huffman.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340
  1. /* Copyright 2013 Google Inc. All Rights Reserved.
  2. Distributed under MIT license.
  3. See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
  4. */
  5. /* Utilities for building Huffman decoding tables. */
  6. #include "huffman.h"
  7. #include <string.h> /* memcpy, memset */
  8. #include "../common/constants.h"
  9. #include "../common/platform.h"
  10. #include <brotli/types.h>
  11. #if defined(__cplusplus) || defined(c_plusplus)
  12. extern "C" {
  13. #endif
  14. #define BROTLI_REVERSE_BITS_MAX 8
  15. #if defined(BROTLI_RBIT)
  16. #define BROTLI_REVERSE_BITS_BASE \
  17. ((sizeof(brotli_reg_t) << 3) - BROTLI_REVERSE_BITS_MAX)
  18. #else
  19. #define BROTLI_REVERSE_BITS_BASE 0
  20. static uint8_t kReverseBits[1 << BROTLI_REVERSE_BITS_MAX] = {
  21. 0x00, 0x80, 0x40, 0xC0, 0x20, 0xA0, 0x60, 0xE0,
  22. 0x10, 0x90, 0x50, 0xD0, 0x30, 0xB0, 0x70, 0xF0,
  23. 0x08, 0x88, 0x48, 0xC8, 0x28, 0xA8, 0x68, 0xE8,
  24. 0x18, 0x98, 0x58, 0xD8, 0x38, 0xB8, 0x78, 0xF8,
  25. 0x04, 0x84, 0x44, 0xC4, 0x24, 0xA4, 0x64, 0xE4,
  26. 0x14, 0x94, 0x54, 0xD4, 0x34, 0xB4, 0x74, 0xF4,
  27. 0x0C, 0x8C, 0x4C, 0xCC, 0x2C, 0xAC, 0x6C, 0xEC,
  28. 0x1C, 0x9C, 0x5C, 0xDC, 0x3C, 0xBC, 0x7C, 0xFC,
  29. 0x02, 0x82, 0x42, 0xC2, 0x22, 0xA2, 0x62, 0xE2,
  30. 0x12, 0x92, 0x52, 0xD2, 0x32, 0xB2, 0x72, 0xF2,
  31. 0x0A, 0x8A, 0x4A, 0xCA, 0x2A, 0xAA, 0x6A, 0xEA,
  32. 0x1A, 0x9A, 0x5A, 0xDA, 0x3A, 0xBA, 0x7A, 0xFA,
  33. 0x06, 0x86, 0x46, 0xC6, 0x26, 0xA6, 0x66, 0xE6,
  34. 0x16, 0x96, 0x56, 0xD6, 0x36, 0xB6, 0x76, 0xF6,
  35. 0x0E, 0x8E, 0x4E, 0xCE, 0x2E, 0xAE, 0x6E, 0xEE,
  36. 0x1E, 0x9E, 0x5E, 0xDE, 0x3E, 0xBE, 0x7E, 0xFE,
  37. 0x01, 0x81, 0x41, 0xC1, 0x21, 0xA1, 0x61, 0xE1,
  38. 0x11, 0x91, 0x51, 0xD1, 0x31, 0xB1, 0x71, 0xF1,
  39. 0x09, 0x89, 0x49, 0xC9, 0x29, 0xA9, 0x69, 0xE9,
  40. 0x19, 0x99, 0x59, 0xD9, 0x39, 0xB9, 0x79, 0xF9,
  41. 0x05, 0x85, 0x45, 0xC5, 0x25, 0xA5, 0x65, 0xE5,
  42. 0x15, 0x95, 0x55, 0xD5, 0x35, 0xB5, 0x75, 0xF5,
  43. 0x0D, 0x8D, 0x4D, 0xCD, 0x2D, 0xAD, 0x6D, 0xED,
  44. 0x1D, 0x9D, 0x5D, 0xDD, 0x3D, 0xBD, 0x7D, 0xFD,
  45. 0x03, 0x83, 0x43, 0xC3, 0x23, 0xA3, 0x63, 0xE3,
  46. 0x13, 0x93, 0x53, 0xD3, 0x33, 0xB3, 0x73, 0xF3,
  47. 0x0B, 0x8B, 0x4B, 0xCB, 0x2B, 0xAB, 0x6B, 0xEB,
  48. 0x1B, 0x9B, 0x5B, 0xDB, 0x3B, 0xBB, 0x7B, 0xFB,
  49. 0x07, 0x87, 0x47, 0xC7, 0x27, 0xA7, 0x67, 0xE7,
  50. 0x17, 0x97, 0x57, 0xD7, 0x37, 0xB7, 0x77, 0xF7,
  51. 0x0F, 0x8F, 0x4F, 0xCF, 0x2F, 0xAF, 0x6F, 0xEF,
  52. 0x1F, 0x9F, 0x5F, 0xDF, 0x3F, 0xBF, 0x7F, 0xFF
  53. };
  54. #endif /* BROTLI_RBIT */
  55. #define BROTLI_REVERSE_BITS_LOWEST \
  56. ((brotli_reg_t)1 << (BROTLI_REVERSE_BITS_MAX - 1 + BROTLI_REVERSE_BITS_BASE))
  57. /* Returns reverse(num >> BROTLI_REVERSE_BITS_BASE, BROTLI_REVERSE_BITS_MAX),
  58. where reverse(value, len) is the bit-wise reversal of the len least
  59. significant bits of value. */
  60. static BROTLI_INLINE brotli_reg_t BrotliReverseBits(brotli_reg_t num) {
  61. #if defined(BROTLI_RBIT)
  62. return BROTLI_RBIT(num);
  63. #else
  64. return kReverseBits[num];
