Synchronous_reactive_systems/beamer/beamer.tex

\documentclass{beamer}
\usepackage{listings}
\lstset{
  basicstyle=\ttfamily,
  columns=fullflexible,
  frame=single,
  breaklines=true,
}
\usepackage{tikz}

\usetikzlibrary{positioning}
\usetheme{Darmstadt}

\begin{document}

\title{Presentation of our Lustre-to-C compiler}
\subtitle{github.com/Benjamin-Loison/Synchronous-reactive-systems}
\date{16 December 2022}
\author{Benjamin Loison, Arnaud Daby-Seesaram, Antoine Grimod}

\frame{\titlepage}

\section{Structure of the compiler}
\begin{frame}{Main ideas}
	\begin{figure}
		\begin{tikzpicture}
			\node (sf) {Source file};
			\node[right =2cm of sf] (ast) {Typed AST};
			\node[right =2cm of ast] (C) {C program};
			\draw
				(sf) edge[->] node[above] {parser} (ast)
				(ast) edge[->] node[above] {compiler} (C);
		\end{tikzpicture}
		\caption{Structure of the compiler}
	\end{figure}
\end{frame}

\begin{frame}{Testing}
	\begin{block}{Passes}
		The passes can be split into:
		\begin{itemize}
			\item those checking the program validity
			\item those modifying the AST of the program
		\end{itemize}
	\end{block}
\end{frame}

\section{Typed AST}
\subsection{First attempt using GADTs}
\begin{frame}
	\begin{block}{Main idea}
		Using GADTs to represent nodes and expressions allows to ensure the
		well-typedness of a program.
	\end{block}
	\begin{figure}
		\centering
		\includegraphics[width=.75\textwidth]{imgs/gadt.png}
	\end{figure}
\end{frame}
\begin{frame}
	\begin{block}{Pros of using GADTs}
		\begin{itemize}
			\item Any term of the GADT represents a well-typed program
			\item Extending the language to support more types consists of adding
				constructors to variables and constants
			\item The types are easy to read and understand
		\end{itemize}
	\end {block}

	\begin{block}{Cons of using GADTs}
		\begin{itemize}
			\item
			They cannot be dynamically generated (hence it is impossible to
			implement a parser that gives back a GADT)
			\item
			One should think about the isomorphism between
			\texttt{a $\ast$ (b $\ast$ c)} and \texttt{(a $\ast$ b) $\ast$ c}.
		\end{itemize}
	\end{block}
\end{frame}

\subsection{Second attempt: using explicit types in the variables, expressions,
\dots{} constructors}
\begin{frame}
	\begin{block}{Idea}
		Explicitly collect typing information while parsing.
	\end{block}
	\begin{figure}
		\centering
		\includegraphics[width=.6\textwidth]{imgs/explicit_types.png}
	\end{figure}
\end{frame}

\begin{frame}
	\begin{block}{Pros of using explicit types}
		\begin{itemize}
			\item Programs can be built dynamically, hence a parser can be
			written
			\item While parsing, the parser has all the required information on
			the sub-variables/nodes/expressions to check the well-typedness
		\end{itemize}
	\end{block}
	\begin{block}{Cons of these definitions}
		\begin{itemize}
			\item The typing information on terms is very redundant.
			\item The rejection of ill-typed programs depends on the correctness
			of the parser
		\end{itemize}
	\end{block}
\end{frame}

\section{Passes}
\begin{frame}{}
    \begin{block}{Classification}
        \begin{itemize}
            \item node-passes: for all nodes, do \texttt{P: t\_node -> t\_node
            option}
            \pause
            \item equation-passes: for all equations of all nodes, do
            \texttt{P: t\_equation -> t\_equation option}
            (the definition uses the node-passes constructor)
            \pause
            \item expression-passes: for all expression of all equations, do
            \texttt{P: t\_expression -> t\_expression option}
            (the definition uses the equation-passes constructor)
        \end{itemize}
    \end{block}
\end{frame}
\begin{frame}{Implemented Passes}
	\begin{block}{Sanity checks}
		\begin{itemize}
			\item Check the well-typedness of a program
            \pause
			\item Check that there are no assignment conflicts in a programs
            (node-pass)
            \pause
		\end{itemize}
	\end{block}
	\begin{block}{AST modification}
		\begin{itemize}
			\item Rewrite automata into \texttt{if-then-else} constructs
            (node-pass)
            \pause
			\item Linearization of the equations:
            \begin{itemize}
                \item for all, \texttt{pre e} add a, intermediate variable
                \item \texttt{v1, v2, v3 = f(i), e3;} is rewritten into
                \texttt{v1, v2 = f(i); v3 = e3;}
            \end{itemize}
            (node-pass)
            \pause
			\item (no longer required) Push the \texttt{pre} to variables
            (expression-pass)
		\end{itemize}
	\end{block}
\end{frame}

