Automated architecture synthesis and application mapping for
ASIP based adaptable MPSoCs
Citation for published version (APA):
Diken, E., Jordans, R., Corvino, R., Jozwiak, L., & Lindwer, M. (2011). Automated architecture synthesis and application mapping for ASIP based adaptable MPSoCs. 135-138. Poster session presented at 7th International Summer School on Advanced Computer Architecture and Compilation for High-Performance and Embedded Systems (ACACES 2011), Fiuggi, Italy.
Document status and date: Published: 01/01/2011
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
providing details and we will investigate your claim.
Automated Architecture Synthesis
and Application Mapping for ASIP
based adaptable MPSoCs
1
Erkan Diken
∗, Roel Jordans
∗,
Rosilde Corvino
∗, Lech Jozwiak
∗,2,
Menno Lindwer
†∗ Eindhoven University of Technology, Eindhoven, The Netherlands †Silicon Hive B.V., Eindhoven, The Netherlands
ABSTRACT
Recent developments in modern embedded system technology have enabled the development of complex heterogeneous multiprocessor systems on single chips (MPSoCs) and created an up surge in high-performance and low-power embedded system design. This paper focuses on the auto-matic architecture synthesis and application restructuring and mapping for customizable ASIP-based MPSoCs. It briefly discusses the exploration methodology proposed for this purpose by the European research project ASAM of the ARTEMIS program.
KEYWORDS: embedded systems, heterogeneous multi-processor system-on-chip (MPSoC), ASIP
syn-thesis, design automation
1
Introduction
Recent developments in modern embedded system technology have enabled the develop-ment of complex heterogeneous multiprocessor systems on single chips (MPSoCs) and cre-ated an up surge in high-performance and low-power embedded system design. An embed-ded system serves a specific aim in a certain larger embedding system through repeatedly executing specific processes required by its application. An embedded system is specially designed to serve the execution of its specific task and needs to satisfy requirements related to attributes such as functional behavior, reaction speed or throughput, energy consump-tion, physical size, price, etc. The recent progress in nano-electronic technology and new applications of embedded systems fostered a situation in which increasingly complex and sophisticated embedded systems are required to fulfill real-time constraints with extremely tight energy, power, area and cost budgets. Moreover, many embedded systems are required to be highly flexible in order to enable hardware reuse among different product versions in
1The information presented in this paper is related to the European project ASAM that is executed in the
framework of the ARTEMIS program
reaction to market shifts, adherence to evolving standards or user requirements, and easy modification during development or even their field use. As a consequence, complex design and development challenges involving multi-objective MPSoC optimization and adequate resolution of many complex design trade-offs, have to be faced.
Our research addresses the development of heterogeneous MPSoCs based on config-urable and extensible application specific instruction-set processors (ASIPs). MPSoC design technology based on adaptable ASIPs provides an extremely flexible platform that is able to deliver high performance and low energy consumption at the same time which makes it applicable for a broad range of applications and is suitable for several implementation tech-nologies. However, despite a decade of research in the field of SoC architecture synthesis, an architecture synthesis methodology and tools for heterogeneous customizable MPSoCs based on adaptable ASIPs have not been built.
This paper focuses on the automatic architecture synthesis and application restructuring and mapping for customizable ASIP-based MPSoCs and discusses the exploration method-ology proposed for this purpose by the European research project ASAM of the ARTEMIS program.
2
Architecture Platform, Applications, Issues and Challenges
The architecture platform targeted in the ASAM project is a configurable and extensible heterogeneous multi-ASIP platform. In particular, the project targets the MPSoC platforms of its industrial partners involving generic ASIPs customizable through instantiation (con-figuration) and extension. Each ASIP forms a VLIW machine capable of executing parallel software with a single thread of control supporting both SIMD and MIMD processing. The numbers, kinds, and parameters of functional units, issue slots, register files, memories, in-terfaces, etc. from an existing library can be selected freely, but also new modules can be developed and exploited for a specific application.
Several such different ASIPs, each customized for a particular part of a complex appli-cation, can be interconnected with global memories and other sub-systems using a config-urable hierarchical interconnection network, and implemented on one chip together with possible hardware accelerators and other digital or analog sub-systems. Several such pow-erful ASIPs with up to 100 issue slots in total, each for 64-way vector processing, can be placed on a single chip implemented in 22nm CMOS technology, can deliver up to 1Tops/s, when operated at 400–600MHz, with a power consumption far below the upper limit of mobile devices.
The MPSoC design technology based on adaptable ASIPs is relevant for a very broad range of application domains (e.g. telecommunications, multi-media, medical, etc.) and for cross-domain convergence products.
