Am Samstag, 18. Januar 2014, 19:26:39 schrieb Andrew: > 18.01.2014 17:53, KP Kirchdörfer пишет: > > Am Samstag, 18. Januar 2014, 17:36:05 schrieb Andrew: > >> 18.01.2014 13:03, KP Kirchdörfer пишет: > >>> Am Donnerstag, 16. Januar 2014, 21:21:04 schrieb Andrew: > >>>> 16.01.2014 19:54, KP Kirchdörfer пишет: > >>>>> Am Donnerstag, 16. Januar 2014, 19:37:50 schrieb Andrew: > >>>>>> Strange, but for me there are some errors on libs linking (like > >>>>>> libnl3) > >>>>>> for arm, for x86 all is OK. Maybe distro-related trouble. Maybe not. > >>>>>> I > >>>>>> checked on gentoo x86_64. > >>>>> > >>>>> I'm building on kubuntu x86_64. > >>>>> > >>>>>> Maybe we should also upgrade gcc/binutils before alpha tagging? gcc > >>>>>> 4.8.2 seems to be stable, and fresh binutils are compiled w/o > >>>>>> troubles. > >>>>> > >>>>> If it solves the errrors for you, I'll test on my platform. IMHO you > >>>>> can > >>>>> reuse branch andrew-headers or whatever branch you'll add for remote > >>>>> access. > >>>>> > >>>>> But, still learning git, I'm not the one to give anyone/you any hints > >>>>> how > >>>>> to work with git :) > >>>>> > >>>>> kp > >>>> > >>>> It doesn't solve errors, but it should break nothing. Also it seems > >>>> like > >>>> gcc 4.8.2 will generate more optimized code. Also gcc-4.6 has MIPS > >>>> regressions which results in non-optimal code. > >>>> So I think that we should update them. > >>> > >>> I'm able to build all toolchains, gdb still fails, but any other issues > >>> are > >>> solved. A drawback is binary size - it's more often bigger than before. > >>> > >>> e.g. > >>> drobear old 87k, dropbear new 110k > >>> dnsmasq old 125k, dnsmasq new 143k > >>> > >>> Looks we need to tweak some more/new optimiziation settings. > >>> > >>> kp > >> > >> Maybe we should define 'resource-hungry' packages which should be > >> assembled with -O2/-O3, and generic packages (like bind, dhcpd, > >> iptables) which usually doesn't consumes much resources, that will be > >> assembled with -Os? > > > > "Resource-hungry" defined as CPU intensive? > > And -Os as default? > > Right. Also -O2 (or -O3) we may use for some libs (incl. system > libs)which seems to be used with that apps.
Ok, I've recompiled the toolchain and dropbear and dnsmasq (unfortunately I have not looked into more packages) and git mixed results: drobear gcc 4.6 -O2 87k dropbear 4.8.2 -O2 110k dropbear gcc 4.8.2 -Os 96k dnsmasq gcc 4.6 -O2 125k dnsmasq gcc 4.8.2 -O2 143k dnsmasq gcc 4.8.2 -Os 117k So mixed results; and probably gcc 4.6 with -Os is also smaller than the results above. > I think we should use something like this into platform include: > ifdef OPT_SPEED > CFLAGS=-O2 > else > CFLAGS=-Os > endif > CFLAGS+=..... Well I prefer to do compile specific packages with -O2 instead of complete toolchains. But then we do have time to decide after alpha-1. But the results above makes me questioning, if moving to gcc 4.8.2 is really a win and the way to go? kp ------------------------------------------------------------------------------ CenturyLink Cloud: The Leader in Enterprise Cloud Services. Learn Why More Businesses Are Choosing CenturyLink Cloud For Critical Workloads, Development Environments & Everything In Between. Get a Quote or Start a Free Trial Today. http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk _______________________________________________ leaf-devel mailing list leaf-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/leaf-devel
