1*3117ece4Schristos /* 2*3117ece4Schristos * Copyright (c) Meta Platforms, Inc. and affiliates. 3*3117ece4Schristos * All rights reserved. 4*3117ece4Schristos * 5*3117ece4Schristos * This source code is licensed under both the BSD-style license (found in the 6*3117ece4Schristos * LICENSE file in the root directory of this source tree) and the GPLv2 (found 7*3117ece4Schristos * in the COPYING file in the root directory of this source tree). 8*3117ece4Schristos * You may select, at your option, one of the above-listed licenses. 9*3117ece4Schristos */ 10*3117ece4Schristos 11*3117ece4Schristos #if defined (__cplusplus) 12*3117ece4Schristos extern "C" { 13*3117ece4Schristos #endif 14*3117ece4Schristos 15*3117ece4Schristos #ifndef ZSTD_ZDICT_H 16*3117ece4Schristos #define ZSTD_ZDICT_H 17*3117ece4Schristos 18*3117ece4Schristos /*====== Dependencies ======*/ 19*3117ece4Schristos #include <stddef.h> /* size_t */ 20*3117ece4Schristos 21*3117ece4Schristos 22*3117ece4Schristos /* ===== ZDICTLIB_API : control library symbols visibility ===== */ 23*3117ece4Schristos #ifndef ZDICTLIB_VISIBLE 24*3117ece4Schristos /* Backwards compatibility with old macro name */ 25*3117ece4Schristos # ifdef ZDICTLIB_VISIBILITY 26*3117ece4Schristos # define ZDICTLIB_VISIBLE ZDICTLIB_VISIBILITY 27*3117ece4Schristos # elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__) 28*3117ece4Schristos # define ZDICTLIB_VISIBLE __attribute__ ((visibility ("default"))) 29*3117ece4Schristos # else 30*3117ece4Schristos # define ZDICTLIB_VISIBLE 31*3117ece4Schristos # endif 32*3117ece4Schristos #endif 33*3117ece4Schristos 34*3117ece4Schristos #ifndef ZDICTLIB_HIDDEN 35*3117ece4Schristos # if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__) 36*3117ece4Schristos # define ZDICTLIB_HIDDEN __attribute__ ((visibility ("hidden"))) 37*3117ece4Schristos # else 38*3117ece4Schristos # define ZDICTLIB_HIDDEN 39*3117ece4Schristos # endif 40*3117ece4Schristos #endif 41*3117ece4Schristos 42*3117ece4Schristos #if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1) 43*3117ece4Schristos # define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBLE 44*3117ece4Schristos #elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1) 45*3117ece4Schristos # define ZDICTLIB_API __declspec(dllimport) ZDICTLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/ 46*3117ece4Schristos #else 47*3117ece4Schristos # define ZDICTLIB_API ZDICTLIB_VISIBLE 48*3117ece4Schristos #endif 49*3117ece4Schristos 50*3117ece4Schristos /******************************************************************************* 51*3117ece4Schristos * Zstd dictionary builder 52*3117ece4Schristos * 53*3117ece4Schristos * FAQ 54*3117ece4Schristos * === 55*3117ece4Schristos * Why should I use a dictionary? 56*3117ece4Schristos * ------------------------------ 57*3117ece4Schristos * 58*3117ece4Schristos * Zstd can use dictionaries to improve compression ratio of small data. 59*3117ece4Schristos * Traditionally small files don't compress well because there is very little 60*3117ece4Schristos * repetition in a single sample, since it is small. But, if you are compressing 61*3117ece4Schristos * many similar files, like a bunch of JSON records that share the same 62*3117ece4Schristos * structure, you can train a dictionary on ahead of time on some samples of 63*3117ece4Schristos * these files. Then, zstd can use the dictionary to find repetitions that are 64*3117ece4Schristos * present across samples. This can vastly improve compression ratio. 65*3117ece4Schristos * 66*3117ece4Schristos * When is a dictionary useful? 67*3117ece4Schristos * ---------------------------- 68*3117ece4Schristos * 69*3117ece4Schristos * Dictionaries are useful when compressing many small files that are similar. 70*3117ece4Schristos * The larger a file is, the less benefit a dictionary will have. Generally, 71*3117ece4Schristos * we don't expect dictionary compression to be effective past 100KB. And the 72*3117ece4Schristos * smaller a file is, the more we would expect the dictionary to help. 73*3117ece4Schristos * 74*3117ece4Schristos * How do I use a dictionary? 