diff --git a/lib/libzstd/Makefile b/lib/libzstd/Makefile index dcf6cd3c0281..5f7cd8248d88 100644 --- a/lib/libzstd/Makefile +++ b/lib/libzstd/Makefile @@ -1,68 +1,68 @@ # $FreeBSD$ LIB= zstd SRCS= entropy_common.c \ error_private.c \ fse_decompress.c \ pool.c \ threading.c \ xxhash.c \ zstd_common.c \ fse_compress.c \ huf_compress.c \ zstd_compress.c \ zstd_compress_literals.c \ zstd_compress_sequences.c \ zstd_compress_superblock.c \ zstdmt_compress.c \ huf_decompress.c \ zstd_ddict.c \ zstd_decompress.c \ zstd_decompress_block.c \ zbuff_common.c \ zbuff_compress.c \ zbuff_decompress.c \ cover.c \ divsufsort.c \ zdict.c \ zstd_fast.c \ zstd_lazy.c \ zstd_ldm.c \ zstd_opt.c \ zstd_double_fast.c \ debug.c \ hist.c \ fastcover.c WARNS?= 2 INCS= zstd.h CFLAGS+= -I${ZSTDDIR}/lib -I${ZSTDDIR}/lib/common -DXXH_NAMESPACE=ZSTD_ \ - -DZSTD_MULTITHREAD=1 + -DZSTD_MULTITHREAD=1 -fvisibility=hidden LIBADD= pthread PRIVATELIB= yes PACKAGE= runtime ZSTDDIR= ${SRCTOP}/sys/contrib/zstd .PATH: ${ZSTDDIR}/lib/common ${ZSTDDIR}/lib/compress \ ${ZSTDDIR}/lib/decompress ${ZSTDDIR}/lib/deprecated \ ${ZSTDDIR}/lib/dictBuilder ${ZSTDDIR}/lib .include # https://github.com/facebook/zstd/commit/812e8f2a [zstd 1.4.1] # "Note that [GCC] autovectorization still does not do a good job on the # optimized version, so it's turned off via attribute and flag. I found # that neither attribute nor command-line flag were entirely successful in # turning off vectorization, which is why there were both." .if ${COMPILER_TYPE} == "gcc" CFLAGS.zstd_decompress_block.c+= -fno-tree-vectorize .endif # Work around lack of __bswap[sd]i2() intrinsics on riscv. Probably should be # added to one of the runtime C libraries instead. .if ${MACHINE_CPUARCH} == "riscv" .PATH: ${ZSTDDIR}/lib/freebsd SRCS+= zstd_kfreebsd.c .endif .include diff --git a/sys/contrib/zstd/lib/common/xxhash.h b/sys/contrib/zstd/lib/common/xxhash.h index eceb55d5e071..ff1fafe4cd22 100644 --- a/sys/contrib/zstd/lib/common/xxhash.h +++ b/sys/contrib/zstd/lib/common/xxhash.h @@ -1,285 +1,288 @@ /* * xxHash - Extremely Fast Hash algorithm * Header File * Copyright (c) 2012-2020, Yann Collet, Facebook, Inc. * * You can contact the author at : * - xxHash source repository : https://github.com/Cyan4973/xxHash * * 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. */ /* Notice extracted from xxHash homepage : xxHash is an extremely fast Hash algorithm, running at RAM speed limits. It also successfully passes all tests from the SMHasher suite. Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz) Name Speed Q.Score Author xxHash 5.4 GB/s 10 CrapWow 3.2 GB/s 2 Andrew MumurHash 3a 2.7 GB/s 10 Austin Appleby SpookyHash 2.0 GB/s 10 Bob Jenkins SBox 1.4 GB/s 9 Bret Mulvey Lookup3 1.2 GB/s 9 Bob Jenkins SuperFastHash 1.2 GB/s 1 Paul Hsieh CityHash64 1.05 GB/s 10 Pike & Alakuijala FNV 0.55 GB/s 5 Fowler, Noll, Vo CRC32 0.43 GB/s 9 MD5-32 0.33 GB/s 10 Ronald L. Rivest SHA1-32 0.28 GB/s 10 Q.Score is a measure of quality of the hash function. It depends on successfully passing SMHasher test set. 10 is a perfect score. A 64-bits version, named XXH64, is available since r35. It offers much better speed, but for 64-bits applications only. Name Speed on 64 bits Speed on 32 bits XXH64 13.8 GB/s 1.9 GB/s XXH32 6.8 GB/s 6.0 GB/s */ #if defined (__cplusplus) extern "C" { #endif #ifndef XXHASH_H_5627135585666179 #define XXHASH_H_5627135585666179 1 /* **************************** * Definitions ******************************/ #include "zstd_deps.h" typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode; /* **************************** * API modifier ******************************/ /** XXH_PRIVATE_API * This is useful if you want to include xxhash functions in `static` mode * in order to inline them, and remove their symbol from the public list. * Methodology : * #define XXH_PRIVATE_API * #include "xxhash.h" * `xxhash.c` is automatically included. * It's not useful to compile and link it as a separate module anymore. */ #ifdef XXH_PRIVATE_API # ifndef XXH_STATIC_LINKING_ONLY # define XXH_STATIC_LINKING_ONLY # endif # if defined(__GNUC__) # define XXH_PUBLIC_API static __inline __attribute__((unused)) # elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) # define XXH_PUBLIC_API static inline # elif defined(_MSC_VER) # define XXH_PUBLIC_API static __inline # else # define XXH_PUBLIC_API static /* this version may generate warnings for unused static functions; disable the relevant warning */ # endif #else # define XXH_PUBLIC_API /* do nothing */ #endif /* XXH_PRIVATE_API */ /*!XXH_NAMESPACE, aka Namespace Emulation : If you want to include _and expose_ xxHash functions from within your own library, but also want to avoid symbol collisions with