Todo

General

• Consider adding a famous arithmetic coder, for comparison
• Provide shannon limit in benchmark samples

Huff0

• Optimize Jump table (4 streams) ?
  • variable size jTable => place at the end => impact on speed
• Investigate delta coding for headers
• Consider changing naming convention for decompression
  • from x2/x4/x6 (based on mem) to x1/x2/x4 (based on nb symbols)

Huff0 Research :

• Dedicated HUF_count(), which just redirects to FSE_count(), for the sake of more consistent API ?
• forward direction ?
• dedicated bistream (for speed ?)
• Stream interleave to avoid jumptable ?
  • Seem to require branching
• Huff0 table construction : reuse previous segments ?
  • more difficult for multi-symbols

FSE

FSE Research :

• Optimize headers
• improve performance at -O2 settings
• Store symbol value into initial state value
  • if no negative impact on speed
  • but then, to detect end : Req nb of symbols to regenerate
• table reuse : differentiate table size vs used table size