ハッシュ Performance Benchmarking リファレンス
リファレンス 用 ハッシュ 関数 performance characteristics, hardware acceleration, および benchmarking methodology.
Hardware Acceleration
Modern x86 CPUs provide SHA-NI instructions accelerating SHA-256 and SHA-1 by 3-4x. ARMv8 Crypto Extensions accelerate SHA-1, SHA-256, and SHA-512. SHA-3 acceleration (KECCAK) is less common but available on some recent processors. BLAKE2/3 use standard integer/vector instructions without dedicated hardware. Hardware acceleration can change the optimal hash choice for a given platform.
- SHA-NI
- x86 SHA-256 accel
- ARMv8 crypto
- SHA-1/256/512
- no HW needed
- BLAKE2/3 use SIMD
3-4x speedupmobile fastportableBenchmarking Methodology
Hash benchmarks must warm up caches, use realistic input sizes, and report throughput (MB/s) and cycles/byte. Small input benchmarks are dominated by setup overhead; large inputs measure steady-state throughput. Memory bandwidth limits parallel hashes beyond a certain core count. SIMD width (SSE2, AVX2, AVX-512, NEON) significantly affects results. Always compare on the same hardware with identical compiler flags.
- warmup
- Prime caches and branch predictor
- input sizes
- Test range small to large
- SIMD
- Vector width matters
avoid cold-startsetup vs steadyAVX2 vs AVX-512