\begin{frame}[fragile]{Translation of automaton}
	\only<1>{\lstinputlisting[language=ml, firstline=1, lastline=7]{code/example_automaton.lus}}
	\only<2>{\lstinputlisting[language=ml, firstline=9, lastline=14]{code/example_automaton.lus}}
\end{frame}
\begin{frame}{Restriction on automaton}
	The patterns that appears on the left of equations must be the
	same in all branches

	\only<2>{\lstinputlisting[language=ml, firstline=16, lastline=22]{code/example_automaton.lus}}
\end{frame}

\begin{frame}{Clock unification}

	Derived from the rules provided in \emph{Clock-directed Modular Code Generation for Synchronous
	Data-flow Languages}
\end{frame}

\section{Translation to C}
\begin{frame}

	\texttt{ast\_to\_c.ml} architecture similar to Arnaud's AST pretty-printer.

	\pause

	For instance, go from \texttt{counting.lus} to \texttt{counting.c}.

    \pause

	Use of three tricks, as our compiler only manages \texttt{bool}s, \texttt{int}s and \texttt{float}s:
		\begin{enumerate}
			\item \texttt{0} can be interpreted as a \texttt{bool}, an \texttt{int} and a \texttt{float}
			\pause
			\item A \texttt{float} correctly encode \texttt{bool}s (\texttt{true} and \texttt{false}) and \texttt{int}s (between $[-2^{24} + 1; 2^{24} + 1]$)
			\pause
			\item To run an assignment of \texttt{value} to \texttt{variable} within the condition of a \texttt{if} and also make the cases of the \texttt{if} depends on a condition \texttt{condition}, we can do \texttt{if(((variable = value) \&\& false) || condition)}\\
			\pause
			We can also use this trick to execute an assignment and \textit{return} a value \texttt{value\_to\_return} without using any \texttt{;}, by using \texttt{((variable = value) || true) ? value\_to\_return : 0} (thanks to the first trick)
		\end{enumerate}
\end{frame}

\begin{frame}
	\begin{itemize}
		\item
			\begin{itemize}
				\item \texttt{pp\_varlist (Base | Arg | Dec)}
				\item \texttt{pp\_retvarlist}
				\item \texttt{pp\_prevarlist}
				\item \texttt{pp\_asnprevarlist}
			\end{itemize}
		\pause
		\item \texttt{when} implementation error: division by zero for instance, so used the first trick (\texttt{when\_merge.node})
		\pause
		\item \texttt{->}: use global variable \texttt{init\_\{NODE\_NAME\}}\\
					\texttt{1 -> 2 -> 3} returns \texttt{0} on first run and only \texttt{3} on the following ones, correct? (\texttt{arrow.node})
		\pause
		\item \texttt{pre}
		\pause
		\item \texttt{reset}% (\texttt{reset})
		\pause
		\item Functions returning tuples thanks to Arnaud's linearization only have to deal with \texttt{(a0, a1, ..., an) = return\_tuple\_a0\_a1\_dots\_an(0)} (\texttt{function\_returning\_tuple.node})
	\end{itemize}
\end{frame}

\section{Tests}
\begin{frame}{Tests}
    \begin{block}{testing methods}
        We thought of three testing methods:
        \begin{itemize}
            \item manual testing of our functionalities
            \item run the sanity-checks-passes after any AST-altering pass
            \item simulation of the nodes (aborted)
        \end{itemize}
    \end{block}
\end{frame}

\section{Possible improvements}
\begin{frame}{Improvements}
    \begin{itemize}
        \item Increase the expressivity of the accepted programs
        \item Improve the complexity of the different passes
        \begin{itemize}
            \item Group neighbour passes of the same type (node-, expression or
            equation-pass).
        \end{itemize}
        \item \dots{}
    \end{itemize}
\end{frame}

\end{document}