OLD | NEW |
1 /* Copyright 2010 Google Inc. All Rights Reserved. | 1 /* Copyright 2010 Google Inc. All Rights Reserved. |
2 | 2 |
3 Distributed under MIT license. | 3 Distributed under MIT license. |
4 See file LICENSE for detail or copy at https://opensource.org/licenses/MIT | 4 See file LICENSE for detail or copy at https://opensource.org/licenses/MIT |
5 */ | 5 */ |
6 | 6 |
7 // Entropy encoding (Huffman) utilities. | 7 /* Entropy encoding (Huffman) utilities. */ |
8 | 8 |
9 #include "./entropy_encode.h" | 9 #include "./entropy_encode.h" |
10 | 10 |
11 #include <algorithm> | 11 #include <string.h> /* memset */ |
12 #include <limits> | |
13 #include <cstdlib> | |
14 | 12 |
15 #include "./histogram.h" | 13 #include "../common/constants.h" |
| 14 #include <brotli/types.h> |
16 #include "./port.h" | 15 #include "./port.h" |
17 #include "./types.h" | |
18 | 16 |
19 namespace brotli { | 17 #if defined(__cplusplus) || defined(c_plusplus) |
| 18 extern "C" { |
| 19 #endif |
20 | 20 |
21 void SetDepth(const HuffmanTree &p, | 21 BROTLI_BOOL BrotliSetDepth( |
22 HuffmanTree *pool, | 22 int p0, HuffmanTree* pool, uint8_t* depth, int max_depth) { |
23 uint8_t *depth, | 23 int stack[16]; |
24 uint8_t level) { | 24 int level = 0; |
25 if (p.index_left_ >= 0) { | 25 int p = p0; |
26 ++level; | 26 assert(max_depth <= 15); |
27 SetDepth(pool[p.index_left_], pool, depth, level); | 27 stack[0] = -1; |
28 SetDepth(pool[p.index_right_or_value_], pool, depth, level); | 28 while (BROTLI_TRUE) { |
29 } else { | 29 if (pool[p].index_left_ >= 0) { |
30 depth[p.index_right_or_value_] = level; | 30 level++; |
| 31 if (level > max_depth) return BROTLI_FALSE; |
| 32 stack[level] = pool[p].index_right_or_value_; |
| 33 p = pool[p].index_left_; |
| 34 continue; |
| 35 } else { |
| 36 depth[pool[p].index_right_or_value_] = (uint8_t)level; |
| 37 } |
| 38 while (level >= 0 && stack[level] == -1) level--; |
| 39 if (level < 0) return BROTLI_TRUE; |
| 40 p = stack[level]; |
| 41 stack[level] = -1; |
31 } | 42 } |
32 } | 43 } |
33 | 44 |
34 // Sort the root nodes, least popular first. | 45 /* Sort the root nodes, least popular first. */ |
35 static inline bool SortHuffmanTree(const HuffmanTree& v0, | 46 static BROTLI_INLINE BROTLI_BOOL SortHuffmanTree( |
36 const HuffmanTree& v1) { | 47 const HuffmanTree* v0, const HuffmanTree* v1) { |
37 if (v0.total_count_ != v1.total_count_) { | 48 if (v0->total_count_ != v1->total_count_) { |
38 return v0.total_count_ < v1.total_count_; | 49 return TO_BROTLI_BOOL(v0->total_count_ < v1->total_count_); |
39 } | 50 } |
40 return v0.index_right_or_value_ > v1.index_right_or_value_; | 51 return TO_BROTLI_BOOL(v0->index_right_or_value_ > v1->index_right_or_value_); |
41 } | 52 } |
42 | 53 |
43 // This function will create a Huffman tree. | 54 /* This function will create a Huffman tree. |
44 // | 55 |
45 // The catch here is that the tree cannot be arbitrarily deep. | 56 The catch here is that the tree cannot be arbitrarily deep. |
46 // Brotli specifies a maximum depth of 15 bits for "code trees" | 57 Brotli specifies a maximum depth of 15 bits for "code trees" |
47 // and 7 bits for "code length code trees." | 58 and 7 bits for "code length code trees." |
48 // | 59 |
49 // count_limit is the value that is to be faked as the minimum value | 60 count_limit is the value that is to be faked as the minimum value |
