1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
|
\begin{table}[htb] %[htbp]
\begin{threeparttable}
\caption{\label{tbl:hybrid} Performance and configuration of hybrid back-projection kernel}
\centering
\noindent
%\resizebox{\columnwidth}{!}{\begin{tabular}{} ... \end{\tabular}}
\begin{tabularx}{\columnwidth}{ | X c | r | l l l l l l | }
\hline
%& & & \multicolumn{5}{c|}{Configuration} \\
%\mhd{|c}{GPU} & \mhd{c|}{Slices} & \mhd{c|}{Perf.} & \mhd{c}{Area} & \mhd{c}{Blocks} & \mhd{c}{L1/SM} & \mhd{c}{CC} & \mhd{c|}{PaO} \\
& & \mhd{c|}{Perf} & \multicolumn{6}{c|}{Configuration} \\
\mhd{|c}{GPU} & \mhd{c|}{$n_v$} & GU/s & \mhd{c}{T/A} &\mhd{c}{$n_q$} & \mhd{c}{$s_d$} & \mhd{c}{U} & \mhd{c}{R} & \mhd{c|}{O} \\
%& & Hybrid & T/A & Px. & Pr. & U & Rnd. & Occ. \\
\hline
\multirow{3}{*}{GTX980}
& 1 & 266 & 3/5 & 16 & 8 & - & SFU &100\% \\
& 2 & 389 & 1/1 & 4 & 16 & - & SFU &100\% \\
\hline
\multirow{3}{*}{Titan}
& 1 & 734 & 3/5 & 16 & 8 & - & SFU &100\% \\
& 2 & 995 & 1/1 & 4 & 16 & - & SFU &100\% \\
\hline
\end{tabularx}
\begin{tablenotes}
\item The table summarizes the performance and optimal configuration for the hybrid back-projection kernel. Both texture engine and ALUs are used to perform interpolation. The configuration specifies: \tblcol{T/A} - is a ratio between the blocks executing \textbf{T}exture-based reconstruction and the blocks running \textbf{A}LU-based algorithm, \tblcol{$n_q$} - a number of pixels per thread, \tblcol{$s_d$} - a number of cached projections, \tblcol{U} - unrolling hint for inner projection loop, \tblcol{R} - the units to perform rounding and type conversions (index is always computed using SFU), \tblcol{O} - the requested occupancy. The caches are configured as specified in \tablename~\ref{tbl:cacheconf}. The number of threads to cache a projection row is determined according to guidelines in \tablename~\ref{tbl:shmemconf}.
\end{tablenotes}
\end{threeparttable}
\end{table}
|