Difference between revisions of "AMD Catalyst"

From ArchWiki
Jump to: navigation, search
m (Installing from the AUR)
(Checking operation)
Line 227: Line 227:
 
{{Note|You can also use:
 
{{Note|You can also use:
 
  $ fgl_glxgears
 
  $ fgl_glxgears
as the fglrx alternative test to glxgears.
+
as the fglrx alternative test to {{ic|glxgears}}.
 
}}
 
}}
  

Revision as of 08:34, 20 October 2013

Template:Article summary start Template:Article summary text Template:Article summary heading Template:Article summary wiki Template:Article summary wiki Template:Article summary heading Template:Article summary link Template:Article summary link Template:Article summary end

Owners of ATI/AMD video cards have a choice between AMD's proprietary driver (catalystAUR) and the open source driver (xf86-video-ati). This article covers the proprietary driver.

AMD's Linux driver package catalyst was previously named fglrx (FireGL and Radeon X). Only the package name has changed, while the kernel module retains its original fglrx.ko filename. Therefore, any mention of fglrx below is specifically in reference to the kernel module, not the package.

As of April 26, 2013, Catalyst packages are no longer offered in the official repositories. In the past, Catalyst has been dropped from official Arch support because of dissatisfaction with the quality and speed of development. This time, it's the incompatibility with Xorg 1.14.

Compared to the open source driver, Catalyst performs worse in 2D graphics, but has a better support for 3D rendering and power management. Supported devices are ATI/AMD Radeon video cards with chipset R600 and newer (Radeon HD 2xxx and newer). See the Xorg decoder ring or this table, to translate model names (X1900, HD4850) to/from chip names (R580, RV770 respectively).

Contents

Installation

There are three ways of installing Catalyst on your system. One way is to use Vi0L0's (Arch's unofficial Catalyst maintainer) repository. This repository contains all the necessary packages. The second method you can use is the AUR; PKGBUILDs offered here are also made by Vi0L0 and are the same he uses to built packages for his repository. Lastly, you can install the driver directly from AMD.

Before choosing the method you prefer, you will have to see which driver you need. Since Catalyst 12.4, AMD has separated its development for Radeon HD 2xxx, 3xxx and 4xxx cards into the legacy Catalyst driver. For Radeon HD 5xxx and newer, there is the regular Catalyst driver. Regardless of the driver you need, you will also need the Catalyst utilities.

Note: After the instructions for every method of installing, you will find general instructions everyone has to perform, regardless of the method you used.

Installing the driver

Installing from the unofficial repository

If you don't fancy building the packages from the AUR, this is the way to go. The repository is maintained by our unofficial Catalyst maintainer, Vi0L0. All packages are signed and are considered safe to use. As you will see later on in this article, Vi0L0 is also responsible for many other packages that will help you get your system working with your ATI graphic cards.

Vi0L0 has three different Catalyst repositories, each having different drivers:

  • [catalyst]; for the regular Catalyst driver needed by Radeon HD 5xxx and up, it contains the latest (stable or beta) Catalyst release;
  • [catalyst-stable]; for the regular Catalyst driver needed by Radeon HD 5xxx and up, with the latest stable driver;
  • [catalyst-hd234k]; for the legacy Catalyst driver needed by Radeon HD 2xxx, 3xxx and 4xxx cards.
Warning: The legacy Catalyst driver does not support Xorg 1.13. Should you want to use this driver, see #Xorg repositories for instructions on how to roll back to Xorg 1.12.
Note: Xorg 1.14 support has been included in the beta driver since 13.6.

To enable one of these, you will have to edit /etc/pacman.conf and add the repository of choice's information above all other repositories in /etc/pacman.conf.

  • For [catalyst], add:
[catalyst]
Server = http://catalyst.wirephire.com/repo/catalyst/$arch
  • For [catalyst-stable], add:
[catalyst-stable]
Server = http://catalyst.wirephire.com/repo/catalyst/$arch
  • For [catalyst-hd234k], add:
[catalyst-hd234k]
Server = http://catalyst.wirephire.com/repo/catalyst-hd234k/$arch

You must also add Vi0L0's GPG key so that pacman trusts the repositories.

# pacman-key --keyserver pgp.mit.edu --recv-keys 0xabed422d653c3094
# pacman-key --lsign-key 0xabed422d653c3094
Tip: Because catalyst.wirephire.com will go down if a certain bandwidth limit is exceeded (this happened in the past) or may be too slow at your location, repository mirrors are provided by yanom at [1] (USA) and rtsinformatique at [2] (France). These mirrors however come with no warranty and are not guaranteed to always be operational:
[catalyst]
#Server = http://70.239.162.206/catalyst-mirror/repo/catalyst/$arch
#Server = http://mirror.rts-informatique.fr/archlinux-catalyst/repo/catalyst/$arch
[catalyst-stable]
#Server = http://70.239.162.206/catalyst-mirror/repo/catalyst/$arch
#Server = http://mirror.rts-informatique.fr/archlinux-catalyst/repo/catalyst/$arch
[catalyst-hd234k]
#Server = http://70.239.162.206/catalyst-mirror/repo/catalyst-hd234k/$arch
#Server = http://mirror.rts-informatique.fr/archlinux-catalyst/repo/catalyst-hd234k/$arch

Uncomment the line for the mirror closest to your location. It is also a good idea to keep alternatives in case of mirror downtime.

Repository mirroring can be easily achieved using rsync://mirror.rts-informatique.fr::archlinux-catalyst.

