jpayne@69: /*
jpayne@69:  * Copyright (c) Meta Platforms, Inc. and affiliates.
jpayne@69:  * All rights reserved.
jpayne@69:  *
jpayne@69:  * This source code is licensed under both the BSD-style license (found in the
jpayne@69:  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
jpayne@69:  * in the COPYING file in the root directory of this source tree).
jpayne@69:  * You may select, at your option, one of the above-listed licenses.
jpayne@69:  */
jpayne@69: 
jpayne@69: #if defined (__cplusplus)
jpayne@69: extern "C" {
jpayne@69: #endif
jpayne@69: 
jpayne@69: #ifndef ZSTD_ZDICT_H
jpayne@69: #define ZSTD_ZDICT_H
jpayne@69: 
jpayne@69: /*======  Dependencies  ======*/
jpayne@69: #include <stddef.h>  /* size_t */
jpayne@69: 
jpayne@69: 
jpayne@69: /* =====   ZDICTLIB_API : control library symbols visibility   ===== */
jpayne@69: #ifndef ZDICTLIB_VISIBLE
jpayne@69:    /* Backwards compatibility with old macro name */
jpayne@69: #  ifdef ZDICTLIB_VISIBILITY
jpayne@69: #    define ZDICTLIB_VISIBLE ZDICTLIB_VISIBILITY
jpayne@69: #  elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
jpayne@69: #    define ZDICTLIB_VISIBLE __attribute__ ((visibility ("default")))
jpayne@69: #  else
jpayne@69: #    define ZDICTLIB_VISIBLE
jpayne@69: #  endif
jpayne@69: #endif
jpayne@69: 
jpayne@69: #ifndef ZDICTLIB_HIDDEN
jpayne@69: #  if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
jpayne@69: #    define ZDICTLIB_HIDDEN __attribute__ ((visibility ("hidden")))
jpayne@69: #  else
jpayne@69: #    define ZDICTLIB_HIDDEN
jpayne@69: #  endif
jpayne@69: #endif
jpayne@69: 
jpayne@69: #if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
jpayne@69: #  define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBLE
jpayne@69: #elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
jpayne@69: #  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.*/
jpayne@69: #else
jpayne@69: #  define ZDICTLIB_API ZDICTLIB_VISIBLE
jpayne@69: #endif
jpayne@69: 
jpayne@69: /*******************************************************************************
jpayne@69:  * Zstd dictionary builder
jpayne@69:  *
jpayne@69:  * FAQ
jpayne@69:  * ===
jpayne@69:  * Why should I use a dictionary?
jpayne@69:  * ------------------------------
jpayne@69:  *
jpayne@69:  * Zstd can use dictionaries to improve compression ratio of small data.
jpayne@69:  * Traditionally small files don't compress well because there is very little
jpayne@69:  * repetition in a single sample, since it is small. But, if you are compressing
jpayne@69:  * many similar files, like a bunch of JSON records that share the same
jpayne@69:  * structure, you can train a dictionary on ahead of time on some samples of
jpayne@69:  * these files. Then, zstd can use the dictionary to find repetitions that are
jpayne@69:  * present across samples. This can vastly improve compression ratio.
jpayne@69:  *
jpayne@69:  * When is a dictionary useful?
jpayne@69:  * ----------------------------
jpayne@69:  *
jpayne@69:  * Dictionaries are useful when compressing many small files that are similar.
jpayne@69:  * The larger a file is, the less benefit a dictionary will have. Generally,
jpayne@69:  * we don't expect dictionary compression to be effective past 100KB. And the
jpayne@69:  * smaller a file is, the more we would expect the dictionary to help.
jpayne@69:  *
jpayne@69:  * How do I use a dictionary?
