LZARI

[Bulat Ziganshin]

A bit offtopic maybe - how difficult is it to make a run-time packer?

i think that upx sources may be reused for this purpose - we should just look how they incorporated lzma compression

[LovePimple]

Quick test LZARI demo...

Test Machine: AMD Sempron 2400+

A10.jpg > 839,690
AcroRd32.exe > 1,793,261
english.dic > 1,394,632
FlashMX.pdf > 3,807,603
FP.LOG > 1,294,712
MSO97.DLL > 2,235,370
ohs.doc > 906,840
rafale.bmp > 1,380,952
vcfiu.hlp > 902,798
world95.txt > 834,023

Total = 15,389,881 bytes


ENWIK8:

Compressed Size = 38,719,322 bytes

Compression Time = 00:05:53.359

Im writing a new engine from scratch

Im looking forward to testing the new engine!

[encode]

Im looking forward to testing the new engine!

Its a new compressor called lzenc (lz encoder) - a new polygon for experiments. The main goal of this compressor is make the LZ-output encoder as simple as it possible (throw away PPM) and keep good enough compression ratio. Of course, decompression will be fast.

[encode]

What I have at the moment:
A hybryd scheme of LZ77 a la LZ77-PM / ROLZ2. Thus a new compressor will be named lzpm.
New scheme finds a much larger number of matches compared to the QUAD and has a simpler LZ-output encoder as a result we have a faster decompression. In addition, this new algorithm:
+ More stable on already compressed data (we can compress A10.jpg)
+ Needs less memory for decompression
However, the compression ratio is not so good in average compared to the QUAD and the compression is really slow even with Lazy Matching, with Flexible Parsing compression is super-slow (much like PAQ6, I guess). Anyway, with Flexible Parsing new scheme already has a higher text compression.
This engine can be regarder as an evolution of my LZ compressors. To be honest, new LZ engine is really stronger than QUAD, but due to the simpler LZ-output coding, in addition to the non-optimal parsing we got worser compression. To increase compression speed I must implement some kind of tree structure...
The work in progress...

[encode]

One important thing - I finally implemented the proper match searching via hashing. However, new scheme needs + 64 MB for compression. I think it's okay. Thanks to the LZ77-PM inventors - they gave me a nice idea...