1.4_beta guinea pig results...

phong · Posted: Wed Aug 07, 2002 2:08 pm Post subject: 1.4_beta guinea pig results...

I decided to be a guinea pig last night and give 1.4_beta a spin from stage 1... Got through the bootstrap, emerge system, kernel compilation just fine, got everything up and running. I don't know yet about the speed of the compiled code, but it seems like everything is compiling noticably faster (in hindsight I wish I had timed things to get some hard numbers).

But then I tried to emerge KDE and it died compiling xfree - here's the error:

dioxmat · Posted: Wed Aug 07, 2002 2:34 pm Post subject:

known gcc3 bug.
I saw it on the gcc mailing lists, dont remember where exactly :(
_________________
mat

paulisdead · Guru Joined: 10 Apr 2002 Posts: 510 Location: Seattle, WA

I had the same error when building X with GCC3.2. I got it to build by adding -mno-mmx -mno-3dnow, and also removing 3dnow from my use variables. Try just removing it from the use variables, as I'm not sure it needs -mno-mmx -mno-3dnow, since when I took 3dnow out of my use variables it compiled and installed fine. I'm going to try rebuilding X without -mno-mmx -mno-3dnow, and just removing it from my use variables and see how that works.

jimmybfan · Posted: Wed Aug 07, 2002 4:32 pm Post subject:

everything compiled fine for me, no errors using

CHOST="i686-pc-linux-gnu"
CFLAGS="-march=athlon-xp -m3dnow -msse -mfpmath=sse -mmmx -O3 -pipe -fomit-frame-pointer"
CXXFLAGS="-march=athlon-xp -m3dnow -msse -mfpmath=sse -mmmx -O3 -pipe -fomit-frame-pointer"

phong · Posted: Wed Aug 07, 2002 4:57 pm Post subject:

Heh, I actually searched the gcc mailing list archive, but somehow failed to find this problem report on the first try. Not to mention this one or this one. I'm going to try just "-mno-mmx" since the bug report seems to link it to mmx... I'll let everybody know how it goes.

Russki · Posted: Wed Aug 07, 2002 5:06 pm Post subject:

Hi, I've compiled my system from stage1 without any errors at all!
I/ve used thise flags:
"-march=athlon-xp -m3dnow -msse -mfpmath=sse -mmmx -O3 -pipe -fforce-addr -fomit-frame-pointer -funroll-loops -frerun-cse-after-loop -frerun-loop-opt -falign-functions=4 -maccumulate-outgoing-args"
The only error I had was mozilla 1.1b, but ithat's something else.
_________________
Facts are stupid things.
-Ronald Reagan
[img]http://www.danasoft.com/vipersig.jpg[/img]

paulisdead · Guru Joined: 10 Apr 2002 Posts: 510 Location: Seattle, WA

Call me crazy, but I'm seeing a bit of a pattern. Seems like it's not working with tbirds, but is with athlon XPs. I know Neeyo's got an XP and it built fine for him as well. Is anyone else with a TBird seeing the same thing?

*edit, BTW I have a tbird too

phong · Posted: Wed Aug 07, 2002 5:30 pm Post subject:

You're probably crazy, but so am I and we see the same pattern which makes it OK. Also, i've noticed a few posts from athlon-xp people who've sucessfully compiled their whole system with gcc 3.1.1 or 3.2_pre with -funroll-loops and/or -maccumulate-outgoing-args at least one of which was a show-stopper for me (with 3.1.1 at least). Everything would be fine until it tried to compile kde at which point gcc would seg fault, but recompiling gcc without those two options fixed the problem. I didn't determine which of the two was at fault, but since there may be an athlon-xp pattern with that as well as this bug, methinks something in the athlon-xp specific scheduler/optimizations makes it immune to these bugs and the rest of us saps have to wait until they get fixed.

That's what I get for not updating to the latest and greatest!

Ard Righ · Posted: Wed Aug 07, 2002 7:12 pm Post subject:

I was going through installing 1.4 from stage1-x86 and it got through ok up until the point where it was doing raidtools, and came up with an error on parser.o being truncated or something stupid.

Now, I am not sure why it wants to install raidtools, I don't have/use raid. So I was pissed because now I have to start again from scratch ;p

I am gonna try using 3.2 the safe way... compile system with 3.1.1 then merge 3.2_pre and see how that goes

phong · Posted: Wed Aug 07, 2002 8:23 pm Post subject:

After trying all the combinations, I got it to compile! You need to specify BOTH -mno-mmx -mno-3dnow when compiling xfree. The USE variable for 3dnow has no effect (which makes sense, because it's an internal compiler error and gcc doesn't care about USE variables). I tried with just one or the other, turning off both was the only way to get it to work. I'm going to take those back out and see if I can get emerge KDE to complete.

For reference, the exact CFLAGS for getting it to work were:
CFLAGS="-march=athlon-tbird -O3 -pipe -mno-mmx -mno-3dnow -fomit-frame-pointer"

This seems like a pretty serious bug, so I'm keeping my fingers crossed that it will be fixed in 3.2 final?!

therobot · Apprentice Joined: 07 Jun 2002 Posts: 256 Location: Canada

I installed X no problem with athlon-xp -pipe -03 -fomit-frame-pointer

silvercat · n00b Joined: 07 Aug 2002 Posts: 54 Location: Canada

hi...

