GCC 3.3

Message

floam · Post by **floam** » Sun May 04, 2003 1:28 am

who uses make xconfig anymore? gconfig baby! (gtk2 goodness)

floam · Post by **floam** » Sun May 04, 2003 3:13 am

I got epiphany working with gcc 3.3, two steps.

1)Edit your epiphany-cvs ebuild and add

Code: Select all

append-flags -fno-strict-aliasing

under the strip-flags line in the ebuild, if you want to be risky you can remove the strip-flags line altogether, but that could give you more problems.

2)Edit /usr/portage/distfiles/cvs-src/epiphany-cvs/epiphany/configure.in and remove the line

Code: Select all

GNOME_COMPILE_WARNING(error)

You only need to do this once, since its cvs itll allow you to have a file thats modified and it wont get overwritten, just parts of the file that are changed get patched.

cyfred · Post by **cyfred** » Sun May 04, 2003 4:29 am

OK here is a link to the patch and new glibc ebuild for glibc to compile with gcc 3.3.
Note: you must put this patch in $PORTAGE_OVERLAY/sys-libs/glibc/files/

http://www.volutin.net/~andrew/GENTOO/g ... -r1.ebuild
http://www.volutin.net/~andrew/GENTOO/glibc-gcc33.patch

OK some people have said that the ebuild works without the FAKE_ROOT used, this is not true as far as my testing has showed. I was working on a -r1 ebuild to fix this up but it did not work so im not going to release that ebuild. I will modify the original to use the cvs eclass and the detection of gcc 3.3 and release that ... next few hours probably.

Lovechild · Post by **Lovechild** » Sun May 04, 2003 7:01 am

arrgghh... it seems GCC might still have a few optimization bugs left - I recompiled GNOME 2.3.0 and it's showing the signs of unstability, acme is very crashy.

ah well...

floam · Post by **floam** » Sun May 04, 2003 7:12 am

Gnome 2.3? I didn't know there were ebuilds yet. Oh well, I'm still married to openbox for the time-being

Lovechild · Post by **Lovechild** » Sun May 04, 2003 7:22 am

floam wrote:Gnome 2.3? I didn't know there were ebuilds yet. Oh well, I'm still married to openbox for the time-being

Real men use experimental compilers on unoffical experimental packages... right?

some guy posted them in one of my threads - it's seems fairly stable, then again not many changes has been going on yet.

fca · Post by **fca** » Sun May 04, 2003 8:13 pm

Must be the Desktop you're using that's causing troubles. I'm using KDE from CVS with GCC 3.3 and rather aggressive compile options, and no problems here.

Just a joke, no wish to start a Desktop war here. To each his own, and I will just stick with KDE for now, as it's something I know intimately now.

GCC 3.3 is quite stable I must say.
XFree86, Mozilla 1.4 from CVS, KDE from CVS, glibc, all run quite well. The testing with Lame revealed some problems with the Lame ebuild, but overall it seems not much has changed with respect to speed. But maybe that's just an incentive to look at the new options GCC 3.3 gives us. Maybe the SSA options are finally stable now?
So much to test, so little time...

bssteph · Post by **bssteph** » Mon May 05, 2003 4:33 pm

floam wrote:yeah, your right, I turned strip-flags off and its better. I was using:
Code: Select all
CFLAGS="-march=athlon-xp -O3 -ffast-math -mfpmath=sse,387 -pipe -fomit-frame-pointer -fprefetch-loop-arrays -fforce-addr -fmerge-all-constants"
so im sure its possible i killed stuff

I've also broken the 3.3 ebuild. I compiled it once with strip-flags enabled, then I took another shot at it, compiling with very aggressive CFLAGS, and it died in the same location. Different reasons though.

Code: Select all

Bootstrap complete - make "quickstrap" to redo last build,
or "cleanstrap" to redo the bootstrap from scratch.
make[1]: Leaving directory `/var/tmp/portage/gcc-3.3/work/build/gcc'
Comparing stage2 and stage3 of the compiler
make[1]: Entering directory `/var/tmp/portage/gcc-3.3/work/build/gcc'
rm -f .bad_compare
case "compare-lean" in compare | compare-lean ) stage=2 ;; * ) stage=`echo compa                          

