phd-defense/slides/60_a72combined/main.tex

66 lines
2 KiB
TeX
Raw Normal View History

2024-12-04 09:59:31 +01:00
\section[Wrapping up:\\the \acombined{} model]{Wrapping up: the \acombined{} model}
2024-11-24 21:07:50 +01:00
\begin{frame}
\begin{minipage}[c]{0.68\textwidth}
\centering
\includegraphics[height=0.9\textheight]{patate_apres.jpg}\\
\end{minipage}\hfill\begin{minipage}[c]{0.30\textwidth}
\Large+ tooled analysis of\\
existing analyzers
\end{minipage} \\
\vfill
\pause{}
\begin{center}
\textbf{\Large$\leadsto$ Let's make a model for the Cortex A72!}
\end{center}
\end{frame}
\begin{frame}{From dependencies to cycles}
\begin{itemize}
\item \staticdeps{}: set to also report \alert{register} dependencies.
\item Unroll $\kerK$ to fill the ROB
\item Build \alert{dependencies graph}: edges are dependencies,
weighted by \alert{source instruction latency} (given by \palmed).
\item Compute \alert{longest path}, divide by repetitions of $\kerK$
\end{itemize}
\begin{center}
\textbf{$\leadsto$ lower bound on execution time}
\end{center}
\vfill
\begin{center}
\textbf{Combine frontend, backend, critical by taking the max}
\end{center}
\end{frame}
\begin{frame}
\begin{columns}
\column{\dimexpr\paperwidth-8pt}
\centering
\begin{minipage}[c]{0.68\textwidth}
\centering
\footnotesize
\begin{tabular}{l r r r r r r}
\toprule
\textbf{Bencher} &
\textbf{Fail} &
\textbf{MAPE} & \textbf{Median} & \textbf{Q1} & \textbf{Q3} & \textbf{$K_\tau$} \\
& (\%) & (\%) & (\%) & (\%) & (\%) & \\
\midrule
2024-11-26 18:37:45 +01:00
A72 combined & 0.51 & 19.26 & 12.98 & 5.57 & \alert{25.38} & 0.75\\
llvm-mca & 0.06 & 32.60 & \alert{25.17} & 8.84 & 59.16 & 0.69\\
2024-11-24 21:07:50 +01:00
Osaca (backend) & 0.17 & 49.33 & 50.19 & 33.53 & 64.94 & 0.67\\
Osaca (crit. path) & 0.17 & 84.02 & 70.39 & 40.37 & 91.47 & 0.24\\
\bottomrule
\end{tabular}
\end{minipage}
\hfill
\begin{minipage}[c]{0.3\textwidth}
\centering
\includegraphics[width=\textwidth]{cesasme_a72combined_boxplot.svg}
\end{minipage}
\end{columns}
\end{frame}