Results 1 to 10 of 11

Thread: LLVM Clang vs. GCC Compilers For AMD's Steamroller

Hybrid View

  1. #1
    Join Date
    Jan 2007
    Posts
    15,138

    Default LLVM Clang vs. GCC Compilers For AMD's Steamroller

    Phoronix: LLVM Clang vs. GCC Compilers For AMD's Steamroller

    Besides the interesting but disappointing AMD Kaveri Gallium3D vs. Catalyst Linux driver benchmarks published this morning, here's some more AMD A10-7850K "Kaveri" benchmarks for your Sunday viewing pleasure...

    http://www.phoronix.com/vr.php?view=MTU3NjE

  2. #2
    Join Date
    Jul 2013
    Posts
    398

    Default

    What's the point of even doing tests that require OpenMP if you know for a fact that Clang doesn't have it?

    On another note, Clang 3.5 seems to have more than one regression, which sucks after the awesome improvements 3.4 brought.

  3. #3
    Join Date
    Sep 2013
    Posts
    14

    Default

    To demonstrate what impact that has perhaps?

  4. #4
    Join Date
    Oct 2012
    Location
    Washington State
    Posts
    488

    Default

    Quote Originally Posted by Daktyl198 View Post
    What's the point of even doing tests that require OpenMP if you know for a fact that Clang doesn't have it?

    On another note, Clang 3.5 seems to have more than one regression, which sucks after the awesome improvements 3.4 brought.
    He must have a soft spot for GCC.

  5. #5
    Join Date
    Oct 2007
    Posts
    1,286

    Default

    Quote Originally Posted by Daktyl198
    What's the point of even doing tests that require OpenMP if you know for a fact that Clang doesn't have it?
    It's important to note the fact when doing a comparison, and that test is part of the compiler suite: http://openbenchmarking.org/suite/pts/compiler

    Quote Originally Posted by Marc Driftmeyer View Post
    He must have a soft spot for GCC.
    Quit trolling.

  6. #6
    Join Date
    Nov 2009
    Location
    Madrid, Spain
    Posts
    398

    Default

    Quote Originally Posted by Marc Driftmeyer View Post
    He must have a soft spot for GCC.
    OpenMP is old technology (the 1.0 revision though), and even Visual Studio 2005 had it. It is in LLVM/Clang's power to do it as it can improve performance of some software packages, if not, it is normal to get some flack for this, doesn't it seem fair for you?

    For comparison when Google's Chrome had the fast JS VM, all people praised them and anyone having an interpreter was bashed in media and everyone have now a full JIT. But similarly when IE had hardware acceleration and Google Chrome didn't, everyone was bashing Google Chrome. Is it fair in a competitive world (as the compilers do compete, as being at least 4 major compilers fighting for improvements: VC++, GCC, Clang, Intel, ...) and missing a major feature for multicore (when today no CPU you can buy, including phones, is single core).

  7. #7
    Join Date
    Aug 2012
    Location
    Pennsylvania, United States
    Posts
    1,911

    Default

    Quote Originally Posted by ciplogic View Post
    OpenMP is old technology (the 1.0 revision though), and even Visual Studio 2005 had it. It is in LLVM/Clang's power to do it as it can improve performance of some software packages, if not, it is normal to get some flack for this, doesn't it seem fair for you?

    For comparison when Google's Chrome had the fast JS VM, all people praised them and anyone having an interpreter was bashed in media and everyone have now a full JIT. But similarly when IE had hardware acceleration and Google Chrome didn't, everyone was bashing Google Chrome. Is it fair in a competitive world (as the compilers do compete, as being at least 4 major compilers fighting for improvements: VC++, GCC, Clang, Intel, ...) and missing a major feature for multicore (when today no CPU you can buy, including phones, is single core).
    To be fair, LLVM chose not to implement OpenMP because a new X.0 spec was coming out, so they just decided to start there and move forward. Which they did. Patches got posted months ago under RFC. So its not that they COULDNT do it, they just decided to wait for the new version and start there instead of trying to do 2 versions at the same time

  8. #8
    Join Date
    Jan 2013
    Posts
    30

    Default

    I think that it is more like some base LLVM main architecture flaw that prevent to catch GCC in code quality. I remember 4+ years ago articles that state LLVM was so good, and we have LLVM based radeon shader that lag behind Catalyst, OpenMP still not implemented, code quality lag behind GCC with exception of few anecdotal special selected cases, etc.

    About comparison, as usual "not apple/ubuntu biased" atricle, for example GCC -O2 have no relation with clang -O2, everybody know that, as very different passes (from hundreds!) choices, like GCC -O2 more like clang -O3 and you compare compiler speed 30 passes vs 200+ passes. Sure it is intentional to continue to post again and again that compiling speed "benchmark".

Posting Permissions

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