Re: question on optimization

4 msg	problem building gcc-4.3 snapshot in home dir
5 msg	mutex failing (stupid C++!)
1 msg	Fwd: g++ source
3 msg	directory with object files
1 msg	Building gcc 4.2.1 on mac os x 10.4 with Xcode 2.2
2 msg	LD_LIBRARY_PATH
3 msg	Re: GNU GCC -m32 Problem?
7 msg	64 bit assignment on a 32 bit platform
1 msg	Re: Optimization of conditional access to globa...
1 msg	4.2.0: Bootstrap comparison failure!
2 msg	Re: Would you like to give me advice about how ...
1 msg	Wir haben etwas gegen Geruch!
1 msg	Would you like to help me?
1 msg	fail to building with -msoft-float

question on optimization
\ Jim Marshall (26 Oct 2007)
. \ Diego Novillo (26 Oct 2007)
. . \ Jim Marshall (27 Oct 2007)
. . . \ NightStrike (27 Oct 2007)

1 msg	Re: GCC4.1.1
1 msg	Creating attachment in bugzilla fails
1 msg	Problem to compile GCC-4.2.2 as a cross-compiler
3 msg	egcs help
4 msg	Gcc-2.95

Subject:	Re: question on optimization
Group:	Gcc-help
From:	NightStrike
Date:	27 Oct 2007

On 10/26/07, Jim Marshall <jim.marshall> wrote:
> Diego Novillo wrote:
> > On 10/26/07, Jim Marshall <jim.marshall> wrote:
> >
> >> gcc <options> module1.c module2.c ... moduleN.c?
> >
> > With the current version of GCC, there is no difference. However, if
> > you *do* compile all the .c files with a single command and use
> > -combine then the compiler will treat all the .c files as a single one
> > and the IPA optimizations will work on all the files at once. This
> > option is currently supported on C only.
> >
> > For future versions of GCC we are working on link-time optimization
> > features which will be more powerful than -combine (See
> > http://gcc.gnu.org/wiki/LinkTimeOptimization for some details). This
> > is not likely to be in GCC for at least a couple of releases, though.
> >
> Great, thanks. I never noticed the --combine option, thanks for the tip!

One of the best ways to test whether certain ways of compiling are
better than others, in general, is to test it. Maybe make up a
specific reproducible test case that you can use as a benchmark.