This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: -funroll-loops experimental in gcc-3.3.2
- From: Kelley Cook <kcook34 at ford dot com>
- To: "Tal Agmon" <Tal dot Agmon at nsc dot com>
- Cc: GCC Mailing List <gcc at gcc dot gnu dot org>
- Date: Tue, 09 Dec 2003 15:48:21 -0500
- Subject: Re: -funroll-loops experimental in gcc-3.3.2
- Hop-count: 1
- References: <OF2F6FDD98.2631A388-ONC2256DF7.005F4BF8-C2256DF7.005FD834@nsc.com>
- Reply-to: Kelley Cook <kelleycook at wideopenwest dot com>
> I understood -funroll-loops is experimental in gcc-3.3.2.
>
> Can you give me details on the problems severity with this option?
> With my port I'm working on it passed many tests with this option.
> It will be very helpfull if you could tell me what is the exact
> meaning of "experimental"? (it has many bug reports, it functions well
> but hasn't been tested for a long time, etc.)
I think you misread the documentation because -funroll-loops is not
experimental and has been a part of GCC for a very long time.
As the documentaion states, it is not enabled with any of the -O[123]
levels since it usually will expand code size and, due to possibly
overrunning CPU cache it is not always a win in performance.
There are currently two bugs listed in the bugzilla against
-funroll-loops: http://tinyurl.com/ygjn