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

Merge pull request #2446 from senhuang42/multiple_ddicts_v3

Browse Source 
[RFC] Support references to multiple DDicts
This commit is contained in:
sen
2021-01-08 16:49:45 -05:00
committed by GitHub
parent 33b73db33c 9ae0dd9336
commit 69085db61c
5 changed files with 334 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lib/decompress/zstd_decompress.c
+202 -1
View File
@@ -62,6 +62,7 @@
#include "../common/fse.h"
#define HUF_STATIC_LINKING_ONLY
#include "../common/huf.h"
#include "../common/xxhash.h" /* XXH64_reset, XXH64_update, XXH64_digest, XXH64 */
#include "../common/zstd_internal.h" /* blockProperties_t */
#include "zstd_decompress_internal.h" /* ZSTD_DCtx */
#include "zstd_ddict.h" /* ZSTD_DDictDictContent */
@@ -72,6 +73,144 @@
#endif
/*************************************
* Multiple DDicts Hashset internals *
*************************************/
#define DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT 4
#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3 /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
* Currently, that means a 0.75 load factor.
* So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
* the load factor of the ddict hash set.
*/
#define DDICT_HASHSET_TABLE_BASE_SIZE 64
#define DDICT_HASHSET_RESIZE_FACTOR 2
/* Hash function to determine starting position of dict insertion within the table
* Returns an index between [0, hashSet->ddictPtrTableSize]
*/
static size_t ZSTD_DDictHashSet_getIndex(const ZSTD_DDictHashSet* hashSet, U32 dictID) {
const U64 hash = XXH64(&dictID, sizeof(U32), 0);
/* DDict ptr table size is a multiple of 2, use size - 1 as mask to get index within [0, hashSet->ddictPtrTableSize) */
return hash & (hashSet->ddictPtrTableSize - 1);
}
/* Adds DDict to a hashset without resizing it.
* If inserting a DDict with a dictID that already exists in the set, replaces the one in the set.
* Returns 0 if successful, or a zstd error code if something went wrong.
*/
static size_t ZSTD_DDictHashSet_emplaceDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict) {
const U32 dictID = ZSTD_getDictID_fromDDict(ddict);
size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
RETURN_ERROR_IF(hashSet->ddictPtrCount == hashSet->ddictPtrTableSize, GENERIC, "Hash set is full!");
DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
while (hashSet->ddictPtrTable[idx] != NULL) {
/* Replace existing ddict if inserting ddict with same dictID */
if (ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]) == dictID) {
DEBUGLOG(4, "DictID already exists, replacing rather than adding");
hashSet->ddictPtrTable[idx] = ddict;
return 0;
}
idx &= idxRangeMask;
idx++;
}
DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
hashSet->ddictPtrTable[idx] = ddict;
hashSet->ddictPtrCount++;
return 0;
}
/* Expands hash table by factor of DDICT_HASHSET_RESIZE_FACTOR and
* rehashes all values, allocates new table, frees old table.
* Returns 0 on success, otherwise a zstd error code.
*/
static size_t ZSTD_DDictHashSet_expand(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
size_t newTableSize = hashSet->ddictPtrTableSize * DDICT_HASHSET_RESIZE_FACTOR;
const ZSTD_DDict** newTable = (const ZSTD_DDict**)ZSTD_customCalloc(sizeof(ZSTD_DDict*) * newTableSize, customMem);
const ZSTD_DDict** oldTable = hashSet->ddictPtrTable;
size_t oldTableSize = hashSet->ddictPtrTableSize;
size_t i;
DEBUGLOG(4, "Expanding DDict hash table! Old size: %zu new size: %zu", oldTableSize, newTableSize);
RETURN_ERROR_IF(!newTable, memory_allocation, "Expanded hashset allocation failed!");
hashSet->ddictPtrTable = newTable;
hashSet->ddictPtrTableSize = newTableSize;
hashSet->ddictPtrCount = 0;
for (i = 0; i < oldTableSize; ++i) {
if (oldTable[i] != NULL) {
FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, oldTable[i]), "");
}
}
ZSTD_customFree((void*)oldTable, customMem);
DEBUGLOG(4, "Finished re-hash");
return 0;
}
/* Fetches a DDict with the given dictID
* Returns the ZSTD_DDict* with the requested dictID. If it doesn't exist, then returns NULL.
