feat(rust): port Huffman compression
Move Huffman table construction, table serialization, and one- and four-stream payload encoding from huf_compress.c into the Rust compatibility archive. The declaration-only C shim preserves the existing internal ABI, so the remaining C compressor can call the migrated implementation unchanged. The translation keeps CTable layouts, workspace checks, repeat-table selection, and bitstream output compatible with the original encoder. This lets native C consumers exercise the Rust implementation through libzstd. Test Plan: - cargo clippy, cargo clippy --benches, and cargo clippy --tests - cargo +nightly fmt and cargo test --all-targets (88 passed) - cargo check --no-default-features --features compression - rebuild and run tests fuzzer, zstreamtest, and invalidDictionaries - compare tables, headers, and 1X/4X streams with a renamed pristine C build Refs: rust/README.md component map
This commit is contained in:
+2
-1463
@@ -1,1464 +1,3 @@
|
||||
/* ******************************************************************
|
||||
* Huffman encoder, part of New Generation Entropy library
|
||||
* Copyright (c) Meta Platforms, Inc. and affiliates.
|
||||
*
|
||||
* You can contact the author at :
|
||||
* - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
|
||||
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
|
||||
*
|
||||
* This source code is licensed under both the BSD-style license (found in the
|
||||
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
|
||||
* in the COPYING file in the root directory of this source tree).
|
||||
* You may select, at your option, one of the above-listed licenses.
|
||||
****************************************************************** */
|
||||
|
||||
/* **************************************************************
|
||||
* Compiler specifics
|
||||
****************************************************************/
|
||||
#ifdef _MSC_VER /* Visual Studio */
|
||||
# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
|
||||
#endif
|
||||
|
||||
|
||||
/* **************************************************************
|
||||
* Includes
|
||||
****************************************************************/
|
||||
#include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memset */
|
||||
#include "../common/compiler.h"
|
||||
#include "../common/bitstream.h"
|
||||
#include "hist.h"
|
||||
#define FSE_STATIC_LINKING_ONLY /* FSE_optimalTableLog_internal */
|
||||
#include "../common/fse.h" /* header compression */
|
||||
/* Huffman compression is implemented in rust/src/huf_compress.rs. */
|
||||
#define FSE_STATIC_LINKING_ONLY
|
||||
#include "../common/huf.h"
|
||||
#include "../common/error_private.h"
|
||||
#include "../common/bits.h" /* ZSTD_highbit32 */
|
||||
|
||||
|
||||
/* **************************************************************
|
||||
* Error Management
|
||||
****************************************************************/
|
||||
#define HUF_isError ERR_isError
|
||||
#define HUF_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c) /* use only *after* variable declarations */
|
||||
|
||||
|
||||
/* **************************************************************
|
||||
* Required declarations
|
||||
****************************************************************/
|
||||
typedef struct nodeElt_s {
|
||||
U32 count;
|
||||
U16 parent;
|
||||
BYTE byte;
|
||||
BYTE nbBits;
|
||||
} nodeElt;
|
||||
|
||||
|
||||
/* **************************************************************
|
||||
* Debug Traces
|
||||
****************************************************************/
|
||||
|
||||
#if DEBUGLEVEL >= 2
|
||||
|
||||
static size_t showU32(const U32* arr, size_t size)
|
||||
{
|
||||
size_t u;
|
||||
for (u=0; u<size; u++) {
|
||||
RAWLOG(6, " %u", arr[u]); (void)arr;
|
||||
}
|
||||
RAWLOG(6, " \n");
|
||||
return size;
|
||||
}
|
||||
|
||||
static size_t HUF_getNbBits(HUF_CElt elt);
|
||||
|
||||
static size_t showCTableBits(const HUF_CElt* ctable, size_t size)
|
||||
{
|
||||
size_t u;
|
||||
for (u=0; u<size; u++) {
|
||||
RAWLOG(6, " %zu", HUF_getNbBits(ctable[u])); (void)ctable;
|
||||
}
|
||||
RAWLOG(6, " \n");
|
||||
return size;
|
||||
|
||||
}
|
||||
|
||||
static size_t showHNodeSymbols(const nodeElt* hnode, size_t size)
|
||||
{
|
||||
size_t u;
|
||||
for (u=0; u<size; u++) {
|
||||
RAWLOG(6, " %u", hnode[u].byte); (void)hnode;
|
||||
}
|
||||
RAWLOG(6, " \n");
|
||||
return size;
|
||||
}
|
||||
|
||||
static size_t showHNodeBits(const nodeElt* hnode, size_t size)
|
||||
{
|
||||
size_t u;
|
||||
for (u=0; u<size; u++) {
|
||||
RAWLOG(6, " %u", hnode[u].nbBits); (void)hnode;
|
||||
}
|
||||
RAWLOG(6, " \n");
|
||||
return size;
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
/* *******************************************************
|
||||
* HUF : Huffman block compression
|
||||
*********************************************************/
|
||||
#define HUF_WORKSPACE_MAX_ALIGNMENT 8
|
||||
|
||||
static void* HUF_alignUpWorkspace(void* workspace, size_t* workspaceSizePtr, size_t align)
|
||||
{
|
||||
size_t const mask = align - 1;
|
||||
size_t const rem = (size_t)workspace & mask;
|
||||
size_t const add = (align - rem) & mask;
|
||||
BYTE* const aligned = (BYTE*)workspace + add;
|
||||
assert((align & (align - 1)) == 0); /* pow 2 */
|
||||
assert(align <= HUF_WORKSPACE_MAX_ALIGNMENT);
|
||||
if (*workspaceSizePtr >= add) {
|
||||
assert(add < align);
|
||||
assert(((size_t)aligned & mask) == 0);
|
||||
*workspaceSizePtr -= add;
|
||||
return aligned;
|
||||
} else {
|
||||
*workspaceSizePtr = 0;
|
||||
return NULL;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
/* HUF_compressWeights() :
|
||||
* Same as FSE_compress(), but dedicated to huff0's weights compression.
|
||||
* The use case needs much less stack memory.
|
||||
* Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
|
||||
*/
|
||||
#define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
|
||||
|
||||
typedef struct {
|
||||
FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
|
||||
U32 scratchBuffer[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(HUF_TABLELOG_MAX, MAX_FSE_TABLELOG_FOR_HUFF_HEADER)];
|
||||
unsigned count[HUF_TABLELOG_MAX+1];
|
||||
S16 norm[HUF_TABLELOG_MAX+1];
|
||||
} HUF_CompressWeightsWksp;
|
||||
|
||||
static size_t
|
||||
HUF_compressWeights(void* dst, size_t dstSize,
|
||||
const void* weightTable, size_t wtSize,
|
||||
void* workspace, size_t workspaceSize)
|
||||
{
|
||||
BYTE* const ostart = (BYTE*) dst;
|
||||
BYTE* op = ostart;
|
||||
BYTE* const oend = ostart + dstSize;
|
||||
|
||||
unsigned maxSymbolValue = HUF_TABLELOG_MAX;
|
||||
U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
|
||||
HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
|
||||
|
||||
if (workspaceSize < sizeof(HUF_CompressWeightsWksp)) return ERROR(GENERIC);
|
||||
|
||||
/* init conditions */
|
||||
if (wtSize <= 1) return 0; /* Not compressible */
|
||||
|
||||
/* Scan input and build symbol stats */
|
||||
{ unsigned const maxCount = HIST_count_simple(wksp->count, &maxSymbolValue, weightTable, wtSize); /* never fails */
|
||||
if (maxCount == wtSize) return 1; /* only a single symbol in src : rle */
|
||||
if (maxCount == 1) return 0; /* each symbol present maximum once => not compressible */
|
||||
}
|
||||
|
||||
tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
|
||||
CHECK_F( FSE_normalizeCount(wksp->norm, tableLog, wksp->count, wtSize, maxSymbolValue, /* useLowProbCount */ 0) );
|
||||
|
||||
/* Write table description header */
|
||||
{ CHECK_V_F(hSize, FSE_writeNCount(op, (size_t)(oend-op), wksp->norm, maxSymbolValue, tableLog) );
|
||||
op += hSize;
|
||||
}
|
||||
|
||||
/* Compress */
|
||||
CHECK_F( FSE_buildCTable_wksp(wksp->CTable, wksp->norm, maxSymbolValue, tableLog, wksp->scratchBuffer, sizeof(wksp->scratchBuffer)) );
|
||||
{ CHECK_V_F(cSize, FSE_compress_usingCTable(op, (size_t)(oend - op), weightTable, wtSize, wksp->CTable) );
|
||||
if (cSize == 0) return 0; /* not enough space for compressed data */
|
||||
op += cSize;
|
||||
}
|
||||
|
||||
return (size_t)(op-ostart);
|
||||
}
|
||||
|
||||
static size_t HUF_getNbBits(HUF_CElt elt)
|
||||
{
|
||||
return elt & 0xFF;
|
||||
}
|
||||
|
||||
static size_t HUF_getNbBitsFast(HUF_CElt elt)
|
||||
{
|
||||
return elt;
|
||||
}
|
||||
|
||||
static size_t HUF_getValue(HUF_CElt elt)
|
||||
{
|
||||
return elt & ~(size_t)0xFF;
|
||||
}
|
||||
|
||||
static size_t HUF_getValueFast(HUF_CElt elt)
|
||||
{
|
||||
return elt;
|
||||
}
|
||||
|
||||
static void HUF_setNbBits(HUF_CElt* elt, size_t nbBits)
|
||||
{
|
||||
assert(nbBits <= HUF_TABLELOG_ABSOLUTEMAX);
|
||||
*elt = nbBits;
|
||||
}
|
||||
|
||||
static void HUF_setValue(HUF_CElt* elt, size_t value)
|
||||
{
|
||||
size_t const nbBits = HUF_getNbBits(*elt);
|
||||
if (nbBits > 0) {
|
||||
assert((value >> nbBits) == 0);
|
||||
*elt |= value << (sizeof(HUF_CElt) * 8 - nbBits);
|
||||
}
|
||||
}
|
||||
|
||||
HUF_CTableHeader HUF_readCTableHeader(HUF_CElt const* ctable)
|
||||
{
|
||||
HUF_CTableHeader header;
|
||||
ZSTD_memcpy(&header, ctable, sizeof(header));
|
||||
return header;
|
||||
}
|
||||
|
||||
static void HUF_writeCTableHeader(HUF_CElt* ctable, U32 tableLog, U32 maxSymbolValue)
|
||||
{
|
||||
HUF_CTableHeader header;
|
||||
HUF_STATIC_ASSERT(sizeof(ctable[0]) == sizeof(header));
|
||||
ZSTD_memset(&header, 0, sizeof(header));
|
||||
assert(tableLog < 256);
|
||||
header.tableLog = (BYTE)tableLog;
|
||||
assert(maxSymbolValue < 256);
|
||||
header.maxSymbolValue = (BYTE)maxSymbolValue;
|
||||
ZSTD_memcpy(ctable, &header, sizeof(header));
|
||||
}
|
||||
|
||||
typedef struct {
|
||||
HUF_CompressWeightsWksp wksp;
|
||||
BYTE bitsToWeight[HUF_TABLELOG_MAX + 1]; /* precomputed conversion table */
|
||||
BYTE huffWeight[HUF_SYMBOLVALUE_MAX];
|
||||
} HUF_WriteCTableWksp;
|
||||
|
||||
size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
|
||||
const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog,
|
||||
void* workspace, size_t workspaceSize)
|
||||
{
|
||||
HUF_CElt const* const ct = CTable + 1;
|
||||
BYTE* op = (BYTE*)dst;
|
||||
U32 n;
|
||||
HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
|
||||
|
||||
HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE >= sizeof(HUF_WriteCTableWksp));
|
||||
|
||||
assert(HUF_readCTableHeader(CTable).maxSymbolValue == maxSymbolValue);
|
||||
assert(HUF_readCTableHeader(CTable).tableLog == huffLog);
|
||||
|
||||
/* check conditions */
|
||||
if (workspaceSize < sizeof(HUF_WriteCTableWksp)) return ERROR(GENERIC);
|
||||
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
|
||||
|
||||
/* convert to weight */
|
||||
wksp->bitsToWeight[0] = 0;
|
||||
for (n=1; n<huffLog+1; n++)
|
||||
wksp->bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
|
||||
for (n=0; n<maxSymbolValue; n++)
|
||||
wksp->huffWeight[n] = wksp->bitsToWeight[HUF_getNbBits(ct[n])];
|
||||
|
||||
/* attempt weights compression by FSE */
|
||||
if (maxDstSize < 1) return ERROR(dstSize_tooSmall);
|
||||
{ CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, wksp->huffWeight, maxSymbolValue, &wksp->wksp, sizeof(wksp->wksp)) );
|
||||
if ((hSize>1) & (hSize < maxSymbolValue/2)) { /* FSE compressed */
|
||||
op[0] = (BYTE)hSize;
|
||||
return hSize+1;
|
||||
} }
|
||||
|
||||
/* write raw values as 4-bits (max : 15) */
|
||||
if (maxSymbolValue > (256-128)) return ERROR(GENERIC); /* should not happen : likely means source cannot be compressed */
|
||||
if (((maxSymbolValue+1)/2) + 1 > maxDstSize) return ERROR(dstSize_tooSmall); /* not enough space within dst buffer */
|
||||
op[0] = (BYTE)(128 /*special case*/ + (maxSymbolValue-1));
|
||||
wksp->huffWeight[maxSymbolValue] = 0; /* to be sure it doesn't cause msan issue in final combination */
|
||||
for (n=0; n<maxSymbolValue; n+=2)
|
||||
op[(n/2)+1] = (BYTE)((wksp->huffWeight[n] << 4) + wksp->huffWeight[n+1]);
|
||||
return ((maxSymbolValue+1)/2) + 1;
|
||||
}
|
||||
|
||||
|
||||
size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* hasZeroWeights)
|
||||
{
|
||||
BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1]; /* init not required, even though some static analyzer may complain */
|
||||
U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1]; /* large enough for values from 0 to 16 */
|
||||
U32 tableLog = 0;
|
||||
U32 nbSymbols = 0;
|
||||
HUF_CElt* const ct = CTable + 1;
|
||||
|
||||
/* get symbol weights */
|
||||
CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
|
||||
*hasZeroWeights = (rankVal[0] > 0);
|
||||
|
||||
/* check result */
|
||||
if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
|
||||
if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
|
||||
|
||||
*maxSymbolValuePtr = nbSymbols - 1;
|
||||
|
||||
HUF_writeCTableHeader(CTable, tableLog, *maxSymbolValuePtr);
|
||||
|
||||
/* Prepare base value per rank */
|
||||
{ U32 n, nextRankStart = 0;
|
||||
for (n=1; n<=tableLog; n++) {
|
||||
U32 curr = nextRankStart;
|
||||
nextRankStart += (rankVal[n] << (n-1));
|
||||
rankVal[n] = curr;
|
||||
} }
|
||||
|
||||
/* fill nbBits */
|
||||
{ U32 n; for (n=0; n<nbSymbols; n++) {
|
||||
const U32 w = huffWeight[n];
|
||||
HUF_setNbBits(ct + n, (BYTE)(tableLog + 1 - w) & -(w != 0));
|
||||
} }
|
||||
|
||||
/* fill val */
|
||||
{ U16 nbPerRank[HUF_TABLELOG_MAX+2] = {0}; /* support w=0=>n=tableLog+1 */
|
||||
U16 valPerRank[HUF_TABLELOG_MAX+2] = {0};
|
||||
{ U32 n; for (n=0; n<nbSymbols; n++) nbPerRank[HUF_getNbBits(ct[n])]++; }
|
||||
/* determine stating value per rank */
|
||||
valPerRank[tableLog+1] = 0; /* for w==0 */
|
||||
{ U16 min = 0;
|
||||
U32 n; for (n=tableLog; n>0; n--) { /* start at n=tablelog <-> w=1 */
|
||||
valPerRank[n] = min; /* get starting value within each rank */
|
||||
min += nbPerRank[n];
|
||||
min >>= 1;
|
||||
} }
|
||||
/* assign value within rank, symbol order */
|
||||
{ U32 n; for (n=0; n<nbSymbols; n++) HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); }
|
||||
}
|
||||
|
||||
return readSize;
|
||||
}
|
||||
|
||||
U32 HUF_getNbBitsFromCTable(HUF_CElt const* CTable, U32 symbolValue)
|
||||
{
|
||||
const HUF_CElt* const ct = CTable + 1;
|
||||
assert(symbolValue <= HUF_SYMBOLVALUE_MAX);
|
||||
if (symbolValue > HUF_readCTableHeader(CTable).maxSymbolValue)
|
||||
return 0;
|
||||
return (U32)HUF_getNbBits(ct[symbolValue]);
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* HUF_setMaxHeight():
|
||||
* Try to enforce @targetNbBits on the Huffman tree described in @huffNode.
|
||||
*
|
||||
* It attempts to convert all nodes with nbBits > @targetNbBits
|
||||
* to employ @targetNbBits instead. Then it adjusts the tree
|
||||
* so that it remains a valid canonical Huffman tree.
|
||||
*
|
||||
* @pre The sum of the ranks of each symbol == 2^largestBits,
|
||||
* where largestBits == huffNode[lastNonNull].nbBits.
|
||||
* @post The sum of the ranks of each symbol == 2^largestBits,
|
||||
* where largestBits is the return value (expected <= targetNbBits).
|
||||
*
|
||||
* @param huffNode The Huffman tree modified in place to enforce targetNbBits.
|
||||
* It's presumed sorted, from most frequent to rarest symbol.
|
||||
* @param lastNonNull The symbol with the lowest count in the Huffman tree.
|
||||
* @param targetNbBits The allowed number of bits, which the Huffman tree
|
||||
* may not respect. After this function the Huffman tree will
|
||||
* respect targetNbBits.
|
||||
* @return The maximum number of bits of the Huffman tree after adjustment.
|
||||
*/
|
||||
static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 targetNbBits)
|
||||
{
|
||||
const U32 largestBits = huffNode[lastNonNull].nbBits;
|
||||
/* early exit : no elt > targetNbBits, so the tree is already valid. */
|
||||
if (largestBits <= targetNbBits) return largestBits;
|
||||
|
||||
DEBUGLOG(5, "HUF_setMaxHeight (targetNbBits = %u)", targetNbBits);
|
||||
|
||||
/* there are several too large elements (at least >= 2) */
|
||||
{ int totalCost = 0;
|
||||
const U32 baseCost = 1 << (largestBits - targetNbBits);
|
||||
int n = (int)lastNonNull;
|
||||
|
||||
/* Adjust any ranks > targetNbBits to targetNbBits.
|
||||
* Compute totalCost, which is how far the sum of the ranks is
|
||||
* we are over 2^largestBits after adjust the offending ranks.
|
||||
*/
|
||||
while (huffNode[n].nbBits > targetNbBits) {
|
||||
totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
|
||||
huffNode[n].nbBits = (BYTE)targetNbBits;
|
||||
n--;
|
||||
}
|
||||
/* n stops at huffNode[n].nbBits <= targetNbBits */
|
||||
assert(huffNode[n].nbBits <= targetNbBits);
|
||||
/* n end at index of smallest symbol using < targetNbBits */
|
||||
while (huffNode[n].nbBits == targetNbBits) --n;
|
||||
|
||||
/* renorm totalCost from 2^largestBits to 2^targetNbBits
|
||||
* note : totalCost is necessarily a multiple of baseCost */
|
||||
assert(((U32)totalCost & (baseCost - 1)) == 0);
|
||||
totalCost >>= (largestBits - targetNbBits);
|
||||
assert(totalCost > 0);
|
||||
|
||||
/* repay normalized cost */
|
||||
{ U32 const noSymbol = 0xF0F0F0F0;
|
||||
U32 rankLast[HUF_TABLELOG_MAX+2];
|
||||
|
||||
/* Get pos of last (smallest = lowest cum. count) symbol per rank */
|
||||
ZSTD_memset(rankLast, 0xF0, sizeof(rankLast));
|
||||
{ U32 currentNbBits = targetNbBits;
|
||||
int pos;
|
||||
for (pos=n ; pos >= 0; pos--) {
|
||||
if (huffNode[pos].nbBits >= currentNbBits) continue;
|
||||
currentNbBits = huffNode[pos].nbBits; /* < targetNbBits */
|
||||
rankLast[targetNbBits-currentNbBits] = (U32)pos;
|
||||
} }
|
||||
|
||||
while (totalCost > 0) {
|
||||
/* Try to reduce the next power of 2 above totalCost because we
|
||||
* gain back half the rank.
|
||||
*/
|
||||
U32 nBitsToDecrease = ZSTD_highbit32((U32)totalCost) + 1;
|
||||
for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
|
||||
U32 const highPos = rankLast[nBitsToDecrease];
|
||||
U32 const lowPos = rankLast[nBitsToDecrease-1];
|
||||
if (highPos == noSymbol) continue;
|
||||
/* Decrease highPos if no symbols of lowPos or if it is
|
||||
* not cheaper to remove 2 lowPos than highPos.
|
||||
*/
|
||||
if (lowPos == noSymbol) break;
|
||||
{ U32 const highTotal = huffNode[highPos].count;
|
||||
U32 const lowTotal = 2 * huffNode[lowPos].count;
|
||||
if (highTotal <= lowTotal) break;
|
||||
} }
|
||||
/* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
|
||||
assert(rankLast[nBitsToDecrease] != noSymbol || nBitsToDecrease == 1);
|
||||
/* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
|
||||
while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))
|
||||
nBitsToDecrease++;
|
||||
assert(rankLast[nBitsToDecrease] != noSymbol);
|
||||
/* Increase the number of bits to gain back half the rank cost. */
|
||||
totalCost -= 1 << (nBitsToDecrease-1);
|
||||
huffNode[rankLast[nBitsToDecrease]].nbBits++;
|
||||
|
||||
/* Fix up the new rank.
|
||||
* If the new rank was empty, this symbol is now its smallest.
|
||||
* Otherwise, this symbol will be the largest in the new rank so no adjustment.
|
||||
*/
|
||||
if (rankLast[nBitsToDecrease-1] == noSymbol)
|
||||
rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];
|
||||
/* Fix up the old rank.
|
||||
* If the symbol was at position 0, meaning it was the highest weight symbol in the tree,
|
||||
* it must be the only symbol in its rank, so the old rank now has no symbols.
|
||||
* Otherwise, since the Huffman nodes are sorted by count, the previous position is now
|
||||
* the smallest node in the rank. If the previous position belongs to a different rank,
|
||||
* then the rank is now empty.
|
||||
*/
|
||||
if (rankLast[nBitsToDecrease] == 0) /* special case, reached largest symbol */
|
||||
rankLast[nBitsToDecrease] = noSymbol;
|
||||
else {
|
||||
rankLast[nBitsToDecrease]--;
|
||||
if (huffNode[rankLast[nBitsToDecrease]].nbBits != targetNbBits-nBitsToDecrease)
|
||||
rankLast[nBitsToDecrease] = noSymbol; /* this rank is now empty */
|
||||
}
|
||||
} /* while (totalCost > 0) */
|
||||
|
||||
/* If we've removed too much weight, then we have to add it back.
|
||||
* To avoid overshooting again, we only adjust the smallest rank.
|
||||
* We take the largest nodes from the lowest rank 0 and move them
|
||||
* to rank 1. There's guaranteed to be enough rank 0 symbols because
|
||||
* TODO.
|
||||
*/
|
||||
while (totalCost < 0) { /* Sometimes, cost correction overshoot */
|
||||
/* special case : no rank 1 symbol (using targetNbBits-1);
|
||||
* let's create one from largest rank 0 (using targetNbBits).
|
||||
*/
|
||||
if (rankLast[1] == noSymbol) {
|
||||
while (huffNode[n].nbBits == targetNbBits) n--;
|
||||
huffNode[n+1].nbBits--;
|
||||
assert(n >= 0);
|
||||
rankLast[1] = (U32)(n+1);
|
||||
totalCost++;
|
||||
continue;
|
||||
}
|
||||
huffNode[ rankLast[1] + 1 ].nbBits--;
|
||||
rankLast[1]++;
|
||||
totalCost ++;
|
||||
}
|
||||
} /* repay normalized cost */
|
||||
} /* there are several too large elements (at least >= 2) */
|
||||
|
||||
return targetNbBits;
|
||||
}
|
||||
|
||||
typedef struct {
|
||||
U16 base;
|
||||
U16 curr;
|
||||
} rankPos;
|
||||
|
||||
typedef nodeElt huffNodeTable[2 * (HUF_SYMBOLVALUE_MAX + 1)];
|
||||
|
||||
/* Number of buckets available for HUF_sort() */
|
||||
#define RANK_POSITION_TABLE_SIZE 192
|
||||
|
||||
typedef struct {
|
||||
huffNodeTable huffNodeTbl;
|
||||
rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
|
||||
} HUF_buildCTable_wksp_tables;
|
||||
|
||||
/* RANK_POSITION_DISTINCT_COUNT_CUTOFF == Cutoff point in HUF_sort() buckets for which we use log2 bucketing.
|
||||
* Strategy is to use as many buckets as possible for representing distinct
|
||||
* counts while using the remainder to represent all "large" counts.
|
||||
*
|
||||
* To satisfy this requirement for 192 buckets, we can do the following:
|
||||
* Let buckets 0-166 represent distinct counts of [0, 166]
|
||||
* Let buckets 166 to 192 represent all remaining counts up to RANK_POSITION_MAX_COUNT_LOG using log2 bucketing.
|
||||
*/
|
||||
#define RANK_POSITION_MAX_COUNT_LOG 32
|
||||
#define RANK_POSITION_LOG_BUCKETS_BEGIN ((RANK_POSITION_TABLE_SIZE - 1) - RANK_POSITION_MAX_COUNT_LOG - 1 /* == 158 */)
|
||||
#define RANK_POSITION_DISTINCT_COUNT_CUTOFF (RANK_POSITION_LOG_BUCKETS_BEGIN + ZSTD_highbit32(RANK_POSITION_LOG_BUCKETS_BEGIN) /* == 166 */)
|
||||
|
||||
/* Return the appropriate bucket index for a given count. See definition of
|
||||
* RANK_POSITION_DISTINCT_COUNT_CUTOFF for explanation of bucketing strategy.
|
||||
*/
|
||||
static U32 HUF_getIndex(U32 const count) {
|
||||
return (count < RANK_POSITION_DISTINCT_COUNT_CUTOFF)
|
||||
? count
|
||||
: ZSTD_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN;
|
||||
}
|
||||
|
||||
/* Helper swap function for HUF_quickSortPartition() */
|
||||
static void HUF_swapNodes(nodeElt* a, nodeElt* b) {
|
||||
nodeElt tmp = *a;
|
||||
*a = *b;
|
||||
*b = tmp;
|
||||
}
|
||||
|
||||
/* Returns 0 if the huffNode array is not sorted by descending count */
|
||||
MEM_STATIC int HUF_isSorted(nodeElt huffNode[], U32 const maxSymbolValue1) {
|
||||
U32 i;
|
||||
for (i = 1; i < maxSymbolValue1; ++i) {
|
||||
if (huffNode[i].count > huffNode[i-1].count) {
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
return 1;
|
||||
}
|
||||
|
||||
/* Insertion sort by descending order */
|
||||
HINT_INLINE void HUF_insertionSort(nodeElt huffNode[], int const low, int const high) {
|
||||
int i;
|
||||
int const size = high-low+1;
|
||||
huffNode += low;
|
||||
for (i = 1; i < size; ++i) {
|
||||
nodeElt const key = huffNode[i];
|
||||
int j = i - 1;
|
||||
while (j >= 0 && huffNode[j].count < key.count) {
|
||||
huffNode[j + 1] = huffNode[j];
|
||||
j--;
|
||||
}
|
||||
huffNode[j + 1] = key;
|
||||
}
|
||||
}
|
||||
|
||||
/* Pivot helper function for quicksort. */
|
||||
static int HUF_quickSortPartition(nodeElt arr[], int const low, int const high) {
|
||||
/* Simply select rightmost element as pivot. "Better" selectors like
|
||||
* median-of-three don't experimentally appear to have any benefit.
|
||||
*/
|
||||
U32 const pivot = arr[high].count;
|
||||
int i = low - 1;
|
||||
int j = low;
|
||||
for ( ; j < high; j++) {
|
||||
if (arr[j].count > pivot) {
|
||||
i++;
|
||||
HUF_swapNodes(&arr[i], &arr[j]);
|
||||
}
|
||||
}
|
||||
HUF_swapNodes(&arr[i + 1], &arr[high]);
|
||||
return i + 1;
|
||||
}
|
||||
|
||||
/* Classic quicksort by descending with partially iterative calls
|
||||
* to reduce worst case callstack size.
|
||||
*/
|
||||
static void HUF_simpleQuickSort(nodeElt arr[], int low, int high) {
|
||||
int const kInsertionSortThreshold = 8;
|
||||
if (high - low < kInsertionSortThreshold) {
|
||||
HUF_insertionSort(arr, low, high);
|
||||
return;
|
||||
}
|
||||
while (low < high) {
|
||||
int const idx = HUF_quickSortPartition(arr, low, high);
|
||||
if (idx - low < high - idx) {
|
||||
HUF_simpleQuickSort(arr, low, idx - 1);
|
||||
low = idx + 1;
|
||||
} else {
|
||||
HUF_simpleQuickSort(arr, idx + 1, high);
|
||||
high = idx - 1;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* HUF_sort():
|
||||
* Sorts the symbols [0, maxSymbolValue] by count[symbol] in decreasing order.
|
||||
* This is a typical bucket sorting strategy that uses either quicksort or insertion sort to sort each bucket.
|
||||
*
|
||||
* @param[out] huffNode Sorted symbols by decreasing count. Only members `.count` and `.byte` are filled.
|
||||
* Must have (maxSymbolValue + 1) entries.
|
||||
* @param[in] count Histogram of the symbols.
|
||||
* @param[in] maxSymbolValue Maximum symbol value.
|
||||
* @param rankPosition This is a scratch workspace. Must have RANK_POSITION_TABLE_SIZE entries.
|
||||
*/
|
||||
static void HUF_sort(nodeElt huffNode[], const unsigned count[], U32 const maxSymbolValue, rankPos rankPosition[]) {
|
||||
U32 n;
|
||||
U32 const maxSymbolValue1 = maxSymbolValue+1;
|
||||
|
||||
/* Compute base and set curr to base.
|
||||
* For symbol s let lowerRank = HUF_getIndex(count[n]) and rank = lowerRank + 1.
|
||||
* See HUF_getIndex to see bucketing strategy.
|
||||
* We attribute each symbol to lowerRank's base value, because we want to know where
|
||||
* each rank begins in the output, so for rank R we want to count ranks R+1 and above.
|
||||
*/
|
||||
ZSTD_memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
|
||||
for (n = 0; n < maxSymbolValue1; ++n) {
|
||||
U32 lowerRank = HUF_getIndex(count[n]);
|
||||
assert(lowerRank < RANK_POSITION_TABLE_SIZE - 1);
|
||||
rankPosition[lowerRank].base++;
|
||||
}
|
||||
|
||||
assert(rankPosition[RANK_POSITION_TABLE_SIZE - 1].base == 0);
|
||||
/* Set up the rankPosition table */
|
||||
for (n = RANK_POSITION_TABLE_SIZE - 1; n > 0; --n) {
|
||||
rankPosition[n-1].base += rankPosition[n].base;
|
||||
rankPosition[n-1].curr = rankPosition[n-1].base;
|
||||
}
|
||||
|
||||
/* Insert each symbol into their appropriate bucket, setting up rankPosition table. */
|
||||
for (n = 0; n < maxSymbolValue1; ++n) {
|
||||
U32 const c = count[n];
|
||||
U32 const r = HUF_getIndex(c) + 1;
|
||||
U32 const pos = rankPosition[r].curr++;
|
||||
assert(pos < maxSymbolValue1);
|
||||
huffNode[pos].count = c;
|
||||
huffNode[pos].byte = (BYTE)n;
|
||||
}
|
||||
|
||||
/* Sort each bucket. */
|
||||
for (n = RANK_POSITION_DISTINCT_COUNT_CUTOFF; n < RANK_POSITION_TABLE_SIZE - 1; ++n) {
|
||||
int const bucketSize = rankPosition[n].curr - rankPosition[n].base;
|
||||
U32 const bucketStartIdx = rankPosition[n].base;
|
||||
if (bucketSize > 1) {
|
||||
assert(bucketStartIdx < maxSymbolValue1);
|
||||
HUF_simpleQuickSort(huffNode + bucketStartIdx, 0, bucketSize-1);
|
||||
}
|
||||
}
|
||||
|
||||
assert(HUF_isSorted(huffNode, maxSymbolValue1));
|
||||
}
|
||||
|
||||
|
||||
/** HUF_buildCTable_wksp() :
|
||||
* Same as HUF_buildCTable(), but using externally allocated scratch buffer.
|
||||
* `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as sizeof(HUF_buildCTable_wksp_tables).
|
||||
*/
|
||||
#define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
|
||||
|
||||
/* HUF_buildTree():
|
||||
* Takes the huffNode array sorted by HUF_sort() and builds an unlimited-depth Huffman tree.
|
||||
*
|
||||
* @param huffNode The array sorted by HUF_sort(). Builds the Huffman tree in this array.
|
||||
* @param maxSymbolValue The maximum symbol value.
|
||||
* @return The smallest node in the Huffman tree (by count).
