faster_keccak

Fast, zero-allocation Keccak-256 for Go with platform-specific assembly.

Go's crypto/sha3 only exposes SHA-3 (domain 0x06), not Keccak-256 (domain 0x01). x/crypto/sha3.NewLegacyKeccak256() provides Keccak-256 but uses a pure-Go permutation on all platforms. This package uses assembly-optimized keccak-f[1600] permutations instead:

• arm64 (Apple Silicon): NEON SHA3 extensions (EOR3, RAX1, XAR, BCAX)
• amd64: Unrolled permutation with complementing lanes optimization
• Fallback: Pure-Go implementation (or with purego build tag)

Usage

import "github.com/Giulio2002/faster_keccak"

// One-shot
digest := keccak.Sum256(data)

// Streaming (zero allocs, stack-allocated)
var h keccak.Hasher
h.Write(part1)
h.Write(part2)
digest := h.Sum256()

Benchmarks

faster_keccak vs x/crypto/sha3

Size	faster_keccak	x/crypto	Speedup
32 B	116.4 ns/op (275 MB/s)	244.8 ns/op (131 MB/s)	2.1x
128 B	121.8 ns/op (1051 MB/s)	244.1 ns/op (524 MB/s)	2.0x
256 B	247.1 ns/op (1036 MB/s)	467.7 ns/op (547 MB/s)	1.9x
1 KB	988.9 ns/op (1035 MB/s)	1801 ns/op (569 MB/s)	1.8x
4 KB	3857 ns/op (1062 MB/s)	6896 ns/op (594 MB/s)	1.8x
500 KB	485.6 us/op (1054 MB/s)	836.8 us/op (612 MB/s)	1.7x

Zero allocations across all sizes (x/crypto allocates 32 B/op).

Testing

go test -v ./...

# Fuzz against x/crypto reference
go test -fuzz FuzzSum256 -fuzztime 30s

Authors: Giulio Rebuffo