Efficient Auto-Tuning of Parallel Programs with Interdependent Tuning Parameters via Auto-Tuning Framework (ATF)

Rasch, Ari; Schulze, Richard; Steuwer, Michel; Gorlatch, Sergei

doi:10.1145/3427093

Cited by 18 publications

(7 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…MetaTune [35] improves auto-tuning parameter selection for systems like AutoTVM. In ATF [31], the authors look at more efficient auto-tuning techniques, with a focus on modeling when parameters have inter-dependencies among them.…”

Section: Related Workmentioning

confidence: 99%

Reusing Auto-Schedules for Efficient DNN Compilation

Gibson¹,

Cano²

2022

Preprint

View full text Add to dashboard Cite

Auto-scheduling is a process where a search algorithm automatically explores candidate schedules (program transformations) for a given tensor program on a given hardware platform to improve its performance. However this can be a very time consuming process, depending on the complexity of the tensor program, and capacity of the target device, with often many thousands of program variants being explored. To address this, in this paper we introduce and demonstrate the idea of tuning-reuse, a novel approach to identify and re-use auto-schedules between tensor programs. We demonstrate this concept using Deep Neural Networks (DNNs), taking sets of auto-schedules from pre-tuned DNNs, and using them to reduce the inference time of a new DNN. Given a set of pretuned schedules, tuning-reuse provides its maximum speedup in less time than auto-scheduling using the state-of-the-art Ansor auto-scheduler. On a set of widely used DNN models, we apply tuning-reuse and achieve maximum speedups between 1.16× and 4.76×, while outperforming Ansor when given limited tuning time.

show abstract

Section: Related Workmentioning

confidence: 99%

Reusing Auto-Schedules for Efficient DNN Compilation

Gibson¹,

Cano²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The last row is the results using autotuning. The best configuration in 200 evaluations has the tile size (50,128,256) for the large dataset and the tile size (64, 50, 256) for the extra-large dataset using GBRT. We observe that autotuning outperforms the other compiling methods to provide the smallest execution time for both datasets.…”

Section: Case Study: Syr2k With Multiple Loop Transformations (Loop T...mentioning

confidence: 99%

“…There exist two kinds of expressions of search space: vector space and tree space. Most of autotuning frameworks present the search space in a vector space, that is, a fixed number of parameter knobs, such as OpenTuner, 42 CLTune, 46 HalideTuner, 44 Orio, 41 KernelTuner, 48 AFT, 45,50 ytopt, 13,24 and so forth. The successor of HalideTuner 51 uses tree search to avoid the limitation of a vector search space, but uses Beam search to explore the space.…”

Section: Related Workmentioning

confidence: 99%

Autotuning PolyBench benchmarks with LLVM Clang/Polly loop optimization pragmas using Bayesian optimization

Kruse

Balaprakash

et al. 2021

Concurrency and Computation

View full text Add to dashboard Cite

We develop a ytopt autotuning framework that leverages Bayesian optimization to explore the parameter space search and compare four different supervised learning methods within Bayesian optimization and evaluate their effectiveness. We select six of the most complex PolyBench benchmarks and apply the newly developed LLVM Clang/Polly loop optimization pragmas to the benchmarks to optimize them. We then use the autotuning framework to optimize the pragma parameters to improve their performance. The experimental results show that our autotuning approach outperforms the other compiling methods to provide the smallest execution time for the benchmarks syr2k, 3mm, heat‐3d, lu, and covariance with two large datasets in 200 code evaluations for effectively searching the parameter spaces with up to 170,368 different configurations. We find that the Floyd–Warshall benchmark did not benefit from autotuning. To cope with this issue, we provide some compiler option solutions to improve the performance. Then we present loop autotuning without a user's knowledge using a simple mctree autotuning framework to further improve the performance of the Floyd–Warshall benchmark. We also extend the ytopt autotuning framework to tune a deep learning application.

show abstract

“…There exist two kinds of expressions of search space: vector space and tree space. Most of autotuning frameworks present the search space in a vector space, that is, a fixed number of parameter knobs, such as OpenTuner 42 , CLTune 46 , HalideTuner 44 , Orio 41 , KernelTuner 48 , AFT 45,50 , ytopt 13,24 , etc. The successor of HalideTuner 51 uses tree search to avoid the limitation of a vector search space, but uses Beam search to explore the space.…”

Section: Related Workmentioning

confidence: 99%

“…We classify the existing autotuning frameworks into four categories:1) enumerate all possible parameter configurations, reject invalid ones, and evaluate the valid ones 32 ; 2) enumerate only valid configurations 45,50 ; 3) sample from the set of possible configurations, and reject invalid ones 41,43,10 during the search; 4) sample only valid configurations and search over them 24 . Our ytopt autotuning framework belongs to Category 4 which overcomes the ineffectiveness of Category 3 by generating valid samples and addresses the limitations of Categories 1 and 2, where enumerating all possible configurations can be computationally expensive for large number of parameters.…”

Section: Related Workmentioning

confidence: 99%

Autotuning PolyBench Benchmarks with LLVM Clang/Polly Loop Optimization Pragmas Using Bayesian Optimization (extended version)

Wu¹,

Kruse²,

Balaprakash³

et al. 2021

Preprint

View full text Add to dashboard Cite

In this paper, we develop a ytopt autotuning framework that leverages Bayesian optimization to explore the parameter space search and compare four different supervised learning methods within Bayesian optimization and evaluate their effectiveness. We select six of the most complex PolyBench benchmarks and apply the newly developed LLVM Clang/Polly loop optimization pragmas to the benchmarks to optimize them.We then use the autotuning framework to optimize the pragma parameters to improve their performance. The experimental results show that our autotuning approach outperforms the other compiling methods to provide the smallest execution time for the benchmarks syr2k, 3mm, heat-3d, lu, and covariance with two large datasets in 200 code evaluations for effectively searching the parameter spaces with up to 170,368 different configurations. We find that the Floyd-Warshall benchmark did not benefit from autotuning because Polly uses heuristics to optimize the benchmark to make it run much slower. To cope with this issue, we provide some compiler option solutions to improve the performance. Then we present loop autotuning without a user's knowledge using a simple mctree autotuning framework to further improve the performance of the Floyd-Warshall benchmark. We also extend the ytopt autotuning framework to tune a deep learning application.

show abstract

Efficient Auto-Tuning of Parallel Programs with Interdependent Tuning Parameters via Auto-Tuning Framework (ATF)

Cited by 18 publications

References 53 publications

Reusing Auto-Schedules for Efficient DNN Compilation

Reusing Auto-Schedules for Efficient DNN Compilation

Autotuning PolyBench benchmarks with LLVM Clang/Polly loop optimization pragmas using Bayesian optimization

Autotuning PolyBench Benchmarks with LLVM Clang/Polly Loop Optimization Pragmas Using Bayesian Optimization (extended version)

Contact Info

Product

Resources

About