Multidimensional Pareto Frontiers Intersection Determination and Processor Optimization Case Study

Multidimensional Pareto Frontiers Intersection Determination and Processor Optimization Case Study

Stať ve sborníku v databázi WoS či Scopus

Almost every today's electronic devices are equipped with a processor. Different applications require and depend on different properties of a processor. For example, the fast growing field of Internet of Things depends on a long operation time of the devices when powered with batteries. Using a general purpose processors has proved ineffective which led to growing usage of Application-Specific Instruction-Set processors (ASIPs) which can be optimized to specific applications using different modifications of their properties (such as the number of registers, cache sizes, instruction set modifications, etc.). A suitable processor configuration can be hand-picked by a designer or by an automatic tool. Such a tool was developed in our previous research. It is able to find a set of Pareto-optimal processor configurations for a specific application which can be a significant help in a device design. The cost of the design process can be cut significantly when a processor is used in multiple designs. The goal of this paper is to introduce a tool able to find a suitable processor configuration for multiple application by constructing a compromise Pareto-optimal frontier of a processor configurations. The paper describes this problem on a theoretical level as well as it introduces a practical implementation and experimental evaluation of constructing a compromise Pareto frontier of a processor configurations for a set of applications. The experiments are based on a parameterizable RISC-V processor.

Almost every today's electronic devices are equipped with a processor. Different applications require and depend on different properties of a processor. For example, the fast growing field of Internet of Things depends on a long operation time of the devices when powered with batteries. Using a general purpose processors has proved ineffective which led to growing usage of Application-Specific Instruction-Set processors (ASIPs) which can be optimized to specific applications using different modifications of their properties (such as the number of registers, cache sizes, instruction set modifications, etc.). A suitable processor configuration can be hand-picked by a designer or by an automatic tool. Such a tool was developed in our previous research. It is able to find a set of Pareto-optimal processor configurations for a specific application which can be a significant help in a device design. The cost of the design process can be cut significantly when a processor is used in multiple designs. The goal of this paper is to introduce a tool able to find a suitable processor configuration for multiple application by constructing a compromise Pareto-optimal frontier of a processor configurations. The paper describes this problem on a theoretical level as well as it introduces a practical implementation and experimental evaluation of constructing a compromise Pareto frontier of a processor configurations for a set of applications. The experiments are based on a parameterizable RISC-V processor.

Pareto optimization, Pareto frontier, processor optimization, ASIP.

Pareto optimization, Pareto frontier, processor optimization, ASIP.

PODIVÍNSKÝ, J.; ČEKAN, O.; KRČMA, M.; BURGET, R.; HRUŠKA, T.; KOTÁSEK, Z.

2020

28.08.2019

Institute of Electrical and Electronics Engineers

Kalithea

978-1-7281-2861-0

Proceedings of the 2019 22nd Euromicro Conference on Digital System Design

597

600

4

https://www.fit.vut.cz/research/publication/11967/

http://hdl.handle.net/

dsd2019_podivinsky