Write a Blog >>
PPoPP 2021
Sat 27 February - Wed 3 March 2021
Tue 2 Mar 2021 11:50 - 12:10 - Session 5. Auto Tuning Chair(s): Saeed Maleki

Manually optimizing the tradeoffs between accuracy, performance and energy for resource-intensive applications with flexible accuracy or precision requirements is extremely difficult. We present ApproxTuner, an automatic framework for accuracy-aware optimization of tensor-based applications while requiring only high-level end-to-end quality specifications. ApproxTuner implements and manages approximations in algorithms, system software, and hardware.

The key contribution in ApproxTuner is a novel three-phase approach to approximation-tuning that consists of development-time, install-time, and run-time phases. Our approach decouples tuning of hardware-independent and hardware-specific approximations, thus providing retargetability across devices. To enable efficient autotuning of approximation choices, we present a novel accuracy-aware tuning technique called predictive approximation-tuning, which speeds up autotuning by analytically predicting the accuracy impacts of approximations.

We evaluate ApproxTuner across 10 convolutional neural networks (CNNs) and a combined CNN and image processing benchmark. For the evaluated CNNs, using only hardware-independent approximation choices we achieve a mean speedup of 2.1x (max 2.7x) on a GPU, and 1.3x mean speedup (max 1.9x) on the CPU, while staying within 1 percentage point of inference accuracy loss. For two different accuracy-prediction models, ApproxTuner speeds up tuning by 12.8x and 20.4x compared to conventional empirical tuning while achieving comparable benefits.

Conference Day
Tue 2 Mar

Displayed time zone: Eastern Time (US & Canada) change

11:10 - 12:10
Session 5. Auto TuningMain Conference
Chair(s): Saeed MalekiMicrosoft Research
11:10
20m
Talk
GPTune: Multitask Learning for Autotuning Exascale Applications
Main Conference
Yang Liu, Wissam M. Sid-LakhdarLawrence Berkeley National Laboratory, Osni MarquesLawrence Berkeley National Laboratory, Xinran ZhuCornell University, Chang MengEmory University, James W. DemmelUC Berkeley, Xiaoye S. LiLawrence Berkeley National Laboratory
Link to publication
11:30
20m
Talk
I/O Lower Bounds for Auto-tuning of Convolutions in CNNs
Main Conference
Xiaoyang ZhangInstitute of Computing Technology, Chinese Academy of Sciences, Junmin XiaoInstitute of Computing Technology, Chinese Academy of Sciences, Guangming TanInstitute of Computing Technology, Chinese Academy of Sciences
Link to publication
11:50
20m
Talk
ApproxTuner: A Compiler and Runtime System for Adaptive Approximations
Main Conference
Hashim SharifUniversity of Illinois at Urbana Champaign, Yifan ZhaoUniversity of Illinois at Urbana Champaign, Maria KotsifakouRuntime Verification, Inc., Akash KothariUniversity of Illinois at Urbana Champaign, Ben SchreiberUniversity of Illinois at Urbana Champaign, Elizabeth WangUniversity of Illinois at Urbana Champaign, Yasmin SaritaCornell University, Nathan ZhaoUniversity of Illinois at Urbana-Champaign, Keyur JoshiUniversity of Illinois at Urbana-Champaign, Vikram S. AdveUniversity of Illinois at Urbana-Champaign, Sasa MisailovicUniversity of Illinois at Urbana-Champaign, Sarita AdveUniversity of Illinois at Urbana-Champaign
Link to publication