Thomas Algorithm Demo application

Dmitry Vodopyanov, Vasily Kumaev
Lobachevsky University, Nizhny Novgorod, Russia

Overview

This application provides four implementations of Thomas algorithm using the technologies below:

• Sequintal (common C++ libraries)
• OpenMP
• Intel(R) Threading Building Blocks
• Intel(R) Cilk(TM) Plus

This algorithm solves the problem of finding the interpolating square spline.

Project structure and launch diagram

----------------------------------------------------------------------------
|                             MatrixGeneration                             |
----------------------------------------------------------------------------
      ||                   ||                  ||                  ||
      \/                   \/                  \/                  \/
---------------    ------------------    ---------------    ----------------
| Version_Seq |    | Version_OpenMP |    | Version_TBB |    | Version_Cilk |
---------------    ------------------    ---------------    ----------------
      ||                   ||                  ||                  ||
      \/                   \/                  \/                  \/
----------------------------------------------------------------------------
|                                  Spline                                  |
----------------------------------------------------------------------------

MatrixGeneration project generates tridiagonal matrices and writes them to file.
Version_Seq, Version_OpenMP, Version_TBB, Version_Cilk projects get the MatrixGeneration's file, make calculations using one of the technologies describing above, and create two files: the file with result vector and the file with execution time of the implementation.
Spline project implements the logic of finding the interpolating square spline and executes each of Version_### project, where ### is Seq, OpenMP, TBB or Cilk.

BKMs

Solution configuration

1. Configure the solution using CMake for Visual Studio
2. Open Visual Studio and make sure that all your projects compile with Intel(R) C++ Compiler
3. For all projects: Go to Properties -> Debugging -> Working Directory and add set the value $(ProjectDir)/../../../bin
4. For all projects: Go to Properties -> C/C++ -> Language [Intel C++] -> OpenMP Support -> Generate Parallel Code (/Qopenmp)
5. For MatrixGeneration project: Go to Properties -> Debugging -> Command Arguments and set value (example) 5 0 1 inputfile, where 5 is dimension of the matrix, 0 and 1 are minimum and maximum elements of the matrix, inputfile is name of the file that will be created
6. For all Version_### projects: Go to Properties -> Debugging -> Command Arguments and set value inputfile outputfile timefile, where ### is Seq, OpenMP, TBB or Cilk

Compile and run using CLI

Automation

1. Execute test/run.py with the following command (example): python run.py 1000 10 15 seq openmp
2. Pros:
3. it is easy to run,
4. it is cross-platform,
5. you can see comparison between versions (execution time, unconsistency in the results)
6. Cons:
7. if your Intel(R) Parallel Studio XE or Intel(R) Sustem Studio installed in non-default folder, you should change the path in run.py. Default paths: Windows - C:\Program Files (x86)\IntelSWTools, Linux - /opt/intel,
8. it is very "hardcoded" script, something can goes wrong

Manual

Windows

1. Set the environment of Intel(R) Parallel Studio XE
   C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\bin\compilervars.bat intel64
2. Go to ThomasAlgorithm\src folder
3. Create file with matrix:
   icl.exe main_gen_matrix.cpp /O2 /D NDEBUG
main_gen_matrix.exe <dimension> <minimum> <maximum> <filename>
4. Compile and run Sequintal, OpenMP, Intel(R) TBB or Intel(R) Cilk(TM) Plus version:
   icl.exe ###_version.cpp main.cpp /O2 /D NDEBUG /Qopenmp /link tbb.lib
###_version.exe inputfile outputfile timefile,
   where ### is seq, openmp, tbb or cilk

Linux and OS X

1. Set the environment of Intel(R) Parallel Studio XE or Intel(R) System Studio (Linux only)
   source /opt/intel/compilers_and_libraries/***/bin/compilervars.sh intel64,
   where *** is linux or mac
2. Go to ThomasAlgorithm/src folder
3. Create file with matrix: icc -O2 -DNDEBUG -m64 -qopenmp -o main_gen_matrix ../src/main_gen_matrix.cpp
./main_gen_matrix <dimension> <minimum> <maximum> <filename>
4. Compile and run Sequintal, OpenMP, Intel(R) TBB or Intel(R) Cilk(TM) Plus version:
   icc ###_version.cpp main.cpp -O2 -DNDEBUG -m64 -qopenmp -ltbb -o ###_version
./###_version inputfile outputfile timefile,
   where ### is seq, openmp, tbb or cilk

Development process

1. AR Vasily: Please make sure that the version of the file you contribute is the latest: as frequently as you can execute git pull