feat(rust): port optimal binary-tree updates
Move the binary-tree table maintenance shared by optimal compression strategies into Rust. A small C projection preserves the private match-state boundary; the dynamic-programming parser and match selection remain C-owned for now. The Rust implementation covers prefix and external-dictionary tree updates and preserves the optional C predictor path. The component map now distinguishes this migrated tree layer from the remaining optimal parser. Test Plan: - cargo test --all-targets - cargo test --target i686-unknown-linux-gnu --all-targets - cargo clippy && cargo clippy --benches && cargo clippy --tests - cargo +nightly fmt - make -B -C tests -j2 fuzzer zstreamtest invalidDictionaries - ./tests/fuzzer -s5346 -i1 --no-big-tests - ./tests/zstreamtest -i3000 -s334462 - ./tests/invalidDictionaries - make -B -C tests fuzzer32 MOREFLAGS=-DZSTD_C_PREDICT - ./tests/fuzzer32 -s5346 -i1 --no-big-tests - byte-compare C and Rust-control CLI output at levels 13, 16, 19, and 22 for random, patterned, zero, and dictionary inputs with and without ZSTD_C_PREDICT Refs: rust/README.md
This commit is contained in:
+50
-138
@@ -16,6 +16,53 @@
|
||||
|| !defined(ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR) \
|
||||
|| !defined(ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR)
|
||||
|
||||
/* The Rust tree updater receives this leaf view instead of the private,
|
||||
* configuration-dependent ZSTD_MatchState_t layout. */
|
||||
typedef struct {
|
||||
U32* hashTable;
|
||||
U32* chainTable;
|
||||
const BYTE* base;
|
||||
const BYTE* dictBase;
|
||||
U32 dictLimit;
|
||||
U32 lowLimit;
|
||||
U32 loadedDictEnd;
|
||||
U32* nextToUpdate;
|
||||
U32 hashLog;
|
||||
U32 chainLog;
|
||||
U32 searchLog;
|
||||
U32 windowLog;
|
||||
int useCPredict;
|
||||
} ZSTD_RustOptTreeState;
|
||||
|
||||
static ZSTD_RustOptTreeState ZSTD_rustOptTreeState(
|
||||
ZSTD_MatchState_t* const ms)
|
||||
{
|
||||
ZSTD_RustOptTreeState state;
|
||||
state.hashTable = ms->hashTable;
|
||||
state.chainTable = ms->chainTable;
|
||||
state.base = ms->window.base;
|
||||
state.dictBase = ms->window.dictBase;
|
||||
state.dictLimit = ms->window.dictLimit;
|
||||
state.lowLimit = ms->window.lowLimit;
|
||||
state.loadedDictEnd = ms->loadedDictEnd;
|
||||
state.nextToUpdate = &ms->nextToUpdate;
|
||||
state.hashLog = ms->cParams.hashLog;
|
||||
state.chainLog = ms->cParams.chainLog;
|
||||
state.searchLog = ms->cParams.searchLog;
|
||||
state.windowLog = ms->cParams.windowLog;
|
||||
#ifdef ZSTD_C_PREDICT
|
||||
state.useCPredict = 1;
|
||||
#else
|
||||
state.useCPredict = 0;
|
||||
#endif
|
||||
return state;
|
||||
}
|
||||
|
||||
void ZSTD_rust_opt_updateTreeInternal(
|
||||
ZSTD_RustOptTreeState* state,
|
||||
const void* ip, const void* iend,
|
||||
U32 mls, int extDict);
|
||||
|
||||
#define ZSTD_LITFREQ_ADD 2 /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
|
||||
#define ZSTD_MAX_PRICE (1<<30)
|
||||
|
||||
@@ -433,130 +480,6 @@ U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_MatchState_t* ms,
|
||||
/*-*************************************
|
||||
* Binary Tree search
|
||||
***************************************/
|
||||
/** ZSTD_insertBt1() : add one or multiple positions to tree.
|
||||
* @param ip assumed <= iend-8 .
