Results 1 to 4 of 4

Thread: Compiler Tuning With Intel Ivy Bridge Processors

  1. #1
    Join Date
    Jan 2007
    Posts
    14,914

    Default Compiler Tuning With Intel Ivy Bridge Processors

    Phoronix: Compiler Tuning With Intel Ivy Bridge Processors

    In this article is a look at the impact that compiler tuning has for the latest-generation Intel Ivy Bridge processors. Being tested is the GNU Compiler Collection (GCC) while passing various architecture flags to look at the impact they have on the results.

    http://www.phoronix.com/vr.php?view=17414

  2. #2
    Join Date
    Jun 2009
    Posts
    1,147

    Default

    LOL is me or bulldozer just trashed ivy bridge massively in C-Ray

    bulldozer 26.92s
    ivy bridge 80.37s

    OOUUUCHH

    or michael made some change to the C-Ray bench profile?? i mean i know bulldozer shines in heavy parallel task but 4 times ivy performance DAMN (<--- im happy i buyed my AMD FX )

  3. #3
    Join Date
    Feb 2008
    Location
    Linuxland
    Posts
    5,132

    Default

    Can anyone explain what the fsgsbase does? Google is not being helpful.

    The other two, half floats and hardware RNG, aren't quite expected to speed up random software. (unless it can go to half-floats with -ffast-math and process twice as much float calculations on the vector unit?)

  4. #4
    Join Date
    Nov 2011
    Posts
    24

    Default

    Quote Originally Posted by curaga View Post
    Can anyone explain what the fsgsbase does? Google is not being helpful.
    http://www.intel.com/idf/library/pdf...CS005_101F.pdf says

    """Fast access of FS & GS base registers
    Useful for user level thread storage by providing 4 new instructions
    for ring-3 access of FS & GS base registers
    """

    AFAICS, it's about reducing the overhead of using TLS variables; IIRC both Linux and Windows use FS/GS to store a per-thread pointer to the (per-thread) TLS descriptor block.

    The other two, half floats and hardware RNG, aren't quite expected to speed up random software. (unless it can go to half-floats with -ffast-math and process twice as much float calculations on the vector unit?)
    No, -ffast-math won't do anything as crazy as that. Note that it doesn't actually support arithmetic on half-floats, the new instructions are only for converting half-floats in memory to/from 32-bit single precision. AFAICS this may be useful for graphics stuff, IIRC some GPU's already support half-float textures.

Posting Permissions

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