  65. #endif
  66. }
  67. /* Stores code in table[0], table[step], table[2*step], ..., table[end] */
  68. /* Assumes that end is an integer multiple of step */
  69. static BROTLI_INLINE void ReplicateValue(HuffmanCode* table,
  70. int step, int end,
  71. HuffmanCode code) {
  72. do {
  73. end -= step;
  74. table[end] = code;
  75. } while (end > 0);
  76. }
  77. /* Returns the table width of the next 2nd level table. |count| is the histogram
  78. of bit lengths for the remaining symbols, |len| is the code length of the
  79. next processed symbol. */
  80. static BROTLI_INLINE int NextTableBitSize(const uint16_t* const count,
  81. int len, int root_bits) {
  82. int left = 1 << (len - root_bits);
  83. while (len < BROTLI_HUFFMAN_MAX_CODE_LENGTH) {
  84. left -= count[len];
  85. if (left <= 0) break;
  86. ++len;
  87. left <<= 1;
  88. }
  89. return len - root_bits;
  90. }
  91. void BrotliBuildCodeLengthsHuffmanTable(HuffmanCode* table,
  92. const uint8_t* const code_lengths,
  93. uint16_t* count) {
  94. HuffmanCode code; /* current table entry */
  95. int symbol; /* symbol index in original or sorted table */
  96. brotli_reg_t key; /* prefix code */
  97. brotli_reg_t key_step; /* prefix code addend */
  98. int step; /* step size to replicate values in current table */
  99. int table_size; /* size of current table */
  100. int sorted[BROTLI_CODE_LENGTH_CODES]; /* symbols sorted by code length */
  101. /* offsets in sorted table for each length */
  102. int offset[BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH + 1];
  103. int bits;
  104. int bits_count;
  105. BROTLI_DCHECK(BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH <=
  106. BROTLI_REVERSE_BITS_MAX);
  107. /* Generate offsets into sorted symbol table by code length. */
  108. symbol = -1;
  109. bits = 1;
  110. BROTLI_REPEAT(BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH, {
  111. symbol += count[bits];
  112. offset[bits] = symbol;
  113. bits++;
  114. });
  115. /* Symbols with code length 0 are placed after all other symbols. */
  116. offset[0] = BROTLI_CODE_LENGTH_CODES - 1;
  117. /* Sort symbols by length, by symbol order within each length. */
  118. symbol = BROTLI_CODE_LENGTH_CODES;
  119. do {
  120. BROTLI_REPEAT(6, {
  121. symbol--;
  122. sorted[offset[code_lengths[symbol]]--] = symbol;
  123. });
  124. } while (symbol != 0);
  125. table_size = 1 << BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH;
  126. /* Special case: all symbols but one have 0 code length. */
  127. if (offset[0] == 0) {
  128. code = ConstructHuffmanCode(0, (uint16_t)sorted[0]);
  129. for (key = 0; key < (brotli_reg_t)table_size; ++key) {
  130. table[key] = code;
  131. }
  132. return;
  133. }
  134. /* Fill in table. */
  135. key = 0;
  136. key_step = BROTLI_REVERSE_BITS_LOWEST;
  137. symbol = 0;
  138. bits = 1;
  139. step = 2;
  140. do {
  141. for (bits_count = count[bits]; bits_count != 0; --bits_count) {
  142. code = ConstructHuffmanCode((uint8_t)bits, (uint16_t)sorted[symbol++]);
  143. ReplicateValue(&table[BrotliReverseBits(key)], step, table_size, code);