75*3117ece4Schristos * -------------------------- 76*3117ece4Schristos * 77*3117ece4Schristos * Simply pass the dictionary to the zstd compressor with 78*3117ece4Schristos * `ZSTD_CCtx_loadDictionary()`. The same dictionary must then be passed to 79*3117ece4Schristos * the decompressor, using `ZSTD_DCtx_loadDictionary()`. There are other 80*3117ece4Schristos * more advanced functions that allow selecting some options, see zstd.h for 81*3117ece4Schristos * complete documentation. 82*3117ece4Schristos * 83*3117ece4Schristos * What is a zstd dictionary? 84*3117ece4Schristos * -------------------------- 85*3117ece4Schristos * 86*3117ece4Schristos * A zstd dictionary has two pieces: Its header, and its content. The header 87*3117ece4Schristos * contains a magic number, the dictionary ID, and entropy tables. These 88*3117ece4Schristos * entropy tables allow zstd to save on header costs in the compressed file, 89*3117ece4Schristos * which really matters for small data. The content is just bytes, which are 90*3117ece4Schristos * repeated content that is common across many samples. 91*3117ece4Schristos * 92*3117ece4Schristos * What is a raw content dictionary? 93*3117ece4Schristos * --------------------------------- 94*3117ece4Schristos * 95*3117ece4Schristos * A raw content dictionary is just bytes. It doesn't have a zstd dictionary 96*3117ece4Schristos * header, a dictionary ID, or entropy tables. Any buffer is a valid raw 97*3117ece4Schristos * content dictionary. 98*3117ece4Schristos * 99*3117ece4Schristos * How do I train a dictionary? 100*3117ece4Schristos * ---------------------------- 101*3117ece4Schristos * 102*3117ece4Schristos * Gather samples from your use case. These samples should be similar to each 103*3117ece4Schristos * other. If you have several use cases, you could try to train one dictionary 104*3117ece4Schristos * per use case. 105*3117ece4Schristos * 106*3117ece4Schristos * Pass those samples to `ZDICT_trainFromBuffer()` and that will train your 107*3117ece4Schristos * dictionary. There are a few advanced versions of this function, but this 108*3117ece4Schristos * is a great starting point. If you want to further tune your dictionary 109*3117ece4Schristos * you could try `ZDICT_optimizeTrainFromBuffer_cover()`. If that is too slow 110*3117ece4Schristos * you can try `ZDICT_optimizeTrainFromBuffer_fastCover()`. 111*3117ece4Schristos * 112*3117ece4Schristos * If the dictionary training function fails, that is likely because you 113*3117ece4Schristos * either passed too few samples, or a dictionary would not be effective 114*3117ece4Schristos * for your data. Look at the messages that the dictionary trainer printed, 115*3117ece4Schristos * if it doesn't say too few samples, then a dictionary would not be effective. 116*3117ece4Schristos * 117*3117ece4Schristos * How large should my dictionary be? 118*3117ece4Schristos * ---------------------------------- 119*3117ece4Schristos * 120*3117ece4Schristos * A reasonable dictionary size, the `dictBufferCapacity`, is about 100KB. 121*3117ece4Schristos * The zstd CLI defaults to a 110KB dictionary. You likely don't need a 122*3117ece4Schristos * dictionary larger than that. But, most use cases can get away with a 123*3117ece4Schristos * smaller dictionary. The advanced dictionary builders can automatically 124*3117ece4Schristos * shrink the dictionary for you, and select the smallest size that doesn't 125*3117ece4Schristos * hurt compression ratio too much. See the `shrinkDict` parameter. 126*3117ece4Schristos * A smaller dictionary can save memory, and potentially speed up 127*3117ece4Schristos * compression. 128*3117ece4Schristos * 129*3117ece4Schristos * How many samples should I provide to the dictionary builder? 130*3117ece4Schristos * ------------------------------------------------------------ 131*3117ece4Schristos * 132*3117ece4Schristos * We generally recommend passing ~100x the size of the dictionary 133*3117ece4Schristos * in samples. A few thousand should suffice. Having too few samples 134*3117ece4Schristos * can hurt the dictionaries effectiveness. Having more samples will 135*3117ece4Schristos * only improve the dictionaries effectiveness. But having too many 136*3117ece4Schristos * samples can slow down the dictionary builder. 