another library which also includes xxHash, you can use XXH_NAMESPACE, to automatically prefix any public symbol from xxhash library with the value of XXH_NAMESPACE (so avoid to keep it NULL and avoid numeric values). Note that no change is required within the calling program as long as it includes `xxhash.h` : regular symbol name will be automatically translated by this header. */ #ifdef XXH_NAMESPACE # define XXH_CAT(A,B) A##B # define XXH_NAME2(A,B) XXH_CAT(A,B) # define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32) # define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64) # define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber) # define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState) # define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState) # define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState) # define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState) # define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset) # define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset) # define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update) # define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update) # define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest) # define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest) # define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState) # define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState) # define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash) # define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash) # define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical) # define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical) #endif /* ************************************* * Version ***************************************/ #define XXH_VERSION_MAJOR 0 #define XXH_VERSION_MINOR 6 #define XXH_VERSION_RELEASE 2 #define XXH_VERSION_NUMBER (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE) XXH_PUBLIC_API unsigned XXH_versionNumber (void); /* **************************** * Simple Hash Functions ******************************/ typedef unsigned int XXH32_hash_t; typedef unsigned long long XXH64_hash_t; XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, unsigned int seed); +/* Begin FreeBSD - This symbol is needed by dll-linked CLI zstd(1). */ +__attribute__((visibility ("default"))) +/* End FreeBSD */ XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t length, unsigned long long seed); /*! XXH32() : Calculate the 32-bits hash of sequence "length" bytes stored at memory address "input". The memory between input & input+length must be valid (allocated and read-accessible). "seed" can be used to alter the result predictably. Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s XXH64() : Calculate the 64-bits hash of sequence of length "len" stored at memory address "input". "seed" can be used to alter the result predictably. This function runs 2x faster on 64-bits systems, but slower on 32-bits systems (see benchmark). */ /* **************************** * Streaming Hash Functions ******************************/ typedef struct XXH32_state_s XXH32_state_t; /* incomplete type */ typedef struct XXH64_state_s XXH64_state_t; /* incomplete type */ /*! State allocation, compatible with dynamic libraries */ XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void); XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr); XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void); XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr); /* hash streaming */ XXH_PUBLIC_API XXH_errorcode XXH32_reset (XXH32_state_t* statePtr, unsigned int seed); XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length); XXH_PUBLIC_API XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr); XXH_PUBLIC_API XXH_errorcode XXH64_reset (XXH64_state_t* statePtr, unsigned long long seed); XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length); XXH_PUBLIC_API XXH64_hash_t XXH64_digest (const XXH64_state_t* statePtr); /* These functions generate the xxHash of an input provided in multiple segments. Note that, for small input, they are slower than single-call functions, due to state management. For small input, prefer `XXH32()` and `XXH64()` . XXH state must first be allocated, using XXH*_createState() . Start a new hash by initializing state with a seed, using XXH*_reset(). Then, feed the hash state by calling XXH*_update() as many times as necessary. Obviously, input must be allocated and read accessible. The function returns an error code, with 0 meaning OK, and any other value meaning there is an error. Finally, a hash value can be produced anytime, by using XXH*_digest(). This