1/* Copyright 2013 Google Inc. All Rights Reserved.
2
3 Distributed under MIT license.
4 See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
5*/
6
7/* Utilities for building Huffman decoding tables. */
8
9#include "huffman.h"
10
11#include <string.h> /* memcpy, memset */
12
13#include <brotli/types.h>
14
15#include "../common/constants.h"
16#include "../common/platform.h"
17
18#if defined(__cplusplus) || defined(c_plusplus)
19extern "C" {
20#endif
21
22#define BROTLI_REVERSE_BITS_MAX 8
23
24#if defined(BROTLI_RBIT)
25#define BROTLI_REVERSE_BITS_BASE \
26 ((sizeof(brotli_reg_t) << 3) - BROTLI_REVERSE_BITS_MAX)
27#else
28#define BROTLI_REVERSE_BITS_BASE 0
29static uint8_t kReverseBits[1 << BROTLI_REVERSE_BITS_MAX] = {
30 0x00, 0x80, 0x40, 0xC0, 0x20, 0xA0, 0x60, 0xE0,
31 0x10, 0x90, 0x50, 0xD0, 0x30, 0xB0, 0x70, 0xF0,
32 0x08, 0x88, 0x48, 0xC8, 0x28, 0xA8, 0x68, 0xE8,
33 0x18, 0x98, 0x58, 0xD8, 0x38, 0xB8, 0x78, 0xF8,
34 0x04, 0x84, 0x44, 0xC4, 0x24, 0xA4, 0x64, 0xE4,
35 0x14, 0x94, 0x54, 0xD4, 0x34, 0xB4, 0x74, 0xF4,
36 0x0C, 0x8C, 0x4C, 0xCC, 0x2C, 0xAC, 0x6C, 0xEC,
37 0x1C, 0x9C, 0x5C, 0xDC, 0x3C, 0xBC, 0x7C, 0xFC,
38 0x02, 0x82, 0x42, 0xC2, 0x22, 0xA2, 0x62, 0xE2,
39 0x12, 0x92, 0x52, 0xD2, 0x32, 0xB2, 0x72, 0xF2,
40 0x0A, 0x8A, 0x4A, 0xCA, 0x2A, 0xAA, 0x6A, 0xEA,
41 0x1A, 0x9A, 0x5A, 0xDA, 0x3A, 0xBA, 0x7A, 0xFA,
42 0x06, 0x86, 0x46, 0xC6, 0x26, 0xA6, 0x66, 0xE6,
43 0x16, 0x96, 0x56, 0xD6, 0x36, 0xB6, 0x76, 0xF6,
44 0x0E, 0x8E, 0x4E, 0xCE, 0x2E, 0xAE, 0x6E, 0xEE,
45 0x1E, 0x9E, 0x5E, 0xDE, 0x3E, 0xBE, 0x7E, 0xFE,
46 0x01, 0x81, 0x41, 0xC1, 0x21, 0xA1, 0x61, 0xE1,
47 0x11, 0x91, 0x51, 0xD1, 0x31, 0xB1, 0x71, 0xF1,
48 0x09, 0x89, 0x49, 0xC9, 0x29, 0xA9, 0x69, 0xE9,
49 0x19, 0x99, 0x59, 0xD9, 0x39, 0xB9, 0x79, 0xF9,
50 0x05, 0x85, 0x45, 0xC5, 0x25, 0xA5, 0x65, 0xE5,
51 0x15, 0x95, 0x55, 0xD5, 0x35, 0xB5, 0x75, 0xF5,
52 0x0D, 0x8D, 0x4D, 0xCD, 0x2D, 0xAD, 0x6D, 0xED,
53 0x1D, 0x9D, 0x5D, 0xDD, 0x3D, 0xBD, 0x7D, 0xFD,
54 0x03, 0x83, 0x43, 0xC3, 0x23, 0xA3, 0x63, 0xE3,
55 0x13, 0x93, 0x53, 0xD3, 0x33, 0xB3, 0x73, 0xF3,
56 0x0B, 0x8B, 0x4B, 0xCB, 0x2B, 0xAB, 0x6B, 0xEB,
57 0x1B, 0x9B, 0x5B, 0xDB, 0x3B, 0xBB, 0x7B, 0xFB,
58 0x07, 0x87, 0x47, 0xC7, 0x27, 0xA7, 0x67, 0xE7,
59 0x17, 0x97, 0x57, 0xD7, 0x37, 0xB7, 0x77, 0xF7,
60 0x0F, 0x8F, 0x4F, 0xCF, 0x2F, 0xAF, 0x6F, 0xEF,
61 0x1F, 0x9F, 0x5F, 0xDF, 0x3F, 0xBF, 0x7F, 0xFF
62};
63#endif /* BROTLI_RBIT */
64
65#define BROTLI_REVERSE_BITS_LOWEST \
66 ((brotli_reg_t)1 << (BROTLI_REVERSE_BITS_MAX - 1 + BROTLI_REVERSE_BITS_BASE))
67
68/* Returns reverse(num >> BROTLI_REVERSE_BITS_BASE, BROTLI_REVERSE_BITS_MAX),
69 where reverse(value, len) is the bit-wise reversal of the len least
70 significant bits of value. */
71static BROTLI_INLINE brotli_reg_t BrotliReverseBits(brotli_reg_t num) {
72#if defined(BROTLI_RBIT)
73 return BROTLI_RBIT(num);
74#else
75 return kReverseBits[num];
76#endif
77}
78