I was trying to emerge kde on an Athlon 1700+ system.. what happened was... when compiling kdelibs... it seemed as though the computer froze... i could ping the machine... but i couldn't ssh to it remotely... and the screen went black... and i couldn't get it to display ne thing... so when i typed "emerge kde" again... after restarting the system... it compiled kdelibs... i had the same problem with kdebase... the compile failed the first time... but got past that point the second time..

I'm just wondering if this non-repeatable compile error could be the result of a bug in gcc 3.2 pre ?... or i'm wondering if it could be an unstable motherboard/system... it's not being overclocked... and i think it's being cooled properly... although i could take the case cover off...

My compiler flags are..
CHOST="i686-pc-linux-gnu"
CFLAGS="-march=athlon-xp -O3 -pipe -fomit-frame-pointer"
CXXFLAGS="-march=athlon-xp -O3 -pipe -fomit-frame-pointer"

O and by the way... this is on a 1.4 beta gentoo system... the bootstrap and emerge system were successfull... and there were no problems emerging XFree..

Dolio · l33t Joined: 17 Jun 2002 Posts: 650

Looks like I'm finally emerging X with the right set of CFLAGS now, then. What a pain (X takes like 3 times as long to compile w/ gcc3, it seems, or at least it did at 6 this morning...).

My question is: is there a way to modify the ebuild so that just that directory gets compiled with -mno-3dnow -mno-mmx? I recall there being a problem with 3dnow in another directory (or maybe it was this same one) and the ebuild maker modified it so that that directory was fixed without having to specifically change your cflags for the whole compile. I don't know what, if any, benefit there is to using mmx and 3dnow instructions in X in general, but it seems a shame to kill those benefits for the whole program when it's just one directory that has a problem.

Anyone know the answer (I looked at the ebuild briefly, but nothing jumped out at me)? It'd be much appreciated.
_________________
They don't have a good bathroom to do coke in.

phong · Posted: Thu Aug 08, 2002 2:01 pm Post subject:

BAH, now it's seg faulting a couple hours into building KDE. I hope 3.2 final is a bit more stable.

Methinks I'm going to try to build a nice stable system and do "experiments" on a spare partition.

paulisdead · Guru Joined: 10 Apr 2002 Posts: 510 Location: Seattle, WA

IWBCMAN · Guru Joined: 25 Jun 2002 Posts: 474

paulisdead · Guru Joined: 10 Apr 2002 Posts: 510 Location: Seattle, WA

That's a different error your getting from the 2 of us with tbirds. With us it's failing in the PEX5 directory.

phong · Posted: Thu Aug 08, 2002 11:39 pm Post subject:

Vanquirius · Posted: Fri Aug 09, 2002 2:36 am Post subject:

I'm having problems emerging X as well.
I have a pentium3 700mhz, and compile failed with flags -O3 -pipe -fomit-frame-pointer funroll-loops -mmmx -m3dnow. I'm giving it a second shot now without funroll-loops, mmmx and m3now.
The system built fine with those flags with GCC3.1.
Anybody has any suggestions that could save me some time?

Details:
p3 700mhz, 128ram+1Gb swap, using ext3fs, built from 1.4beta stage1 tarball.
_________________
Hello.

Vanquirius · Posted: Fri Aug 09, 2002 3:53 am Post subject:

In case you're wondering, my second attempt to emerge xfree worked...
_________________
Hello.

mksoft · l33t Joined: 28 May 2002 Posts: 844

From the ChangeLog for xfree looks like a patch was added fixing this yesterday:

CoronaLover · Posted: Fri Aug 09, 2002 9:11 am Post subject:

paulisdead · Guru Joined: 10 Apr 2002 Posts: 510 Location: Seattle, WA

Hey phong, could you try re-building Xfree with -mno-mmx -m3dnow. I was playing around with some different Cflags and finally got it to build with 3dnow. Just curious if this'll work for you, since your the only other person I know of running gentoo with GCC 3.2 on an old school T-Bird.

these are my normal cflags, I just replaced -mmmx with -mno-mmx and it actually built.

-march=athlon-tbird -O3 -pipe -fomit-frame-pointer -ffast-math -frerun-cse-after-loop -frerun-loop-opt -funroll-all-loops -falign-functions=4 -fno-exceptions -fexpensive-optimizations -mmmx -m3dnow

phong · Posted: Fri Aug 09, 2002 5:03 pm Post subject:

paulisdead · Guru Joined: 10 Apr 2002 Posts: 510 Location: Seattle, WA

This was my first experiment with fast math, and it's actually working pretty well. Only thing it broke was the gimp, text was all messed up and, and erasing smudged instead of erasing. I'll have to look into if all the other flags are in -O3. I did all this on my second hard drive, so it wouldn't have been too big of a deal if it didn't work, but everything seems to be pretty solid. I might try to compile KDE, just to see if I get a segfault, but Gnome's working flawlessly.