Bugzilla Bug 147841

Hello, I upgraded xorg-server to the new release 1.1.1-r1 this morning, and I
noticed a clear performance drop with EXA enabled. This is visible not only
when I move transparent windows around but also when I scroll web pages in
firefox, documents in evince or text in gnome-terminal, when I switch virtual
desktop and when I resize windows etc. etc.

I tried to compile xorg without the 0* patches to find if they're the culprit,
but the build fails. 

My video card is a Radeon 7500 (R100), with the DRI driver. I also noticed that
Xorg.0.log included this new warnings:

(WW) AIGLX: 3D driver claims to not support visual 0x23
(WW) AIGLX: 3D driver claims to not support visual 0x24
(WW) AIGLX: 3D driver claims to not support visual 0x25
(WW) AIGLX: 3D driver claims to not support visual 0x26
(WW) AIGLX: 3D driver claims to not support visual 0x27
(WW) AIGLX: 3D driver claims to not support visual 0x28
(WW) AIGLX: 3D driver claims to not support visual 0x29
(WW) AIGLX: 3D driver claims to not support visual 0x2a
(WW) AIGLX: 3D driver claims to not support visual 0x2b
(WW) AIGLX: 3D driver claims to not support visual 0x2c
(WW) AIGLX: 3D driver claims to not support visual 0x2d
(WW) AIGLX: 3D driver claims to not support visual 0x2e
(WW) AIGLX: 3D driver claims to not support visual 0x2f
(WW) AIGLX: 3D driver claims to not support visual 0x30
(WW) AIGLX: 3D driver claims to not support visual 0x31
(WW) AIGLX: 3D driver claims to not support visual 0x32

emerge --info output:

Portage 2.1.1 (default-linux/x86/2006.1/desktop, gcc-3.4.6, glibc-2.3.6-r4,
2.6.17-ck1-r3 i686)
=================================================================
System uname: 2.6.17-ck1-r3 i686 AMD Athlon(tm) XP 1700+
Gentoo Base System version 1.12.5
Last Sync: Sat, 16 Sep 2006 15:30:06 +0000
app-admin/eselect-compiler: [Not Present]
dev-java/java-config: 1.3.3-r1, 2.0.28-r1
dev-lang/python:     2.4.3-r1
dev-python/pycrypto: 2.0.1-r5
dev-util/ccache:     [Not Present]
dev-util/confcache:  [Not Present]
sys-apps/sandbox:    1.2.17
sys-devel/autoconf:  2.13, 2.59-r7
sys-devel/automake:  1.4_p6, 1.5, 1.6.3, 1.7.9-r1, 1.8.5-r3, 1.9.6-r2
sys-devel/binutils:  2.16.1-r3
sys-devel/gcc-config: 1.3.13-r3
sys-devel/libtool:   1.5.22
virtual/os-headers:  2.6.17-r1
ACCEPT_KEYWORDS="x86"
AUTOCLEAN="yes"
CBUILD="i686-pc-linux-gnu"
CFLAGS="-march=athlon-xp -O2 -pipe -fomit-frame-pointer"
CHOST="i686-pc-linux-gnu"
CONFIG_PROTECT="/etc /usr/kde/3.5/env /usr/kde/3.5/share/config
/usr/kde/3.5/shutdown /usr/share/X11/xkb /usr/share/config"
CONFIG_PROTECT_MASK="/etc/env.d /etc/env.d/java/ /etc/gconf
/etc/java-config/vms/ /etc/revdep-rebuild /etc/terminfo /etc/texmf/web2c"
CXXFLAGS="-march=athlon-xp -O2 -pipe -fomit-frame-pointer"
DISTDIR="/usr/portage/distfiles"
EMERGE_DEFAULT_OPTS="--alphabetical"
FEATURES="autoconfig distlocks metadata-transfer sandbox sfperms strict"
GENTOO_MIRRORS="http://www.die.unipd.it/pub/Linux/distributions/gentoo-sources/
http://sunsite.cnlab-switch.ch/ftp/mirror/gentoo/
http://ftp.belnet.be/mirror/rsync.gentoo.org/gentoo/"
LANG="it_IT"
LDFLAGS="-Wl,-O1 -Wl,--sort-common -s"
LINGUAS="it en"
PKGDIR="/usr/portage/packages"
PORTAGE_RSYNC_OPTS="--recursive --links --safe-links --perms --times --compress
--force --whole-file --delete --delete-after --stats --timeout=180
--exclude='/distfiles' --exclude='/local' --exclude='/packages'"
PORTAGE_TMPDIR="/dati/wd31gb/portage-tmp/"
PORTDIR="/usr/portage"
PORTDIR_OVERLAY="/usr/local/portage"
SYNC="rsync://rsync.gentoo.org/gentoo-portage"
USE="x86 3dnow 3dnowext X a52 aac acpi alsa bash-completion berkdb bitmap-fonts
bzip2 cairo cdr cli crypt cups curl dbus dlloader dri dts dvd dvdr eds
elibc_glibc emboss encode exif fam fbcon ffmpeg firefox flac fortran gdbm gif
glitz gnome gpm gstreamer gtk hal imagemagick input_devices_keyboard
input_devices_mouse isdnlog java jce jpeg kdeenablefinal kernel_linux lcms
libg++ linguas_en linguas_it mad mmx mmxext mng mono mp3 mpeg musepack mysql
ncurses nls nntp nptl nptlonly nsplugin ogg opengl pam pcre pdf perl png ppds
pppd python quicktime readline real reflection samba sdl session sndfile spell
spl sse ssl startup-notification svg svga tcpd tetex theora threads tiff
truetype truetype-fonts type1-fonts udev unicode usb userland_GNU
video_cards_radeon vorbis win32codecs wmf x264 xcomposite xml xorg xv xvid
zlib"
Unset:  CTARGET, INSTALL_MASK, LC_ALL, MAKEOPTS, PORTAGE_RSYNC_EXTRA_OPTS