|
||||
*/
|
||||
static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
|
||||
{
|
||||
nodeElt* const huffNode0 = huffNode - 1;
|
||||
int nonNullRank;
|
||||
int lowS, lowN;
|
||||
int nodeNb = STARTNODE;
|
||||
int n, nodeRoot;
|
||||
DEBUGLOG(5, "HUF_buildTree (alphabet size = %u)", maxSymbolValue + 1);
|
||||
/* init for parents */
|
||||
nonNullRank = (int)maxSymbolValue;
|
||||
while(huffNode[nonNullRank].count == 0) nonNullRank--;
|
||||
lowS = nonNullRank; nodeRoot = nodeNb + lowS - 1; lowN = nodeNb;
|
||||
huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count;
|
||||
huffNode[lowS].parent = huffNode[lowS-1].parent = (U16)nodeNb;
|
||||
nodeNb++; lowS-=2;
|
||||
for (n=nodeNb; n<=nodeRoot; n++) huffNode[n].count = (U32)(1U<<30);
|
||||
huffNode0[0].count = (U32)(1U<<31); /* fake entry, strong barrier */
|
||||
|
||||
/* create parents */
|
||||
while (nodeNb <= nodeRoot) {
|
||||
int const n1 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
|
||||
int const n2 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
|
||||
huffNode[nodeNb].count = huffNode[n1].count + huffNode[n2].count;
|
||||
huffNode[n1].parent = huffNode[n2].parent = (U16)nodeNb;
|
||||
nodeNb++;
|
||||
}
|
||||
|
||||
/* distribute weights (unlimited tree height) */
|
||||
huffNode[nodeRoot].nbBits = 0;
|
||||
for (n=nodeRoot-1; n>=STARTNODE; n--)
|
||||
huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
|
||||
for (n=0; n<=nonNullRank; n++)
|
||||
huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
|
||||
|
||||
DEBUGLOG(6, "Initial distribution of bits completed (%zu sorted symbols)", showHNodeBits(huffNode, maxSymbolValue+1));
|
||||
|
||||
return nonNullRank;
|
||||
}
|
||||
|
||||
/**
|
||||
* HUF_buildCTableFromTree():
|
||||
* Build the CTable given the Huffman tree in huffNode.
|
||||
*
|
||||
* @param[out] CTable The output Huffman CTable.
|
||||
* @param huffNode The Huffman tree.
|
||||
* @param nonNullRank The last and smallest node in the Huffman tree.
|
||||
* @param maxSymbolValue The maximum symbol value.
|
||||
* @param maxNbBits The exact maximum number of bits used in the Huffman tree.
|
||||
*/
|
||||
static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, int nonNullRank, U32 maxSymbolValue, U32 maxNbBits)
|
||||
{
|
||||
HUF_CElt* const ct = CTable + 1;
|
||||
/* fill result into ctable (val, nbBits) */
|
||||
int n;
|
||||
U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
|
||||
U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
|
||||
int const alphabetSize = (int)(maxSymbolValue + 1);
|
||||
for (n=0; n<=nonNullRank; n++)
|
||||
nbPerRank[huffNode[n].nbBits]++;
|
||||
/* determine starting value per rank */
|
||||
{ U16 min = 0;
|
||||
for (n=(int)maxNbBits; n>0; n--) {
|
||||
valPerRank[n] = min; /* get starting value within each rank */
|
||||
min += nbPerRank[n];
|
||||
min >>= 1;
|
||||
} }
|
||||
for (n=0; n<alphabetSize; n++)
|
||||
HUF_setNbBits(ct + huffNode[n].byte, huffNode[n].nbBits); /* push nbBits per symbol, symbol order */
|
||||
for (n=0; n<alphabetSize; n++)
|
||||
HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); /* assign value within rank, symbol order */
|
||||
|
||||
HUF_writeCTableHeader(CTable, maxNbBits, maxSymbolValue);
|
||||
}
|
||||
|
||||
size_t
|
||||
HUF_buildCTable_wksp(HUF_CElt* CTable, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
|
||||
void* workSpace, size_t wkspSize)
|
||||
{
|
||||
HUF_buildCTable_wksp_tables* const wksp_tables =
|
||||
(HUF_buildCTable_wksp_tables*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(U32));
|
||||
nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
|
||||
nodeElt* const huffNode = huffNode0+1;
|
||||
int nonNullRank;
|
||||
|
||||
HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE == sizeof(HUF_buildCTable_wksp_tables));
|
||||
|
||||
DEBUGLOG(5, "HUF_buildCTable_wksp (alphabet size = %u)", maxSymbolValue+1);
|
||||
|
||||
/* safety checks */
|
||||
if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
|
||||
return ERROR(workSpace_tooSmall);
|
||||
if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
|
||||
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
|
||||
return ERROR(maxSymbolValue_tooLarge);
|
||||
ZSTD_memset(huffNode0, 0, sizeof(huffNodeTable));
|
||||
|
||||
/* sort, decreasing order */
|
||||
HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
|
||||
DEBUGLOG(6, "sorted symbols completed (%zu symbols)", showHNodeSymbols(huffNode, maxSymbolValue+1));
|
||||
|
||||
/* build tree */
|
||||
nonNullRank = HUF_buildTree(huffNode, maxSymbolValue);
|
||||
|
||||
/* determine and enforce maxTableLog */
|
||||
maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
|
||||
if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC); /* check fit into table */
|
||||
|
||||
HUF_buildCTableFromTree(CTable, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
|
||||
|
||||
return maxNbBits;
|
||||
}
|
||||
|
||||
size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
|
||||
{
|
||||
HUF_CElt const* ct = CTable + 1;
|
||||
size_t nbBits = 0;
|
||||
int s;
|
||||
for (s = 0; s <= (int)maxSymbolValue; ++s) {
|
||||
nbBits += HUF_getNbBits(ct[s]) * count[s];
|
||||
}
|
||||
return nbBits >> 3;
|
||||
}
|
||||
|
||||
int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
|
||||
HUF_CTableHeader header = HUF_readCTableHeader(CTable);
|
||||
HUF_CElt const* ct = CTable + 1;
|
||||
int bad = 0;
|
||||
int s;
|
||||
|
||||
assert(header.tableLog <= HUF_TABLELOG_ABSOLUTEMAX);
|
||||
|
||||
if (header.maxSymbolValue < maxSymbolValue)
|
||||
return 0;
|
||||
|
||||
for (s = 0; s <= (int)maxSymbolValue; ++s) {
|
||||
bad |= (count[s] != 0) & (HUF_getNbBits(ct[s]) == 0);
|
||||
}
|
||||
return !bad;
|
||||
}
|
||||
|
||||
size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
|
||||
|
||||
/** HUF_CStream_t:
|
||||
* Huffman uses its own BIT_CStream_t implementation.
|
||||
* There are three major differences from BIT_CStream_t:
|
||||
* 1. HUF_addBits() takes a HUF_CElt (size_t) which is
|
||||
* the pair (nbBits, value) in the format:
|
||||
* format:
|
||||
* - Bits [0, 4) = nbBits
|
||||
* - Bits [4, 64 - nbBits) = 0
|
||||
* - Bits [64 - nbBits, 64) = value
|
||||
* 2. The bitContainer is built from the upper bits and
|
||||
* right shifted. E.g. to add a new value of N bits
|
||||
* you right shift the bitContainer by N, then or in
|
||||
* the new value into the N upper bits.
|
||||
* 3. The bitstream has two bit containers. You can add
|
||||
* bits to the second container and merge them into
|
||||
* the first container.
|
||||
*/
|
||||
|
||||
#define HUF_BITS_IN_CONTAINER (sizeof(size_t) * 8)
|
||||
|
||||
typedef struct {
|
||||
size_t bitContainer[2];
|
||||
size_t bitPos[2];
|
||||
|
||||
BYTE* startPtr;
|
||||
BYTE* ptr;
|
||||
BYTE* endPtr;
|
||||
} HUF_CStream_t;
|
||||
|
||||
/**! HUF_initCStream():
|
||||
* Initializes the bitstream.
|
||||
* @returns 0 or an error code.
|
||||
*/
|
||||
static size_t HUF_initCStream(HUF_CStream_t* bitC,
|
||||
void* startPtr, size_t dstCapacity)
|
||||
{
|
||||
ZSTD_memset(bitC, 0, sizeof(*bitC));
|
||||
bitC->startPtr = (BYTE*)startPtr;
|
||||
bitC->ptr = bitC->startPtr;
|
||||
bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer[0]);
|
||||
if (dstCapacity <= sizeof(bitC->bitContainer[0])) return ERROR(dstSize_tooSmall);
|
||||
return 0;
|
||||
}
|
||||
|
||||
/*! HUF_addBits():
|
||||
* Adds the symbol stored in HUF_CElt elt to the bitstream.
|
||||
*
|
||||
* @param elt The element we're adding. This is a (nbBits, value) pair.
|
||||
* See the HUF_CStream_t docs for the format.
|
||||
* @param idx Insert into the bitstream at this idx.
|
||||
* @param kFast This is a template parameter. If the bitstream is guaranteed
|
||||
* to have at least 4 unused bits after this call it may be 1,
|
||||
* otherwise it must be 0. HUF_addBits() is faster when fast is set.
|
||||
*/
|
||||
FORCE_INLINE_TEMPLATE void HUF_addBits(HUF_CStream_t* bitC, HUF_CElt elt, int idx, int kFast)
|
||||
{
|
||||
assert(idx <= 1);
|
||||
assert(HUF_getNbBits(elt) <= HUF_TABLELOG_ABSOLUTEMAX);
|
||||
/* This is efficient on x86-64 with BMI2 because shrx
|
||||
* only reads the low 6 bits of the register. The compiler
|
||||
* knows this and elides the mask. When fast is set,
|
||||
* every operation can use the same value loaded from elt.
|
||||
*/
|
||||
bitC->bitContainer[idx] >>= HUF_getNbBits(elt);
|
||||
bitC->bitContainer[idx] |= kFast ? HUF_getValueFast(elt) : HUF_getValue(elt);
|
||||
/* We only read the low 8 bits of bitC->bitPos[idx] so it
|
||||
* doesn't matter that the high bits have noise from the value.
|
||||
*/
|
||||
bitC->bitPos[idx] += HUF_getNbBitsFast(elt);
|
||||
assert((bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
|
||||
/* The last 4-bits of elt are dirty if fast is set,
|
||||
* so we must not be overwriting bits that have already been
|
||||
* inserted into the bit container.
|
||||
*/
|
||||
#if DEBUGLEVEL >= 1
|
||||
{
|
||||
size_t const nbBits = HUF_getNbBits(elt);
|
||||
size_t const dirtyBits = nbBits == 0 ? 0 : ZSTD_highbit32((U32)nbBits) + 1;
|
||||
(void)dirtyBits;
|
||||
/* Middle bits are 0. */
|
||||
assert(((elt >> dirtyBits) << (dirtyBits + nbBits)) == 0);
|
||||
/* We didn't overwrite any bits in the bit container. */
|
||||
assert(!kFast || (bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
|
||||
(void)dirtyBits;
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
FORCE_INLINE_TEMPLATE void HUF_zeroIndex1(HUF_CStream_t* bitC)
|
||||
{
|
||||
bitC->bitContainer[1] = 0;
|
||||
bitC->bitPos[1] = 0;
|
||||
}
|
||||
|
||||
/*! HUF_mergeIndex1() :
|
||||
* Merges the bit container @ index 1 into the bit container @ index 0
|
||||
* and zeros the bit container @ index 1.
|
||||
*/
|
||||
FORCE_INLINE_TEMPLATE void HUF_mergeIndex1(HUF_CStream_t* bitC)
|
||||
{
|
||||
assert((bitC->bitPos[1] & 0xFF) < HUF_BITS_IN_CONTAINER);
|
||||
bitC->bitContainer[0] >>= (bitC->bitPos[1] & 0xFF);
|
||||
bitC->bitContainer[0] |= bitC->bitContainer[1];
|
||||
bitC->bitPos[0] += bitC->bitPos[1];
|
||||
assert((bitC->bitPos[0] & 0xFF) <= HUF_BITS_IN_CONTAINER);
|
||||
}
|
||||
|
||||
/*! HUF_flushBits() :
|
||||
* Flushes the bits in the bit container @ index 0.
|
||||
*
|
||||
* @post bitPos will be < 8.
|
||||
* @param kFast If kFast is set then we must know a-priori that
|
||||
* the bit container will not overflow.
|
||||
*/
|
||||
FORCE_INLINE_TEMPLATE void HUF_flushBits(HUF_CStream_t* bitC, int kFast)
|
||||
{
|
||||
/* The upper bits of bitPos are noisy, so we must mask by 0xFF. */
|
||||
size_t const nbBits = bitC->bitPos[0] & 0xFF;
|
||||
size_t const nbBytes = nbBits >> 3;
|
||||
/* The top nbBits bits of bitContainer are the ones we need. */
|
||||
size_t const bitContainer = bitC->bitContainer[0] >> (HUF_BITS_IN_CONTAINER - nbBits);
|
||||
/* Mask bitPos to account for the bytes we consumed. */
|
||||
bitC->bitPos[0] &= 7;
|
||||
assert(nbBits > 0);
|
||||
assert(nbBits <= sizeof(bitC->bitContainer[0]) * 8);
|
||||
assert(bitC->ptr <= bitC->endPtr);
|
||||
MEM_writeLEST(bitC->ptr, bitContainer);
|
||||
bitC->ptr += nbBytes;
|
||||
assert(!kFast || bitC->ptr <= bitC->endPtr);
|
||||
if (!kFast && bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
|
||||
/* bitContainer doesn't need to be modified because the leftover
|
||||
* bits are already the top bitPos bits. And we don't care about
|
||||
* noise in the lower values.
|
||||
*/
|
||||
}
|
||||
|
||||
/*! HUF_endMark()
|
||||
* @returns The Huffman stream end mark: A 1-bit value = 1.
|
||||
*/
|
||||
static HUF_CElt HUF_endMark(void)
|
||||
{
|
||||
HUF_CElt endMark;
|
||||
HUF_setNbBits(&endMark, 1);
|
||||
HUF_setValue(&endMark, 1);
|
||||
return endMark;
|
||||
}
|
||||
|
||||
/*! HUF_closeCStream() :
|
||||
* @return Size of CStream, in bytes,
|
||||
* or 0 if it could not fit into dstBuffer */
|
||||
static size_t HUF_closeCStream(HUF_CStream_t* bitC)
|
||||
{
|
||||
HUF_addBits(bitC, HUF_endMark(), /* idx */ 0, /* kFast */ 0);
|
||||
HUF_flushBits(bitC, /* kFast */ 0);
|
||||
{
|
||||
size_t const nbBits = bitC->bitPos[0] & 0xFF;
|
||||
if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
|
||||
return (size_t)(bitC->ptr - bitC->startPtr) + (nbBits > 0);
|
||||
}
|
||||
}
|
||||
|
||||
FORCE_INLINE_TEMPLATE void
|
||||
HUF_encodeSymbol(HUF_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable, int idx, int fast)
|
||||
{
|
||||
HUF_addBits(bitCPtr, CTable[symbol], idx, fast);
|
||||
}
|
||||
|
||||
FORCE_INLINE_TEMPLATE void
|
||||
HUF_compress1X_usingCTable_internal_body_loop(HUF_CStream_t* bitC,
|
||||
const BYTE* ip, size_t srcSize,
|
||||
const HUF_CElt* ct,
|
||||
int kUnroll, int kFastFlush, int kLastFast)
|
||||
{
|
||||
/* Join to kUnroll */
|
||||
int n = (int)srcSize;
|
||||
int rem = n % kUnroll;
|
||||
if (rem > 0) {
|
||||
for (; rem > 0; --rem) {
|
||||
HUF_encodeSymbol(bitC, ip[--n], ct, 0, /* fast */ 0);
|
||||
}
|
||||
HUF_flushBits(bitC, kFastFlush);
|
||||
}
|
||||
assert(n % kUnroll == 0);
|
||||
|
||||
/* Join to 2 * kUnroll */
|
||||
if (n % (2 * kUnroll)) {
|
||||
int u;
|
||||
for (u = 1; u < kUnroll; ++u) {
|
||||
HUF_encodeSymbol(bitC, ip[n - u], ct, 0, 1);
|
||||
}
|
||||
HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, 0, kLastFast);
|
||||
HUF_flushBits(bitC, kFastFlush);
|
||||
n -= kUnroll;
|
||||
}
|
||||
assert(n % (2 * kUnroll) == 0);
|
||||
|
||||
for (; n>0; n-= 2 * kUnroll) {
|
||||
/* Encode kUnroll symbols into the bitstream @ index 0. */
|
||||
int u;
|
||||
for (u = 1; u < kUnroll; ++u) {
|
||||
HUF_encodeSymbol(bitC, ip[n - u], ct, /* idx */ 0, /* fast */ 1);
|
||||
}
|
||||
HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, /* idx */ 0, /* fast */ kLastFast);
|
||||
HUF_flushBits(bitC, kFastFlush);
|
||||
/* Encode kUnroll symbols into the bitstream @ index 1.
|
||||
* This allows us to start filling the bit container
|
||||
* without any data dependencies.
