MIT EECS - Keel Foundation Undergraduate Research and Innovation Scholar
“ Proposal for work on high-performance physics simulation language
Saman P. Amarasinghe
Many applications could benefit from the development of a language designed for high-performance physics simulations: machine learning, physics research, real-time game rendering, etc. The goal of this project is to work with Fredrik Kjolstad on the development and validation of such a language based on parallelized computations over a hyper-graph of data. I will develop visualizations to represent the data and operations and also work with the team to implement collisions in this system without losing the performance benefits of parallelizing the computations on each element of the data. On the compiler side, I will look into developing a parser for the language to move beyond building queries using the programmtic API.
I develop a sanbox-style particle physics Android application, “The Elements: Sand Game”. A lot of the ideas of developing a fast time-stepping physics system for that game are relevant to this research. In particular, I spent time looking into developing a shader-based alternative to the physics for that game, which would have required a collision system implemented in a highly parallelized environment.