79/* Stores code in table[0], table[step], table[2*step], ..., table[end] */
80/* Assumes that end is an integer multiple of step */
81static BROTLI_INLINE void ReplicateValue(HuffmanCode* table,
82 int step, int end,
83 HuffmanCode code) {
84 do {
85 end -= step;
86 table[end] = code;
87 } while (end > 0);
88}
89
90/* Returns the table width of the next 2nd level table. |count| is the histogram
91 of bit lengths for the remaining symbols, |len| is the code length of the
92 next processed symbol. */
93static BROTLI_INLINE int NextTableBitSize(const uint16_t* const count,
94 int len, int root_bits) {
95 int left = 1 << (len - root_bits);
96 while (len < BROTLI_HUFFMAN_MAX_CODE_LENGTH) {
97 left -= count[len];
98 if (left <= 0) break;
99 ++len;
100 left <<= 1;
101 }
102 return len - root_bits;
103}
104
105void BrotliBuildCodeLengthsHuffmanTable(HuffmanCode* table,
106 const uint8_t* const code_lengths,
107 uint16_t* count) {
108 HuffmanCode code; /* current table entry */
109 int symbol; /* symbol index in original or sorted table */
110 brotli_reg_t key; /* prefix code */
111 brotli_reg_t key_step; /* prefix code addend */
112 int step; /* step size to replicate values in current table */
113 int table_size; /* size of current table */
114 int sorted[BROTLI_CODE_LENGTH_CODES]; /* symbols sorted by code length */
115 /* offsets in sorted table for each length */
116 int offset[BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH + 1];
117 int bits;
118 int bits_count;
119 BROTLI_DCHECK(BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH <=
120 BROTLI_REVERSE_BITS_MAX);
121 BROTLI_DCHECK(BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH == 5);
122
123 /* Generate offsets into sorted symbol table by code length. */
124 symbol = -1;
125 bits = 1;
126 /* BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH == 5 */
127 BROTLI_REPEAT_5({
128 symbol += count[bits];
129 offset[bits] = symbol;
130 bits++;
131 });
132 /* Symbols with code length 0 are placed after all other symbols. */
133 offset[0] = BROTLI_CODE_LENGTH_CODES - 1;
134
135 /* Sort symbols by length, by symbol order within each length. */
136 symbol = BROTLI_CODE_LENGTH_CODES;
137 do {
138 BROTLI_REPEAT_6({
139 symbol--;
140 sorted[offset[code_lengths[symbol]]--] = symbol;
141 });
142 } while (symbol != 0);
143
144 table_size = 1 << BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH;
145
146 /* Special case: all symbols but one have 0 code length. */
147 if (offset[0] == 0) {
148 code = ConstructHuffmanCode(0, (uint16_t)sorted[0]);
149 for (key = 0; key < (brotli_reg_t)table_size; ++key) {
150 table[key] = code;
151 }
152 return;
153 }
154
155 /* Fill in table. */
156 key = 0;
157 key_step = BROTLI_REVERSE_BITS_LOWEST;
158 symbol = 0;
159 bits = 1;
160 step = 2;
161 do {
162 for (bits_count = count[bits]; bits_count != 0; --bits_count) {
163 code = ConstructHuffmanCode((uint8_t)bits, (uint16_t)sorted[symbol++]);