Once you have done this, update pacman's database and install the packages:

# pacman -Syu
# pacman -S catalyst catalyst-utils
Note: If pacman asks you about removing libgl you can safely do so.
Warning: Catalyst from Vi0L0's repository is compiled against one kernel and one kernel only. For Vi0L0, it's getting more and more difficult to keep up with kernel's pace and as such, we have seen a lot of users asking for what to do in the forums. Often, the solution is to a) wait for Vi0L0 to rebuilt the Catalyst package against those newer versions, b) do so yourself or c) hold back the Linux kernel update. The easiest solution is to do so yourself, not manually, but by using Vi0L0's excellent #Catalyst-hook.

If you are on 64-bit and also need 32-bit OpenGL support:

# pacman -S lib32-catalyst-utils

Repositories also contain other packages, that can replace the catalyst package:

  • catalyst-generator; this package is able to generate fglrx modules packed into pacman compliant packages - most secure and KISS-compatible package in this side-note, although it has to be operated manually. It's described in #Catalyst-generator
  • catalyst-hook; a systemd service which will automatically update the fglrx module whilst the system shuts down or reboots. It's described in #Catalyst-hook

You will find more details about those packages in #Tools. Lastly, both repositories also contain the xvba-videoAUR package, which enables video acceleration, described in #Video acceleration, the AMDOverdriveCtrl package, which is a GUI to control over- and underclocking, also described in #GPU/Mem frequency, Temperature, Fan speed, Overclocking utilities, and a pack of tools for OpenCL/OpenGL developers, described in #OpenCL and OpenGL development

Installing from the AUR

The second way to install Catalyst is from the AUR. If you want to built the packages specifically for your computer, this is your way to go. Note that this is also the most tedious way to install Catalyst; it requires the most work and also requires manual updates upon every kernel update.

Warning: If you install the Catalyst package from the AUR, you will have to rebuild Catalyst every time the kernel is updated. Otherwise X will fail to start.

All packages mentioned above in Vi0L0's unofficial repository are also available on the AUR:

The AUR also holds some packages that are not found in any of the repositories. These packages contain the so-called Catalyst-total packages and the beta versions:

The catalyst-totalAUR packages are made to make the lives of AUR users easier. It builds the driver, the kernel utilities and the 32 bit kernel utilities. It also builds the catalyst-hookAUR package, which is explained briefly above and in more detail in its own section below.

catalyst-total-pxpAUR builds Catalyst with experimental powerXpress support.

Configuring the driver

After you have installed the driver via your chosen method, you will have to configure X to work with Catalyst. Also, you will have to make sure the module gets loaded at boot. Also, one should disable kernel mode setting.

Configuring X

To configure X, you will have to create an xorg.conf file. Catalyst provides its own aticonfig tool to create and/or modify this file. It also can configure virtually every aspect of the card for it also accesses the /etc/ati/amdpcsdb file. For a complete list of aticonfig options, run:

# aticonfig --help | less
Warning: Use the --output option before committing to /etc/X11/ as an xorg.conf file will override anything in /etc/X11/xorg.conf.d/
Note: To adhere to the new config location use # aticonfig [...] --output to adapt the Device section to /etc/X11/xorg.conf.d/20-radeon.conf. The drawback is that many aticonfig options rely on an xorg.conf, and will be unavailable.

Now, to configure Catalyst. If you have only one monitor, run this:

# aticonfig --initial
Note: If you have a PowerXpress problem you should probably install catalyst-total-pxpAUR.

However, if you have two monitors and want to use both of them, you can run the command stated below. Note that this will generate a dual head configuration with the second screen located above the first screen.

# aticonfig --initial=dual-head --screen-layout=above
Note: See #Double Screen (Dual Head / Dual Screen / Xinerama) for more information on setting up dual monitors.

You can compare the generated file to one of the Sample Xorg.conf examples listed on the Xorg page.

Although the current Xorg versions auto-detect most options when started, you may want to specify some in case the defaults change between versions.

Here is an example (with notes) for reference. Entries with # should be required, add entries with ## as needed:

/etc/X11/xorg.conf
Section "ServerLayout"
        Identifier     "Arch"
        Screen      0  "Screen0" 0 0          # 0's are necessary.
EndSection
Section "Module"
        Load [...]
        [...]
EndSection
Section "Monitor"
        Identifier   "Monitor0"
        [...]
EndSection
Section "Device"
        Identifier  "Card0"
        Driver      "fglrx"                         # Essential.
        BusID       "PCI:1:0:0"                     # Recommended if autodetect fails.
        Option      "OpenGLOverlay" "0"             ##
        Option      "XAANoOffscreenPixmaps" "false" ##
EndSection
Section "Screen"
        Identifier "Screen0"
        Device     "Card0"
        Monitor    "Monitor0"
        DefaultDepth    24
        SubSection "Display"
                Viewport   0 0
                Depth     24                        # Should not change from '24'
                Modes "1280x1024" "2048x1536"       ## 1st value=default resolution, 2nd=maximum.
                Virtual 1664 1200                   ## (x+64, y) to workaround potential OGL rect. artifacts/
        EndSubSection                               ## fixed in Catalyst 9.8
EndSection
Section "DRI"
        Mode 0666                                   # May help enable direct rendering.
EndSection
Note: With every Catalyst update you should remove amdpcsdb file in this way: kill X, remove /etc/ati/amdpcsdb, start X and then run amdcccle - otherwise the version of Catalyst may display wrongly in amdcccle.

If you need more information on Catalyst, visit this thread.

Loading the module at boot

We have to blacklist the radeon module to prevent it from auto-loading. To do so, blacklist radeon in /etc/modprobe.d/modprobe.conf. Also, make sure that it is not loaded by any file under /etc/modules-load.d/. For more information, see kernel modules#Blacklisting.

Then we will have to make sure that the fglrx module gets auto-loaded. Either add fglrx on a new line of an existing module file located under /etc/modules-load.d/, or create a new file and add fglrx.