jpayne@69:  * --------------------------
jpayne@69:  *
jpayne@69:  * Simply pass the dictionary to the zstd compressor with
jpayne@69:  * `ZSTD_CCtx_loadDictionary()`. The same dictionary must then be passed to
jpayne@69:  * the decompressor, using `ZSTD_DCtx_loadDictionary()`. There are other
jpayne@69:  * more advanced functions that allow selecting some options, see zstd.h for
jpayne@69:  * complete documentation.
jpayne@69:  *
jpayne@69:  * What is a zstd dictionary?
jpayne@69:  * --------------------------
jpayne@69:  *
jpayne@69:  * A zstd dictionary has two pieces: Its header, and its content. The header
jpayne@69:  * contains a magic number, the dictionary ID, and entropy tables. These
jpayne@69:  * entropy tables allow zstd to save on header costs in the compressed file,
jpayne@69:  * which really matters for small data. The content is just bytes, which are
jpayne@69:  * repeated content that is common across many samples.
jpayne@69:  *
jpayne@69:  * What is a raw content dictionary?
jpayne@69:  * ---------------------------------
jpayne@69:  *
jpayne@69:  * A raw content dictionary is just bytes. It doesn't have a zstd dictionary
jpayne@69:  * header, a dictionary ID, or entropy tables. Any buffer is a valid raw
jpayne@69:  * content dictionary.
jpayne@69:  *
jpayne@69:  * How do I train a dictionary?
jpayne@69:  * ----------------------------
jpayne@69:  *
jpayne@69:  * Gather samples from your use case. These samples should be similar to each
jpayne@69:  * other. If you have several use cases, you could try to train one dictionary
jpayne@69:  * per use case.
jpayne@69:  *
jpayne@69:  * Pass those samples to `ZDICT_trainFromBuffer()` and that will train your
jpayne@69:  * dictionary. There are a few advanced versions of this function, but this
jpayne@69:  * is a great starting point. If you want to further tune your dictionary
jpayne@69:  * you could try `ZDICT_optimizeTrainFromBuffer_cover()`. If that is too slow
jpayne@69:  * you can try `ZDICT_optimizeTrainFromBuffer_fastCover()`.
jpayne@69:  *
jpayne@69:  * If the dictionary training function fails, that is likely because you
jpayne@69:  * either passed too few samples, or a dictionary would not be effective
jpayne@69:  * for your data. Look at the messages that the dictionary trainer printed,
jpayne@69:  * if it doesn't say too few samples, then a dictionary would not be effective.
jpayne@69:  *
jpayne@69:  * How large should my dictionary be?
jpayne@69:  * ----------------------------------
jpayne@69:  *
jpayne@69:  * A reasonable dictionary size, the `dictBufferCapacity`, is about 100KB.
jpayne@69:  * The zstd CLI defaults to a 110KB dictionary. You likely don't need a
jpayne@69:  * dictionary larger than that. But, most use cases can get away with a
jpayne@69:  * smaller dictionary. The advanced dictionary builders can automatically
jpayne@69:  * shrink the dictionary for you, and select the smallest size that doesn't
jpayne@69:  * hurt compression ratio too much. See the `shrinkDict` parameter.
jpayne@69:  * A smaller dictionary can save memory, and potentially speed up
jpayne@69:  * compression.
jpayne@69:  *
jpayne@69:  * How many samples should I provide to the dictionary builder?
jpayne@69:  * ------------------------------------------------------------
jpayne@69:  *
jpayne@69:  * We generally recommend passing ~100x the size of the dictionary
jpayne@69:  * in samples. A few thousand should suffice. Having too few samples
jpayne@69:  * can hurt the dictionaries effectiveness. Having more samples will
jpayne@69:  * only improve the dictionaries effectiveness. But having too many
jpayne@69:  * samples can slow down the dictionary builder.
jpayne@69:  *
jpayne@69:  * How do I determine if a dictionary will be effective?
jpayne@69:  * -----------------------------------------------------
jpayne@69:  *
jpayne@69:  * Simply train a dictionary and try it out. You can use zstd's built in
jpayne@69:  * benchmarking tool to test the dictionary effectiveness.