re-lean | sed -e 's,^compare\([0-9][0-9]*\).*,\1,'` ;; esac; \
for file in *.o; do \
  tail +16c ./$file > tmp-foo1; \
  tail +16c stage$stage/$file > tmp-foo2 \
    && (cmp tmp-foo1 tmp-foo2 > /dev/null 2>&1 || echo $file differs >> .bad_com                          
pare) || true; \
done
case "compare-lean" in compare | compare-lean ) stage=2 ;; * ) stage=`echo compa                          
re-lean | sed -e 's,^compare\([0-9][0-9]*\).*,\1,'` ;; esac; \
for dir in tmp-foo intl cp f java objc; do \
  if [ "`echo $dir/*.o`" != "$dir/*.o" ] ; then \
    for file in $dir/*.o; do \
      tail +16c ./$file > tmp-foo1; \
      tail +16c stage$stage/$file > tmp-foo2 \
        && (cmp tmp-foo1 tmp-foo2 > /dev/null 2>&1 || echo $file differs >> .bad                          
_compare) || true; \
    done; \
  else true; fi; \
done
rm -f tmp-foo*
case "compare-lean" in compare | compare-lean ) stage=2 ;; * ) stage=`echo compa                          
re-lean | sed -e 's,^compare\([0-9][0-9]*\).*,\1,'` ;; esac; \
if [ -f .bad_compare ]; then \
  echo "Bootstrap comparison failure!"; \
  cat .bad_compare; \
  exit 1; \
else \
  case "compare-lean" in \
    *-lean ) rm -rf stage$stage ;; \
    *) ;; \
  esac; true; \
fi
Bootstrap comparison failure!
reload.o differs
f/com.o differs
make[1]: *** [compare-lean] Error 1
make[1]: Leaving directory `/var/tmp/portage/gcc-3.3/work/build/gcc'
make: *** [bootstrap-lean] Error 2

!!! ERROR: sys-devel/gcc-3.3 failed.
!!! Function src_compile, Line 216, Exitcode 2
!!! (no error message)

And here are my CFLAGS

Code: Select all

CFLAGS="-march=athlon-tbird -mmmx -m3dnow -O3 -pipe -fomit-frame-pointer -fforce-addr -falign-functions=64 -falign-jumps=4 -falign-loops=4 -ffast-math -fprefetch-loop-arrays -maccumulate-outgoing-args -funroll-loops -finline-functions"

I kind of suspect -ffast-math as being the culprit, although I have nothing to really validate that, just intuition I guess. I'm trying the compile again w/out that flag.

Lovechild · Post by **Lovechild** » Mon May 05, 2003 4:44 pm

DO NOT compile you compiler with optimizations, if you break that, you are sure to break the rest of your system - strip-flags is there for a reason.

The allowed flags are somewhat safish, DO NOT, I repeat, DO NOT remove strip-flags from an ebuild and complain if it is breaks - it's your own fault.

bssteph · Post by **bssteph** » Mon May 05, 2003 5:07 pm

Oh, I know. :)
I'm just curious what's gumming it all up, and while I'm going nuts with an "unsafe" compiler, I might as well go all out and see how well I can break things.
I follow the rules... usually.

Besides, 3.2.2 is there in case I really hose 3.3.

[Edit: forgot to mention that so far, so good without -ffast-math, although I forget exactly where in the process my last attempt broke, so who knows if I'm in the clear or not. Anyway, I'll report when it's done. Heck, maybe we can chalk this up to -really- finding out what flags break the new compiler.]

fca · Post by **fca** » Mon May 05, 2003 6:07 pm

OK, gcc 3.3 won't compile Openoffice 1.0.3 and 1.1_beta getting the following error:

Code: Select all

conditn.c:292:97: pasting "to" and "." does not give a valid preprocessing token
conditn.c:292:97: pasting "to" and "." does not give a valid preprocessing token
conditn.c:292:97: pasting "to" and "." does not give a valid preprocessing token
conditn.c:292:97: pasting "to" and "." does not give a valid preprocessing token
conditn.c:292:97: pasting "to" and "." does not give a valid preprocessing token
dmake:  Error code 1, while making '../../unxlngi4.pro/obj/conditn.obj'
---* TG_SLO.MK *---

ERROR: Error 65280 occurred while making /var/tmp/portage/openoffice-1.1_beta/work/oo_1.1beta_src/sal/osl/unx

!!! ERROR: app-office/openoffice-1.1_beta failed.
!!! Function src_compile, Line 356, Exitcode 1
!!! Build failed!

I'll have to try with less aggressive optimizations I'm afraid.
You have to change the ebuilds to get it even start compiling though, since the ebuilds want gcc 3.2, and nothing else.

And the gcc 3.3 ebuild has been mentioned in the GWN..
though no link. Maybe they're afraid of how many people would like to try it, and break all kinds of things?

bssteph · Post by **bssteph** » Mon May 05, 2003 6:11 pm

Hells yes. gcc compiled fine with my above CFLAGS, sans -ffast-math which I removed. I'm compiling libogg, libvorbis, and xmms (all with -ffast-math) right now as a test. I would do more but I have to leave for classes in about 10 minutes.

--

Rarr. They're done already. Everything runs fine. I'm blasting my favorite j-pop right now and it's all peachy-keen. The libraries don't seem to strip any flags, xmms filters -fforce-addr. I wish I had some sort of benchmarks to give you (and the time to do it), but this insignificant little test will have to do for now.

Tonight I try something big.

bssteph · Post by **bssteph** » Mon May 05, 2003 6:14 pm

fca wrote: And the gcc 3.3 ebuild has been mentioned in the GWN..
though no link. Maybe they're afraid of how many people would like to try it, and break all kinds of things?

