Skip to content

BENCHMARKS.md

Latest commit

 

History

History
112 lines (92 loc) · 3.66 KB

BENCHMARKS.md

File metadata and controls

112 lines (92 loc) · 3.66 KB

Benchmarks

Release measurements for native vs zig engines.

Environment

  • Android: ARM64 (AArch64), Android 15
  • iOS: Apple A15 (arm64), iOS 26.3
  • Build: Release
  • Values below are medians (ms, lower is better)

Key takeaway

  • Main Android bottleneck is SHA-2 in raw generic Zig builds.
  • Example: SHA-256 / 200 MiB is 220 ms (native) vs 12818 ms (Zig raw).
  • Current package routes Android SHA-224 / SHA-256 and HMAC-SHA-224 / HMAC-SHA-256 through native fallback in Zig mode.
  • If Zig is built with -Dcpu=baseline+sha2, SHA-256 / 200 MiB becomes 219 ms (near native 220 ms).

Why keep Zig engine in React Native

  • One portable hashing core with a stable C ABI, so behavior does not drift between Swift/Kotlin/other bindings as the library grows.
  • Faster delivery of new algorithms/features: implement once in Zig core, then expose through RN bindings without duplicating full crypto logic per platform.
  • Performance hedge instead of lock-in: keep both engines available and select the better path per platform/algorithm based on real measurements.
  • Current data already shows Zig wins in part of the matrix (for example iOS SHA-256, SHA-224, HMAC-SHA-256, XXH3-64) while native wins others.
  • Android deployment strategy is flexible: current package uses generic Zig + native SHA-2 fallback; an alternative is shipping two Zig variants (generic and +sha2) with runtime routing.

Quick comparison

Android

Scenario Native Zig raw (no fallback) Zig (-Dcpu=baseline+sha2)
SHA-256, file 16 MiB 23 1038 20
SHA-256, file 200 MiB 220 12818 219
SHA-512/224, file 200 MiB 3490 725 723
BLAKE3, file 200 MiB 298 555 551
XXH3-64, file 200 MiB 79 73 76

iOS

Scenario Native Zig
SHA-256, file 16 MiB 31 29
SHA-256, file 200 MiB 262 221
BLAKE3, file 200 MiB 339 728
XXH3-64, file 200 MiB 120 81

Full matrix: Android 200 MiB

Algorithm Native Zig raw (no fallback) Zig (-Dcpu=baseline+sha2)
SHA-256 220 12818 219
MD5 561 536 523
SHA-1 221 509 501
SHA-224 215 12785 218
SHA-384 538 723 723
SHA-512 543 724 736
SHA-512/224 3490 725 723
SHA-512/256 3484 724 720
HMAC-SHA-224 233 12803 218
HMAC-SHA-256 227 12764 221
HMAC-SHA-384 549 728 725
HMAC-SHA-512 540 728 723
HMAC-MD5 566 537 526
HMAC-SHA-1 223 508 501
BLAKE3 298 555 551
XXH3-64 79 73 76

Full matrix: iOS 200 MiB

Algorithm Native Zig
SHA-256 262 221
MD5 627 607
SHA-1 258 602
SHA-224 265 240
SHA-384 344 1236
SHA-512 354 1239
SHA-512/224 343 1236
SHA-512/256 343 1236
HMAC-SHA-224 260 255
HMAC-SHA-256 258 253
HMAC-SHA-384 342 1240
HMAC-SHA-512 343 1246
HMAC-MD5 621 616
HMAC-SHA-1 256 601
BLAKE3 339 728
XXH3-64 120 81

Size impact

  • Android APK: +0.49% (+81,920 bytes) with Zig engine prebuilts
  • Android AAB: +0.42% (+51,551 bytes)
  • iOS app dir: -0.20% (-60 KiB)
  • iOS main binary: -0.30% (-61,952 bytes)

Notes

  • Zig raw (no fallback) shows pure Zig-path timing without Android runtime fallback and without target-specific CPU features.
  • In real app flow with current Android fallback, SHA-224 / SHA-256 and matching HMAC variants run through native, so expected latency there is near the Native column, not Zig raw.
  • XXH3-128 is native-only and not included in Zig comparisons.