Changeset 97 for anr/section-2.tex


Ignore:
Timestamp:
Feb 5, 2010, 10:16:22 PM (15 years ago)
Author:
alain
Message:

Alain : refonte de la section 2

File:
1 edited

Legend:

Unmodified
Added
Removed

  • anr/section-2.tex

    r69 r97  
    1 The emerging complex and integrated heterogeneous embedded system platforms require
    2 adequate design methods to efficiently model, explore, analyze and design the ever complex software
    3 and hardware architectures. In order to rapidly meet the increasing performance requirements and a pressure
    4 to lower development cost and shorten time-to-market, future embedded systems suppliers
    5 will have to adopt new design methodologies and flows in order to keep pace with the increasing
    6 complexity of design problems. Such methods, addressing these challenges starting from high levels of
    7 abstraction, will have to perform large solution space explorations both for software and (possibly
    8 reconfigurable) hardware, reducing the design effort and offering a high predictability of results
    9 with respect to cost and performance objectives.
    10 \\
    11 Current design methodologies provide quite low-level abstraction capabilities. However in a few years
    12 from now, tens of programmable processors will be embedded in an IC with more than 100M
    13 transistors, therefore adding to the complexity of the problem of designing such systems.
    14 Taking into account that the complexity of the software part is increasing at an even
    15 faster rate, current solutions for design space exploration, mainly manually based, by no
    16 means do supply an adequate efficiency.
    17 Consequently, there is an urgent need to leverage system level
    18 exploration through the use of a high-level specification of the application and an early design
    19 space exploration step. The first system oriented approaches are appearing, among which those
    20 based on C/C++ and SystemC are the most popular. Such approaches can take place before and/or after
    21 the co-design or architecture refinement steps and target the design space pruning in order to fully
    22 exploit potential solutions that meet design and application constraints (power, latency,
    23 throughput) within the design and market timeframe.
    24 \\
    25 Thus, new system-level design flows need to be developed, enabling the exploration of an application
    26 independently of the implementation, almost at the beginning of the design process.
    27 A fundamental element of this evolution is the definition of abstraction layers that should allow the
    28 performance driven re-use of software and hardware components at the system level.
    29 In this context, COACH will combine modeling and estimation methods and compilers and
    30 design space exploration techniques. This approach will be a radical innovation in
    31 embedded system design methodology.
    32 \\
    33 The reason is that the COACH framework is applied before high-level design tools in the embedded
    34 systems design flow. In that way, it will make possible a real and efficiently combined
    35 exploitation of high-level synthesis tools, parallelizing approaches and compilers, already
    36 available on the market. These tools and approaches are not yet massively adopted, precisely
    37 because this preliminary design step is missing. COACH will indeed permit (i) to predict and
    38 control implementation optimizations, (ii) to target multiple implementation technologies
    39 (and thus the associated tools) from a unique specification and (iii) to efficiently integrate high
    40 and low-level design tools in a unique seamless design flow.
    41 \\
    42 The performance estimation methods combined with the design space exploration techniques will
    43 finally allow the design process to start from system level specification and automatically explore the
    44 potential architectures in order to find out the optimal implementation in a shorter design time and at
    45 a lower global cost.
     1The first objective of COACH is to provide SMEs (Small and Medium Enterprises) an open-source framework to
     2design embedded system on FPGA devices.
     3
     4Due to the exploding fabrication costs, the ASIC technology (Application Specific Integrated Circuit)
     5is not an option for most SMEs. Fortunately, the new FPGA (Field Programmable Gate Array) components,
     6such as the Virtex5 family from Xilinx, or the Stratix4 family from Altera can implement a complete
     7multi-processor architecture on a single chip.
     8
     9%But the design of a SoC (System on Chip) or MPSoC (Multi-Processor System on Chip) is a complex
     10%task, requiring adequate design methods to efficiently model, explore, and analyze the
     11%interactions between the software application and the hardware architectures. Moreover, most SMEs do not have
