Announcement

Collapse
No announcement yet.

AMD Radeon GPUs Run Great With Linux 3.11 Kernel, Mesa 9.3-devel

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Just tried flightgear and running it with rembrandt results in weird rendering issues which probably are a result of the following errors:

    Code:
    EE r600_shader.c:158 r600_pipe_shader_create - translation from TGSI failed !
    EE r600_state_common.c:754 r600_shader_select - Failed to build shader variant (type=1) -1
    Code:
    FRAGMENT glCompileShader "/usr/share/flightgear/Shaders/urban-gbuffer.frag" FAILED
    FRAGMENT Shader "/usr/share/flightgear/Shaders/urban-gbuffer.frag" infolog:
    0:91(22): preprocessor error: syntax error, unexpected IDENTIFIER, expecting NEWLINE
    0:88(1): preprocessor error: Unterminated #if
    glLinkProgram "" FAILED
    Program "" infolog:
    error: linking with uncompiled shader
    Code:
    FRAGMENT glCompileShader "/usr/share/flightgear/Shaders/ubershader-gbuffer.frag" FAILED
    FRAGMENT Shader "/usr/share/flightgear/Shaders/ubershader-gbuffer.frag" infolog:
    0:134(24): error: could not implicitly convert operands to relational operator
    0:134(24): error: if-statement condition must be scalar boolean
    glLinkProgram "" FAILED
    Program "" infolog:
    error: linking with uncompiled shader
    The first one seems to be a driver problem. I can't tell if the following ones are errors in the shader or in the driver, but on a first look the shaders looks ok and the driver simply has a problem with the preprocessor stuff. The second ones compares an integer to an float without conversation so this might be an error in the shader.
    Last edited by droste; 08-15-2013, 02:52 AM.

    Comment


    • #32
      Originally posted by xtachx View Post
      Sounds like the open source driver made a lot of progress and I can finally ditch catalyst!
      Remeber that you need to use the latest mesa 9.3 devel, x driver xf86-ati, and libdrm besides kernel 3.11 for best performance. Look out for bugs too, because there can be various issues like random freezing and such.
      The driver works very well woth Opengl apps, on my A8-5500 TF2 is smoother and loads faster than with catalyst.

      Comment


      • #33
        @droste

        elseif is not a valid preprocessor keyword, it's "elif", just like in cpp. So it's the shader that is buggy.

        From GLSL 1.50 spec:
        The complete list of preprocessor directives is as follows.
        #
        #define
        #undef
        #if
        #ifdef
        #ifndef
        #else
        #elif
        #endif
        #error
        #pragma
        #extension
        #version
        #line

        Comment


        • #34
          Sooo bottom line is: FlightGear with project rembrandt is quite broken

          Probably the only one where the driver is to blame is the first error, which results from a too complex shader.

          Comment


          • #35
            Originally posted by xtachx View Post
            Yes exactly, I have 2 monitors. I do not want to deactivate any of them because i use them a lot (usually while coding, i have my references on one and code in another). The fan sounds like a jet engine.
            Can you check the debugfs output to see if this is actually a duplicate of that bug? It could just be that the OEM that built your board used a crappy fan or didn't use a fan controller.

            Comment


            • #36
              Originally posted by agd5f View Post
              Can you check the debugfs output to see if this is actually a duplicate of that bug? It could just be that the OEM that built your board used a crappy fan or didn't use a fan controller.
              Sure. I will be home tomorrow, I can post a debugfs output. However, I have to mention that catalyst works fine - and the fan is quiet with it. IF the hardware was faulty, then it wouldnt work with catalyst as well.

              Comment


              • #37
                Originally posted by droste View Post
                Sooo bottom line is: FlightGear with project rembrandt is quite broken

                Probably the only one where the driver is to blame is the first error, which results from a too complex shader.
                Its still an open source project. Instead of blaming rembrandt, we can submit a fix! I will look at the source when I am home tomorrow.

                Comment


                • #38
                  Originally posted by droste View Post
                  Sooo bottom line is: FlightGear with project rembrandt is quite broken

                  Probably the only one where the driver is to blame is the first error, which results from a too complex shader.
                  I remember urban shader used to work in FG 2.6 and mesa when set to level 3. Now it's just broken. Maybe it's time to report a bug?
                  BTW can you check what's the performance compared to catalyst? My current rig (Samsung R522 - Radeon 4330 HD mobile) runs at about 50% of fps compared to Win7 (same fg settings).

