source: anr/section-4.1.tex @ 23

\begin{figure}\leavevmode\center
\includegraphics[width=.8\linewidth]{architecture-csg}
\caption{\label{archi-csg} software architecture for digital system generation}
%\end{figure}\begin{figure}\leavevmode\center
\mbox{}\vspace*{1ex}\\
\includegraphics[width=1.0\linewidth]{architecture-hls}
\caption{\label{archi-hls} software architecture of hardware accellerator synthesis}
%\end{figure}\begin{figure}\leavevmode\center
\mbox{}\vspace*{1ex}\\
\includegraphics[width=.8\linewidth]{architecture-hpc}
\caption{\label{archi-hpc} software architecture of HPC}
\end{figure}
%
The figures~\ref{archi-csg}, \ref{archi-hls} and \ref{archi-hpc}
summarize the software architecture of COACH framework we plan to develop.
In figures, the dotted boxes are the softwares or formats that COACH
has to provide.
\vspace*{.75ex}\par
For the system genration presented in figure~\ref{archi-csg}, the conductor
is the tool \verb!CSG! (COACH System Generator). Its inputs are a process
network describing the application to design and the synthesis parameters.
The main parameters are the target hardware architectural template
with its instanciation parameters, the hardware/software mapping of the
tasks, the FPGA device and design constraints.
\verb+CSG+ thus requires an architectural template library, a operating system
library, two system hardware component (CPU, memories, BUS...) libraries
(one for synthesis, one for simulation).
For generating the coprocessor of a task mapped as hardware, \verb+CSG+
controls the HAS tools described below.
From these inputs \verb!CSG! can generate the entire system (both software \&
hardware) either as a SystemC simulator to prototype and explore quickly the
design space or as a bitstream\footnote{COACH generates synthesizable VHDL, and
launch the Xilinx or Altera RTL synthesis tools.} directly downloadable on the
FPGA device.
\\
%To proove CSG that COACH is open and CSG is really configurable, COACH will
%basically support 3 architecture template (the COACH template based on a
%MIPS processors and a VCI token ring, the Altera template based on the NIOS
%and AVALON bus, the Xilinx template based on the MICROBLAZE and OPB bus)
%and 2 operating systems (DNA/OS and MUTEK). Furthermore, thus is enforced
%by the \mustbecompleted{FIXME:zied} contribution that consists in
%implementing an other hardware target.
%\\
%Finally, it is important to notice that this work is a strong
%enhancement of the SocLib software.
\vspace*{.75ex}\par
The software architecture for HAS is presented in figure~\ref{archi-hls}.
The input is a single task of the process network. The HAS tools do not work
directly on the C++ task description but on an internal format called
\xcoach generated by a the GNU C compiler (GCC) tainted by a COACH
driver. This allows on the one hand to insure that all the tools will
accept the same C++ description and on the other hand to make possible
their chaining. The front-end tools read a \xcoach description and generate
a new \xcoach description that exibits more parallelism or implement
specific instructions for ASIP. The back-end tools read a \xcoach
description and generate a \xcoachplus description. This is a \xcoach
description anotated with hardware information (scheduling, binding) required by
the VHDL and systemC drivers.
Furthermore, the back-end tools uses a macro-cell library (functional and memory
unit).
\vspace*{.75ex}\par
In addition to digital system design, HPC requires a supplementary
partitioning step presented in figure~\ref{archi-hpc}. The designer
splits the initial application (tag 1) in two parts: one still on the PC and the
other running in a FPGA plugged on the PCI/X PC bus. The two parts exchange data
through communication primitives (tag 2) implemented in a library.
To evaluate if the relevance of the partitioning, the designer can build a
simulator. Once the partitioning is validated, the design of the FPGA part
is done through \verb!CSG! (figure~\ref{archi-csg}).
\vspace*{.75ex}\par
The project is splitted into 8 tasks numbered from 0 to 7.
The first task (task 0) is the project management, the last one (task 7) is
the dissemination the other task are listed below:
\begin{enumerate}
\item\textbf{\backbone:} This task tackles the fundamental points of the
        project such as the defintion of the COACH inputs and outputs,
    the internal formats (e.g. \xcoach), the architectural templates and
    the design flow.
\item\textbf{system generation:} This task addresses the prototyping and
    the generation of digital system. Apart from HAS that belong to the task 3
    and 4, its components are those presented figure~\ref{archi-csg}
    (e.g.  \verb!CSG!, operating systems).
\item\textbf{HAS front-end:} This task mainly focusses on four functionalities:
    optimization of the memory usage, parallelism enhancement through loop
    transformations, coarse grain parallelization and ASIP generation.
\item\textbf{HAS back-end:} This task groups two functionalities:
    High-Level Synthesis of data dominated description and HLS of control
    dominated description.
    This task contains also the development of a frequency adaptator
    that will allow the coprocessors to respect the processor \& the bus
    frequency.
\item\textbf{Communication between PC \& FPGA-SoC:}
    This task pools the features dedicated to HPC. The main are the
    partitioning validation (see figure~\ref{archi-hpc}, the sytem drivers for
    both PC and FPGA-SoC sides, the hardware communication components.
\item\textbf{Demonstrators:}
    This task groups the demostrators of the COACH project.
    \mustbecompleted{FIXME}
\end{enumerate}
%
\begin{figure}\leavevmode\center
%\includegraphics[width=.4\linewidth]{dependence-task}
\includegraphics[width=0.70\linewidth]{dependence-task-h}
\caption{\label{dependence-task}Task dependencies}
\end{figure}
Figure~\ref{dependence-task} presents the dependencies between the tasks.
"$task-N \longrightarrow task-M$" means that $task-N$ requires $task-M$
to work and be demonstrated. The more bold is the arrow, the more important is
the dependency.
The graph shows:
\begin{itemize}
\item Even that $T3$ and $T4$ functionalities are complementary, their
developments are independent (thanks to \xcoach internal format).
\item $T2$ depends slightly from $T3$ and $T4$. Indeed, $T2$ may works
without $T3$ and $T4$ if we limit to digital systems without hardware
accellerators.
\item $T5$  strongly depends on $T2$ but, $T2$ does not depend at all on
$T5$. So demonstrators ($T6$) of embedded system would not be impacted if
$T5$ would fail.  
\item $T1$ drives all the tasks ($T2$, $T3$, $T4$, $T5$) at the heart of
the COACH project. 
\item $T7$ and $T0$ respectively depends on and impacts all the other tasks.
\end{itemize}
So this organisation offers enough robustness for insure the success of the
project except for the specification task $T1$. However, the partners meet
10 times (a one day meeting per month) during the last years to prepare the
specification and the project proposal.