==============
Loop Unrolling                                                        [TOC]
==============

We optimize our previous implementation by manual loop unrolling.  Again, the
performance achieved here I would have expected from a smart compiler for the
`demo-pure-c` branch.  At least if we provide enough hints through compiler
flags and attributes.


Select the demo-naive-sse-with-intrinsics-unrolled Branch
=========================================================
Check out the `demo-naive-sse-with-intrinsics` branch:

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

Then we compile the project

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


Unrolling
=========

TODO:

    - Add some of the lecture notes about pipelining, branch prediction and
      prefecting.
    - Add pictures on how we realize this in the implementation.


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-naive-sse-with-intrinsics-unrolled`
branch:

    *--[SHELL(path=ulmBLAS/bench)]--------------------------------------*
    |                                                                   |
    |  grep PASS report > demo-naive-sse-with-intrinsics-unrolled       |
    |                                                                   |
    *-------------------------------------------------------------------*

With the gnuplot script

:import: ulmBLAS/bench/bench4.gps

we feed gnuplot

    *--[SHELL(path=ulmBLAS/bench)]--------------------------------------*
    |                                                                   |
    |  gnuplot bench4.gps                                               |
    |                                                                   |
    *-------------------------------------------------------------------*

and get

    ---- IMAGE -------------
    ulmBLAS/bench/bench4.svg
    ------------------------



:navigate: __up__    -> doc:index
           __back__  -> doc:page03/index
           __next__  -> doc:page05/index