Disable kernel mode setting

Disabling kernel mode setting is important, as the driver doesn't take advantage of KMS yet. If you do not deactivate KMS, your system might freeze when trying to switch to a TTY or even when shutting down via your DE.

To disable kernel mode setting, add nomodeset to your kernel parameters.

Checking operation

Assuming that a reboot to your login was successful, you can check if fglrx is running properly with the following commands:

$ lsmod | grep fglrx
$ fglrxinfo

If you get output, it works. Finally, run X with $ startx or by using GDM/KDM and verify that direct rendering is enabled by running the following command in a terminal:

$ glxinfo | grep "direct rendering"

If it says "direct rendering: yes" then you're good to go! If the $ glxinfo command is not found install the mesa-demos package.

Note: You can also use:
$ fgl_glxgears

as the fglrx alternative test to glxgears.

Warning: In recent versions of Xorg, the paths of libs are changed. So, sometimes libGL.so cannot be correctly loaded even if it's installed. Check this if your GL is not working. Please read #Troubleshooting section for details.

Custom kernels

Note: If you are at all uncomfortable or inexperienced with making packages, read up the ABS wiki page first so things go smoothly.

To install catalyst for a custom kernel, you'll need to build your own catalyst-$kernel package:

  1. Obtain the PKGBUILD and catalyst.install files from Catalyst.
  2. Editing the PKGBUILD. Two changes need to be made here:
    1. Change pkgname=catalyst to pkgname=catalyst-$kernel_name, where $kernel_name is whatever you want (e.g. custom, mm, themostawesomekernelever).
    2. Change the dependency of linux to $kernel_name.
  3. Build your package and install; run makepkg -i or makepkg followed by # pacman -U pkgname.pkg.tar.gz
Note:
  • If you run multiple kernels, you have to install the catalyst-utils packages for all kernels. They won't conflict with one another.
  • catalyst-generator is able to build catalyst-{kernver} packages for you so you do not actually need to perform all those steps manually. For more information, see Tools section.

PowerXpress support

PowerXpress technology allows switching notebooks with dual-graphic capability from integrated graphics (IGP) to discrete graphics either to increase battery life or to achieve better 3D rendering capabilities.

To use such functionality on Arch you will have to:

  • Get and build catalyst-total-pxpAUR package from the AUR, or
  • Install catalyst-utils-pxp package from the [catalyst] repository (plus additional lib32-catalyst-utils-pxp, if needed).

To perform a switch into Intel's IGP you will also have to install the mesa-libgl package and Intel's drivers: xf86-video-intel and intel-dri.

Note: With the latest version of Catalyst, version 13.1 (not Catalyst legacy) ChrisXY was able to work on the newest xorg-server (version 1.13.1), mesa 9.0.1 and xf86-video-intel 2.20.18.

On any version of Catalyst below 13.1 (and all versions of Catalyst legacy) there are some problems with the new Intel drivers and the last noted working version of xf86-video-intel is 2.20.2-2, so you will probably have to downgrade from the latest version that you have gotten from Arch's repositories (although we recommend to test the newest one before downgrading - there's always some possibility that it will work).

xf86-video-intel 2.20.2-2 works only with xorg-server 1.12 and so it is a part of the xorg112 repository. If you want to use it you will have to downgrade xorg-server as well. For information on this, see #Xorg repositories.

Now you can switch between the integrated and the discrete GPU, using these commands:

# aticonfig --px-igpu    #for integrated GPU
# aticonfig --px-dgpu    #for discrete GPU

Just remember that fglrx needs /etc/X11/xorg.conf configured for AMD's card with fglrx inside.

You can also use the pxp_switch_catalyst switching script that will perform some additional usefull operations:

  • Switching xorg.conf - it will rename xorg.conf into xorg.conf.cat (if there's fglrx inside) or xorg.conf.oth (if there's intel inside) and then it will create a symlink to xorg.conf, depending on what you chose.
  • Running aticonfig --px-Xgpu.
  • Running switchlibGL.
  • Adding/removing fglrx into/from /etc/modules-load.d/catalyst.conf.

Usage:

# pxp_switch_catalyst amd
# pxp_switch_catalyst intel

If you have got problems when you try to run X on Intel's driver you may try to force "UXA" acceleration. Just make sure you got Option "AccelMethod" "uxa" in xorg.conf:

/etc/X11/xorg.conf
Section "Device"
        Identifier  "Intel Graphics"
        Driver      "intel"
        #Option      "AccelMethod"  "sna"
        Option      "AccelMethod"  "uxa"
        #Option      "AccelMethod"  "xaa"
EndSection

Running two X servers (one using the Intel driver, another one using fglrx) simultaneously

Because fglrx is crash-prone (regarding PowerXpress), it could be a good idea to use the Intel driver in the main X server and have a secondary X server using fglrx when 3D acceleration is needed. However, simply switching to the discrete GPU from the integrated GPU using aticonfig or amdcccle will cause all sorts of weird bugs when starting the second X.

To run two X servers at the same time (each using different drivers), you should firstly set up a fully working X with Catalyst and then move its xorg.conf to a temporary place (for example, /etc/X11/xorg.conf.fglrx. The next time X is started, it will use the Intel driver by default instead of fglrx.

To start a second X server using fglrx, simply move xorg.conf back to the proper place (/etc/X11/xorg.conf) before starting X. This method even allows you to switch between running X sessions. When you are done using fglrx, move xorg.conf somewhere else again.

The only disadvantage of this method is not having 3D acceleration using the Intel driver. 2D acceleration, however, is fully functional. Other than that, this will provide us with a completely stable desktop.

Xorg repositories

Catalyst is notorious for its slow update process. As such, it is common that a new Xorg version is pushed down from upstream that will break compatibility for Catalyst. This means that Catalyst users either have to hold the Xorg packages from updating, or use a backported repository that only contains the Xorg packages that should be hold back. Vi0L0 has stepped in to fulfil this task and provides several backported repositories.

