Announcement

Collapse
No announcement yet.

Progress On SPIR-V For The Nouveau Driver Is Moving Slow

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

  • Progress On SPIR-V For The Nouveau Driver Is Moving Slow

    Phoronix: Progress On SPIR-V For The Nouveau Driver Is Moving Slow

    Back in July I wrote about someone working on a SPIR-V to NV50 IR Nouveau translator so that this intermediate representation for Vulkan and OpenCL 2.1+ could then be fed into this open-source NVIDIA driver. A brief, indirect update was shared this weekend and so far it appears the work is progressing slowly...

    http://www.phoronix.com/scan.php?pag...o-NV50-IR-Slow

  • #2
    There will be closed-source nVidia Vulkan driver for NV50? Really?

    Comment


    • #3
      For the near future at least, NVIDIA card owners will likely need to rely upon NVIDIA's closed-source Vulkan driver for suitable support.
      Vulkan isn't really need Nouveau and for OpenCL... do Nvidia support OpenCL 2.0 at least?

      Comment


      • #4
        Originally posted by SXX⁣ View Post
        Vulkan isn't really need Nouveau and for OpenCL... do Nvidia support OpenCL 2.0 at least?
        No it's not necessary, but there could be some aadvantages for open source drivers if they do handle vulkan, assuming that vulkan becomes effective in the market

        Right now OSS driver devs have to focus on 2 things: 1. hardware drivers that control the hardware, and pass execution code to the hardware, and 2. software implementations of the GL spec that software likes to use to create graphics. The first is quite difficult, but the latter requires plenty of effort and expertise.
        Now cooperative efforts like galium and mesa try to share a lot of the effort for the second item, but the driver devs still end up working a lot to meet growing spec needs (and taking flak for being behind.)

        If vulkan and opencl become primary targets, then the driver devs can focus a lot more on the first part, and invest heavily in SPIR compiling efforts and optimizations, instead of spending a lot of time trying to chase the ever growing opengl spec, plenty of which is rarely used.

        Writing the vulkan/opencl handlers will have it's own overhead, but assuming that the SPIRV decision isn't a mistake, it should lead to a situation where driver devs can focus on hardware, and optimizing their spirv consumption for the real world.

        It might what driver devs would really rather be doing, but I'm not one of them, so I can't say.

        Ilia's comment about core differences in SPIRVand N50 lead me to wonderif SPIRV will actually make Vulkan and OpenCL hard to use.

        Comment


        • #5
          Originally posted by jaxxed View Post
          Ilia's comment about core differences in SPIRVand N50 lead me to wonderif SPIRV will actually make Vulkan and OpenCL hard to use.
          My comment is more about converting from SPIR-V to nv50 ir (which, btw, is used as the basis of the codegen compiler, used by the nv50 and nvc0 drivers). Just a bit of annoyance that anyone familiar with compilers is used to. And you're not going to be writing SPIR-V by hand -- a compiler will be doing it for you.

          Comment

          Working...
          X