ddidderr/zstd-rs
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

implemented fractional bit cost evaluation

Browse Source 
for FSE symbols.

While it seems to work, the gains are negligible compared to rough maxNbBits evaluation.
There are even a few losses sometimes, that still need to be explained.
Furthermode, there are still cases where btlazy2 does a better job than btopt,
which seems rather strange too.
This commit is contained in:
Yann Collet
2018-05-08 17:43:13 -07:00
parent 1aff63b114
commit ba2ad9b6b9
3 changed files with 24 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lib/common/fse.h
+14 -1
View File
@@ -577,10 +577,23 @@ MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePt
MEM_STATIC U32 FSE_getMaxNbBits(const FSE_symbolCompressionTransform* symbolTT, U32 symbolValue)
{
assert(symbolValue <= FSE_MAX_SYMBOL_VALUE);
return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16;
}
/* FSE_bitCost_b256() :
* Approximate symbol cost,
* provide fractional value, using fixed-point format (8 bit) */
MEM_STATIC U32 FSE_bitCost_b256(const FSE_symbolCompressionTransform* symbolTT, U32 tableLog, U32 symbolValue)
{
U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16;
U32 const threshold = (minNbBits+1) << 16;
assert(symbolTT[symbolValue].deltaNbBits + (1<<tableLog) <= threshold);
U32 const deltaFromThreshold = threshold - (symbolTT[symbolValue].deltaNbBits + (1 << tableLog));
U32 const normalizedDeltaFromThreshold = (deltaFromThreshold << 8) >> tableLog; /* linear interpolation (very approximate) */
assert(normalizedDeltaFromThreshold <= 256);
return (minNbBits+1)*256 - normalizedDeltaFromThreshold;
}
/* ====== Decompression ====== */