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