Results 1 to 8 of 8

Thread: POCL 0.9 Implements OpenCL 1.2 On The CPU

  1. #1
    Join Date
    Jan 2007
    Posts
    13,466

    Default POCL 0.9 Implements OpenCL 1.2 On The CPU

    Phoronix: POCL 0.9 Implements OpenCL 1.2 On The CPU

    Version 0.9 of the Portable Computing Language (POCL) is now available as an open-source implementation of OpenCL 1.2...

    http://www.phoronix.com/vr.php?view=MTU5MDQ

  2. #2

    Default

    Thats all fine and good, but whole point of OpenCL is to be able to pull in the GPU, DSP and anything else to accelerate it.

  3. #3
    Join Date
    Jan 2013
    Posts
    97

    Default

    Quote Originally Posted by Kivada View Post
    Thats all fine and good, but whole point of OpenCL is to be able to pull in the GPU, DSP and anything else to accelerate it.
    Having a CPU-based implementation which doesn't rely on any of that hardware means you can write code that uses OpenCL (and so can be accelerated by specialised hardware if available) without having to bother with a fallback path to do the same thing on older/lower-spec computers.

  4. #4

    Default

    Quote Originally Posted by FLHerne View Post
    Having a CPU-based implementation which doesn't rely on any of that hardware means you can write code that uses OpenCL (and so can be accelerated by specialised hardware if available) without having to bother with a fallback path to do the same thing on older/lower-spec computers.
    Outside of people on old Intel IGPs most boxes still in use today have a GPU that supports OpenCL since it was introduced with DirectX10/OpenGL3 class hardware back in 2006 on the dedicated GPU side and 2008 on the IGP side.

    In mobile space with ARM systems, just write off the older hardware users, they upgrade their phones on about a 2 year cycle as most plans offer this, as well many people break their phones well before that.

    So should a developer still be building for a non accelerated environment when the odds are that there is an OpenCL capable GPU?

    This is basically the 64 bit vs 32 bit argument again, like 64 bit, we shouldn't be dictated by how slow the Windows world is at adopting new tech.

  5. #5
    Join Date
    Jan 2013
    Posts
    97

    Default

    Quote Originally Posted by Kivada View Post
    Outside of people on old Intel IGPs most boxes still in use today have a GPU that supports OpenCL since it was introduced with DirectX10/OpenGL3 class hardware back in 2006 on the dedicated GPU side and 2008 on the IGP side.
    You're ruling out every Core2 IGP - there are a ton of those around here (fairly well-off part of the UK), both laptops and low-end desktops, which don't look like they're going away any time soon. There are a fair few Pentium M laptops about too, although those can't have that long left. If OpenCL support is over 50% now - possibly, but not by that much I suspect - it's still far below a level where you can just assume that anything you develop for will have it.

    Quote Originally Posted by Kivada View Post
    So should a developer still be building for a non accelerated environment when the odds are that there is an OpenCL capable GPU?.
    That's...the point. Having a CPU-based OpenCL library avoids the need to care about non-accelerated environments - you can just write your OpenCL code, have it accelerated where the hardware's available, and transparently run it on older machines without having to write a whole lot of non-OCL code to fall back onto.

  6. #6
    Join Date
    Mar 2011
    Posts
    18

    Default

    Quote Originally Posted by Kivada View Post
    Outside of people on old Intel IGPs most boxes still in use today have a GPU that supports OpenCL since it was introduced with DirectX10/OpenGL3 class hardware back in 2006 on the dedicated GPU side and 2008 on the IGP side.

    In mobile space with ARM systems, just write off the older hardware users, they upgrade their phones on about a 2 year cycle as most plans offer this, as well many people break their phones well before that.

    So should a developer still be building for a non accelerated environment when the odds are that there is an OpenCL capable GPU?

    This is basically the 64 bit vs 32 bit argument again, like 64 bit, we shouldn't be dictated by how slow the Windows world is at adopting new tech.
    And you're ruling out 1st and 2nd gen Core i-family processors...

  7. #7
    Join Date
    Oct 2008
    Posts
    2,913

    Default

    Quote Originally Posted by Kivada View Post
    Outside of people on old Intel IGPs most boxes still in use today have a GPU that supports OpenCL since it was introduced with DirectX10/OpenGL3 class hardware back in 2006 on the dedicated GPU side and 2008 on the IGP side.
    Have you seen how many people complain on here about how AMD dropped the r600/r700 support from their drivers? r700 doesn't have OpenCL support.

    Now imagine trying to use OpenCL on a common piece of software. Like LibreOffice/Calc. Having it run only on new computers with OpenCL support would be completely unacceptable. Which means they have to make any new OpenCL code completely optional, and use normal C/whatever code to do the same thing. This library means you can just have the OpenCL code only, and not have to worry about putting in backwards compatibility code for older computers.

  8. #8
    Join Date
    Sep 2010
    Posts
    421

    Default

    A big advantage of a CPU OpenCL implementation is the testing possibilities.
    An application developer is able to test the program written.
    Another is that the OpenCL implementation can be easier tested for implementation issues.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •