CFLAGS Central (Part 1)

Houdini · Posted: Fri Jun 21, 2002 4:32 pm Post subject:

Ahem, from the gcc Man page:

(made to look a little more neat than the man page)

AutoBot · l33t Joined: 22 Apr 2002 Posts: 968 Location: Usually Out

Nice post, will save pacman and others alot of time.
_________________
This message self destructed a long time ago.

DArtagnan · Posted: Fri Jun 21, 2002 4:53 pm Post subject: 03

delta407 · Posted: Fri Jun 21, 2002 4:56 pm Post subject:

slik · Posted: Fri Jun 21, 2002 6:19 pm Post subject:

-fomit-frame-pointer is not turned on when I pass an -O option to my duron 1200, I doubt it is for early pentiums either.

See this thread for some more optimizing discussion (I touch on -fomit-frame-pointer more there)

Houdini · Posted: Fri Jun 21, 2002 8:32 pm Post subject:

No, X86 cannot. -O3 won't turn it on because it horribly, horribly breaks debugging. It is recommended that you don't do that. Meh.

Has anyone checked to see what sort of speed increase it gives?
_________________
^]:wq

delta407 · Posted: Fri Jun 21, 2002 8:36 pm Post subject:

Well, it does free up a register, which is a very limited resource. I've found my system becomes more responsive with -fomit-frame-pointer than without. I don't have numbers, but it really does "feel" faster.

DArtagnan · Posted: Fri Jun 21, 2002 8:43 pm Post subject: x86

Well, you said that x86 can not use these optimizations but i see that mine appends those flags and compiles fast...even it's very fast.
_________________
All for one and one for All
--

MACPRO machine...

delta407 · Posted: Fri Jun 21, 2002 8:51 pm Post subject:

No, he said that x86 can't debug when these optimizations are on. But, since you're not debugging, it doesn't much matter, does it?

Sivar · Apprentice Joined: 25 May 2002 Posts: 266 Location: USA

I compile all my apps with:

CFLAGS="-s -march=i686 -O2 -pipe -fomit-frame-pointer -fexpensive-optimizations -frerun-cse-after-loop -frerun-loop-opt -malign-functions=4"

I do not use -O3 because the overhead of the 'pusha' and 'popa' instructions plus the overhead of the jump instruction isn't really that big of a deal for most software. The -O3 can potentially make software slower because the code size is larger. Really, what takes the longest to load a program?
- Limitation of a modern CPU at processing instructions
- Limitation of a modern hard drive physically moving the head to physically read data in all of the physical places which it is stored in.
I'll give you two guesses.

I find that the loading of software is the slowest part, at least on my computer, and that once it is running it generally works fine. There are a few specific things I build with -O3, such as QT, GCC, gzip, bzip2, GTK+, and a few others.

Note that any programmer worth their salt will manually inline small and important functions in their code using the "inline" keyword. -O3 makes no distinction between fast and slow code and may very well inline a function 20 times that is executed exactlyt 20 times, such as in init code. This will definitely make the program load more slowly, will use more memory, and will make it larger without any increase in speed. Note that functions inlined in the code are inlined by GCC if you use ANY optimization level otehr than -Os and -O0.
_________________
The greatest deeds are still undone, the greatest songs are still unsung...

Dolio · l33t Joined: 17 Jun 2002 Posts: 650

I tried recompiling my system with Autobot's list of fun flags. Then after about 3 hours, the build failed at xscreensaver. I figured it was just the flags, so I unmerged it, and started the build over. Everything only took 6 hours (less than I was expecting

). However, after all was said and done, I tried taking out flags one by one, without much success. So I decided to change the flags to just "-march=i686 -O2 -pipe", and found out that it doesn't compile at all anymore under any kind of circumstances that I can make (I've tried rebooting, as well).

The errors it seems to fail with are all kinds of undefined references (to _XmStrings, XmCreateSeparatorGadget, ..., blurb, etc) mostly in demo-Xm-widgets.o, but also in prefs.o, remote.o, dpms.o, and exec.o.

Have you encountered this Autobot (or anyone for that matter)? Are these references to the stuff in XFree86, or somewhere else, does anyone know (so if possible I can avoid recompiling a whole bunch of stuff needlessly)? I guess if worse comes to worst I can just do emerge -e xscreensaver and just recompile the stuff it directly depends on (and hopefully avoid some of the 6 hours of rebuilding). Thanks for the help.
_________________
They don't have a good bathroom to do coke in.

Sivar · Apprentice Joined: 25 May 2002 Posts: 266 Location: USA

Has anyone timed differences in execution time with and without the "frerun-cse-after-loop -frerun-loop-opt" flags?
_________________
The greatest deeds are still undone, the greatest songs are still unsung...

Sivar · Apprentice Joined: 25 May 2002 Posts: 266 Location: USA

Dolio · l33t Joined: 17 Jun 2002 Posts: 650

Very sorry, but I edited out a couple of my USE="-..." variables and now it compiles fine. My face is officially red. Apologies.
_________________
They don't have a good bathroom to do coke in.

AutoBot · l33t Joined: 22 Apr 2002 Posts: 968 Location: Usually Out

I haven't had any problems with my flags, I don't have anything KDE related either

_________________
This message self destructed a long time ago.

Lovechild · Posted: Sat Jun 22, 2002 9:54 am Post subject:

GCC 3.1-r6
CFLAGS="-march=athlon-tbird -O3 -pipe -fomit-frame-pointer"
CXXFLAGS=$CFLAGS

CoronaLover · Posted: Sat Jun 22, 2002 11:10 am Post subject:

AutoBot · l33t Joined: 22 Apr 2002 Posts: 968 Location: Usually Out

Yeah I wouldn't get too (flag happy) with gcc3 yet, those look good to me also.
_________________
This message self destructed a long time ago.

Zu` · l33t Joined: 26 May 2002 Posts: 716 Location: BE

Zu` · l33t Joined: 26 May 2002 Posts: 716 Location: BE

AutoBot · l33t Joined: 22 Apr 2002 Posts: 968 Location: Usually Out

Malakin · Posted: Sun Jun 23, 2002 7:05 am Post subject:

kdemultimedia has crashing problems if you use "-fomit-frame-pointer", so you may want to emerge everything with it, then re-emerge that one ebuild without it.

DArtagnan · Posted: Sun Jun 23, 2002 7:08 am Post subject:

Did u tried it by your self and saw the error?
Or just heared about it?
_________________
All for one and one for All
--

MACPRO machine...

Lovechild · Posted: Sun Jun 23, 2002 7:31 am Post subject:

Lovechild · Posted: Sun Jun 23, 2002 8:15 am Post subject:

Following Sivar' advice I'm now trying to rebuild the entire thing with