ACEAPEX 2.6.0

ACEAPEX 2.6.0 — 2026-10-30

The fast profile's level 11 (aceapex_fast -1 in lzbench) compresses faster than zxc -1 at a better ratio. Format and
decoder are unchanged: level-11 archives differ from 2.5.0, levels 12-18 are byte-identical, every older archive decodes.

• Level 11 encoder: blocks encoded straight into the output buffer with one thread, branch-free table inserts (ENC3);
  a 10-byte probe key and the literal-run step after a key hit rejected for length (SKIP10).
• lzbench on Zen 5 (Google Cloud C4D, AMD EPYC 9B45; lzbench 1cda8a2, gcc 13.3, silesia.tar, one core, median of 5, zxc in the same run): aceapex_fast -1 61.57 %, 969 / 15858 MB/s; zxc 0.15.0 -1 61.67 %,
  869 / 11135 MB/s (results/skip10-zen5-lzbench-2026-10-30.log).
• Ratio of level 11 against zxc -1: silesia 61.57 / 61.67 %, enwik8 83.37 / 89.23 %, chr1 88.39 / 73.54 %
  (results/skip10-ratio-l11-2026-10-30.md). chr1 loses: no DNA packing in the fast profile. Against 2.5.0 level 11 is
  0.19 / 0.69 / 0.72 percentage points larger on silesia / enwik8 / chr1.
• Checked: HEAD judge, fast-profile tests (new fixture text_fast_l11_16K), portability gate, round trip levels 11..18
  at 1 and 4 threads on 9 inputs, ASan+UBSan with a negative control.

Binary

aceapex-2.6.0-linux-x86_64.tar.gz (static, x86-64-v2, zstd 1.5.7; scripts/package_linux.sh, built on the tagged commit e104741).

1305f91098b4c7e3805d8f6bdf9fda5f9c46996c2a067301d5eaf913ae92e28e  aceapex-2.6.0-linux-x86_64.tar.gz
087e886a11e82e1796576d021a76b5194404193b68d3b21bd9067a49ff6f4d05  aceapex (binary inside the tarball)