|
||||
*/
|
||||
HUF_zeroIndex1(bitC);
|
||||
for (u = 1; u < kUnroll; ++u) {
|
||||
HUF_encodeSymbol(bitC, ip[n - kUnroll - u], ct, /* idx */ 1, /* fast */ 1);
|
||||
}
|
||||
HUF_encodeSymbol(bitC, ip[n - kUnroll - kUnroll], ct, /* idx */ 1, /* fast */ kLastFast);
|
||||
/* Merge bitstream @ index 1 into the bitstream @ index 0 */
|
||||
HUF_mergeIndex1(bitC);
|
||||
HUF_flushBits(bitC, kFastFlush);
|
||||
}
|
||||
assert(n == 0);
|
||||
|
||||
}
|
||||
|
||||
/**
|
||||
* Returns a tight upper bound on the output space needed by Huffman
|
||||
* with 8 bytes buffer to handle over-writes. If the output is at least
|
||||
* this large we don't need to do bounds checks during Huffman encoding.
|
||||
*/
|
||||
static size_t HUF_tightCompressBound(size_t srcSize, size_t tableLog)
|
||||
{
|
||||
return ((srcSize * tableLog) >> 3) + 8;
|
||||
}
|
||||
|
||||
|
||||
FORCE_INLINE_TEMPLATE size_t
|
||||
HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
const HUF_CElt* CTable)
|
||||
{
|
||||
U32 const tableLog = HUF_readCTableHeader(CTable).tableLog;
|
||||
HUF_CElt const* ct = CTable + 1;
|
||||
const BYTE* ip = (const BYTE*) src;
|
||||
BYTE* const ostart = (BYTE*)dst;
|
||||
BYTE* const oend = ostart + dstSize;
|
||||
HUF_CStream_t bitC;
|
||||
|
||||
/* init */
|
||||
if (dstSize < 8) return 0; /* not enough space to compress */
|
||||
{ BYTE* op = ostart;
|
||||
size_t const initErr = HUF_initCStream(&bitC, op, (size_t)(oend-op));
|
||||
if (HUF_isError(initErr)) return 0; }
|
||||
|
||||
if (dstSize < HUF_tightCompressBound(srcSize, (size_t)tableLog) || tableLog > 11)
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ MEM_32bits() ? 2 : 4, /* kFast */ 0, /* kLastFast */ 0);
|
||||
else {
|
||||
if (MEM_32bits()) {
|
||||
switch (tableLog) {
|
||||
case 11:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 0);
|
||||
break;
|
||||
case 10: ZSTD_FALLTHROUGH;
|
||||
case 9: ZSTD_FALLTHROUGH;
|
||||
case 8:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 1);
|
||||
break;
|
||||
case 7: ZSTD_FALLTHROUGH;
|
||||
default:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 3, /* kFastFlush */ 1, /* kLastFast */ 1);
|
||||
break;
|
||||
}
|
||||
} else {
|
||||
switch (tableLog) {
|
||||
case 11:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 0);
|
||||
break;
|
||||
case 10:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 1);
|
||||
break;
|
||||
case 9:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 6, /* kFastFlush */ 1, /* kLastFast */ 0);
|
||||
break;
|
||||
case 8:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 7, /* kFastFlush */ 1, /* kLastFast */ 0);
|
||||
break;
|
||||
case 7:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 8, /* kFastFlush */ 1, /* kLastFast */ 0);
|
||||
break;
|
||||
case 6: ZSTD_FALLTHROUGH;
|
||||
default:
|
||||
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 9, /* kFastFlush */ 1, /* kLastFast */ 1);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
assert(bitC.ptr <= bitC.endPtr);
|
||||
|
||||
return HUF_closeCStream(&bitC);
|
||||
}
|
||||
|
||||
#if DYNAMIC_BMI2
|
||||
|
||||
static BMI2_TARGET_ATTRIBUTE size_t
|
||||
HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
const HUF_CElt* CTable)
|
||||
{
|
||||
return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
|
||||
}
|
||||
|
||||
static size_t
|
||||
HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
const HUF_CElt* CTable)
|
||||
{
|
||||
return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
|
||||
}
|
||||
|
||||
static size_t
|
||||
HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
const HUF_CElt* CTable, const int flags)
|
||||
{
|
||||
if (flags & HUF_flags_bmi2) {
|
||||
return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
|
||||
}
|
||||
return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
|
||||
}
|
||||
|
||||
#else
|
||||
|
||||
static size_t
|
||||
HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
const HUF_CElt* CTable, const int flags)
|
||||
{
|
||||
(void)flags;
|
||||
return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
|
||||
{
|
||||
return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
|
||||
}
|
||||
|
||||
static size_t
|
||||
HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
const HUF_CElt* CTable, int flags)
|
||||
{
|
||||
size_t const segmentSize = (srcSize+3)/4; /* first 3 segments */
|
||||
const BYTE* ip = (const BYTE*) src;
|
||||
const BYTE* const iend = ip + srcSize;
|
||||
BYTE* const ostart = (BYTE*) dst;
|
||||
BYTE* const oend = ostart + dstSize;
|
||||
BYTE* op = ostart;
|
||||
|
||||
if (dstSize < 6 + 1 + 1 + 1 + 8) return 0; /* minimum space to compress successfully */
|
||||
if (srcSize < 12) return 0; /* no saving possible : too small input */
|
||||
op += 6; /* jumpTable */
|
||||
|
||||
assert(op <= oend);
|
||||
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
|
||||
if (cSize == 0 || cSize > 65535) return 0;
|
||||
MEM_writeLE16(ostart, (U16)cSize);
|
||||
op += cSize;
|
||||
}
|
||||
|
||||
ip += segmentSize;
|
||||
assert(op <= oend);
|
||||
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
|
||||
if (cSize == 0 || cSize > 65535) return 0;
|
||||
MEM_writeLE16(ostart+2, (U16)cSize);
|
||||
op += cSize;
|
||||
}
|
||||
|
||||
ip += segmentSize;
|
||||
assert(op <= oend);
|
||||
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
|
||||
if (cSize == 0 || cSize > 65535) return 0;
|
||||
MEM_writeLE16(ostart+4, (U16)cSize);
|
||||
op += cSize;
|
||||
}
|
||||
|
||||
ip += segmentSize;
|
||||
assert(op <= oend);
|
||||
assert(ip <= iend);
|
||||
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, flags) );
|
||||
if (cSize == 0 || cSize > 65535) return 0;
|
||||
op += cSize;
|
||||
}
|
||||
|
||||
return (size_t)(op-ostart);
|
||||
}
|
||||
|
||||
size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
|
||||
{
|
||||
return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
|
||||
}
|
||||
|
||||
typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
|
||||
|
||||
static size_t HUF_compressCTable_internal(
|
||||
BYTE* const ostart, BYTE* op, BYTE* const oend,
|
||||
const void* src, size_t srcSize,
|
||||
HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int flags)
|
||||
{
|
||||
size_t const cSize = (nbStreams==HUF_singleStream) ?
|
||||
HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags) :
|
||||
HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags);
|
||||
if (HUF_isError(cSize)) { return cSize; }
|
||||
if (cSize==0) { return 0; } /* uncompressible */
|
||||
op += cSize;
|
||||
/* check compressibility */
|
||||
assert(op >= ostart);
|
||||
if ((size_t)(op-ostart) >= srcSize-1) { return 0; }
|
||||
return (size_t)(op-ostart);
|
||||
}
|
||||
|
||||
typedef struct {
|
||||
unsigned count[HUF_SYMBOLVALUE_MAX + 1];
|
||||
HUF_CElt CTable[HUF_CTABLE_SIZE_ST(HUF_SYMBOLVALUE_MAX)];
|
||||
union {
|
||||
HUF_buildCTable_wksp_tables buildCTable_wksp;
|
||||
HUF_WriteCTableWksp writeCTable_wksp;
|
||||
U32 hist_wksp[HIST_WKSP_SIZE_U32];
|
||||
} wksps;
|
||||
} HUF_compress_tables_t;
|
||||
|
||||
#define SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE 4096
|
||||
#define SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO 10 /* Must be >= 2 */
|
||||
|
||||
unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue)
|
||||
{
|
||||
unsigned cardinality = 0;
|
||||
unsigned i;
|
||||
|
||||
for (i = 0; i < maxSymbolValue + 1; i++) {
|
||||
if (count[i] != 0) cardinality += 1;
|
||||
}
|
||||
|
||||
return cardinality;
|
||||
}
|
||||
|
||||
unsigned HUF_minTableLog(unsigned symbolCardinality)
|
||||
{
|
||||
U32 minBitsSymbols = ZSTD_highbit32(symbolCardinality) + 1;
|
||||
return minBitsSymbols;
|
||||
}
|
||||
|
||||
unsigned HUF_optimalTableLog(
|
||||
unsigned maxTableLog,
|
||||
size_t srcSize,
|
||||
unsigned maxSymbolValue,
|
||||
void* workSpace, size_t wkspSize,
|
||||
HUF_CElt* table,
|
||||
const unsigned* count,
|
||||
int flags)
|
||||
{
|
||||
assert(srcSize > 1); /* Not supported, RLE should be used instead */
|
||||
assert(wkspSize >= sizeof(HUF_buildCTable_wksp_tables));
|
||||
|
||||
if (!(flags & HUF_flags_optimalDepth)) {
|
||||
/* cheap evaluation, based on FSE */
|
||||
return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
|
||||
}
|
||||
|
||||
{ BYTE* dst = (BYTE*)workSpace + sizeof(HUF_WriteCTableWksp);
|
||||
size_t dstSize = wkspSize - sizeof(HUF_WriteCTableWksp);
|
||||
size_t hSize, newSize;
|
||||
const unsigned symbolCardinality = HUF_cardinality(count, maxSymbolValue);
|
||||
const unsigned minTableLog = HUF_minTableLog(symbolCardinality);
|
||||
size_t optSize = ((size_t) ~0) - 1;
|
||||
unsigned optLog = maxTableLog, optLogGuess;
|
||||
|
||||
DEBUGLOG(6, "HUF_optimalTableLog: probing huf depth (srcSize=%zu)", srcSize);
|
||||
|
||||
/* Search until size increases */
|
||||
for (optLogGuess = minTableLog; optLogGuess <= maxTableLog; optLogGuess++) {
|
||||
DEBUGLOG(7, "checking for huffLog=%u", optLogGuess);
|
||||
|
||||
{ size_t maxBits = HUF_buildCTable_wksp(table, count, maxSymbolValue, optLogGuess, workSpace, wkspSize);
|
||||
if (ERR_isError(maxBits)) continue;
|
||||
|
||||
if (maxBits < optLogGuess && optLogGuess > minTableLog) break;
|
||||
|
||||
hSize = HUF_writeCTable_wksp(dst, dstSize, table, maxSymbolValue, (U32)maxBits, workSpace, wkspSize);
|
||||
}
|
||||
|
||||
if (ERR_isError(hSize)) continue;
|
||||
|
||||
newSize = HUF_estimateCompressedSize(table, count, maxSymbolValue) + hSize;
|
||||
|
||||
if (newSize > optSize + 1) {
|
||||
break;
|
||||
}
|
||||
|
||||
if (newSize < optSize) {
|
||||
optSize = newSize;
|
||||
optLog = optLogGuess;
|
||||
}
|
||||
}
|
||||
assert(optLog <= HUF_TABLELOG_MAX);
|
||||
return optLog;
|
||||
}
|
||||
}
|
||||
|
||||
/* HUF_compress_internal() :
|
||||
* `workSpace_align4` must be aligned on 4-bytes boundaries,
|
||||
* and occupies the same space as a table of HUF_WORKSPACE_SIZE_U64 unsigned */
|
||||
static size_t
|
||||
HUF_compress_internal (void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
unsigned maxSymbolValue, unsigned huffLog,
|
||||
HUF_nbStreams_e nbStreams,
|
||||
void* workSpace, size_t wkspSize,
|
||||
HUF_CElt* oldHufTable, HUF_repeat* repeat, int flags)
|
||||
{
|
||||
HUF_compress_tables_t* const table = (HUF_compress_tables_t*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(size_t));
|
||||
BYTE* const ostart = (BYTE*)dst;
|
||||
BYTE* const oend = ostart + dstSize;
|
||||
BYTE* op = ostart;
|
||||
|
||||
DEBUGLOG(5, "HUF_compress_internal (srcSize=%zu)", srcSize);
|
||||
HUF_STATIC_ASSERT(sizeof(*table) + HUF_WORKSPACE_MAX_ALIGNMENT <= HUF_WORKSPACE_SIZE);
|
||||
|
||||
/* checks & inits */
|
||||
if (wkspSize < sizeof(*table)) return ERROR(workSpace_tooSmall);
|
||||
if (!srcSize) return 0; /* Uncompressed */
|
||||
if (!dstSize) return 0; /* cannot fit anything within dst budget */
|
||||
if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong); /* current block size limit */
|
||||
if (huffLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
|
||||
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
|
||||
if (!maxSymbolValue) maxSymbolValue = HUF_SYMBOLVALUE_MAX;
|
||||
if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
|
||||
|
||||
/* Heuristic : If old table is valid, use it for small inputs */
|
||||
if ((flags & HUF_flags_preferRepeat) && repeat && *repeat == HUF_repeat_valid) {
|
||||
return HUF_compressCTable_internal(ostart, op, oend,
|
||||
src, srcSize,
|
||||
nbStreams, oldHufTable, flags);
|
||||
}
|
||||
|
||||
/* If uncompressible data is suspected, do a smaller sampling first */
|
||||
DEBUG_STATIC_ASSERT(SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO >= 2);
|
||||
if ((flags & HUF_flags_suspectUncompressible) && srcSize >= (SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO)) {
|
||||
size_t largestTotal = 0;
|
||||
DEBUGLOG(5, "input suspected incompressible : sampling to check");
|
||||
{ unsigned maxSymbolValueBegin = maxSymbolValue;
|
||||
CHECK_V_F(largestBegin, HIST_count_simple (table->count, &maxSymbolValueBegin, (const BYTE*)src, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
|
||||
largestTotal += largestBegin;
|
||||
}
|
||||
{ unsigned maxSymbolValueEnd = maxSymbolValue;
|
||||
CHECK_V_F(largestEnd, HIST_count_simple (table->count, &maxSymbolValueEnd, (const BYTE*)src + srcSize - SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
|
||||
largestTotal += largestEnd;
|
||||
}
|
||||
if (largestTotal <= ((2 * SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) >> 7)+4) return 0; /* heuristic : probably not compressible enough */
|
||||
}
|
||||
|
||||
/* Scan input and build symbol stats */
|
||||
{ CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->wksps.hist_wksp, sizeof(table->wksps.hist_wksp)) );
|
||||
if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; } /* single symbol, rle */
|
||||
if (largest <= (srcSize >> 7)+4) return 0; /* heuristic : probably not compressible enough */
|
||||
}
|
||||
DEBUGLOG(6, "histogram detail completed (%zu symbols)", showU32(table->count, maxSymbolValue+1));
|
||||
|
||||
/* Check validity of previous table */
|
||||
if ( repeat
|
||||
&& *repeat == HUF_repeat_check
|
||||
&& !HUF_validateCTable(oldHufTable, table->count, maxSymbolValue)) {
|
||||
*repeat = HUF_repeat_none;
|
||||
}
|
||||
/* Heuristic : use existing table for small inputs */
|
||||
if ((flags & HUF_flags_preferRepeat) && repeat && *repeat != HUF_repeat_none) {
|
||||
return HUF_compressCTable_internal(ostart, op, oend,
|
||||
src, srcSize,
|
||||
nbStreams, oldHufTable, flags);
|
||||
}
|
||||
|
||||
/* Build Huffman Tree */
|
||||
huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, &table->wksps, sizeof(table->wksps), table->CTable, table->count, flags);
|
||||
{ size_t const maxBits = HUF_buildCTable_wksp(table->CTable, table->count,
|
||||
maxSymbolValue, huffLog,
|
||||
&table->wksps.buildCTable_wksp, sizeof(table->wksps.buildCTable_wksp));
|
||||
CHECK_F(maxBits);
|
||||
huffLog = (U32)maxBits;
|
||||
DEBUGLOG(6, "bit distribution completed (%zu symbols)", showCTableBits(table->CTable + 1, maxSymbolValue+1));
|
||||
}
|
||||
|
||||
/* Write table description header */
|
||||
{ CHECK_V_F(hSize, HUF_writeCTable_wksp(op, dstSize, table->CTable, maxSymbolValue, huffLog,
|
||||
&table->wksps.writeCTable_wksp, sizeof(table->wksps.writeCTable_wksp)) );
|
||||
/* Check if using previous huffman table is beneficial */
|
||||
if (repeat && *repeat != HUF_repeat_none) {
|
||||
size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, table->count, maxSymbolValue);
|
||||
size_t const newSize = HUF_estimateCompressedSize(table->CTable, table->count, maxSymbolValue);
|
||||
if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
|
||||
return HUF_compressCTable_internal(ostart, op, oend,
|
||||
src, srcSize,
|
||||
nbStreams, oldHufTable, flags);
|
||||
} }
|
||||
|
||||
/* Use the new huffman table */
|
||||
if (hSize + 12ul >= srcSize) { return 0; }
|
||||
op += hSize;
|
||||
if (repeat) { *repeat = HUF_repeat_none; }
|
||||
if (oldHufTable)
|
||||
ZSTD_memcpy(oldHufTable, table->CTable, sizeof(table->CTable)); /* Save new table */
|
||||
}
|
||||
return HUF_compressCTable_internal(ostart, op, oend,
|
||||
src, srcSize,
|
||||
nbStreams, table->CTable, flags);
|
||||
}
|
||||
|
||||
size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
unsigned maxSymbolValue, unsigned huffLog,
|
||||
void* workSpace, size_t wkspSize,
|
||||
HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
|
||||
{
|
||||
DEBUGLOG(5, "HUF_compress1X_repeat (srcSize = %zu)", srcSize);
|
||||
return HUF_compress_internal(dst, dstSize, src, srcSize,
|
||||
maxSymbolValue, huffLog, HUF_singleStream,
|
||||
workSpace, wkspSize, hufTable,
|
||||
repeat, flags);
|
||||
}
|
||||
|
||||
/* HUF_compress4X_repeat():
|
||||
* compress input using 4 streams.
