Hash: SHA1

On 25.04.2014 20:31, Adrien Nader wrote:
> On Fri, Apr 25, 2014, JonY wrote:
>> On 4/25/2014 21:17, Dongsheng Song wrote:
>>> On Fri, Apr 25, 2014 at 6:50 PM, JonY <jo...@users.sourceforge.net> 
>>> wrote:
>>>> On 4/25/2014 17:47, Dongsheng Song wrote:
>>>>> I'm glad to upload my daily builds to mingw-w64, if someone 
>>>>> intersting and give me (dongsheng) permission.
>>>> What is your configuration like? And what exactly do you have in 
>>>> your download?
>>> For win64: ${GCC_SRC_ROOT}/configure \ --prefix=${SYS_ROOT} \ 
>>> --with-sysroot=${SYS_ROOT} \ --build=${BUILD_TRIPLET} 
>>> --host=${TARGET_TRIPLET} --target=${TARGET_TRIPLET} \ 
>>> --disable-multilib --disable-nls --disable-win32-registry \ 
>>> --enable-checking=release --enable-languages=c,c++,fortran \ 
>>> --with-arch=core2 --with-tune=generic
>>> For win32: ${GCC_SRC_ROOT}/configure \ --prefix=${SYS_ROOT} \ 
>>> --with-sysroot=${SYS_ROOT} \ --build=${BUILD_TRIPLET} 
>>> --host=${TARGET_TRIPLET} --target=${TARGET_TRIPLET} \ 
>>> --disable-multilib --disable-nls --disable-win32-registry \ 
>>> --enable-checking=release --enable-languages=c,c++,fortran \ 
>>> --with-arch=i686 --with-tune=generic
>> Add --enable-fully-dynamic-string and 
>> --enable-version-specific-runtime-libs and everything is good.
>> You may need to use winpthread if you want to use --enable-libgomp.
>> I have now added your login to allow uploads, please upload to somewhere
>> like "Toolchains targetting Win{32,64}/Personal Builds/dongsheng-daily".
>> probably a good idea to rotate the files weekly to prevent uploads from
>> getting too big (if there is such a thing as too big).
> I believe --with-arch=core2 is a big issue for generic toolchains. It
> will create troubles which will be very annoying to pinpoint.
> Any x86_64/amd64/EM64T (I love how Intel always makes up awful names) 
> already has SSE2 and there is little value in restricting this except in 
> very specific situation which are better dealt on a case-by-case basis.


desrt: https://bugzilla.gnome.org/show_bug.cgi?id=722604#c7
Services: Bug 722604: normal, Normal, ---, gtkdev, NEW, Various tests are
failing with 2.39.3
desrt: mclasen: the test is failing because of a gcc bug
desrt: but gcc is 'buggy by design' according to upstream
desrt: since following the spec would be expensive on machines that don't
have SSE instructions
desrt: (ie: chips made in the 1990s)
desrt: and since most distros have -march=i686 or worse, gcc has no choice
but to either (a) be slow or (b) violate spec
desrt: and they pick (b)
desrt: imho we should insist on a standards-complying C compiler -- and it's
quite possible to get gcc to do the right thing with a CFLAG or two...
desrt: upshot is that we will get distros providing some custom CFLAGS when
building glib -- but hopefully eventually evaluating their choice of default
CFLAGS and making changes there
desrt: either that or they just patch out the failing testcases...
mclasen: so, you're expecting distros to choose correct over fast, when
upstream gcc does ?
mclasen: does not, I should say
desrt: i'm expecting distros to reevaluate if they still want to support
chips from the 90s
desrt: i don't honestly think anyone would want to pick the "let's be slow"
desrt: but i also think people are generally unaware of the issue...
desrt: so mostly i aim to raise the visibility of the problem
desrt: also making people fully aware of the costs of supporting correct ISO
C on these old chips...
LRN: so, -match=i686 is expensive?
desrt: LRN: expensive or incorrect. take your pick.
desrt: LRN: gcc picks 'incorrect', for the record
LRN: i build glib with -march=i686 :(
desrt: LRN: ya... so do all the distros...
***desrt thinks that nobody really knows about this issue
***desrt recommends -march=pentium4 -mfpmath=sse
LRN: no, i mean i *specifically* build *glib* with -march=i686, because it
won't compile without that (i.e. with the default). At least it didn't back
when i made a first build
desrt: LRN: i think we depend on at least -march=i486 for atomic
compare-and-exchange instructions
LRN: yes, that's the reason.
daniels: desrt: ouch, if you have -march=pentium4, you probably want a -mtune
for a less-horrific microarch
desrt: daniels: ya.. -mtune=generic might make sense
desrt: in general, having access to sse/sse2 would be wonderful

- -- 
O< ascii ribbon - stop html email! - www.asciiribbon.org
Version: GnuPG v1.4.11 (MingW32)


Start Your Social Network Today - Download eXo Platform
Build your Enterprise Intranet with eXo Platform Software
Java Based Open Source Intranet - Social, Extensible, Cloud Ready
Get Started Now And Turn Your Intranet Into A Collaboration Platform
Mingw-w64-public mailing list

Reply via email to