Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 11 Nov 2003 17:54:51 +0100
From:      Alexander Leidinger <Alexander@Leidinger.net>
To:        Guido van Rooij <guido@gvr.org>, Marius Strobl <marius@alchemy.franken.de>
Cc:        arch@freebsd.org
Subject:   Re: Patches to compile the kernel with Intel's C/C++ compiler
Message-ID:  <20031111175451.7b836abf.Alexander@Leidinger.net>
In-Reply-To: <20031111153142.GA42684@gvr.gvr.org>
References:  <20031110222221.36a11979.Alexander@Leidinger.net> <20031111153142.GA42684@gvr.gvr.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, 11 Nov 2003 16:31:42 +0100
Guido van Rooij <guido@gvr.org> wrote:

> What are the benefits of running an icc compiled kernel?

The same as every other program gets.

a) Icc knows better to schedule instructions for a P4 than gcc does.
b) If you have an application which stresses some critical path in the
kernel, icc may be able to optimize parts of it and allows you to get a
little bit more out of it.
c) Someone is working on getting the icc profiling information out of
the linux kernel back into icc to optimize the system based upon the
measured values (think about branch prediction and probabilities). When
this matures, it's very likely that we also get support for it.
d) If 2 compilers agree on the behavior of the same code, either both
compilers a buggy, or your code is fine (yes, an overly optimistic point
of view, but a pragmatic one). If one of 2 compilers disagrees,
something is wrong with one of the compilers, or with your code.

Or perhaps you're just religious and want to use as less (L)GPLed code
as possible. ;-)

Obviously not everyone needs to run an icc compiled kernel or userland,
but it doesn't hurts (modulo bugs in icc (as FreeBSD has a commercial
license of icc, we get commercial support, and Intel fixed already some
ICE issues for us) or in the code, but there could be/are bugs in gcc
too) to run or to be able to run one.

BTW.: even if Intel doesn't like it, but: icc produces the fastest 32bit
code for AMD CPUs too (in specific applications of course).

Bye,
Alexander.

-- 
                   It's not a bug, it's tradition!

http://www.Leidinger.net                       Alexander @ Leidinger.net
  GPG fingerprint = C518 BC70 E67F 143F BE91  3365 79E2 9C60 B006 3FE7



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20031111175451.7b836abf.Alexander>