Parallel Programming Course

The following parallel programming technologies are considered in practice:

• MPI
• OpenMP
• TBB
• std::thread

Rules for submissions

1. You are not supposed to trigger CI jobs by frequent updates of your pull request. First you should test you work locally with all the scripts (code style).
   • Respect others time and don't slow down the job queue
2. Carefully check if the program can hang.

0. Download all submodules

git submodule update --init --recursive --depth=1

1. Set up your environment

Code style analysis

Please, follow Google C++ Style Guide.

Code style is checked using clang-format tool.

Parallel programming technologies

MPI

• Windows (MSVC):

Installers link. You have to install msmpisdk.msi and msmpisetup.exe.

• Linux (gcc and clang):

sudo apt install -y mpich openmpi-bin libopenmpi-dev

• MacOS (apple clang):

brew install open-mpi

OpenMP

OpenMP is included into gcc and msvc, but some components should be installed additionally:

• Linux (gcc and clang):

sudo apt install -y libomp-dev

• MacOS (llvm):

brew install llvm
brew install libomp

TBB

• Windows (MSVC), Linux (gcc and clang), MacOS (apple clang):
  • Build as 3rdparty in the current project

std::thread

• std::thread is included into STL libraries.

2. Build the project with CMake

Navigate to a source code folder.

1. Configure the build: Makefile, .sln, etc.

mkdir build && cd build
cmake -D USE_SEQ=ON -D USE_MPI=ON -D USE_OMP=ON -D USE_TBB=ON -D USE_STL=ON -D USE_FUNC_TESTS=ON -D USE_PERF_TESTS=ON -D CMAKE_BUILD_TYPE=Release ..

Help on CMake keys:

• -D USE_SEQ=ON enable Sequential labs (based on OpenMP's CMakeLists.txt).
• -D USE_MPI=ON enable MPI labs.
• -D USE_OMP=ON enable OpenMP labs.
• -D USE_TBB=ON enable TBB labs.
• -D USE_STL=ON enable std::thread labs.
• -D USE_FUNC_TESTS=ON enable functional tests.
• -D USE_PERF_TESTS=ON enable performance tests.
• -D CMAKE_BUILD_TYPE=Release required parameter for stable work of repo.

A corresponding flag can be omitted if it's not needed.

2. Build the project:

cmake --build . --config RELEASE

3. Check the task

• Run <project's folder>/build/bin

3. How to submit you work

• There are mpi, omp, seq, stl, tbb folders in tasks directory. Move to a folder of your task. Make a directory named <last name>_<first letter of name>_<short task name>. Example: seq/nesterov_a_vector_sum. Please name all tasks same name directory. If seq task named seq/nesterov_a_vector_sum then omp task need to be named omp/nesterov_a_vector_sum.

• Go into the newly created folder and begin you work on the task. There must be only 4 directory with files:

  • func_tests - google tests directory with files for the functional tests of task.
  • include - a header files directory with function prototypes.
  • perf_tests - google tests directory with files for the performance tests of task. The number of tests must be 2 - run_task and run_pipeline.
  • src - a source files directory with functions realization.

• We need to know that exist 10 executable files for running:

  • <mpi, omp, seq, stl, tbb>_<func, perf>_tests e.g. omp_perf_tests - executable file for performance tests of OpenMP practice tasks.

• All prototypes and classes in the include directory must be namespace escaped, name your namespace in the following way:

  namespace <last name>_<first letter of name>_<short task name>_<technology> {
  ...
  }
  e.g.
  namespace nesterov_a_test_task_seq {
  
  std::vector<int> getRandomVector(int sz);
  class TestTaskSequential : public ppc::core::Task {
  ...
  };
  
  }  // namespace nesterov_a_test_task_seq
  

• Name your group of tests and current test in the following way:

  • for functional tests:

  TEST(<last name>_<first letter of name>_<short task name>_<technology>, <any_name_of_test_case>) {
  ...
  }
  e.g.
  TEST(nesterov_a_vector_sum_omp, test_sum) {
  ...
  }
  

  • for performance tests:

  TEST(<last name>_<first letter of name>_<short task name>_<technology>, <type_of_performance_validation>) {
  ...
  }
  e.g.
  TEST(nesterov_a_vector_sum_stl, test_pipeline_run) {
  ...
  }
  TEST(nesterov_a_vector_sum_stl, test_task_run) {
  ...
  }
  

• All tests need to be written without main() function

• Name your pull request in the following way:

  • for tasks:

  <Last Name First Name>. Task <Number of task>. Variant <Number of variant>. <Full name of task>.
  Нестеров Александр. Задача 1. Вариант 123. Сумма элементов вектора.
  

• Provide the full task definition in pull request's description.

  Example pull request is located in repo's pull requests.

• Work on your fork-repository. Keep your work on a separate branch and NOT on master!!!. Name you branch in the same way as your task's folder. To create a branch run:

  git checkout -b nesterov_a_vector_sum
  

Failing to follow the rules makes the project build red.

And finally,