137*3117ece4Schristos * 138*3117ece4Schristos * How do I determine if a dictionary will be effective? 139*3117ece4Schristos * ----------------------------------------------------- 140*3117ece4Schristos * 141*3117ece4Schristos * Simply train a dictionary and try it out. You can use zstd's built in 142*3117ece4Schristos * benchmarking tool to test the dictionary effectiveness. 143*3117ece4Schristos * 144*3117ece4Schristos * # Benchmark levels 1-3 without a dictionary 145*3117ece4Schristos * zstd -b1e3 -r /path/to/my/files 146*3117ece4Schristos * # Benchmark levels 1-3 with a dictionary 147*3117ece4Schristos * zstd -b1e3 -r /path/to/my/files -D /path/to/my/dictionary 148*3117ece4Schristos * 149*3117ece4Schristos * When should I retrain a dictionary? 150*3117ece4Schristos * ----------------------------------- 151*3117ece4Schristos * 152*3117ece4Schristos * You should retrain a dictionary when its effectiveness drops. Dictionary 153*3117ece4Schristos * effectiveness drops as the data you are compressing changes. Generally, we do 154*3117ece4Schristos * expect dictionaries to "decay" over time, as your data changes, but the rate 155*3117ece4Schristos * at which they decay depends on your use case. Internally, we regularly 156*3117ece4Schristos * retrain dictionaries, and if the new dictionary performs significantly 157*3117ece4Schristos * better than the old dictionary, we will ship the new dictionary. 158*3117ece4Schristos * 159*3117ece4Schristos * I have a raw content dictionary, how do I turn it into a zstd dictionary? 160*3117ece4Schristos * ------------------------------------------------------------------------- 161*3117ece4Schristos * 162*3117ece4Schristos * If you have a raw content dictionary, e.g. by manually constructing it, or 163*3117ece4Schristos * using a third-party dictionary builder, you can turn it into a zstd 164*3117ece4Schristos * dictionary by using `ZDICT_finalizeDictionary()`. You'll also have to 165*3117ece4Schristos * provide some samples of the data. It will add the zstd header to the 166*3117ece4Schristos * raw content, which contains a dictionary ID and entropy tables, which 167*3117ece4Schristos * will improve compression ratio, and allow zstd to write the dictionary ID 168*3117ece4Schristos * into the frame, if you so choose. 169*3117ece4Schristos * 170*3117ece4Schristos * Do I have to use zstd's dictionary builder? 171*3117ece4Schristos * ------------------------------------------- 172*3117ece4Schristos * 173*3117ece4Schristos * No! You can construct dictionary content however you please, it is just 174*3117ece4Schristos * bytes. It will always be valid as a raw content dictionary. If you want 175*3117ece4Schristos * a zstd dictionary, which can improve compression ratio, use 176*3117ece4Schristos * `ZDICT_finalizeDictionary()`. 177*3117ece4Schristos * 178*3117ece4Schristos * What is the attack surface of a zstd dictionary? 179*3117ece4Schristos * ------------------------------------------------ 180*3117ece4Schristos * 181*3117ece4Schristos * Zstd is heavily fuzz tested, including loading fuzzed dictionaries, so 182*3117ece4Schristos * zstd should never crash, or access out-of-bounds memory no matter what 183*3117ece4Schristos * the dictionary is. However, if an attacker can control the dictionary 184*3117ece4Schristos * during decompression, they can cause zstd to generate arbitrary bytes, 185*3117ece4Schristos * just like if they controlled the compressed data. 186*3117ece4Schristos * 187*3117ece4Schristos ******************************************************************************/ 188*3117ece4Schristos 189*3117ece4Schristos 190*3117ece4Schristos /*! ZDICT_trainFromBuffer(): 191*3117ece4Schristos * Train a dictionary from an array of samples. 192*3117ece4Schristos * Redirect towards ZDICT_optimizeTrainFromBuffer_fastCover() single-threaded, with d=8, steps=4, 193*3117ece4Schristos * f=20, and accel=1. 