GCC 4.4.0 testing

Message

Basso Buffo · Post by **Basso Buffo** » Wed Jan 28, 2009 5:00 pm

media-sound/wavpack fails on GCC 4.4 with a flood of messages similar to the following:

extra2.c: In function 'decorr_stereo_pass':
extra2.c:85: error: incompatible type for argument 2 of '__builtin_ia32_psrld'
extra2.c:85: note: expected 'int __vector__' but argument is of type 'int'

I've gotten to be pretty good at fixing the C++-missing-headers GCC 4.4 bugaboos, but this one has me stumped. It compiles cleanly with GCC 4.2.4.

Unfortunately, I can't seem to locate wavpack's bugzilla. What is it with open source projects that make it impossible to report bugs? I couldn't find dev-libs/fcgi's bugtracker either (I filed a Gentoo bug for fcgi anyway).

unK · Post by **unK** » Wed Jan 28, 2009 5:07 pm

media-gfx/xsane doesn't run if compiled with gcc-4.4, cpu usage goes to 100% and nothing happens. (infinite loop maybe? no idea, I didn't do further investigation). works ok with 4.3.2.

alex.blackbit · Post by **alex.blackbit** » Wed Feb 11, 2009 12:46 pm

i think there is a problem in the new graphite enabled ebuild with the ecj file.
of course this problem only applies when the gcj use flag is enabled.
it is in the distfiles directory, but is not copied for some reason.
anybody with the same situation?

Basso Buffo · Post by **Basso Buffo** » Thu Feb 19, 2009 10:03 pm

Just FYI -- while the new ebuild does enable graphite, the new graphite features are not workable. If you add the new graphite-enabling flags to your CFLAGS, gcc will crash and burn when trying to build even simple programs.

Code: Select all

CFLAGS="${CFLAGS} -floop-interchange -floop-strip-mine -floop-block"

rhill · Post by **rhill** » Thu Feb 19, 2009 11:14 pm

if you mail me a log i'll have a look.

Basso Buffo · Post by **Basso Buffo** » Fri Feb 27, 2009 4:37 am

-floop-interchange is tentatively working (it worked before as long as you specified -O0; crashed with -O1, -O2, -O3 or -Os) with the patch listed here:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39308
You'll have to edit the patch to prepend "gcc/" to the filenames. To my knowledge this patch hasn't made it into trunk, but the change was made in the 'graphite' branch with revision 144455.

Edit:
As of rev 144470, the patch has now been applied to trunk.

rhill · Post by **rhill** » Fri Mar 06, 2009 7:50 pm

i'm going to be doing some unrelated wxGTK work with my overlay soon and don't want to break stuff for you guys, so I'm moving the gcc ebuilds to the gcc-porting overlay, which, now that i think about it, makes a hell of a lot more sense. i'll put a message or something in the dirtyepic overlay gcc to tell people to switch.

pdw_hu · Post by **pdw_hu** » Wed Mar 11, 2009 8:41 am

dirtyepic wrote:i'm going to be doing some unrelated wxGTK work with my overlay soon and don't want to break stuff for you guys, so I'm moving the gcc ebuilds to the gcc-porting overlay, which, now that i think about it, makes a hell of a lot more sense. i'll put a message or something in the dirtyepic overlay gcc to tell people to switch.

Could you please copy dev-libs/cloog-ppl over to the new overlay also? It's needed for the graphite use-flag

rhill · Post by **rhill** » Wed Mar 11, 2009 10:21 pm

whoops. i'll do so now. thanks.

rufnut · Post by **rufnut** » Sat Mar 14, 2009 3:13 am

Basso Buffo wrote:Just FYI -- while the new ebuild does enable graphite, the new graphite features are not workable. If you add the new graphite-enabling flags to your CFLAGS, gcc will crash and burn when trying to build even simple programs.
Code: Select all
CFLAGS="${CFLAGS} -floop-interchange -floop-strip-mine -floop-block"

Just compiled the svn GCC 4.4.x and have built quite a few things with these flags.