50 // and this minimum value is raised until the tree matches the | 61 and this minimum value is raised until the tree matches the |
51 // maximum length requirement. | 62 maximum length requirement. |
52 // | 63 |
53 // This algorithm is not of excellent performance for very long data blocks, | 64 This algorithm is not of excellent performance for very long data blocks, |
54 // especially when population counts are longer than 2**tree_limit, but | 65 especially when population counts are longer than 2**tree_limit, but |
55 // we are not planning to use this with extremely long blocks. | 66 we are not planning to use this with extremely long blocks. |
56 // | 67 |
57 // See http://en.wikipedia.org/wiki/Huffman_coding | 68 See http://en.wikipedia.org/wiki/Huffman_coding */ |
58 void CreateHuffmanTree(const uint32_t *data, | 69 void BrotliCreateHuffmanTree(const uint32_t *data, |
59 const size_t length, | 70 const size_t length, |
60 const int tree_limit, | 71 const int tree_limit, |
61 HuffmanTree* tree, | 72 HuffmanTree* tree, |
62 uint8_t *depth) { | 73 uint8_t *depth) { |
63 // For block sizes below 64 kB, we never need to do a second iteration | 74 uint32_t count_limit; |
64 // of this loop. Probably all of our block sizes will be smaller than | 75 HuffmanTree sentinel; |
65 // that, so this loop is mostly of academic interest. If we actually | 76 InitHuffmanTree(&sentinel, BROTLI_UINT32_MAX, -1, -1); |
66 // would need this, we would be better off with the Katajainen algorithm. | 77 /* For block sizes below 64 kB, we never need to do a second iteration |
67 for (uint32_t count_limit = 1; ; count_limit *= 2) { | 78 of this loop. Probably all of our block sizes will be smaller than |
| 79 that, so this loop is mostly of academic interest. If we actually |
| 80 would need this, we would be better off with the Katajainen algorithm. */ |
| 81 for (count_limit = 1; ; count_limit *= 2) { |
68 size_t n = 0; | 82 size_t n = 0; |
69 for (size_t i = length; i != 0;) { | 83 size_t i; |
| 84 size_t j; |
| 85 size_t k; |
| 86 for (i = length; i != 0;) { |
70 --i; | 87 --i; |
71 if (data[i]) { | 88 if (data[i]) { |
72 const uint32_t count = std::max(data[i], count_limit); | 89 const uint32_t count = BROTLI_MAX(uint32_t, data[i], count_limit); |
73 tree[n++] = HuffmanTree(count, -1, static_cast<int16_t>(i)); | 90 InitHuffmanTree(&tree[n++], count, -1, (int16_t)i); |
74 } | 91 } |
75 } | 92 } |
76 | 93 |
77 if (n == 1) { | 94 if (n == 1) { |
78 depth[tree[0].index_right_or_value_] = 1; // Only one element. | 95 depth[tree[0].index_right_or_value_] = 1; /* Only one element. */ |
79 break; | 96 break; |
80 } | 97 } |
81 | 98 |
82 std::sort(tree, tree + n, SortHuffmanTree); | 99 SortHuffmanTreeItems(tree, n, SortHuffmanTree); |
83 | 100 |
84 // The nodes are: | 101 /* The nodes are: |
85 // [0, n): the sorted leaf nodes that we start with. | 102 [0, n): the sorted leaf nodes that we start with. |
86 // [n]: we add a sentinel here. | 103 [n]: we add a sentinel here. |
87 // [n + 1, 2n): new parent nodes are added here, starting from | 104 [n + 1, 2n): new parent nodes are added here, starting from |
88 // (n+1). These are naturally in ascending order. | 105 (n+1). These are naturally in ascending order. |
