wiki: v1.6.12-pre - levels renumbered, fast tier, level guard, byte-delta plane lanes, LZMA-family comparison
Getting Started: level table with the exact v1.6.12-pre Silesia sizes of all 15 rows and the new CLI/API numbering. Seq Stream Format: the delta bit of
the plane lane mode byte. Architecture, Decoder Internals, Development: the guard, delta lanes, the XDIFF_EXT gate mode, the noise rule. Benchmarks: v1.6.12-pre
against v1.6.11-pre and against xz, lzma, fast-lzma2, uf-lzma2, zpaq and RAR. Home and Roadmap: state and next candidates.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
wiki: v1.6.11-pre - decoder internals rewritten, same-session benchmarks, format rules (extra-bit widths, rep stack, exact size), roadmap
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Benchmarks/Home/Roadmap: compare with the whole zstd ladder; measured causes of the single-thread gap
The Silesia table only had zstd -1, -5 and -19. A complementary run adds the
levels in between; on one thread zstd's ladder is ahead of lz6 over most of the
range. Roadmap now records the measured causes (instructions per sequence,
stack traffic in the decode loop, LLC misses and kernel time in the encoder).
Also fixes the lizard line in Home and the ooffice "loss" in Roadmap, which the
x86 filter turned into a win in v1.6.9-pre.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
v1.6.10-pre: multithreaded command line (-T#): architecture, benchmarks, usage, roadmap
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
v1.6.9-pre: benchmarks (pinned sweep), level table, Home, Roadmap
v1.6.8-pre: benchmarks, level table (single-pass L1-L2), architecture, roadmap
v1.6.7-pre: benchmarks, level table (row hash at L3-L5), architecture, roadmap
Benchmarks: fix table cells
Benchmarks/Home: v1.6.6-pre sweep (Silesia L15 25.43%, gap to zstd -19 0.47 points)
wiki: benchmarks and progress for v1.6.5-pre
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: uf-lzma2 level 11 is 1.2.0
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: uf-lzma2 level 11; corrected AIT per-file reading
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: drop the AIT rows from the progress table (mixed bases)
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: benchmarks from the seq v2 sweep; decoder progress table
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: correct AIT numbers (parser dropped file D from lz6's totals); add the without-D comparison
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: benchmarks add uf-lzma2 (asm decoder), fastlzma2 and memlz
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: benchmarks and roadmap from the 2026-09-22 full lzbench sweep
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: project pages (architecture, seq stream format, decoder internals, benchmarks, development, roadmap)
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>