     12%in-home expertise in the field of hardware design or VHDL/Verilog modeling.
     13%In order to meet the increasing performance requirements, to decrease the development cost, and to
     14%shorten the time-to-market, they need new design methodologies.
     15
     16%Current design methodologies provide quite low-level abstraction capabilities, and
     17%there is an urgent need to leverage system level exploration through the use of a high-level
     18%specification of the application and  design space exploration tools.
     19
     20%The first system oriented approaches are appearing, among which those
     21%based on C/C++ and SystemC are the most popular, but few of them are specifically targetting FPGAs.
     22
     23The COACH project will leverage on the expertise gained in the field of virtual prototyping
     24with the SoCLib platform, to propose a new design flow based on a small number of architectural templates.
     25An architectural template is a generic, parametrized architecture, relying on a predefined library
     26of IP cores.
     27Besides using a specific collection of general purpose IP cores (such as processors cores,
     28embedded memory controllers, system bus controllers, I/O and peripheral controllers), each architectural
     29template can be enriched by dedicated hardware coprocessors, obtained by high level synthesis (HLS) tools.
     30During this project, the COACH partners will develop three different architectural templates:
     31
     32\begin{enumerate}
     33\item An \altera architectural template based on the \altera IP core library and the AVALON system bus.
     34\item A \xilinx architectural template based on the Xlinx IP core library and the OPB system bus.
     35\item A Neutral architectural template based on the SoCLib IP core library and the VCI/OCP communication infrastructure.
     36\end{enumerate}
     37
     38The proposed design flow starts from a high level description of the application, specified as a set of
     39parallel tasks written in C, without any assumption on the hardware or software implementation
     40of these tasks. It let the system
     41designer in charge of expessing the coarse grain parallelism of the application, gives the designer
     42the possibility to explore various mapping of the application on the selected template architecture,
     43and offers a high predictability of results with respect to cost and performance objectives.
     44
     45When this interactive, system level, design space exploration is completed (converging to
     46a specific mapping on a specific version of the selected architectural template), the rest of the flow
     47is fully automated: The synthesisable VHDL models for the various hardware components, as well as the binary
     48code for the software running on the embedded processors, and the bit-stream to program the the target FPGA
     49will be automatically generated by the COACH tools.
     50
    4651\par
    47 To get an efficient embedded system, the system designer has to take into account
    48 application characteristics when it chooses one of the available technologies.
    49 This choice is not easy and in most cases the designer has to try different
    50 technologies to retain the most adapted one.
    51 \\
    52 The first objective of COACH is to provide an open-source framework to
    53 design embedded system on FPGA devices.
    54 The COACH framework allows the designer to explore various software/hardware
    55 partitions of the target application, to run timing and functional
    56 simulations and to generate automatically both the software and the
    57 synthesizable description of the hardware.
    58 The main topics of the project are:
    59 \begin{itemize}
    60 \item
    61 \textbf{Design space exploration}: It consists in analysing the application running
    62 on FPGA, defining the target technology (SoC, MPSoC, ASIP, ...) and
    63 hardware/software partitioning of tasks depending on technology choice.
    64 This exploration is driven basically by throughput, latency and power
    65 consumption criteria.
    66 \item
    67 \textbf{Micro-architectural exploration}: When hardware components are required, the
    68 HLS tools of the framework generate them automatically. At this stage the
    69 framework provides various HLS tools that allow the micro-architectural space
    70 design exploration. The exploration criteria also are throughput, latency
    71 and power consumption.
    72 At this stage, preliminary source-level transformations will be
    73 required to improve the efficiency of the target component.
    74 For instance, one may transform a loop nest to expose parallelism,
    75 or shrink an array to promote it to a register or reduce a memory footprint.
