DeepCuts: A Deep Learning Optimization Framework for Versatile GPU Workloads
Fri 25 Jun 2021 21:00 - 21:05 at PLDI-A - Talks 5A: Machine Learning and Probabilistic Programming
Widely used Deep Learning (DL) frameworks, such as TensorFlow, PyTorch, and MXNet, heavily rely on the NVIDIA cuDNN for performance. However, using cuDNN does not always give the best performance. One reason is that it is hard to handle every case of versatile DNN models and GPU architectures with a library that has a fixed implementation. Another reason is that cuDNN lacks kernel fusion functionality that gives a lot of chances to improve performance. In this paper, we propose a DL optimization framework for versatile GPU workloads, called DeepCuts. It considers both kernel implementation parameters and GPU architectures. It analyzes the DL workload, groups multiple DL operations into a single GPU kernel, and generates optimized GPU kernels considering kernel implementation parameters and GPU architecture parameters. The evaluation result with various DL workloads for inference and training indicates that DeepCuts outperforms cuDNN/cuBLAS-based implementations and the state-of-the-art DL optimization frameworks, such as TVM, TensorFlow XLA, and TensorRT.
Fri 25 JunDisplayed time zone: Eastern Time (US & Canada) change
09:00 - 09:40 | |||
09:00 5mTalk | DeepCuts: A Deep Learning Optimization Framework for Versatile GPU Workloads PLDI Wookeun Jung Seoul National University, Thanh Tuan Dao Seoul National University, Jaejin Lee Seoul National University DOI | ||
09:05 5mTalk | Provable Repair of Deep Neural Networks PLDI Matthew Sotoudeh University of California at Davis, Aditya V. Thakur University of California at Davis DOI Pre-print Media Attached | ||
09:10 5mTalk | DreamCoder: Bootstrapping Inductive Program Synthesis with Wake-Sleep Library Learning PLDI Kevin Ellis Cornell University, Lionel Wong Massachusetts Institute of Technology, Maxwell Nye Massachusetts Institute of Technology, Mathias Sablé-Meyer PSL University; Collège de France; NeuroSpin, Lucas Morales Massachusetts Institute of Technology, Luke Hewitt Massachusetts Institute of Technology, Luc Cary Massachusetts Institute of Technology, Armando Solar-Lezama Massachusetts Institute of Technology, Joshua B. Tenenbaum Massachusetts Institute of Technology DOI | ||
09:15 5mTalk | Specification Synthesis with Constrained Horn Clauses PLDI Sumanth Prabhu TCS Research, Grigory Fedyukovich Florida State University, Kumar Madhukar TCS Research, Deepak D'Souza IISc Bangalore DOI | ||
09:20 5mTalk | Compiling Stan to Generative Probabilistic Languages and Extension to Deep Probabilistic Programming PLDI Guillaume Baudart Inria, Javier Burroni University of Massachusetts Amherst, Martin Hirzel IBM Research, Louis Mandel IBM Research, USA, Avraham Shinnar IBM Research DOI | ||
09:25 5mTalk | Sound Probabilistic Inference via Guide Types PLDI Di Wang Carnegie Mellon University, Jan Hoffmann Carnegie Mellon University, Thomas Reps University of Wisconsin DOI | ||
09:30 5mTalk | SPPL: Probabilistic Programming with Fast Exact Symbolic Inference PLDI Feras Saad Massachusetts Institute of Technology, Martin C. Rinard Massachusetts Institute of Technology, Vikash K. Mansinghka Massachusetts Institute of Technology DOI | ||
09:35 5mTalk | Quantitative Analysis of Assertion Violations in Probabilistic Programs PLDI Jinyi Wang Shanghai Jiao Tong University, Yican Sun Peking University, Hongfei Fu Shanghai Jiao Tong University, Krishnendu Chatterjee IST Austria, Amir Kafshdar Goharshady Hong Kong University of Science and Technology DOI |
21:00 - 21:40 | |||
21:00 5mTalk | DeepCuts: A Deep Learning Optimization Framework for Versatile GPU Workloads PLDI Wookeun Jung Seoul National University, Thanh Tuan Dao Seoul National University, Jaejin Lee Seoul National University DOI | ||
21:05 5mTalk | Provable Repair of Deep Neural Networks PLDI Matthew Sotoudeh University of California at Davis, Aditya V. Thakur University of California at Davis DOI Pre-print Media Attached | ||
21:10 5mTalk | DreamCoder: Bootstrapping Inductive Program Synthesis with Wake-Sleep Library Learning PLDI Kevin Ellis Cornell University, Lionel Wong Massachusetts Institute of Technology, Maxwell Nye Massachusetts Institute of Technology, Mathias Sablé-Meyer PSL University; Collège de France; NeuroSpin, Lucas Morales Massachusetts Institute of Technology, Luke Hewitt Massachusetts Institute of Technology, Luc Cary Massachusetts Institute of Technology, Armando Solar-Lezama Massachusetts Institute of Technology, Joshua B. Tenenbaum Massachusetts Institute of Technology DOI | ||
21:15 5mTalk | Specification Synthesis with Constrained Horn Clauses PLDI Sumanth Prabhu TCS Research, Grigory Fedyukovich Florida State University, Kumar Madhukar TCS Research, Deepak D'Souza IISc Bangalore DOI | ||
21:20 5mTalk | Compiling Stan to Generative Probabilistic Languages and Extension to Deep Probabilistic Programming PLDI Guillaume Baudart Inria, Javier Burroni University of Massachusetts Amherst, Martin Hirzel IBM Research, Louis Mandel IBM Research, USA, Avraham Shinnar IBM Research DOI | ||
21:25 5mTalk | Sound Probabilistic Inference via Guide Types PLDI Di Wang Carnegie Mellon University, Jan Hoffmann Carnegie Mellon University, Thomas Reps University of Wisconsin DOI | ||
21:30 5mTalk | SPPL: Probabilistic Programming with Fast Exact Symbolic Inference PLDI Feras Saad Massachusetts Institute of Technology, Martin C. Rinard Massachusetts Institute of Technology, Vikash K. Mansinghka Massachusetts Institute of Technology DOI | ||
21:35 5mTalk | Quantitative Analysis of Assertion Violations in Probabilistic Programs PLDI Jinyi Wang Shanghai Jiao Tong University, Yican Sun Peking University, Hongfei Fu Shanghai Jiao Tong University, Krishnendu Chatterjee IST Austria, Amir Kafshdar Goharshady Hong Kong University of Science and Technology DOI |