projects/linux·blob master · refs

278 lines · 13.8 KiB

1/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */
2/* ******************************************************************
3 * huff0 huffman codec,
4 * part of Finite State Entropy library
5 * Copyright (c) Meta Platforms, Inc. and affiliates.
6 *
7 * You can contact the author at :
8 * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
9 *
10 * This source code is licensed under both the BSD-style license (found in the
11 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
12 * in the COPYING file in the root directory of this source tree).
13 * You may select, at your option, one of the above-listed licenses.
14****************************************************************** */
15
16#ifndef HUF_H_298734234
17#define HUF_H_298734234
18
19/* *** Dependencies *** */
20#include "zstd_deps.h" /* size_t */
21#include "mem.h" /* U32 */
22#define FSE_STATIC_LINKING_ONLY
23#include "fse.h"
24
25/* *** Tool functions *** */
26#define HUF_BLOCKSIZE_MAX (128 * 1024) /*< maximum input size for a single block compressed with HUF_compress */
27size_t HUF_compressBound(size_t size); /*< maximum compressed size (worst case) */
28
29/* Error Management */
30unsigned HUF_isError(size_t code); /*< tells if a return value is an error code */
31const char* HUF_getErrorName(size_t code); /*< provides error code string (useful for debugging) */
32
33
34#define HUF_WORKSPACE_SIZE ((8 << 10) + 512 /* sorting scratch space */)
35#define HUF_WORKSPACE_SIZE_U64 (HUF_WORKSPACE_SIZE / sizeof(U64))
36
37/* *** Constants *** */
38#define HUF_TABLELOG_MAX 12 /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_TABLELOG_ABSOLUTEMAX */
39#define HUF_TABLELOG_DEFAULT 11 /* default tableLog value when none specified */
40#define HUF_SYMBOLVALUE_MAX 255
41
42#define HUF_TABLELOG_ABSOLUTEMAX 12 /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
43#if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
44# error "HUF_TABLELOG_MAX is too large !"
45#endif
46
47
48/* ****************************************
49* Static allocation
50******************************************/
51/* HUF buffer bounds */
52#define HUF_CTABLEBOUND 129
53#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8) /* only true when incompressible is pre-filtered with fast heuristic */
54#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size)) /* Macro version, useful for static allocation */
55
56/* static allocation of HUF's Compression Table */
57/* this is a private definition, just exposed for allocation and strict aliasing purpose. never EVER access its members directly */
58typedef size_t HUF_CElt; /* consider it an incomplete type */
59#define HUF_CTABLE_SIZE_ST(maxSymbolValue) ((maxSymbolValue)+2) /* Use tables of size_t, for proper alignment */
60#define HUF_CTABLE_SIZE(maxSymbolValue) (HUF_CTABLE_SIZE_ST(maxSymbolValue) * sizeof(size_t))
61#define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
62 HUF_CElt name[HUF_CTABLE_SIZE_ST(maxSymbolValue)] /* no final ; */
63
64/* static allocation of HUF's DTable */
65typedef U32 HUF_DTable;
66#define HUF_DTABLE_SIZE(maxTableLog) (1 + (1<<(maxTableLog)))
67#define HUF_CREATE_STATIC_DTABLEX1(DTable, maxTableLog) \
68 HUF_DTable DTable[HUF_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1) * 0x01000001) }
69#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
70 HUF_DTable DTable[HUF_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog) * 0x01000001) }
71
72
73/* ****************************************
74* Advanced decompression functions
75******************************************/
76
77/*
78 * Huffman flags bitset.
79 * For all flags, 0 is the default value.
80 */
81typedef enum {
82 /*
83 * If compiled with DYNAMIC_BMI2: Set flag only if the CPU supports BMI2 at runtime.
84 * Otherwise: Ignored.
85 */
86 HUF_flags_bmi2 = (1 << 0),
87 /*
88 * If set: Test possible table depths to find the one that produces the smallest header + encoded size.