89 // [2n]: we add a sentinel at the end as well. | 106 [2n]: we add a sentinel at the end as well. |
90 // There will be (2n+1) elements at the end. | 107 There will be (2n+1) elements at the end. */ |
91 const HuffmanTree sentinel(std::numeric_limits<uint32_t>::max(), -1, -1); | |
92 tree[n] = sentinel; | 108 tree[n] = sentinel; |
93 tree[n + 1] = sentinel; | 109 tree[n + 1] = sentinel; |
94 | 110 |
95 size_t i = 0; // Points to the next leaf node. | 111 i = 0; /* Points to the next leaf node. */ |
96 size_t j = n + 1; // Points to the next non-leaf node. | 112 j = n + 1; /* Points to the next non-leaf node. */ |
97 for (size_t k = n - 1; k != 0; --k) { | 113 for (k = n - 1; k != 0; --k) { |
98 size_t left, right; | 114 size_t left, right; |
99 if (tree[i].total_count_ <= tree[j].total_count_) { | 115 if (tree[i].total_count_ <= tree[j].total_count_) { |
100 left = i; | 116 left = i; |
101 ++i; | 117 ++i; |
102 } else { | 118 } else { |
103 left = j; | 119 left = j; |
104 ++j; | 120 ++j; |
105 } | 121 } |
106 if (tree[i].total_count_ <= tree[j].total_count_) { | 122 if (tree[i].total_count_ <= tree[j].total_count_) { |
107 right = i; | 123 right = i; |
108 ++i; | 124 ++i; |
109 } else { | 125 } else { |
110 right = j; | 126 right = j; |
111 ++j; | 127 ++j; |
112 } | 128 } |
113 | 129 |
114 // The sentinel node becomes the parent node. | 130 { |
115 size_t j_end = 2 * n - k; | 131 /* The sentinel node becomes the parent node. */ |
116 tree[j_end].total_count_ = | 132 size_t j_end = 2 * n - k; |
117 tree[left].total_count_ + tree[right].total_count_; | 133 tree[j_end].total_count_ = |
118 tree[j_end].index_left_ = static_cast<int16_t>(left); | 134 tree[left].total_count_ + tree[right].total_count_; |
119 tree[j_end].index_right_or_value_ = static_cast<int16_t>(right); | 135 tree[j_end].index_left_ = (int16_t)left; |
| 136 tree[j_end].index_right_or_value_ = (int16_t)right; |
120 | 137 |
121 // Add back the last sentinel node. | 138 /* Add back the last sentinel node. */ |
122 tree[j_end + 1] = sentinel; | 139 tree[j_end + 1] = sentinel; |
| 140 } |
123 } | 141 } |
124 SetDepth(tree[2 * n - 1], &tree[0], depth, 0); | 142 if (BrotliSetDepth((int)(2 * n - 1), &tree[0], depth, tree_limit)) { |
125 | 143 /* We need to pack the Huffman tree in tree_limit bits. If this was not |
126 // We need to pack the Huffman tree in tree_limit bits. | 144 successful, add fake entities to the lowest values and retry. */ |
127 // If this was not successful, add fake entities to the lowest values | |
128 // and retry. | |
129 if (*std::max_element(&depth[0], &depth[length]) <= tree_limit) { | |
130 break; | 145 break; |
131 } | 146 } |
132 } | 147 } |
133 } | 148 } |
134 | 149 |
135 static void Reverse(uint8_t* v, size_t start, size_t end) { | 150 static void Reverse(uint8_t* v, size_t start, size_t end) { |
136 --end; | 151 --end; |
137 while (start < end) { | 152 while (start < end) { |
138 uint8_t tmp = v[start]; | 153 uint8_t tmp = v[start]; |
139 v[start] = v[end]; | 154 v[start] = v[end]; |
140 v[end] = tmp; | 155 v[end] = tmp; |
141 ++start; | 156 ++start; |
142 --end; | 157 --end; |
143 } | 158 } |
144 } | 159 } |
145 | 160 |
146 static void WriteHuffmanTreeRepetitions( | 161 static void BrotliWriteHuffmanTreeRepetitions( |
147 const uint8_t previous_value, | 162 const uint8_t previous_value, |
148 const uint8_t value, | 163 const uint8_t value, |
