Update DXT1 timings table.

Clang is able to get much better times on ARM (around 430 us for both
ARM32 and ARM64 NEON). The reference implementation is 1.13 ms on clang.
This commit is contained in:
Bartosz Taudul 2019-07-13 20:24:58 +02:00
parent f65373ece7
commit ff9637e884

View File

@ -459,17 +459,17 @@ To further reduce image data size, frame images are internally compressed using
\centering \centering
\begin{tabular}[h]{c|c|c} \begin{tabular}[h]{c|c|c}
\textbf{Implementation} & \textbf{Required define} & \textbf{Time} \\ \hline \textbf{Implementation} & \textbf{Required define} & \textbf{Time} \\ \hline
x86 Reference & --- & 228 \si{\micro\second} \\ x86 Reference & --- & 218.7 \si{\micro\second} \\
x86 SSE4.1\textsuperscript{a} & \texttt{\_\_SSE4\_1\_\_} & 35.8 \si{\micro\second} \\ x86 SSE4.1\textsuperscript{a} & \texttt{\_\_SSE4\_1\_\_} & 33.9 \si{\micro\second} \\
x86 AVX2 & \texttt{\_\_AVX2\_\_} & 26.5 \si{\micro\second} \\ x86 AVX2 & \texttt{\_\_AVX2\_\_} & 23.8 \si{\micro\second} \\
ARM Reference & --- & 1.23 \si{\milli\second} \\ ARM Reference & --- & 1.23 \si{\milli\second} \\
ARM32 NEON\textsuperscript{b} & \texttt{\_\_ARM\_NEON} & 561 \si{\micro\second} \\ ARM32 NEON\textsuperscript{b} & \texttt{\_\_ARM\_NEON} & 561 \si{\micro\second} \\
ARM64 NEON & \texttt{\_\_ARM\_NEON} & 473 \si{\micro\second} ARM64 NEON & \texttt{\_\_ARM\_NEON} & 469 \si{\micro\second}
\end{tabular} \end{tabular}
\vspace{1em} \vspace{1em}
\textsuperscript{a)} VEX encoding; \hspace{0.5em} \textsuperscript{b)} ARM32 NEON code compiled for ARM64 \textsuperscript{a)} VEX encoding; \hspace{0.5em} \textsuperscript{b)} ARM32 NEON code compiled for ARM64
\caption{Compression time of $320\times180$ image. x86: i7 8700K; ARM: ODROID-C2.} \caption{Compression time of $320\times180$ image. x86: i7 8700K (MSVC); ARM: ODROID-C2 (gcc).}
\label{EtcSimd} \label{EtcSimd}
\end{table} \end{table}