It could be the change to mesa 6.5.1. If you're comfortable editing the ebuild,
try removing xorg-x11-server-1.1.1-mesa-6.5.1.patch from PATCHES and changing
MESA_PV="6.5.1" to "6.5" instead.

(In reply to comment #1)
> It could be the change to mesa 6.5.1. If you're comfortable editing the ebuild,
> try removing xorg-x11-server-1.1.1-mesa-6.5.1.patch from PATCHES and changing
> MESA_PV="6.5.1" to "6.5" instead.
> 

I've just compiled xorg-server with those changes: it is still slow, so the
change to mesa 6.5.1 is not the cause of the problem. The AIGLX warnings
disappeared.

Now that you've made those changes, you could try removing all the 0[0-9]*
patches from PATCHES again..

(In reply to comment #3)
> Now that you've made those changes, you could try removing all the 0[0-9]*
> patches from PATCHES again..
> 

Oh, sorry, I didn't read the message. However, I think I have found the
problem: I recompiled xorg-server WITH mesa 6.5.1 and all the patches EXCEPT
02-dont-backfill-bg-none.patch and now everything is fast again. 

There is only a problem, when I set the transparency of a window with transset
CPU usage grows and its movements are jerky for a while (~1s). This happened
with 1.1.0 as well but not with 1.1.1-r1.

The aiglx patches (all the 0[0-9]* ones) come from Kristian H

The aiglx patches (all the 0[0-9]* ones) come from Kristian Høgsberg
<krh@bitplanet.net>. You'll have to contact him personally with issues. Please
keep us posted on it. Note that he's not a Gentoo user, so talking about
portage won't get you far. =)

I just received an answer from Kristian H

I just received an answer from Kristian Høgsberg. He says at the patches
assume XAA and under EXA there are at least "a couple of things" that don't
work and that everything is untested. He says also that a better compatibility
with EXA at the moment isn't a priority, and "for now it remains an XAA
specific hack."

I did some tests, and I can confirm that compiz (the portage version) crashes
the X server when I try to run it with EXA enabled and it works decently with
XAA and xorg-server without the 02 patch.

My suggestions are:

1) add an einfo to the ebuilds (xorg-server and compiz) warning the users that
they have to use XAA acceleration and disable EXA in xorg.conf
2) remove the 02 patch from the xorg-server ebuild, so that EXA/xcompmgr users
have a properly working server (I don't know what is the performance impact of
this change to compiz users, but since compiz is alpha software I think that we
should favour the stable EXA)

Any news on this? Now Gentoo supplies a broken xorg to its users: upstream
supports and advertises EXA as stable and working on the radeon cards, but a
Gentoo local patch, known faulty, breaks it.

