Date: Sat, 26 May 2007 16:23:53 +0200 From: Stefan Ehmann <shoesoft@gmx.net> To: Roman Divacky <rdivacky@freebsd.org> Cc: freebsd-current@freebsd.org, Ivan Voras <ivoras@fer.hr> Subject: Re: size of kernel after gcc4.2 upgrade Message-ID: <200705261623.54752.shoesoft@gmx.net> In-Reply-To: <20070526140421.GA97649@freebsd.org> References: <20070525095146.GA45288@freebsd.org> <200705261554.38444.shoesoft@gmx.net> <20070526140421.GA97649@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Saturday 26 May 2007 16:04:21 Roman Divacky wrote: > On Sat, May 26, 2007 at 03:54:37PM +0200, Stefan Ehmann wrote: > > On Saturday 26 May 2007 15:15:34 Ivan Voras wrote: > > > Roman Divacky wrote: > > > > well.. I dont think that 60% increase of size when you are optimizing > > > > for size is normal. I even think its a "bug" in that sense that > > > > something wrong is set somewhere which causes this. > > > > > > > > I certainly dont believe this is normal > > > > > > I'm using the default make flags. In my case, it's release, > > > non-debugging 6-RELEASE kernel of 26 MB vs debugging kernel of > > > 7-CURRENT of 106 MB :) > > > > I guess that's due to /boot/kernel/*.symbols which are not present in my > > 6.2 installation. gcc42 file size increase shouldn't be that much. > > thats the problem. with -Os and gcc42 the kernel increases A LOT (60%), > other optimization levels doesnt seem to be affected Yes, but Ivan said he used default flags. So I think in his case the symbol files are mainly responsible for the 106MB vs 26MB. There's definitely something fishy about the -Os build. As I wrote earlier today: It looks like -Os does excessive inlining. Which might possibly be caused by our gcc options or a gcc bug. Setting a very low -finline-limit for testing purposes gives reasonable filesizes. Any gcc experts around to comment on this issue? Stefan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200705261623.54752.shoesoft>