People talking crap about source compiles

ScriptDevil · 2006-05-16 09:29:24

I was in a Gentoo chatroomthe other da just to see what the people of clearly the only other distro worth commenting about were saying. They said-> 'Source builds cant be matched by even i686'. I believed them, yes source builds would optimize it for P$ in markt but these pacman binaries were 1686 ish. Lets look at stats.
1) I did the sick gentoo install, the compiled OpenOffice.org from source. OOwriter took abut 3.86 seconds to startup.
2) Now i reinstalled arch then put open office. Surprise it took 3.7 secs
That wasw faster. Thanks to various modifications i suppose before compilataion of pacman binaries on arch.
Any other satisfactory explanation??

myst · 2006-05-16 09:32:02

It depends. I believe that for MPlayer it can make difference.

Gullible Jones · 2006-05-16 10:57:13

Eh, MPlayer can autodetect what stuff your CPU has.

FUBAR · 2006-05-16 11:28:14

Gullible Jones wrote:

Eh, MPlayer can autodetect what stuff your CPU has.

Yes, but when doing so it also throws a warning saying it's not the optimal way.

ScriptDevil · 2006-05-16 12:55:37

Yes, but when doing so it also throws a warning saying it's not the optimal way.

it does, but have u noticed any difference??

FUBAR · 2006-05-16 13:00:54

ScriptDevil wrote:

Yes, but when doing so it also throws a warning saying it's not the optimal way.
it does, but have u noticed any difference??

It works fine for me. But I'm not ABS'ing so I can rebuild Mplayer to get rid of the warning.

nightfrost · 2006-05-16 14:33:06

I've noticed differences in mainly multimedia apps when compiling for pentium-m. I think it's mostly the -sse and -sse2 stuff that speed things up a little.

One interesting thing is, under ubuntu, for example, dragging windows around under gnome lags on my laptop, it's noticeably jerky. Under archlinux they're a little jerky (they sort of leave trails). Xorg & gnome compiled with -march=pentium-m gets rid of those things. It's also very noticable with xcompmgr. When I use xcompmgr to add drop shadows, window movement is sort of slow under i686-builds, but are perfectly useable after a re-compile.

But I hardly think openoffice would get much faster.

phrakture · 2006-05-16 14:38:33

From a sheer technical perspective, building from source with all the wonky USE flags and specialized -O9 CFLAGS is more optimized, yes. But there's a point where that doesn't matter.

If you can have a choice of a car that:
a) Goes 500 km/h and you can have it tomorrow
b) Goes 507 km/h and you can have it in 3 days

Which would you choose? Going "faster" isn't always what a given app needs. Most apps depend on user input anyway, so compiling them with --go-faster isn't going to make *you* move or click the mouse faster, or type faster.

nightfrost · 2006-05-16 14:53:09

phrakture wrote:

Which would you choose? Going "faster" isn't always what a given app needs. Most apps depend on user input anyway, so compiling them with --go-faster isn't going to make *you* move or click the mouse faster, or type faster.

From my experience, you're definitely right on this point. But where "speed" really matters, it's actually more about responsiveness. Like the window moving crap I mentioned above.

Dusty · 2006-05-16 14:53:27

Phrakture, I think you deserve the analogy-of-the century award for that one!!

Dusty

myst · 2006-05-16 14:55:14

phrakture, sure. It's why I tell "it depends". For 0.(0)1% of all progs it makes difference, but it takes too much time, to wait for it again and again.

user · 2006-05-16 15:11:53

Source compile is not only good for speed.
Though, i never ever used gentoo, but arch will do the source compile job for me.

JGC · 2006-05-16 15:16:05

Another thing with source compiles: you spend 4 hours compiling openoffice from source, need 8GB of diskspace to compile it, and in the end, your harddisk is fragmented as hell... then 10 days later a new version or revision of the OpenOffice ebuild is out and you need to re-do the whole process.

Is it worth the problems? I'd rather have a package monkey compiling it for me, so I can install it in a few minutes. Hmm, damn, I happen to be such a thing

Cerebral · 2006-05-16 15:48:33

JGC wrote:

Another thing with source compiles: you spend 4 hours compiling openoffice from source, need 8GB of diskspace to compile it, and in the end, your harddisk is fragmented as hell... then 10 days later a new version or revision of the OpenOffice ebuild is out and you need to re-do the whole process.
Is it worth the problems? I'd rather have a package monkey compiling it for me, so I can install it in a few minutes. Hmm, damn, I happen to be such a thing

*whistles* Bitter much? j/k
We appreciate your work JGC, though, come to think of it, aren't I a package monkey too? :mrgreen:

WillySilly · 2006-05-16 15:54:21

Being a package monkey is a hard life...

In any case, I've never noticed a speed increase when compiling from source

user · 2006-05-16 16:20:11

Arch is the result of source compile(as 5-years debian(i386) user's perspective).
And without doing the source compile, no arch.

And one more thing, i omit previous post,
when you need to use cross hardware that has much performance than x86 or something, but there is no linux on the hardware,

Your choice is source compile.