Nothing has failed.

Yet...

dodo1122 · Post by **dodo1122** » Sun Mar 15, 2009 12:33 am

Stuff fails rarely, but often miscompiles -- example is firefox, which compiles just fine, but rendering is completely broken. Same goes for webkit. On the other hand, compiling paludis with gcc-4.4 yields visible speed improvements -- and none of the tests failed

rhill · Post by **rhill** » Sun Mar 15, 2009 4:53 am

a lot of those are caused by strict aliasing violations. -fno-strict-aliasing usually does the trick. with firefox it's actually a problem in nspr. you can find a patched version in the overlay.

neuron · Post by **neuron** » Thu Mar 19, 2009 12:56 pm

anyone crazy enough to run it outside chroot yet?

Was bored and googled a bit and found some performance tests here http://encode.ru/forum/showthread.php?t=278 which look very impressive.

unK · Post by **unK** » Thu Mar 19, 2009 2:36 pm

neuron wrote:anyone crazy enough to run it outside chroot yet?

I have world compiled with gcc-4.4.0_pre9999 for couple of months ;p Everything runs ok, except two packages (wavpack doesn't compile and if glibc is compiled with gcc-4.4, threads debugging in gdb doesn't work).

neuron · Post by **neuron** » Thu Mar 19, 2009 2:53 pm

nice, was expecting more things to break.

I'm not too worried about compile failure (which chroots test nicely), but compile ok and run unstable/crash is really annoying.

Basso Buffo · Post by **Basso Buffo** » Thu Mar 19, 2009 3:54 pm

I'm crazy enough to have bootstrapped with 4.4 (it took some doing), and I'm crazy enough to have the experimental graphite-related optimization flags in my CFLAGS. I have GCC 4.2.4 installed for the few times when 4.4 fails (such as wavpack and nspr up until this last month). I haven't had any major crashes, though I have had reason to acquaint myself with the gcc bugzilla.

neuron · Post by **neuron** » Fri Mar 20, 2009 11:01 am

Could you post your cflags and packages you compiled with 4.2 if you got a log of em?

Gonna clone my root and try to boot off 4.4 and see how it goes, I'll stick with not using graphite for the system though, I might toy with it on specific programs however

FallenWizard · Post by **FallenWizard** » Fri Mar 20, 2009 1:00 pm

I made a little patch to get OpenLDAP working with GCC 4.4:

http://rafb.net/p/cv1Bc347.html

I tested it with OpenLDAP 2.3.43 and it works fine.

neuron · Post by **neuron** » Fri Mar 20, 2009 1:04 pm

might wanna put that where people can see it, especially the openldap people and http://bugs.gentoo.org/show_bug.cgi?id=249226

rhill · Post by **rhill** » Fri Mar 20, 2009 10:44 pm

what is this chroot you speak of?

rufnut · Post by **rufnut** » Sat Mar 21, 2009 1:30 am

I think it stands for CHange ROOT.

You can always "man chroot"

I rolled back to 4.3.2 for a while as I had some errors on my machine.

I don't think it was 4.4.0's fault but it was best to be safe as I don't know a lot about GCC and programming.

Xake · Post by **Xake** » Sat Mar 21, 2009 8:30 am

rufnut wrote:I think it stands for CHange ROOT.

You can always "man chroot"

I am pretty sure dirty epic was ironic. After all he is involved somehow with the toolchain-maintainers...

rufnut · Post by **rufnut** » Sat Mar 21, 2009 9:31 am

Cool , and he still has time to play with a Phased Array like in his Picture.

kernelOfTruth · Post by **kernelOfTruth** » Sat Mar 21, 2009 4:00 pm

how about the greatest challenge:

kernel-compilation

was it successful for anyone ?

for me it produced an internal compiler error (2.6.29-rc8-zen1)

ComaWhite · Post by **ComaWhite** » Sat Mar 21, 2009 5:38 pm

yeah I was able to compile the gentoo-sources and zen-sources fine