On Fri, Dec 23, 2011 at 08:04:32PM +0100, Dimitry Andric wrote: > On 2011-12-23 18:55, Kostik Belousov wrote: > >On Fri, Dec 23, 2011 at 06:03:42PM +0100, Dimitry Andric wrote: > ... > >>The only thing my patch makes sure of, is that amd64 does the same thing > >>as all other arches, e.g.: compile with a low optimization settings for > >>debug (-O, which is equivalent to -O1), compile with arch-specific high > >>optimization settings for release (-O2 plus whatever is required for the > >>arch, or lower if optimization breaks things). > > > >Release is built with -g for long time, this is where the symbol files > >in /boot/kernel comes from. > > Ah, that is done via 'makeoptions DEBUG=-g' in the kernel configuration > file, right? I didn't realize that was kept in for a release. But even > in that case, amd64 is somehow different from the other arches, which > all get compiled with -O instead. Yes. > > If people prefer that to stay as it is, I'll change the diff so only > -frename-registers gets removed when clang is used, as clang does not > support this flag. This question cannot be answered without measurement. I think that even the 'default' benchmark of buildworld over -O and -O2 kernels can be useful to continue the discussion.
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:22 UTC