Exploiting Errors for Efficiency

Stanley-Marbell, Phillip; Alaghi, Armin; Carbin, Michael; Darulová, Eva; Dolecek, Lara; Gerstlauer, Andreas; Gillani, G. A.; Jevdjic, Djordje; Moreau, Thierry; Cacciotti, Mattia; Daglis, Alexandros; Jerger, Natalie Enright; Falsafi, Babak; Misailovíc, Saša; Sampson, Adrian; Zufferey, Damien

doi:10.1145/3394898

Cited by 38 publications

(5 citation statements)

References 195 publications

(141 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Error metrics were evaluated for a single adder accelerator, for five different configuration sets (T = 5). The adder variants considered for this evaluation were GeAr (8,1,4), GeAr(8,1,5), GeAr(8,2,2), GeAr(9,2,3) and GeAr (10,2,4). The configurations are shown in Figure 11.…”

Section: Comparison With State-of-the-art [23]mentioning

confidence: 99%

“…Approximate computing can be employed to gain benefits in terms of performance, energy and/or area, by relaxing the bounds of precise computing [1][2][3][4][5]. Recent investigations suggest that there are a number of computationally intensive applications that can tolerate approximation errors while still producing outputs that are of acceptable quality for the end-users [6][7][8][9][10].…”

Section: Introductionmentioning

confidence: 99%

“…Recent investigations suggest that there are a number of computationally intensive applications that can tolerate approximation errors while still producing outputs that are of acceptable quality for the end-users [6][7][8][9][10]. Specifically, applications such as image/video processing, multimedia, machine learning, big data analytics and recognition are intrinsically prone to noise and/or are resilient to error because of the perceptual limitations of the end-users and, hence, are suitable candidates for approximate computing [2,4,11].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

DAEM: A Data- and Application-Aware Error Analysis Methodology for Approximate Adders

Hanif,

Hafiz,

Shafique

2023

Information

View full text Add to dashboard Cite

Approximate adders are some of the fundamental arithmetic operators that are being employed in error-resilient applications, to achieve performance/energy/area gains. This improvement usually comes at the cost of some accuracy and, therefore, requires prior error analysis, to select an approximate adder variant that provides acceptable accuracy. Most of the state-of-the-art error analysis techniques for approximate adders assume input bits and operands to be independent of one another, while some also assume the operands to be uniformly distributed. In this paper, we analyze the impact of these assumptions on the accuracy of error estimation techniques, and we highlight the need to address these assumptions, to achieve better and more realistic quality estimates. Based on our analysis, we propose DAEM, a data- and application-aware error analysis methodology for approximate adders. Unlike existing error analysis models, we neither assume the adder operands to be uniformly distributed nor assume them to be independent. Specifically, we use 2D joint input probability mass functions (PMFs), populated using sample data, in order to incorporate the data and application knowledge in the analysis. These 2D joint input PMFs, along with 2D error maps of approximate adders, are used to estimate the error PMF of an adder network. The error PMF is then utilized to compute different error measures, such as the mean squared error (MSE) and mean error distance (MED). We evaluate the proposed error analysis methodology on audio and video processing applications, and we demonstrate that our methodology provides error estimates having a better correlation with the simulation results, as compared to the state-of-the-art techniques.

show abstract

Section: Comparison With State-of-the-art [23]mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

DAEM: A Data- and Application-Aware Error Analysis Methodology for Approximate Adders

Hanif,

Hafiz,

Shafique

2023

Information

View full text Add to dashboard Cite

show abstract

“…In these cases, it is acceptable to relax the output accuracy to a certain degree to increase power, area, or delay savings. The design techniques trading the accuracy of results for power, delay, or area are referred to as Approximate Computing (AxC) [1].…”

Section: Introductionmentioning

confidence: 99%

A Design Space Exploration of Power-efficient Gaussian Filter Architectures using Logical Optimization and Approximated Adders

Monteiro,

Seidel,

Güntzel

et al. 2023

JICS

View full text Add to dashboard Cite

Gaussian filtering is an important step in many image-processing applications because it reduces image noise.However, this step is also compute-intensive, so power-optimized hardware architectures are necessary to allow its adoption in embedded devices. This work presents a design space exploration of power-efficient Gaussian filter architectures.Differently from related work, this work shows the impacts of the design decisions on two target applications: the Canny Algorithm and an Automatic License Plate Recognition system.To explore the design space, the 3x3, 5x5, and 7x7kernels were logically refactored using Multiplierless Constant Multiplication and Common Sub-expression Exploration. The adders were approximated using the copy strategy to further reduce power consumption. Systematic experiments show the effects of the adopted strategies on power savings and quality of results compared to exact baselines using the two applications.The approximate strategy reached up to 51 dB of Peak Signal-to-Noise Ratio with power reductions of up to 48% in the best-case scenario of the standalone Gaussian filters. Also, the total power of the Gaussian filter in the Canny Algorithm can be reduced down to 34% while maintaining the precision of results between 57% and 90%. Finally, the proposed strategies reduce up to 64% of the Gaussian filter power consumption when adopted in the plate detection solution with a similar detection rate compared to the exact filter architecture.

show abstract

“…Mixed Precision is a branch of a more general class of techniques, known as Approximate Computing, which aim at trading off computation accuracy for other quality metrics, including performance and energy. Recent surveys [4,9] show that a significant number of tools have been developed to automatically analyze and transform codebases to exploit Approximate Computing.…”

Section: Introductionmentioning

confidence: 99%

Mixed Precision in Heterogeneous Parallel Computing Platforms via Delayed Code Analysis

Cattaneo,

Maggioli,

Magnani

et al. 2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Mixed Precision techniques have been successfully applied to improve the performance and energy efficiency of computation in embedded and high performance systems. However, few solutions have been proposed that address precision tuning of both GPGPU code and its corresponding CPU code, limiting the gains achievable by mixed precision. We propose an extension to the taffo precision tuning toolset that enables Mixed Precision across the space of floating and fixed point data types on GPGPUs, leveraging static analysis and providing seamless interface adaptation between host and GPGPU kernel code. The proposed tool achieves speedups exceeding 2× by exploiting the optimization of both kernel and host code.

show abstract

Exploiting Errors for Efficiency

Cited by 38 publications

References 195 publications

DAEM: A Data- and Application-Aware Error Analysis Methodology for Approximate Adders

DAEM: A Data- and Application-Aware Error Analysis Methodology for Approximate Adders

A Design Space Exploration of Power-efficient Gaussian Filter Architectures using Logical Optimization and Approximated Adders

Mixed Precision in Heterogeneous Parallel Computing Platforms via Delayed Code Analysis

Contact Info

Product

Resources

About