3

AsyncTaichi: Whole-Program Optimizations for Megakernel Sparse Computation and Differentiable Programming

We present a whole-program optimization framework for the Taichi programming language, leading to 1.73−2.76× speed up on our benchmarks.