If you want to use pacman to hold back packages from updating, see pacman#Skip package from being upgraded. Packages you should hold back, are:

  • xorg-server-*
  • xf86-input-*
  • xf86-video-*

If you want to use the backported repositories, you have to edit /etc/pacman.conf and add the information of the repository above all other repositories, even above your Catalyst repository, should you use one.

[xorg113]

Catalyst < 13.8 doesn't support xorg-server 1.14.

[xorg113]
Server = http://catalyst.wirephire.com/repo/xorg113/$arch

[xorg112]

Catalyst < 12.10 and Catalyst Legacy do not support xorg-server 1.13.

[xorg112]
Server = http://catalyst.wirephire.com/repo/xorg112/$arch

[xorg111]

Catalyst < 12.6 doesn't support xorg-server 1.12.

[xorg111]
Server = http://catalyst.wirephire.com/repo/xorg111/$arch

Tools

Catalyst-hook

Catalyst-hookAUR is a systemd service that will automatically rebuild the fglrx modules while the system shuts down or reboots, but only if it's necessary (e.g. after an update).

Before using this package make sure that both the base-devel group and the linux-headers package (the one specific to the kernel you use) are installed.

To enable the automatic update, enable the catalyst-hook.service:

# systemctl enable catalyst-hook
# systemctl start catalyst-hook

You can also use this package to build the fglrx module manually. Simply run the catalyst_build_module script after the kernel has been updated:

# catalyst_build_module all

A few more technical details:

The catalyst-hook.service is stopping the systemd "river" and is forcing systemd to wait until catalyst-hook finishes its job.

The catalyst-hook.service is calling the catalyst_build_module check function which checks if fglrx rebuilds are really necessary.

The check function is checking if the fglrx module exists, if it:

  • doesn't exist, it will build it;
  • does exist, it will compare the two values to be sure that a rebuild is necessary.

These values are md5sums of the /usr/lib/modules/<kernel_version>/build/Module.symvers file (because I, Vi0L0, noticed that this file is unique and different for every kernel's release). The first value is the md5sum of the existing Module.symvers file. The second value is the md5sum of the Module.symvers file which existed in a moment of the fglrx module creation. This value was compiled into the fglrx module by a catalyst_build_module script.

If the values are different, it will compile the new fglrx module.

The check is checking the whole /usr/lib/modules/ directory and building modules for all of the installed kernels if it's necessary. If the build or rebuild isn't necessary, the whole process takes only some milliseconds to complete before it gets killed by systemd.

Catalyst-generator

catalyst-generatorAUR is a package that is able to build and install the fglrx module packed into pacman compliant catalyst-${kernver} packages. The basic difference from #Catalyst-hook is that you will have to trigger this command manually, whereas Catalyst-hook will do this automatically at boot when a new kernel got installed.

It creates catalyst-${kernver} packages using makepkg and installs them with pacman. ${kernver} is the kernel version for which each package was built (e.g. catalyst-2.6.35-ARCH package was built for 2.6.35-ARCH kernel).

To build and install catalyst-${kernver} package for a currently booted kernel as an unprivileged user (non-root; safer way), use catalyst_build_module. You will be asked for your root password to proceed to package installation.

A short summary on how to use this package:

  1. As root: catalyst_build_module remove. This will remove all unused catalyst-{kernver} packages.
  2. As unprivileged user: catalyst_build_module ${kernver}, where ${kernver} is the version of the kernel to which you just updated. For example: catalyst_build_module 2.6.36-ARCH. You can also build catalyst-{kernver} for all installed kernels by using catalyst_build_module all.
  3. If you want to remove catalyst-generator, it's best to run this as root before removing catalyst-generator: catalyst_build_module remove_all. This will remove all catalyst-{kernver} packages from the system.

Catalyst-generator isn't able to remove all those catalyst-{kernver} packages automatically while being removed because there can not be more than one instance of pacman running. If you forget to run # catalyst_build_module remove_all before using # pacman -R catalyst-generator catalyst-generator will tell you which catalyst-{kernver} packages you will have to remove manually after removing catalyst-generator itself.

Catalyst-generator is most safe and KISS-friendly solution because:

  1. you can use unprivileged user to build the package;
  2. it is building modules in a fakeroot environment;
  3. it is not throwing files here and there, pacman always knows where they are;
  4. all you have to do is to remember to use it
Note: If you see those warnings:
WARNING: Package contains reference to $srcdir
WARNING: '.pkg' is not a valid archive extension.
while building catalyst-{kernver} package, do not be concerned, it's normal.

OpenCL and OpenGL development

Since years AMD is working on tools for OpenCL and OpenGL developement.

Now under the banner of "Heterogeneous Computing" AMD is providing even more of them, fortunately most of their computing tools are available also for Linux.

In the AUR and the [catalyst] repositories you will find packages that represent the most important work from AMD, namely; amdapp-aparapi, amdapp-sdk and amdapp-codexl.

APP shortcut stands for Accelerated Parallel Processing.

amdapp-aparapi

AMD's Aparapi is an API for expressing data parallel workloads in Java and a runtime component capable of converting the Java bytecode of compatible workloads into OpenCL so that it can be executed on a variety of GPU devices. If Aparapi can’t execute on the GPU, it will execute in a Java thread pool.

You can find more information about Aparapi here.

amdapp-sdk (formerly known as amdstream)

The AMD APP Software Development Kit (SDK) is a complete development platform created by AMD to allow you to quickly and easily develop applications accelerated by AMD APP technology. The SDK provides samples, documentation and other materials to quickly get you started leveraging accelerated compute using OpenCL, Bolt, or C++ AMP in your C/C++ application.