|
||||
* @param target The target of ZSTD_updateTree_internal() - we are filling to this position
|
||||
* @return : nb of positions added */
|
||||
static
|
||||
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
|
||||
U32 ZSTD_insertBt1(
|
||||
const ZSTD_MatchState_t* ms,
|
||||
const BYTE* const ip, const BYTE* const iend,
|
||||
U32 const target,
|
||||
U32 const mls, const int extDict)
|
||||
{
|
||||
const ZSTD_compressionParameters* const cParams = &ms->cParams;
|
||||
U32* const hashTable = ms->hashTable;
|
||||
U32 const hashLog = cParams->hashLog;
|
||||
size_t const h = ZSTD_hashPtr(ip, hashLog, mls);
|
||||
U32* const bt = ms->chainTable;
|
||||
U32 const btLog = cParams->chainLog - 1;
|
||||
U32 const btMask = (1 << btLog) - 1;
|
||||
U32 matchIndex = hashTable[h];
|
||||
size_t commonLengthSmaller=0, commonLengthLarger=0;
|
||||
const BYTE* const base = ms->window.base;
|
||||
const BYTE* const dictBase = ms->window.dictBase;
|
||||
const U32 dictLimit = ms->window.dictLimit;
|
||||
const BYTE* const dictEnd = dictBase + dictLimit;
|
||||
const BYTE* const prefixStart = base + dictLimit;
|
||||
const BYTE* match;
|
||||
const U32 curr = (U32)(ip-base);
|
||||
const U32 btLow = btMask >= curr ? 0 : curr - btMask;
|
||||
U32* smallerPtr = bt + 2*(curr&btMask);
|
||||
U32* largerPtr = smallerPtr + 1;
|
||||
U32 dummy32; /* to be nullified at the end */
|
||||
/* windowLow is based on target because
|
||||
* we only need positions that will be in the window at the end of the tree update.
|
||||
*/
|
||||
U32 const windowLow = ZSTD_getLowestMatchIndex(ms, target, cParams->windowLog);
|
||||
U32 matchEndIdx = curr+8+1;
|
||||
size_t bestLength = 8;
|
||||
U32 nbCompares = 1U << cParams->searchLog;
|
||||
#ifdef ZSTD_C_PREDICT
|
||||
U32 predictedSmall = *(bt + 2*((curr-1)&btMask) + 0);
|
||||
U32 predictedLarge = *(bt + 2*((curr-1)&btMask) + 1);
|
||||
predictedSmall += (predictedSmall>0);
|
||||
predictedLarge += (predictedLarge>0);
|
||||
#endif /* ZSTD_C_PREDICT */
|
||||
|
||||
DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
|
||||
|
||||
assert(curr <= target);
|
||||
assert(ip <= iend-8); /* required for h calculation */
|
||||
hashTable[h] = curr; /* Update Hash Table */
|
||||
|
||||
assert(windowLow > 0);
|
||||
for (; nbCompares && (matchIndex >= windowLow); --nbCompares) {
|
||||
U32* const nextPtr = bt + 2*(matchIndex & btMask);
|
||||
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
|
||||
assert(matchIndex < curr);
|
||||
|
||||
#ifdef ZSTD_C_PREDICT /* note : can create issues when hlog small <= 11 */
|
||||
const U32* predictPtr = bt + 2*((matchIndex-1) & btMask); /* written this way, as bt is a roll buffer */
|
||||
if (matchIndex == predictedSmall) {
|
||||
/* no need to check length, result known */
|
||||
*smallerPtr = matchIndex;
|
||||
if (matchIndex <= btLow) { smallerPtr=&dummy32; break; } /* beyond tree size, stop the search */
|
||||
smallerPtr = nextPtr+1; /* new "smaller" => larger of match */
|
||||
matchIndex = nextPtr[1]; /* new matchIndex larger than previous (closer to current) */
|
||||
predictedSmall = predictPtr[1] + (predictPtr[1]>0);
|
||||
continue;
|
||||
}
|
||||
if (matchIndex == predictedLarge) {
|
||||
*largerPtr = matchIndex;
|
||||
if (matchIndex <= btLow) { largerPtr=&dummy32; break; } /* beyond tree size, stop the search */
|
||||
largerPtr = nextPtr;
|
||||
matchIndex = nextPtr[0];