Hardware also change, it's alive and flexible.

user · 2006-05-16 16:29:37

ugh i hate the fucking number #155
so post this message.

PS: in a hurry.

ScriptDevil · 2006-05-16 16:45:53

numbers like 155 are the silliest things u can hate. But stuff like that make u real c00l

Neuro · 2006-05-16 16:59:48

ScriptDevil wrote:

Yes, but when doing so it also throws a warning saying it's not the optimal way.
it does, but have u noticed any difference??

Actually yes. I had a PIII 650 laptop. After compiling mplayer from source it became much more responsive (seeking through the movie was noticeably faster). But I guess with ~2GHz machines it doesn't make a noticeable difference.

iBertus · 2006-05-16 21:44:55

Yes, Arch is the product of a source build, as is every other software product you can imagine. The difference is that we waste fewer processor cycles on end-user machines than the so-called sourced based distros. If you've got extra processor cycles sitting around, go fold protiens.

Gullible Jones · 2006-05-16 23:52:37

Neuro wrote:

ScriptDevil wrote:
Yes, but when doing so it also throws a warning saying it's not the optimal way.
it does, but have u noticed any difference??
Actually yes. I had a PIII 650 laptop. After compiling mplayer from source it became much more responsive (seeking through the movie was noticeably faster). But I guess with ~2GHz machines it doesn't make a noticeable difference.

Interesting, doing that on a P3 450 desktop didn't change anything... How much RAM does the laptop have? Perhaps certain optimizations affect RAM consumption? Also, was this with an earlier version of GCC (3.3.x) and -fomit-frame-pointer present among your optimizations?

test1000 · 2006-05-17 11:40:59

@Gullbile Jones: I think a source compile should speedup mplayers start atleast because at the start it spends some time doing this:

MPlayer 1.0pre7try2-4.0.3 (C) 2000-2005 MPlayer Team
CPU: Intel Pentium 4/Xeon/Celeron Foster (Family: 8, Stepping: 4)
Detected cache-line size is 64 bytes
CPUflags: MMX: 1 MMX2: 1 3DNow: 0 3DNow2: 0 SSE: 1 SSE2: 1
Compiled with runtime CPU detection - WARNING - this is not optimal!
To get best performance, recompile MPlayer with --disable-runtime-cpudetection.

So if you compile it with --disable-runtime-cpudetection then atleast it will start faster.

Gullible Jones · 2006-05-17 19:47:34

I suppose it might, but I have a hard time imagining MPlayer starting up any faster.

(Also remember... The HDD is the biggest bottleneck, not the CPU.)

user · 2006-05-18 00:27:51

here is my compiled mplayer on arch.

MPlayer dev-CVS-060427-01:45-4.1.0 (C) 2000-2006 MPlayer Team
CPU: Intel Pentium 4/Celeron 4 Northwood; Pentium 4 EE/Xeon Prestonia,Gallatin (Family: 15, Stepping: 7)
CPUflags: MMX: 1 MMX2: 1 3DNow: 0 3DNow2: 0 SSE: 1 SSE2: 1
Compiled for x86 CPU with extensions: MMX MMX2 SSE SSE2

Arch Linux

#1 2006-05-16 09:29:24

People talking crap about source compiles

#2 2006-05-16 09:32:02

Re: People talking crap about source compiles

#3 2006-05-16 10:57:13

Re: People talking crap about source compiles

#4 2006-05-16 11:28:14

Re: People talking crap about source compiles

#5 2006-05-16 12:55:37

Re: People talking crap about source compiles

#6 2006-05-16 13:00:54

Re: People talking crap about source compiles

#7 2006-05-16 14:33:06

Re: People talking crap about source compiles

#8 2006-05-16 14:38:33

Re: People talking crap about source compiles

#9 2006-05-16 14:53:09

Re: People talking crap about source compiles

#10 2006-05-16 14:53:27

Re: People talking crap about source compiles

#11 2006-05-16 14:55:14

Re: People talking crap about source compiles

#12 2006-05-16 15:11:53

Re: People talking crap about source compiles

#13 2006-05-16 15:16:05

Re: People talking crap about source compiles

#14 2006-05-16 15:48:33

Re: People talking crap about source compiles

#15 2006-05-16 15:54:21

Re: People talking crap about source compiles

#16 2006-05-16 16:20:11

Re: People talking crap about source compiles

#17 2006-05-16 16:29:37

Re: People talking crap about source compiles

#18 2006-05-16 16:45:53

Re: People talking crap about source compiles

#19 2006-05-16 16:59:48

Re: People talking crap about source compiles

#20 2006-05-16 21:44:55

Re: People talking crap about source compiles

#21 2006-05-16 23:52:37

Re: People talking crap about source compiles

#22 2006-05-17 11:40:59

Re: People talking crap about source compiles

#23 2006-05-17 19:47:34

Re: People talking crap about source compiles

#24 2006-05-18 00:27:51

Re: People talking crap about source compiles

Board footer