|
||||
* consider skipping quickly
|
||||
* reuse an existing huffman compression table */
|
||||
size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
|
||||
const void* src, size_t srcSize,
|
||||
unsigned maxSymbolValue, unsigned huffLog,
|
||||
void* workSpace, size_t wkspSize,
|
||||
HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
|
||||
{
|
||||
DEBUGLOG(5, "HUF_compress4X_repeat (srcSize = %zu)", srcSize);
|
||||
return HUF_compress_internal(dst, dstSize, src, srcSize,
|
||||
maxSymbolValue, huffLog, HUF_fourStreams,
|
||||
workSpace, wkspSize,
|
||||
hufTable, repeat, flags);
|
||||
}
|
||||
|
||||
@@ -22,6 +22,8 @@ zstd ABI:
|
||||
- `fse_decompress` builds FSE decoding tables and decodes FSE streams.
|
||||
- `fse_compress` normalizes counts, writes FSE headers, builds compression
|
||||
tables, and encodes FSE streams.
|
||||
- `huf_compress` builds Huffman compression tables, writes table headers,
|
||||
and encodes one- and four-stream Huffman payloads.
|
||||
- `huf_decompress` builds Huffman decoding tables and decodes X1 and X2
|
||||
Huffman streams.
|
||||
- Compression primitives
|
||||
|
||||
@@ -0,0 +1,1546 @@
|
||||
#![allow(non_snake_case)]
|
||||
|
||||
//! Huffman compression tables, headers, and single/four-stream encoders.
|
||||
//!
|
||||
//! This is the Rust implementation of `lib/compress/huf_compress.c`. The
|
||||
//! externally supplied workspaces remain part of the C ABI: their alignment
|
||||
//! and size checks are preserved even though the translation keeps temporary
|
||||
//! state in fixed Rust arrays.
|
||||
|
||||
use crate::bits::ZSTD_highbit32;
|
||||
use crate::entropy_common::{HUF_readStats, HUF_TABLELOG_MAX};
|
||||
use crate::errors::{ERR_isError, ZstdErrorCode, ERROR};
|
||||
use crate::fse_compress::{
|
||||
FSE_buildCTable_wksp, FSE_compress_usingCTable, FSE_normalizeCount, FSE_optimalTableLog,
|
||||
FSE_optimalTableLog_internal, FSE_writeNCount,
|
||||
};
|
||||
use crate::hist::{HIST_count_simple, HIST_count_wksp, HIST_WKSP_SIZE_U32};
|
||||
use crate::mem::MEM_writeLE16;
|
||||
use std::mem::size_of;
|
||||
use std::os::raw::{c_int, c_short, c_uint, c_void};
|
||||
|
||||
const HUF_BLOCKSIZE_MAX: usize = 128 * 1024;
|
||||
const HUF_TABLELOG_DEFAULT: u32 = 11;
|
||||
const HUF_TABLELOG_ABSOLUTEMAX: u32 = 12;
|
||||
const HUF_SYMBOLVALUE_MAX: u32 = 255;
|
||||
const HUF_CTABLE_SIZE_ST: usize = HUF_SYMBOLVALUE_MAX as usize + 2;
|
||||
const HUF_CTABLE_WORKSPACE_SIZE: usize = (4 * (HUF_SYMBOLVALUE_MAX as usize + 1) + 192) * 4;
|
||||
const HUF_WORKSPACE_SIZE: usize = (8 << 10) + 512;
|
||||
const HUF_WORKSPACE_MAX_ALIGNMENT: usize = 8;
|
||||
const MAX_FSE_TABLELOG_FOR_HUFF_HEADER: u32 = 6;
|
||||
const HUF_FLAGS_OPTIMAL_DEPTH: c_int = 1 << 1;
|
||||
const HUF_FLAGS_PREFER_REPEAT: c_int = 1 << 2;
|
||||
const HUF_FLAGS_SUSPECT_UNCOMPRESSIBLE: c_int = 1 << 3;
|
||||
const HUF_REPEAT_NONE: c_int = 0;
|
||||
const HUF_REPEAT_CHECK: c_int = 1;
|
||||
const HUF_REPEAT_VALID: c_int = 2;
|
||||
const SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE: usize = 4096;
|
||||
const SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO: usize = 10;
|
||||
const STARTNODE: usize = HUF_SYMBOLVALUE_MAX as usize + 1;
|
||||
const RANK_POSITION_TABLE_SIZE: usize = 192;
|
||||
const RANK_POSITION_LOG_BUCKETS_BEGIN: u32 = (RANK_POSITION_TABLE_SIZE as u32 - 1) - 32 - 1;
|
||||
const RANK_POSITION_DISTINCT_COUNT_CUTOFF: u32 = RANK_POSITION_LOG_BUCKETS_BEGIN + 7;
|
||||
|
||||
#[repr(C)]
|
||||
#[derive(Copy, Clone)]
|
||||
pub struct HUF_CTableHeader {
|
||||
pub tableLog: u8,
|
||||
pub maxSymbolValue: u8,
|
||||
pub unused: [u8; size_of::<usize>() - 2],
|
||||
}
|
||||
|
||||
#[repr(C)]
|
||||
#[derive(Copy, Clone, Default)]
|
||||
struct NodeElt {
|
||||
count: u32,
|
||||
parent: u16,
|
||||
byte: u8,
|
||||
nb_bits: u8,
|
||||
}
|
||||
|
||||
#[repr(C)]
|
||||
#[derive(Copy, Clone, Default)]
|
||||
struct RankPos {
|
||||
base: u16,
|
||||
curr: u16,
|
||||
}
|
||||
|
||||
#[inline]
|
||||
unsafe fn ctable_read(table: *const usize, index: usize) -> usize {
|
||||
table.add(index).read_unaligned()
|
||||
}
|
||||
|
||||
#[inline]
|
||||
unsafe fn ctable_write(table: *mut usize, index: usize, value: usize) {
|
||||
table.add(index).write_unaligned(value);
|
||||
}
|
||||
|
||||
#[inline]
|
||||
unsafe fn huf_read_ctable_header(table: *const usize) -> HUF_CTableHeader {
|
||||
let mut header = HUF_CTableHeader {
|
||||
tableLog: 0,
|
||||
maxSymbolValue: 0,
|
||||
unused: [0; size_of::<usize>() - 2],
|
||||
};
|
||||
std::ptr::copy_nonoverlapping(
|
||||
table.cast::<u8>(),
|
||||
(&mut header as *mut HUF_CTableHeader).cast::<u8>(),
|
||||
size_of::<HUF_CTableHeader>(),
|
||||
);
|
||||
header
|
||||
}
|
||||
|
||||
#[inline]
|
||||
unsafe fn huf_write_ctable_header(table: *mut usize, table_log: u32, max_symbol_value: u32) {
|
||||
let header = HUF_CTableHeader {
|
||||
tableLog: table_log as u8,
|
||||
maxSymbolValue: max_symbol_value as u8,
|
||||
unused: [0; size_of::<usize>() - 2],
|
||||
};
|
||||
std::ptr::copy_nonoverlapping(
|
||||
(&header as *const HUF_CTableHeader).cast::<u8>(),
|
||||
table.cast::<u8>(),
|
||||
size_of::<HUF_CTableHeader>(),
|
||||
);
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn huf_get_nb_bits(elt: usize) -> usize {
|
||||
elt & 0xff
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn huf_get_value(elt: usize) -> usize {
|
||||
elt & !0xffusize
|
||||
}
|
||||
|
||||
#[inline]
|
||||
unsafe fn huf_set_nb_bits(table: *mut usize, index: usize, nb_bits: usize) {
|
||||
debug_assert!(nb_bits <= HUF_TABLELOG_ABSOLUTEMAX as usize);
|
||||
ctable_write(table, index, nb_bits);
|
||||
}
|
||||
|
||||
#[inline]
|
||||
unsafe fn huf_set_value(table: *mut usize, index: usize, value: usize) {
|
||||
let elt = ctable_read(table, index);
|
||||
let nb_bits = huf_get_nb_bits(elt);
|
||||
if nb_bits != 0 {
|
||||
debug_assert!(value >> nb_bits == 0);
|
||||
ctable_write(
|
||||
table,
|
||||
index,
|
||||
elt | (value << (usize::BITS as usize - nb_bits)),
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn huf_aligned_workspace_size(
|
||||
workspace: *mut c_void,
|
||||
workspace_size: usize,
|
||||
align: usize,
|
||||
) -> usize {
|
||||
debug_assert!(align.is_power_of_two());
|
||||
debug_assert!(align <= HUF_WORKSPACE_MAX_ALIGNMENT);
|
||||
let add = (align - ((workspace as usize) & (align - 1))) & (align - 1);
|
||||
workspace_size.saturating_sub(add)
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_readCTableHeader(ctable: *const usize) -> HUF_CTableHeader {
|
||||
huf_read_ctable_header(ctable)
|
||||
}
|
||||
|
||||
unsafe fn huf_compress_weights(
|
||||
dst: *mut u8,
|
||||
dst_size: usize,
|
||||
weight_table: *const u8,
|
||||
wt_size: usize,
|
||||
) -> usize {
|
||||
if wt_size <= 1 {
|
||||
return 0;
|
||||
}
|
||||
|
||||
let mut count = [0u32; HUF_TABLELOG_MAX as usize + 1];
|
||||
let mut max_symbol_value = HUF_TABLELOG_MAX;
|
||||
let max_count = HIST_count_simple(
|
||||
count.as_mut_ptr(),
|
||||
&mut max_symbol_value,
|
||||
weight_table.cast::<c_void>(),
|
||||
wt_size,
|
||||
);
|
||||
if max_count as usize == wt_size {
|
||||
return 1;
|
||||
}
|
||||
if max_count == 1 {
|
||||
return 0;
|
||||
}
|
||||
|
||||
let table_log =
|
||||
FSE_optimalTableLog(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, wt_size, max_symbol_value);
|
||||
let mut norm = [0i16; HUF_TABLELOG_MAX as usize + 1];
|
||||
let result = FSE_normalizeCount(
|
||||
norm.as_mut_ptr().cast::<c_short>(),
|
||||
table_log,
|
||||
count.as_ptr(),
|
||||
wt_size,
|
||||
max_symbol_value,
|
||||
0,
|
||||
);
|
||||
if ERR_isError(result) {
|
||||
return result;
|
||||
}
|
||||
|
||||
let h_size = FSE_writeNCount(
|
||||
dst.cast::<c_void>(),
|
||||
dst_size,
|
||||
norm.as_ptr().cast::<c_short>(),
|
||||
max_symbol_value,
|
||||
table_log,
|
||||
);
|
||||
if ERR_isError(h_size) {
|
||||
return h_size;
|
||||
}
|
||||
if h_size > dst_size {
|
||||
return ERROR(ZstdErrorCode::DstSizeTooSmall);
|
||||
}
|
||||
|
||||
// FSE_CTABLE_SIZE_U32(6, HUF_TABLELOG_MAX) and the corresponding build
|
||||
// workspace macro from fse.h.