function returns the nn-bits hash as an int or long long. It's still possible to continue inserting input into the hash state after a digest, and generate some new hashes later on, by calling again XXH*_digest(). When done, free XXH state space if it was allocated dynamically. */ /* ************************** * Utils ****************************/ #if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)) /* ! C99 */ # define restrict /* disable restrict */ #endif XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dst_state, const XXH32_state_t* restrict src_state); XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dst_state, const XXH64_state_t* restrict src_state); /* ************************** * Canonical representation ****************************/ /* Default result type for XXH functions are primitive unsigned 32 and 64 bits. * The canonical representation uses human-readable write convention, aka big-endian (large digits first). * These functions allow transformation of hash result into and from its canonical format. * This way, hash values can be written into a file / memory, and remain comparable on different systems and programs. */ typedef struct { unsigned char digest[4]; } XXH32_canonical_t; typedef struct { unsigned char digest[8]; } XXH64_canonical_t; XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash); XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash); XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src); XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src); #endif /* XXHASH_H_5627135585666179 */ /* ================================================================================================ This section contains definitions which are not guaranteed to remain stable. They may change in future versions, becoming incompatible with a different version of the library. They shall only be used with static linking. Never use these definitions in association with dynamic linking ! =================================================================================================== */ #if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXH_STATIC_H_3543687687345) #define XXH_STATIC_H_3543687687345 /* These definitions are only meant to allow allocation of XXH state statically, on stack, or in a struct for example. Do not use members directly. */ struct XXH32_state_s { unsigned total_len_32; unsigned large_len; unsigned v1; unsigned v2; unsigned v3; unsigned v4; unsigned mem32[4]; /* buffer defined as U32 for alignment */ unsigned memsize; unsigned reserved; /* never read nor write, will be removed in a future version */ }; /* typedef'd to XXH32_state_t */ struct XXH64_state_s { unsigned long long total_len; unsigned long long v1; unsigned long long v2; unsigned long long v3; unsigned long long v4; unsigned long long mem64[4]; /* buffer defined as U64 for alignment */ unsigned memsize; unsigned reserved[2]; /* never read nor write, will be removed in a future version */ }; /* typedef'd to XXH64_state_t */ # ifdef XXH_PRIVATE_API # include "xxhash.c" /* include xxhash functions as `static`, for inlining */ # endif #endif /* XXH_STATIC_LINKING_ONLY && XXH_STATIC_H_3543687687345 */ #if defined (__cplusplus) } #endif diff --git a/sys/contrib/zstd/lib/dictBuilder/zdict.c b/sys/contrib/zstd/lib/dictBuilder/zdict.c index 79c522ef3679..057ff3f16da4 100644 --- a/sys/contrib/zstd/lib/dictBuilder/zdict.c +++ b/sys/contrib/zstd/lib/dictBuilder/zdict.c @@ -1,1134 +1,1137 @@ /* * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc. * All rights reserved. * * 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. */ /*-************************************** * Tuning parameters ****************************************/ #define MINRATIO 4 /* minimum nb of apparition to be selected in dictionary */ #define ZDICT_MAX_SAMPLES_SIZE (2000U << 20) #define ZDICT_MIN_SAMPLES_SIZE (ZDICT_CONTENTSIZE_MIN * MINRATIO) /*-************************************** * Compiler Options ****************************************/ /* Unix Large Files support (>4GB) */ #define _FILE_OFFSET_BITS 64 #if (defined(__sun__) && (!defined(__LP64__))) /* Sun Solaris 32-bits requires specific definitions */ # define _LARGEFILE_SOURCE #elif ! defined(__LP64__) /* No point defining Large file for 64 bit */ # define _LARGEFILE64_SOURCE #endif /*-************************************* * Dependencies ***************************************/ #include /* malloc, free */ #include /* memset */ #include /* fprintf, fopen, ftello64 */ #include /* clock */ #include "../common/mem.h" /* read */ #include "../common/fse.h" /* FSE_normalizeCount, FSE_writeNCount */ #define HUF_STATIC_LINKING_ONLY #include "../common/huf.h" /* HUF_buildCTable, HUF_writeCTable */ #include "../common/zstd_internal.h" /* includes zstd.h */ #include "../common/xxhash.h" /* XXH64 */ #include "divsufsort.h" #ifndef ZDICT_STATIC_LINKING_ONLY # define ZDICT_STATIC_LINKING_ONLY #endif #include "zdict.h" #include "../compress/zstd_compress_internal.h" /* ZSTD_loadCEntropy() */ /*-************************************* * Constants ***************************************/ #define KB *(1 <<10) #define MB *(1 <<20) #define GB *(1U<<30) #define DICTLISTSIZE_DEFAULT 10000 #define NOISELENGTH 32 static const U32 g_selectivity_default = 9; /*-************************************* * Console display ***************************************/ #undef DISPLAY #define DISPLAY(...) { fprintf(stderr, __VA_ARGS__); fflush( stderr ); } #undef DISPLAYLEVEL #define DISPLAYLEVEL(l, ...) if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); } /* 0 : no display; 1: errors; 2: default; 3: details; 4: debug */ static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; } static void ZDICT_printHex(const void* ptr, size_t length) { const BYTE* const b = (const BYTE*)ptr; size_t u; for (u=0; u126) c = '.'; /* non-printable char */ DISPLAY("%c", c); } } /*-******************************************************** * Helper functions **********************************************************/ unsigned ZDICT_isError(size_t errorCode) { return ERR_isError(errorCode); } const char* ZDICT_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); } unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize) { if (dictSize < 8) return 0; if (MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return 0; return MEM_readLE32((const char*)dictBuffer + 4); } size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize) { size_t headerSize; if (dictSize <= 8 || MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return ERROR(dictionary_corrupted); { ZSTD_compressedBlockState_t* bs = (ZSTD_compressedBlockState_t*)malloc(sizeof(ZSTD_compressedBlockState_t)); U32* wksp = (U32*)malloc(HUF_WORKSPACE_SIZE); if (!bs || !wksp) { headerSize = ERROR(memory_allocation); } else { ZSTD_reset_compressedBlockState(bs); headerSize = ZSTD_loadCEntropy(bs, wksp, dictBuffer, dictSize); } free(bs); free(wksp); } return headerSize; } /*-******************************************************** * Dictionary training functions **********************************************************/ static unsigned ZDICT_NbCommonBytes (size_t val) { if (MEM_isLittleEndian()) { if (MEM_64bits()) { # if defined(_MSC_VER) && defined(_WIN64) unsigned long r = 0; _BitScanForward64( &r, (U64)val ); return (unsigned)(r>>3); # elif defined(__GNUC__) && (__GNUC__ >= 3) return (__builtin_ctzll((U64)val) >> 3); # else static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2, 0, 3, 1, 3, 1, 4, 2, 7, 0, 2, 3, 6, 1, 5, 3, 5, 1, 3, 4, 4, 2, 5, 6, 7, 7, 0, 1, 2, 3, 3, 4, 6, 2, 6, 5, 5, 3, 4, 5, 6, 7, 1, 2, 4, 6, 4, 4, 5, 7, 2, 6, 5, 7, 6, 7, 7 }; return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58]; # endif } else { /* 32 bits */ # if defined(_MSC_VER) unsigned long r=0; _BitScanForward( &r, (U32)val ); return (unsigned)(r>>3); # elif defined(__GNUC__) && (__GNUC__ >= 3) return (__builtin_ctz((U32)val) >> 3); # else static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0, 3, 2, 2, 1, 3, 2, 0, 1, 3, 3, 1, 2, 2, 2, 2, 0, 3, 1, 2, 0, 1, 0, 1, 1 }; return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27]; # endif } } else { /* Big Endian CPU */ if (MEM_64bits()) { # if defined(_MSC_VER) && defined(_WIN64) unsigned long r = 0; _BitScanReverse64( &r, val ); return (unsigned)(r>>3); # elif defined(__GNUC__) && (__GNUC__ >= 3) return (__builtin_clzll(val) >> 3); # else unsigned r; const unsigned n32 = sizeof(size_t)*4; /* calculate this way due to compiler complaining in 32-bits mode */ if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; } if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; } r += (!val); return r; # endif } else { /* 32 bits */ # if defined(_MSC_VER) unsigned long r = 0; _BitScanReverse( &r, (unsigned long)val ); return (unsigned)(r>>3); # elif defined(__GNUC__) && (__GNUC__ >= 3) return (__builtin_clz((U32)val) >> 3); # else unsigned r; if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; } r += (!val); return r; # endif } } } /*! ZDICT_count() : Count the nb of common bytes between 2 pointers. Note : this function presumes end of buffer followed by noisy guard band. */ static size_t ZDICT_count(const void* pIn, const void* pMatch) { const char* const pStart = (const char*)pIn; for (;;) { size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn); if (!diff) { pIn = (const char*)pIn+sizeof(size_t); pMatch = (const char*)pMatch+sizeof(size_t); continue; } pIn = (const