149 size_t repetitions, | 164 size_t repetitions, |
150 size_t* tree_size, | 165 size_t* tree_size, |
151 uint8_t* tree, | 166 uint8_t* tree, |
152 uint8_t* extra_bits_data) { | 167 uint8_t* extra_bits_data) { |
153 assert(repetitions > 0); | 168 assert(repetitions > 0); |
154 if (previous_value != value) { | 169 if (previous_value != value) { |
155 tree[*tree_size] = value; | 170 tree[*tree_size] = value; |
156 extra_bits_data[*tree_size] = 0; | 171 extra_bits_data[*tree_size] = 0; |
157 ++(*tree_size); | 172 ++(*tree_size); |
158 --repetitions; | 173 --repetitions; |
159 } | 174 } |
160 if (repetitions == 7) { | 175 if (repetitions == 7) { |
161 tree[*tree_size] = value; | 176 tree[*tree_size] = value; |
162 extra_bits_data[*tree_size] = 0; | 177 extra_bits_data[*tree_size] = 0; |
163 ++(*tree_size); | 178 ++(*tree_size); |
164 --repetitions; | 179 --repetitions; |
165 } | 180 } |
166 if (repetitions < 3) { | 181 if (repetitions < 3) { |
167 for (size_t i = 0; i < repetitions; ++i) { | 182 size_t i; |
| 183 for (i = 0; i < repetitions; ++i) { |
168 tree[*tree_size] = value; | 184 tree[*tree_size] = value; |
169 extra_bits_data[*tree_size] = 0; | 185 extra_bits_data[*tree_size] = 0; |
170 ++(*tree_size); | 186 ++(*tree_size); |
171 } | 187 } |
172 } else { | 188 } else { |
| 189 size_t start = *tree_size; |
173 repetitions -= 3; | 190 repetitions -= 3; |
174 size_t start = *tree_size; | 191 while (BROTLI_TRUE) { |
175 while (true) { | 192 tree[*tree_size] = BROTLI_REPEAT_PREVIOUS_CODE_LENGTH; |
176 tree[*tree_size] = 16; | |
177 extra_bits_data[*tree_size] = repetitions & 0x3; | 193 extra_bits_data[*tree_size] = repetitions & 0x3; |
178 ++(*tree_size); | 194 ++(*tree_size); |
179 repetitions >>= 2; | 195 repetitions >>= 2; |
180 if (repetitions == 0) { | 196 if (repetitions == 0) { |
181 break; | 197 break; |
182 } | 198 } |
183 --repetitions; | 199 --repetitions; |
184 } | 200 } |
185 Reverse(tree, start, *tree_size); | 201 Reverse(tree, start, *tree_size); |
186 Reverse(extra_bits_data, start, *tree_size); | 202 Reverse(extra_bits_data, start, *tree_size); |
187 } | 203 } |
188 } | 204 } |
189 | 205 |
190 static void WriteHuffmanTreeRepetitionsZeros( | 206 static void BrotliWriteHuffmanTreeRepetitionsZeros( |
191 size_t repetitions, | 207 size_t repetitions, |
192 size_t* tree_size, | 208 size_t* tree_size, |
193 uint8_t* tree, | 209 uint8_t* tree, |
194 uint8_t* extra_bits_data) { | 210 uint8_t* extra_bits_data) { |
195 if (repetitions == 11) { | 211 if (repetitions == 11) { |
196 tree[*tree_size] = 0; | 212 tree[*tree_size] = 0; |
197 extra_bits_data[*tree_size] = 0; | 213 extra_bits_data[*tree_size] = 0; |
198 ++(*tree_size); | 214 ++(*tree_size); |
199 --repetitions; | 215 --repetitions; |
200 } | 216 } |
201 if (repetitions < 3) { | 217 if (repetitions < 3) { |
202 for (size_t i = 0; i < repetitions; ++i) { | 218 size_t i; |
| 219 for (i = 0; i < repetitions; ++i) { |
203 tree[*tree_size] = 0; | 220 tree[*tree_size] = 0; |
204 extra_bits_data[*tree_size] = 0; | 221 extra_bits_data[*tree_size] = 0; |
205 ++(*tree_size); | 222 ++(*tree_size); |
206 } | 223 } |
207 } else { | 224 } else { |
| 225 size_t start = *tree_size; |
208 repetitions -= 3; | 226 repetitions -= 3; |
209 size_t start = *tree_size; | 227 while (BROTLI_TRUE) { |