Traditionally, embedded systems development involves largely disjoint processes, where the computation platform and the application software are developed largely independently by different teams using different tools. This leads to inefficiency, errors, and costly reitera-tions of the design process. Moreover, high-performance embedded applicareitera-tions require an application specific computation platform to satisfy the stringent requirements. Therefore, the computing platform and application software have to be developed largely in paral-lel. Unfortunately, the efficiency of the required parallel development technology is much too low currently, due to the lack of effective automated methods of industrial strength, and
weak interoperability of the architecture, algorithm, HW/SW, and hardware synthesis tools. While most of the existing methods and tools for ASIP construction are devoted to a single processor design, in a customizable multi-ASIP MPSoC its various ASIPs have to be cus-tomized in combination, and in a strict relation to the selection of the number of ASIPs, as well as to the scheduling and mapping of the application’s required computations on the particular ASIPs. The issues and challenges in the design of massively parallel heteroge-neous MPSoCs based on adaptable ASIPs for modern highly-demanding applications are discussed in more detail in [JL11].
3
Approach
The ASAM project aims to tackle the problems and challenges, as briefly explained in the previous section, by proposing an uniform design flow for efficient exploration of the archi-tecture alternatives and application mapping trade-offs. The ASAM uniform design flow consists of two main sub-tasks: Macro-architecture (MPSoC-level) synthesis, and micro-architecture (ASIP-level) synthesis. Coarsely speaking, the macro-micro-architecture synthesis ac-counts for the decision on the number and type of the ASIP processors, the task-level par-titioning of an application, and adequate global memory and interconnection structures, while the micro-architecture synthesis takes care of the actual ASIP synthesis. In contrast to the traditional approaches (separating these two tasks, as well as, hardware and software development) followed in the state-of-the-art methods, ASAM design flow proposes one co-herent and complete approach due to the strong interrelationships between the macro- and micro-architecture sub-tasks. To create feedback for both architecture design levels, a simu-lation and FPGA emusimu-lation prototyping environment will be used. Figure 1 demonstrates the high-level view of the ASAM design flow.
Realization of the proposed uniform design flow requires an unified synthesis and proto-typing environment that involves the collaborating processes of the multi-objective macro-architecture design space exploration (DSE) and construction of the multi-ASIP system-level architecture, and the multi-objective micro-architecture exploration and architecture cus-tomization of particular generic ASIPs. The exploration and synthesis decisions are based on the application and platform analysis and restructuring(for parallelism exploitation), it produces feedbacks to the DSE processes on physical characteristics of the proposed solu-tions (area, delay/throughput, power) and is refined against the results of the simulation and a FPGA-based emulation. The macro-architecture synthesis proposes to use a certain number of customizable ASIP cores and assigns them parts of the application to be exe-cuted. The micro-architecture synthesis customizes each of the ASIP cores together with its local memories to execute the assigned application part as effective and efficient as possible. After several exploration iterations an optimized MPSoC architecture is constructed.
4
Conclusion
Highly effective, efficient and flexible parallel systems for demanding embedded appli-cations usually involve small numbers of highly optimized powerful ASIPs. The ASAM project will facilitate the creation of such systems by developing a consistent highly effi-cient automatic synthesis flow from the algorithmic application specification down to its
Tech-aware solution analy. & evaluation micro-level abstraction Input requirements specification mapping SW synthesis A1 A2 A3 Application model abstraction Platform HW1 ASIP2 ASIP1 Network Selection ASIP synthesis P1 P2 P3 HW synthesis Library HW1 ASIP2 ASIP1 HW2 ASIP3 ASIP2 ASIP1 Network A1 A2 A3 model Application requirements macro-level SW synthesis Application software A1 A2 A3 Architecture refinement Decision models Application model Platform Platform model HW Platform
instance HW & SW Actual Synthesis FPGA/SoC SW Code
Simulation FPGA Emulation/impl.
Rapid Prototyping & Actual Implementation
HW
Techn. Independent
description
Platform refinement
Figure 1: High-level ASAM design flow
hardware/software implementation at the circuit/code level. The new coherent automated design environment will enable the system and algorithm designers to perform rapid ex-ploration of the high-level algorithm and architecture solution spaces, and in consequence, quickly develop high-quality designs. The ASAM webpage [web] provides more informa-tion about the research of the ASAM project and its up-to-date status overview.
Acknowledgement
The authors of this paper are indebted to all partners of the ASAM project.
References
[JL11] L. Jó´zwiak and M. Lindwer. Issues and challenges in development of massively-parallel heterogeneous mpsocs based on adaptable asips. In Parallel, Distributed and Network-Based Processing (PDP), 2011 19th Euromicro International Conference on, pages 483–487, 2 2011.