char*)pIn+ZDICT_NbCommonBytes(diff); return (size_t)((const char*)pIn - pStart); } } typedef struct { U32 pos; U32 length; U32 savings; } dictItem; static void ZDICT_initDictItem(dictItem* d) { d->pos = 1; d->length = 0; d->savings = (U32)(-1); } #define LLIMIT 64 /* heuristic determined experimentally */ #define MINMATCHLENGTH 7 /* heuristic determined experimentally */ static dictItem ZDICT_analyzePos( BYTE* doneMarks, const int* suffix, U32 start, const void* buffer, U32 minRatio, U32 notificationLevel) { U32 lengthList[LLIMIT] = {0}; U32 cumulLength[LLIMIT] = {0}; U32 savings[LLIMIT] = {0}; const BYTE* b = (const BYTE*)buffer; size_t maxLength = LLIMIT; size_t pos = suffix[start]; U32 end = start; dictItem solution; /* init */ memset(&solution, 0, sizeof(solution)); doneMarks[pos] = 1; /* trivial repetition cases */ if ( (MEM_read16(b+pos+0) == MEM_read16(b+pos+2)) ||(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 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= 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) { U32 idx; for(idx=start; idx= %i at pos %7u ", (unsigned)(end-start), MINMATCHLENGTH, (unsigned)pos); DISPLAYLEVEL(4, "\n"); for (mml = MINMATCHLENGTH ; ; mml++) { BYTE currentChar = 0; U32 currentCount = 0; U32 currentID = refinedStart; U32 id; U32 selectedCount = 0; U32 selectedID = currentID; for (id =refinedStart; id < refinedEnd; id++) { if (b[suffix[id] + mml] != currentChar) { if (currentCount > selectedCount) { selectedCount = currentCount; selectedID = currentID; } currentID = id; currentChar = b[ suffix[id] + mml]; currentCount = 0; } currentCount ++; } if (currentCount > selectedCount) { /* for last */ selectedCount = currentCount; selectedID = currentID; } if (selectedCount < minRatio) break; refinedStart = selectedID; refinedEnd = refinedStart + selectedCount; } /* evaluate gain based on new dict */ start = refinedStart; pos = suffix[refinedStart]; end = start; 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 */ { 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)); cumulLength[maxLength-1] = lengthList[maxLength-1]; for (i=(int)(maxLength-2); i>=0; i--) cumulLength[i] = cumulLength[i+1] + lengthList[i]; for (i=LLIMIT-1; i>=MINMATCHLENGTH; i--) if (cumulLength[i]>=minRatio) break; maxLength = i; /* reduce maxLength in case of final into repetitive data */ { U32 l = (U32)maxLength; BYTE const c = b[pos + maxLength-1]; while (b[pos+l-2]==c) l--; maxLength = l; } if (maxLength < MINMATCHLENGTH) return solution; /* skip : no long-enough solution */ /* calculate savings */ savings[5] = 0; for (i=MINMATCHLENGTH; i<=(int)maxLength; i++) savings[i] = savings[i-1] + (lengthList[i] * (i-3)); DISPLAYLEVEL(4, "Selected dict at position %u, of length %u : saves %u (ratio: %.2f) \n", (unsigned)pos, (unsigned)maxLength, (unsigned)savings[maxLength], (double)savings[maxLength] / maxLength); solution.pos = (U32)pos; solution.length = (U32)maxLength; solution.savings = savings[maxLength]; /* mark positions done */ { U32 id; for (id=start; id solution.length) length = solution.length; } pEnd = (U32)(testedPos + length); for (p=testedPos; ppos; const U32 eltEnd = elt.pos + elt.length; const char* const buf = (const char*) buffer; /* tail overlap */ U32 u; for (u=1; u elt.pos) && (table[u].pos <= eltEnd)) { /* overlap, existing > new */ /* append */ U32 const addedLength = table[u].pos - elt.pos; table[u].length += addedLength; table[u].pos = elt.pos; table[u].savings += elt.savings * addedLength / elt.length; /* rough approx */ table[u].savings += elt.length / 8; /* rough approx bonus */ elt = table[u]; /* sort : improve rank */ while ((u>1) && (table[u-1].savings < elt.savings)) table[u] = table[u-1], u--; table[u] = elt; return u; } } /* front overlap */ for (u=1; u= elt.pos) && (table[u].pos < elt.pos)) { /* overlap, existing < new */ /* append */ int const addedLength = (int)eltEnd - (table[u].pos + table[u].length); table[u].savings += elt.length / 8; /* rough approx bonus */ if (addedLength > 0) { /* otherwise, elt fully included into existing */ table[u].length += addedLength; table[u].savings += elt.savings * addedLength / elt.length; /* rough approx */ } /* sort : improve rank */ elt = table[u]; while ((u>1) && (table[u-1].savings < elt.savings)) table[u] = table[u-1], u--; table[u] = elt; return u; } if (MEM_read64(buf + table[u].pos) == MEM_read64(buf + elt.pos + 1)) { if (isIncluded(buf + table[u].pos, buf + elt.pos + 1, table[u].length)) { size_t const addedLength = MAX( (int)elt.length - (int)table[u].length , 1 ); table[u].pos = elt.pos; table[u].savings += (U32)(elt.savings * addedLength / elt.length); table[u].length = MIN(elt.length, table[u].length + 1); return u; } } } return 0; } static void