Announcement

Collapse
No announcement yet.

Improving OpenCL On CPUs, Building Linux

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

  • #16
    Originally posted by curaga View Post
    Do the slides show Intel's implementation beating AMD's, a completely opposite result to Michael's?

    Units aren't exactly clear nor is the hw, but it appears that way.
    reading the PDF it does seem that way but then
    Michael only needs a headline to grab the ad revenue
    http://llvm.org/devmtg/2012-04-12/Sl...Karrenberg.pdf

    see: Evaluation II: WFVOpenCL vs. Intel/AMD (milliseconds)

    the meat of it being... they say that their WFVOpenCL tested code algorithm's beats the tested Intel OpenCL SDK v1.1 / AMD APP SDK v2.5 by an Average of : 2.5x (Intel), 40x (AMD)
    lower being better OC.

    and then there's the Conclusion

    "OpenCL benefits from both multi-threading and WFV on CPUs"

    WFV being "whole function vectorisation" or if you prefer whole function SIMD Optimization ,OC they could have known that if they had just asked the x264 devs and looked at it's assembly and C code functions to know that Intel SIMD beats AMD clock for clock for a very long time now

    in fact the WFVOpenCL guys would probably get a lot more speed in their CPU Optimizations if they just took the x264 code examples and modified them and the general framework to their needs and pay special attention to the supplied
    checkasm as The tool can be used to perform function-level benchmarks
    Last edited by popper; 06-18-2012, 03:26 PM.

    Comment

    Working...
    X