|
||||
let mut ctable = [0u32; 59];
|
||||
let mut scratch = [0u32; 41];
|
||||
let result = FSE_buildCTable_wksp(
|
||||
ctable.as_mut_ptr(),
|
||||
norm.as_ptr().cast::<c_short>(),
|
||||
max_symbol_value,
|
||||
table_log,
|
||||
scratch.as_mut_ptr().cast::<c_void>(),
|
||||
size_of::<[u32; 41]>(),
|
||||
);
|
||||
if ERR_isError(result) {
|
||||
return result;
|
||||
}
|
||||
|
||||
let c_size = FSE_compress_usingCTable(
|
||||
dst.add(h_size).cast::<c_void>(),
|
||||
dst_size - h_size,
|
||||
weight_table.cast::<c_void>(),
|
||||
wt_size,
|
||||
ctable.as_ptr(),
|
||||
);
|
||||
if ERR_isError(c_size) {
|
||||
return c_size;
|
||||
}
|
||||
if c_size == 0 {
|
||||
return 0;
|
||||
}
|
||||
h_size + c_size
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_writeCTable_wksp(
|
||||
dst: *mut c_void,
|
||||
max_dst_size: usize,
|
||||
ctable: *const usize,
|
||||
max_symbol_value: c_uint,
|
||||
huff_log: c_uint,
|
||||
workspace: *mut c_void,
|
||||
workspace_size: usize,
|
||||
) -> usize {
|
||||
let header = huf_read_ctable_header(ctable);
|
||||
debug_assert_eq!(u32::from(header.maxSymbolValue), max_symbol_value);
|
||||
debug_assert_eq!(u32::from(header.tableLog), huff_log);
|
||||
|
||||
if huf_aligned_workspace_size(workspace, workspace_size, size_of::<u32>())
|
||||
< HUF_CTABLE_WORKSPACE_SIZE
|
||||
{
|
||||
return ERROR(ZstdErrorCode::Generic);
|
||||
}
|
||||
if max_symbol_value > HUF_SYMBOLVALUE_MAX {
|
||||
return ERROR(ZstdErrorCode::MaxSymbolValueTooLarge);
|
||||
}
|
||||
if max_dst_size < 1 {
|
||||
return ERROR(ZstdErrorCode::DstSizeTooSmall);
|
||||
}
|
||||
|
||||
let mut bits_to_weight = [0u8; HUF_TABLELOG_MAX as usize + 1];
|
||||
for (n, weight) in bits_to_weight
|
||||
.iter_mut()
|
||||
.enumerate()
|
||||
.take(huff_log as usize + 1)
|
||||
.skip(1)
|
||||
{
|
||||
*weight = (huff_log as usize + 1 - n) as u8;
|
||||
}
|
||||
let mut huff_weight = [0u8; HUF_SYMBOLVALUE_MAX as usize + 1];
|
||||
for (n, weight) in huff_weight
|
||||
.iter_mut()
|
||||
.enumerate()
|
||||
.take(max_symbol_value as usize)
|
||||
{
|
||||
let bits = huf_get_nb_bits(ctable_read(ctable, n + 1));
|
||||
if bits > huff_log as usize {
|
||||
return ERROR(ZstdErrorCode::TableLogTooLarge);
|
||||
}
|
||||
*weight = bits_to_weight[bits];
|
||||
}
|
||||
|
||||
let output = dst.cast::<u8>();
|
||||
let h_size = huf_compress_weights(
|
||||
output.add(1),
|
||||
max_dst_size - 1,
|
||||
huff_weight.as_ptr(),
|
||||
max_symbol_value as usize,
|
||||
);
|
||||
if ERR_isError(h_size) {
|
||||
return h_size;
|
||||
}
|
||||
if h_size > 1 && h_size < max_symbol_value as usize / 2 {
|
||||
*output = h_size as u8;
|
||||
return h_size + 1;
|
||||
}
|
||||
|
||||
if max_symbol_value > 256 - 128 {
|
||||
return ERROR(ZstdErrorCode::Generic);
|
||||
}
|
||||
let raw_size = (max_symbol_value as usize).div_ceil(2) + 1;
|
||||
if raw_size > max_dst_size {
|
||||
return ERROR(ZstdErrorCode::DstSizeTooSmall);
|
||||
}
|
||||
*output = (128 + (max_symbol_value - 1)) as u8;
|
||||
huff_weight[max_symbol_value as usize] = 0;
|
||||
for n in (0..max_symbol_value as usize).step_by(2) {
|
||||
*output.add(n / 2 + 1) = (huff_weight[n] << 4).wrapping_add(huff_weight[n + 1]);
|
||||
}
|
||||
raw_size
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_readCTable(
|
||||
ctable: *mut usize,
|
||||
max_symbol_value_ptr: *mut c_uint,
|
||||
src: *const c_void,
|
||||
src_size: usize,
|
||||
has_zero_weights: *mut c_uint,
|
||||
) -> usize {
|
||||
let mut huff_weight = [0u8; HUF_SYMBOLVALUE_MAX as usize + 1];
|
||||
let mut rank_val = [0u32; HUF_TABLELOG_ABSOLUTEMAX as usize + 1];
|
||||
let mut table_log = 0u32;
|
||||
let mut nb_symbols = 0u32;
|
||||
let read_size = HUF_readStats(
|
||||
huff_weight.as_mut_ptr(),
|
||||
huff_weight.len(),
|
||||
rank_val.as_mut_ptr(),
|
||||
&mut nb_symbols,
|
||||
&mut table_log,
|
||||
src,
|
||||
src_size,
|
||||
);
|
||||
if ERR_isError(read_size) {
|
||||
return read_size;
|
||||
}
|
||||
*has_zero_weights = u32::from(rank_val[0] > 0);
|
||||
if table_log > HUF_TABLELOG_MAX {
|
||||
return ERROR(ZstdErrorCode::TableLogTooLarge);
|
||||
}
|
||||
if nb_symbols > *max_symbol_value_ptr + 1 {
|
||||
return ERROR(ZstdErrorCode::MaxSymbolValueTooSmall);
|
||||
}
|
||||
|
||||
*max_symbol_value_ptr = nb_symbols - 1;
|
||||
huf_write_ctable_header(ctable, table_log, *max_symbol_value_ptr);
|
||||
|
||||
let mut next_rank_start = 0u32;
|
||||
for (rank, rank_value) in rank_val
|
||||
.iter_mut()
|
||||
.enumerate()
|
||||
.take(table_log as usize + 1)
|
||||
.skip(1)
|
||||
{
|
||||
let current = next_rank_start;
|
||||
next_rank_start = next_rank_start.wrapping_add(*rank_value << (rank - 1));
|
||||
*rank_value = current;
|
||||
}
|
||||
|
||||
for (symbol, weight) in huff_weight
|
||||
.iter()
|
||||
.copied()
|
||||
.take(nb_symbols as usize)
|
||||
.enumerate()
|
||||
{
|
||||
let weight = weight as u32;
|
||||
let bits = if weight == 0 {
|
||||
0
|
||||
} else {
|
||||
table_log + 1 - weight
|
||||
};
|
||||
huf_set_nb_bits(ctable, symbol + 1, bits as usize);
|
||||
}
|
||||
|
||||
let mut nb_per_rank = [0u16; HUF_TABLELOG_MAX as usize + 2];
|
||||
let mut val_per_rank = [0u16; HUF_TABLELOG_MAX as usize + 2];
|
||||
for symbol in 0..nb_symbols as usize {
|
||||
let bits = huf_get_nb_bits(ctable_read(ctable, symbol + 1));
|
||||
nb_per_rank[bits] = nb_per_rank[bits].wrapping_add(1);
|
||||
}
|
||||
val_per_rank[table_log as usize + 1] = 0;
|
||||
let mut min = 0u16;
|
||||
for rank in (1..=table_log as usize).rev() {
|
||||
val_per_rank[rank] = min;
|
||||
min = min.wrapping_add(nb_per_rank[rank]);
|
||||
min >>= 1;
|
||||
}
|
||||
for symbol in 0..nb_symbols as usize {
|
||||
let bits = huf_get_nb_bits(ctable_read(ctable, symbol + 1));
|
||||
let value = val_per_rank[bits];
|
||||
huf_set_value(ctable, symbol + 1, value as usize);
|
||||
val_per_rank[bits] = val_per_rank[bits].wrapping_add(1);
|
||||
}
|
||||
read_size
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_getNbBitsFromCTable(ctable: *const usize, symbol_value: u32) -> u32 {
|
||||
if symbol_value > HUF_SYMBOLVALUE_MAX {
|
||||
return 0;
|
||||
}
|
||||
if symbol_value > u32::from(huf_read_ctable_header(ctable).maxSymbolValue) {
|
||||
return 0;
|
||||
}
|
||||
huf_get_nb_bits(ctable_read(ctable, symbol_value as usize + 1)) as u32
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn huf_get_index(count: u32) -> usize {
|
||||
if count < RANK_POSITION_DISTINCT_COUNT_CUTOFF {
|
||||
count as usize
|
||||
} else {
|
||||
(ZSTD_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN) as usize
|
||||
}
|
||||
}
|
||||
|
||||
fn huf_insertion_sort(nodes: &mut [NodeElt], low: i32, high: i32) {
|
||||
let size = high - low + 1;
|
||||
if size <= 1 {
|
||||
return;
|
||||
}
|
||||
for offset in 1..size {
|
||||
let key = nodes[(low + offset) as usize];
|
||||
let mut index = offset - 1;
|
||||
while index >= 0 && nodes[(low + index) as usize].count < key.count {
|
||||
nodes[(low + index + 1) as usize] = nodes[(low + index) as usize];
|
||||
index -= 1;
|
||||
}
|
||||
nodes[(low + index + 1) as usize] = key;
|
||||
}
|
||||
}
|
||||
|
||||
fn huf_quick_sort_partition(nodes: &mut [NodeElt], low: i32, high: i32) -> i32 {
|
||||
let pivot = nodes[high as usize].count;
|
||||
let mut i = low - 1;
|
||||
for j in low..high {
|
||||
if nodes[j as usize].count > pivot {
|
||||
i += 1;
|
||||
nodes.swap(i as usize, j as usize);
|
||||
}
|
||||
}
|
||||
nodes.swap((i + 1) as usize, high as usize);
|
||||
i + 1
|
||||
}
|
||||
|
||||
fn huf_simple_quick_sort(nodes: &mut [NodeElt], mut low: i32, mut high: i32) {
|
||||
const INSERTION_SORT_THRESHOLD: i32 = 8;
|
||||
if high - low < INSERTION_SORT_THRESHOLD {
|
||||
huf_insertion_sort(nodes, low, high);
|
||||
return;
|
||||
}
|
||||
while low < high {
|
||||
let index = huf_quick_sort_partition(nodes, low, high);
|
||||
if index - low < high - index {
|
||||
huf_simple_quick_sort(nodes, low, index - 1);
|
||||
low = index + 1;
|
||||
} else {
|
||||
huf_simple_quick_sort(nodes, index + 1, high);
|
||||
high = index - 1;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
fn huf_sort(
|
||||
nodes: &mut [NodeElt; 2 * (HUF_SYMBOLVALUE_MAX as usize + 1)],
|
||||
count: *const c_uint,
|
||||
max_symbol_value: u32,
|
||||
rank_position: &mut [RankPos; RANK_POSITION_TABLE_SIZE],
|
||||
) {
|
||||
let max_symbol_value1 = max_symbol_value as usize + 1;
|
||||
*rank_position = [RankPos::default(); RANK_POSITION_TABLE_SIZE];
|
||||
for symbol in 0..max_symbol_value1 {
|
||||
let lower_rank = huf_get_index(unsafe { *count.add(symbol) });
|
||||
debug_assert!(lower_rank < RANK_POSITION_TABLE_SIZE - 1);
|
||||
rank_position[lower_rank].base = rank_position[lower_rank].base.wrapping_add(1);
|
||||
}
|
||||
|
||||
for rank in (1..RANK_POSITION_TABLE_SIZE).rev() {
|
||||
rank_position[rank - 1].base = rank_position[rank - 1]
|
||||
.base
|
||||
.wrapping_add(rank_position[rank].base);
|
||||
rank_position[rank - 1].curr = rank_position[rank - 1].base;
|
||||
}
|
||||
|
||||
for symbol in 0..max_symbol_value1 {
|
||||
let current_count = unsafe { *count.add(symbol) };
|
||||
let rank = huf_get_index(current_count) + 1;
|
||||
let position = rank_position[rank].curr as usize;
|
||||
debug_assert!(position < max_symbol_value1);
|
||||
nodes[position + 1].count = current_count;
|
||||
nodes[position + 1].byte = symbol as u8;
|
||||
rank_position[rank].curr = rank_position[rank].curr.wrapping_add(1);
|
||||
}
|
||||
|
||||
for position in rank_position
|
||||
.iter()
|
||||
.take(RANK_POSITION_TABLE_SIZE - 1)
|
||||
.skip(RANK_POSITION_DISTINCT_COUNT_CUTOFF as usize)
|
||||
{
|
||||
let bucket_size = i32::from(position.curr) - i32::from(position.base);
|
||||
let bucket_start = position.base as usize;
|
||||
if bucket_size > 1 {
|
||||
huf_simple_quick_sort(
|
||||
&mut nodes[bucket_start + 1..bucket_start + 1 + bucket_size as usize],
|
||||
0,
|
||||
bucket_size - 1,
|
||||
);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn huff_node_index(index: usize) -> usize {
|
||||
index + 1
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn huff_node_index_signed(index: i32) -> usize {
|
||||
debug_assert!(index >= -1);
|
||||
(index + 1) as usize
|
||||
}
|
||||
|
||||
fn huf_build_tree(
|
||||
nodes: &mut [NodeElt; 2 * (HUF_SYMBOLVALUE_MAX as usize + 1)],
|
||||
max_symbol_value: u32,
|
||||
) -> Option<usize> {
|
||||
let mut non_null_rank = max_symbol_value as i32;
|
||||
while non_null_rank >= 0 && nodes[huff_node_index(non_null_rank as usize)].count == 0 {
|
||||
non_null_rank -= 1;
|
||||
}
|
||||
if non_null_rank < 1 {
|
||||
return None;
|
||||
}
|
||||
|
||||
let mut low_s = non_null_rank;
|
||||
let node_root = STARTNODE as i32 + low_s - 1;
|
||||
let mut low_n = STARTNODE as i32;
|
||||
let mut node_nb = STARTNODE as i32;
|
||||
let first_count = nodes[huff_node_index_signed(low_s)].count;
|
||||
let second_count = nodes[huff_node_index_signed(low_s - 1)].count;
|
||||
nodes[huff_node_index(node_nb as usize)].count = first_count.wrapping_add(second_count);
|
||||
nodes[huff_node_index_signed(low_s)].parent = node_nb as u16;
|
||||
nodes[huff_node_index_signed(low_s - 1)].parent = node_nb as u16;
|
||||
node_nb += 1;
|
||||
low_s -= 2;
|
||||
|
||||
for node in node_nb..=node_root {
|
||||
nodes[huff_node_index(node as usize)].count = 1 << 30;
|
||||
}
|
||||
nodes[0].count = 1 << 31;
|
||||
|
||||
while node_nb <= node_root {
|
||||
let n1 = if nodes[huff_node_index_signed(low_s)].count
|
||||
< nodes[huff_node_index_signed(low_n)].count
|
||||
{
|
||||
let node = low_s;
|
||||
low_s -= 1;
|
||||
node
|
||||
} else {
|
||||
let node = low_n;
|
||||
low_n += 1;
|
||||
node
|
||||
};
|
||||
let n2 = if nodes[huff_node_index_signed(low_s)].count
|
||||
< nodes[huff_node_index_signed(low_n)].count
|
||||
{
|
||||
let node = low_s;
|
||||
low_s -= 1;
|
||||
node
|
||||
} else {
|
||||
let node = low_n;
|
||||
low_n += 1;
|
||||
node
|
||||
};
|
||||
nodes[huff_node_index(node_nb as usize)].count = nodes[huff_node_index_signed(n1)]
|
||||
.count
|
||||
.wrapping_add(nodes[huff_node_index_signed(n2)].count);
|
||||
nodes[huff_node_index_signed(n1)].parent = node_nb as u16;
|
||||
nodes[huff_node_index_signed(n2)].parent = node_nb as u16;
|
||||
node_nb += 1;
|
||||
}
|
||||
|
||||
nodes[huff_node_index(node_root as usize)].nb_bits = 0;
|
||||
for node in (STARTNODE as i32..node_root).rev() {
|
||||
let parent = nodes[huff_node_index(node as usize)].parent as usize;
|
||||
nodes[huff_node_index(node as usize)].nb_bits =
|
||||
nodes[huff_node_index(parent)].nb_bits.wrapping_add(1);
|
||||
}
|
||||
for node in 0..=non_null_rank as usize {
|
||||
let parent = nodes[huff_node_index(node)].parent as usize;
|
||||
nodes[huff_node_index(node)].nb_bits =
|
||||
nodes[huff_node_index(parent)].nb_bits.wrapping_add(1);
|
||||
}
|
||||
Some(non_null_rank as usize)
|
||||
}
|
||||
|
||||
fn huf_set_max_height(
|
||||
nodes: &mut [NodeElt; 2 * (HUF_SYMBOLVALUE_MAX as usize + 1)],
|
||||
last_non_null: usize,
|
||||
target_nb_bits: u32,
|
||||
) -> u32 {
|
||||
let largest_bits = u32::from(nodes[huff_node_index(last_non_null)].nb_bits);
|
||||
if largest_bits <= target_nb_bits {
|
||||
return largest_bits;
|
||||
}
|
||||
|
||||
let mut total_cost = 0i32;
|
||||
let base_cost = 1i32 << (largest_bits - target_nb_bits);
|
||||
let mut n = last_non_null as i32;
|
||||
while u32::from(nodes[huff_node_index(n as usize)].nb_bits) > target_nb_bits {
|
||||
let bits = u32::from(nodes[huff_node_index(n as usize)].nb_bits);
|
||||
total_cost += base_cost - (1i32 << (largest_bits - bits));
|
||||
nodes[huff_node_index(n as usize)].nb_bits = target_nb_bits as u8;
|
||||
n -= 1;
|
||||
}
|
||||
while n >= 0 && u32::from(nodes[huff_node_index(n as usize)].nb_bits) == target_nb_bits {
|
||||
n -= 1;
|
||||
}
|
||||
if n < 0 {
|
||||
return target_nb_bits;
|
||||
}
|
||||
|
||||
