Thread: ROFZ Compressor
View Single Post
  #3  
Old Today, 11:51
sajmon83's Avatar
sajmon83 sajmon83 is online now
Registered User
 
Join Date: Jun 2017
Location: Poland
Posts: 47
Thanks: 252
Thanked 57 Times in 35 Posts
sajmon83 is on a distinguished road
Thanks! Levels 1–4 (the default) keep decompression at ~200 MB/s, and the ADPCM model decodes at only ~6 MB/s per thread, so on those levels ADPCM is left to LZ. For ADPCM-heavy data use -l5…-l8 or -ds 0. 0.17.1 also fixes WAV/ADPCM headers that claim more data than the file holds.
I am working on a lighter version of the ADPCM model for -l1 … -l4

0.17.1 (stream format v17: 0.17.0 streams decode; streams made with -nw need 0.17.1)
switches for the single models and the trained models.
default level 4 (was 3)
fix: audio from a WAV / ADPCM header that claims more data than follows it (a cut or broken file) ran on over the next data and coded it as audio. Textures and images found by their headers are now taken out of the audio, and the last 32 KB of every audio stream are coded by the audio model and by LZ; where LZ wins, a binary search in 64 KB steps finds where the stream turns into other data and the audio ends there (cut-WAV test files: −2.8% at -l4, −15.5% at -l7). Normal audio codes as before.

Last edited by sajmon83; Today at 12:15.
Reply With Quote