OpenCL fuzzing

Discussion in 'GPGPU Technology & Programming' started by RecessionCone, Apr 16, 2015.

  1. RecessionCone

    Regular Subscriber

    Joined:
    Feb 27, 2010
    Messages:
    499
    Likes Received:
    177
  2. Dade

    Newcomer

    Joined:
    Dec 20, 2009
    Messages:
    206
    Likes Received:
    20
    From the paper: "Many OpenCL implementations are not robust Many of the configurations we tested exhibit basic compiler bugs (see Figure 1), miscompilations caused machine crashes for some configurations, and we found that some configurations are incapable of running standard OpenCL benchmarks. These issues undermine the portability aim of the OpenCL effort."

    Vendors have unable to provide even vaguely bug free compilers for years. The Khronos solution of providing generic to SPIR-V compilers for OpenCL v2.1 looks like the only way to solve this problem to me.
     
  3. Jawed

    Legend

    Joined:
    Oct 2, 2004
    Messages:
    10,873
    Likes Received:
    767
    Location:
    London
    Ah, that'll be why my OpenCL works: I avoid structs.

    I don't think SPIR-V will be a step improvement. The report seems to strongly imply that optimising vectors (or structs with vectors inside :mrgreen:) and optimisation while respecting memory ordering are alien concepts to extant compiler technology. So the people building these compilers are trying to shoehorn those concepts into existing compilation paradigms.

    It never ceases to amaze me how painfully behind compiler technology is. It's like compilers are stuck in the 1950s.
     
  4. AlexV

    AlexV Heteroscedasticitate
    Moderator Veteran

    Joined:
    Mar 15, 2005
    Messages:
    2,528
    Likes Received:
    107
    Or maybe there is a relatively wide world outside of our small secluded corner, and (quality) compiler writers have different concerns, associated with products with far wider TAM. I find your closing statement to be somewhat too narrow in its focus, and unfair. Surprising as it might seem, (quality) compiler writers actually have a clue.
     
  5. entity279

    Veteran Regular Subscriber

    Joined:
    May 12, 2008
    Messages:
    1,229
    Likes Received:
    422
    Location:
    Romania
    Jawed probably just expressed its frustration with current compiler/language tech.

    I certainly can echo that, there's many things about which we (we = the state of the art) kinda know how they should be really done, yet we're seeing none of that. ( there are reasons for this and they are mostly political/financial and not directly related to the technical expertise of the writers themselves, most likely )

    My most meaningful experience is regarding JVM languages so not directly relevant, just saying.
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...