Skip to content
Optimal LZ4 compressor, that produces files that decompress faster while keeping the best compression ratio
Branch: master
Clone or download
Type Name Latest commit message Commit time
Failed to load latest commit information.
VS2017 Add in-memory decompression benchmark for internal decompressor May 16, 2019
src Mention -l in usage May 20, 2019
.gitignore Support all LZ4 block sizes, and independent blocks compression Apr 12, 2019
LICENSE Update LICENSE May 11, 2019 Initial checkin Apr 8, 2019 Initial checkin Apr 8, 2019
Makefile Add in-memory decompression benchmark for internal decompressor May 16, 2019

lz4ultra -- Optimal LZ4 packer with faster decompression

lz4ultra is a command-line optimal compression utility that produces compressed files in the lz4 format created by Yann Collet.

The tool creates optimally compressed files, like lz4 in optimal compression mode ("lz4hc"), smallLZ4, blz4 and lz4x. The files created with lz4ultra decompress faster.

lz4ultra beats lz4 1.9.1 --12 --favor-decSpeed in both size and decompression speed. With enwik9 (1,000,000,000 bytes):

                                  Compr.size    Tokens      Decomp.time (μs, Core i7-6700)
lz4 1.9.1 --12 (favor ratio)      371,680,440   95,708,169  345,272
smalLZ4 1.3                       371,680,328   93,172,985  343,571
lz4ultra 1.1.3 (favor ratio)      371,687,509   85,910,002  342,159
lz4 1.9.1 --12 --favor-decSpeed   376,408,347   92,105,212  304,911
lz4ultra 1.1.3 --favor-decSpeed   376,118,380   88,521,891  289,355 <--------------

The produced files are meant to be decompressed with the lz4 tool and library. While lz4ultra includes a decompressor, it is mostly meant to verify the output of the compressor and isn't as optimized as Yann Collet's lz4 proper.

lz4ultra works by performing the usual optimal compression (using a suffix array), and then applying a forward peephole optimization pass, that breaks ties (sequences of compression commands that result in an identical number of bytes added to the compressed data), in favor of outputting less commands.

In other words, it will go over the compressed data before it is flattened into the output bytestream, and possibly replace match->match and match->literal sequences by a literal->match or just literal sequences, if the output size is unchanged.

Everything else being equal, having less commands speeds up decompression as the decompressor is required to do less mode switches.

The peephole optimizer is located in src/shrink.c, as lz4ultra_optimize_command_count(). This pass is extremely fast and cache-friendly as it just goes over the compressed data once. The logic isn't tied to compressing with a suffix array, it can be adapted to any compressor.

lz4ultra is an offshoot of the lzsa compressor as we thought it would be interesting to the wider compression community, to users of LZ4 that compress data once and require the fastest decompression time without a trade-off in compression ratio, and also to retrocomputing developers as this will speed decompression up further on 8/16-bit systems.

The tool defaults to 4 Mb blocks with inter-block dependencies but can be configured to output all of the LZ4 block sizes (64 Kb to 4 Mb) and to compress independent blocks, using command-line switches.

lz4ultra is developed by Emmanuel Marty with the help of spke.

You can’t perform that action at this time.