=============================
Using Optimized Micro-Kernels                                   [TOC:2]
=============================

Changes:

- The benchmark program `matprod.cc` was not modified.
- In `gemm.hpp` memory for buffers now gets aligned.
- If you compile with `-DHAVE_AVX` the file `avx.hpp` gets
  included.  This contains inline-assembly code for AVX.
- If you compile with `-DHAVE_FMA` the file `fma.hpp` gets
  included.  This contains inline-assembly code for FMA.
- Default block sizes can be overwritten with:
    - `-DBS_D_MC=512`
    - `-DBS_D_KC=512`
    - ...

Tar-Ball for this Session
=========================
The tar-ball __session2.tgz__ contains the files:

---- SHELL ---------------------------------------------------------------------
tar tfvz session2.tgz
--------------------------------------------------------------------------------

:links: session2.tgz -> http://www.mathematik.uni-ulm.de/~lehn/test_ublas/session2.tgz


Compile and Run Benchmark
=========================
For a quick benchmark we again limit $m$ to $500$ with `-DM_MAX=500`. This
time we also enable the optimized micro kernel for AVX with `-DHAVE_AVX`:

---- SHELL (path=session2,hostname=heim) ---------------------------------------
g++ -Ofast -mavx -Wall -std=c++11 -DNDEBUG -DHAVE_AVX -DM_MAX=500 -I ../boost_1_60_0 matprod.cc
./a.out
--------------------------------------------------------------------------------

Main Program with Benchmark and Test
====================================
:import: session2/matprod.cc

Core Function for Matrix-Matrix Produkt
=======================================
:import: session2/gemm.hpp

Micro-Kernel for AVX
====================
:import: session2/avx.hpp


Micro-Kernel for FMA
====================
:import: session2/fma.hpp

Benchmark Results
=================

---- SHELL (path=session2,hostname=heim) ---------------------------------------
g++ -Ofast -mavx -Wall -std=c++11 -DHAVE_AVX -DNDEBUG -I ../boost_1_60_0 matprod.cc
./a.out  > report.session2
cat report.session2
gnuplot plot.session2.mflops
gnuplot plot.session2.time
gnuplot plot.session2.time_log
--------------------------------------------------------------------------------


MFLOPS
------
---- IMAGE -----------------------
session2/bench.session2.mflops.svg
----------------------------------


Time
----

---- IMAGE -----------------------
session2/bench.session2.time.svg
----------------------------------

Time with Logarithmic scale
---------------------------

---- IMAGE -------------------------
session2/bench.session2.time_log.svg
------------------------------------


:navigate: back  -> doc:session1/page01
           next  -> doc:session3/page01