Announcement

Collapse
No announcement yet.

RadeonSI Primitive Culling Lands In Mesa 19.2

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

  • #21
    Originally posted by ihatemichael
    Mesa 19.0.4, I also tested git (as of today) and the issues are still there.

    Vega 11 (Raven Ridge).
    Vega 11 that's an desktop RR like2400G .
    I use an 2500u with Vega8 to write this but i'm not using i3wm, is it the same with other VM/DE?
    With my Vega8 i don't have that problem unless i need to do something to trigger it.
    How frequent does the artifacts or missing objects appear?
    I had things like that earlier on older hardware with glamor but new hardware should work except 19.0.0 that was broken, have you reported the bugs?
    Have you tried some live distros from usb and if so was it also broken?
    Have it always been like that or is it an regression?
    I had some various other problems earlier with this laptop but the latest bios/uefi made things work as expected.

    Comment


    • #22
      Originally posted by ihatemichael

      Mesa 19.0.4, I also tested git (as of today) and the issues are still there.



      Vega 11 (Raven Ridge).
      mmm, as far as i understand Raven Ridge should work its best with kernel 5.0+, LLVM 9.0(don't remember the exact commits but any recent version should have the improved support) and mesa git(which need latest LLVM to work correctly).

      about I3wm i'm not sure since i don't use it but you can try maybe switch to wayland with sway, in that case you may save yourselves from the LLVM upgrade if it is too hard on your distro.

      If you are open to suggestions, i would recommend ArchLinux with laurent carlier mesa-git repo

      Comment


      • #23
        Originally posted by stalkerg View Post
        NO. 'Your' bug points to an additional commit.
        winsys/radeon: implement ctx_query_reset_status by copying radeonsi

        And it ended that it realy was #78e35df52aa2f7d770f929a0866a0faa89c261a9
        "radeonsi: update buffer descriptors in all contexts after buffer invalidation"

        FIXED by:
        radeonsi: fix a regression in si_rebind_buffer
        https://cgit.freedesktop.org/mesa/me...97d2f35f0e9eae
        Last edited by nuetzel; 05-21-2019, 07:55 PM. Reason: Added real wrong commit and fix.

        Comment


        • #24
          Originally posted by ihatemichael

          I already use Arch Linux and mesa git does not help with my issues, if it did, I wouldn't be complaining here.

          sway makes some of the issues go away (corruption) but some issues are still there, like invisible objects with some programs.
          Then send a detailed bug report for Vega 11 Raven Ridge.

          Comment


          • #25
            Originally posted by ihatemichael

            I already use Arch Linux and mesa git does not help with my issues, if it did, I wouldn't be complaining here.

            sway makes some of the issues go away (corruption) but some issues are still there, like invisible objects with some programs.
            mmm, i see, then we can conclude AMD is not at fault and their support GLAMOR is actually just fine but it seems they actually have severe issues still on Raven Ridge class hardware given other AMD hardware don't present this issue.

            now, for your specific problem, you should open a bug report for Raven Ridge if you haven't already also i can think about 2 more fixes to try

            1.) check your memory XMP profile(memory bandwith can push very weird behaviour on APU's) and upgrade to latest bios(i had issues with AGESA back on kabini age and a BIOS update did wonders)

            2.) Do you have enabled amdgpu.ppfeaturemask=0xxxxxxxxx on your boot process? (/etc/default/grub or /boot/loader/entries/xxxxx.conf) because at least on certain POLARIS10 cards like mine all hell break loose when enabled outside the amd-staging branch on a5gdf repo(and it varies wildly)

            also can you post an glxinfo -B

            Comment


            • #26
              ihatemichael Try setting AMD_DEBUG=nodpbb.

              Comment

              Working...
              X