Hey hanno, could you test without the 02 patch to xorg-server and make sure
things still work fine w aiglx+compiz, or CC someone else who uses it?

Donnie: Removing the 02-patch causes some screen flickering on gpu-intensive
effects with compiz.
Thus still everything works, performance lacks a bit more. I leave it up to you
if you remove it, but if you could wait 1-2 days, I'll try to get kristian on
irc and ask him if there's a better solution, maybe I'll hack something
together to conditionally enable the code section on exa enabled.

02-dont-backfill-bg-none.patch should not cause any slowdowns.  It was added to
prevent the X server from copying from framebuffer to host memory pixmaps when
exposing bg=none window.  Without this patch, exposure of bg=none windows will
cause an expensive readback from framebuffer.  Either way, the patch removes a
copy operation, so it should never cause a slowdown.

I didn't wrote before that I've been using xcompmgr in the last months, because
it is (was?) the only working compositing manager on my machine. 
I don't know if it's useful, but I tried to remove the 02 patch because I
noticed that when I switched virtual desktops or windows the newly created
windows became black for a moment before being actually painted, and the patch
had a name which seemed to be "correlated" to that behaviour.

I'm gonna have to hide this patch behind USE=aiglx before we go stable.

Reporter: Can you catch me on irc or jabber so we can try to get this thing
resolved? (jabber@hboeck.de or hanno on freenode)

Tomorrow I should be able to hang on IRC, I hope. By chance today I stumbled on
this thread on debian-x:
http://www.mail-archive.com/debian-x@lists.debian.org/msg51055.html. Apparently
the "02" patch doesn't make so much difference, and more importantly, someone
is working on making the patches EXA-friendly. You should contact them, I
think.

AIGLX patches are now only applied with USE="aiglx".  Thanks for the report.

Interestingly enough, I have the exact _opposite_ behaviour. Since -r1 hid the
patches behind "aiglx", my Composite (with EXA) is slow as molasses. I am on
the "radeon" driver with a X700 pro. I had to enable "aiglx" now to get
unsluggish behaviour. Maybe I am missing the big point?!

Perhaps try re-building your driver X and see if that fixes it.

Why? I added the patches (in like adding aiglx for the new ebuild) and it's
fine again ...

I mean that you should try rebuilding the driver without the patches and see if
you resolve your speed problem that way.  You shouldn't have to have the
patches for decent EXA support.

(In reply to comment #19)
> I mean that you should try rebuilding the driver without the patches and see if
> you resolve your speed problem that way.  You shouldn't have to have the
> patches for decent EXA support.
> 

Well, for me, it runs like _butt_ without the patches.  And with the patches,
it crashes (see bug #151666). 

(In reply to comment #19)
> I mean that you should try rebuilding the driver without the patches and see if
> you resolve your speed problem that way.  You shouldn't have to have the
> patches for decent EXA support.

I may seem a bit dumb here, but since the patches apply to xorg-server, not 
x11-drivers/xf86-video-ati, what exactly do you suggest? It's not really urgent
anymore since it works again with USE="aiglx", but interesting nevertheless.

(In reply to comment #21)
> I may seem a bit dumb here, but since the patches apply to xorg-server, not 
> x11-drivers/xf86-video-ati, what exactly do you suggest? It's not really urgent
> anymore since it works again with USE="aiglx", but interesting nevertheless.
> 

My thought was that the patches were changing ABI somehow, meaning that the
driver would have to be rebuilt against the server to fix the issue.  I agree
that it's not a huge deal since it's functioning now.

Reporter, can you have a look at the patch on bug #151666 together with all
aiglx-patches and see if that helps?
If exa runs fine with it then we could get rid of the aiglx-flag and apply the
aiglx-patches always.

(In reply to comment #23)
> Reporter, can you have a look at the patch on bug #151666 together with all
> aiglx-patches and see if that helps?
> If exa runs fine with it then we could get rid of the aiglx-flag and apply the
> aiglx-patches always.
> 

I'm sorry but I can't help anymore with this bug, I recently upgraded my
motherboard/cpu and I had to buy a new PCI-E videocard to replace my old AGP
card

There's not really much evidence for or against these patches, and I believe
they're all in the mainline now in some form.  With any luck, the aiglx USE
flag will be gone in X 7.2.