     52The strength of the COACH approach is the strong integration of the high-level synthesis tools
     53in a plat-form based design flow supporting virtual prototyping and design space exploration.
     54Most building blocks already exist (resulting from previous projects): the GAUT
     55or UGH synthesis tools, the MutekH or DNA embedded operating systems, the ASIP technology,
     56the DSX exploration tool, the MWMR hardware/software communication middleware, the BEE parallelisation tool,
     57as well as the SoCLib library of systemC simulation models. They must now be integrated in
     58a consistent design flow.
     59%The five academic laboratories worked very closely during more than one year (one monthly meeting
     60%in Paris from january 2009 to february 2010, to analyse the issues of interfacing and integrating
     61%those various technologies, and to define the detailed architecture of the proposed design flow.
     62\par
    7663
    77 \item
    78 \textbf{Performance measurement}: For each point in the design space,
    79 figures of merit are available such as throughput, latency, power
    80 consumption, area, memory allocation and data locality. They are evaluated
    81 using virtual prototyping, estimation or analyzing methodologies.
    82 \item
    83 \textbf{Targeted hardware technology}: The COACH description of a system is
    84 independent of the FPGA family.  Every point of the design
    85 space can be implemented on any FPGA having the required resources.
    86 Basically, COACH handles both Altera and Xilinx FPGA families.
    87 \end{itemize}
    88 As an extension of embedded system design, COACH deals also with High
    89 Performance Computing (HPC).
    90 In HPC, the kind of targeted application is an existing one running on a PC.
    91 The COACH framework helps designer to accelerate it by migrating critical parts into a
    92 SoC implemented on an FPGA plugged to the PC bus.
    93 \par
    94 COACH is the result of the will of several laboratories to unify their knowhow
    95 and skills in the following domains: Operating system and hardware
    96 communication (\tima, \upmc), SoC and MPSoC (\upmc and \tima), ASIP (\irisa) and
    97 HLS (\upmc, \ubs) and compilation (\irisa, \lip).
    98 The project objective is to integrate these various domains into a unique
    99 free framework (licence ...) masking as much as possible these domains and
    100 its different tools to the user.
     64In summary, the COACH project is clearly oriented toward industry, even if most technology building blocks
     65have been previously developed by academic laboratories.
    10166
     67
     68%Finally, the key points of the proposed design flow are :
     69%\begin{itemize}
     70%\item
     71%\textbf{System level exploration}: The application coarse grain parallelism
     72%is explicitely described as a Tasks and Communication Graph (TCG).
     73%A template architecture is selected, and the performances are evaluated
     74%on various variant of this architecture using the SoCLib virtual protyping
     75%environment. This result in a specific hardware/software partitioning. 
     76%This system level exploration is fully controlled by the system designer, and is driven
     77%by cost, throughput, latency and power consumption criteria.
     78%
     79%\item
     80%\textbf{High Level Synthesis}: When dedicated hardware coprocessors have been
     81%identified as mandatory, they will be generated by the high level synthesis (HLS) tools.
     82%The Coach framework will integrate various HLS tools, supporting the micro-architectural space
     83%design exploration. Here again, the exploration criteria are cost, throughput, latency
     84%and power consumption.
     85%At this stage, preliminary source-level transformations and optimisations by front-end
     86%tools will be required to improve the efficiency of the back-end HLS tools.
     87%
     88%\item
     89%\textbf{Early performance evaluation}: For each point in the design space,
     90%figures of merit must be available such as throughput, latency, power
     91%consumption, area, memory allocation and data locality. They are evaluated
     92%by reliable estimators obtained by running the actual multi-task software
     93%application on the virtual prototype.
     94%
     95%\item
     96%\textbf{Independance from the Target FPGA}: The COACH description of the system
     97%(both hardware and software) should be independent of the FPGA family. 
     98%Every point of the design space can be implemented on any FPGA component,
     99%as long as it contains the hardware ressources required by the selected architectural template.
     100%Basically, COACH will support both Altera and Xilinx FPGA families.
     101%\end{itemize}
     102%
     103
     104
     105
Note: See TracChangeset for help on using the changeset viewer.