Since version 2.8 amdapp-sdk is providing aparapiUtil as well as aparapi's samples. A package is available on the [catalyst] repository; it depends on the amdapp-aparapiAUR package. The AUR's package will let you decide whether you want aparapi's additions or not.

Version 2.8 does not provide Profiler functionality, it has been moved to CodeXL.

You can find more information about AMD APP SDK here.

amdapp-codexl

CodeXL is an OpenCL and OpenGL Debugger and Profiler, with a static OpenCL kernel analyzer. It's a GUI application written atop of the well known gDEBugger and is available only for x86_64 systems.

You can find more informations about CodeXL here.

Features

Tear Free Rendering

Presented in Catalyst 11.1, the Tear Free Desktop feature reduces tearing in 2D, 3D and video applications. This likely adds triple-buffering and v-sync. Do note that it requires additional GPU processing.

To enable 'Tear Free Desktop' run amdcccle and go to: Display OptionsTear Free.

Or as root run:

# aticonfig --set-pcs-u32=DDX,EnableTearFreeDesktop,1

To disable, again use amdcccle or run as root:

# aticonfig --del-pcs-key=DDX,EnableTearFreeDesktop

Video acceleration

Video Acceleration API (VA API) is an open source software library (libVA) and API specification which provides GPU acceleration for video processing on Linux/UNIX based operating systems. The process works by enabling hardware accelerated video decode at various entry-points (VLD, IDCT, Motion Compensation, deblocking) for common encoding standards (MPEG-2, MPEG-4 ASP/H.263, MPEG-4 AVC/H.264, and VC-1/WMV3).

VA-API gained a proprietary backend (in November 2009) called xvba-videoAUR, that allows VA-API programmed applications to take advantage of AMD Radeons UVD2 chipsets via the XvBA (X-Video Bitstream Acceleration API designed by AMD) library.

XvBA support and xvba-video is still under development, however it is working very well in most cases. Build (or install from Vi0L0's repo) the proprietary xvba-videoAUR package, or if you have problems with that version, install either xvba-video-openAUR or libva-xvba-driverAUR; and also install mplayer-vaapi and libva. Then just set your video player to use vaapi:gl as video output:

$ mplayer -vo vaapi:gl movie.avi

These options can be added to your mplayer configuration file, see MPlayer.

For smplayer:

Options → Preferences → General → Video (tab) → Output driver: User Defined : vaapi:gl
Options → Preferences → General → Video (tab) → Double buffering on
Options → Preferences → General → General → Screenshots → Turn screenshots off
Options → Preferences → Performance → Threads for decoding: 1 (to turn off -lavdopts parameter)
Note: If Tear Free Desktop is enabled it's better to use:
Options → Preferences → General → Video (tab) → Output driver: vaapi

If Video Output vaapi:gl isn't working - please check:

vaapi, vaapi:gl2 or simply xv(0 - AMD Radeon AVIVO Video).

For VLC:

Tools → Preferences → Input & Codecs → Use GPU accelerated decoding

It might help to enable v-sync in amdcccle:

3D → More Settings → Wait for vertical refresh = Always On
Note: If you are using Compiz/KWin, the only way to avoid video flickering is to watch videos in full-screen and only when Unredirect Fullscreen is off. In compiz you need to set Redirected Direct Rendering in General Options of ccsm. If it is still flickering, try to disable this option in CCSM. It's off by default in KWin, but if you see flickering try to turn "Suspend desktop effects for fullscreen windows" on or off in System SettingsDesktop EffectsAdvanced.

GPU/Mem frequency, Temperature, Fan speed, Overclocking utilities

You can get the GPU/Mem clocks with: $ aticonfig --od-getclocks.

You can get the fan speed with: $ aticonfig --pplib-cmd "get fanspeed 0"

You can get the temperature with: $ aticonfig --odgt

To set the fanspeed with: $ aticonfig --pplib-cmd "set fanspeed 0 50" Query Index: 50, Speed in percent

To overclock and/or underclock it's easier to use a GUI, like ATi Overclocking Utility, which is very simple and requires qt to work. It might be out of date/old, but you can get it here.

Another, more complex utility to perform such operations is AMDOverdriveCtrl. Its homepage is here and you can build amdoverdrivectrlAUR from the AUR or from Vi0L0's unofficial repositories.

Double Screen (Dual Head / Dual Screen / Xinerama)

Introduction

Warning: you should know that there isn't one specific solution because each setup differs and needs its own configuration. That's why you will have to adapt the steps below to your own needs. It is possible that you have to try more than once. Therefore, you should save your working /etc/X11/xorg.conf before you start modifying and you must be able to recover from a command-line environment.
  • In this chapter, we will describe the installation of two different-sized screens on only one graphics card with two different output ports (DVI + HDMI) using a "BIG Desktop" configuration.
  • The Xinerama solution has some inconveniences, especially because it is not compatible with XrandR. For that very reason, you should not use this solution, because XrandR is a must for our later configuration.
  • The Dual Head solution would allow you to have 2 different sessions (one for each screen). It could be what you want, but you will not be able to move windows from one screen to another. If you have only one screen, you will have to define the mouse inside your Xorg session for each of the two sessions inside the Server Layout section.

ATI Documentation

ATI Catalyst Control Center

The GUI tool shipped by ATI is very useful and we will try to use it as much as we can. To launch it, open a terminal and use the following command:

$ {kdesu/gksu} amdcccle
Warning: Do not use sudo directly with a GUI. Sudo gives you admin rights with user account information. Instead, use gksu (GNOME) or kdesu (KDE).