jpayne@69:  *
jpayne@69:  *   # Benchmark levels 1-3 without a dictionary
jpayne@69:  *   zstd -b1e3 -r /path/to/my/files
jpayne@69:  *   # Benchmark levels 1-3 with a dictionary
jpayne@69:  *   zstd -b1e3 -r /path/to/my/files -D /path/to/my/dictionary
jpayne@69:  *
jpayne@69:  * When should I retrain a dictionary?
jpayne@69:  * -----------------------------------
jpayne@69:  *
jpayne@69:  * You should retrain a dictionary when its effectiveness drops. Dictionary
jpayne@69:  * effectiveness drops as the data you are compressing changes. Generally, we do
jpayne@69:  * expect dictionaries to "decay" over time, as your data changes, but the rate
jpayne@69:  * at which they decay depends on your use case. Internally, we regularly
jpayne@69:  * retrain dictionaries, and if the new dictionary performs significantly
jpayne@69:  * better than the old dictionary, we will ship the new dictionary.
jpayne@69:  *
jpayne@69:  * I have a raw content dictionary, how do I turn it into a zstd dictionary?
jpayne@69:  * -------------------------------------------------------------------------
jpayne@69:  *
jpayne@69:  * If you have a raw content dictionary, e.g. by manually constructing it, or
jpayne@69:  * using a third-party dictionary builder, you can turn it into a zstd
jpayne@69:  * dictionary by using `ZDICT_finalizeDictionary()`. You'll also have to
jpayne@69:  * provide some samples of the data. It will add the zstd header to the
jpayne@69:  * raw content, which contains a dictionary ID and entropy tables, which
jpayne@69:  * will improve compression ratio, and allow zstd to write the dictionary ID
jpayne@69:  * into the frame, if you so choose.
jpayne@69:  *
jpayne@69:  * Do I have to use zstd's dictionary builder?
jpayne@69:  * -------------------------------------------
jpayne@69:  *
jpayne@69:  * No! You can construct dictionary content however you please, it is just
jpayne@69:  * bytes. It will always be valid as a raw content dictionary. If you want
jpayne@69:  * a zstd dictionary, which can improve compression ratio, use
jpayne@69:  * `ZDICT_finalizeDictionary()`.
jpayne@69:  *
jpayne@69:  * What is the attack surface of a zstd dictionary?
jpayne@69:  * ------------------------------------------------
jpayne@69:  *
jpayne@69:  * Zstd is heavily fuzz tested, including loading fuzzed dictionaries, so
jpayne@69:  * zstd should never crash, or access out-of-bounds memory no matter what
jpayne@69:  * the dictionary is. However, if an attacker can control the dictionary
jpayne@69:  * during decompression, they can cause zstd to generate arbitrary bytes,
jpayne@69:  * just like if they controlled the compressed data.
jpayne@69:  *
jpayne@69:  ******************************************************************************/
jpayne@69: 
jpayne@69: 
jpayne@69: /*! ZDICT_trainFromBuffer():
jpayne@69:  *  Train a dictionary from an array of samples.
jpayne@69:  *  Redirect towards ZDICT_optimizeTrainFromBuffer_fastCover() single-threaded, with d=8, steps=4,
jpayne@69:  *  f=20, and accel=1.
jpayne@69:  *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
jpayne@69:  *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
jpayne@69:  *  The resulting dictionary will be saved into `dictBuffer`.
jpayne@69:  * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
jpayne@69:  *          or an error code, which can be tested with ZDICT_isError().
jpayne@69:  *  Note:  Dictionary training will fail if there are not enough samples to construct a
jpayne@69:  *         dictionary, or if most of the samples are too small (< 8 bytes being the lower limit).
