pentium4 gcc options

Robert Connolly robert at linuxfromscratch.org
Wed Mar 2 09:43:02 PST 2005


On March 2, 2005 12:12 pm, Mike Hernandez wrote:
> > I'm also using -O3 and --enable-omitfp with good results btw.
> >
> > robert
>
> is --enable-omitfp the same as -fomit-frame-pointer ?

--enable-omitfp also uses -O99, and defines string_inlines. 
-fomit-frame-pointer can't be added to CFLAGS for glibc because a handfull of 
files can't use it, the configure option enables it where ever it can.

> And does it still hold true that -O2 combined with
> -fomit-frame-pointer produces faster executables than -O3? (as stated
> here:
> http://archives.linuxfromscratch.org/mail-archives/blfs-dev/2002-October/00
>1419.html)

Not in my experience. And it looks like the glibc folks think -O99 (basically 
-O3) works well with -fomit-frame-pointer.

-O2 vs -O3 totally depends on the cpu. On my duron -O3 is very bad because 
there's only 64kb of L2 cache. I still use --enable-omitfp but I hack 
Makeconfig and s/-O99/-O2/. I use -finline-functions because it still helps. 
I also covert packages to gzip on that system because bzip2 is heavy on the 
L2 cache. I expect Celeron's would have the same problem. On this pentium4 I 
have much more L2 cache, and -O3 is usable.

Optimizations also vary in benefits from one package to another. -O3 would 
work better on the GNU packages, and kernel, because they're coded for GCC. 
Software like ash, and procps, are not programed for GCC specifically, so -O3 
would make them slower because -finline-functions, etc, would not be taken 
advantage of. Others like mplayer use -O4 because they have done everything 
they can think of to take advantage of GCC.

robert



More information about the lfs-chat mailing list