=============================
Inline Assembler Micro Kernel                                            [TOC]
=============================

We use the __Intel SSE intrinsics guide__ and translate the micro kernel for
the previous page into assembler code ourself.  Declaring the assembler block
as `volatile` ensures that the compiler will not reorder any instruction.  This
way we are in full control to optimize pipelining!

Select the demo-sse-asm Branch
==============================
Check out the `demo-sse-asm` branch:

    *--[SHELL(path=ulmBLAS)]--------------------------------------------*
    |                                                                   |
    |  git branch -a                                                    |
    |  git checkout -B demo-sse-asm                                  +++|
    |                  remotes/origin/demo-sse-asm                      |
    |                                                                   |
    *-------------------------------------------------------------------*

Then we compile the project

    *--[SHELL(path=ulmBLAS,height=15)]----------------------------------*
    |                                                                   |
    |  make                                                             |
    |                                                                   |
    *-------------------------------------------------------------------*


The Micro Kernel
================
We bascially just translate the intrinsics one by one into assembler
instructions.  Using the __Intel SSE intrinsics guide__ this is fairly easy.
In the comments we have left the original intrinsics.  Note that we decalre
the `asm` block as volatile.  Otherwise we again have the  problem that the
smart ass compiler reorders the instructions.


The dgemm_nn Code
=================
:import: ulmBLAS/src/level3/dgemm_nn.c [linenumbers]


Benchmark Results
=================
We run the benchmarks

    *--[SHELL(path=ulmBLAS)]--------------------------------------------*
    |                                                                   |
    |  cd bench                                                         |
    |  ./xdl3blastst > report                                           |
    |  cat report                                                       |
    |                                                                   |
    *-------------------------------------------------------------------*

and filter out the results for the `demo-sse-asm` branch:

    *--[SHELL(path=ulmBLAS/bench)]--------------------------------------*
    |                                                                   |
    |  grep PASS report > demo-sse-asm                                  |
    |                                                                   |
    *-------------------------------------------------------------------*

With the gnuplot script

:import: ulmBLAS/bench/bench9.gps

we feed gnuplot

    *--[SHELL(path=ulmBLAS/bench)]--------------------------------------*
    |                                                                   |
    |  gnuplot bench9.gps                                               |
    |                                                                   |
    *-------------------------------------------------------------------*

and get

    ---- IMAGE -------------
    ulmBLAS/bench/bench9.svg
    ------------------------


:links: Intel SSE intrinsics guide -> https://software.intel.com/sites/landingpage/IntrinsicsGuide/

:navigate: __up__    -> doc:index
           __back__  -> doc:page07/index
           __next__  -> doc:page09/index