Announcement

Collapse
No announcement yet.

NVIDIA's NVPTX Support For GCC Is Close To Being Merged

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

  • NVIDIA's NVPTX Support For GCC Is Close To Being Merged

    Phoronix: NVIDIA's NVPTX Support For GCC Is Close To Being Merged

    The NVPTX back-end code for GCC that's going to allow OpenACC 2.0 offloading support for NVIDIA GPUs with GCC is close to materializing within the mainline code-base...

    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
    I don't see why they should merge NVPTX support into GCC.

    NVPTX and CUDA is some Nvidia-only proprietary technology.
    GCC is free software.

    Anything merged in should be usable by everyone.
    Else its just extra bloat and extra maintenance for everyone.

    Comment


    • #3
      Originally posted by uid313 View Post
      NVPTX and CUDA is some Nvidia-only proprietary technology.
      GCC is free software.
      But aren't all GCC targets some processor manufacturer's proprietary technology?

      Comment


      • #4
        Yeah they are. May as well object to x86 in GCC because it x86 is a virtual target that is compiled by the CPU into microcode.

        Comment


        • #5
          Originally posted by Zan Lynx View Post
          Yeah they are. May as well object to x86 in GCC because it x86 is a virtual target that is compiled by the CPU into microcode.
          With the minor difference that one doesn't need an additional driver to run the compiled x86 code. With this NVPTX stuff you have to rely on nvidia to support your platform/OS to be able to use it. And if they don't, well bad luck sucker....

          Comment


          • #6
            Originally posted by log0 View Post
            With the minor difference that one doesn't need an additional driver to run the compiled x86 code. With this NVPTX stuff you have to rely on nvidia to support your platform/OS to be able to use it. And if they don't, well bad luck sucker....
            It doesn't have anything to do with "luck". Something tells me anybody who is going to seriously invest in CUDA is going to know what hardware and platforms are necessary to make the thing go. But this is just how the business world works.

            Comment


            • #7
              Originally posted by johnc View Post
              It doesn't have anything to do with "luck". Something tells me anybody who is going to seriously invest in CUDA is going to know what hardware and platforms are necessary to make the thing go. But this is just how the business world works.
              The zealots here seem to think that only 100% open fpga designs from kickstarter projects should only be supported even though they might only have a handful of devices available, ever. I know few guys from academia who do GPU programming and basically CUDA is the only thing that works. OpenCL introduces that same dynamic parallelism, but the drivers suck in general. Time is money, they say. OpenCL drivers still don't work and CUDA has been a commercial success for a decade soon.

              Comment


              • #8
                When using nVIdia's NVPTX plugin for GCC on a Haswell-E CPU and GTX 980 GPU setup, does it take 10 seconds to compile Linus' kernel rather than just 45 seconds without NVPTX?

                Comment


                • #9
                  Originally posted by drone4four View Post
                  When using nVIdia's NVPTX plugin for GCC on a Haswell-E CPU and GTX 980 GPU setup, does it take 10 seconds to compile Linus' kernel rather than just 45 seconds without NVPTX?
                  GCC doesn't use offloading to speed up compilation, but it allows to write programs with offloading to GPU/MIC/etc.

                  Comment

                  Working...
                  X