rscrypto 0.1.0

Rust crypto with zero default deps: BLAKE3, Ed25519/X25519, hashes, MACs, KDFs, AEADs, and checksums with SIMD/ASM acceleration.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

//! Blake2s kernel dispatch with `OnceCache` caching.
//!
//! # Non-x86 SIMD policy
//!
//! The AArch64 NEON, POWER VSX, and s390x vector kernels are correct, but
//! single-block SIMD loses to the portable scalar path in CI. The 2026-04-27
//! Linux bench run confirmed AArch64 NEON loses to portable on Graviton3/4, so
//! do not select those kernels in production dispatch.
//!
//! Regaining dominance on those platforms requires multi-block kernels
//! (Blake3-style 2-way or 4-way), not more single-block shuffle tuning.

use super::kernels::{Blake2sKernelId, CompressBlocksFn, CompressFn, compress_blocks_fn, compress_fn};
define_blake2_dispatch! {
  kernel_id: Blake2sKernelId,
  compress_fn_ty: CompressFn,
  compress_blocks_fn_ty: CompressBlocksFn,
  portable_kernel: Blake2sKernelId::Portable,
  compress_fn: compress_fn,
  compress_blocks_fn: compress_blocks_fn,
  required_caps: super::kernels::required_caps,
  candidates: [
    #[cfg(target_arch = "x86_64")]
    Blake2sKernelId::X86Avx512vl,
    #[cfg(target_arch = "x86_64")]
    Blake2sKernelId::X86Avx2,
    #[cfg(target_arch = "riscv64")]
    Blake2sKernelId::Riscv64V,
    #[cfg(target_arch = "wasm32")]
    Blake2sKernelId::WasmSimd128,
  ],
}