jpayne@69:  *         If dictionary training fails, you should use zstd without a dictionary, as the dictionary
jpayne@69:  *         would've been ineffective anyways. If you believe your samples would benefit from a dictionary
jpayne@69:  *         please open an issue with details, and we can look into it.
jpayne@69:  *  Note: ZDICT_trainFromBuffer()'s memory usage is about 6 MB.
jpayne@69:  *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
jpayne@69:  *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
jpayne@69:  *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
jpayne@69:  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
jpayne@69:  */
jpayne@69: ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
jpayne@69:                                     const void* samplesBuffer,
jpayne@69:                                     const size_t* samplesSizes, unsigned nbSamples);
jpayne@69: 
jpayne@69: typedef struct {
jpayne@69:     int      compressionLevel;   /**< optimize for a specific zstd compression level; 0 means default */
jpayne@69:     unsigned notificationLevel;  /**< Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
jpayne@69:     unsigned dictID;             /**< force dictID value; 0 means auto mode (32-bits random value)
jpayne@69:                                   *   NOTE: The zstd format reserves some dictionary IDs for future use.
jpayne@69:                                   *         You may use them in private settings, but be warned that they
jpayne@69:                                   *         may be used by zstd in a public dictionary registry in the future.
jpayne@69:                                   *         These dictionary IDs are:
jpayne@69:                                   *           - low range  : <= 32767
jpayne@69:                                   *           - high range : >= (2^31)
jpayne@69:                                   */
jpayne@69: } ZDICT_params_t;
jpayne@69: 
jpayne@69: /*! ZDICT_finalizeDictionary():
jpayne@69:  * Given a custom content as a basis for dictionary, and a set of samples,
jpayne@69:  * finalize dictionary by adding headers and statistics according to the zstd
jpayne@69:  * dictionary format.
jpayne@69:  *
jpayne@69:  * Samples must be stored concatenated in a flat buffer `samplesBuffer`,
jpayne@69:  * supplied with an array of sizes `samplesSizes`, providing the size of each
jpayne@69:  * sample in order. The samples are used to construct the statistics, so they
jpayne@69:  * should be representative of what you will compress with this dictionary.
jpayne@69:  *
jpayne@69:  * The compression level can be set in `parameters`. You should pass the
jpayne@69:  * compression level you expect to use in production. The statistics for each
jpayne@69:  * compression level differ, so tuning the dictionary for the compression level
jpayne@69:  * can help quite a bit.
jpayne@69:  *
jpayne@69:  * You can set an explicit dictionary ID in `parameters`, or allow us to pick
jpayne@69:  * a random dictionary ID for you, but we can't guarantee no collisions.
jpayne@69:  *
jpayne@69:  * The dstDictBuffer and the dictContent may overlap, and the content will be
jpayne@69:  * appended to the end of the header. If the header + the content doesn't fit in
jpayne@69:  * maxDictSize the beginning of the content is truncated to make room, since it
jpayne@69:  * is presumed that the most profitable content is at the end of the dictionary,
jpayne@69:  * since that is the cheapest to reference.
jpayne@69:  *
jpayne@69:  * `maxDictSize` must be >= max(dictContentSize, ZSTD_DICTSIZE_MIN).
jpayne@69:  *
jpayne@69:  * @return: size of dictionary stored into `dstDictBuffer` (<= `maxDictSize`),
jpayne@69:  *          or an error code, which can be tested by ZDICT_isError().
jpayne@69:  * Note: ZDICT_finalizeDictionary() will push notifications into stderr if
jpayne@69:  *       instructed to, using notificationLevel>0.
