paragroup / dac

Divide-and-Conquer Parallel Pattern Implementation in FastFlow

DAC

This repository contains the source code used to perform the experimental evaluations in the paper entitled A Divide-and-Conquer Parallel Pattern Implementation for Multicores, presented at SEPS 2016.

In the repository you can find the applications used for the evaluations and the backend implementation of the Parallel Divide and Conquer pattern in OpenMP, Intel TBB and Fastflow.

The pattern (and related backend implementations) can be used to easily parallelize other Divide and Conquer algorithms. Details on the interface can be found in the paper.

Applications

To understand how the pattern works and its interface a basic example for the n-th fibonacci number computation is provided.

The main applications used for the evaluation are essentially three: the merge- and quick-sort algorithms and the Strassen algorithm for matrix multiplication. It is important to notice that for the three applications the main program is the same for the different backends (can be found under the src/ folder). The different backends can be selected by using proper compiler directives (USE_OPENMP for OpenMPUSE_TBB for the Intel TBB version and USE_FF for the FastFlow version).

In addition, to compare the pattern based version with third-party algorithms are present hand-made parallelizations of the aforementioned applications (for the merge-sort comparison we used the stable sort implementation provided by Intel here).

Usage

Required software

The artifact uses external software. In particular:

• FastFlow: a C++ parallel programming framework targeting shared-memory architectures. Website: http://calvados.di.unipi.it/
• Intel Stable Sort: a C++11 implementation of a Stable Merge Sort provided by Intel. Website: https://software.intel.com/en-us/articles/a-parallel-stable-sort-using-c11-for-tbb-cilk-plus-and-openmp

In the sequel we will provide a brief description for their downloading

Fastflow

It is an header-only library. Therefore, it is only required to download it from the website or the SVN. To download the latest version and save it into the fastflow directory, run the following command in the shell:

$ svn checkout svn://svn.code.sf.net/p/mc-fastflow/code/ fastflow

Intel Stable Sort

The source code can be downloaded at https://software.intel.com/en-us/articles/a-parallel-stable-sort-using-c11-for-tbb-cilk-plus-and-openmp. Once decompressed and compiled it is ready to be used.

Compilation

Before compiling the user must provide two diffent environment variables: FASTFLOW_DIR that points to the FastFlow library directory and INTEL_STABLESORT_DIR that points to the directory that contains the Intel source code (used for the comparison), After that, the code can be compiled. The set of command is the following:

 $ export FASTFLOW_DIR=<...path to fastflow...>
 $ export INTEL_STABLESORT_DIR=<... path to intel source code...>
 $ make -j

This will produce different executables:

• fibonacci_dac_{openmp,tbb,ff}: are the the parallel pattern based implementations of the fibonacci problem that use the OpenMP, Intel TBB and FastFlow backends respectively;
• mergesort_dac_{openmp,tbb,ff}: that are the parallel pattern based implementations of the mergesort problem;
• quicksort_dac_{openmp,tbb,ff}: the implementations for the quicksort problems for the different backends;
• strassen_dac_{openmp,tbb,ff}: implementations for the Strassen matrices multiplication algorithm;
• stable_mergesort_dac_{openmp,tbb,ff}: implementation of the Intel Stable Sort algorithm used for the comparison. It is essentially the same algorithm (with the same classes and data types) provided by Intel whose divide-and-conquer part is parallelized using the proposed pattern;
• quicksort_hm_{openmp,tbb} and strassen_hm_{openmp,tbb}: hand made parallelizations for OpenMP and TBB
• intel_sort_{openmp,tbb}: the intel version of the program. Can be compiled directly from the source codes provided in the Intel WebSite.

Each of these programs require certain parameters. To see the right sequence it is sufficient to invoke the program without arguments.

How to Cite

If our work is useful for your research, please cite the following paper:

@inproceedings{10.1145/3002125.3002128,
 author = {Danelutto, Marco and De Matteis, Tiziano and Mencagli, Gabriele and Torquati, Massimo},
 title = {A Divide-and-Conquer Parallel Pattern Implementation for Multicores},
 year = {2016},
 isbn = {9781450346412},
 publisher = {Association for Computing Machinery},
 address = {New York, NY, USA},
 url = {https://doi.org/10.1145/3002125.3002128},
 doi = {10.1145/3002125.3002128},
 booktitle = {Proceedings of the 3rd International Workshop on Software Engineering for Parallel Systems},
 pages = {10–19},
 numpages = {10},
 keywords = {Divide and Conquer, High-level parallel patterns},
 location = {Amsterdam, Netherlands},
 series = {SEPS 2016}
}

Contributors

DAC has been developed by Tiziano De Matteis.