GCC 4.4.0 testing

Letharion · Veteran Joined: 13 Jun 2005 Posts: 1344 Location: Sweden

My amarok 2 was crashing before too, so I have no reason to blame that on graphite.
I've had some crashes from vlc. I can't recall if I've had them before or not, could possibly be an issue with 4.4/graphite/2.10. If it happens frequently enough I'll investige, but currently it doesn't.
Haven't noticed anything else.

hirakendu · Guru Joined: 24 Jan 2007 Posts: 386 Location: san diego

Letharion · Veteran Joined: 13 Jun 2005 Posts: 1344 Location: Sweden

robnotts · Posted: Mon Jun 01, 2009 3:43 pm Post subject:

Right, after the success on the laptop... i moved my main machine over to GCC4.4.0, but with very sane CFLAGS!

If you are using MYTHTV with MYSQL, I wouldn't recommend compiling MYSQL with GCC4.4.0, as it corrupted the database. Luckily I had a backup. Recompiling MYSQL with GCC4.3.3-r2, all back up and running.

Was driving me mad for a few hours!
_________________
---

Gentoo Phenom][ X4 955 on AMD790 + Geforce 220GT 8GB/1.75TB (Desktop)
+ MythTV (3xFreeview,1xFreesat HD) on 1080p
Gentoo Turion64 X2 Geforce 6150 2GB/120GB (Laptop)

hirakendu · Guru Joined: 24 Jan 2007 Posts: 386 Location: san diego

Alright, after all the emerge -e world, I actually started using stuff. Things seem to be going nice but for few things. Firefox is extremely crashy - especially while scrolling (actually it hangs). It could be due to nvidia-driver too, but I doubt that is the case (I remember 185.18.14 was working fine the other day). Did anyone else experience it too. Might have something to do with xulrunner. Happens with both 3.0.10 from portage and the binary (x86_64) minefield (3.6a1pre). Also got a kernel oops on my notebook, possibly due to the broadcom-sta driver (wl module). Anyway, the frequent firefox hangs are annoying and unacceptable and make it almost unusable.
_________________
Helium Sources || Gentoo Minimal Livecd

neuron · Advocate Joined: 28 May 2002 Posts: 2371

hm, I've actually built xulrunner and firefox with custom-optimizations and graphite + tree vector + fast-math and it seems fine

engineermdr · Posted: Tue Jun 02, 2009 4:44 pm Post subject:

I think the firefox crashes might be due to flash: Flash plugin crashes Firefox. I have uninstalled adobe-flash and no more crashes for me.

unK · l33t Joined: 06 Feb 2007 Posts: 769

robnotts · Posted: Tue Jun 02, 2009 11:05 pm Post subject:

Apart from the above mentioned problem with mythtv/mysql, I now have 2 machines compiled with gcc-4.4.0, and can report that all is stable. The laptop has the mad cflags, the desktop the nice stable cflags... and both very usable!
_________________
---

Gentoo Phenom][ X4 955 on AMD790 + Geforce 220GT 8GB/1.75TB (Desktop)
+ MythTV (3xFreeview,1xFreesat HD) on 1080p
Gentoo Turion64 X2 Geforce 6150 2GB/120GB (Laptop)

Letharion · Veteran Joined: 13 Jun 2005 Posts: 1344 Location: Sweden

tranquilcool · Veteran Joined: 25 Mar 2005 Posts: 1245

hirakendu · Guru Joined: 24 Jan 2007 Posts: 386 Location: san diego

ppurka · Advocate Joined: 26 Dec 2004 Posts: 3256

hirakendu · Guru Joined: 24 Jan 2007 Posts: 386 Location: san diego

I think I have for now narrowed down my problem of browser hangs to adobe-flash - perhaps not playing well with glibc-2.10. But for me, no flash is pretty much a stripped down browser

. For now I'll stick to my older backup image with gcc-4.3 and glibc-2.9, and will later use the new image with gcc-4.4 and glibc-2.10. If I find time, I'll try to confirm that indeed it is adobe-flash by trying ubuntu karmic alpha's. Perhaps that only reflects the importance of open source and freedom - after fixing trivial breakings of most packages with gcc-4.4/glibc-2.10, here I am, almost helplessly stuck at a 1mb binary library file.

Aside, I have four closed source programs (that I can think of right now) on my machine - nvidia-drivers, adobe-reader, adobe-flash and skype - all very important. And two of them are 32-bit

. (That is discounting sun-java, in the hope of using iced-tea some day.)
_________________
Helium Sources || Gentoo Minimal Livecd

Gregoire · Apprentice Joined: 15 Apr 2006 Posts: 292

dvgrab-3.4 fails with (didn't find a bug on bgo) :

hirakendu · Guru Joined: 24 Jan 2007 Posts: 386 Location: san diego

Ah, hell freezes over

. All that firefox crashing due to adobe-flash was due to latest update to gnutls-2.8.0. Downgrading to gnutls-2.6.6 fixes it all. (It is a little disconcerting that almost every major upgrade of a system package causes one breakage or the other

.) It turns out that crashing problem happened also with gcc-4.3 and glibc-2.9 (as I figured out in last few days using the backup image and then scratching my head). Still, Opera happens to handle it almost flawlessly.

