AuroraMiddleware/zstd
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request #982 from facebook/fix304

Browse Source 
Fix for #304 and #977 : error during dictionary creation
This commit is contained in:
Yann Collet 2018-01-11 13:20:59 -08:00 committed by GitHub
commit 04c00f9388
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
7 changed files with 155 additions and 106 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
lib/common/huf.h
View File

@ -206,10 +206,10 @@ The following API allows targeting specific sub-functions for advanced tasks.
For example, it's possible to compress several blocks using the same 'CTable',
or to save and regenerate 'CTable' using external methods.
*/
/* FSE_count() : find it within "fse.h" */
/* FSE_count() : exposed within "fse.h" */
unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
typedef struct HUF_CElt_s HUF_CElt; /* incomplete type */
size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits);
size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits); /* @return : maxNbBits; CTable and count can overlap, in which case, CTable will overwrite count content */
size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);

1
lib/compress/huf_compress.c
View File

@ -405,6 +405,7 @@ size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValu
}
/** HUF_buildCTable() :
* @return : maxNbBits
* Note : count is used before tree is written, so they can safely overlap
*/
size_t HUF_buildCTable (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits)

2
lib/compress/zstd_compress.c
View File

@ -2869,7 +2869,7 @@ size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
if (params.nbThreads > 1) {
if (cctx->mtctx == NULL || (params.nbThreads != ZSTDMT_getNbThreads(cctx->mtctx))) {
DEBUGLOG(4, "ZSTD_compress_generic: creating new mtctx for nbThreads=%u (previous: %u)",
params.nbThreads, ZSTDMT_getNbThreads(cctx->mtctx));
params.nbThreads, (U32)ZSTDMT_getNbThreads(cctx->mtctx));
ZSTDMT_freeCCtx(cctx->mtctx);
cctx->mtctx = ZSTDMT_createCCtx_advanced(params.nbThreads, cctx->customMem);
if (cctx->mtctx == NULL) return ERROR(memory_allocation);

19
lib/dictBuilder/cover.c
View File

@ -537,8 +537,8 @@ static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
/* Checks */
if (totalSamplesSize < MAX(d, sizeof(U64)) ||
totalSamplesSize >= (size_t)COVER_MAX_SAMPLES_SIZE) {
DISPLAYLEVEL(1, "Total samples size is too large, maximum size is %u MB\n",
(COVER_MAX_SAMPLES_SIZE >> 20));
DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
(U32)(totalSamplesSize>>20), (COVER_MAX_SAMPLES_SIZE >> 20));
return 0;
}
/* Zero the context */
@ -651,12 +651,16 @@ static size_t COVER_buildDictionary(const COVER_ctx_t *ctx, U32 *freqs,
}
ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
const size_t *samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t parameters) {
BYTE *const dict = (BYTE *)dictBuffer;
void *dictBuffer, size_t dictBufferCapacity,
const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t parameters)
{
BYTE* const dict = (BYTE*)dictBuffer;
COVER_ctx_t ctx;
COVER_map_t activeDmers;
/* Initialize global data */
g_displayLevel = parameters.zParams.notificationLevel;
/* Checks */
if (!COVER_checkParameters(parameters, dictBufferCapacity)) {
DISPLAYLEVEL(1, "Cover parameters incorrect\n");
@ -671,8 +675,6 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
ZDICT_DICTSIZE_MIN);
return ERROR(dstSize_tooSmall);
}
/* Initialize global data */
g_displayLevel = parameters.zParams.notificationLevel;
/* Initialize context and activeDmers */
if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
parameters.d)) {
@ -947,6 +949,7 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
unsigned k;
COVER_best_t best;
POOL_ctx *pool = NULL;
/* Checks */
if (kMinK < kMaxD || kMaxK < kMinK) {
LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");

