High Performance Computing Course Project (CS610)

Course: CS610 – High Performance Computing
Instructor: Prof. Swarnendu Biswas
Institute: IIT Kanpur
Duration: Aug 2025 – Dec 2025

This repository contains optimized CPU and GPU implementations of core numerical workloads, focusing on instruction-level, thread-level, and accelerator-based parallelism.

---

🚀 Implemented Projects

1. Serial Matrix Multiplication with AVX2 & SSE4

• File: matmul.cpp
• Optimized using:
  • AVX2 and SSE4 vector intrinsics
  • Loop unrolling and cache-friendly access
• Achieved up to 6× speedup over naive implementation

---

2. Grid Search (Baseline – Serial)

• File: gridsearch_original.cpp
• Reference serial implementation
• Used for correctness and performance comparison

---

3. Grid Search (Parallel – OpenMP)

• File: gridsearch_openmp.cpp
• Parallelized using OpenMP
• Achieved ~6.5× speedup on multi-core CPU

---

4. 2D/3D Convolution using CUDA

• File: convolution_gpu.cu
• GPU implementation using CUDA
• Achieved ~4× speedup compared to serial CPU version

---

🛠️ Build Instructions

Requirements

• g++ (with OpenMP support)
• nvcc (CUDA Toolkit)
• Linux environment recommended

Compile All Programs

make