89 * If unset: Use heuristic to find the table depth.
90 */
91 HUF_flags_optimalDepth = (1 << 1),
92 /*
93 * If set: If the previous table can encode the input, always reuse the previous table.
94 * If unset: If the previous table can encode the input, reuse the previous table if it results in a smaller output.
95 */
96 HUF_flags_preferRepeat = (1 << 2),
97 /*
98 * If set: Sample the input and check if the sample is uncompressible, if it is then don't attempt to compress.
99 * If unset: Always histogram the entire input.
100 */
101 HUF_flags_suspectUncompressible = (1 << 3),
102 /*
103 * If set: Don't use assembly implementations
104 * If unset: Allow using assembly implementations
105 */
106 HUF_flags_disableAsm = (1 << 4),
107 /*
108 * If set: Don't use the fast decoding loop, always use the fallback decoding loop.
109 * If unset: Use the fast decoding loop when possible.
110 */
111 HUF_flags_disableFast = (1 << 5)
112} HUF_flags_e;
113
114
115/* ****************************************
116 * HUF detailed API
117 * ****************************************/
118#define HUF_OPTIMAL_DEPTH_THRESHOLD ZSTD_btultra
119
120/*! HUF_compress() does the following:
121 * 1. count symbol occurrence from source[] into table count[] using FSE_count() (exposed within "fse.h")
122 * 2. (optional) refine tableLog using HUF_optimalTableLog()
123 * 3. build Huffman table from count using HUF_buildCTable()
124 * 4. save Huffman table to memory buffer using HUF_writeCTable()
125 * 5. encode the data stream using HUF_compress4X_usingCTable()
126 *
127 * The following API allows targeting specific sub-functions for advanced tasks.
128 * For example, it's possible to compress several blocks using the same 'CTable',
129 * or to save and regenerate 'CTable' using external methods.
130 */
131unsigned HUF_minTableLog(unsigned symbolCardinality);
132unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue);
133unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, void* workSpace,
134 size_t wkspSize, HUF_CElt* table, const unsigned* count, int flags); /* table is used as scratch space for building and testing tables, not a return value */
135size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog, void* workspace, size_t workspaceSize);
136size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
137size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
138int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
139
140typedef enum {
141 HUF_repeat_none, /*< Cannot use the previous table */
142 HUF_repeat_check, /*< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
143 HUF_repeat_valid /*< Can use the previous table and it is assumed to be valid */
144 } HUF_repeat;
145
146/* HUF_compress4X_repeat() :
147 * Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
148 * If it uses hufTable it does not modify hufTable or repeat.
149 * If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
150 * If preferRepeat then the old table will always be used if valid.
151 * If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
152size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
153 const void* src, size_t srcSize,
154 unsigned maxSymbolValue, unsigned tableLog,
155 void* workSpace, size_t wkspSize, /*< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
156 HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
157
158/* HUF_buildCTable_wksp() :
159 * Same as HUF_buildCTable(), but using externally allocated scratch buffer.
160 * `workSpace` must be aligned on 4-bytes boundaries, and its size must be >= HUF_CTABLE_WORKSPACE_SIZE.
161 */
162#define HUF_CTABLE_WORKSPACE_SIZE_U32 ((4 * (HUF_SYMBOLVALUE_MAX + 1)) + 192)
163#define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
164size_t HUF_buildCTable_wksp (HUF_CElt* tree,
165 const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
166 void* workSpace, size_t wkspSize);
167
168/*! HUF_readStats() :
169 * Read compact Huffman tree, saved by HUF_writeCTable().
170 * `huffWeight` is destination buffer.
171 * @return : size read from `src` , or an error Code .