85
lib/dictBuilder/zdict.c
View File

@ -207,7 +207,6 @@ static dictItem ZDICT_analyzePos(
U32 cumulLength[LLIMIT] = {0};
U32 savings[LLIMIT] = {0};
const BYTE* b = (const BYTE*)buffer;
size_t length;
size_t maxLength = LLIMIT;
size_t pos = suffix[start];
U32 end = start;
@ -222,26 +221,30 @@ static dictItem ZDICT_analyzePos(
||(MEM_read16(b+pos+1) == MEM_read16(b+pos+3))
||(MEM_read16(b+pos+2) == MEM_read16(b+pos+4)) ) {
/* skip and mark segment */
U16 u16 = MEM_read16(b+pos+4);
U32 u, e = 6;
while (MEM_read16(b+pos+e) == u16) e+=2 ;
if (b[pos+e] == b[pos+e-1]) e++;
for (u=1; u<e; u++)
U16 const pattern16 = MEM_read16(b+pos+4);
U32 u, patternEnd = 6;
while (MEM_read16(b+pos+patternEnd) == pattern16) patternEnd+=2 ;
if (b[pos+patternEnd] == b[pos+patternEnd-1]) patternEnd++;
for (u=1; u<patternEnd; u++)
doneMarks[pos+u] = 1;
return solution;
}
/* look forward */
{ size_t length;
do {
end++;
length = ZDICT_count(b + pos, b + suffix[end]);
} while (length >=MINMATCHLENGTH);
} while (length >= MINMATCHLENGTH);
}
/* look backward */
{ size_t length;
do {
length = ZDICT_count(b + pos, b + *(suffix+start-1));
if (length >=MINMATCHLENGTH) start--;
} while(length >= MINMATCHLENGTH);
}
/* exit if not found a minimum nb of repetitions */
if (end-start < minRatio) {
@ -268,7 +271,7 @@ static dictItem ZDICT_analyzePos(
U32 selectedCount = 0;
U32 selectedID = currentID;
for (id =refinedStart; id < refinedEnd; id++) {
if (b[ suffix[id] + searchLength] != currentChar) {
if (b[suffix[id] + searchLength] != currentChar) {
if (currentCount > selectedCount) {
selectedCount = currentCount;
selectedID = currentID;
@ -297,21 +300,24 @@ static dictItem ZDICT_analyzePos(
memset(lengthList, 0, sizeof(lengthList));
/* look forward */
{ size_t length;
do {
end++;
length = ZDICT_count(b + pos, b + suffix[end]);
if (length >= LLIMIT) length = LLIMIT-1;
lengthList[length]++;
} while (length >=MINMATCHLENGTH);
}
/* look backward */
length = MINMATCHLENGTH;
{ size_t length = MINMATCHLENGTH;
while ((length >= MINMATCHLENGTH) & (start > 0)) {
length = ZDICT_count(b + pos, b + suffix[start - 1]);
if (length >= LLIMIT) length = LLIMIT - 1;
lengthList[length]++;
if (length >= MINMATCHLENGTH) start--;
}
}
/* largest useful length */
memset(cumulLength, 0, sizeof(cumulLength));
@ -345,12 +351,12 @@ static dictItem ZDICT_analyzePos(
/* mark positions done */
{ U32 id;
for (id=start; id<end; id++) {
U32 p, pEnd;
U32 p, pEnd, length;
U32 const testedPos = suffix[id];
if (testedPos == pos)
length = solution.length;
else {
length = ZDICT_count(b+pos, b+testedPos);
length = (U32)ZDICT_count(b+pos, b+testedPos);
if (length > solution.length) length = solution.length;
}
pEnd = (U32)(testedPos + length);
@ -575,8 +581,8 @@ static void ZDICT_fillNoise(void* buffer, size_t length)
typedef struct
{
ZSTD_CCtx* ref;
ZSTD_CCtx* zc;
ZSTD_CCtx* ref; /* contains reference to dictionary */
ZSTD_CCtx* zc; /* working context */
void* workPlace; /* must be ZSTD_BLOCKSIZE_MAX allocated */
} EStats_ress_t;
@ -584,7 +590,8 @@ typedef struct
static void ZDICT_countEStats(EStats_ress_t esr, ZSTD_parameters params,
U32* countLit, U32* offsetcodeCount, U32* matchlengthCount, U32* litlengthCount, U32* repOffsets,
const void* src, size_t srcSize, U32 notificationLevel)
const void* src, size_t srcSize,
U32 notificationLevel)
{
size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_MAX, 1 << params.cParams.windowLog);
size_t cSize;
@ -597,7 +604,7 @@ static void ZDICT_countEStats(EStats_ress_t esr, ZSTD_parameters params,
if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (U32)srcSize); return; }
if (cSize) { /* if == 0; block is not compressible */
const seqStore_t* seqStorePtr = ZSTD_getSeqStore(esr.zc);
const seqStore_t* const seqStorePtr = ZSTD_getSeqStore(esr.zc);
