Task: add runtime-dispatched AVX2 merkle hashing

[metaphorics]

Parent

#16

What to build

Process independent SHA-256d merkle pairs in 8-way AVX2 batches on supported x86-64 hosts, with the current scalar path as the general tail and unsupported-host implementation. Preserve Bitcoin's exact odd-node duplication and mutated-tree result.

Use current upstream bitcoin_hashes and Bitcoin Core source as the API and algorithm authority before deciding whether project code is necessary. Do not add a dependency or a second abstraction when the standard library intrinsics and the existing merkle interface suffice.

Blocked by

• Task: calibrate allocator parity and freeze the matched 0-150k baseline #24

Acceptance criteria

• Runtime feature detection selects AVX2 safely; non-AVX2 builds and hosts use a scalar fallback with no illegal instruction risk.
• Scalar and AVX2 outputs are byte-identical for one leaf, every batch tail, odd levels, duplicate adjacent hashes, and the Bitcoin mutated-pair rule.
• Corpus-wide parity covers every merkle level in blocks 0→150,000 and preserves the accepted stop hash.
• The hot path performs no per-node heap allocation, avoids needless copies, and batches homogeneous 64-byte pair inputs.
• A microbenchmark records the hash-kernel gain, but promotion depends on three interleaved full-replay control/candidate pairs under the Task: calibrate allocator parity and freeze the matched 0-150k baseline #24 custody packet.
• Run the full-replay gate with fjall, rocksdb, and redb to catch contention or backend interaction.
• Promote only if one whole-process primary axis clears 1.05× and the other axis plus RSS do not materially regress. Otherwise revert and record the falsifier.

[coderabbitai]

🔗 Related PRs

#7 - Parallelize block input script checks for performance improvement [merged]
#14 - feat: header timestamp consensus rules, index reorg rollback, mempool policy, fuzz targets [open]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[metaphorics]

Implemented in #30.

The candidate reuses prepared txids, batches eight independent SHA256d64 parent pairs behind runtime AVX2 detection, and retains scalar tails and unsupported-host fallback. Corpus instrumentation matched the scalar reducer at every Merkle level for all blocks 0 through 150,000.

Three-run paired wall medians improved by 1.177x on fjall, 1.171x on RocksDB, and 1.112x on redb. Separate validation runs for all three backends matched the exact tip, UTXO count, total amount, MuHash, and serialized UTXO hash. The final interleaved Core panel measured bitcoin-rs at 49.356s / 390.542 CPU-s against Core at 64.914s / 481.092 CPU-s: 1.315x wall and 1.232x CPU, with 1.042x Core RSS.

The committed custody packet is docs/benchmarks/data/end-to-end-sync/avx2-merkle-custody-v1.json.

[metaphorics]

Reopened until PR #30 merges. The implementation and evidence are complete in the open PR; its existing Closes #27 directive will close this ticket when the code reaches the target branch.