194*3117ece4Schristos * Samples must be stored concatenated in a single flat buffer `samplesBuffer`, 195*3117ece4Schristos * supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order. 196*3117ece4Schristos * The resulting dictionary will be saved into `dictBuffer`. 197*3117ece4Schristos * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) 198*3117ece4Schristos * or an error code, which can be tested with ZDICT_isError(). 199*3117ece4Schristos * Note: Dictionary training will fail if there are not enough samples to construct a 200*3117ece4Schristos * dictionary, or if most of the samples are too small (< 8 bytes being the lower limit). 201*3117ece4Schristos * If dictionary training fails, you should use zstd without a dictionary, as the dictionary 202*3117ece4Schristos * would've been ineffective anyways. If you believe your samples would benefit from a dictionary 203*3117ece4Schristos * please open an issue with details, and we can look into it. 204*3117ece4Schristos * Note: ZDICT_trainFromBuffer()'s memory usage is about 6 MB. 205*3117ece4Schristos * Tips: In general, a reasonable dictionary has a size of ~ 100 KB. 206*3117ece4Schristos * It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`. 207*3117ece4Schristos * In general, it's recommended to provide a few thousands samples, though this can vary a lot. 208*3117ece4Schristos * It's recommended that total size of all samples be about ~x100 times the target size of dictionary. 209*3117ece4Schristos */ 210*3117ece4Schristos ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity, 211*3117ece4Schristos const void* samplesBuffer, 212*3117ece4Schristos const size_t* samplesSizes, unsigned nbSamples); 213*3117ece4Schristos 214*3117ece4Schristos typedef struct { 215*3117ece4Schristos int compressionLevel; /**< optimize for a specific zstd compression level; 0 means default */ 216*3117ece4Schristos unsigned notificationLevel; /**< Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */ 217*3117ece4Schristos unsigned dictID; /**< force dictID value; 0 means auto mode (32-bits random value) 218*3117ece4Schristos * NOTE: The zstd format reserves some dictionary IDs for future use. 219*3117ece4Schristos * You may use them in private settings, but be warned that they 220*3117ece4Schristos * may be used by zstd in a public dictionary registry in the future. 221*3117ece4Schristos * These dictionary IDs are: 222*3117ece4Schristos * - low range : <= 32767 223*3117ece4Schristos * - high range : >= (2^31) 224*3117ece4Schristos */ 225*3117ece4Schristos } ZDICT_params_t; 226*3117ece4Schristos 227*3117ece4Schristos /*! ZDICT_finalizeDictionary(): 228*3117ece4Schristos * Given a custom content as a basis for dictionary, and a set of samples, 229*3117ece4Schristos * finalize dictionary by adding headers and statistics according to the zstd 230*3117ece4Schristos * dictionary format. 231*3117ece4Schristos * 232*3117ece4Schristos * Samples must be stored concatenated in a flat buffer `samplesBuffer`, 233*3117ece4Schristos * supplied with an array of sizes `samplesSizes`, providing the size of each 234*3117ece4Schristos * sample in order. The samples are used to construct the statistics, so they 235*3117ece4Schristos * should be representative of what you will compress with this dictionary. 236*3117ece4Schristos * 237*3117ece4Schristos * The compression level can be set in `parameters`. You should pass the 238*3117ece4Schristos * compression level you expect to use in production. The statistics for each 239*3117ece4Schristos * compression level differ, so tuning the dictionary for the compression level 240*3117ece4Schristos * can help quite a bit. 241*3117ece4Schristos * 242*3117ece4Schristos * You can set an explicit dictionary ID in `parameters`, or allow us to pick 243*3117ece4Schristos * a random dictionary ID for you, but we can't guarantee no collisions. 