164 ReplicateValue(&table[BrotliReverseBits(key)], step, table_size, code);
165 key += key_step;
166 }
167 step <<= 1;
168 key_step >>= 1;
169 } while (++bits <= BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH);
170}
171
172uint32_t BrotliBuildHuffmanTable(HuffmanCode* root_table,
173 int root_bits,
174 const uint16_t* const symbol_lists,
175 uint16_t* count) {
176 HuffmanCode code; /* current table entry */
177 HuffmanCode* table; /* next available space in table */
178 int len; /* current code length */
179 int symbol; /* symbol index in original or sorted table */
180 brotli_reg_t key; /* prefix code */
181 brotli_reg_t key_step; /* prefix code addend */
182 brotli_reg_t sub_key; /* 2nd level table prefix code */
183 brotli_reg_t sub_key_step; /* 2nd level table prefix code addend */
184 int step; /* step size to replicate values in current table */
185 int table_bits; /* key length of current table */
186 int table_size; /* size of current table */
187 int total_size; /* sum of root table size and 2nd level table sizes */
188 int max_length = -1;
189 int bits;
190 int bits_count;
191
192 BROTLI_DCHECK(root_bits <= BROTLI_REVERSE_BITS_MAX);
193 BROTLI_DCHECK(BROTLI_HUFFMAN_MAX_CODE_LENGTH - root_bits <=
194 BROTLI_REVERSE_BITS_MAX);
195
196 while (symbol_lists[max_length] == 0xFFFF) max_length--;
197 max_length += BROTLI_HUFFMAN_MAX_CODE_LENGTH + 1;
198
199 table = root_table;
200 table_bits = root_bits;
201 table_size = 1 << table_bits;
202 total_size = table_size;
203
204 /* Fill in the root table. Reduce the table size to if possible,
205 and create the repetitions by memcpy. */
206 if (table_bits > max_length) {
207 table_bits = max_length;
208 table_size = 1 << table_bits;
209 }
210 key = 0;
211 key_step = BROTLI_REVERSE_BITS_LOWEST;
212 bits = 1;
213 step = 2;
214 do {
215 symbol = bits - (BROTLI_HUFFMAN_MAX_CODE_LENGTH + 1);
216 for (bits_count = count[bits]; bits_count != 0; --bits_count) {
217 symbol = symbol_lists[symbol];
218 code = ConstructHuffmanCode((uint8_t)bits, (uint16_t)symbol);
219 ReplicateValue(&table[BrotliReverseBits(key)], step, table_size, code);
220 key += key_step;
221 }
222 step <<= 1;
223 key_step >>= 1;
224 } while (++bits <= table_bits);
225
226 /* If root_bits != table_bits then replicate to fill the remaining slots. */
227 while (total_size != table_size) {
228 memcpy(&table[table_size], &table[0],
229 (size_t)table_size * sizeof(table[0]));
230 table_size <<= 1;
231 }
232
233 /* Fill in 2nd level tables and add pointers to root table. */
234 key_step = BROTLI_REVERSE_BITS_LOWEST >> (root_bits - 1);
235 sub_key = (BROTLI_REVERSE_BITS_LOWEST << 1);
236 sub_key_step = BROTLI_REVERSE_BITS_LOWEST;
237 for (len = root_bits + 1, step = 2; len <= max_length; ++len) {
238 symbol = len - (BROTLI_HUFFMAN_MAX_CODE_LENGTH + 1);
239 for (; count[len] != 0; --count[len]) {
240 if (sub_key == (BROTLI_REVERSE_BITS_LOWEST << 1U)) {
