wiki: Roadmap, v1.6.11-pre is released 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>
Getting-Started: measured ratios for levels 6/9/11/14 in v1.6.9-pre
v1.6.9-pre: benchmarks (pinned sweep), level table, Home, Roadmap
Seq Stream Format: x86 filter block (flag 0x10); Architecture: x86 branch filter
Getting-Started: measured ratios for levels 6/9/11/14 (not in the sweep)
v1.6.8-pre: benchmarks, level table (single-pass L1-L2), architecture, roadmap
Seq Stream Format: literal mode 7 (4-stream Huffman, v1.6.8-pre)
v1.6.7-pre: benchmarks, level table (row hash at L3-L5), architecture, roadmap
Getting-Started: L7-L15 encode speeds from one run
Getting-Started: level table after the L3-L7 retune
Decoder Internals: rich 8-byte entries, fill/steady/drain loop, what the latency chain means
Roadmap: state at v1.6.6-pre (done items, remaining decode/ratio/encode work)
Benchmarks: fix table cells
Benchmarks/Home: v1.6.6-pre sweep (Silesia L15 25.43%, gap to zstd -19 0.47 points)
Architecture: offset coding with low bits
Seq stream format: offset codes with low bits, varint table header; level table ratios after the offset model
Getting-Started: level table after the encode-speed work (L1-3 ~2.5x)
wiki: benchmarks and progress for v1.6.5-pre Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: level 2 is not the CLI default (-0 is) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: seq level ladder and entropy-priced parser 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: portable profile reference version (v1.6.4-pre, win32 fast-encoder fix) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wiki: BMI2 dispatch in Decoder Internals 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: seq v2 format (tANS + one backward bitstream) and decoder internals Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>