Installation

Before we start, make sure that your hardware is plugged in correctly, that power is on and that you know your hardware characteristics (screen dimensions, sizes, refreshment rates, etc.) Normally, both screens are recognized during boot time but not necessarily identified properly, especially if you are not using any Xorg base configuration file (/etc/X11/xorg.conf) but relying on the hot-plugging feature.

The first step is to make sure that you screens will be recognized by your DE and by X. For this, you need to generate a basic Xorg configuration file for your two screens:

# aticonfig --initial --desktop-setup=horizontal --overlay-on=1

or

# aticonfig --initial=dual-head --screen-layout=left
Note: overlay is important because it allows you to have 1 pixel (or more) shared between the 2 screens.
Tip: For the other possible and available options, do not hesitate to type aticonfig --help inside a terminal to display all available command lines.

Now you should have a basic Xorg configuration file that you can edit to add your screen resolutions. It is important to use the precise resolution, especially if you have screens of different sizes. These resolutions have to be added in the "Screen" section:

SubSection "Display"
    Depth     24
    Modes     "X-resolution screen 1xY-resolution screen 1" "Xresolution screen 2xY-resolution screen 2"
EndSubSection

From now on, instead of editing the xorg.conf file manually, let us use the ATI GUI tool. Restart X to be sure that your two screens are properly supported and that the resolutions are properly recognized (Screens must be independent, not mirrored).

Configuration

Now you will only have to launch the ATI control center with root privileges, go to the display menu and choose how you would like to set your configuration (small arrow of the drop down menu). A last restart of X and you should be done!

Before you restart X, do not hesitate to verify your new xorg.conf file. At this stage, inside the "Display" sub-section of the "Screen" section, you should see a "Virtual" command line, of which the resolution should be the sum of both screens. The "Server Layout" section says all the rest.

Uninstallation

If for any reason this driver is not working for you or if you simply want to try out the open source driver, remove the catalyst and catalyst-utils packages. Also you should remove catalyst-generatorAUR, catalyst-hookAUR and lib32-catalyst-utilsAUR packages if they have been installed on your system.

Warning:
  • You may need to use # pacman -Rdd to remove catalyst-utilsAUR (and/or lib32-catalyst-utilsAUR) because that package contains gl related files and many of your installed packages depend on them. These dependencies will be satisfied again when you install xf86-video-ati.
  • You may need to remove /etc/profile.d/ati-flgrx.sh and /etc/profile.d/lib32-catalyst (if it exists on your system), otherwise r600_dri.so will fail to load and you would not have 3D support.
Note: You should remove unofficial repositories from your /etc/pacman.conf and run # pacman -Syu, because those repositories include out-dated Xorg packages to allow use of catalyst and the xf86-video-ati package needs up-to-date Xorg packages from the Official repositories.

Also follow these steps:

  • If you have the /etc/modprobe.d/blacklist-radeon.conf file remove it or comment the line blacklist radeon in that file.
  • If you have a file in /etc/modules-load.d to load the fglrx module on boot, remove it or comment the line containing fglrx.
  • Make sure to remove or backup /etc/X11/xorg.conf.
  • If you have installed the catalyst-hookAUR package, make sure to disable the systemd service.
  • If you used the nomodeset option in your kernel parameters and plan to use KMS, remove it.
  • Reboot before installing another driver.

Troubleshooting

If you can still boot to command-line, then the problem probably lies in /etc/X11/xorg.conf

You can parse the whole /var/log/Xorg.0.log or, for clues:

$ grep '(EE)' /var/log/Xorg.0.log
$ grep '(WW)' /var/log/Xorg.0.log

If you are still confused about what is going on, search the forums first. Then post a message in the thread specific to ATI/AMD. Provide the information from xorg.conf and both commands mentioned above.

3D Wine applications freeze

If you use a 3D Wine application and it hangs, you have to disable TLS. To do this, either use aticonfig or edit /etc/X11/xorg.conf. To use aticonfig:

# aticonfig --tls=off

Or, to edit /etc/X11/xorg.conf; first open the file in an editor as root and then add Option "UseFastTLS" "off" to the Device section of this file.

After applying either of the solutions, restart X for it to take effect.

Problems with video colours

You may still use vaapi:gl to avoid video flickering, but without video acceleration:

  • Run mplayer without -vo vaapi switch.
  • Run smplayer remove -vo vaapi from Options → Preferences → Advanced → Options for MPlayer → Options: -vo vaapi

Plus for smplayer you may now safely turn screenshots on.

KWin and composite

You may use XRender if the rendering with OpenGL is slow. However, XRender might also be slower than OpenGL depending on your card. XRender also solves artefact issues in some cases, for instance when resizing Konsole.

Black screen with complete lockups and/or hangs after reboot or startx

Ensure you have added the nomodeset option to the kernel options line in your bootloader (see #Disable kernel mode setting).

If you are using the legacy driver (catalyst-hd234k) and get a black screen, try downgrading xorg-server to 1.11 by using the #[xorg111] repository.

Faulty ACPI hardware calls

It is possible that fglrx doesn't cooperate well with the system's ACPI hardware calls, so it auto-disables itself and there is no screen output.

If so, try to run this:

$ aticonfig --acpi-services=off

KDM disappears after logout

If you are running Catalyst proprietary driver and you get a console (tty1) instead of the expected KDM greeting when you log out, you must instruct KDM to restart the X server after each logout. Uncomment the following line under the section titled [X-:*-Core]

/usr/share/config/kdm/kdmrc
TerminateServer=True

KDM should now appear when you log out of KDE.

Direct Rendering does not work

This problem may occur when using the proprietary Catalyst driver.

Warning: This error would also appear if you have not rebooted your system after the installation or upgrade of catalyst. The system needs to load the fglrx.ko module in order to make the driver work.

If you have problem with direct rendering, run:

$ LIBGL_DEBUG=verbose glxinfo > /dev/null

at the command prompt. At the very start of the output, it'll usually give you a nice error message saying why you do not have direct rendering.

