Update on Overleaf.

1 files changed, 14 insertions, 8 deletions

diff --git a/evaluation.tex b/evaluation.tex
index 89866ef..544cafa 100644
--- a/evaluation.tex
+++ b/evaluation.tex
@@ -17,18 +17,19 @@ Our evaluation is designed to answer the following three research questions.
 
 \subsection{RQ1: How fast is CoqUp-generated hardware?}
 
-Figure~\ref{fig:comparison_time} shows.....
-
 \begin{figure}
 \begin{tikzpicture}
 \begin{axis}[
     xmode=log,
     ymode=log,
-    height=60mm,
-    width=60mm,
+    height=80mm,
+    width=80mm,
     xlabel={LegUp (ms)},
     ylabel={CoqUp (ms)},
+    xmin=0.01,
     xmax=10,
+    ymax=500,
+    ymin=0.5,
     log ticks with fixed point,
   ]
   
@@ -36,23 +37,28 @@ Figure~\ref{fig:comparison_time} shows.....
   table [x=legupwallclockMS, y=coqupwallclockMS, col sep=comma]
   {results/comparison.csv};
   
+\addplot[dashed, domain=0.01:10]{50*x};
+  
 \end{axis}
 \end{tikzpicture}
-\caption{A comparison of the execution time of hardware designs generated by CoqUp and by LegUp}
+\caption{A comparison of the execution time of hardware designs generated by CoqUp and by LegUp. The diagonal is at $y=50x$.}
 \label{fig:comparison_time}
 \end{figure}
 
+Figure~\ref{fig:comparison_time} compares the execution time of hardware designs generated by CoqUp and by LegUp, with each data point corresponding to one of the PolyBench/C benchmarks. It uses logarithmic scales so that the data points are spread out more evenly. The execution times are quite well correlated between the two tools, but LegUp generates hardware designs that execute roughly fifty times faster, as illustrated by the diagonal line at $y=50x$. There are two reasons for this. First, because CoqUp-generated hardware has all of its operations serialised, it requires more clock cycles to perform its computation -- about 4$\times$ as many as LegUp-generated hardware. Second, CoqUp generates hardware with a much lower clock frequency than LegUp (roughly 22 MHz compared to roughly 250 MHz) so each of those clock cycles takes about 11$\times$ longer. The main reason for the low clock frequency is \JW{could somebody else finish this bit off?}.
+
 \subsection{RQ2: How area-efficient is CoqUp-generated hardware?}
 
-Figure~\ref{fig:comparison_area} shows....
+Figure~\ref{fig:comparison_area} compares the size of the hardware designs generated by CoqUp and by LegUp, with each data point corresponding to one of the PolyBench/C benchmarks. We see that CoqUp designs use between 1\% and 30\% of the available logic on the FPGA, averaging at around 10\%, whereas LegUp designs all use less than 1\% of the FPGA, averaging at around 0.45\%. The main reason for this is \JW{blah blah}.
 
 \begin{figure}
 \begin{tikzpicture}
 \begin{axis}[
-    height=60mm,
-    width=60mm,
+    height=80mm,
+    width=80mm,
     xlabel={LegUp (\%)},
     ylabel={CoqUp (\%)},
+    xmin=0, ymin=0,
   ]
   
 \addplot[draw=none, mark=*, draw opacity=0, fill opacity=0.3]