244*3117ece4Schristos * 245*3117ece4Schristos * The dstDictBuffer and the dictContent may overlap, and the content will be 246*3117ece4Schristos * appended to the end of the header. If the header + the content doesn't fit in 247*3117ece4Schristos * maxDictSize the beginning of the content is truncated to make room, since it 248*3117ece4Schristos * is presumed that the most profitable content is at the end of the dictionary, 249*3117ece4Schristos * since that is the cheapest to reference. 250*3117ece4Schristos * 251*3117ece4Schristos * `maxDictSize` must be >= max(dictContentSize, ZSTD_DICTSIZE_MIN). 252*3117ece4Schristos * 253*3117ece4Schristos * @return: size of dictionary stored into `dstDictBuffer` (<= `maxDictSize`), 254*3117ece4Schristos * or an error code, which can be tested by ZDICT_isError(). 255*3117ece4Schristos * Note: ZDICT_finalizeDictionary() will push notifications into stderr if 256*3117ece4Schristos * instructed to, using notificationLevel>0. 257*3117ece4Schristos * NOTE: This function currently may fail in several edge cases including: 258*3117ece4Schristos * * Not enough samples 259*3117ece4Schristos * * Samples are uncompressible 260*3117ece4Schristos * * Samples are all exactly the same 261*3117ece4Schristos */ 262*3117ece4Schristos ZDICTLIB_API size_t ZDICT_finalizeDictionary(void* dstDictBuffer, size_t maxDictSize, 263*3117ece4Schristos const void* dictContent, size_t dictContentSize, 264*3117ece4Schristos const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples, 265*3117ece4Schristos ZDICT_params_t parameters); 266*3117ece4Schristos 267*3117ece4Schristos 268*3117ece4Schristos /*====== Helper functions ======*/ 269*3117ece4Schristos ZDICTLIB_API unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize); /**< extracts dictID; @return zero if error (not a valid dictionary) */ 270*3117ece4Schristos ZDICTLIB_API size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize); /* returns dict header size; returns a ZSTD error code on failure */ 271*3117ece4Schristos ZDICTLIB_API unsigned ZDICT_isError(size_t errorCode); 272*3117ece4Schristos ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode); 273*3117ece4Schristos 274*3117ece4Schristos #endif /* ZSTD_ZDICT_H */ 275*3117ece4Schristos 276*3117ece4Schristos #if defined(ZDICT_STATIC_LINKING_ONLY) && !defined(ZSTD_ZDICT_H_STATIC) 277*3117ece4Schristos #define ZSTD_ZDICT_H_STATIC 278*3117ece4Schristos 279*3117ece4Schristos /* This can be overridden externally to hide static symbols. */ 280*3117ece4Schristos #ifndef ZDICTLIB_STATIC_API 281*3117ece4Schristos # if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1) 282*3117ece4Schristos # define ZDICTLIB_STATIC_API __declspec(dllexport) ZDICTLIB_VISIBLE 283*3117ece4Schristos # elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1) 284*3117ece4Schristos # define ZDICTLIB_STATIC_API __declspec(dllimport) ZDICTLIB_VISIBLE 285*3117ece4Schristos # else 286*3117ece4Schristos # define ZDICTLIB_STATIC_API ZDICTLIB_VISIBLE 287*3117ece4Schristos # endif 288*3117ece4Schristos #endif 289*3117ece4Schristos 290*3117ece4Schristos /* ==================================================================================== 291*3117ece4Schristos * The definitions in this section are considered experimental. 292*3117ece4Schristos * They should never be used with a dynamic library, as they may change in the future. 293*3117ece4Schristos * They are provided for advanced usages. 294*3117ece4Schristos * Use them only in association with static linking. 295*3117ece4Schristos * ==================================================================================== */ 296*3117ece4Schristos 297*3117ece4Schristos #define ZDICT_DICTSIZE_MIN 256 298*3117ece4Schristos /* Deprecated: Remove in v1.6.0 */ 299*3117ece4Schristos #define ZDICT_CONTENTSIZE_MIN 128 300*3117ece4Schristos 301*3117ece4Schristos /*! ZDICT_cover_params_t: 302*3117ece4Schristos * k and d are the only required parameters. 303*3117ece4Schristos * For others, value 0 means default. 