Performance assessment of the cube Remapper

This project allows to assess the performance of the Cube plugin: "remapper".
We mainly compare the deployed implementation (refered to as trunk) with our custom one (refered to as AIO).
For more details on this project or the cube remapper custom implementation, please refer to my thesis.

Prerequisite

Before installing or running any of the cited projects, one must insure that the following tools are installed:

1. Score-P(see the INSTALL file within the Score-P project). The "orphaned pthreads" version needs to be used.
   If PAPI is not recognized by Score-P during the installation, one must restart the whole process by previously running the command:

./configure --without-shmem --with-papi-lib=<path to PAPI>/lib --with-papi-header=<path to PAPI>/include --with-papi-header=/homeb/zam/sidlakhr/papi-5.5.1/src/_install/include/ \

2. Cube (see the INSTALL file within Cube project). We will re install different version of it. But this basic version is required for internal profiling needs.
   
3. JUBE
   
4. PAPI
   

Install

Once the two previously cited project ([trunk](https://svn.version.fz-juelich.de/scalasca_soft/Cube2.0/CubeLib/trunk] and AIO have been downloaded and set as neighbor directories to the current project (outside the current project), the installation process of these two projects might be done through the Makefile within the current project. One might achieve a basic installation through the following commands (from the current project):

Install the trunk version of Cube

make preCompileTrunk
make compileTrunkScorep

Install the custom (AIO) version of Cube

make preCompileAio
make compileAioScorep

Run the performance evaluation

The performance evaluation might be runed through the command

make runAllBenchmark

Then the results might be ploted through different graphical representations using the command

make plotPointCompare
make plotPoint

To plot some previous performance results (obtained on an HPC Intel Xeon CPU E5-2680 v3 @ 2.50GHz), you can run the command

make plotPointCompareArchive

An HMI will allow you to choose among different archives that we obtained.

The custom pthread wrapper

TODO

The custom alligned allocation

This thread safe version of posix_memalign has beed designed specifically for our memory access pattern: TODO