Re: About -On optimization flag

[Ian Lance Taylor <iant at google dot com>]

Josà Luis GarcÃa Pallero <jgpallero@gmail.com> writes:

> Using the -On optimization flag, what is the greatest "n" value that
> can I use? In the gcc manual (4.5.1, 4.4.5, 3.4.6) the greatest level
> documented is -O3
> (http://gcc.gnu.org/onlinedocs/gcc-4.5.1/gcc/Optimize-Options.html#Optimize-Options),
> but if you tray for example with -O4, -O5, -O9, etc. no warning nor
> error is emitted. I've asket some people and all says that -O4 and -O5
> are valid and higher level optimizators than the documented -O3.
> What is the correct answer? Is -O3 the higher optimization level as is
> documented in the manual or -O4, -O5 performs additional
> optimizations?

In current gcc values greater than -O3 are equivalent to -O3.

> If -O3 is the higher why the use of -O4, -O5 not emits
> a warning or error?

If -O4 is ever defined, it will certainly include all the optimizations
at -O3, so there is no particular reason to warn about using -O4 today.

In practice I think -O3 is not all that useful.  The distinction between
-O1 and -O2 is reasonably clear: -O2 ignores compilation time and also
enables strict adherence to language standards where that helps
optimizations (-fstrict-aliasing, -fstrict-overflow).  Going to -O3
enables more speculative optimizations, so in some cases code compiled
with -O3 can be slower than code compiled with -O2.

Mainline gcc introduces a -Ofast option, which enables -O3 plus
disregarding language standards where that helps optimiations
(basically, it enables -ffast-math).

Ian