/* literals stats */
{ const BYTE* bytePtr;
@ -659,6 +666,18 @@ static void ZDICT_insertSortCount(offsetCount_t table[ZSTD_REP_NUM+1], U32 val,
}
}
/* ZDICT_flatLit() :
* rewrite `countLit` to contain a mostly flat but still compressible distribution of literals.
* necessary to avoid generating a non-compressible distribution that HUF_writeCTable() cannot encode.
*/
static void ZDICT_flatLit(U32* countLit)
{
int u;
for (u=1; u<256; u++) countLit[u] = 2;
countLit[0] = 4;
countLit[253] = 1;
countLit[254] = 1;
}
#define OFFCODE_MAX 30 /* only applicable to first block */
static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
@ -688,6 +707,7 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
BYTE* dstPtr = (BYTE*)dstBuffer;
/* init */
DEBUGLOG(4, "ZDICT_analyzeEntropy");
esr.ref = ZSTD_createCCtx();
esr.zc = ZSTD_createCCtx();
esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
@ -713,7 +733,7 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
goto _cleanup;
} }
/* collect stats on all files */
/* collect stats on all samples */
for (u=0; u<nbFiles; u++) {
ZDICT_countEStats(esr, params,
countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
@ -722,14 +742,21 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
pos += fileSizes[u];
}
/* analyze */
errorCode = HUF_buildCTable (hufTable, countLit, 255, huffLog);
if (HUF_isError(errorCode)) {
/* analyze, build stats, starting with literals */
{ size_t maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
if (HUF_isError(maxNbBits)) {
eSize = ERROR(GENERIC);
DISPLAYLEVEL(1, "HUF_buildCTable error \n");
DISPLAYLEVEL(1, " HUF_buildCTable error \n");
goto _cleanup;
}
huffLog = (U32)errorCode;
if (maxNbBits==8) { /* not compressible : will fail on HUF_writeCTable() */
DISPLAYLEVEL(2, "warning : pathological dataset : literals are not compressible : samples are noisy or too regular \n");
ZDICT_flatLit(countLit); /* replace distribution by a fake "mostly flat but still compressible" distribution, that HUF_writeCTable() can encode */
maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
assert(maxNbBits==9);
}
huffLog = (U32)maxNbBits;
}
/* looking for most common first offsets */
{ U32 offset;
@ -850,6 +877,7 @@ size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
U32 const notificationLevel = params.notificationLevel;
/* check conditions */
DEBUGLOG(4, "ZDICT_finalizeDictionary");
if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
if (dictContentSize < ZDICT_CONTENTSIZE_MIN) return ERROR(srcSize_wrong);
if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
@ -1025,8 +1053,9 @@ size_t ZDICT_trainFromBuffer_unsafe_legacy(
}
/* issue : samplesBuffer need to be followed by a noisy guard band.
* work around : duplicate the buffer, and add the noise */
/* ZDICT_trainFromBuffer_legacy() :
* issue : samplesBuffer need to be followed by a noisy guard band.
* work around : duplicate the buffer, and add the noise */
size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_legacy_params_t params)
@ -1054,14 +1083,18 @@ size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
{
ZDICT_cover_params_t params;
DEBUGLOG(3, "ZDICT_trainFromBuffer");
memset(&params, 0, sizeof(params));
params.d = 8;
params.steps = 4;
/* Default to level 6 since no compression level information is avaialble */
/* Default to level 6 since no compression level information is available */
params.zParams.compressionLevel = 6;
#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
params.zParams.notificationLevel = ZSTD_DEBUG;
#endif
return ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, dictBufferCapacity,
samplesBuffer, samplesSizes,
nbSamples, &params);
samplesBuffer, samplesSizes, nbSamples,
&params);
}
size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,