*/
static const ZSTD_DDict* ZSTD_DDictHashSet_getDDict(ZSTD_DDictHashSet* hashSet, U32 dictID) {
size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
for (;;) {
size_t currDictID = ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]);
if (currDictID == dictID || currDictID == 0) {
/* currDictID == 0 implies a NULL ddict entry */
break;
} else {
idx &= idxRangeMask; /* Goes to start of table when we reach the end */
idx++;
}
}
DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
return hashSet->ddictPtrTable[idx];
}
/* Allocates space for and returns a ddict hash set
* The hash set's ZSTD_DDict* table has all values automatically set to NULL to begin with.
* Returns NULL if allocation failed.
*/
static ZSTD_DDictHashSet* ZSTD_createDDictHashSet(ZSTD_customMem customMem) {
ZSTD_DDictHashSet* ret = (ZSTD_DDictHashSet*)ZSTD_customMalloc(sizeof(ZSTD_DDictHashSet), customMem);
DEBUGLOG(4, "Allocating new hash set");
ret->ddictPtrTable = (const ZSTD_DDict**)ZSTD_customCalloc(DDICT_HASHSET_TABLE_BASE_SIZE * sizeof(ZSTD_DDict*), customMem);
ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
ret->ddictPtrCount = 0;
if (!ret || !ret->ddictPtrTable) {
return NULL;
}
return ret;
}
/* Frees the table of ZSTD_DDict* within a hashset, then frees the hashset itself.
* Note: The ZSTD_DDict* within the table are NOT freed.
*/
static void ZSTD_freeDDictHashSet(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
DEBUGLOG(4, "Freeing ddict hash set");
if (hashSet && hashSet->ddictPtrTable) {
ZSTD_customFree((void*)hashSet->ddictPtrTable, customMem);
}
if (hashSet) {
ZSTD_customFree(hashSet, customMem);
}
}
/* Public function: Adds a DDict into the ZSTD_DDictHashSet, possibly triggering a resize of the hash set.
* Returns 0 on success, or a ZSTD error.
*/
static size_t ZSTD_DDictHashSet_addDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict, ZSTD_customMem customMem) {
DEBUGLOG(4, "Adding dict ID: %u to hashset with - Count: %zu Tablesize: %zu", ZSTD_getDictID_fromDDict(ddict), hashSet->ddictPtrCount, hashSet->ddictPtrTableSize);
if (hashSet->ddictPtrCount * DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT / hashSet->ddictPtrTableSize * DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT != 0) {
FORWARD_IF_ERROR(ZSTD_DDictHashSet_expand(hashSet, customMem), "");
}
FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, ddict), "");
return 0;
}
/*-*************************************************************
* Context management
***************************************************************/
@@ -101,6 +240,7 @@ static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
dctx->outBufferMode = ZSTD_bm_buffered;
dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
dctx->refMultipleDDicts = ZSTD_rmd_refSingleDDict;
}
static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
@@ -120,8 +260,8 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
dctx->noForwardProgress = 0;
dctx->oversizedDuration = 0;
dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
dctx->ddictSet = NULL;
ZSTD_DCtx_resetParameters(dctx);
dctx->validateChecksum = 1;
#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
dctx->dictContentEndForFuzzing = NULL;
#endif
@@ -178,6 +318,10 @@ size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
if (dctx->legacyContext)
ZSTD_freeLegacyStreamContext(dctx->legacyContext, dctx->previousLegacyVersion);
#endif
if (dctx->ddictSet) {
ZSTD_freeDDictHashSet(dctx->ddictSet, cMem);
dctx->ddictSet = NULL;
}
ZSTD_customFree(dctx, cMem);
return 0;
}
@@ -190,6 +334,29 @@ void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
ZSTD_memcpy(dstDCtx, srcDCtx, toCopy); /* no need to copy workspace */
}
/* Given a dctx with a digested frame params, re-selects the correct ZSTD_DDict based on
* the requested dict ID from the frame. If there exists a reference to the correct ZSTD_DDict, then
* accordingly sets the ddict to be used to decompress the frame.
*
* If no DDict is found, then no action is taken, and the ZSTD_DCtx::ddict remains as-is.
*
* ZSTD_d_refMultipleDDicts must be enabled for this function to be called.
