mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
abf99d05fc
Restructured from flat .tex/.pdf files into per-paper subdirectories. Added Antje Worring as co-author on all papers. Synced PDFs to pdfs/ for download site.
47 lines
1.4 KiB
TeX
47 lines
1.4 KiB
TeX
\section{Benchmarks}
|
|
|
|
|
|
\subsection{NTT Performance}
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\small
|
|
\begin{tabular}{@{}lrrr@{}}
|
|
\toprule
|
|
\textbf{Implementation} & \textbf{NTT-256} & \textbf{NTT-512} & \textbf{NTT-1024} \\
|
|
\midrule
|
|
Naive multiply (Go) & 25.1\,$\mu$s & 98.3\,$\mu$s & 401\,$\mu$s \\
|
|
NTT pure Go & 3.1\,$\mu$s & 7.2\,$\mu$s & 16.8\,$\mu$s \\
|
|
NTT Go + AVX2 & 0.9\,$\mu$s & 2.1\,$\mu$s & 4.9\,$\mu$s \\
|
|
Reference C (\texttt{-O3}) & 0.85\,$\mu$s & 2.0\,$\mu$s & 4.6\,$\mu$s \\
|
|
\midrule
|
|
\textbf{Speedup (AVX2 vs.\ naive)} & \textbf{28x} & \textbf{47x} & \textbf{82x} \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{NTT performance on AMD Zen~4 (5.0~GHz, single core). Median of 10,000 runs.}
|
|
\end{table}
|
|
|
|
\subsection{Impact on Cryptographic Operations}
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\small
|
|
\begin{tabular}{@{}lrr@{}}
|
|
\toprule
|
|
\textbf{Operation} & \textbf{Pure Go} & \textbf{Go + AVX2} \\
|
|
\midrule
|
|
ML-KEM-768 KeyGen & 83\,$\mu$s & 28\,$\mu$s \\
|
|
ML-KEM-768 Encaps & 101\,$\mu$s & 34\,$\mu$s \\
|
|
ML-DSA-65 Sign (avg.) & 542\,$\mu$s & 186\,$\mu$s \\
|
|
ML-DSA-65 Verify & 162\,$\mu$s & 56\,$\mu$s \\
|
|
FHE Bootstrapping & 48\,ms & 16\,ms \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{SIMD impact on full cryptographic operations.}
|
|
\end{table}
|
|
|
|
The 3x speedup from SIMD is consistent across operations because NTT dominates the
|
|
computation in all cases.
|
|
|
|
%% ============================================================
|