MIT EECS | Keel Foundation Undergraduate Research and Innovation Scholar
Improving the Performance of the Tensor Algebra Compiler
- Computer Systems
Saman P. Amarasinghe
Tensor algebra is a powerful tool with applications in machine learning, data analytics, engineering, and the physical sciences. The combinations of possible tensor operations are infinite, so it is impossible to manually implement and optimize kernels for every operation of interest. The Tensor Algebra Compiler ( taco ) is an open-source C++ library that automatically generates performant kernels to compute any compound tensor algebra operation with the use of compiler techniques. My research will focus on improving the performance of taco by exploring different performance-engineering approaches, such as vectorization and parallelism. The goal of taco is to provide performance competitive kernels for any possible tensor operation.
I am participating in SuperUROP to get a feel for what it is like to do computer science research. As my time at college comes to an end, I have to decide whether I want to continue my studies through graduate school or focus my career elsewhere. I believe SuperUROP is the perfect opportunity to do that while contributing to an important open-source project. I look forward to learning what compilers are all about.