41
lib/dictBuilder/zdict.h
View File

@ -39,7 +39,7 @@ extern "C" {
/*! ZDICT_trainFromBuffer():
* Train a dictionary from an array of samples.
* Uses ZDICT_optimizeTrainFromBuffer_cover() single-threaded, with d=8 and steps=4.
* Redirect towards ZDICT_optimizeTrainFromBuffer_cover() single-threaded, with d=8 and steps=4.
* Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
* supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
* The resulting dictionary will be saved into `dictBuffer`.
@ -47,8 +47,8 @@ extern "C" {
* or an error code, which can be tested with ZDICT_isError().
* Note: ZDICT_trainFromBuffer() requires about 9 bytes of memory for each input byte.
* Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
* It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, but this can vary a lot.
* It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, though this can vary a lot.
* It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
*/
ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
@ -72,14 +72,14 @@ ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
* ==================================================================================== */
typedef struct {
int compressionLevel; /* 0 means default; target a specific zstd compression level */
unsigned notificationLevel; /* Write to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
unsigned dictID; /* 0 means auto mode (32-bits random value); other : force dictID value */
int compressionLevel; /* optimize for a specific zstd compression level; 0 means default */
unsigned notificationLevel; /* Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
unsigned dictID; /* force dictID value; 0 means auto mode (32-bits random value) */
} ZDICT_params_t;
/*! ZDICT_cover_params_t:
* For all values 0 means default.
* k and d are the only required parameters.
* For others, value 0 means default.
*/
typedef struct {
unsigned k; /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
@ -99,20 +99,20 @@ typedef struct {
* or an error code, which can be tested with ZDICT_isError().
* Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte.
* Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
* It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, but this can vary a lot.
* It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, though this can vary a lot.
* It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
*/
ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
const size_t *samplesSizes, unsigned nbSamples,
void *dictBuffer, size_t dictBufferCapacity,
const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t parameters);
/*! ZDICT_optimizeTrainFromBuffer_cover():
* The same requirements as above hold for all the parameters except `parameters`.
* This function tries many parameter combinations and picks the best parameters.
* `*parameters` is filled with the best parameters found, and the dictionary
* constructed with those parameters is stored in `dictBuffer`.
* `*parameters` is filled with the best parameters found,
* dictionary constructed with those parameters is stored in `dictBuffer`.
*
* All of the parameters d, k, steps are optional.
* If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
@ -125,9 +125,9 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
* Note: ZDICT_optimizeTrainFromBuffer_cover() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
*/
ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
const size_t *samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t *parameters);
void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t* parameters);
/*! ZDICT_finalizeDictionary():
* Given a custom content as a basis for dictionary, and a set of samples,
@ -165,14 +165,15 @@ typedef struct {
* @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
* or an error code, which can be tested with ZDICT_isError().
* Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
* It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, but this can vary a lot.
* It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, though this can vary a lot.
* It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
* Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
*/
ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
const size_t *samplesSizes, unsigned nbSamples, ZDICT_legacy_params_t parameters);
void *dictBuffer, size_t dictBufferCapacity,
const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
ZDICT_legacy_params_t parameters);
/* Deprecation warnings */
/* It is generally possible to disable deprecation warnings from compiler,

17
tests/fuzzer.c
View File

@ -666,12 +666,13 @@ static int basicUnitTests(U32 seed, double compressibility)
/* Dictionary and dictBuilder tests */
{ ZSTD_CCtx* const cctx = ZSTD_createCCtx();
size_t dictSize = 16 KB;
void* dictBuffer = malloc(dictSize);
size_t const dictBufferCapacity = 16 KB;
void* dictBuffer = malloc(dictBufferCapacity);
size_t const totalSampleSize = 1 MB;
size_t const sampleUnitSize = 8 KB;
U32 const nbSamples = (U32)(totalSampleSize / sampleUnitSize);
size_t* const samplesSizes = (size_t*) malloc(nbSamples * sizeof(size_t));
size_t dictSize;
U32 dictID;
if (dictBuffer==NULL || samplesSizes==NULL) {
@ -680,9 +681,19 @@ static int basicUnitTests(U32 seed, double compressibility)
goto _output_error;
}
DISPLAYLEVEL(4, "test%3i : dictBuilder on cyclic data : ", testNb++);
assert(compressedBufferSize >= totalSampleSize);
{ U32 u; for (u=0; u<totalSampleSize; u++) ((BYTE*)decodedBuffer)[u] = (BYTE)u; }
{ U32 u; for (u=0; u<nbSamples; u++) samplesSizes[u] = sampleUnitSize; }
{ size_t const sDictSize = ZDICT_trainFromBuffer(dictBuffer, dictBufferCapacity,
decodedBuffer, samplesSizes, nbSamples);
if (ZDICT_isError(sDictSize)) goto _output_error;
DISPLAYLEVEL(4, "OK, created dictionary of size %u \n", (U32)sDictSize);
}
DISPLAYLEVEL(4, "test%3i : dictBuilder : ", testNb++);
{ U32 u; for (u=0; u<nbSamples; u++) samplesSizes[u] = sampleUnitSize; }
dictSize = ZDICT_trainFromBuffer(dictBuffer, dictSize,
dictSize = ZDICT_trainFromBuffer(dictBuffer, dictBufferCapacity,
CNBuffer, samplesSizes, nbSamples);
if (ZDICT_isError(dictSize)) goto _output_error;
DISPLAYLEVEL(4, "OK, created dictionary of size %u \n", (U32)dictSize);