*/
static void ZSTD_DCtx_selectFrameDDict(ZSTD_DCtx* dctx) {
assert(dctx->refMultipleDDicts && dctx->ddictSet);
DEBUGLOG(4, "Adjusting DDict based on requested dict ID from frame");
if (dctx->ddict) {
const ZSTD_DDict* frameDDict = ZSTD_DDictHashSet_getDDict(dctx->ddictSet, dctx->fParams.dictID);
if (frameDDict) {
DEBUGLOG(4, "DDict found!");
ZSTD_clearDict(dctx);
dctx->dictID = dctx->fParams.dictID;
dctx->ddict = frameDDict;
dctx->dictUses = ZSTD_use_indefinitely;
}
}
}
/*-*************************************************************
* Frame header decoding
@@ -441,12 +608,19 @@ unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
/** ZSTD_decodeFrameHeader() :
* `headerSize` must be the size provided by ZSTD_frameHeaderSize().
* If multiple DDict references are enabled, also will choose the correct DDict to use.
* @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
{
size_t const result = ZSTD_getFrameHeader_advanced(&(dctx->fParams), src, headerSize, dctx->format);
if (ZSTD_isError(result)) return result; /* invalid header */
RETURN_ERROR_IF(result>0, srcSize_wrong, "headerSize too small");
/* Reference DDict requested by frame if dctx references multiple ddicts */
if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts && dctx->ddictSet) {
ZSTD_DCtx_selectFrameDDict(dctx);
}
#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
/* Skip the dictID check in fuzzing mode, because it makes the search
* harder.
@@ -1391,6 +1565,16 @@ size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
if (ddict) {
dctx->ddict = ddict;
dctx->dictUses = ZSTD_use_indefinitely;
if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts) {
if (dctx->ddictSet == NULL) {
dctx->ddictSet = ZSTD_createDDictHashSet(dctx->customMem);
if (!dctx->ddictSet) {
RETURN_ERROR(memory_allocation, "Failed to allocate memory for hash set!");
}
}
assert(!dctx->staticSize); /* Impossible: ddictSet cannot have been allocated if static dctx */
FORWARD_IF_ERROR(ZSTD_DDictHashSet_addDDict(dctx->ddictSet, ddict, dctx->customMem), "");
}
}
return 0;
}
@@ -1436,6 +1620,10 @@ ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
bounds.lowerBound = (int)ZSTD_d_validateChecksum;
bounds.upperBound = (int)ZSTD_d_ignoreChecksum;
return bounds;
case ZSTD_d_refMultipleDDicts:
bounds.lowerBound = (int)ZSTD_rmd_refSingleDDict;
bounds.upperBound = (int)ZSTD_rmd_refMultipleDDicts;
return bounds;
default:;
}
bounds.error = ERROR(parameter_unsupported);
@@ -1473,6 +1661,9 @@ size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value
case ZSTD_d_forceIgnoreChecksum:
*value = (int)dctx->forceIgnoreChecksum;
return 0;
case ZSTD_d_refMultipleDDicts:
*value = (int)dctx->refMultipleDDicts;
return 0;
default:;
}
RETURN_ERROR(parameter_unsupported, "");
@@ -1499,6 +1690,13 @@ size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value
CHECK_DBOUNDS(ZSTD_d_forceIgnoreChecksum, value);
dctx->forceIgnoreChecksum = (ZSTD_forceIgnoreChecksum_e)value;
return 0;
case ZSTD_d_refMultipleDDicts:
CHECK_DBOUNDS(ZSTD_d_refMultipleDDicts, value);
if (dctx->staticSize != 0) {
RETURN_ERROR(parameter_unsupported, "Static dctx does not support multiple DDicts!");
}
dctx->refMultipleDDicts = (ZSTD_refMultipleDDicts_e)value;
return 0;
default:;
}
RETURN_ERROR(parameter_unsupported, "");
@@ -1680,6 +1878,9 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
} }
#endif
{ size_t const hSize = ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
if (zds->refMultipleDDicts && zds->ddictSet) {
ZSTD_DCtx_selectFrameDDict(zds);
}
DEBUGLOG(5, "header size : %u", (U32)hSize);
if (ZSTD_isError(hSize)) {
#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
lib/decompress/zstd_decompress_internal.h
+9
View File
@@ -99,6 +99,13 @@ typedef enum {
ZSTD_use_once = 1 /* Use the dictionary once and set to ZSTD_dont_use */