ZDICT_removeDictItem(dictItem* table, U32 id) { /* convention : table[0].pos stores nb of elts */ U32 const max = table[0].pos; U32 u; if (!id) return; /* protection, should never happen */ for (u=id; upos--; } static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt, const void* buffer) { /* merge if possible */ U32 mergeId = ZDICT_tryMerge(table, elt, 0, buffer); if (mergeId) { U32 newMerge = 1; while (newMerge) { newMerge = ZDICT_tryMerge(table, table[mergeId], mergeId, buffer); if (newMerge) ZDICT_removeDictItem(table, mergeId); mergeId = newMerge; } return; } /* insert */ { U32 current; U32 nextElt = table->pos; if (nextElt >= maxSize) nextElt = maxSize-1; current = nextElt-1; while (table[current].savings < elt.savings) { table[current+1] = table[current]; current--; } table[current+1] = elt; table->pos = nextElt+1; } } static U32 ZDICT_dictSize(const dictItem* dictList) { U32 u, dictSize = 0; for (u=1; u=l) { \ if (ZDICT_clockSpan(displayClock) > refreshRate) \ { displayClock = clock(); DISPLAY(__VA_ARGS__); \ if (notificationLevel>=4) fflush(stderr); } } /* init */ DISPLAYLEVEL(2, "\r%70s\r", ""); /* clean display line */ if (!suffix0 || !reverseSuffix || !doneMarks || !filePos) { result = ERROR(memory_allocation); goto _cleanup; } if (minRatio < MINRATIO) minRatio = MINRATIO; memset(doneMarks, 0, bufferSize+16); /* limit sample set size (divsufsort limitation)*/ if (bufferSize > ZDICT_MAX_SAMPLES_SIZE) DISPLAYLEVEL(3, "sample set too large : reduced to %u MB ...\n", (unsigned)(ZDICT_MAX_SAMPLES_SIZE>>20)); while (bufferSize > ZDICT_MAX_SAMPLES_SIZE) bufferSize -= fileSizes[--nbFiles]; /* sort */ DISPLAYLEVEL(2, "sorting %u files of total size %u MB ...\n", nbFiles, (unsigned)(bufferSize>>20)); { int const divSuftSortResult = divsufsort((const unsigned char*)buffer, suffix, (int)bufferSize, 0); if (divSuftSortResult != 0) { result = ERROR(GENERIC); goto _cleanup; } } suffix[bufferSize] = (int)bufferSize; /* leads into noise */ suffix0[0] = (int)bufferSize; /* leads into noise */ /* build reverse suffix sort */ { size_t pos; for (pos=0; pos < bufferSize; pos++) reverseSuffix[suffix[pos]] = (U32)pos; /* note filePos tracks borders between samples. It's not used at this stage, but planned to become useful in a later update */ filePos[0] = 0; for (pos=1; pos> 21); } } typedef struct { ZSTD_CDict* dict; /* dictionary */ ZSTD_CCtx* zc; /* working context */ void* workPlace; /* must be ZSTD_BLOCKSIZE_MAX allocated */ } EStats_ress_t; #define MAXREPOFFSET 1024 static void ZDICT_countEStats(EStats_ress_t esr, const ZSTD_parameters* params, unsigned* countLit, unsigned* offsetcodeCount, unsigned* matchlengthCount, unsigned* litlengthCount, U32* repOffsets, const void* src, size_t srcSize, U32 notificationLevel) { size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_MAX, 1 << params->cParams.windowLog); size_t cSize; if (srcSize > blockSizeMax) srcSize = blockSizeMax; /* protection vs large samples */ { size_t const errorCode = ZSTD_compressBegin_usingCDict(esr.zc, esr.dict); if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_compressBegin_usingCDict failed \n"); return; } } cSize = ZSTD_compressBlock(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_MAX, src, srcSize); if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (unsigned)srcSize); return; } if (cSize) { /* if == 0; block is not compressible */ const seqStore_t* const seqStorePtr = ZSTD_getSeqStore(esr.zc); /* literals stats */ { const BYTE* bytePtr; for(bytePtr = seqStorePtr->litStart; bytePtr < seqStorePtr->lit; bytePtr++) countLit[*bytePtr]++; } /* seqStats */ { U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart); ZSTD_seqToCodes(seqStorePtr); { const BYTE* codePtr = seqStorePtr->ofCode; U32 u; for (u=0; umlCode; U32 u; for (u=0; ullCode; U32 u; for (u=0; u= 2) { /* rep offsets */ const seqDef* const seq = seqStorePtr->sequencesStart; U32 offset1 = seq[0].offset - 3; U32 offset2 = seq[1].offset - 3; if (offset1 >= MAXREPOFFSET) offset1 = 0; if (offset2 >= MAXREPOFFSET) offset2 = 0; repOffsets[offset1] += 3; repOffsets[offset2] += 1; } } } } static size_t ZDICT_totalSampleSize(const size_t* fileSizes, unsigned nbFiles) { size_t total=0; unsigned u; for (u=0; u0; u--) { offsetCount_t tmp; if (table[u-1].count >= table[u].count) break; tmp = table[u-1]; table[u-1] = table[u]; table[u] = tmp; } } /* 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(unsigned* 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, int compressionLevel, const void* srcBuffer, const size_t* fileSizes, unsigned nbFiles, const void* dictBuffer, size_t dictBufferSize, unsigned notificationLevel) { unsigned countLit[256]; HUF_CREATE_STATIC_CTABLE(hufTable, 255); unsigned offcodeCount[OFFCODE_MAX+1]; short