To -Os or not to -Os?

[/usr/Bananaman]

How come everyone on the forums seems to love optimizations with -03 -pipe -fomit....etc X 4 lines?...Is there actually noticable performance enhancements? Nobody really talks about -Os? which seems a little strange...I would have thought this would be a better switch to be flicking - I think I read somewhere that the gentoo sources are compiled with this? - yet in CFlags Central all the go is -03 etc....?

Anyhoo, My real question is does anyone know about -Os? I've been compiling with -Os and I'd like to know if the switches that you would normally use with -03 (like -fomit-frame-pointer) would make -Os more responsive? The reason I ask is because I like a quick responsive system - and programs compiled with -Os seem to load pretty damn quick! or would these switches just bloat out the compile?

[zhenlin]

-Os -- optimize for size:

man gcc:

[Lovechild]

I used to think that smaller binaries loaded faster, but I found that lib. lookup count makes more of a difference than binary size. So optimize the hell out of it and use ELF prelinking, get the best of both worlds.

[zypher]

I have two systems here, one Duron 700@900/512MB and one PII300/96MB.
Both of them running gentoo.-1.4-r1 compiled with -0s -pipe -ffast-math.
In my personal view and feeling the difference is noticeable on the pII-Laptop (small though).
The Duron does not seem to be more responsive (lots of mem, UDMA100-Hdds). And -O3 -flotsaflags was not faster on that machine.
YMMV as always, but I'll stay with the -Os flag.
And btw, check out freebench (iirc), there you can check your settings.
_________________
linux user 65882

[AlterEgo]

I think zypher's got it right:
on a big and modern PC the difference is too small to be noticed: you'd probably need the linux equivalent of Bapco's Sysmark or PCMark to be able to measure the difference.

I use:
CFLAGS="-march=athlon-xp -mmmx -m3dnow -msse -mfpmath=sse -O3 -pipe -fomit-frame-pointer -funroll-loops -fforce-addr -frerun-cse-after-loop -frerun-loop-opt -falign-functions=4"

Say it's 2% faster than:
CFLAGS="-march=athlon-xp -mmmx -m3dnow -msse -O2
it certainly takes close to twice as long to compile stuff.

Therefore, altogether, it just might be a very bad "investment" to use these high optimizations (I'm forgetting the occasional ebuild breaking under high optimizations).

The good thing is: it's all up to your own choice and taste.

[/usr/Bananaman]

Thanks for clearing that up guyz!
For my slower machine I'm going to keep compiling with -Os and the usual flags I used to use (which was my q) and I'll switch my main computer back to -03 -flotsflags...

[/usr/Bananaman]

If anyone is interested this site cleared up a little bit for me http://www.goof.com/pcg/os2/tips.html I know it says for os2 but it talks about gcc and explainsa couple of useful tags including extras to use with -Os