210 while (true) { | 228 tree[*tree_size] = BROTLI_REPEAT_ZERO_CODE_LENGTH; |
211 tree[*tree_size] = 17; | |
212 extra_bits_data[*tree_size] = repetitions & 0x7; | 229 extra_bits_data[*tree_size] = repetitions & 0x7; |
213 ++(*tree_size); | 230 ++(*tree_size); |
214 repetitions >>= 3; | 231 repetitions >>= 3; |
215 if (repetitions == 0) { | 232 if (repetitions == 0) { |
216 break; | 233 break; |
217 } | 234 } |
218 --repetitions; | 235 --repetitions; |
219 } | 236 } |
220 Reverse(tree, start, *tree_size); | 237 Reverse(tree, start, *tree_size); |
221 Reverse(extra_bits_data, start, *tree_size); | 238 Reverse(extra_bits_data, start, *tree_size); |
222 } | 239 } |
223 } | 240 } |
224 | 241 |
225 void OptimizeHuffmanCountsForRle(size_t length, uint32_t* counts, | 242 void BrotliOptimizeHuffmanCountsForRle(size_t length, uint32_t* counts, |
226 uint8_t* good_for_rle) { | 243 uint8_t* good_for_rle) { |
227 size_t nonzero_count = 0; | 244 size_t nonzero_count = 0; |
228 size_t stride; | 245 size_t stride; |
229 size_t limit; | 246 size_t limit; |
230 size_t sum; | 247 size_t sum; |
231 const size_t streak_limit = 1240; | 248 const size_t streak_limit = 1240; |
232 // Let's make the Huffman code more compatible with rle encoding. | 249 /* Let's make the Huffman code more compatible with RLE encoding. */ |
233 size_t i; | 250 size_t i; |
234 for (i = 0; i < length; i++) { | 251 for (i = 0; i < length; i++) { |
235 if (counts[i]) { | 252 if (counts[i]) { |
236 ++nonzero_count; | 253 ++nonzero_count; |
237 } | 254 } |
238 } | 255 } |
239 if (nonzero_count < 16) { | 256 if (nonzero_count < 16) { |
240 return; | 257 return; |
241 } | 258 } |
242 while (length != 0 && counts[length - 1] == 0) { | 259 while (length != 0 && counts[length - 1] == 0) { |
243 --length; | 260 --length; |
244 } | 261 } |
245 if (length == 0) { | 262 if (length == 0) { |
246 return; // All zeros. | 263 return; /* All zeros. */ |
247 } | 264 } |
248 // Now counts[0..length - 1] does not have trailing zeros. | 265 /* Now counts[0..length - 1] does not have trailing zeros. */ |
249 { | 266 { |
250 size_t nonzeros = 0; | 267 size_t nonzeros = 0; |
251 uint32_t smallest_nonzero = 1 << 30; | 268 uint32_t smallest_nonzero = 1 << 30; |
252 for (i = 0; i < length; ++i) { | 269 for (i = 0; i < length; ++i) { |
253 if (counts[i] != 0) { | 270 if (counts[i] != 0) { |
254 ++nonzeros; | 271 ++nonzeros; |
255 if (smallest_nonzero > counts[i]) { | 272 if (smallest_nonzero > counts[i]) { |
256 smallest_nonzero = counts[i]; | 273 smallest_nonzero = counts[i]; |
257 } | 274 } |
258 } | 275 } |
259 } | 276 } |
260 if (nonzeros < 5) { | 277 if (nonzeros < 5) { |
261 // Small histogram will model it well. | 278 /* Small histogram will model it well. */ |
262 return; | 279 return; |
263 } | 280 } |
264 size_t zeros = length - nonzeros; | |
265 if (smallest_nonzero < 4) { | 281 if (smallest_nonzero < 4) { |
| 282 size_t zeros = length - nonzeros; |
266 if (zeros < 6) { | 283 if (zeros < 6) { |
267 for (i = 1; i < length - 1; ++i) { | 284 for (i = 1; i < length - 1; ++i) { |
268 if (counts[i - 1] != 0 && counts[i] == 0 && counts[i + 1] != 0) { | 285 if (counts[i - 1] != 0 && counts[i] == 0 && counts[i + 1] != 0) { |
269 counts[i] = 1; | 286 counts[i] = 1; |
270 } | 287 } |
271 } | 288 } |
272 } | 289 } |
273 } | 290 } |