total_cost >>= largest_bits - target_nb_bits;
|
||||
const NO_SYMBOL: u32 = 0xf0f0_f0f0;
|
||||
let mut rank_last = [NO_SYMBOL; HUF_TABLELOG_MAX as usize + 2];
|
||||
let mut current_nb_bits = target_nb_bits;
|
||||
for position in (0..=n as usize).rev() {
|
||||
let bits = u32::from(nodes[huff_node_index(position)].nb_bits);
|
||||
if bits >= current_nb_bits {
|
||||
continue;
|
||||
}
|
||||
current_nb_bits = bits;
|
||||
rank_last[(target_nb_bits - current_nb_bits) as usize] = position as u32;
|
||||
}
|
||||
|
||||
while total_cost > 0 {
|
||||
let mut bits_to_decrease = ZSTD_highbit32(total_cost as u32) + 1;
|
||||
while bits_to_decrease > 1 {
|
||||
let high_position = rank_last[bits_to_decrease as usize];
|
||||
let low_position = rank_last[(bits_to_decrease - 1) as usize];
|
||||
if high_position == NO_SYMBOL {
|
||||
bits_to_decrease -= 1;
|
||||
continue;
|
||||
}
|
||||
if low_position == NO_SYMBOL {
|
||||
break;
|
||||
}
|
||||
let high_total = nodes[huff_node_index(high_position as usize)].count;
|
||||
let low_total = nodes[huff_node_index(low_position as usize)]
|
||||
.count
|
||||
.wrapping_mul(2);
|
||||
if high_total <= low_total {
|
||||
break;
|
||||
}
|
||||
bits_to_decrease -= 1;
|
||||
}
|
||||
while bits_to_decrease <= HUF_TABLELOG_MAX
|
||||
&& rank_last[bits_to_decrease as usize] == NO_SYMBOL
|
||||
{
|
||||
bits_to_decrease += 1;
|
||||
}
|
||||
if bits_to_decrease > HUF_TABLELOG_MAX + 1 {
|
||||
return target_nb_bits;
|
||||
}
|
||||
let position = rank_last[bits_to_decrease as usize] as usize;
|
||||
total_cost -= 1i32 << (bits_to_decrease - 1);
|
||||
nodes[huff_node_index(position)].nb_bits =
|
||||
nodes[huff_node_index(position)].nb_bits.wrapping_add(1);
|
||||
|
||||
if rank_last[(bits_to_decrease - 1) as usize] == NO_SYMBOL {
|
||||
rank_last[(bits_to_decrease - 1) as usize] = rank_last[bits_to_decrease as usize];
|
||||
}
|
||||
if rank_last[bits_to_decrease as usize] == 0 {
|
||||
rank_last[bits_to_decrease as usize] = NO_SYMBOL;
|
||||
} else {
|
||||
rank_last[bits_to_decrease as usize] -= 1;
|
||||
let previous = rank_last[bits_to_decrease as usize] as usize;
|
||||
if u32::from(nodes[huff_node_index(previous)].nb_bits)
|
||||
!= target_nb_bits - bits_to_decrease
|
||||
{
|
||||
rank_last[bits_to_decrease as usize] = NO_SYMBOL;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
while total_cost < 0 {
|
||||
if rank_last[1] == NO_SYMBOL {
|
||||
while n >= 0 && u32::from(nodes[huff_node_index(n as usize)].nb_bits) == target_nb_bits
|
||||
{
|
||||
n -= 1;
|
||||
}
|
||||
if n < 0 {
|
||||
return target_nb_bits;
|
||||
}
|
||||
let position = n as usize + 1;
|
||||
nodes[huff_node_index(position)].nb_bits =
|
||||
nodes[huff_node_index(position)].nb_bits.wrapping_sub(1);
|
||||
rank_last[1] = position as u32;
|
||||
total_cost += 1;
|
||||
continue;
|
||||
}
|
||||
let position = rank_last[1] as usize + 1;
|
||||
nodes[huff_node_index(position)].nb_bits =
|
||||
nodes[huff_node_index(position)].nb_bits.wrapping_sub(1);
|
||||
rank_last[1] = rank_last[1].wrapping_add(1);
|
||||
total_cost += 1;
|
||||
}
|
||||
target_nb_bits
|
||||
}
|
||||
|
||||
unsafe fn huf_build_ctable_from_tree(
|
||||
ctable: *mut usize,
|
||||
nodes: &[NodeElt; 2 * (HUF_SYMBOLVALUE_MAX as usize + 1)],
|
||||
non_null_rank: usize,
|
||||
max_symbol_value: u32,
|
||||
max_nb_bits: u32,
|
||||
) {
|
||||
let mut nb_per_rank = [0u16; HUF_TABLELOG_MAX as usize + 1];
|
||||
let mut val_per_rank = [0u16; HUF_TABLELOG_MAX as usize + 1];
|
||||
for node in 0..=non_null_rank {
|
||||
let bits = nodes[huff_node_index(node)].nb_bits as usize;
|
||||
nb_per_rank[bits] = nb_per_rank[bits].wrapping_add(1);
|
||||
}
|
||||
let mut min = 0u16;
|
||||
for rank in (1..=max_nb_bits as usize).rev() {
|
||||
val_per_rank[rank] = min;
|
||||
min = min.wrapping_add(nb_per_rank[rank]);
|
||||
min >>= 1;
|
||||
}
|
||||
for node in 0..=max_symbol_value as usize {
|
||||
let symbol = nodes[huff_node_index(node)].byte as usize;
|
||||
huf_set_nb_bits(
|
||||
ctable,
|
||||
symbol + 1,
|
||||
nodes[huff_node_index(node)].nb_bits as usize,
|
||||
);
|
||||
}
|
||||
for symbol in 0..=max_symbol_value as usize {
|
||||
let bits = huf_get_nb_bits(ctable_read(ctable, symbol + 1));
|
||||
let value = val_per_rank[bits];
|
||||
huf_set_value(ctable, symbol + 1, value as usize);
|
||||
val_per_rank[bits] = val_per_rank[bits].wrapping_add(1);
|
||||
}
|
||||
huf_write_ctable_header(ctable, max_nb_bits, max_symbol_value);
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_buildCTable_wksp(
|
||||
ctable: *mut usize,
|
||||
count: *const c_uint,
|
||||
max_symbol_value: u32,
|
||||
mut max_nb_bits: u32,
|
||||
workspace: *mut c_void,
|
||||
workspace_size: usize,
|
||||
) -> usize {
|
||||
if huf_aligned_workspace_size(workspace, workspace_size, size_of::<u32>())
|
||||
< HUF_CTABLE_WORKSPACE_SIZE
|
||||
{
|
||||
return ERROR(ZstdErrorCode::WorkSpaceTooSmall);
|
||||
}
|
||||
if max_nb_bits == 0 {
|
||||
max_nb_bits = HUF_TABLELOG_DEFAULT;
|
||||
}
|
||||
if max_symbol_value > HUF_SYMBOLVALUE_MAX {
|
||||
return ERROR(ZstdErrorCode::MaxSymbolValueTooLarge);
|
||||
}
|
||||
|
||||
let mut nodes = [NodeElt::default(); 2 * (HUF_SYMBOLVALUE_MAX as usize + 1)];
|
||||
let mut rank_position = [RankPos::default(); RANK_POSITION_TABLE_SIZE];
|
||||
huf_sort(&mut nodes, count, max_symbol_value, &mut rank_position);
|
||||
let non_null_rank = match huf_build_tree(&mut nodes, max_symbol_value) {
|
||||
Some(rank) => rank,
|
||||
None => return ERROR(ZstdErrorCode::Generic),
|
||||
};
|
||||
max_nb_bits = huf_set_max_height(&mut nodes, non_null_rank, max_nb_bits);
|
||||
if max_nb_bits > HUF_TABLELOG_MAX {
|
||||
return ERROR(ZstdErrorCode::Generic);
|
||||
}
|
||||
huf_build_ctable_from_tree(ctable, &nodes, non_null_rank, max_symbol_value, max_nb_bits);
|
||||
max_nb_bits as usize
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_estimateCompressedSize(
|
||||
ctable: *const usize,
|
||||
count: *const c_uint,
|
||||
max_symbol_value: c_uint,
|
||||
) -> usize {
|
||||
let mut bits = 0usize;
|
||||
for symbol in 0..=max_symbol_value as usize {
|
||||
bits = bits.wrapping_add(
|
||||
huf_get_nb_bits(ctable_read(ctable, symbol + 1))
|
||||
.wrapping_mul(*count.add(symbol) as usize),
|
||||
);
|
||||
}
|
||||
bits >> 3
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_validateCTable(
|
||||
ctable: *const usize,
|
||||
count: *const c_uint,
|
||||
max_symbol_value: c_uint,
|
||||
) -> c_int {
|
||||
let header = huf_read_ctable_header(ctable);
|
||||
if u32::from(header.maxSymbolValue) < max_symbol_value {
|
||||
return 0;
|
||||
}
|
||||
for symbol in 0..=max_symbol_value as usize {
|
||||
if *count.add(symbol) != 0 && huf_get_nb_bits(ctable_read(ctable, symbol + 1)) == 0 {
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
1
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub extern "C" fn HUF_compressBound(size: usize) -> usize {
|
||||
129usize
|
||||
.wrapping_add(size)
|
||||
.wrapping_add(size >> 8)
|
||||
.wrapping_add(8)
|
||||
}
|
||||
|
||||
struct HufCStream {
|
||||
bit_container: usize,
|
||||
bit_pos: usize,
|
||||
dst: *mut u8,
|
||||
ptr: usize,
|
||||
end: usize,
|
||||
}
|
||||
|
||||
impl HufCStream {
|
||||
unsafe fn init(dst: *mut u8, dst_capacity: usize) -> Option<Self> {
|
||||
let container_size = size_of::<usize>();
|
||||
if dst_capacity <= container_size {
|
||||
return None;
|
||||
}
|
||||
Some(Self {
|
||||
bit_container: 0,
|
||||
bit_pos: 0,
|
||||
dst,
|
||||
ptr: 0,
|
||||
end: dst_capacity - container_size,
|
||||
})
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn add_bits(&mut self, elt: usize) {
|
||||
let nb_bits = huf_get_nb_bits(elt);
|
||||
self.bit_container >>= nb_bits;
|
||||
self.bit_container |= huf_get_value(elt);
|
||||
self.bit_pos = self.bit_pos.wrapping_add(nb_bits);
|
||||
}
|
||||
|
||||
unsafe fn flush_bits(&mut self) {
|
||||
let nb_bits = self.bit_pos & 0xff;
|
||||
if nb_bits == 0 {
|
||||
return;
|
||||
}
|
||||
let nb_bytes = nb_bits >> 3;
|
||||
let bit_container = self.bit_container >> (usize::BITS as usize - nb_bits);
|
||||
for byte in 0..nb_bytes {
|
||||
// The C encoder writes a full native word at `ptr`, then clamps the
|
||||
// logical cursor. Only materialize the bytes that are logically
|
||||
// part of the stream; `ptr <= end` guarantees those are in bounds.
|
||||
*self.dst.add(self.ptr + byte) = (bit_container >> (8 * byte)) as u8;
|
||||
}
|
||||
self.ptr = self.ptr.saturating_add(nb_bytes);
|
||||
if self.ptr > self.end {
|
||||
self.ptr = self.end;
|
||||
}
|
||||
self.bit_pos &= 7;
|
||||
}
|
||||
|
||||
unsafe fn close(mut self) -> usize {
|
||||
let end_mark = 1usize | (1usize << (usize::BITS as usize - 1));
|
||||
self.add_bits(end_mark);
|
||||
self.flush_bits();
|
||||
if self.ptr >= self.end {
|
||||
return 0;
|
||||
}
|
||||
if self.bit_pos & 0xff != 0 {
|
||||
let partial = self.bit_container >> (usize::BITS as usize - (self.bit_pos & 0xff));
|
||||
*self.dst.add(self.ptr) = partial as u8;
|
||||
}
|
||||
self.ptr + usize::from((self.bit_pos & 0xff) > 0)
|
||||
}
|
||||
}
|
||||
|
||||
unsafe fn huf_compress1x_using_ctable_internal(
|
||||
dst: *mut u8,
|
||||
dst_size: usize,
|
||||
src: *const u8,
|
||||
src_size: usize,
|
||||
ctable: *const usize,
|
||||
) -> usize {
|
||||
if dst_size < 8 {
|
||||
return 0;
|
||||
}
|
||||
let mut stream = match HufCStream::init(dst, dst_size) {
|
||||
Some(stream) => stream,
|
||||
None => return 0,
|
||||
};
|
||||
for index in (0..src_size).rev() {
|
||||
let symbol = *src.add(index) as usize;
|
||||
stream.add_bits(ctable_read(ctable, symbol + 1));
|
||||
// Keeping the pending tail below one byte makes the portable Rust
|
||||
// stream independent of the C implementation's BMI2-oriented unroll
|
||||
// schedule while retaining the exact HUF bitstream representation.
|
||||
stream.flush_bits();
|
||||
}
|
||||
stream.close()
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_compress1X_usingCTable(
|
||||
dst: *mut c_void,
|
||||
dst_size: usize,
|
||||
src: *const c_void,
|
||||
src_size: usize,
|
||||
ctable: *const usize,
|
||||
_flags: c_int,
|
||||
) -> usize {
|
||||
huf_compress1x_using_ctable_internal(dst.cast(), dst_size, src.cast(), src_size, ctable)
|
||||
}
|
||||
|
||||
unsafe fn huf_compress4x_using_ctable_internal(
|
||||
dst: *mut u8,
|
||||
dst_size: usize,
|
||||
src: *const u8,
|
||||
src_size: usize,
|
||||
ctable: *const usize,
|
||||
) -> usize {
|
||||
if dst_size < 6 + 1 + 1 + 1 + 8 || src_size < 12 {
|
||||
return 0;
|
||||
}
|
||||
let segment_size = src_size.div_ceil(4);
|
||||
let mut input_offset = 0usize;
|
||||
let mut output_offset = 6usize;
|
||||
for segment in 0..4 {
|
||||
let input_size = if segment < 3 {
|
||||
segment_size
|
||||
} else {
|
||||
src_size - input_offset
|
||||
};
|
||||
let c_size = huf_compress1x_using_ctable_internal(
|
||||
dst.add(output_offset),
|
||||
dst_size - output_offset,
|
||||
src.add(input_offset),
|
||||
input_size,
|
||||
ctable,
|
||||
);
|
||||
if c_size == 0 || c_size > u16::MAX as usize {
|
||||
return 0;
|
||||
}
|
||||
if segment < 3 {
|
||||
MEM_writeLE16(dst.add(segment * 2).cast::<c_void>(), c_size as u16);
|
||||
}
|
||||
output_offset += c_size;
|
||||
input_offset += input_size;
|
||||
}
|
||||
output_offset
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_compress4X_usingCTable(
|
||||
dst: *mut c_void,
|
||||
dst_size: usize,
|
||||
src: *const c_void,
|
||||
src_size: usize,
|
||||
ctable: *const usize,
|
||||
_flags: c_int,
|
||||
) -> usize {
|
||||
huf_compress4x_using_ctable_internal(dst.cast(), dst_size, src.cast(), src_size, ctable)
|
||||
}
|
||||
|
||||
#[allow(clippy::too_many_arguments)]
|
||||
unsafe fn huf_compress_ctable_internal(
|
||||
dst: *mut u8,
|
||||
dst_size: usize,
|
||||
prefix_size: usize,
|
||||
src: *const u8,
|
||||
src_size: usize,
|
||||
four_streams: bool,
|
||||
ctable: *const usize,
|
||||
flags: c_int,
|
||||
) -> usize {
|
||||
if prefix_size > dst_size {
|
||||
return 0;
|
||||
}
|
||||
let c_size = if four_streams {
|
||||
HUF_compress4X_usingCTable(
|
||||
dst.add(prefix_size).cast::<c_void>(),
|
||||
dst_size - prefix_size,
|
||||
src.cast::<c_void>(),
|
||||
src_size,
|
||||
ctable,
|
||||
flags,
|
||||
)
|
||||
} else {
|
||||
HUF_compress1X_usingCTable(
|
||||
dst.add(prefix_size).cast::<c_void>(),
|
||||
dst_size - prefix_size,
|
||||
src.cast::<c_void>(),
|
||||
src_size,
|
||||
ctable,
|
||||
flags,
|
||||
)
|
||||
};
|
||||
if ERR_isError(c_size) {
|
||||
return c_size;
|
||||
}
|
||||
let total_size = prefix_size.wrapping_add(c_size);
|
||||
if c_size == 0 || total_size >= src_size.saturating_sub(1) {
|
||||
return 0;
|
||||
}
|
||||
total_size
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_cardinality(count: *const c_uint, max_symbol_value: c_uint) -> c_uint {
|
||||
let mut cardinality = 0u32;
|
||||
for symbol in 0..=max_symbol_value as usize {
|
||||
if *count.add(symbol) != 0 {
|
||||
cardinality += 1;
|
||||
}
|
||||
}
|
||||
cardinality
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub extern "C" fn HUF_minTableLog(symbol_cardinality: c_uint) -> c_uint {
|
||||
ZSTD_highbit32(symbol_cardinality) + 1
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_optimalTableLog(
|
||||
max_table_log: c_uint,
|
||||
src_size: usize,
|
||||
max_symbol_value: c_uint,
|
||||
workspace: *mut c_void,
|
||||
workspace_size: usize,
|
||||
table: *mut usize,
|
||||
count: *const c_uint,
|
||||
flags: c_int,
|
||||
) -> c_uint {
|
||||
if flags & HUF_FLAGS_OPTIMAL_DEPTH == 0 {
|
||||
return FSE_optimalTableLog_internal(max_table_log, src_size, max_symbol_value, 1);
|
||||
}
|
||||
|
||||
let symbol_cardinality = HUF_cardinality(count, max_symbol_value);
|
||||
if symbol_cardinality == 0 {
|
||||
return max_table_log;
|
||||
}
|
||||
let min_table_log = HUF_minTableLog(symbol_cardinality);
|
||||
let mut opt_size = usize::MAX - 1;
|