Very special thanks to the bug 272388 to figuring it out. Looks like I wasn't looking at the gdb output carefully when I was doing it

.

So with this, I can declare gcc-4.4 and glibc-2.10 to be completely rock solid for me.

** goes back to 'emerge -e world' with graphite cflags of gcc-4.4

**

Meanwhile,

Gregoire · Apprentice Joined: 15 Apr 2006 Posts: 292

hirakendu you are my hero !!!

I didn't look forward in compiling dvgrab (partly because my old DV camera isn't here) because I tried to understood why my firefox crashed again and again, even when returning to gcc-4.4 without graphite, then to gcc-4.3.3...

So thank you very muchh !!!

hirakendu · Guru Joined: 24 Jan 2007 Posts: 386 Location: san diego

Yeah, thanks to bgo and got lucky with google (yet again)

.
_________________
Helium Sources || Gentoo Minimal Livecd

MageSlayer · Posted: Mon Jun 29, 2009 2:34 pm Post subject:

For those who experience strange gdb behavior.
It maybe related to https://bugs.gentoo.org/show_bug.cgi?id=269307
It's basically about gcc4.4 optimize out several glibc functions. I attached ebuild+patch for temporary workaround.

ComaWhite · Tux's lil' helper Joined: 07 Oct 2008 Posts: 125

yeah I get a problem with gdb not giving me no backtrace anymore with GCC-4.4.1

devsk · Posted: Thu Jul 02, 2009 11:11 pm Post subject:

Are there packages still broken with -ftracer? My benchmarks indicate that gcc-4.4 has much better -O3 performance (runtime performance of the app being compiled) compared to gcc-4.3.3. Also, in gcc 4.4, -O2 plus -ftracer is almost same as -O3 in performance (tested python math, x264). The compile time is much better with -O2 -ftracer than with -O3. Package size is also better with -O2 -ftracer. Just plain -O2 is about 9% slower in python benchmarks compared to -O2 -ftracer.

So, net result is that I think for 4.4, I plan to use -O2 -ftracer. But I wanted to find out from people if they know of a breakage with -ftracer.

mv · Watchman Joined: 20 Apr 2005 Posts: 6780

randomlychosen · n00b Joined: 07 Jul 2009 Posts: 7

I'm not sure why is everyone so excited about both GCC 4.3.x and GCC 4.4.x.

First of all, GCC 4.4.0 is known to produce mplayer binaries which segfault playing some video files (that doesn't happen if you use GCC 4.2.x or GCC 4.3.x).

Secondly, according to my own measurements amongst GCC 4.2.4/4.3.3/4.4.0, GCC 4.2.4 is a definite winner.

I have tested some applications and the most useful of them is unrar decompressor. In all my tests GCC 4.2.x produces the least amount of code (read smallest binaries) which works faster than code produced with the later GCC versions. I have not tested graphite flags introduced in GCC 4.4.0 because they are known to cause ICE's and these flags are still considered highly experimental (as they are not enabled at any -O level).

I have Athlon 64 x2 K8 CPU (AMD Athlon(tm) 64 X2 Dual Core Processor 5200+ with 1MB L2 cache specifically) and I'm running a 32bit Linux distro.

At third, some tests on the net indicate that GCC 4.4.0 is still lagging behind older GCC versions regardless new major release and number.

At fourth, as I've mentioned earlier, GCC 4.4.0 produces binaries which are usually larger (2-5%) than the ones which are produced by GCC 4.2.x without significant performance boost.

I would be glad if someone posted his/her verifiable results (with compilation flags and PC details) of GCC 4.4.0 winning over GCC 4.2.4 considerably.

alex.blackbit · Advocate Joined: 26 Jul 2005 Posts: 2397

randomlychosen,

welcome to the gentoo forums.
it is quite unusual that new users join a forum and do not introduce themselves, nor ask a question, because they need help with something.
IMHO it is not productive to just question the work of other people.

since 4.4.0 is the current version of gcc, it is onlyl reasonable to try to get it in a stable state on a linux distribution, in our case gentoo.
ignoring it would lead to beeing out of date.
if you want to compare the performance of gcc versions, compare 4.2.0, 4.3.0 and 4.4.0, not higher point releases with a point zero release.

and, if you do not like 4.4.0, why not just stick with an older version? nobody forces you to upgrade.

randomlychosen · n00b Joined: 07 Jul 2009 Posts: 7

alex.blackbit

I have many years of Linux experience and I haven't come here to ask questions or introduce myself because it would only be detrimental to this technical discussion.

My point of posting here is that gentoo users are the most tech-savvy users in a Linux user base so I wanted to share my thoughts with you, hoping that someone could dispel my doubts. I see no point of comparing GCC 4.4.0 with older .0 releases since newer subreleases in absolutely most cases only resolve bugs and introduce fixes.

> you did not post anything reproducible yourself, are you aware of that?

Yes, I'm aware of that, and I'll post my results later. As some people in this thread have indicated a major performance boost gained from using GCC 4.4.0 I decided to hear them first.

Added later:

Here comes the first whopping evidence:

Qt 4.5.2 /lib directory (without *.debug files) occupies