Results 1 to 10 of 10

Thread: LLVM 3.3 To Introduce SLP Vectorizer

  1. #1
    Join Date
    Jan 2007
    Posts
    14,788

    Default LLVM 3.3 To Introduce SLP Vectorizer

    Phoronix: LLVM 3.3 To Introduce SLP Vectorizer

    One of the prominent features to be introduced with the LLVM 3.3 release this summer is the SLP Vectorizer. Introduced in the LLVM 3.2 release was the LLVM Loop Vectorizer for vectorizing loops while the new SLP Vectorizer is about optimizing straight-line code by merging multiple scalars into vectors.

    http://www.phoronix.com/vr.php?view=18694

  2. #2
    Join Date
    May 2012
    Posts
    811

    Default

    There's little difference probably because modern CPUs parallelize certain micro operations by default.
    Last edited by mark45; 05-07-2013 at 04:21 AM.

  3. #3
    Join Date
    Dec 2011
    Posts
    2,048

    Default Itanium

    Intel ought to add Itanium (IA-64) support to LLVM.
    With all this vectorizing it sounds like a good fit.
    Heard one of the problems with Itanium was bad compilers.

  4. #4
    Join Date
    Jul 2008
    Location
    Berlin, Germany
    Posts
    822

    Default

    Quote Originally Posted by uid313 View Post
    Intel ought to add Itanium (IA-64) support to LLVM.
    I don't think there is a lot of commercial interest left in Itanium. For example, X has been broken for several releases and nobody at Intel bothered to react.
    http://lists.x.org/archives/xorg-dev...er/034867.html

  5. #5
    Join Date
    Dec 2011
    Posts
    2,048

    Default

    Quote Originally Posted by chithanh View Post
    I don't think there is a lot of commercial interest left in Itanium. For example, X has been broken for several releases and nobody at Intel bothered to react.
    http://lists.x.org/archives/xorg-dev...er/034867.html
    Well X is pretty much for workstations and Itanium for servers, mainframes and HPC.
    I know Oracle and other companies lost commercial interest in Itanium, but have Intel too?

    Itanium got a shit reputation because but it was a promising architecture.

    It got bad reputation because:
    * The processors (and perhaps micro architectures) were bad, this does not mean the instruction set architecture is bad.
    * The compilers for IA-64 were bad and didn't output bytecode that were fast.
    * Itanium contained x86-compatibility and people whined that Itanium in x86-mode were slower than real x86 processors.

    The architecture seems pretty interesting though. EPIC, VLIW.

  6. #6
    Join Date
    Jul 2008
    Location
    Greece
    Posts
    3,791

    Default

    Itanium is dead. What does it matter if it's an interesting architecture? It's dead, period. So support for it must be pretty low on the priority list (if it's in the list at all.)

  7. #7
    Join Date
    Jan 2011
    Location
    Bathurst, NSW, Australia
    Posts
    47

    Default

    Itanium failed because it assumed that compile-time optimisations were sufficient for high performance. In reality you'll always need at least some degree of run-time re-ordering, etc.

    Back on topic, would this vectorizer benefit GPUs? Or are they no longer so vectorized?

  8. #8
    Join Date
    Sep 2009
    Posts
    114

    Default

    Quote Originally Posted by uid313 View Post
    Well X is pretty much for workstations and Itanium for servers, mainframes and HPC.
    I know Oracle and other companies lost commercial interest in Itanium, but have Intel too?

    Itanium got a shit reputation because but it was a promising architecture.

    It got bad reputation because:
    * The processors (and perhaps micro architectures) were bad, this does not mean the instruction set architecture is bad.
    * The compilers for IA-64 were bad and didn't output bytecode that were fast.
    * Itanium contained x86-compatibility and people whined that Itanium in x86-mode were slower than real x86 processors.

    The architecture seems pretty interesting though. EPIC, VLIW.
    What I think you're missing here is the reason why the compilers were "bad." EPIC and VLIW architectures demand so much more from the compiler. A good optimizing compiler is already hard to make for x86...now lets ramp up the difficulty much much further with EPIC and VLIW and see what kind of a compiler we can make. EPIC and VLIW, like so many computer science ideas, are great in theory but turn out no so good in practice.

  9. #9
    Join Date
    Oct 2012
    Location
    Washington State
    Posts
    458

    Default

    Quote Originally Posted by chithanh View Post
    I don't think there is a lot of commercial interest left in Itanium. For example, X has been broken for several releases and nobody at Intel bothered to react.
    http://lists.x.org/archives/xorg-dev...er/034867.html
    There is most certainly specific interest in IA-64 and it involves also making LLVM/Clang work out-of-the-box on Windows. There is plenty of discussion on the LLVM/Clang dev lists to express the need for this support.

    Intel should pony it up. They're getting far more out of using LLVM/Clang than what they'll ever contribute.

  10. #10
    Join Date
    Oct 2008
    Posts
    3,129

    Default

    Quote Originally Posted by Marc Driftmeyer View Post
    Intel should pony it up. They're getting far more out of using LLVM/Clang than what they'll ever contribute.
    Intel just fought a lawsuit to allow them to stop spending money on Itanium. They aren't going to turn around and start throwing money porting a new compiler after all that work to let them stop working on it.

Posting Permissions

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