  144. key += key_step;
  145. }
  146. step <<= 1;
  147. key_step >>= 1;
  148. } while (++bits <= BROTLI_HUFFMAN_MAX_CODE_LENGTH_CODE_LENGTH);
  149. }
  150. uint32_t BrotliBuildHuffmanTable(HuffmanCode* root_table,
  151. int root_bits,
  152. const uint16_t* const symbol_lists,
  153. uint16_t* count) {
  154. HuffmanCode code; /* current table entry */
  155. HuffmanCode* table; /* next available space in table */
  156. int len; /* current code length */
  157. int symbol; /* symbol index in original or sorted table */
  158. brotli_reg_t key; /* prefix code */
  159. brotli_reg_t key_step; /* prefix code addend */
  160. brotli_reg_t sub_key; /* 2nd level table prefix code */
  161. brotli_reg_t sub_key_step; /* 2nd level table prefix code addend */
  162. int step; /* step size to replicate values in current table */
  163. int table_bits; /* key length of current table */
  164. int table_size; /* size of current table */
  165. int total_size; /* sum of root table size and 2nd level table sizes */
  166. int max_length = -1;
  167. int bits;
  168. int bits_count;
  169. BROTLI_DCHECK(root_bits <= BROTLI_REVERSE_BITS_MAX);
  170. BROTLI_DCHECK(BROTLI_HUFFMAN_MAX_CODE_LENGTH - root_bits <=
  171. BROTLI_REVERSE_BITS_MAX);
  172. while (symbol_lists[max_length] == 0xFFFF) max_length--;
  173. max_length += BROTLI_HUFFMAN_MAX_CODE_LENGTH + 1;
  174. table = root_table;
  175. table_bits = root_bits;
  176. table_size = 1 << table_bits;
  177. total_size = table_size;
  178. /* Fill in the root table. Reduce the table size to if possible,
  179. and create the repetitions by memcpy. */
  180. if (table_bits > max_length) {
  181. table_bits = max_length;
  182. table_size = 1 << table_bits;
  183. }
  184. key = 0;
  185. key_step = BROTLI_REVERSE_BITS_LOWEST;
  186. bits = 1;
  187. step = 2;
  188. do {
  189. symbol = bits - (BROTLI_HUFFMAN_MAX_CODE_LENGTH + 1);
  190. for (bits_count = count[bits]; bits_count != 0; --bits_count) {
  191. symbol = symbol_lists[symbol];
  192. code = ConstructHuffmanCode((uint8_t)bits, (uint16_t)symbol);
  193. ReplicateValue(&table[BrotliReverseBits(key)], step, table_size, code);
  194. key += key_step;
  195. }
  196. step <<= 1;
  197. key_step >>= 1;
  198. } while (++bits <= table_bits);
  199. /* If root_bits != table_bits then replicate to fill the remaining slots. */
  200. while (total_size != table_size) {
  201. memcpy(&table[table_size], &table[0],
  202. (size_t)table_size * sizeof(table[0]));
  203. table_size <<= 1;
  204. }
  205. /* Fill in 2nd level tables and add pointers to root table. */
  206. key_step = BROTLI_REVERSE_BITS_LOWEST >> (root_bits - 1);
  207. sub_key = (BROTLI_REVERSE_BITS_LOWEST << 1);
  208. sub_key_step = BROTLI_REVERSE_BITS_LOWEST;
  209. for (len = root_bits + 1, step = 2; len <= max_length; ++len) {
  210. symbol = len - (BROTLI_HUFFMAN_MAX_CODE_LENGTH + 1);
  211. for (; count[len] != 0; --count[len]) {
  212. if (sub_key == (BROTLI_REVERSE_BITS_LOWEST << 1U)) {
  213. table += table_size;
  214. table_bits = NextTableBitSize(count, len, root_bits);