offcodeNCount[OFFCODE_MAX+1]; U32 offcodeMax = ZSTD_highbit32((U32)(dictBufferSize + 128 KB)); unsigned matchLengthCount[MaxML+1]; short matchLengthNCount[MaxML+1]; unsigned litLengthCount[MaxLL+1]; short litLengthNCount[MaxLL+1]; U32 repOffset[MAXREPOFFSET]; offsetCount_t bestRepOffset[ZSTD_REP_NUM+1]; EStats_ress_t esr = { NULL, NULL, NULL }; ZSTD_parameters params; U32 u, huffLog = 11, Offlog = OffFSELog, mlLog = MLFSELog, llLog = LLFSELog, total; size_t pos = 0, errorCode; size_t eSize = 0; size_t const totalSrcSize = ZDICT_totalSampleSize(fileSizes, nbFiles); size_t const averageSampleSize = totalSrcSize / (nbFiles + !nbFiles); BYTE* dstPtr = (BYTE*)dstBuffer; /* init */ DEBUGLOG(4, "ZDICT_analyzeEntropy"); if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; } /* too large dictionary */ for (u=0; u<256; u++) countLit[u] = 1; /* any character must be described */ for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1; for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1; for (u=0; u<=MaxLL; u++) litLengthCount[u] = 1; memset(repOffset, 0, sizeof(repOffset)); repOffset[1] = repOffset[4] = repOffset[8] = 1; memset(bestRepOffset, 0, sizeof(bestRepOffset)); if (compressionLevel==0) compressionLevel = ZSTD_CLEVEL_DEFAULT; params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize); esr.dict = ZSTD_createCDict_advanced(dictBuffer, dictBufferSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, params.cParams, ZSTD_defaultCMem); esr.zc = ZSTD_createCCtx(); esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX); if (!esr.dict || !esr.zc || !esr.workPlace) { eSize = ERROR(memory_allocation); DISPLAYLEVEL(1, "Not enough memory \n"); goto _cleanup; } /* collect stats on all samples */ for (u=0; u dictBufferCapacity) dictContentSize = dictBufferCapacity - hSize; { size_t const dictSize = hSize + dictContentSize; char* dictEnd = (char*)dictBuffer + dictSize; memmove(dictEnd - dictContentSize, customDictContent, dictContentSize); memcpy(dictBuffer, header, hSize); return dictSize; } } static size_t ZDICT_addEntropyTablesFromBuffer_advanced( void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity, const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples, ZDICT_params_t params) { int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel; U32 const notificationLevel = params.notificationLevel; size_t hSize = 8; /* calculate entropy tables */ DISPLAYLEVEL(2, "\r%70s\r", ""); /* clean display line */ DISPLAYLEVEL(2, "statistics ... \n"); { size_t const eSize = ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize, compressionLevel, samplesBuffer, samplesSizes, nbSamples, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, notificationLevel); if (ZDICT_isError(eSize)) return eSize; hSize += eSize; } /* add dictionary header (after entropy tables) */ MEM_writeLE32(dictBuffer, ZSTD_MAGIC_DICTIONARY); { U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0); U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768; U32 const dictID = params.dictID ? params.dictID : compliantID; MEM_writeLE32((char*)dictBuffer+4, dictID); } if (hSize + dictContentSize < dictBufferCapacity) memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize); return MIN(dictBufferCapacity, hSize+dictContentSize); } /* Hidden declaration for dbio.c */ +/* Begin FreeBSD - This symbol is needed by dll-linked CLI zstd(1). */ +ZSTDLIB_API +/* End FreeBSD */ size_t ZDICT_trainFromBuffer_unsafe_legacy( void* dictBuffer, size_t maxDictSize, const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples, ZDICT_legacy_params_t params); /*! ZDICT_trainFromBuffer_unsafe_legacy() : * Warning : `samplesBuffer` must be followed by noisy guard band. * @return : size of dictionary, or an error code which can be tested with ZDICT_isError() */ size_t ZDICT_trainFromBuffer_unsafe_legacy( void* dictBuffer, size_t maxDictSize, const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples, ZDICT_legacy_params_t params) { U32 const dictListSize = MAX(MAX(DICTLISTSIZE_DEFAULT, nbSamples), (U32)(maxDictSize/16)); dictItem* const dictList = (dictItem*)malloc(dictListSize * sizeof(*dictList)); unsigned const selectivity = params.selectivityLevel == 0 ? g_selectivity_default : params.selectivityLevel; unsigned const minRep = (selectivity > 30) ? MINRATIO : nbSamples >> selectivity; size_t const targetDictSize = maxDictSize; size_t const samplesBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples); size_t dictSize = 0; U32 const notificationLevel = params.zParams.notificationLevel; /* checks */ if (!dictList) return ERROR(memory_allocation); if (maxDictSize < ZDICT_DICTSIZE_MIN) { free(dictList); return ERROR(dstSize_tooSmall); } /* requested dictionary size is too