Common errors and their solutions, are:

libGL error: XF86DRIQueryDirectRenderingCapable returned false
  • Ensure that you are loading the correct agp modules for your AGP chipset before you load the fglrx kernel module. To determine which agp modules you'll need, run # hwdetect --show-agp. Then open your /etc/modules-load.d/fglrx.conf and add the agp module on a line before the fglrx line.
libGL error: failed to open DRM: Operation not permitted
libGL error: reverting to (slow) indirect rendering
libGL: OpenDriver: trying /usr/lib/xorg/modules/dri//fglrx_dri.so
libGL error: dlopen /usr/lib/xorg/modules/dri//fglrx_dri.so failed
(/usr/lib/xorg/modules/dri//fglrx_dri.so: cannot open shared object file: No such file or directory)
libGL error: unable to find driver: fglrx_dri.so
  • Something has not been installed correctly. If the paths in the error message are /usr/X11R6/lib/modules/dri/fglrx_dri.so, then ensure you've logged completely out of your system, then back in. If you're using a graphical login manager (gdm, kdm, xdm), ensure that /etc/profile is sourced every time you log in. This is usually accomplished by adding source /etc/profile into ~/.xsession or ~/.xinitrc, but this may vary between login managers.
  • If the paths above in your error message are /usr/lib/xorg/modules/dri/fglrx_dri.so, then something has not been correctly installed. Try reinstalling the catalystAUR package.

Errors such as:

fglrx: libGL version undetermined - OpenGL module is using glapi fallback

could be caused by having multiple versions of libGL.so on your system. The command below should return the following output:

$ locate libGL.s
/usr/lib/libGL.so
/usr/lib/libGL.so.1
/usr/lib/libGL.so.1.2

These are the only three libGL.so files you should have on your system. If you have any more (e.g. /usr/X11R6/lib/libGL.so.1.2), then remove them. This should fix your problem.

You might not get any error to indicate that this is a problem. If you are using X11R7, make sure you do not have these files on your system:

/usr/X11R6/lib/libGL.so.1.2
/usr/X11R6/lib/libGL.so.1

Hibernate/Sleep issues

Video fails to resume from suspend2ram

ATI's proprietary Catalyst driver cannot resume from suspend if the framebuffer is enabled. To disable the framebuffer, add vga=0 to your kernel kernel parameters.

To see where you need to add this with other bootloaders, see #Disable kernel mode setting.

System freezes/Hard locks

  • The radeonfb framebuffer drivers have been known in the past to cause problems of this nature. If your kernel has radeonfb support compiled in, you may want to try a different kernel and see if this helps.
  • If you experience system freezes when exiting your DE (shut down, suspend, switching to tty etc.) you probably forgot to deactivate KMS. (See #Disable kernel mode setting)

Hardware conflicts

Radeon cards used in conjunction with some versions of the nForce3 chipset (e.g. nForce 3 250Gb) won't have 3D acceleration. Currently the cause of this issue is unknown, but some sources indicate that it may be possible to get acceleration with this combination of hardware by booting Windows with the drivers from nVIDIA and then rebooting the system. This can be verified by getting output something similar to this (using an nForce3-based system):

$ dmesg | grep agp
agpgart: Detected AGP bridge 0
agpgart: Setting up Nforce3 AGP.
agpgart: aperture base > 4G

and also if issuing the following command gets you the following output:

$ tail -n 100 /var/log/Xorg.0.log | grep agp
 (EE) fglrx(0): [agp] unable to acquire AGP, error "xf86_ENODEV"

you have this bug.

Some sources indicate that in some situations, downgrading the motherboard BIOS may help, but this cannot be verified in all cases. Also, a bad BIOS downgrade can render your hardware useless, so beware.

See this bugreport for more information and a potential fix.

Temporary hangs when playing video

This problem may occur when using the proprietary Catalyst.

If you experience temporary hangs lasting from a few seconds to several minutes occuring randomly during playback with mplayer, check /var/log/messages.log for output like:

Nov 28 18:31:56 pandemonium [<c01c64a6>] ? proc_get_sb+0xc6/0x160
Nov 28 18:31:56 pandemonium [<c01c64a6>] ? proc_get_sb+0xc6/0x160
Nov 28 18:31:56 pandemonium [<f8bc628c>] ? ip_firegl_ioctl+0x1c/0x30 [fglrx]
Nov 28 18:31:56 pandemonium [<c01c64a6>] ? proc_get_sb+0xc6/0x160
Nov 28 18:31:56 pandemonium [<c0197038>] ? vfs_ioctl+0x78/0x90
Nov 28 18:31:56 pandemonium [<c01970b7>] ? do_vfs_ioctl+0x67/0x2f0
Nov 28 18:31:56 pandemonium [<c01973a6>] ? sys_ioctl+0x66/0x70
Nov 28 18:31:56 pandemonium [<c0103ef3>] ? sysenter_do_call+0x12/0x33
Nov 28 18:31:56 pandemonium [<c01c64a6>] ? proc_get_sb+0xc6/0x160
Nov 28 18:31:56 pandemonium =======================

Adding the nopat kernel option to your kernel options in your bootloader and rebooting fixed the problem at least for me. To see how to do this for different bootloaders, see #Disable kernel mode setting.

"aticonfig: No supported adapters detected"

If you get:

# aticonfig --initial
aticonfig: No supported adaptaters detected

but you do have an AMD GPU (or APU), it may still be possible to get Catalyst working by manually setting the device in your your etc/X11/xorg.conf file or by copying an older working /etc/ati/control file (preferred - this also fixes the watermark issue).