jpayne@69:  * NOTE: This function currently may fail in several edge cases including:
jpayne@69:  *         * Not enough samples
jpayne@69:  *         * Samples are uncompressible
jpayne@69:  *         * Samples are all exactly the same
jpayne@69:  */
jpayne@69: ZDICTLIB_API size_t ZDICT_finalizeDictionary(void* dstDictBuffer, size_t maxDictSize,
jpayne@69:                                 const void* dictContent, size_t dictContentSize,
jpayne@69:                                 const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
jpayne@69:                                 ZDICT_params_t parameters);
jpayne@69: 
jpayne@69: 
jpayne@69: /*======   Helper functions   ======*/
jpayne@69: ZDICTLIB_API unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize);  /**< extracts dictID; @return zero if error (not a valid dictionary) */
jpayne@69: ZDICTLIB_API size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize);  /* returns dict header size; returns a ZSTD error code on failure */
jpayne@69: ZDICTLIB_API unsigned ZDICT_isError(size_t errorCode);
jpayne@69: ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
jpayne@69: 
jpayne@69: #endif   /* ZSTD_ZDICT_H */
jpayne@69: 
jpayne@69: #if defined(ZDICT_STATIC_LINKING_ONLY) && !defined(ZSTD_ZDICT_H_STATIC)
jpayne@69: #define ZSTD_ZDICT_H_STATIC
jpayne@69: 
jpayne@69: /* This can be overridden externally to hide static symbols. */
jpayne@69: #ifndef ZDICTLIB_STATIC_API
jpayne@69: #  if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
jpayne@69: #    define ZDICTLIB_STATIC_API __declspec(dllexport) ZDICTLIB_VISIBLE
jpayne@69: #  elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
jpayne@69: #    define ZDICTLIB_STATIC_API __declspec(dllimport) ZDICTLIB_VISIBLE
jpayne@69: #  else
jpayne@69: #    define ZDICTLIB_STATIC_API ZDICTLIB_VISIBLE
jpayne@69: #  endif
jpayne@69: #endif
jpayne@69: 
jpayne@69: /* ====================================================================================
jpayne@69:  * The definitions in this section are considered experimental.
jpayne@69:  * They should never be used with a dynamic library, as they may change in the future.
jpayne@69:  * They are provided for advanced usages.
jpayne@69:  * Use them only in association with static linking.
jpayne@69:  * ==================================================================================== */
jpayne@69: 
jpayne@69: #define ZDICT_DICTSIZE_MIN    256
jpayne@69: /* Deprecated: Remove in v1.6.0 */
jpayne@69: #define ZDICT_CONTENTSIZE_MIN 128
jpayne@69: 
jpayne@69: /*! ZDICT_cover_params_t:
jpayne@69:  *  k and d are the only required parameters.
jpayne@69:  *  For others, value 0 means default.
jpayne@69:  */
jpayne@69: typedef struct {
jpayne@69:     unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
jpayne@69:     unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
jpayne@69:     unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
jpayne@69:     unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
jpayne@69:     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 */
jpayne@69:     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  */
jpayne@69:     unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */
jpayne@69:     ZDICT_params_t zParams;
jpayne@69: } ZDICT_cover_params_t;
jpayne@69: 
jpayne@69: typedef struct {
jpayne@69:     unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
jpayne@69:     unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
jpayne@69:     unsigned f;                  /* log of size of frequency array : constraint: 0 < f <= 31 : 1 means default(20)*/
jpayne@69:     unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
jpayne@69:     unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
jpayne@69:     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 */
jpayne@69:     unsigned accel;              /* Acceleration level: constraint: 0 < accel <= 10, higher means faster and less accurate, 0 means default(1) */
jpayne@69:     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  */
jpayne@69:     unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */
jpayne@69: 
jpayne@69:     ZDICT_params_t zParams;
jpayne@69: } ZDICT_fastCover_params_t;
jpayne@69: 
jpayne@69: /*! ZDICT_trainFromBuffer_cover():
jpayne@69:  *  Train a dictionary from an array of samples using the COVER algorithm.
jpayne@69:  *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
jpayne@69:  *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
jpayne@69:  *  The resulting dictionary will be saved into `dictBuffer`.