304*3117ece4Schristos */ 305*3117ece4Schristos typedef struct { 306*3117ece4Schristos unsigned k; /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */ 307*3117ece4Schristos unsigned d; /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */ 308*3117ece4Schristos unsigned steps; /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */ 309*3117ece4Schristos unsigned nbThreads; /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */ 310*3117ece4Schristos double splitPoint; /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (1.0), 1.0 when all samples are used for both training and testing */ 311*3117ece4Schristos unsigned shrinkDict; /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking */ 312*3117ece4Schristos unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */ 313*3117ece4Schristos ZDICT_params_t zParams; 314*3117ece4Schristos } ZDICT_cover_params_t; 315*3117ece4Schristos 316*3117ece4Schristos typedef struct { 317*3117ece4Schristos unsigned k; /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */ 318*3117ece4Schristos unsigned d; /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */ 319*3117ece4Schristos unsigned f; /* log of size of frequency array : constraint: 0 < f <= 31 : 1 means default(20)*/ 320*3117ece4Schristos unsigned steps; /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */ 321*3117ece4Schristos unsigned nbThreads; /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */ 322*3117ece4Schristos double splitPoint; /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (0.75), 1.0 when all samples are used for both training and testing */ 323*3117ece4Schristos unsigned accel; /* Acceleration level: constraint: 0 < accel <= 10, higher means faster and less accurate, 0 means default(1) */ 324*3117ece4Schristos unsigned shrinkDict; /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking */ 325*3117ece4Schristos unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */ 326*3117ece4Schristos 327*3117ece4Schristos ZDICT_params_t zParams; 328*3117ece4Schristos } ZDICT_fastCover_params_t; 329*3117ece4Schristos 330*3117ece4Schristos /*! ZDICT_trainFromBuffer_cover(): 331*3117ece4Schristos * Train a dictionary from an array of samples using the COVER algorithm. 332*3117ece4Schristos * Samples must be stored concatenated in a single flat buffer `samplesBuffer`, 333*3117ece4Schristos * supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order. 334*3117ece4Schristos * The resulting dictionary will be saved into `dictBuffer`. 335*3117ece4Schristos * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) 336*3117ece4Schristos * or an error code, which can be tested with ZDICT_isError(). 337*3117ece4Schristos * See ZDICT_trainFromBuffer() for details on failure modes. 338*3117ece4Schristos * Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte. 339*3117ece4Schristos * Tips: In general, a reasonable dictionary has a size of ~ 100 KB. 340*3117ece4Schristos * It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`. 341*3117ece4Schristos * In general, it's recommended to provide a few thousands samples, though this can vary a lot. 342*3117ece4Schristos * It's recommended that total size of all samples be about ~x100 times the target size of dictionary. 343*3117ece4Schristos */ 344*3117ece4Schristos ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_cover( 345*3117ece4Schristos void *dictBuffer, size_t dictBufferCapacity, 346*3117ece4Schristos const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples, 347*3117ece4Schristos ZDICT_cover_params_t parameters); 348*3117ece4Schristos 349*3117ece4Schristos /*! ZDICT_optimizeTrainFromBuffer_cover(): 350*3117ece4Schristos * The same requirements as above hold for all the parameters except `parameters`. 351*3117ece4Schristos * This function tries many parameter combinations and picks the best parameters. 352*3117ece4Schristos * `*parameters` is filled with the best parameters found, 353*3117ece4Schristos * dictionary constructed with those parameters is stored in `dictBuffer`. 354*3117ece4Schristos * 355*3117ece4Schristos * All of the parameters d, k, steps are optional. 356*3117ece4Schristos * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}. 357*3117ece4Schristos * if steps is zero it defaults to its default value. 358*3117ece4Schristos * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000]. 