High-Performance AEADs Benchmarks

This repository contains cryptographic benchmarking implementations for various AEAD (Authenticated Encryption with Associated Data) algorithms designed for high-performance, both on ARM and Intel CPUs.

Included Algorithms

• AEGIS-128x2 variants (AES-NI, VAES, ARM)
• AEGIS-128x4 (AVX-512 optimized)
• AES-128-GCM (OpenSSL backend)
• HiAE variants (HiAE, HiAEx2, HiAEx4)
• ROCCA-S cipher

Quick Start

Using the Top-Level Makefile

# Build all algorithms (auto-detects architecture)
make all

# Build with optimal performance (recommended)
CC="clang -march=native" make all

# Run all tests
make test

# Run all benchmarks
make benchmark

# Clean all build artifacts
make clean

Building Individual Algorithms

Each algorithm is in its own directory with consistent build system:

cd <algorithm-directory>
make                    # Build test and benchmark executables
./<algorithm>_test      # Run functionality tests
./<algorithm>_benchmark # Run performance benchmarks

Performance Recommendation: For optimal performance, use clang as your compiler:

CC="clang -march=native" make

Clang typically produces faster cryptographic code than GCC due to better optimization of vectorized operations and loop unrolling.

Architecture Support

The build system automatically detects your architecture and builds appropriate implementations:

• x86_64/Intel: AEGIS-128x2 (AES-NI, VAES), AEGIS-128x4 (AVX-512), AES-128-GCM, HiAE, HiAEx2, HiAEx4, ROCCA-S
• ARM64: AEGIS-128x2 (ARM crypto extensions), AES-128-GCM, HiAE, HiAEx2, HiAEx4, ROCCA-S

Note: Some implementations require specific CPU features:

• AEGIS-128x2-aesni: AES-NI, AVX
• AEGIS-128x2-vaes: AES-NI, AVX2, VAES
• AEGIS-128x4-avx512: AVX-512F, VAES
• AES-128-GCM: OpenSSL library

Performance Testing

Benchmarks test multiple message sizes (16B to 64KB) and measure:

• Throughput (Gbps/Mbps)
• Cycles per byte
• Cross-platform performance characteristics

Running Benchmarks

# Run all benchmarks
make benchmark

# Run individual benchmark
cd <algorithm-directory>
./<algorithm>_benchmark

# Generate CSV output for analysis
./<algorithm>_benchmark > results.csv

Performance Results

Apple Silicon M4

AMD Zen 4

Benchmark Visualization

The repository includes a Python tool for visualizing benchmark results:

cd benchmark-visualizer
pip install pandas matplotlib numpy  # or: uv sync
python plot_performance.py path/to/csv/files/

This generates comparative plots for:

• Throughput comparison across algorithms
• Efficiency (cycles per byte) analysis
• Performance trends across message sizes

Dependencies

• C compiler (clang recommended for performance)
• OpenSSL development libraries (for AES-GCM)
• Python 3.x with pandas, matplotlib, numpy (for visualization tools)

Contributing

When adding new algorithms, follow the existing directory structure and implement the standard AEAD interface defined in crypto_aead.h.