Optimising Performance of Quadrature Methods with Reduced Precision

Tse, Anson H.T.; Chow, Gary C.T.; Jin, Qiwei; Thomas, David B.; Luk, Wayne

doi:10.1007/978-3-642-28365-9_21

Cited by 12 publications

(16 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We use our approach to optimize throughput of quadrature method-based financial application with varied precision [6] implemented on the Maxeler platform. We show how analytical models can aid MLO.…”

Section: Discussionmentioning

confidence: 99%

“…In [6] an optimization model (traditional step 3) is proposed along with an associated algorithm (traditional step 4) that improves design throughput (integrations per second φ int ) by varying the number representation in quadrature based applications. The application is parameterized with density factor d f of integral estimation and the number mantissa width m w .…”

Section: Discussionmentioning

confidence: 99%

“…Varying d f only involves software execution, as long as the bit-stream has already been evaluated. As a reference design we use m w = 53, d f = 32 and fix FPGA frequency to 100 MHz, allowing us to meet timings (In [6] d f = 20 is used to produce reference values). We follow with either a two dimensional optimization over m w and d f , or a three dimensional including cores.…”

Section: Discussionmentioning

confidence: 99%

“…• We use our approach to optimize throughput of a quadrature based financial application with varied precision [6]. We show how MLO can significantly decrease benchmark evaluations (up to 85%) and can be used in combination with analytical models (Section III).…”

Section: Introductionmentioning

confidence: 99%

“…We classify the parameter space using SVMs to identify designs that would fail constraints; over-map on resources, produce inaccurate results or other. Our contributions:• The proposed approach has two new aspects: (a) surrogate models of application benchmarks (b) automated application optimization scheme based on the surrogate models and MLO (Section II).• We use our approach to optimize throughput of a quadrature based financial application with varied precision [6]. We show how MLO can significantly decrease benchmark evaluations (up to 85%) and can be used in combination with analytical models (Section III).…”

mentioning

confidence: 99%

See 4 more Smart Citations

Parametric reconfigurable designs with Machine Learning Optimizer

Kurek

Luk

2012

2012 International Conference on Field-Programmable Technology

Self Cite

View full text Add to dashboard Cite

Abstract-We investigate the use of meta-heuristics and machine learning to automate reconfigurable application parameter optimization. The traditional approach involves two steps: (a) analyzing the application in order to create models and tools for exploration of the parameter space, and (b) exploring the parameter space using such tools. The proposed approach, called the Machine Learning Optimizer (MLO), involves a Particle Swarm Optimization (PSO) algorithm with an underlying surrogate fitness function model based on Gaussian Process (GP) and Support Vector Machines (SVMs). We present a case study of a quadrature based financial application with varied precision. We evaluate our approach by comparing the amount of benchmark evaluations and bit-stream generations when using MLO and when using the traditional approach. I. INTRODUCTIONThe optimization of heterogeneous computing applications often requires substantial effort from the designer who has to analyze the application, create models and benchmarks and subsequently use them to optimize the application. One could try to employ exhaustive search of the application parameter space to carry out optimization yet it is unrealistic since benchmark evaluations involve bit-stream generation and code execution which takes hours of computing time. Recently it has been shown useful to use surrogate models combined with fitness functions for computationally expensive optimization problems in various fields [1], [2], [3], [4], [5]. As these models are orders of magnitude cheaper, they can substantially decrease optimization cost thus allowing for an automated approach. This is the motivation behind MLO which we apply to non-linear and multi-modal problem of heterogeneous application parameter optimization. We use GPs to model performance of the design like execution time or throughput, while searching for the global optimum using PSO. We classify the parameter space using SVMs to identify designs that would fail constraints; over-map on resources, produce inaccurate results or other. Our contributions:• The proposed approach has two new aspects: (a) surrogate models of application benchmarks (b) automated application optimization scheme based on the surrogate models and MLO (Section II).• We use our approach to optimize throughput of a quadrature based financial application with varied precision [6]. We show how MLO can significantly decrease benchmark evaluations (up to 85%) and can be used in combination with analytical models (Section III).

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations

Parametric reconfigurable designs with Machine Learning Optimizer

Kurek

Luk

2012

2012 International Conference on Field-Programmable Technology

Self Cite

View full text Add to dashboard Cite

show abstract

A study of reduced numerical precision to make superparameterization more competitive using a hardware emulator in the OpenIFS model

Dueben

Subramanian

Dawson

et al. 2017

J Adv Model Earth Syst

View full text Add to dashboard Cite

The use of reduced numerical precision to reduce computing costs for the cloud resolving model of superparameterized simulations of the atmosphere is investigated. An approach to identify the optimal level of precision for many different model components is presented, and a detailed analysis of precision is performed. This is nontrivial for a complex model that shows chaotic behavior such as the cloud resolving model in this paper. It is shown not only that numerical precision can be reduced significantly but also that the results of the reduced precision analysis provide valuable information for the quantification of model uncertainty for individual model components. The precision analysis is also used to identify model parts that are of less importance thus enabling a reduction of model complexity. It is shown that the precision analysis can be used to improve model efficiency for both simulations in double precision and in reduced precision. Model simulations are performed with a superparameterized single‐column model version of the OpenIFS model that is forced by observational data sets. A software emulator was used to mimic the use of reduced precision floating point arithmetic in simulations.

show abstract