Aart Bik
distinguished engineer at NVIDIA
Actions
Aart J.C. Bik received his master’s degree from Utrecht University and his PhD from Leiden University. He was a principal engineer at Intel, where he was the lead compiler architect of automatic vectorization. At Google he worked on the large-scale graph processing system Pregel, Google Glass, the Android optimizing compiler, and the MLIR Sparsifier. He is now a distinguished engineer at NVIDIA, researching sparsity in AI.
Exhaustive Sparse Kernel Search with the Universal Sparse Tensor
The Universal Sparse Tensor (UST) decouples a tensor's sparsity from its actual memory layout for greater flexibility and performance. A tensor format DSL (Domain Specific Language) describes how the sparse tensor should be represented. Type polymorphism on a small set of base operations defines the vast space of instances for these operations. This talk demonstrates how the UST can be used to perform an exhaustive state space search over all CUDA kernels for PyTorch operations, accounting for many different iteration orderings and sparse storage formats. When integrated with heuristic pruning techniques and potential agent-assisted optimization, this methodology facilitates the acceleration of sparsity in a manner that was previously unattainable for most model developers.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top