jpayne@69:  * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
jpayne@69:  *          or an error code, which can be tested with ZDICT_isError().
jpayne@69:  *          See ZDICT_trainFromBuffer() for details on failure modes.
jpayne@69:  *  Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte.
jpayne@69:  *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
jpayne@69:  *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
jpayne@69:  *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
jpayne@69:  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
jpayne@69:  */
jpayne@69: ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_cover(
jpayne@69:           void *dictBuffer, size_t dictBufferCapacity,
jpayne@69:     const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
jpayne@69:           ZDICT_cover_params_t parameters);
jpayne@69: 
jpayne@69: /*! ZDICT_optimizeTrainFromBuffer_cover():
jpayne@69:  * The same requirements as above hold for all the parameters except `parameters`.
jpayne@69:  * This function tries many parameter combinations and picks the best parameters.
jpayne@69:  * `*parameters` is filled with the best parameters found,
jpayne@69:  * dictionary constructed with those parameters is stored in `dictBuffer`.
jpayne@69:  *
jpayne@69:  * All of the parameters d, k, steps are optional.
jpayne@69:  * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
jpayne@69:  * if steps is zero it defaults to its default value.
jpayne@69:  * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
jpayne@69:  *
jpayne@69:  * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
jpayne@69:  *          or an error code, which can be tested with ZDICT_isError().
jpayne@69:  *          On success `*parameters` contains the parameters selected.
jpayne@69:  *          See ZDICT_trainFromBuffer() for details on failure modes.
jpayne@69:  * 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.
jpayne@69:  */
jpayne@69: ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_cover(
jpayne@69:           void* dictBuffer, size_t dictBufferCapacity,
jpayne@69:     const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
jpayne@69:           ZDICT_cover_params_t* parameters);
jpayne@69: 
jpayne@69: /*! ZDICT_trainFromBuffer_fastCover():
jpayne@69:  *  Train a dictionary from an array of samples using a modified version of COVER algorithm.
jpayne@69:  *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
jpayne@69:  *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
jpayne@69:  *  d and k are required.
jpayne@69:  *  All other parameters are optional, will use default values if not provided
jpayne@69:  *  The resulting dictionary will be saved into `dictBuffer`.
jpayne@69:  * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
jpayne@69:  *          or an error code, which can be tested with ZDICT_isError().
jpayne@69:  *          See ZDICT_trainFromBuffer() for details on failure modes.
jpayne@69:  *  Note: ZDICT_trainFromBuffer_fastCover() requires 6 * 2^f bytes of memory.
jpayne@69:  *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
jpayne@69:  *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
jpayne@69:  *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
jpayne@69:  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
jpayne@69:  */
jpayne@69: ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer,
jpayne@69:                     size_t dictBufferCapacity, const void *samplesBuffer,
jpayne@69:                     const size_t *samplesSizes, unsigned nbSamples,
jpayne@69:                     ZDICT_fastCover_params_t parameters);
jpayne@69: 
jpayne@69: /*! ZDICT_optimizeTrainFromBuffer_fastCover():
jpayne@69:  * The same requirements as above hold for all the parameters except `parameters`.
jpayne@69:  * This function tries many parameter combinations (specifically, k and d combinations)
jpayne@69:  * and picks the best parameters. `*parameters` is filled with the best parameters found,
jpayne@69:  * dictionary constructed with those parameters is stored in `dictBuffer`.
jpayne@69:  * All of the parameters d, k, steps, f, and accel are optional.
jpayne@69:  * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
jpayne@69:  * if steps is zero it defaults to its default value.
jpayne@69:  * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
jpayne@69:  * If f is zero, default value of 20 is used.
jpayne@69:  * If accel is zero, default value of 1 is used.
jpayne@69:  *
jpayne@69:  * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
jpayne@69:  *          or an error code, which can be tested with ZDICT_isError().
jpayne@69:  *          On success `*parameters` contains the parameters selected.