  215. table_size = 1 << table_bits;
  216. total_size += table_size;
  217. sub_key = BrotliReverseBits(key);
  218. key += key_step;
  219. root_table[sub_key] = ConstructHuffmanCode(
  220. (uint8_t)(table_bits + root_bits),
  221. (uint16_t)(((size_t)(table - root_table)) - sub_key));
  222. sub_key = 0;
  223. }
  224. symbol = symbol_lists[symbol];
  225. code = ConstructHuffmanCode((uint8_t)(len - root_bits), (uint16_t)symbol);
  226. ReplicateValue(
  227. &table[BrotliReverseBits(sub_key)], step, table_size, code);
  228. sub_key += sub_key_step;
  229. }
  230. step <<= 1;
  231. sub_key_step >>= 1;
  232. }
  233. return (uint32_t)total_size;
  234. }
  235. uint32_t BrotliBuildSimpleHuffmanTable(HuffmanCode* table,
  236. int root_bits,
  237. uint16_t* val,
  238. uint32_t num_symbols) {
  239. uint32_t table_size = 1;
  240. const uint32_t goal_size = 1U << root_bits;
  241. switch (num_symbols) {
  242. case 0:
  243. table[0] = ConstructHuffmanCode(0, val[0]);
  244. break;
  245. case 1:
  246. if (val[1] > val[0]) {
  247. table[0] = ConstructHuffmanCode(1, val[0]);
  248. table[1] = ConstructHuffmanCode(1, val[1]);
  249. } else {
  250. table[0] = ConstructHuffmanCode(1, val[1]);
  251. table[1] = ConstructHuffmanCode(1, val[0]);
  252. }
  253. table_size = 2;
  254. break;
  255. case 2:
  256. table[0] = ConstructHuffmanCode(1, val[0]);
  257. table[2] = ConstructHuffmanCode(1, val[0]);
  258. if (val[2] > val[1]) {
  259. table[1] = ConstructHuffmanCode(2, val[1]);
  260. table[3] = ConstructHuffmanCode(2, val[2]);
  261. } else {
  262. table[1] = ConstructHuffmanCode(2, val[2]);
  263. table[3] = ConstructHuffmanCode(2, val[1]);
  264. }
  265. table_size = 4;
  266. break;
  267. case 3: {
  268. int i, k;
  269. for (i = 0; i < 3; ++i) {
  270. for (k = i + 1; k < 4; ++k) {
  271. if (val[k] < val[i]) {
  272. uint16_t t = val[k];
  273. val[k] = val[i];
  274. val[i] = t;
  275. }
  276. }
  277. }
  278. table[0] = ConstructHuffmanCode(2, val[0]);
  279. table[2] = ConstructHuffmanCode(2, val[1]);
  280. table[1] = ConstructHuffmanCode(2, val[2]);
  281. table[3] = ConstructHuffmanCode(2, val[3]);
  282. table_size = 4;
  283. break;
  284. }
  285. case 4: {
  286. if (val[3] < val[2]) {
  287. uint16_t t = val[3];
  288. val[3] = val[2];
  289. val[2] = t;
  290. }
  291. table[0] = ConstructHuffmanCode(1, val[0]);
  292. table[1] = ConstructHuffmanCode(2, val[1]);
  293. table[2] = ConstructHuffmanCode(1, val[0]);
  294. table[3] = ConstructHuffmanCode(3, val[2]);
  295. table[4] = ConstructHuffmanCode(1, val[0]);
  296. table[5] = ConstructHuffmanCode(2, val[1]);
  297. table[6] = ConstructHuffmanCode(1, val[0]);
  298. table[7] = ConstructHuffmanCode(3, val[3]);
  299. table_size = 8;
  300. break;
  301. }
  302. }
  303. while (table_size != goal_size) {
  304. memcpy(&table[table_size], &table[0],
  305. (size_t)table_size * sizeof(table[0]));
  306. table_size <<= 1;
  307. }
  308. return goal_size;
  309. }
  310. #if defined(__cplusplus) || defined(c_plusplus)
  311. } /* extern "C" */
  312. #endif