||||
let mut opt_log = max_table_log;
|
||||
let mut probe_dst = [0u8; HUF_WORKSPACE_SIZE];
|
||||
let write_workspace_size = workspace_size.saturating_sub(748);
|
||||
|
||||
for guess in min_table_log..=max_table_log {
|
||||
let max_bits = HUF_buildCTable_wksp(
|
||||
table,
|
||||
count,
|
||||
max_symbol_value,
|
||||
guess,
|
||||
workspace,
|
||||
workspace_size,
|
||||
);
|
||||
if ERR_isError(max_bits) {
|
||||
continue;
|
||||
}
|
||||
if max_bits < guess as usize && guess > min_table_log {
|
||||
break;
|
||||
}
|
||||
let h_size = HUF_writeCTable_wksp(
|
||||
probe_dst.as_mut_ptr().cast::<c_void>(),
|
||||
write_workspace_size.min(probe_dst.len()),
|
||||
table,
|
||||
max_symbol_value,
|
||||
max_bits as u32,
|
||||
workspace,
|
||||
workspace_size,
|
||||
);
|
||||
if ERR_isError(h_size) {
|
||||
continue;
|
||||
}
|
||||
let new_size =
|
||||
HUF_estimateCompressedSize(table, count, max_symbol_value).wrapping_add(h_size);
|
||||
if new_size > opt_size.wrapping_add(1) {
|
||||
break;
|
||||
}
|
||||
if new_size < opt_size {
|
||||
opt_size = new_size;
|
||||
opt_log = guess;
|
||||
}
|
||||
}
|
||||
opt_log
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn huf_compress_tables_size() -> usize {
|
||||
(HUF_SYMBOLVALUE_MAX as usize + 1) * size_of::<u32>()
|
||||
+ HUF_CTABLE_SIZE_ST * size_of::<usize>()
|
||||
+ HUF_CTABLE_WORKSPACE_SIZE
|
||||
}
|
||||
|
||||
#[allow(clippy::too_many_arguments)]
|
||||
unsafe fn huf_compress_internal(
|
||||
dst: *mut u8,
|
||||
dst_size: usize,
|
||||
src: *const u8,
|
||||
src_size: usize,
|
||||
mut max_symbol_value: u32,
|
||||
mut huff_log: u32,
|
||||
four_streams: bool,
|
||||
workspace: *mut c_void,
|
||||
workspace_size: usize,
|
||||
old_huf_table: *mut usize,
|
||||
repeat: *mut c_int,
|
||||
flags: c_int,
|
||||
) -> usize {
|
||||
if huf_aligned_workspace_size(workspace, workspace_size, size_of::<usize>())
|
||||
< huf_compress_tables_size()
|
||||
{
|
||||
return ERROR(ZstdErrorCode::WorkSpaceTooSmall);
|
||||
}
|
||||
if src_size == 0 || dst_size == 0 {
|
||||
return 0;
|
||||
}
|
||||
if src_size > HUF_BLOCKSIZE_MAX {
|
||||
return ERROR(ZstdErrorCode::SrcSizeWrong);
|
||||
}
|
||||
if huff_log > HUF_TABLELOG_MAX {
|
||||
return ERROR(ZstdErrorCode::TableLogTooLarge);
|
||||
}
|
||||
if max_symbol_value > HUF_SYMBOLVALUE_MAX {
|
||||
return ERROR(ZstdErrorCode::MaxSymbolValueTooLarge);
|
||||
}
|
||||
if max_symbol_value == 0 {
|
||||
max_symbol_value = HUF_SYMBOLVALUE_MAX;
|
||||
}
|
||||
if huff_log == 0 {
|
||||
huff_log = HUF_TABLELOG_DEFAULT;
|
||||
}
|
||||
|
||||
if flags & HUF_FLAGS_PREFER_REPEAT != 0
|
||||
&& !repeat.is_null()
|
||||
&& *repeat == HUF_REPEAT_VALID
|
||||
&& !old_huf_table.is_null()
|
||||
{
|
||||
return huf_compress_ctable_internal(
|
||||
dst,
|
||||
dst_size,
|
||||
0,
|
||||
src,
|
||||
src_size,
|
||||
four_streams,
|
||||
old_huf_table,
|
||||
flags,
|
||||
);
|
||||
}
|
||||
|
||||
let mut count = [0u32; HUF_SYMBOLVALUE_MAX as usize + 1];
|
||||
if flags & HUF_FLAGS_SUSPECT_UNCOMPRESSIBLE != 0
|
||||
&& src_size >= SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO
|
||||
{
|
||||
let mut largest_total = 0usize;
|
||||
let mut sample_max = max_symbol_value;
|
||||
largest_total += HIST_count_simple(
|
||||
count.as_mut_ptr(),
|
||||
&mut sample_max,
|
||||
src.cast::<c_void>(),
|
||||
SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE,
|
||||
) as usize;
|
||||
sample_max = max_symbol_value;
|
||||
largest_total += HIST_count_simple(
|
||||
count.as_mut_ptr(),
|
||||
&mut sample_max,
|
||||
src.add(src_size - SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE)
|
||||
.cast::<c_void>(),
|
||||
SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE,
|
||||
) as usize;
|
||||
if largest_total <= ((2 * SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) >> 7).wrapping_add(4) {
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
let mut hist_workspace = [0u32; HIST_WKSP_SIZE_U32];
|
||||
let largest = HIST_count_wksp(
|
||||
count.as_mut_ptr(),
|
||||
&mut max_symbol_value,
|
||||
src.cast::<c_void>(),
|
||||
src_size,
|
||||
hist_workspace.as_mut_ptr().cast::<c_void>(),
|
||||
size_of::<[u32; HIST_WKSP_SIZE_U32]>(),
|
||||
);
|
||||
if ERR_isError(largest) {
|
||||
return largest;
|
||||
}
|
||||
if largest == src_size {
|
||||
*dst = *src;
|
||||
return 1;
|
||||
}
|
||||
if largest <= (src_size >> 7).wrapping_add(4) {
|
||||
return 0;
|
||||
}
|
||||
|
||||
if !repeat.is_null()
|
||||
&& *repeat == HUF_REPEAT_CHECK
|
||||
&& (old_huf_table.is_null()
|
||||
|| HUF_validateCTable(old_huf_table, count.as_ptr(), max_symbol_value) == 0)
|
||||
{
|
||||
*repeat = HUF_REPEAT_NONE;
|
||||
}
|
||||
if flags & HUF_FLAGS_PREFER_REPEAT != 0
|
||||
&& !repeat.is_null()
|
||||
&& *repeat != HUF_REPEAT_NONE
|
||||
&& !old_huf_table.is_null()
|
||||
{
|
||||
return huf_compress_ctable_internal(
|
||||
dst,
|
||||
dst_size,
|
||||
0,
|
||||
src,
|
||||
src_size,
|
||||
four_streams,
|
||||
old_huf_table,
|
||||
flags,
|
||||
);
|
||||
}
|
||||
|
||||
let mut ctable = [0usize; HUF_CTABLE_SIZE_ST];
|
||||
huff_log = HUF_optimalTableLog(
|
||||
huff_log,
|
||||
src_size,
|
||||
max_symbol_value,
|
||||
workspace,
|
||||
HUF_CTABLE_WORKSPACE_SIZE,
|
||||
ctable.as_mut_ptr(),
|
||||
count.as_ptr(),
|
||||
flags,
|
||||
);
|
||||
let max_bits = HUF_buildCTable_wksp(
|
||||
ctable.as_mut_ptr(),
|
||||
count.as_ptr(),
|
||||
max_symbol_value,
|
||||
huff_log,
|
||||
workspace,
|
||||
HUF_CTABLE_WORKSPACE_SIZE,
|
||||
);
|
||||
if ERR_isError(max_bits) {
|
||||
return max_bits;
|
||||
}
|
||||
huff_log = max_bits as u32;
|
||||
|
||||
let header_size = HUF_writeCTable_wksp(
|
||||
dst.cast::<c_void>(),
|
||||
dst_size,
|
||||
ctable.as_ptr(),
|
||||
max_symbol_value,
|
||||
huff_log,
|
||||
workspace,
|
||||
HUF_CTABLE_WORKSPACE_SIZE,
|
||||
);
|
||||
if ERR_isError(header_size) {
|
||||
return header_size;
|
||||
}
|
||||
if !repeat.is_null() && *repeat != HUF_REPEAT_NONE && !old_huf_table.is_null() {
|
||||
let old_size = HUF_estimateCompressedSize(old_huf_table, count.as_ptr(), max_symbol_value);
|
||||
let new_size =
|
||||
HUF_estimateCompressedSize(ctable.as_ptr(), count.as_ptr(), max_symbol_value);
|
||||
if old_size <= header_size.wrapping_add(new_size)
|
||||
|| header_size.wrapping_add(12) >= src_size
|
||||
{
|
||||
return huf_compress_ctable_internal(
|
||||
dst,
|
||||
dst_size,
|
||||
0,
|
||||
src,
|
||||
src_size,
|
||||
four_streams,
|
||||
old_huf_table,
|
||||
flags,
|
||||
);
|
||||
}
|
||||
}
|
||||
if header_size.wrapping_add(12) >= src_size {
|
||||
return 0;
|
||||
}
|
||||
if !repeat.is_null() {
|
||||
*repeat = HUF_REPEAT_NONE;
|
||||
}
|
||||
if !old_huf_table.is_null() {
|
||||
std::ptr::copy_nonoverlapping(
|
||||
ctable.as_ptr().cast::<u8>(),
|
||||
old_huf_table.cast::<u8>(),
|
||||
size_of::<[usize; HUF_CTABLE_SIZE_ST]>(),
|
||||
);
|
||||
}
|
||||
huf_compress_ctable_internal(
|
||||
dst,
|
||||
dst_size,
|
||||
header_size,
|
||||
src,
|
||||
src_size,
|
||||
four_streams,
|
||||
ctable.as_ptr(),
|
||||
flags,
|
||||
)
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_compress1X_repeat(
|
||||
dst: *mut c_void,
|
||||
dst_size: usize,
|
||||
src: *const c_void,
|
||||
src_size: usize,
|
||||
max_symbol_value: c_uint,
|
||||
huff_log: c_uint,
|
||||
workspace: *mut c_void,
|
||||
workspace_size: usize,
|
||||
huf_table: *mut usize,
|
||||
repeat: *mut c_int,
|
||||
flags: c_int,
|
||||
) -> usize {
|
||||
huf_compress_internal(
|
||||
dst.cast(),
|
||||
dst_size,
|
||||
src.cast(),
|
||||
src_size,
|
||||
max_symbol_value,
|
||||
huff_log,
|
||||
false,
|
||||
workspace,
|
||||
workspace_size,
|
||||
huf_table,
|
||||
repeat,
|
||||
flags,
|
||||
)
|
||||
}
|
||||
|
||||
#[no_mangle]
|
||||
pub unsafe extern "C" fn HUF_compress4X_repeat(
|
||||
dst: *mut c_void,
|
||||
dst_size: usize,
|
||||
src: *const c_void,
|
||||
src_size: usize,
|
||||
max_symbol_value: c_uint,
|
||||
huff_log: c_uint,
|
||||
workspace: *mut c_void,
|
||||
workspace_size: usize,
|
||||
huf_table: *mut usize,
|
||||
repeat: *mut c_int,
|
||||
flags: c_int,
|
||||
) -> usize {
|
||||
huf_compress_internal(
|
||||
dst.cast(),
|
||||
dst_size,
|
||||
src.cast(),
|
||||
src_size,
|
||||
max_symbol_value,
|
||||
huff_log,
|
||||
true,
|
||||
workspace,
|
||||
workspace_size,
|
||||
huf_table,
|
||||
repeat,
|
||||
flags,
|
||||
)
|
||||
}
|
||||
|
||||
#[cfg(all(test, feature = "decompression"))]
|
||||
mod tests {
|
||||
use super::*;
|
||||
use crate::huf_decompress::{
|
||||
HUF_decompress1X_usingDTable, HUF_decompress4X_usingDTable, HUF_readDTableX1_wksp,
|
||||
};
|
||||
|
||||
const WORKSPACE_SIZE: usize = HUF_WORKSPACE_SIZE;
|
||||
|
||||
fn source() -> Vec<u8> {
|
||||
let mut source = Vec::with_capacity(4096);
|
||||
for index in 0..4096 {
|
||||
source.push(match index % 17 {
|
||||
0..=8 => b'a',
|
||||
9..=12 => b'b',
|
||||
13..=14 => b'c',
|
||||
15 => b'd',
|
||||
_ => b'e',
|
||||
});
|
||||
}
|
||||
source
|
||||
}
|
||||
|
||||
unsafe fn make_table(
|
||||
source: &[u8],
|
||||
table: &mut [usize; HUF_CTABLE_SIZE_ST],
|
||||
workspace: &mut [u64; WORKSPACE_SIZE / size_of::<u64>()],
|
||||
) -> u32 {
|
||||
let mut count = [0u32; HUF_SYMBOLVALUE_MAX as usize + 1];
|
||||
for byte in source {
|
||||
count[*byte as usize] += 1;
|
||||
}
|
||||
let mut max_symbol = HUF_SYMBOLVALUE_MAX;
|
||||
while count[max_symbol as usize] == 0 {
|
||||
max_symbol -= 1;
|
||||
}
|
||||
let bits = HUF_buildCTable_wksp(
|
||||
table.as_mut_ptr(),
|
||||
count.as_ptr(),
|
||||
max_symbol,
|
||||
0,
|
||||
workspace.as_mut_ptr().cast::<c_void>(),
|
||||
std::mem::size_of_val(workspace),
|
||||
);
|
||||
assert!(!ERR_isError(bits));
|
||||
bits as u32
|
||||
}
|
||||
|
||||
unsafe fn dtable_from_header(
|
||||
header: &[u8],
|
||||
workspace: &mut [u64; WORKSPACE_SIZE / size_of::<u64>()],
|
||||
) -> [u32; 1 + (1 << 11)] {
|
||||
let mut dtable = [0u32; 1 + (1 << 11)];
|
||||
dtable[0] = 11 * 0x0100_0001;
|
||||
let read = HUF_readDTableX1_wksp(
|
||||
dtable.as_mut_ptr(),
|
||||
header.as_ptr().cast::<c_void>(),
|
||||
header.len(),
|
||||
workspace.as_mut_ptr().cast::<c_void>(),
|
||||
std::mem::size_of_val(workspace),
|
||||
0,
|
||||
);
|
||||
assert!(!ERR_isError(read));
|
||||
dtable
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn tables_and_single_stream_round_trip() {
|
||||
let source = source();
|
||||
let mut workspace = [0u64; WORKSPACE_SIZE / size_of::<u64>()];
|
||||
let mut table = [0usize; HUF_CTABLE_SIZE_ST];
|
||||
let bits = unsafe { make_table(&source, &mut table, &mut workspace) };
|
||||
assert!((1..=HUF_TABLELOG_MAX).contains(&bits));
|
||||
|
||||
let max_symbol = unsafe { HUF_readCTableHeader(table.as_ptr()).maxSymbolValue as u32 };
|
||||
let mut header = [0u8; 512];
|
||||
let header_size = unsafe {
|
||||
HUF_writeCTable_wksp(
|
||||
header.as_mut_ptr().cast::<c_void>(),
|
||||
header.len(),
|
||||
table.as_ptr(),
|
||||
max_symbol,
|
||||
bits,
|
||||
workspace.as_mut_ptr().cast::<c_void>(),
|
||||
std::mem::size_of_val(&workspace),
|
||||
)
|
||||
};
|
||||
assert!(!ERR_isError(header_size));
|
||||
let dtable = unsafe { dtable_from_header(&header[..header_size], &mut workspace) };
|
||||
|
||||
let mut compressed = vec![0u8; source.len() + 64];
|
||||
let c_size = unsafe {
|
||||
HUF_compress1X_usingCTable(
|
||||
compressed.as_mut_ptr().cast::<c_void>(),
|
||||
compressed.len(),
|
||||
source.as_ptr().cast::<c_void>(),
|
||||
source.len(),
|
||||
table.as_ptr(),
|
||||
0,
|
||||
)
|
||||
};
|
||||
assert!(c_size > 0);
|
||||
let mut decoded = vec![0u8; source.len()];
|
||||
let d_size = unsafe {
|
||||
HUF_decompress1X_usingDTable(
|
||||
decoded.as_mut_ptr().cast::<c_void>(),
|
||||
decoded.len(),
|
||||
compressed.as_ptr().cast::<c_void>(),
|
||||
c_size,
|
||||
dtable.as_ptr(),
|
||||
0,
|
||||
)
|
||||
};
|
||||
assert_eq!(d_size, source.len());
|
||||
assert_eq!(decoded, source);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn four_stream_and_repeat_round_trip() {
|
||||
let source = source();
|
||||
let mut workspace = [0u64; WORKSPACE_SIZE / size_of::<u64>()];
|
||||
let mut old_table = [0usize; HUF_CTABLE_SIZE_ST];
|
||||
let mut repeat = HUF_REPEAT_NONE;
|
||||
let mut compressed = vec![0u8; source.len() + 256];
|
||||
let c_size = unsafe {
|
||||
HUF_compress4X_repeat(
|
||||
compressed.as_mut_ptr().cast::<c_void>(),
|
||||
compressed.len(),
|
||||
source.as_ptr().cast::<c_void>(),
|
||||
source.len(),
|
||||
HUF_SYMBOLVALUE_MAX,
|
||||
HUF_TABLELOG_DEFAULT,
|
||||
workspace.as_mut_ptr().cast::<c_void>(),
|
||||
std::mem::size_of_val(&workspace),
|
||||
old_table.as_mut_ptr(),
|
||||
&mut repeat,
|
||||
HUF_FLAGS_OPTIMAL_DEPTH,
|
||||
)
|
||||
};
|
||||
assert!(c_size > 0);
|
||||
let mut dtable = [0u32; 1 + (1 << 11)];
|
||||
dtable[0] = 11 * 0x0100_0001;
|
||||
let header_size = unsafe {
|
||||
HUF_readDTableX1_wksp(
|
||||
dtable.as_mut_ptr(),
|
||||
compressed.as_ptr().cast::<c_void>(),
|
||||
c_size,
|
||||
workspace.as_mut_ptr().cast::<c_void>(),
|
||||
std::mem::size_of_val(&workspace),
|
||||
0,
|
||||
)
|
||||
};
|
||||
assert!(!ERR_isError(header_size));
|
||||
let mut decoded = vec![0u8; source.len()];
|
||||
let d_size = unsafe {
|
||||
HUF_decompress4X_usingDTable(
|
||||
decoded.as_mut_ptr().cast::<c_void>(),
|
||||
decoded.len(),
|
||||
compressed.as_ptr().add(header_size).cast::<c_void>(),
|
||||
c_size - header_size,
|
||||
dtable.as_ptr(),
|
||||
0,
|
||||
)
|
||||
};
|
||||
assert_eq!(d_size, source.len());
|
||||
assert_eq!(decoded, source);
|
||||
}
|
||||
}
|
||||
@@ -12,6 +12,8 @@ pub mod fse_compress;
|
||||
pub mod fse_decompress;
|
||||
#[cfg(feature = "compression")]
|
||||
pub mod hist;
|
||||
#[cfg(feature = "compression")]
|
||||
pub mod huf_compress;
|
||||
#[cfg(feature = "decompression")]
|
||||
pub mod huf_decompress;
|
||||
pub mod mem;
|
||||
|
||||
Reference in New Issue
Block a user