StarPlat Graph DSL - Program Analysis

During my final year at IIT Madras, I joined the StarPlat team led by Prof. Rupesh Nasre, aiming to develop a graph DSL supporting several backends including threaded, distributed, and GPU-accelerated computation As a part of my bachelor thesis with the group, I implemented multiple compiler-level optimizations for the DSL compiler using AST-level analyses, such as barriers in the parallel PageRank algorithm, and converting between a push-based implementation of the parallel SSSP algorithm and a pull-based implementation, to enhance the efficiency of generated C/C++ code. We rigorously tested these optimizations on real datasets containing up to 90 million nodes on several backends across algorithms such as the Single-Source Shortest Path, PageRank, and Strongly Connected Component among others, demonstrating how smart compiler optimizations and hardware-specific design can enhance the performance and scalability of graph algorithms on large datasets. The project led to my first patent(missing reference)!