Blocked Ellpack Sparse Matrix Support in PyTorch

This repository contains the code, experiments, and documentation for my thesis project at ACSAI, Sapienza University of Rome, focused on extending PyTorch's sparse matrix capabilities.

Thesis Overview

Modern machine learning workloads often deal with large sparse matrices, but efficient support for custom sparse formats in PyTorch is still limited. This project introduces — Blocked ELLPACK (BELL) — format into the PyTorch ecosystem.

---

Key Features

• BELL Format Support: Blocked ELLPACK implementation in C++ using LibTorch (PyTorch C++ frontend) and CUDA.
• Automatic Block Size Selection: Heuristic that selects the optimal block size based on matrix structure.
• Performance Analysis: Benchmarks against existing sparse formats (COO, CSR, etc.).
• Multithreading with OpenMP: Optimized for parallel computation.
• Debugging and Visualization Tools: Tools to inspect and print tensor blocks and layouts.