|
||||
predictedLarge = predictPtr[0] + (predictPtr[0]>0);
|
||||
continue;
|
||||
}
|
||||
#endif
|
||||
|
||||
if (!extDict || (matchIndex+matchLength >= dictLimit)) {
|
||||
assert(matchIndex+matchLength >= dictLimit); /* might be wrong if actually extDict */
|
||||
match = base + matchIndex;
|
||||
matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
|
||||
} else {
|
||||
match = dictBase + matchIndex;
|
||||
matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
|
||||
if (matchIndex+matchLength >= dictLimit)
|
||||
match = base + matchIndex; /* to prepare for next usage of match[matchLength] */
|
||||
}
|
||||
|
||||
if (matchLength > bestLength) {
|
||||
bestLength = matchLength;
|
||||
if (matchLength > matchEndIdx - matchIndex)
|
||||
matchEndIdx = matchIndex + (U32)matchLength;
|
||||
}
|
||||
|
||||
if (ip+matchLength == iend) { /* equal : no way to know if inf or sup */
|
||||
break; /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
|
||||
}
|
||||
|
||||
if (match[matchLength] < ip[matchLength]) { /* necessarily within buffer */
|
||||
/* match is smaller than current */
|
||||
*smallerPtr = matchIndex; /* update smaller idx */
|
||||
commonLengthSmaller = matchLength; /* all smaller will now have at least this guaranteed common length */
|
||||
if (matchIndex <= btLow) { smallerPtr=&dummy32; break; } /* beyond tree size, stop searching */
|
||||
smallerPtr = nextPtr+1; /* new "candidate" => larger than match, which was smaller than target */
|
||||
matchIndex = nextPtr[1]; /* new matchIndex, larger than previous and closer to current */
|
||||
} else {
|
||||
/* match is larger than current */
|
||||
*largerPtr = matchIndex;
|
||||
commonLengthLarger = matchLength;
|
||||
if (matchIndex <= btLow) { largerPtr=&dummy32; break; } /* beyond tree size, stop searching */
|
||||
largerPtr = nextPtr;
|
||||
matchIndex = nextPtr[0];
|
||||
} }
|
||||
|
||||
*smallerPtr = *largerPtr = 0;
|
||||
{ U32 positions = 0;
|
||||
if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384)); /* speed optimization */
|
||||
assert(matchEndIdx > curr + 8);
|
||||
return MAX(positions, matchEndIdx - (curr + 8));
|
||||
}
|
||||
}
|
||||
|
||||
FORCE_INLINE_TEMPLATE
|
||||
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
|
||||
void ZSTD_updateTree_internal(
|
||||
@@ -564,20 +487,9 @@ void ZSTD_updateTree_internal(
|
||||
const BYTE* const ip, const BYTE* const iend,
|
||||
const U32 mls, const ZSTD_dictMode_e dictMode)
|
||||
{
|
||||
const BYTE* const base = ms->window.base;
|
||||
U32 const target = (U32)(ip - base);
|
||||
U32 idx = ms->nextToUpdate;
|
||||
DEBUGLOG(7, "ZSTD_updateTree_internal, from %u to %u (dictMode:%u)",
|
||||
idx, target, dictMode);
|
||||
|
||||
while(idx < target) {
|
||||
U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, target, mls, dictMode == ZSTD_extDict);
|
||||
assert(idx < (U32)(idx + forward));
|
||||
idx += forward;
|
||||
}
|
||||
assert((size_t)(ip - base) <= (size_t)(U32)(-1));
|
||||
assert((size_t)(iend - base) <= (size_t)(U32)(-1));
|
||||
ms->nextToUpdate = target;
|
||||
ZSTD_RustOptTreeState state = ZSTD_rustOptTreeState(ms);
|
||||
ZSTD_rust_opt_updateTreeInternal(&state, ip, iend, mls,
|
||||
dictMode == ZSTD_extDict);
|
||||
}
|
||||
|
||||
void ZSTD_updateTree(ZSTD_MatchState_t* ms, const BYTE* ip, const BYTE* iend) {
|
||||
|
||||
Reference in New Issue
Block a user