} ZSTD_dictUses_e;
/* Hashset for storing references to multiple ZSTD_DDict within ZSTD_DCtx */
typedef struct {
const ZSTD_DDict** ddictPtrTable;
size_t ddictPtrTableSize;
size_t ddictPtrCount;
} ZSTD_DDictHashSet;
struct ZSTD_DCtx_s
{
const ZSTD_seqSymbol* LLTptr;
@@ -136,6 +143,8 @@ struct ZSTD_DCtx_s
U32 dictID;
int ddictIsCold; /* if == 1 : dictionary is "new" for working context, and presumed "cold" (not in cpu cache) */
ZSTD_dictUses_e dictUses;
ZSTD_DDictHashSet* ddictSet; /* Hash set for multiple ddicts */
ZSTD_refMultipleDDicts_e refMultipleDDicts; /* User specified: if == 1, will allow references to multiple DDicts. Default == 0 (disabled) */
/* streaming */
ZSTD_dStreamStage streamStage;
lib/zstd.h
+40 -1
View File
@@ -546,12 +546,14 @@ typedef enum {
* ZSTD_d_format
* ZSTD_d_stableOutBuffer
* ZSTD_d_forceIgnoreChecksum
* ZSTD_d_refMultipleDDicts
* Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
* note : never ever use experimentalParam? names directly
*/
ZSTD_d_experimentalParam1=1000,
ZSTD_d_experimentalParam2=1001,
ZSTD_d_experimentalParam3=1002
ZSTD_d_experimentalParam3=1002,
ZSTD_d_experimentalParam4=1003
} ZSTD_dParameter;
@@ -999,6 +1001,13 @@ ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, s
/*! ZSTD_DCtx_refDDict() :
* Reference a prepared dictionary, to be used to decompress next frames.
* The dictionary remains active for decompression of future frames using same DCtx.
*
* If called with ZSTD_d_refMultipleDDicts enabled, repeated calls of this function
* will store the DDict references in a table, and the DDict used for decompression
* will be determined at decompression time, as per the dict ID in the frame.
* The memory for the table is allocated on the first call to refDDict, and can be
* freed with ZSTD_freeDCtx().
*
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
* Note 1 : Currently, only one dictionary can be managed.
* Referencing a new dictionary effectively "discards" any previous one.
@@ -1205,6 +1214,12 @@ typedef enum {
ZSTD_d_ignoreChecksum = 1
} ZSTD_forceIgnoreChecksum_e;
typedef enum {
/* Note: this enum controls ZSTD_d_refMultipleDDicts */
ZSTD_rmd_refSingleDDict = 0,
ZSTD_rmd_refMultipleDDicts = 1
} ZSTD_refMultipleDDicts_e;
typedef enum {
/* Note: this enum and the behavior it controls are effectively internal
* implementation details of the compressor. They are expected to continue
@@ -2003,6 +2018,30 @@ ZSTDLIB_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param
*/
#define ZSTD_d_forceIgnoreChecksum ZSTD_d_experimentalParam3
/* ZSTD_d_refMultipleDDicts
* Experimental parameter.
* Default is 0 == disabled. Set to 1 to enable
*
* If enabled and dctx is allocated on the heap, then additional memory will be allocated
* to store references to multiple ZSTD_DDict. That is, multiple calls of ZSTD_refDDict()
* using a given ZSTD_DCtx, rather than overwriting the previous DDict reference, will instead
* store all references. At decompression time, the appropriate dictID is selected
* from the set of DDicts based on the dictID in the frame.
*
* Usage is simply calling ZSTD_refDDict() on multiple dict buffers.
*
* Param has values of byte ZSTD_refMultipleDDicts_e
*
* WARNING: Enabling this parameter and calling ZSTD_DCtx_refDDict(), will trigger memory
* allocation for the hash table. ZSTD_freeDCtx() also frees this memory.
* Memory is allocated as per ZSTD_DCtx::customMem.
*
* Although this function allocates memory for the table, the user is still responsible for
* memory management of the underlying ZSTD_DDict* themselves.
*/
#define ZSTD_d_refMultipleDDicts ZSTD_d_experimentalParam4
/*! ZSTD_DCtx_setFormat() :
* Instruct the decoder context about what kind of data to decode next.
* This instruction is mandatory to decode data without a fully-formed header,