What's Changed
- zstd: two-pass decode with match-source prefetch on arm64 (largest win decoding SpeedBestCompression-class data) by @lizthegrey in #1212
- zstd: speed up better encoder hashing, primes in registers on arm64 by @lizthegrey in #1230
- zstd: reuse dictionary sequence tables in the encoder by @lizthegrey in #1233
- zstd: close most of the decode gap to libzstd on real corpora by @lizthegrey in #1232
- huff0: align the decode inner loops to 64 bytes by @lizthegrey in #1236
- zstd: keep default-encoder hash primes in registers on arm64 by @lizthegrey in #1241
- zstd: align the sequence decode and execute loops to 64 bytes by @lizthegrey in #1240
- zstd, huff0: cheaper per-block table setup by @lizthegrey in #1234
- zstd, s2, flate: smoke-test Silesia in CI, add real-corpus benchmarks by @lizthegrey in #1237
- zip: reject entries named "." in ReadDir to stop infinite fs.WalkDir by @Raghuvardhan-Reddy78 in #1249
- zstd: decode at full speed into exact-size destinations by @lizthegrey in #1246
- s2: Reject short index chunk in RemoveIndexHeaders by @Raghuvardhan-Reddy78 in #1254
- zstd, huff0: keep the encoder bit writers in registers by @lizthegrey in #1245
- huff0: reuse Scratch across inputs in the fuzzers by @lizthegrey in #1257
New Contributors
- @Raghuvardhan-Reddy78 made their first contribution in #1235
- @SulimanAbdulrazzaq made their first contribution in #1255
Full Changelog: v1.20.1...v1.20.3