Announcement

Collapse
No announcement yet.

AMD's opensource lies exposed

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

  • Originally posted by not.sure View Post
    yeah, although your link shows a double/triple cycle 16x16 MAC so slower and not very good for you

    in the case of the IRC log they are referring to DSP SIMD instructions, for instance ARM cortex has among others the Single-cycle 16x16 and 32x16 MAC implementations.

    "Compilers targeting the [ARM] architecture can use these DSP extensions to improve code-generation for standard C and C++ software, or allow software developers to explicitly request use of these extension via intrinsics or inline assembly code.

    Performance
    The ARM DSP extensions enable increased DSP performance without the need for very high clock frequencies. This performance is achieved with almost no increase in power consumption on a typical implementation."

    http://www.arm.com/products/processo...s/dsp-simd.php

    Comment


    • Originally posted by popper View Post
      "2011-01-28 05:36:46 < pengvado> but there is a vpcomltb and there is no vpcmpltb
      2011-01-28 05:36:57 < Dark_Shikari> oh you mean vcmp has gt only
      2011-01-28 05:36:58 < Dark_Shikari> ic
      2011-01-28 05:37:24 < Dark_Shikari> vphsubbw might be interesting?
      2011-01-28 05:38:44 < Dark_Shikari> hmm. VPMACSDD would be useful for high bit depth quant perhaps
      2011-01-28 05:38:53 < Dark_Shikari> 32 * 32 + 32 -> 32
      2011-01-28 05:39:25 < Dark_Shikari> and VPMACSSDD (saturating)
      2011-01-28 05:39:55 < Dark_Shikari> ... they have 16x16 MAC too
      2011-01-28 05:39:56 < Dark_Shikari> for integer
      2011-01-28 05:39:58 < Dark_Shikari> this might be useful.
      "
      Are they actually communicating or that's just gibberish to make it look like they know what they're talking about to non-tech people? How do you know they are even talking about AVX?

      Comment


      • Originally posted by devius View Post
        Are they actually communicating or that's just gibberish to make it look like they know what they're talking about to non-tech people? How do you know they are even talking about AVX?
        LOL read the log in sections sometime ,it goes back a long time so start at say
        2010-10-30 22:15:27 < kierank> is x264 participating in google code-in?

        http://akuvian.org/src/x264/freenode-x264dev.log.bz2

        its interesting for even non assembly reader's and you get good advice/idea's you can take to other projects etc, how you might improve audio codecs AAC etc being the latest insight for instance, not to mention why even professional GFX coders seem to have an aversion to writing working Gfx cuda/opencl etc x264 encoder patches once they ask on x264dev, all people who try are eaten by the cuda monster


        and OC if you goggled vphsubbw for instance you would find
        https://docs.google.com/viewer?url=h...Docs/26568.pdf

        "AMD64 Architecture
        Programmerís Manual

        Volume 4:
        128-Bit and 256-Bit
        Media Instructions"

        AKA SIMD/AVX etc

        and OC if you look above that section of the log i posted you would see they are writing AVX and mention AVX pathes etc...

        Comment

        Working...
        X