Announcement

Collapse
No announcement yet.

r6xx 3D games

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

  • #46
    OpenArena runs fine with indirect rendering, but it tends to crash the X server in certain occasions.

    All the issues with direct rendering and indirect crashing are still present, tested them with latest drivers yesterday.

    Comment


    • #47
      There seem to have been some regressions recently.

      Neverball and other games flicker again in direct mode. Less than before, but it was gone completely until recently.

      I guess the devs are still hunting the bug around.

      Comment


      • #48
        git changes and reversion allow dri to work again but, was playing eduke32 for a very short while and suddenly hard drive starts running constantly and PC doesn't respond. Finally after holding down ctrl-alt-bkspc, X kicks out. dmesg gives various memory related messages.

        Is this what is meant by a race condition? Is there any feedback I can post to help?

        Edit: this was with radeonhd. I'll try with radeon shortly.

        Edit2: happens with radeon also.
        Last edited by forum1793; 08-25-2009, 08:24 PM.

        Comment


        • #49
          Quote from the original commit:
          "There is a possbile race that _glapi_Context is reset by another thread after it is tested in GET_CURRENT_CONTEXT but before it is returned."
          That's what it means, hard to speculate what it might cause though.

          Comment


          • #50
            Originally posted by forum1793 View Post
            git changes and reversion allow dri to work again but, was playing eduke32 for a very short while and suddenly hard drive starts running constantly and PC doesn't respond. Finally after holding down ctrl-alt-bkspc, X kicks out. dmesg gives various memory related messages.

            Is this what is meant by a race condition? Is there any feedback I can post to help?
            No, that is a leak. I saw it too when running neverball (since I run it windowed I saw my mem usage applet raising to fill my 4GB). And IIRC, I saw it before the race condition patch was reverted, so I don't think it's related. I haven't tested with the recent updates, will test now.

            (edit) The leak is still there. Neverball was using 2.3GB after 2 minutes of play (and increasing).

            (edit2) Ok, I've found the culprit. It's commit 8dd151b947c36100f38cf83eca674bd427b23e47 (r600 code cleanup). Something there is allocating memory and not freeing it. Until that is fixed, the last good version is dea10381864269bd6251a8b0af7a88de88296b32.
            Last edited by Fran; 08-26-2009, 04:41 AM.

            Comment


            • #51
              Ah!

              I had OpenArena slow down to the crawl and hang on me too. I guess I'm experiencing the same issue as everyone else then?

              Comment


              • #52
                Originally posted by forum1793 View Post
                Is this what is meant by a race condition?
                No; as Fran points out, you've described a memory leak. A race condition is (basically) when an error can be caused by related events happening in an unexpected order. It's called a race condition because the outcome depends on which process/signal/event "wins the race". The solution is usually to either explicitly force synchronization or to tweak the system such that the appropriate process is guaranteed to "win" (the latter tends to be used in hardware, where delays/sequencing can be built physically).

                Comment


                • #53
                  Nice read guys! Cant wait until September 10th or is it 14th? I have monies stashed away for a little upgrade.

                  Comment


                  • #54
                    Cooper has pushed a patch which should help with the corruption which previously forced you to run "indirect" (eg openarena). Haven't tried it myself yet though.

                    Comment


                    • #55
                      Originally posted by bridgman View Post
                      Cooper has pushed a patch which should help with the corruption which previously forced you to run "indirect" (eg openarena). Haven't tried it myself yet though.
                      Doesn't help me. See http://bugs.freedesktop.org/show_bug.cgi?id=23469

                      Comment


                      • #56
                        Yeah..

                        Originally posted by bridgman View Post
                        Cooper has pushed a patch which should help with the corruption which previously forced you to run "indirect" (eg openarena). Haven't tried it myself yet though.
                        I can't believe things are moving so fast Is it Google earth working already?
                        Bridgman, I always wanted to ask you something...as you a working in AMD, can your team exchange if not code, but ideas with fglrx team.

                        Comment


                        • #57
                          We discuss ideas with the proprietary driver team (not just fglrx) and pick up code snippets for programming specific hardware blocks. What we don't do is pick up million-line code blocks and try to fit them into a 20,000 line driver

                          Comment


                          • #58
                            Why is fglrx SO huge?

                            Because it covers all the chipsets? Because it implements its own memory manager and half of X Server? Because it implements all of Mesa basically?

                            I mean, the new open source 3D drivers are looking mean and lean, other than the spread among Mesa/DDX/KMS/GEM etc, I don't see where the discrepancy comes from.

                            Comment


                            • #59
                              Probably because it is full of

                              switch (app) {
                              case IDTECH3:
                              ...
                              case IDTECH4:
                              ...
                              case MAYA3D:
                              ...
                              case GLXGEARS:
                              //uh!
                              case BIGKNOWN3DBENCH:
                              //100000 line of code
                              case ...
                              }

                              Comment


                              • #60
                                Originally posted by pingufunkybeat View Post
                                Why is fglrx SO huge?

                                Because it covers all the chipsets? Because it implements its own memory manager and half of X Server? Because it implements all of Mesa basically?

                                I mean, the new open source 3D drivers are looking mean and lean, other than the spread among Mesa/DDX/KMS/GEM etc, I don't see where the discrepancy comes from.
                                take a peak at nvidia's driver then you'll see that fglrx (the kernel module) isn't too big at all

                                I guess it's that big because of what Mickabouille worte - optimization for all different kinds of apps

                                @Drago:

                                last time I tried (some days ago) there still was massive corruption with it

                                Comment

                                Working...
                                X