jpayne@69:  *          See ZDICT_trainFromBuffer() for details on failure modes.
jpayne@69:  * Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 6 * 2^f bytes of memory for each thread.
jpayne@69:  */
jpayne@69: ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer,
jpayne@69:                     size_t dictBufferCapacity, const void* samplesBuffer,
jpayne@69:                     const size_t* samplesSizes, unsigned nbSamples,
jpayne@69:                     ZDICT_fastCover_params_t* parameters);
jpayne@69: 
jpayne@69: typedef struct {
jpayne@69:     unsigned selectivityLevel;   /* 0 means default; larger => select more => larger dictionary */
jpayne@69:     ZDICT_params_t zParams;
jpayne@69: } ZDICT_legacy_params_t;
jpayne@69: 
jpayne@69: /*! ZDICT_trainFromBuffer_legacy():
jpayne@69:  *  Train a dictionary from an array of samples.
jpayne@69:  *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
jpayne@69:  *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
jpayne@69:  *  The resulting dictionary will be saved into `dictBuffer`.
jpayne@69:  * `parameters` is optional and can be provided with values set to 0 to mean "default".
jpayne@69:  * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
jpayne@69:  *          or an error code, which can be tested with ZDICT_isError().
jpayne@69:  *          See ZDICT_trainFromBuffer() for details on failure modes.
jpayne@69:  *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
jpayne@69:  *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
jpayne@69:  *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
jpayne@69:  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
jpayne@69:  *  Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
jpayne@69:  */
jpayne@69: ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_legacy(
jpayne@69:     void* dictBuffer, size_t dictBufferCapacity,
jpayne@69:     const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
jpayne@69:     ZDICT_legacy_params_t parameters);
jpayne@69: 
jpayne@69: 
jpayne@69: /* Deprecation warnings */
jpayne@69: /* It is generally possible to disable deprecation warnings from compiler,
jpayne@69:    for example with -Wno-deprecated-declarations for gcc
jpayne@69:    or _CRT_SECURE_NO_WARNINGS in Visual.
jpayne@69:    Otherwise, it's also possible to manually define ZDICT_DISABLE_DEPRECATE_WARNINGS */
jpayne@69: #ifdef ZDICT_DISABLE_DEPRECATE_WARNINGS
jpayne@69: #  define ZDICT_DEPRECATED(message) /* disable deprecation warnings */
jpayne@69: #else
jpayne@69: #  define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
jpayne@69: #  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
jpayne@69: #    define ZDICT_DEPRECATED(message) [[deprecated(message)]]
jpayne@69: #  elif defined(__clang__) || (ZDICT_GCC_VERSION >= 405)
jpayne@69: #    define ZDICT_DEPRECATED(message) __attribute__((deprecated(message)))
jpayne@69: #  elif (ZDICT_GCC_VERSION >= 301)
jpayne@69: #    define ZDICT_DEPRECATED(message) __attribute__((deprecated))
jpayne@69: #  elif defined(_MSC_VER)
jpayne@69: #    define ZDICT_DEPRECATED(message) __declspec(deprecated(message))
jpayne@69: #  else
jpayne@69: #    pragma message("WARNING: You need to implement ZDICT_DEPRECATED for this compiler")
jpayne@69: #    define ZDICT_DEPRECATED(message)
jpayne@69: #  endif
jpayne@69: #endif /* ZDICT_DISABLE_DEPRECATE_WARNINGS */
jpayne@69: 
jpayne@69: ZDICT_DEPRECATED("use ZDICT_finalizeDictionary() instead")
jpayne@69: ZDICTLIB_STATIC_API
jpayne@69: size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
jpayne@69:                                   const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
jpayne@69: 
jpayne@69: 
jpayne@69: #endif   /* ZSTD_ZDICT_H_STATIC */
jpayne@69: 
jpayne@69: #if defined (__cplusplus)
jpayne@69: }
jpayne@69: #endif