Hello

I just wonder how much difference the -O argument CFLAG really makes. Usually -O2 is standard, as I understand, for most systems i486. I have used -O3 , but I cant make any difference in performance. I have even read somewhere that it's possible to use up to -O9 Optimization. I know that some optimization only implies on newer CPUs.

I just wonder if it realy is worth putting effort to this matter in reality.

Amd Athlon XP 2000+

-O2 and -O3 does usually make a difference.

In general:
-O1 performs only optimisations that don't effect the compile time much.
-O2 performs optimisations without trading space for speed
-O3 performs maximal speed optimisation
-Os tries to make the executable as small as possible

-O4 through -O9 are just legacy synonyms for -O3

Obviously, the effects that you will get performance wise depend largely on what code you are optimising. Programs that read and write a lot to hardware (like a disk) without doing much computation won't generally notice much difference (as the bottle-neck here is waiting for the hardware).

If you want speed, use scalable algorithms (merge-sort rather than bubble-sort). For real speed, use assembler on the bits that get run a lot.

— Robert J. Lee

yea it does make a difference -- also you should be using
-march=athlon-xp

compiler optimizations are actually quite application specific as far as performance goes.

-g is to be avoided at all costs unless you need debugging symbols for some reason

The downside to all this fun is some programs will not compile and or run properly with agressive optimizations..

example:
i compiled cvs with
-march=pentium4 -O3 -funroll-loops -ftracer -momit-leaf-frame-pointer -fprefetch-loop-arrays
which is generally what i use for my P4 system all the time
later when i go to use it everything i do make it segfault and i have to recompile without the flags.

for some applications that are specific to a certain type of calculations -O3 can actually be slower than
-O2
or -O2 slower than -O1 with just some of the flags included in but not all.
for instance this is alot better (30% better) at math on my system than what i put above
-march=pentium4 -O1 -finline-functions -mieee-fp -fforce-mem -fschedule-insns -fstrict-aliasing -fomit-frame-pointer -fmove-all-movables -funroll-loops -ftracer -momit-leaf-frame-pointer -fprefetch-loop-arrays

for some real fun use this
to figure out what exactly works best on your system
it is time well spent (if you'r a real nerd and like this sort of thing)

What version of GCC are you planning on using? If you're using 3.4, don't combine -O3 and -ftracer, as it causes problems