Announcement

Collapse
No announcement yet.

MSAA Anti-Aliasing For AMD Radeon Evergreen GPUs

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

  • #11
    Originally posted by agd5f View Post
    It will take a little longer because it's a little more complex to support on 6xx/7xx GPUs. Evergreen+ supports uncompressed AA buffers, but earlier asics only support compressed buffers. Once support for CB compression is enabled, r6xx/7xx MSAA will work. Compression support also improves performance for MSAA.

    See the r600g todo page:
    http://dri.freedesktop.org/wiki/R600ToDo
    Thanks for the update and all of the hard work Alex.

    Comment


    • #12
      Originally posted by curaga View Post
      Would that help non-MSAA rendering at all?
      No. Not on r6xx/r7xx. On evergreen+, it can use used for fast color clears which save memory bandwidth for color buffer clears in some cases, similar to fast Z clears for depth/stencil.

      Comment


      • #13
        Originally posted by Veerappan View Post
        I seem to remember that this is the last thing that r600g needs to pass the WebGL conformance tests in Firefox. Intel already passes, and I'm not sure about Nouveau (I seem to think that it does).

        Also, according to RadeonFeature this is possibly the last feature needed for r600g to report (and actually support) OpenGL 3.0 on this hardware.
        When i run the kronos conformance test it results in the screen going blank for a few seconds and gpu lockup.

        Comment


        • #14
          Originally posted by Veerappan View Post
          I seem to remember that this is the last thing that r600g needs to pass the WebGL conformance tests in Firefox. Intel already passes, and I'm not sure about Nouveau (I seem to think that it does).

          Also, according to RadeonFeature this is possibly the last feature needed for r600g to report (and actually support) OpenGL 3.0 on this hardware.

          Also, many thanks to airlied and mareo (and everyone else contributing to Mesa) for this feature.
          Nouveau sure does not pass on my 460GTX

          Comment


          • #15
            Originally posted by agd5f View Post
            No. Not on r6xx/r7xx. On evergreen+, it can use used for fast color clears which save memory bandwidth for color buffer clears in some cases, similar to fast Z clears for depth/stencil.
            Ah, the joys of owning a bw-limited R700

            Comment


            • #16
              Originally posted by curaga View Post
              Note the above was not just for Michael. Anybody doing a feature that affects image quality, you must have comparison shots and benches already to submit it as working, so include links in your patch submission
              Marek already claimed that this series makes most of the MSAA piglit tests pass. Getting nearer to OGL conformance is way more important than any screenshot.

              Comment


              • #17
                Originally posted by Lynxeye View Post
                Marek already claimed that this series makes most of the MSAA piglit tests pass. Getting nearer to OGL conformance is way more important than any screenshot.
                Saying it passes technical tests ("does it sample the points correctly", "if I blit from a MSAA buffer to a regular buffer, does it crap out") is good, but not relevant to users. Even if it's technically correct, it could be slow and ugly.

                Comment


                • #18
                  Originally posted by curaga View Post
                  Saying it passes technical tests ("does it sample the points correctly", "if I blit from a MSAA buffer to a regular buffer, does it crap out") is good, but not relevant to users. Even if it's technically correct, it could be slow and ugly.
                  well to be honest is not easy to meet the technical reference implementation and high performance at the same time, so i would assume once marek get it to render correctly then the optimization process will begin[which is perfectly normal] especially with a hard beast to tame like MSAA.

                  so maybe in mesa 9.0 it will kill FPS but render properly so they can achieve OpenGL 3.1 by the book and in 9.1 performance will go up[and weird cornercases fixed]

                  for now 1000 kudos to marek and all the mesa team cuz the great work XD

                  btw LLVM shaders in r600g are shaping great, now i can play Lineage2 Tauti in maximum setting[except MSAA] with HDR totally fluid [wine 1.5.10 + mesa git].

                  btw Michael Opengl 3.2+ is not used by any big proyect yet[taking out the awesome Unigine engine] so for now with good GL 2.1/3.1 mesa can support 99% of the 3d stuff out there including wine since DX10/DX11 don't have enough marketshare yet to remove the DX9 codepath[and you can barely notice any difference in most engines out there], so ofc will be good to get up to date with khronos but would be interesting to discuss the importance of Optimal GL3.X/ vs catch up to khronos since most game makers out there will take as reference Mac OS X GL support for their linux ports[if apple only support GL 3.2(not sure fix me if wrong) is very unlikely that valve for example will use GL 4.3 only on linux] but if mesa can provide optimal speed for the same GL version in OS X then we can kill zombies the FOSS way

                  hint: maybe an more deep analysis about it can bring some nice discussions and maybe suggest some goals for mesa in the short term

                  Comment


                  • #19
                    Originally posted by Veerappan View Post
                    I seem to remember that this is the last thing that r600g needs to pass the WebGL conformance tests in Firefox. Intel already passes, and I'm not sure about Nouveau (I seem to think that it does).

                    Also, according to RadeonFeature this is possibly the last feature needed for r600g to report (and actually support) OpenGL 3.0 on this hardware.

                    Also, many thanks to airlied and mareo (and everyone else contributing to Mesa) for this feature.
                    There's only 1 test on r600g that's failing now, tracked here: https://bugs.freedesktop.org/show_bug.cgi?id=44912

                    Intel passes all their tests, and so do the NVidia binary drivers.

                    For the MSAA problem, i think Firefox updated their code to only use that on Mesa 8.1 or greater, where it's been fixed, so it works with Intel 8.0 by disabling that.

                    Comment

                    Working...
                    X