Announcement

Collapse
No announcement yet.

Understanding More Of AMD's Open-Source HSA Plans

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

  • Understanding More Of AMD's Open-Source HSA Plans

    Phoronix: Understanding More Of AMD's Open-Source HSA Plans

    On Monday I wrote about AMD adding native object code support to their Radeon Gallium3D drivers and Clover. Besides being a huge performance win for OpenCL kernel compile times, this work is also instrumental as part of AMD's open-source HSA Linux plans...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Originally posted by phoronix View Post
    Phoronix: Understanding More Of AMD's Open-Source HSA Plans

    On Monday I wrote about AMD adding native object code support to their Radeon Gallium3D drivers and Clover. Besides being a huge performance win for OpenCL kernel compile times, this work is also instrumental as part of AMD's open-source HSA Linux plans...

    http://www.phoronix.com/vr.php?view=MTgwNzY


    D3D is a language API.

    Comment


    • #3
      I've recently been putting together a Kaveri system, and noticed something. Since it's by definition UMA, 1G of my 8G of RAM has been reserved for the graphics adapter. Presumably HSA will make this memory-division nonsense unnecessary, and memory can all be one big happy blob. Graphics space could be managed to-need using classical memory allocation methods, and returned when unused. That of course would require support in the video drivers. Is this on anyone's radar screen?

      Comment


      • #4
        Originally posted by artivision View Post
        D3D is a language API.
        Where is Direct3D mentioned in the article?
        Or did you mean G3D?
        And what do you mean by 'language API'?

        Comment


        • #5
          Originally posted by CrystalGamma View Post
          Where is Direct3D mentioned in the article?
          Or did you mean G3D?
          And what do you mean by 'language API'?

          I mean if it will be easier for us to port a state tracker (shading language compiler rules). D3D API = Shading programing language.
          Last edited by artivision; 08 October 2014, 10:03 AM.

          Comment


          • #6
            Originally posted by artivision View Post
            I mean if it will be easier for us to port a state tracker (shading language compiler rules). D3D API = Shading programing language.
            Strictly speaking D3D API = D3D API, HLSL = Shading programming language.
            Test signature

            Comment


            • #7
              Originally posted by phred14 View Post
              I've recently been putting together a Kaveri system, and noticed something. Since it's by definition UMA, 1G of my 8G of RAM has been reserved for the graphics adapter. Presumably HSA will make this memory-division nonsense unnecessary, and memory can all be one big happy blob. Graphics space could be managed to-need using classical memory allocation methods, and returned when unused. That of course would require support in the video drivers. Is this on anyone's radar screen?
              AFAIK, it would still be good to reserve memory for the GPU, so that there is less allocator overhead. I recall someone explained this a while ago, but I don't remember who nor where.

              Comment


              • #8
                Is HSA a AMD thing only?
                Or can HSA be used by Intel and Nvidia too?
                Can HSA be used by Qualcomm too?

                Can HSA be used on POWER, SPARC, MIPS and ARM too?

                Comment


                • #9


                  ARM and Qualcomm are founding members along with TI, Samsung, ImgTek and Mediatek.
                  Last edited by bridgman; 08 October 2014, 02:05 PM.
                  Test signature

                  Comment


                  • #10
                    Originally posted by bridgman View Post
                    Strictly speaking D3D API = D3D API, HLSL = Shading programming language.

                    Strictly speaking each programing language has its own programing modes and examples. C type languages for example have multidimensional tables, tale, if/else -type, as long is valid -type and goes on. 3D languages have some less and some more (textures for example). A language is not only the typos, so D3D-API + HLSL = D3D graphics language. The difference of low level APIs and high level like D3D and OGL is that with low level, graphics are distributed in low level shader model 3-4-5 assembly (no Representation, API, SL compiler, API calls, needed), only HW compiling needed. With high level APIs, AMD, MS, NV and Intel have created a paradox, the first programing language in the world that everything is distributed in shader language bytecode, unmodeled, lowered source code. Even OCL and Java don't do that. Why? AMD's share holders are criminals as the rest above, that enforce legally (with paid laws) and illegally a graphics monopoly. If you had a company Mr Bridgman, you would have code of another corporation's standard inside your driver or you would have only your own?
                    Last edited by artivision; 08 October 2014, 02:39 PM.

                    Comment

                    Working...
                    X