mpyCUDA

Unifying MPI, C++, CUDA, Apache Spark, and Python for maximum parallel processing power!

1st Project: Image Convolution Engine	2nd Project: AWACS Simulation

---

Caution

🛑 ENABLE YOUR GPU!

These projects REQUIRE a CUDA-capable GPU (NVIDIA T4 or better). If you run this on a standard CPU runtime, the code will fail or run extremely slowly.

How to fix in Colab:

1. Go to Runtime → Change runtime type
2. Select T4 GPU (or better)
3. Click Save

---

🚀 Quick Start — Google Colab

This is the recommended way for lecturers and reviewers.

1. Click the Colab Badge above for the project you wish to evaluate.
2. In Colab, go to Runtime → Change runtime type → GPU (T4 is sufficient).
3. Run all cells from top to bottom (Runtime → Run all).

---

📁 Projects Overview

This repository contains two parallel programming projects for the Parallel Programming A course.

1. Image Convolution Engine (ACTIVE)
   • High-performance image filtering (Blur, Sobel, Sharpen, Emboss).
   • Uses MPI + CUDA + OpenMP hybrid parallelism.
   • Implements Guided Scheduling for dynamic load balancing.
2. Ace Combat AWACS Simulation (ACTIVE)
   • Distributed IFF (Identification Friend or Foe) threat sorting using Apache Spark MLlib (Random Forest).
   • Advanced Feature Engineering (vectorized closest-base tracking and angle-difference trigonometry) over 1 million radar sweeps.
   • Parallel tri-base interception trajectory math solved concurrently inside custom CUDA GPGPU kernels.
   • Dynamic scramble matchmaking (Allied F-15 Eagle dashes for MiG-29s, F-14 Tomcat scrambles for Su-27s, SAM batteries for unknown threats).
   • Atomic GPU operations for thread-safe concurrent target evasion logging.
   • High-fidelity interactive 3D tactical map animating 20 dogfights, complete with solid square bases, cyan/red delta-wing markers, expanding orange hit explosions, orbiting CAP squads, and smooth glass-like Mesh3d radar domes (zero grid spirals).
   • Graceful nearest-base localized evasion callouts.

Project Structure

mpyCUDA/
├── Kamen-Rider-Image-Convolution/
│   ├── colab_notebook.ipynb                 ← 1st Project Colab Entry Point
│   ├── images/                              ← Input/output image assets
│   ├── scripts/
│   │   ├── Makefile                         ← Linux build script
│   │   ├── parallel_image.cu                ← The Hybrid MPI+CUDA Engine
│   │   └── parallel_conv.cu                 ← Matrix Convolution Sandbox
│   ├── README.md                            ← Detailed Project Docs
│   └── presentation_guide.md                ← Presentation Outline
│
└── Ace-Combat-AWACS-Simulation/
    ├── colab_notebook.ipynb                 ← Colab Notebook (Linux)
    ├── windows_notebook.ipynb               ← Local Notebook (Windows/Anaconda)
    ├── create_notebook.py                   ← Generator for Colab version
    ├── create_windows_notebook.py           ← Generator for Windows version
    ├── scripts/
    │   ├── build_cuda.bat                   ← Windows MSVC compile script
    │   ├── data_gen.py                      ← 1 Million Record Radar Synthesizer
    │   └── trajectory_math.cu               ← Parallel SAM Trajectory Engine
    └── presentation_guide.md                ← AWACS Presentation Outline

🛠️ Local Build Instructions

Project 1: Image Convolution (Linux / WSL)

cd Kamen-Rider-Image-Convolution/scripts
make all

# Run the primary Image Engine (sobel mode)
mpirun --oversubscribe -n 4 ./parallel_image ../images/input.jpg ../images/output.jpg sobel

Project 2: AWACS Simulation (Windows)

cd Ace-Combat-AWACS-Simulation
# Compile the CUDA engine into a dynamic link library (DLL)
./scripts/build_cuda.bat

# Launch the interactive notebook
jupyter notebook windows_notebook.ipynb

⚡ Core Features

Hybrid Parallel Architecture (Project 1)

• Guided Scheduling: Dynamic load balancing via MPI Master-Worker pattern to prevent idle cores.
• Hybrid Parallelism: CUDA GPU kernels + OpenMP CPU threads + MPI Clusters working in tandem.
• Color Image Support: Full BGR channel-wise processing.

Big Data & GPU Interop (Project 2)

• Spark MLlib: Distributed processing of 1,000,000 radar records to predict critical threats.
• PySpark Feature Engineering: Vectorized trigonometry (atan2) for accurate model training.
• CUDA Atomics: Thread-safe global evasion counters natively processed on the GPU.
• Ctypes Integration: Seamless passing of data arrays from Python directly to CUDA VRAM.