KarpelesLab / anyhash / 24333417795

88%

push

github

Speed up gost and snefru: 4.5x and 1.8x on 64 KiB throughput

gost (13 → 59 MB/s):
- Fold the GOST 28147-89 S-box substitution with the 11-bit left rotation
  into four precomputed 256-entry uint32 tables per S-box, so each round
  collapses to four byte-indexed lookups OR'd together instead of eight
  variable-shift 4-bit lookups followed by a rotate.
- Rewrite gostPsi as four uint64 shift/or ops, eliminating the 30-byte
  memmove that ran 74 times per compression.
- Inline the 74 psi iterations, the psi(m XOR ...) step, and the h XOR
  into gostShuffle64 so the 256-bit shuffle state stays in registers for
  the whole shuffle.

snefru (25 → 45 MB/s):
- Remove the (i/2)&1 branch in the 16-word inner loop by picking the two
  S-boxes into locals once per round and referring to them directly in the
  unrolled body.
- Unroll the 4-pass byteInWord loop and use bits.RotateLeft32 with a
  literal shift so the compiler emits a ROR instruction per word instead
  of a register-held shift+or.

All existing tests pass.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

119 of 119 new or added lines in 2 files covered. (100.0%)

2847 of 3264 relevant lines covered (87.22%)

695.82 hits per line