274 if (nonzeros < 28) { | 291 if (nonzeros < 28) { |
275 return; | 292 return; |
276 } | 293 } |
277 } | 294 } |
278 // 2) Let's mark all population counts that already can be encoded | 295 /* 2) Let's mark all population counts that already can be encoded |
279 // with an rle code. | 296 with an RLE code. */ |
280 memset(good_for_rle, 0, length); | 297 memset(good_for_rle, 0, length); |
281 { | 298 { |
282 // Let's not spoil any of the existing good rle codes. | 299 /* Let's not spoil any of the existing good RLE codes. |
283 // Mark any seq of 0's that is longer as 5 as a good_for_rle. | 300 Mark any seq of 0's that is longer as 5 as a good_for_rle. |
284 // Mark any seq of non-0's that is longer as 7 as a good_for_rle. | 301 Mark any seq of non-0's that is longer as 7 as a good_for_rle. */ |
285 uint32_t symbol = counts[0]; | 302 uint32_t symbol = counts[0]; |
286 size_t step = 0; | 303 size_t step = 0; |
287 for (i = 0; i <= length; ++i) { | 304 for (i = 0; i <= length; ++i) { |
288 if (i == length || counts[i] != symbol) { | 305 if (i == length || counts[i] != symbol) { |
289 if ((symbol == 0 && step >= 5) || | 306 if ((symbol == 0 && step >= 5) || |
290 (symbol != 0 && step >= 7)) { | 307 (symbol != 0 && step >= 7)) { |
291 size_t k; | 308 size_t k; |
292 for (k = 0; k < step; ++k) { | 309 for (k = 0; k < step; ++k) { |
293 good_for_rle[i - k - 1] = 1; | 310 good_for_rle[i - k - 1] = 1; |
294 } | 311 } |
295 } | 312 } |
296 step = 1; | 313 step = 1; |
297 if (i != length) { | 314 if (i != length) { |
298 symbol = counts[i]; | 315 symbol = counts[i]; |
299 } | 316 } |
300 } else { | 317 } else { |
301 ++step; | 318 ++step; |
302 } | 319 } |
303 } | 320 } |
304 } | 321 } |
305 // 3) Let's replace those population counts that lead to more rle codes. | 322 /* 3) Let's replace those population counts that lead to more RLE codes. |
306 // Math here is in 24.8 fixed point representation. | 323 Math here is in 24.8 fixed point representation. */ |
307 stride = 0; | 324 stride = 0; |
308 limit = 256 * (counts[0] + counts[1] + counts[2]) / 3 + 420; | 325 limit = 256 * (counts[0] + counts[1] + counts[2]) / 3 + 420; |
309 sum = 0; | 326 sum = 0; |
310 for (i = 0; i <= length; ++i) { | 327 for (i = 0; i <= length; ++i) { |
311 if (i == length || good_for_rle[i] || | 328 if (i == length || good_for_rle[i] || |
312 (i != 0 && good_for_rle[i - 1]) || | 329 (i != 0 && good_for_rle[i - 1]) || |
313 (256 * counts[i] - limit + streak_limit) >= 2 * streak_limit) { | 330 (256 * counts[i] - limit + streak_limit) >= 2 * streak_limit) { |
314 if (stride >= 4 || (stride >= 3 && sum == 0)) { | 331 if (stride >= 4 || (stride >= 3 && sum == 0)) { |
315 size_t k; | 332 size_t k; |
316 // The stride must end, collapse what we have, if we have enough (4). | 333 /* The stride must end, collapse what we have, if we have enough (4). */ |
317 size_t count = (sum + stride / 2) / stride; | 334 size_t count = (sum + stride / 2) / stride; |
318 if (count == 0) { | 335 if (count == 0) { |
319 count = 1; | 336 count = 1; |
320 } | 337 } |
321 if (sum == 0) { | 338 if (sum == 0) { |
322 // Don't make an all zeros stride to be upgraded to ones. | 339 /* Don't make an all zeros stride to be upgraded to ones. */ |
323 count = 0; | 340 count = 0; |
324 } | 341 } |
325 for (k = 0; k < stride; ++k) { | 342 for (k = 0; k < stride; ++k) { |
