Announcement

Collapse
No announcement yet.

Radeon's ROCm OpenCL Runtime Finally Open-Sourced

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

  • wizard69
    replied
    This is really going off the deep end here.

    Originally posted by dungeon View Post

    It is about respect to some group of people in this case Free/Libre, because who does not give respect to others should not expect any kind respect in return also
    This has nothing to do with respect!!! I really don't understand people that claim disrespect when they don't get their way. There simply isn't enough info here to claim to know AMD's motivation for any blob that might be there. Considering their efforts so far with open source, standards and the like, i really don't believe that AMD is willfully thumbing their noses at the open source community.

    Stating something as friendly to Free/Libre, but everybody knows their stance on blob matters is already is total disrespect
    That is total garbage. What do you want next, the source code to the hardware? There are real limits to what a company like AMD can do and continue to survive as a company. Lets face it some really bad moves over the last few years have put them in bad shape. Simply put they have to keep certain aspects of their GPUs secret to survive. This has nothing to do with respect of a community.

    The Libre movement should be happy that AMD is doing as much as they are. More is better of course but who can we go to that is dramatically better?

    Leave a comment:


  • wizard69
    replied
    Originally posted by Serafean View Post
    1) Driver is useless without firmware, so it effectively is.
    That mahes about as much sense as saying that a CPUs instruction set is part of a driver.
    2) yes. At that point it becomes hardware.
    Personally id rather see blob free myself, however your arguements make no sense to me.

    Leave a comment:


  • Marc.2377
    replied
    There is enough (imo) C++ support in recent OpenCL versions to be productive: https://en.wikipedia.org/wiki/OpenCL#OpenCL_2.2


    Originally posted by Adarion View Post
    Nice, really nice. Even though my few RadeonSI chips won't benefit from it but this is very welcome. I hope this will gain more momentum. It takes a while sometimes, but in the end it's being delivered. I wonder what's next.
    I too have a SI card (the R9 280x) but I'm optimistic that ROCm can be made to work with it, given enough time.


    Originally posted by [email protected] View Post
    Thank you AMD. I guess I will go from Evergreen to Polaris or Vega pretty soon

    Though I hope the community is still working on clover, since it is gallium-based
    Last I checked it was abandoned...

    Leave a comment:


  • bridgman
    replied
    Not sure I understand. HCC takes C++ and generates GPU code from the parallelized parts, CPU code from the serial parts.

    IIRC it can also run in a dual-source mode (serial & parallel parts not mixed) where kernel code is converted directly to GPU code.

    Leave a comment:


  • artivision
    replied
    Originally posted by bridgman View Post

    For C++ you want to be looking at HCC rather than OpenCL, although recent versions of OpenCL have been picking up more C++ features.

    http://gpuopen.com/compute-product/h...pute-compiler/
    I mean GPU only. Not as a CPU amplifier.

    Leave a comment:


  • bridgman
    replied
    Originally posted by artivision View Post
    As far as i understand it OCL is C and CUDA is C++. I don't think that is a performance problem, but that AMD doesn't have complete hardware C++. There is no room today on HPC with OCL.
    For C++ you want to be looking at HCC rather than OpenCL, although recent versions of OpenCL have been picking up more C++ features.

    http://gpuopen.com/compute-product/h...pute-compiler/

    Leave a comment:


  • artivision
    replied
    As far as i understand it OCL is C and CUDA is C++. I don't think that is a performance problem, but that AMD doesn't have complete hardware C++. There is no room today on HPC with OCL.

    Leave a comment:


  • tiwake
    replied
    Leave harping on closed firmware blobs for another day. Today is the day for celebrating Open Source videocard drivers. May linux nerds and basement dwellers dance around computers and get drunk tonight.

    I know I will.

    Leave a comment:


  • lucrus
    replied
    Originally posted by dungeon View Post
    I disagree, because that way we can claim also how some blob software is effectively more open then opensource in some cases
    Yes, that's theoretically possible to pretend, but common sense suggests the cases when it's a reasonable statement to make or a off the wall one.

    Originally posted by dungeon View Post
    To me these firmwares are indeed software and it is shipped as blob software, so that is where all story ended up
    Yes, they are software and they are proprietary software. In the AMD case I feel like they don't make the free/libre driver any less free/libre.

    Originally posted by dungeon View Post
    And it beggins at moment when someone conviced you and now you think something is not what obviosly is
    Thanks for considering me that stupid, but, your english grammar aside, have you ever considered your opinion as such, e.g. an opinion like many others? I understand your opinion would be the most perfect way of doing things in a perfect world, but hey, look out of your window for a while, it's not perfect. And in a many-shadows-of-friendlyness not-so-perfect world we live in, AMD is making good things and deserves to be called free/libre friendly. And that's just my opinion, whether you laugh for it or not.

    Originally posted by dungeon View Post
    So lets check it, is it software? Yes. Is it shipped as blob? Yes. ... So what else is needed for clarification?
    I don't know what YOU need for clarification of my point. Maybe if you ask what is not clear to you, I can explain better.

    Let's make a simple math. Let's name the libre friendlyness level of a driver "the F factor", a number between 0 and 1. Then let's multiply F by the average FPS of the same driver as reported by some PTS benchmark, and let's call P the result. Then we go and buy a graphics card supported by the driver with maximum value of P.

    Using your perfect metrics, F is zero for all and any driver out there as of today. You end up waiting for the perfect driver and firmware (F=1) to magically come up, and if everyone were to follow your example, the world would be perfect. However things are a bit different.

    Using my approximate metrics, F is 0 for nvidia and fglrx, and it is 0.8 for nouveau, intel (whatever is called the driver), radeon and amdgpu. The FPS value makes me choose a Polaris card and tell others to do the same. It's that easy.
    Last edited by lucrus; 05-13-2017, 10:21 AM.

    Leave a comment:


  • dungeon
    replied
    Originally posted by lucrus View Post
    it does not mean that, at least to me. There are different levels of friendlyness, and firmware can be considered effectively hardware in some cases.
    I disagree, because that way we can claim also how some blob software is effectively more open then opensource in some cases

    To me these firmwares are indeed software and it is shipped as blob software, so that is where all story ended up

    And it beggins at moment when someone conviced you and now you think something is not what obviosly is

    So lets check it, is it software? Yes. Is it shipped as blob only? Yes. ... So what else is needed for clarification?
    Last edited by dungeon; 05-13-2017, 09:38 AM.

    Leave a comment:

Working...
X