If you are "l33t h4x0r", you can see that they have the URL in the source code, just that it's not being interpreted as a link for browsers. I don't know if it's an error on their behalf or done purposefully, but the URL works fine. In fact that's how I got to this page.

floam · Post by **floam** » Tue May 06, 2003 2:53 am

couple more things that wont compile

iptables and opensp both fail.

Rhino · Post by **Rhino** » Tue May 06, 2003 5:49 am

iptables also fail with gcc-3.2.3-r1.

Chaos · Post by **Chaos** » Tue May 06, 2003 4:45 pm

so far I've had debianutils, X-4.3-r2 and openjade fail. Stil rebuilding world...

Rhino · Post by **Rhino** » Tue May 06, 2003 4:50 pm

bootstraped in one machine here just fine, with glibc-2.3.2-r1 with the patch applied.
cflags :

Code: Select all

CFLAGS="-march=pentium4 -O2 -pipe -mmmx -msse -msse2 -mfpmath=sse -fomit-frame-pointer -frename-registers 
-ffast-math -fforce-addr -frerun-loop-opt -frerun-cse-after-loop 
-fprefetch-loop-arrays -falign-functions=4 -falign-jumps=4 -falign-loops=4"

i'm not optimizing the gcc itself, but optimizing glibc filtering some of those flags.

edited
in fact i'm optimizing the gcc-3.3 with -march=pentium4 -O2 -pipe.

bssteph · Post by **bssteph** » Tue May 06, 2003 5:38 pm

Reporting in, xfree-4.3.0-r2 compiled fine. I didn't edit the ebuild, so it stripped my crazy flags, but it included -march=athlon-tbird, which implies -mmmx and -m3dnow, so what I really wanted was included.

According to a friend who recently rebuilt world, only the following three packages broke: debianutils, koffice, and openjade.
I think he patched glibc and compiled that w/out problems.

Next for me is a kernel compile, although I don't think I'll be able to boot it until tonight... another day of labs and busyness at school.

Post by **dioxmat** » Tue May 06, 2003 6:45 pm

bssteph wrote: but it included -march=athlon-tbird, which implies -mmmx and -m3dnow, so what I really wanted was included.

gcc 3.2.? had to have a bug that made him always disable mmx and 3dnow, even when using -march=idontrememberwhat, even if you specified it on the command line, so I suggest you check if it really added the flags :)

bssteph · Post by **bssteph** » Tue May 06, 2003 11:10 pm

Oh, I did, I did. I was very pleased. :) I found out that athlon-tbird also implies -maccumulate-outgoing-args.

[Edit: Turns out -maccumulate-outgoing-args is used all the time by gcc 3.3.]

Bad news though. I can't compile gaming-kernel-2.4.20-r3 with gcc 3.3. It craps out not too far in with the following error:

Code: Select all

In file included from ide-cd.c:320:
ide-cd.h:440: error: long, short, signed or unsigned used invalidly for `slot_tablelen'
make[3]: *** [ide-cd.o] Error 1
make[3]: Leaving directory `/usr/src/linux-2.4.20-gaming-r3/drivers/ide'
make[2]: *** [first_rule] Error 2
make[2]: Leaving directory `/usr/src/linux-2.4.20-gaming-r3/drivers/ide'
make[1]: *** [_subdir_ide] Error 2
make[1]: Leaving directory `/usr/src/linux-2.4.20-gaming-r3/drivers'
make: *** [_dir_drivers] Error 2

The nature of the error (and that it compiled before, of course) leads me to believe that GCC is being a bit more strict with its variables. I'm no kernel hacker but I'm going to take a look at it when I get home.

[Edit: I just thought that since I'm going to be on campus for a while, I might as well try -something-... so back to a gcc compile with strip-flags enabled. Maybe that will do the trick.]

floam · Post by **floam** » Tue May 06, 2003 11:14 pm

2.5 kernels compile nicely on gcc 3.3

bssteph · Post by **bssteph** » Tue May 06, 2003 11:23 pm

Yeah, I saw that Lovechild got 2.5 to compile, and the above-mentioned friend got 2.5.69-mm1 to go through, so maybe it's an issue with 2.4 kernels in general, which is a bit of a drag. I want to use gcc 3.3 for everything, but I really like the gaming-sources (and not having to worry about a 2.5 kernel)... decisions, decisions.

floam · Post by **floam** » Tue May 06, 2003 11:25 pm

give mm-sources a try, it blows ck-sources/gaming-sources/all other old kernels out of the water

bssteph · Post by **bssteph** » Tue May 06, 2003 11:33 pm

I'm seriously thinking of it.. but it will have to wait until I get home, I don't want to monkey with -everything- through ssh.

OT, but how are the emu10k1 ALSA drivers in the kernel?

floam · Post by **floam** » Tue May 06, 2003 11:49 pm

their great, to get the kernel alsa working unmerge inject the latest alsa-driver, remove alsasound from runlevel boot, and compile alsa and oss emulation into kernel. be sure to emerge module-init-tools before trying 2.5 kernels