Announcement

Collapse
No announcement yet.

In-Fighting Continues Over OpenACC In GCC

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

  • #31
    I'm guessing GPL licensed gcc allows NVIDIA to maintain a out-of-tree version while it's being only usable to their binary blob.
    No reason or wide benefit to pull it into main line before that.

    If I remember correctly, Kernel DRM has similar policy: a DRM module can only be accepted into main line if there's at least a open source user-space driver.

    Comment


    • #32
      From my understanding (which is rather sparse in this case) are the complaints against this in some way similar (not exactly) to AMD's Mantle?

      IIRC AMD said that it is possible for Mantle code to be run on other GPU's, IE those from nVidia, Intel or the ARM ones, if those vendors (or people coding drivers for such) wrote a driver that could interpret Mantle code and execute it on other hardware? The problem being that while the Mantle API itself is open, the design is for AMD GPU's, and works best for them.

      Similarly, the PTX ISA that is converted to nVidia hardware ISA by the nVidia drivers could be converted instead by Intel's or AMD's drivers into an ISA that is compatible with their hardware. Again the problem being that the hardware itself has different designs, and the PTX ISA would work best with nVidia hardware/drivers

      IIRC Most people were against Mantle because it was 'Open' mainly in name and technical definition of the word, but not in practical use. Is this not a similar situation to the Mantle one?

      IMO a separate patch/branch maintained alongside the main GCC would be the best solution, as previously suggested.

      Comment


      • #33
        Originally posted by 0xBADCODE View Post
        So now we can see: "Open"ACC is all about closed-source driver and proprietary CUDA, just rebranded, isn't it?
        What could you possibly be talking about? If you want to create some other backend for OpenACC you're more than welcome to. CUDA is not a requirement.

        This place can be a complete trainwreck sometimes.

        I get that some of you people have a read hard-on for Intel and AMD for supposedly being "open", but at least base your anti-NVIDIA froth on some legit complaints.

        Comment


        • #34
          Originally posted by zeealpal View Post
          From my understanding (which is rather sparse in this case) are the complaints against this in some way similar (not exactly) to AMD's Mantle?
          Until Mantle is getting support by the free drivers, or becomes an open standard, including a Mantle backend would put GCC into the same situation as it is now with PTX. I however do not think that adding Mantle to GCC is proposed by anybody at this time.

          Comment


          • #35
            Originally posted by johnc View Post
            What other vendor is selling CUDA-capable hardware?
            Technically any x86 vendor as well since Cuda can be ran on CPUs. Splitting hairs I know but it is the truth.

            Comment


            • #36
              Originally posted by zeealpal View Post
              From my understanding (which is rather sparse in this case) are the complaints against this in some way similar (not exactly) to AMD's Mantle?

              IIRC AMD said that it is possible for Mantle code to be run on other GPU's, IE those from nVidia, Intel or the ARM ones, if those vendors (or people coding drivers for such) wrote a driver that could interpret Mantle code and execute it on other hardware? The problem being that while the Mantle API itself is open, the design is for AMD GPU's, and works best for them.

              Similarly, the PTX ISA that is converted to nVidia hardware ISA by the nVidia drivers could be converted instead by Intel's or AMD's drivers into an ISA that is compatible with their hardware. Again the problem being that the hardware itself has different designs, and the PTX ISA would work best with nVidia hardware/drivers

              IIRC Most people were against Mantle because it was 'Open' mainly in name and technical definition of the word, but not in practical use. Is this not a similar situation to the Mantle one?

              IMO a separate patch/branch maintained alongside the main GCC would be the best solution, as previously suggested.
              This is exactly the problem PTX is just like mantel, it's a single vendor solution which that vendor is trying to sell as a vendor neutral solution.

              Comment

              Working...
              X