To get an older control file, download a previous version of fglrx from AMD and run it with --extract driver parameter. You'll find the control file in driver/common/etc/ati/control. Copy the extracted file over the system file and restart Xorg. You can try different versions of the file.

To set your model in xorg.conf, edit the device section of /etc/X11/xorg.conf to:

/etc/X11/xorg.conf
Section "Device"
        Identifier "ATI radeon ****"
        Driver     "fglrx"
EndSection

Where **** should be replaced with your device's marketing number (e.g. 6870 for the HD 6870 and 6310 for the E-350 APU).

Xorg will start and it is possible to use amdcccle instead of aticonfig. There will be an "AMD Unsupported hardware" watermark.

You can remove this watermark using the following script:

#!/bin/sh
DRIVER=/usr/lib/xorg/modules/drivers/fglrx_drv.so
for x in $(objdump -d $DRIVER|awk '/call/&&/EnableLogo/{print "\\x"$2"\\x"$3"\\x"$4"\\x"$5"\\x"$6}'); do
   sed -i "s/$x/\x90\x90\x90\x90\x90/g" $DRIVER
done

and rebooting.

WebGL support in Chromium

Google has blacklisted Linux's Catalyst driver from supporting webGL in their Chromium/Chrome browsers.

You can turn webGL on by editing and adding --ignore-gpu-blacklist flag into the Exec line so it looks like this:

/usr/share/applications/chromium.desktop
Exec=chromium %U --ignore-gpu-blacklist

You can also run chromium from console with the same --ignore-gpu-blacklist flag:

$ chromium --ignore-gpu-blacklist
Warning: Catalyst does not support the GL_ARB_robustness extension, so it is possible that a malicious site could use WebGL to perform a DoS attack on your graphic card. For more info, read this.

Laggs/freezes when watching flash videos via Adobe's flashplugin

Edit:

/etc/adobe/mms.cfg
#EnableLinuxHWVideoDecode=1
OverrideGPUValidation=true

If you are using KDE make sure that "Suspend desktop effects for fullscreen windows" is unchecked under System SettingsDesktop EffectsAdvanced.

Laggs/slow windows movement in GNOME3

You can try this solution out, it's working for many people.

Add this line into ~/.profile or into /etc/profile:

export CLUTTER_VBLANK=none

Restart X server or reboot your system.

Not using fullscreen resolution at 1920x1080 (underscanning)

Using the amdcccle GUI you can select the display, go to adjustments, and set Underscan to 0% (aticonfig defaults to 15% underscan).

Alternatively, you can use aticonfig to disable underscanning as well:

# aticonfig --set-pcs-val=MCIL,DigitalHDTVDefaultUnderscan,0

For newer version (for example, 12.11), if Catalyst control center repeatedly fails to save the overscan setting, edit /etc/ati/amdpcsdb like so:

/etc/ati/amdpcsdb
TVEnableOverscan=V0

Then logout and login.

Dual Screen Setup: general problems with acceleration, OpenGL, compositing, performance

Try to disable xinerama and xrandr12. Check out ie. this way:

Type those commands:

# aticonfig --initial
# aticonfig --set-pcs-str="DDX,EnableRandR12,FALSE"

Then reboot your system. In /etc/X11/xorg.conf check that xinerama is disabled, if it's not disable it and reboot your system.

Next run amdcccle and pick up amdcccle → display manager → multi-display → multidisplay desktop with display(s) 2.

Reboot again and set up your display layout whatever you desire.

Disabling VariBright feature

Type the following command to disable VariBright:

# aticonfig --set-pcs-u32=MCIL,PP_UserVariBrightEnable,0

Hybrid/PowerXpress: turning off discrete GPU

When you are using catalyst-total-pxpAUR or catalyst-utils-pxpAUR and you are switching to integrated GPU you may notice that discrete GPU is still working, consuming power and making your system's temperature higher.

Sometimes ie. when your integrated GPU is intel's one you can use vgaswitcheroo to turn the discrete GPU off. Sometimes unfortunatelly, it's not working.

Then you may check out acpi_call. MrDeepPurple has prepared the script which he's using to perform 'turn off' task, he's calling script via systemd service while booting and resuming his system. Here's his script:

#!/bin/sh
libglx=$(/usr/lib/fglrx/switchlibglx query)
modprobe acpi_call
if [ "$libglx" = "intel" ]; then
    echo '\_SB.PCI0.PEG0.PEGP._OFF' > /proc/acpi/call
fi

Switching from X session to TTYs gives blank screen

Workaround for this "feature", which appeared in catalyst 13.2 betas, is to use vga= kernel option, like vga=792. You can get the list of supported resolutions with

$ hwinfo --framebuffer

command. Pick up the one from the very bottom, and copy-paste it into kernel line of your bootloader, so it could look like ie. vga=0x03d4

TV showing a black border around the screen

This may happen when displaying on a TV through HDMI. When using the catalyst driver this cannot be solved via xrandr (which could with the xf86-video-ati). The solution is to disable overscan.

# aticonfig --set-pcs-val=MCIL,DigitalHDTVDefaultUnderscan,0

30 fps / tear free / v-sync bug

Bug introduced in catalyst 13.6 beta, not fixed till now (13.9).

After enabling "Tear Free" functionality every freshly started OpenGL application is lagging, often generates only 30 fps, it also touches composited desktop.

Workaround is pretty simple and was found by M132. Here are the steps, do everything in "AMD Catalyst Control Center" (amdcccle) application:

1. Enable Tear-Free, it will set 3D V-Sync option to Always on.
2. Set 3D V-Sync to Always Off.
3. Make sure Tear-Free is still on.
4. Restart X / Re-login.

It's working well also on KDE 4.11.x but in case of problems M132 suggests: "try disabling "Detect refresh rate" and specify monitor's refresh rate in Composite plugin."