You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Client throughput: compress and decompress at least as fast as xz #105
On 1 MB samples, the client from #104 takes 1.7 s to compress and 1.9 s to decompress Rust source, and 5.4 s / 5.3 s for an ELF binary, which is about 0.2–0.6 MB/s. xz -9 compresses at a few MB/s and decompresses at tens of MB/s; gzip is faster still.
Work
Profile the exported Rust (generated/rust/uorc_{compress,decompress}) on the benchmark corpus.
Known hot spots: the SHA-256 chain, the canonical re-encode check on decode, and whole-buffer Bytes copies in the exported code.
Where the fix belongs in LexLean's Rust output, PR it upstream and pin to the patch until it is released.
Done when
On the release benchmark (#51), compress throughput is at least xz -9's and decompress throughput is at least xz -d's.
Activity
uorc: native codec fast path and framed streaming (#105, #106)
uorc: native codec fast path and framed streaming (#105, #106)
uorc: native codec fast path (faster than xz) and framed streaming of…
uorc: ncm context-mixing level (--best) and the standard kernel basis (…
uorc: ncm context-mixing level (--best) and the standard kernel basis…
uorc: window frames, so matches cross 8 MiB frame boundaries (#117)
uorc: ncm v2, 3.9% smaller on Canterbury (--best 355,559 B vs xz -9e …
ulz: hash4 chain finder, 3.7x faster and 2.1% smaller on a 128 MB tar…
uorc: ncm-only compress, ncm window frames, x86 filter, parallel bloc…
uorc: store identical frames once, as κ-labelled reference frames (#124)