An input-centric paradigm for program dynamic optimizations

Tian, Kai; Jiang, Yunlian; Zhang, Eddy Z.; Shen, Xipeng

doi:10.1145/1869459.1869471

Cited by 30 publications

(28 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Software multi-versioning was proposed to reduce the overhead of code instrumentation [4,8,10,14,16,30], for checking program correctness [14], for loop parallelization, for automatic, speculative optimizations [7,19,29], or to optimize the execution for different inputs [41,46]. The technique periodically switches between highoverhead instrumenting versions and more efficient original or optimized versions [4,8,10,16].…”

Section: Related Workmentioning

confidence: 99%

Multiversioned decoupled access-execute: the key to energy-efficient compilation of general-purpose programs

Koukos

Ekemark

Zacharopoulos

et al. 2016

Proceedings of the 25th International Conference on Compiler Construction

View full text Add to dashboard Cite

Computer architecture design faces an era of great challenges in an attempt to simultaneously improve performance and energy efficiency. Previous hardware techniques for energy management become severely limited, and thus, compilers play an essential role in matching the software to the more restricted hardware capabilities. One promising approach is software decoupled access-execute (DAE), in which the compiler transforms the code into coarsegrain phases that are well-matched to the Dynamic Voltage and Frequency Scaling (DVFS) capabilities of the hardware. While this method is proved efficient for statically analyzable codes, generalpurpose applications pose significant challenges due to pointer aliasing, complex control flow and unknown runtime events. We propose a universal compile-time method to decouple generalpurpose applications, using simple but efficient heuristics. Our solutions overcome the challenges of complex code and show that automatic decoupled execution significantly reduces the energy expenditure of irregular or memory-bound applications and even yields slight performance boosts. Overall, our technique achieves over 20% on average energy-delay-product (EDP) improvements (energy over 15% and performance over 5%) across 14 benchmarks from SPEC CPU 2006 and Parboil benchmark suites, with peak EDP improvements surpassing 70%.

show abstract

Section: Related Workmentioning

confidence: 99%

Multiversioned decoupled access-execute: the key to energy-efficient compilation of general-purpose programs

Koukos

Ekemark

Zacharopoulos

et al. 2016

Proceedings of the 25th International Conference on Compiler Construction

View full text Add to dashboard Cite

show abstract

“…They concentrate on some specific library functions (e.g., FFT, sorting) while the algorithmic choices in these studies are limited. Input-centric program optimization [42] showed the benefits in enhancing Just-In-Time compilation. Jung and others have considered inputs when selecting the appropriate data structures to use [30].…”

Section: Related Workmentioning

confidence: 99%

“…For many problems, no single optimized program exists which can match the performance of a collection of optimized programs autotuned for different subsets of the input space. A common solution is to search for good optimizations on every training input, based on which, it builds a machine learning model that predicts the best optimization to use in a new run according to the features of the new input [22,30,33,39,42].…”

Section: Introductionmentioning

confidence: 99%

Autotuning algorithmic choice for input sensitivity

et al. 2015

Self Cite

View full text Add to dashboard Cite

A daunting challenge faced by program performance autotuning is input sensitivity, where the best autotuned configuration may vary with different input sets. This paper presents a novel two-level input learning algorithm to tackle the challenge for an important class of autotuning problems, algorithmic autotuning. The new approach uses a two-level input clustering method to automatically refine input grouping, feature selection, and classifier construction. Its design solves a series of open issues that are particularly essential to algorithmic autotuning, including the enormous optimization space, complex influence by deep input features, high cost in feature extraction, and variable accuracy of algorithmic choices. Experimental results show that the new solution yields up to a 3x speedup over using a single configuration for all inputs, and a 34x speedup over a traditional one-level method for addressing input sensitivity in program optimizations.

show abstract

“…Previous work (e.g. [3]) on selecting good features for input aware optimizations can be plugged into our framework with appropriate modifications. We suggest if the bitwidths of variables vary in a highly random manner, it could be advantageous to use learning algorithms such as Gaussian regression.…”

Section: A Training Phasementioning

confidence: 99%

Model-based input-adaptive vectorization

Sundararajah

Jayasena

2016

2016 Moratuwa Engineering Research Conference (MERCon)

View full text Add to dashboard Cite

In a program, not all the bits of a variable are always used during execution. Identifying the minimum number of bits necessary to represent a variable in a program can potentially provide optimization opportunities. Providing the knowledge of bitwidths to a compilation and execution framework will be advantageous if it could use that information to optimize the execution of the program, for instance, being able to select instructions for SIMD vectorization. This paper introduces a framework to exploit the potential vectorizations hidden in a program which is not exposed during static compilation time. Our framework unlocks instruction level data parallelism by using the bitwidths of array like variables that depend on runtime input. Our framework shows a maximum achievable performance gain of 37% and a mean achievable performance gain of 11% against the ICC compiler on our micro benchmark suite.

show abstract

An input-centric paradigm for program dynamic optimizations

Cited by 30 publications

References 38 publications

Multiversioned decoupled access-execute: the key to energy-efficient compilation of general-purpose programs

Multiversioned decoupled access-execute: the key to energy-efficient compilation of general-purpose programs

Autotuning algorithmic choice for input sensitivity

Model-based input-adaptive vectorization

Contact Info

Product

Resources

About