ExTensor
Kartik Hegde, Hadi Asghari-Moghaddam, Michael Pellauer et al. · Proceedings of the ... annual International Symposium on Microarchitecture/Proceedings of the annual International Symposium on Microarchitecture · 2019
Generalized tensor algebra is a prime candidate for acceleration via customized ASICs. Modern tensors feature a wide range of data sparsity, with the density of non-zero elements ranging from 10-6% to 50%. This paper proposes a novel approach to accelerate tensor kernels based on the principle of hierarchical elimination of computation in the presence of sparsity. This approach relies on rapidly finding intersections---situations where both operands of a multiplication are non-zero---enabling new data fetching mechanisms and avoiding memory latency overheads associated with sparse kernels implemented in software.
Cited by 235 works. doi:10.1145/3352460.3358275
Authors
- Kartik Hegde
- Hadi Asghari-Moghaddam
- Michael Pellauer
- Neal Clayton Crago
- Aamer Jaleel
- Edgar Solomonik
- Joel Emer
- Christopher W. Fletcher
References
- Deep Residual Learning for Image Recognition
- ImageNet classification with deep convolutional neural networks
- TensorFlow: A system for large-scale machine learning
- Tensor Decompositions and Applications
- The Netflix Prize
- In-Datacenter Performance Analysis of a Tensor Processing Unit
- The university of Florida sparse matrix collection
- Hidden factors and hidden topics
- EIE: Efficient Inference Engine on Compressed Deep Neural Network
- Eyeriss