Automatically Exploring Tradeoffs Between Software Output Fidelity and Energy Costs

Dorn, Jonathan; Lacomis, Jeremy; Weimer, Westley; Forrest, Stephanie

doi:10.1109/tse.2017.2775634

Cited by 19 publications

(22 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other works [18,63] constrain EC's search space for improved energy consumption of Java programs by asking users to provide predefined locations or equivalent functions or class implementations. These, and several subsequent works [16,29], demonstrate the potential for stochastic search methods such as EC to find machine-or architecture-specific optimizations that improve a program's performance or energy efficiency. However, these methods are not yet mature or carefully analyzed.…”

Section: Evolutionary Computation (Ec)mentioning

confidence: 88%

Gevo

Liou

Wang

Forrest

et al. 2020

ACM Trans. Archit. Code Optim.

Self Cite

View full text Add to dashboard Cite

GPUs are a key enabler of the revolution in machine learning and high-performance computing, functioning as de facto co-processors to accelerate large-scale computation. As the programming stack and tool support have matured, GPUs have also become accessible to programmers, who may lack detailed knowledge of the underlying architecture and fail to fully leverage the GPU's computation power. GEVO (Gpu optimization using EVOlutionary computation) is a tool for automatically discovering optimization opportunities and tuning the performance of GPU kernels in the LLVM representation. GEVO uses population-based search to find edits to GPU code compiled to LLVM-IR and improves performance on desired criteria while retaining required functionality. We demonstrate that GEVO improves the execution time of general-purpose GPU programs and machine learning (ML) models on NVIDIA Tesla P100. For the Rodinia benchmarks, GEVO improves GPU kernel runtime performance by an average of 49.48% and by as much as 412% over the fully compiler-optimized baseline. If kernel output accuracy is relaxed to tolerate up to 1% error, GEVO can find kernel variants that outperform the baseline by an average of 51.08%. For the ML workloads, GEVO achieves kernel performance improvement for SVM on the MNIST handwriting recognition (3.24×) and the a9a income prediction (2.93×) datasets with no loss of model accuracy. GEVO achieves 1.79× kernel performance improvement on image classification using ResNet18/CIFAR-10, with less than 1% model accuracy reduction.

show abstract

Section: Evolutionary Computation (Ec)mentioning

confidence: 88%

Gevo

Liou

Wang

Forrest

et al. 2020

ACM Trans. Archit. Code Optim.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Other GI work on non-functional fitness have targeted for example memory [5] or energy [6] usage. We chose to focus on software speed only, at the risk of less general conclusions, because of its universal relevance and to better control our observations on a single type of scenario: the most common [1].…”

Section: B Fitnessmentioning

confidence: 99%

“…Efficiency of alternative GenProg components (representation, mutation, crossover) have been studied in several studies [22], [23]. There has been an exponential increase in the number of publications on GI in recent years, improving various GI components, and expanding to new programming languages and application domains ( [6], [17], [18], and many others). Very few though compare just the search component.…”

mentioning

confidence: 99%

Empirical Comparison of Search Heuristics for Genetic Improvement of Software

Blot

Petke

2021

IEEE Trans. Evol. Computat.

View full text Add to dashboard Cite

Genetic improvement uses automated search to improve existing software. It has been successfully used to optimise various program properties, such as runtime or energy consumption, as well as for the purpose of bug fixing. Genetic improvement typically navigates a space of thousands of patches in search for the program mutation that best improves the desired software property. While genetic programming has been dominantly used as the search strategy, more recently other search strategies, such as local search, have been tried. It is, however, still unclear which strategy is the most effective and efficient. In this paper, we conduct an in-depth empirical comparison of a total of 18 search processes using a set of 8 improvement scenarios. Additionally, we also provide new genetic improvement benchmarks and we report on new software patches found. Our results show that, overall, local search approaches achieve better effectiveness and efficiency than genetic programming approaches. Moreover, improvements were found in all scenarios (between 15% and 68%).

show abstract

“…The second category represents frameworks that aim to apply code or binary transformations to introduce the possibility of exploiting accuracy-throughput tradeoffs. Quick-Step [31], Paraprox [43], and PowerGAUGE [12] are some examples in this category. The main focus of these works is on how to expose tradeoffs by introducing softwareknobs.…”

Section: State-of-the-artmentioning

confidence: 99%

On-line Application Autotuning Exploiting Ensemble Models

Martinovic,

Gadioli,

Palermo

et al. 2019

Preprint

View full text Add to dashboard Cite

Application autotuning is a promising path investigated in literature to improve computation efficiency. In this context, the end-users define high-level requirements and an autonomic manager is able to identify and seize optimization opportunities by leveraging trade-offs between extra-functional properties of interest, such as execution time, power consumption or quality of results. The relationship between an application configuration and the extra-functional properties might depend on the underlying architecture, on the system workload and on features of the current input. For these reasons, autotuning frameworks rely on application's knowledge to drive the adaptation strategies. The autotuning task is typically done offline because having it in production requires significant effort to reduce its overhead. In this paper, we enhance a dynamic autotuning framework with a module for learning the application knowledge during the production phase, in a distributed fashion. We leverage two strategies to limit the overhead introduced at the production phase. On one hand, we use a scalable infrastructure capable of leveraging the parallelism of the underlying platform. On the other hand, we use ensemble models to speed up the predictive capabilities, while iteratively gathering production data. Experimental results on synthetic applications and on a use case show how the proposed approach is able to learn the application knowledge, by exploring a small fraction of the design space.

show abstract

Automatically Exploring Tradeoffs Between Software Output Fidelity and Energy Costs

Cited by 19 publications

References 51 publications

Gevo

Gevo

Empirical Comparison of Search Heuristics for Genetic Improvement of Software

On-line Application Autotuning Exploiting Ensemble Models

Contact Info

Product

Resources

About