172 * Note : Needed by HUF_readCTable() and HUF_readDTableXn() . */
173size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
174 U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
175 const void* src, size_t srcSize);
176
177/*! HUF_readStats_wksp() :
178 * Same as HUF_readStats() but takes an external workspace which must be
179 * 4-byte aligned and its size must be >= HUF_READ_STATS_WORKSPACE_SIZE.
180 * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
181 */
182#define HUF_READ_STATS_WORKSPACE_SIZE_U32 FSE_DECOMPRESS_WKSP_SIZE_U32(6, HUF_TABLELOG_MAX-1)
183#define HUF_READ_STATS_WORKSPACE_SIZE (HUF_READ_STATS_WORKSPACE_SIZE_U32 * sizeof(unsigned))
184size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
185 U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
186 const void* src, size_t srcSize,
187 void* workspace, size_t wkspSize,
188 int flags);
189
190/* HUF_readCTable() :
191 * Loading a CTable saved with HUF_writeCTable() */
192size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
193
194/* HUF_getNbBitsFromCTable() :
195 * Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
196 * Note 1 : If symbolValue > HUF_readCTableHeader(symbolTable).maxSymbolValue, returns 0
197 * Note 2 : is not inlined, as HUF_CElt definition is private
198 */
199U32 HUF_getNbBitsFromCTable(const HUF_CElt* symbolTable, U32 symbolValue);
200
201typedef struct {
202 BYTE tableLog;
203 BYTE maxSymbolValue;
204 BYTE unused[sizeof(size_t) - 2];
205} HUF_CTableHeader;
206
207/* HUF_readCTableHeader() :
208 * @returns The header from the CTable specifying the tableLog and the maxSymbolValue.
209 */
210HUF_CTableHeader HUF_readCTableHeader(HUF_CElt const* ctable);
211
212/*
213 * HUF_decompress() does the following:
214 * 1. select the decompression algorithm (X1, X2) based on pre-computed heuristics
215 * 2. build Huffman table from save, using HUF_readDTableX?()
216 * 3. decode 1 or 4 segments in parallel using HUF_decompress?X?_usingDTable()
217 */
218
219/* HUF_selectDecoder() :
220 * Tells which decoder is likely to decode faster,
221 * based on a set of pre-computed metrics.
222 * @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
223 * Assumption : 0 < dstSize <= 128 KB */
224U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
225
226/*
227 * The minimum workspace size for the `workSpace` used in
228 * HUF_readDTableX1_wksp() and HUF_readDTableX2_wksp().
229 *
230 * The space used depends on HUF_TABLELOG_MAX, ranging from ~1500 bytes when
231 * HUF_TABLE_LOG_MAX=12 to ~1850 bytes when HUF_TABLE_LOG_MAX=15.
232 * Buffer overflow errors may potentially occur if code modifications result in
233 * a required workspace size greater than that specified in the following
234 * macro.
235 */
236#define HUF_DECOMPRESS_WORKSPACE_SIZE ((2 << 10) + (1 << 9))
237#define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
238
239
240/* ====================== */
241/* single stream variants */
242/* ====================== */
243
244size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
245/* HUF_compress1X_repeat() :
246 * Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
247 * If it uses hufTable it does not modify hufTable or repeat.
248 * If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
249 * If preferRepeat then the old table will always be used if valid.
250 * If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
251size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
252 const void* src, size_t srcSize,
253 unsigned maxSymbolValue, unsigned tableLog,
254 void* workSpace, size_t wkspSize, /*< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
255 HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
256
257size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
258#ifndef HUF_FORCE_DECOMPRESS_X1
259size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags); /*< double-symbols decoder */
260#endif
261
262/* BMI2 variants.
263 * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
264 */
265size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
266#ifndef HUF_FORCE_DECOMPRESS_X2
267size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
268#endif
269size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
270size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
271#ifndef HUF_FORCE_DECOMPRESS_X2
272size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
273#endif
274#ifndef HUF_FORCE_DECOMPRESS_X1
275size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
276#endif
277
278#endif /* HUF_H_298734234 */