small */ if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return ERROR(dictionaryCreation_failed); } /* not enough source to create dictionary */ /* init */ ZDICT_initDictItem(dictList); /* build dictionary */ ZDICT_trainBuffer_legacy(dictList, dictListSize, samplesBuffer, samplesBuffSize, samplesSizes, nbSamples, minRep, notificationLevel); /* display best matches */ if (params.zParams.notificationLevel>= 3) { unsigned const nb = MIN(25, dictList[0].pos); unsigned const dictContentSize = ZDICT_dictSize(dictList); unsigned u; DISPLAYLEVEL(3, "\n %u segments found, of total size %u \n", (unsigned)dictList[0].pos-1, dictContentSize); DISPLAYLEVEL(3, "list %u best segments \n", nb-1); for (u=1; u samplesBuffSize) || ((pos + length) > samplesBuffSize)) { free(dictList); return ERROR(GENERIC); /* should never happen */ } DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |", u, length, pos, (unsigned)dictList[u].savings); ZDICT_printHex((const char*)samplesBuffer+pos, printedLength); DISPLAYLEVEL(3, "| \n"); } } /* create dictionary */ { unsigned dictContentSize = ZDICT_dictSize(dictList); if (dictContentSize < ZDICT_CONTENTSIZE_MIN) { free(dictList); return ERROR(dictionaryCreation_failed); } /* dictionary content too small */ if (dictContentSize < targetDictSize/4) { DISPLAYLEVEL(2, "! warning : selected content significantly smaller than requested (%u < %u) \n", dictContentSize, (unsigned)maxDictSize); if (samplesBuffSize < 10 * targetDictSize) DISPLAYLEVEL(2, "! consider increasing the number of samples (total size : %u MB)\n", (unsigned)(samplesBuffSize>>20)); if (minRep > MINRATIO) { DISPLAYLEVEL(2, "! consider increasing selectivity to produce larger dictionary (-s%u) \n", selectivity+1); DISPLAYLEVEL(2, "! note : larger dictionaries are not necessarily better, test its efficiency on samples \n"); } } if ((dictContentSize > targetDictSize*3) && (nbSamples > 2*MINRATIO) && (selectivity>1)) { unsigned proposedSelectivity = selectivity-1; while ((nbSamples >> proposedSelectivity) <= MINRATIO) { proposedSelectivity--; } DISPLAYLEVEL(2, "! note : calculated dictionary significantly larger than requested (%u > %u) \n", dictContentSize, (unsigned)maxDictSize); DISPLAYLEVEL(2, "! consider increasing dictionary size, or produce denser dictionary (-s%u) \n", proposedSelectivity); DISPLAYLEVEL(2, "! always test dictionary efficiency on real samples \n"); } /* limit dictionary size */ { U32 const max = dictList->pos; /* convention : nb of useful elts within dictList */ U32 currentSize = 0; U32 n; for (n=1; n targetDictSize) { currentSize -= dictList[n].length; break; } } dictList->pos = n; dictContentSize = currentSize; } /* build dict content */ { U32 u; BYTE* ptr = (BYTE*)dictBuffer + maxDictSize; for (u=1; upos; u++) { U32 l = dictList[u].length; ptr -= l; if (ptr<(BYTE*)dictBuffer) { free(dictList); return ERROR(GENERIC); } /* should not happen */ memcpy(ptr, (const char*)samplesBuffer+dictList[u].pos, l); } } dictSize = ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, maxDictSize, samplesBuffer, samplesSizes, nbSamples, params.zParams); } /* clean up */ free(dictList); return dictSize; } /* 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) { size_t result; void* newBuff; size_t const sBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples); if (sBuffSize < ZDICT_MIN_SAMPLES_SIZE) return 0; /* not enough content => no dictionary */ newBuff = malloc(sBuffSize + NOISELENGTH); if (!newBuff) return ERROR(memory_allocation); memcpy(newBuff, samplesBuffer, sBuffSize); ZDICT_fillNoise((char*)newBuff + sBuffSize, NOISELENGTH); /* guard band, for end of buffer condition */ result = ZDICT_trainFromBuffer_unsafe_legacy(dictBuffer, dictBufferCapacity, newBuff, samplesSizes, nbSamples, params); free(newBuff); return result; } size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity, const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples) { ZDICT_fastCover_params_t params; DEBUGLOG(3, "ZDICT_trainFromBuffer"); memset(¶ms, 0, sizeof(params)); params.d = 8; params.steps = 4; /* Use default level since no compression level information is available */ params.zParams.compressionLevel = ZSTD_CLEVEL_DEFAULT; #if defined(DEBUGLEVEL) && (DEBUGLEVEL>=1) params.zParams.notificationLevel = DEBUGLEVEL; #endif return ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, dictBufferCapacity, samplesBuffer, samplesSizes, nbSamples, ¶ms); } size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity, const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples) { ZDICT_params_t params; memset(¶ms, 0, sizeof(params)); return ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, dictBufferCapacity, samplesBuffer, samplesSizes, nbSamples, params); }