359*3117ece4Schristos * 360*3117ece4Schristos * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) 361*3117ece4Schristos * or an error code, which can be tested with ZDICT_isError(). 362*3117ece4Schristos * On success `*parameters` contains the parameters selected. 363*3117ece4Schristos * See ZDICT_trainFromBuffer() for details on failure modes. 364*3117ece4Schristos * 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. 365*3117ece4Schristos */ 366*3117ece4Schristos ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_cover( 367*3117ece4Schristos void* dictBuffer, size_t dictBufferCapacity, 368*3117ece4Schristos const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples, 369*3117ece4Schristos ZDICT_cover_params_t* parameters); 370*3117ece4Schristos 371*3117ece4Schristos /*! ZDICT_trainFromBuffer_fastCover(): 372*3117ece4Schristos * Train a dictionary from an array of samples using a modified version of COVER algorithm. 373*3117ece4Schristos * Samples must be stored concatenated in a single flat buffer `samplesBuffer`, 374*3117ece4Schristos * supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order. 375*3117ece4Schristos * d and k are required. 376*3117ece4Schristos * All other parameters are optional, will use default values if not provided 377*3117ece4Schristos * The resulting dictionary will be saved into `dictBuffer`. 378*3117ece4Schristos * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) 379*3117ece4Schristos * or an error code, which can be tested with ZDICT_isError(). 380*3117ece4Schristos * See ZDICT_trainFromBuffer() for details on failure modes. 381*3117ece4Schristos * Note: ZDICT_trainFromBuffer_fastCover() requires 6 * 2^f bytes of memory. 382*3117ece4Schristos * Tips: In general, a reasonable dictionary has a size of ~ 100 KB. 383*3117ece4Schristos * It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`. 384*3117ece4Schristos * In general, it's recommended to provide a few thousands samples, though this can vary a lot. 385*3117ece4Schristos * It's recommended that total size of all samples be about ~x100 times the target size of dictionary. 386*3117ece4Schristos */ 387*3117ece4Schristos ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer, 388*3117ece4Schristos size_t dictBufferCapacity, const void *samplesBuffer, 389*3117ece4Schristos const size_t *samplesSizes, unsigned nbSamples, 390*3117ece4Schristos ZDICT_fastCover_params_t parameters); 391*3117ece4Schristos 392*3117ece4Schristos /*! ZDICT_optimizeTrainFromBuffer_fastCover(): 393*3117ece4Schristos * The same requirements as above hold for all the parameters except `parameters`. 394*3117ece4Schristos * This function tries many parameter combinations (specifically, k and d combinations) 395*3117ece4Schristos * and picks the best parameters. `*parameters` is filled with the best parameters found, 396*3117ece4Schristos * dictionary constructed with those parameters is stored in `dictBuffer`. 397*3117ece4Schristos * All of the parameters d, k, steps, f, and accel are optional. 398*3117ece4Schristos * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}. 399*3117ece4Schristos * if steps is zero it defaults to its default value. 400*3117ece4Schristos * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000]. 401*3117ece4Schristos * If f is zero, default value of 20 is used. 402*3117ece4Schristos * If accel is zero, default value of 1 is used. 403*3117ece4Schristos * 404*3117ece4Schristos * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) 405*3117ece4Schristos * or an error code, which can be tested with ZDICT_isError(). 406*3117ece4Schristos * On success `*parameters` contains the parameters selected. 407*3117ece4Schristos * See ZDICT_trainFromBuffer() for details on failure modes. 408*3117ece4Schristos * Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 6 * 2^f bytes of memory for each thread. 