While in practice one might not see much improvement, if one would make even a couple of percent improvement on something like a context switch one could see significant systen performance improvement (read measurable as a large fraction of a percent)
secondly it is intersting just from the perspective of having a kernel with correct c code victor pelt Andrew Gaul <[EMAIL PROTECTED]> wrote: __________ >On Tue, Dec 02, 2003 at 07:10:55PM -0600, Big Mike Forsberg wrote: >> I have some free time and I'm looking into rebuilding my kernel with >> the non-comercial Intel optimized compiler. > >I recommend against doing this, if only for reliability concerns (no one >seriously tests Linux/icc and there is little incentive to do so while >icc remains proprietary). It is not clear what benefit you hope to >realize, but icc's speed advantage becomes less pronounced the further >one moves away from code that looks like SPECmark. Even if you doubled >kernel performance, the effect on overall system performance would be >small unless you were doing something unusual such as using a complex >iptables ruleset while saturating a 10 Gb Ethernet link (constantly >context switching and servicing IO). > >Playing compiler optimization games is not as straightforward as it >seems. A cautionary tale of optimization adventures was posted to >gentoo-user a month ago: > > http://www.debian.org/News/weekly/2003/44/ > >Unless you are willing to fiddle with many permutations of flags, you >are better off using the packager's recommendations with -march and >-mcpu set to your CPU type (and usually -fomit-frame-pointer on x86-32). > > >On Tue, Dec 02, 2003 at 07:38:21PM -0600, Anthony Liguori wrote: >> The Linux kernel only compiles with GCC. It makes use of GCC C >> extensions that are not supported by the Intel compiler (or any other >> compiler for that matter). > >My understanding is that icc has been able to compile Linux for some >time. Some GCC extensions were rolled into C99 and are thus standard C >(alas, not labels as values). icc supports these and others. > >-- >Andrew Gaul >http://gaul.org/ >_______________________________________________ >Siglinux mailing list >[EMAIL PROTECTED] >http://machito.utacm.org/mailman/listinfo/siglinux _______________________________________________ Siglinux mailing list [EMAIL PROTECTED] http://machito.utacm.org/mailman/listinfo/siglinux