                  Comment


                  • #39
                    Originally posted by Xeno View Post
                    I remember urban shader used to work in FG 2.6 and mesa when set to level 3. Now it's just broken. Maybe it's time to report a bug?
                    BTW can you check what's the performance compared to catalyst? My current rig (Samsung R522 - Radeon 4330 HD mobile) runs at about 50% of fps compared to Win7 (same fg settings).
                    I can compare the linux catalyst vs radeon. Cannot compare to windows. (Sorry, I havent used windows in the past 16 years!).


                    Now I have to figure out how to run kmod-catalyst and radeon side by side with different kernels. If anyone has any experiance on it, any tips?

                    Comment


                    • #40
                      Unfortunatelly it's not so straight-forward. Depends on config you might need two different initramfs images, add/remove radeon kernel module to blacklist and reboot to switch. Another question if fglrx libs won't interfere with radeon driver (it may try to load proprietary OpenGL libs instead of mesa ones). Maybe with custom lib paths magic and handcrafted xorg.conf files.

                      Usually if I wanted to try fglrx, I install driver/kmod, do initramfs/blacklist dance (bl radeon, backup current init img, create new one) reboot an test.
                      When i'm done, I uninstall catalyst and its kmod, remove radeon from blacklist then restore initarmfs image, reboot.

                      Also I always use driver rpms from RPMFusion (my distro is Fedora). Driver pack from AMD site overwrite xorg and mesa libs.

                      Comment


                      • #41
                        Originally posted by Xeno View Post
                        Unfortunatelly it's not so straight-forward.
                        As an ATI card owner, that word has a whole new meaning to it. Making an initramfs is the easy part. (you know what I am saying you have an ati card as well)

                        Depends on config you might need two different initramfs images, add/remove radeon kernel module to blacklist and reboot to switch. Another question if fglrx libs won't interfere with radeon driver (it may try to load proprietary OpenGL libs instead of mesa ones). Maybe with custom lib paths magic and handcrafted xorg.conf files.
                        I have a dual screen, and about 500 backups of my hand crafted and mastered xorg.conf (where is that xkcd comic again?). However I should be able to remove fglrx from kernel 3.11 initramfs with dracut, and have a custom entry in xorg.conf to not blacklist radeon. (and possibly /etc/modprobe.d/blaclist.conf). The question is, can I automate this based on which kernel the system boots...


                        Usually if I wanted to try fglrx, I install driver/kmod, do initramfs/blacklist dance (bl radeon, backup current init img, create new one) reboot an test.
                        When i'm done, I uninstall catalyst and its kmod, remove radeon from blacklist then restore initarmfs image, reboot.

                        Also I always use driver rpms from RPMFusion (my distro is Fedora). Driver pack from AMD site overwrite xorg and mesa libs.

                        I use fedora as well. I am afraid to remove kmod-catalyst - there is no guarantee that it will start again when I install it. I am scared of ATI drivers really. IT can make or break my weekend! I would rather try it on a separate kernel or an USB live image.

                        Comment


                        • #42
                          Originally posted by Xeno View Post
                          BTW can you check what's the performance compared to catalyst?
                          Sorry I don't have windows installed (well I have, but only as a VM) and I don't want to install fglrx on my perfectly fine running system

                          Comment


                          • #43
                            Originally posted by droste View Post
                            Do you have 2 monitors attached?

                            If that's the case see: https://bugzilla.kernel.org/show_bug.cgi?id=60523

                            My current workaround: Deactivate one screen when I don't use it.
                            Droste, I can confirm now after checking that it is the exact same bug. I am stuck in power level 2. (I have tested with a liveUSB, so things were slow). No matter what I do, the power level doesnt change at all!


                            Is there any more test you would like me to run?

                            Comment


                            • #44
                              You could add yourself to the CC list of the bug in case Alex Deucher (agd5f) has something new to test.
                              Right now there's nothing else you could do... Well other than trying to fix the bug by yourself

                              Comment


                              • #45
                                BTW, AMD guys, is there anything still missing from the radeon kernel module that would prevent fglrx from using it instead of its own kernel module? Except for the will of the Catalyst team to rewrite the user space part to use the DRM API to talk to kernel, of course.

                                Comment

                                Working...
                                X