241 table += table_size;
242 table_bits = NextTableBitSize(count, len, root_bits);
243 table_size = 1 << table_bits;
244 total_size += table_size;
245 sub_key = BrotliReverseBits(key);
246 key += key_step;
247 root_table[sub_key] = ConstructHuffmanCode(
248 (uint8_t)(table_bits + root_bits),
249 (uint16_t)(((size_t)(table - root_table)) - sub_key));
250 sub_key = 0;
251 }
252 symbol = symbol_lists[symbol];
253 code = ConstructHuffmanCode((uint8_t)(len - root_bits), (uint16_t)symbol);
254 ReplicateValue(
255 &table[BrotliReverseBits(sub_key)], step, table_size, code);
256 sub_key += sub_key_step;
257 }
258 step <<= 1;
259 sub_key_step >>= 1;
260 }
261 return (uint32_t)total_size;
262}
263
264uint32_t BrotliBuildSimpleHuffmanTable(HuffmanCode* table,
265 int root_bits,
266 uint16_t* val,
267 uint32_t num_symbols) {
268 uint32_t table_size = 1;
269 const uint32_t goal_size = 1U << root_bits;
270 switch (num_symbols) {
271 case 0:
272 table[0] = ConstructHuffmanCode(0, val[0]);
273 break;
274 case 1:
275 if (val[1] > val[0]) {
276 table[0] = ConstructHuffmanCode(1, val[0]);
277 table[1] = ConstructHuffmanCode(1, val[1]);
278 } else {
279 table[0] = ConstructHuffmanCode(1, val[1]);
280 table[1] = ConstructHuffmanCode(1, val[0]);
281 }
282 table_size = 2;
283 break;
284 case 2:
285 table[0] = ConstructHuffmanCode(1, val[0]);
286 table[2] = ConstructHuffmanCode(1, val[0]);
287 if (val[2] > val[1]) {
288 table[1] = ConstructHuffmanCode(2, val[1]);
289 table[3] = ConstructHuffmanCode(2, val[2]);
290 } else {
291 table[1] = ConstructHuffmanCode(2, val[2]);
292 table[3] = ConstructHuffmanCode(2, val[1]);
293 }
294 table_size = 4;
295 break;
296 case 3: {
297 int i, k;
298 for (i = 0; i < 3; ++i) {
299 for (k = i + 1; k < 4; ++k) {
300 if (val[k] < val[i]) {
301 uint16_t t = val[k];
302 val[k] = val[i];
303 val[i] = t;
304 }
305 }
306 }
307 table[0] = ConstructHuffmanCode(2, val[0]);
308 table[2] = ConstructHuffmanCode(2, val[1]);
309 table[1] = ConstructHuffmanCode(2, val[2]);
310 table[3] = ConstructHuffmanCode(2, val[3]);
311 table_size = 4;
312 break;
313 }
314 case 4: {
315 if (val[3] < val[2]) {
316 uint16_t t = val[3];
317 val[3] = val[2];
318 val[2] = t;
319 }
320 table[0] = ConstructHuffmanCode(1, val[0]);
321 table[1] = ConstructHuffmanCode(2, val[1]);
322 table[2] = ConstructHuffmanCode(1, val[0]);
323 table[3] = ConstructHuffmanCode(3, val[2]);
324 table[4] = ConstructHuffmanCode(1, val[0]);
325 table[5] = ConstructHuffmanCode(2, val[1]);
326 table[6] = ConstructHuffmanCode(1, val[0]);
327 table[7] = ConstructHuffmanCode(3, val[3]);
328 table_size = 8;
329 break;
330 }
331 }
332 while (table_size != goal_size) {
333 memcpy(&table[table_size], &table[0],
334 (size_t)table_size * sizeof(table[0]));
335 table_size <<= 1;
336 }
337 return goal_size;
338}
339
340#if defined(__cplusplus) || defined(c_plusplus)
341} /* extern "C" */
342#endif
343