amethyst

Files

T

Claude eb9527d793 perf: hardware SHA-256 acceleration (SHA-NI x86_64, CE ARM64)

Add hardware-accelerated SHA-256 using platform crypto extensions:

x86_64 (SHA-NI, Intel Goldmont+/AMD Zen+):
  SHA256RNDS2 — processes 2 SHA-256 rounds per instruction
  SHA256MSG1/SHA256MSG2 — message schedule expansion
  Result: SHA-256(160B) 749ns → 165ns (4.5x faster)

ARM64 (Crypto Extensions, all Android ARMv8 phones):
  SHA256H/SHA256H2 — hash update (4 rounds per instruction)
  SHA256SU0/SHA256SU1 — message schedule
  Expected: similar 4-5x speedup on phone

Impact on secp256k1 operations:
  signSchnorr:    33→31µs (8%, 3-4 SHA-256 calls per sign)
  signXOnly:      17→16µs (8%)
  verifyFast:     34→33µs (4%, 1 SHA-256 call)
  batch(200):     7.2→6.6µs/event (10%, 200 SHA-256 calls)
  batch(200) now 5.1x faster than ACINQ individual verify

Native C-to-C (with SHA-NI + all optimizations):
  pubkeyCreate:  ACINQ 15.9  Ours 14.9  1.07x faster
  sign (cached): ACINQ 17.7  Ours 15.1  1.18x faster
  sign (full):   ACINQ 33.2  Ours 30.6  1.08x faster
  verifyFast:    ACINQ 32.1  Ours 33.7  0.95x (tied)
  batch(200):    ACINQ 33.5  Ours  6.6  5.1x faster

https://claude.ai/code/session_011KVZhDcV2G7idNWEBz12GY

2026-04-11 13:19:46 +00:00

.idea

Switches to our own version of the Url Detector

2026-03-06 10:58:59 -05:00

benchmarks

feat: upgrade negentropy-kmp to v1.0.2, enable macosArm64 target

2026-04-11 00:44:33 +00:00

src

perf: hardware SHA-256 acceleration (SHA-NI x86_64, CE ARM64)

2026-04-11 13:19:46 +00:00

.gitignore

…