feat(pz2d): ship the dict tier (-p pz2d) — blob 30.5%, best LZ ratio in pz#148
Merged
Conversation
…in pz Pz2d (Pipeline id 14): one container block = one 32 MiB segment whose first 16 MiB doubles as a shared dictionary for the segment's remaining 2 MiB blocks. Rides the existing container/scheduler unchanged. Library: pz2::encode_segment / decode_segment — the dict region is ONE parse split at block boundaries (split match halves keep their offsets, still valid against earlier region content; sub-MIN_MATCH remnants become literals), dicted blocks use the frozen finder with an arena reused across blocks. Took the 16 MiB-dict blob encode 70.5 -> 38.6 s ST vs the frozen probe (no partial-dict rebuilds, region parsed once). Container: inner framing [num_inner][orig,comp]*N + wires inside the segment payload; 2-wave decode in decompress_block_pz2d (dict region = sequential prefix chain, remaining blocks fan out across scoped threads against the completed region). Auto/greedy/lazy parse mapping matches Pz2 (auto gate applied at segment granularity). CLI: -p pz2d, -l, list. Measured (blob, CLI e2e) and documented honestly in section 11b: - ratio 30.475% — best LZ-family ratio in pz, 0.92pp under pzstd-3, 0.56pp under pz2; round-trip verified - decode 42.5 ms v1 — memory-traffic-bound (saturates at 4 threads) by naive per-block dict priming (~3.3 GB of zero+memcpy), exactly the failure mode section 11's arithmetic predicted; arena decode projected ~17-20 ms, deferred because it touches the proven unsafe splice - encode 23.7 s — concurrent segment encodes inflate the ST probe 2.9x (16 MiB dict + 64 MB frozen prev walks; the 4 MiB-block cache lesson at segment scale); levers documented Pz2d added to all four cross-pipeline test matrices + segment round-trip unit test (boundary-crossing matches, unaligned blocks, dict > segment, degenerate tiny blocks). 742 + 595 tests, fmt, clippy clean. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Ships the Pz2d dict tier end-to-end:
-p pz2d(id 14), one container block = one 32 MiB segment whose first 16 MiB doubles as the shared dictionary for the segment's remaining 2 MiB blocks. Zero changes to the container/scheduler machinery.Pieces
pz2::encode_segment/decode_segment— dict region encoded as ONE parse split at block boundaries (split match halves keep their offsets — still valid against earlier region content); dicted blocks via the frozen finder (feat(pz2): frozen shared match-finder — dict-tier encode at 3x spike speed, same ratio #147) with arena reuse. Blob encode for the 16 MiB tier: 70.5 → 38.6 s ST vs the frozen probe.[num_inner][orig,comp]*N+ wires) inside the segment payload; 2-wave decode: dict region is a sequential prefix chain, the remaining blocks fan out across scoped threads against the completed region.-p pz2d,-l,--list-pipelines), parse-strategy mapping identical to Pz2 (auto-greedy gate at segment granularity), all four cross-pipeline test matrices + a dedicated segment round-trip test (boundary-crossing matches, unaligned blocks, dict > segment, degenerate tiny blocks).Measured (blob, CLI e2e) — the honest ledger (§11b)
Best LZ-family ratio in pz, 0.9pp under pzstd-3, still 3.3× faster decode than zstd ST — an opt-in max-ratio tier. Two known v1 bottlenecks, both measured and documented with fixes:
prevwalks across 7 segments) — same cache physics as the 4 MiB block rejection; tuning levers listed.Test plan
🤖 Generated with Claude Code