AKG: Automatic Kernel Generation for Neural Processing Units using Polyhedral Transformations
Thu 24 Jun 2021 02:00 - 02:05 at PLDI-A - Talks 2A: Machine Learning
Existing tensor compilers have proven their effectiveness in deploying deep neural networks on general-purpose hardware like CPU and GPU, but optimizing for neural processing units (NPUs) is still challenging due to the heterogeneous compute units and complicated memory hierarchy.
In this paper, we present AKG, a tensor compiler for NPUs. AKG first lowers the tensor expression language to a polyhedral representation, which is used to automate the memory management of NPUs. Unlike existing approaches that resort to manually written schedules, AKG leverages polyhedral schedulers to perform a much wider class of transformations, and extends the semantics of the polyhedral representation to combine complex tiling techniques and hierarchical fusion strategies. We also implement the domain-specific optimization of convolution in AKG. Moreover, to achieve the optimal performance, we introduce complementary optimizations in code generation, which is followed by an auto-tuner.
We conduct extensive experiments on benchmarks ranging from single operators to end-to-end networks. The experimental results show that AKG can obtain superior performance to both manual scheduling approaches and vendor provided libraries. We believe AKG will cast a light on the follow-up compiler works on NPUs.
Wed 23 JunDisplayed time zone: Eastern Time (US & Canada) change
13:30 - 14:05 | |||
13:30 5mTalk | Learning to Find Naming Issues with Big Code and Small Supervision PLDI DOI | ||
13:35 5mTalk | Fast and Precise Certification of Transformers PLDI Gregory Bonaert ETH Zurich, Dimitar I. Dimitrov ETH Zurich, Maximilian Baader ETH Zurich, Martin Vechev ETH Zurich DOI | ||
13:40 5mTalk | Web Question Answering with Neurosymbolic Program Synthesis PLDI Qiaochu Chen University of Texas at Austin, USA, Aaron Lamoreaux University of Texas at Austin, Xinyu Wang University of Michigan, Greg Durrett University of Texas at Austin, USA, Osbert Bastani University of Pennsylvania, Işil Dillig University of Texas at Austin DOI | ||
13:45 5mTalk | Robustness Certification with Generative Models PLDI Matthew Mirman ETH Zurich, Alexander Hägele ETH Zurich, Timon Gehr ETH Zurich, Pavol Bielik ETH Zurich, Martin Vechev ETH Zurich Link to publication DOI | ||
13:50 5mTalk | DNNFusion: Accelerating Deep Neural Networks Execution with Advanced Operator Fusion PLDI Wei Niu College of William & Mary, Jiexiong Guan College of William & Mary, Yanzhi Wang Northeastern University, Gagan Agrawal Augusta University, Bin Ren College of William & Mary DOI | ||
13:55 5mTalk | Vectorized Secure Evaluation of Decision Forests PLDI Raghav Malik Purdue University, Vidush Singhal Purdue University, Benjamin Gottfried Purdue University, Milind Kulkarni Purdue University DOI Pre-print | ||
14:00 5mTalk | AKG: Automatic Kernel Generation for Neural Processing Units using Polyhedral Transformations PLDI Jie Zhao State Key Laboratory of Mathematical Engineering and Advanced Computing, Bojie Li Huawei Technologies, Wang Nie Huawei Technologies, Zhen Geng Huawei Technologies, Renwei Zhang Huawei Technologies, Xiong Gao Huawei Technologies, Bin Cheng Huawei Technologies, Chen Wu Huawei, Yun Cheng Huawei Technologies, Zheng Li Huawei Technologies, Peng Di Huawei Technologies, Kun Zhang Huawei Technologies, Xuefeng Jin Huawei Technologies DOI |
Thu 24 JunDisplayed time zone: Eastern Time (US & Canada) change
01:30 - 02:05 | |||
01:30 5mTalk | Learning to Find Naming Issues with Big Code and Small Supervision PLDI DOI | ||
01:35 5mTalk | Fast and Precise Certification of Transformers PLDI Gregory Bonaert ETH Zurich, Dimitar I. Dimitrov ETH Zurich, Maximilian Baader ETH Zurich, Martin Vechev ETH Zurich DOI | ||
01:40 5mTalk | Web Question Answering with Neurosymbolic Program Synthesis PLDI Qiaochu Chen University of Texas at Austin, USA, Aaron Lamoreaux University of Texas at Austin, Xinyu Wang University of Michigan, Greg Durrett University of Texas at Austin, USA, Osbert Bastani University of Pennsylvania, Işil Dillig University of Texas at Austin DOI | ||
01:45 5mTalk | Robustness Certification with Generative Models PLDI Matthew Mirman ETH Zurich, Alexander Hägele ETH Zurich, Timon Gehr ETH Zurich, Pavol Bielik ETH Zurich, Martin Vechev ETH Zurich Link to publication DOI | ||
01:50 5mTalk | DNNFusion: Accelerating Deep Neural Networks Execution with Advanced Operator Fusion PLDI Wei Niu College of William & Mary, Jiexiong Guan College of William & Mary, Yanzhi Wang Northeastern University, Gagan Agrawal Augusta University, Bin Ren College of William & Mary DOI | ||
01:55 5mTalk | Vectorized Secure Evaluation of Decision Forests PLDI Raghav Malik Purdue University, Vidush Singhal Purdue University, Benjamin Gottfried Purdue University, Milind Kulkarni Purdue University DOI Pre-print | ||
02:00 5mTalk | AKG: Automatic Kernel Generation for Neural Processing Units using Polyhedral Transformations PLDI Jie Zhao State Key Laboratory of Mathematical Engineering and Advanced Computing, Bojie Li Huawei Technologies, Wang Nie Huawei Technologies, Zhen Geng Huawei Technologies, Renwei Zhang Huawei Technologies, Xiong Gao Huawei Technologies, Bin Cheng Huawei Technologies, Chen Wu Huawei, Yun Cheng Huawei Technologies, Zheng Li Huawei Technologies, Peng Di Huawei Technologies, Kun Zhang Huawei Technologies, Xuefeng Jin Huawei Technologies DOI |