409*3117ece4Schristos */ 410*3117ece4Schristos ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer, 411*3117ece4Schristos size_t dictBufferCapacity, const void* samplesBuffer, 412*3117ece4Schristos const size_t* samplesSizes, unsigned nbSamples, 413*3117ece4Schristos ZDICT_fastCover_params_t* parameters); 414*3117ece4Schristos 415*3117ece4Schristos typedef struct { 416*3117ece4Schristos unsigned selectivityLevel; /* 0 means default; larger => select more => larger dictionary */ 417*3117ece4Schristos ZDICT_params_t zParams; 418*3117ece4Schristos } ZDICT_legacy_params_t; 419*3117ece4Schristos 420*3117ece4Schristos /*! ZDICT_trainFromBuffer_legacy(): 421*3117ece4Schristos * Train a dictionary from an array of samples. 422*3117ece4Schristos * Samples must be stored concatenated in a single flat buffer `samplesBuffer`, 423*3117ece4Schristos * supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order. 424*3117ece4Schristos * The resulting dictionary will be saved into `dictBuffer`. 425*3117ece4Schristos * `parameters` is optional and can be provided with values set to 0 to mean "default". 426*3117ece4Schristos * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) 427*3117ece4Schristos * or an error code, which can be tested with ZDICT_isError(). 428*3117ece4Schristos * See ZDICT_trainFromBuffer() for details on failure modes. 429*3117ece4Schristos * Tips: In general, a reasonable dictionary has a size of ~ 100 KB. 430*3117ece4Schristos * It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`. 431*3117ece4Schristos * In general, it's recommended to provide a few thousands samples, though this can vary a lot. 432*3117ece4Schristos * It's recommended that total size of all samples be about ~x100 times the target size of dictionary. 433*3117ece4Schristos * Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0. 434*3117ece4Schristos */ 435*3117ece4Schristos ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_legacy( 436*3117ece4Schristos void* dictBuffer, size_t dictBufferCapacity, 437*3117ece4Schristos const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples, 438*3117ece4Schristos ZDICT_legacy_params_t parameters); 439*3117ece4Schristos 440*3117ece4Schristos 441*3117ece4Schristos /* Deprecation warnings */ 442*3117ece4Schristos /* It is generally possible to disable deprecation warnings from compiler, 443*3117ece4Schristos for example with -Wno-deprecated-declarations for gcc 444*3117ece4Schristos or _CRT_SECURE_NO_WARNINGS in Visual. 445*3117ece4Schristos Otherwise, it's also possible to manually define ZDICT_DISABLE_DEPRECATE_WARNINGS */ 446*3117ece4Schristos #ifdef ZDICT_DISABLE_DEPRECATE_WARNINGS 447*3117ece4Schristos # define ZDICT_DEPRECATED(message) /* disable deprecation warnings */ 448*3117ece4Schristos #else 449*3117ece4Schristos # define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__) 450*3117ece4Schristos # if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */ 451*3117ece4Schristos # define ZDICT_DEPRECATED(message) [[deprecated(message)]] 452*3117ece4Schristos # elif defined(__clang__) || (ZDICT_GCC_VERSION >= 405) 453*3117ece4Schristos # define ZDICT_DEPRECATED(message) __attribute__((deprecated(message))) 454*3117ece4Schristos # elif (ZDICT_GCC_VERSION >= 301) 455*3117ece4Schristos # define ZDICT_DEPRECATED(message) __attribute__((deprecated)) 456*3117ece4Schristos # elif defined(_MSC_VER) 457*3117ece4Schristos # define ZDICT_DEPRECATED(message) __declspec(deprecated(message)) 458*3117ece4Schristos # else 459*3117ece4Schristos # pragma message("WARNING: You need to implement ZDICT_DEPRECATED for this compiler") 460*3117ece4Schristos # define ZDICT_DEPRECATED(message) 461*3117ece4Schristos # endif 462*3117ece4Schristos #endif /* ZDICT_DISABLE_DEPRECATE_WARNINGS */ 463*3117ece4Schristos 464*3117ece4Schristos ZDICT_DEPRECATED("use ZDICT_finalizeDictionary() instead") 465*3117ece4Schristos ZDICTLIB_STATIC_API 466*3117ece4Schristos size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity, 467*3117ece4Schristos const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples); 468*3117ece4Schristos 469*3117ece4Schristos 470*3117ece4Schristos #endif /* ZSTD_ZDICT_H_STATIC */ 471*3117ece4Schristos 472*3117ece4Schristos #if defined (__cplusplus) 473*3117ece4Schristos } 474*3117ece4Schristos #endif 475