down in Cornwall
writing up their memoirs for a paper-back edition
of the Boy Scout Manual.
Offline
/me claps
"Be conservative in what you send; be liberal in what you accept." -- Postel's Law
"tacos" -- Cactus' Law
"t̥͍͎̪̪͗a̴̻̩͈͚ͨc̠o̩̙͈ͫͅs͙͎̙͊ ͔͇̫̜t͎̳̀a̜̞̗ͩc̗͍͚o̲̯̿s̖̣̤̙͌ ̖̜̈ț̰̫͓ạ̪͖̳c̲͎͕̰̯̃̈o͉ͅs̪ͪ ̜̻̖̜͕" -- -̖͚̫̙̓-̺̠͇ͤ̃ ̜̪̜ͯZ͔̗̭̞ͪA̝͈̙͖̩L͉̠̺͓G̙̞̦͖O̳̗͍
Offline
/me claps
holy crap, you're alive?
Offline
maybe...
"Be conservative in what you send; be liberal in what you accept." -- Postel's Law
"tacos" -- Cactus' Law
"t̥͍͎̪̪͗a̴̻̩͈͚ͨc̠o̩̙͈ͫͅs͙͎̙͊ ͔͇̫̜t͎̳̀a̜̞̗ͩc̗͍͚o̲̯̿s̖̣̤̙͌ ̖̜̈ț̰̫͓ạ̪͖̳c̲͎͕̰̯̃̈o͉ͅs̪ͪ ̜̻̖̜͕" -- -̖͚̫̙̓-̺̠͇ͤ̃ ̜̪̜ͯZ͔̗̭̞ͪA̝͈̙͖̩L͉̠̺͓G̙̞̦͖O̳̗͍
Offline
actually, I'm working on a repo optimized for pentium-m, so far including gtk, qt, X, e17, (and a custom kernel made for lifebook p7010s). I might add kdelibs and kdebase when they hit final, as well as some essential gnome packages (don't know which). When all that's done I'll go public with the repo. I have some questions though, if you guys might feel like answering:
should I add glibc, as well?
would -march=pentium-m packages work flawlessly under pentium4?
which gnome packages would be most beneficial to tweak?
would there be any interest in a repo like this? I've been curious whether it would be frowned upon...
Offline
dpc wrote:If only there were glibc, qt and kde compiled for pentium4 somewhere out there ...
you want someone else to do your work for you?
Is this so suprising or bad? I've got a feeling that you're looking at me with aversion just because I haven't said "Well here is my repo where I'm compiling stuff optimized for pentium4. Go and use it."
If my intuition is right please tell me what is wrong in looking for people who do such thing anyway and would like to share their work with others (including me).
What I'm tring to achive is (in order):
- get some opinions;
- find people who like the idea;
- in bigger group find out what next (repo hosting, cpus for compiling and such);
If someone is repeting "your tring to find someone to do your job, your tring to find someone to do your job" then I think he missed whole Open Source idea ...
Anyhow, I still belive that I've just misunderstood your intentions and so please ignore my post in such case.
Offline
should I add glibc, as well?
Yes, why not. It's very essential lib.
would -march=pentium-m packages work flawlessly under pentium4?
I can only guess that yes, but more googlin' is needed.
would there be any interest in a repo like this?
If you'll find attention then sure. Pentium-M are used in mobile computers and on such machines long compilations are things that many would like to avoid. The more people you (we) find, the more hands there will be to divide work on.
If march=pentium-m will work on my machine then you've got one user already. Till I'll find/create pentium4-repo of course pentium-m would be better than plain i686.
Offline
thanks for the feedback. Let's see what we can do. I've been googling alot on pentium-m code on pentium4 but I haven't found anything useful (that I understand). As soon as I'm done with glibc, I'll upload glibc, qt, gtk, xorg6.9R2 for testing if you would want to. They're all based on the standard PKGBUILDs.
Offline
srcpac is an easy solution for this. pacman -S srcpac
If you wanted you could use another vt during an Arch install to install srcpac and then maybe alias 'pacman' to it or a middle-man type script to install everything from source. Or even better, you could install Gentoo!
Offline
Superb! but ...
As I can see I've got only sse sse2, no sse3 extension in /proc/cpuinfo.
Looking here http://en.wikipedia.org/wiki/Pentium_M and there http://en.wikipedia.org/wiki/Pentium_4 I've see my cpu is lacking those SSE3.
I can try, but I expect failure. Don't worry - I belive there are many who would like to try.
Offline
The sse3 extension isn't coming until january, with the yonah CPUs. My old Dothan only has sse and sse2 so it should be all good
Offline
srcpac is an easy solution for this. pacman -S srcpac
If you wanted you could use another vt during an Arch install to install srcpac and then maybe alias 'pacman' to it or a middle-man type script to install everything from source. Or even better, you could install Gentoo!
You're missing the point. Read through posts (yes I know my English isn't perfect).
"Gentoo users are most important reason of global warming." Why to compile everyting when 10/90 rule says that only compilation of most important parts will give you noticable effect. And why to compile anything when someone could have done it before. We are using very cool binary distro and let use this strength - not making redundant work over and over.
The vision I see is using standard Arch i686 base and few CPU-specific repos (only for those who are not afraid of problems) with essential packages that every "tweaker" would compile for himself anyway.
Offline
The sse3 extension isn't coming until january, with the yonah CPUs. My old Dothan only has sse and sse2 so it should be all good
Cool! So I'm waiting with great hope for your move. (but don't be in haste - this hope is very patient).
Offline
great! I've noticed there are new versions of gtk2 and qt in the official repos, so I'm gonna rebuild them too. I'll post here as soon as it's done.
Offline
Just for clarify - my CPU is Intel(R) Celeron(R) CPU 2.60GHz. I'm not an expert but I think (hope) this changes only performance, because it's based on pentium4 core (no HT & lower cache). Which core - I don't know - how to see that?
Full list of features:
flags: fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr
Offline
I've got "est" and "tm2", but you lack them. And you've "pse36" "ht" "cid" "xtpr", but I lack them.
I have no clue as of what they are.
Offline
srcpac is an easy solution for this. pacman -S srcpac
If you wanted you could use another vt during an Arch install to install srcpac and then maybe alias 'pacman' to it or a middle-man type script to install everything from source. Or even better, you could install Gentoo!
But we all left gentoo for that reason didn't we ;-)
I think what we want is a binary repo optimized for pentium4 or pentium-m or whatever. As it so happened, I'm doing the compiling now. But I wouldn't expect everyone else to compile as well; that would just be a waste of global cpu power, wouldn't it
Offline