ParallelOpenMP

A set of C++ parallel-computing labs built with OpenMP and MPI. Each lab is a self-contained program that compares a sequential implementation against a parallel one and reports execution times, covering multithreading, SIMD vectorization, recursive task parallelism, and message passing between processes.

Labs

Lab	Topic	Demonstrates
1-lab	Hello world	Basic #pragma omp parallel region with a critical section, thread count taken from std::thread::hardware_concurrency().
2-lab	Uneven workload split	Manual chunking of an array across 8 threads using master and barrier; per-thread fill and stencil-calculation timing.
3-lab	Matrix multiplication	parallel for collapse(3) with atomic accumulation over a 500x500 matrix; sequential vs. 1/2/4/8-thread comparison.
4-lab	Recursive fractals	Koch snowflake and Towers of Hanoi parallelized with #pragma omp task ... if(...); Koch output rendered to an SVG file.
5-lab	SIMD reduction	Series summation vectorized with #pragma omp simd reduction(...), sequential vs. parallel timing.
6-lab	SIMD benchmarking	Same SIMD reduction kernel as lab 5 run in a repeated benchmark loop; ships with a Linux compile script and Intel icc optimization-flag examples.
7-lab-mpi	MPI message passing	Two ranks generate array halves and exchange them with MPI_Send/MPI_Recv, then each sums its portion and reports MPI_Wtime.
control-task	Layered network pass	Multi-layer feed-forward pass using parallel for reduction and a tanh-style activation, with if-toggled parallelism.

Tech stack

• C++ with OpenMP (#pragma omp directives, omp_* runtime calls).
• MPI for the message-passing lab (mpi.h; the Visual Studio project links msmpi.lib).
• Visual Studio solutions/projects (.sln, .vcxproj) per lab, with OpenMP support enabled (<OpenMPSupport>true</OpenMPSupport>).
• Timing via omp_get_wtime() and MPI_Wtime().

Build and run

Visual Studio (Windows)

Open the Parallel_OpenMP.sln inside any lab folder and build/run it. OpenMP support is already enabled in the project settings; the MPI lab additionally requires MS-MPI (msmpi.lib).

Command line (Linux/macOS)

Most labs compile directly with an OpenMP-capable compiler:

c++ -o out -fopenmp Parallel_OpenMP.cpp
./out

6-lab includes a helper script that uses the system compiler by default and Intel icc when arguments are passed:

cd 6-lab
./compile.sh             # c++ -o out -fopenmp Parallel_OpenMP.cpp
./compile.sh -O1 -ipo    # icc -O1 -ipo -fopenmp Parallel_OpenMP.cpp -o out

See 6-lab/commands.txt for further icc optimization-flag examples (-O0, -ipo, -prof-gen/-prof-use, -parallel).

The MPI lab is compiled and launched with the MPI toolchain (built for two ranks):

cd 7-lab-mpi
mpic++ -o out Parallel_OpenMP.cpp
mpirun -np 2 ./out

Project structure

.
├── 1-lab/            # OpenMP hello world (parallel + critical)
├── 2-lab/            # Manual array chunking across threads
├── 3-lab/            # Matrix multiplication (collapse + atomic)
├── 4-lab/            # Koch snowflake & Hanoi towers (omp task)
│   ├── koch.cpp
│   ├── hanoi.cpp
│   └── Parallel_OpenMP.cpp
├── 5-lab/            # SIMD reduction
├── 6-lab/            # SIMD reduction benchmark + compile.sh / commands.txt
├── 7-lab-mpi/        # MPI send/recv between two ranks
└── control-task/     # Layered feed-forward pass

Each lab folder contains a Parallel_OpenMP.cpp source file alongside its Visual Studio solution and project files. Lab 4 additionally splits its workloads into koch.cpp and hanoi.cpp.