326 // We don't want to change value at counts[i], | 343 /* We don't want to change value at counts[i], |
327 // that is already belonging to the next stride. Thus - 1. | 344 that is already belonging to the next stride. Thus - 1. */ |
328 counts[i - k - 1] = static_cast<uint32_t>(count); | 345 counts[i - k - 1] = (uint32_t)count; |
329 } | 346 } |
330 } | 347 } |
331 stride = 0; | 348 stride = 0; |
332 sum = 0; | 349 sum = 0; |
333 if (i < length - 2) { | 350 if (i < length - 2) { |
334 // All interesting strides have a count of at least 4, | 351 /* All interesting strides have a count of at least 4, */ |
335 // at least when non-zeros. | 352 /* at least when non-zeros. */ |
336 limit = 256 * (counts[i] + counts[i + 1] + counts[i + 2]) / 3 + 420; | 353 limit = 256 * (counts[i] + counts[i + 1] + counts[i + 2]) / 3 + 420; |
337 } else if (i < length) { | 354 } else if (i < length) { |
338 limit = 256 * counts[i]; | 355 limit = 256 * counts[i]; |
339 } else { | 356 } else { |
340 limit = 0; | 357 limit = 0; |
341 } | 358 } |
342 } | 359 } |
343 ++stride; | 360 ++stride; |
344 if (i != length) { | 361 if (i != length) { |
345 sum += counts[i]; | 362 sum += counts[i]; |
346 if (stride >= 4) { | 363 if (stride >= 4) { |
347 limit = (256 * sum + stride / 2) / stride; | 364 limit = (256 * sum + stride / 2) / stride; |
348 } | 365 } |
349 if (stride == 4) { | 366 if (stride == 4) { |
350 limit += 120; | 367 limit += 120; |
351 } | 368 } |
352 } | 369 } |
353 } | 370 } |
354 } | 371 } |
355 | 372 |
356 static void DecideOverRleUse(const uint8_t* depth, const size_t length, | 373 static void DecideOverRleUse(const uint8_t* depth, const size_t length, |
357 bool *use_rle_for_non_zero, | 374 BROTLI_BOOL *use_rle_for_non_zero, |
358 bool *use_rle_for_zero) { | 375 BROTLI_BOOL *use_rle_for_zero) { |
359 size_t total_reps_zero = 0; | 376 size_t total_reps_zero = 0; |
360 size_t total_reps_non_zero = 0; | 377 size_t total_reps_non_zero = 0; |
361 size_t count_reps_zero = 1; | 378 size_t count_reps_zero = 1; |
362 size_t count_reps_non_zero = 1; | 379 size_t count_reps_non_zero = 1; |
363 for (size_t i = 0; i < length;) { | 380 size_t i; |
| 381 for (i = 0; i < length;) { |
364 const uint8_t value = depth[i]; | 382 const uint8_t value = depth[i]; |
365 size_t reps = 1; | 383 size_t reps = 1; |
366 for (size_t k = i + 1; k < length && depth[k] == value; ++k) { | 384 size_t k; |
| 385 for (k = i + 1; k < length && depth[k] == value; ++k) { |
367 ++reps; | 386 ++reps; |
368 } | 387 } |
369 if (reps >= 3 && value == 0) { | 388 if (reps >= 3 && value == 0) { |
370 total_reps_zero += reps; | 389 total_reps_zero += reps; |
371 ++count_reps_zero; | 390 ++count_reps_zero; |
372 } | 391 } |
373 if (reps >= 4 && value != 0) { | 392 if (reps >= 4 && value != 0) { |
374 total_reps_non_zero += reps; | 393 total_reps_non_zero += reps; |
375 ++count_reps_non_zero; | 394 ++count_reps_non_zero; |
376 } | 395 } |
377 i += reps; | 396 i += reps; |
378 } | 397 } |
379 *use_rle_for_non_zero = total_reps_non_zero > count_reps_non_zero * 2; | 398 *use_rle_for_non_zero = |
380 *use_rle_for_zero = total_reps_zero > count_reps_zero * 2; | 399 TO_BROTLI_BOOL(total_reps_non_zero > count_reps_non_zero * 2); |
| 400 *use_rle_for_zero = TO_BROTLI_BOOL(total_reps_zero > count_reps_zero * 2); |
381 } | 401 } |
382 | 402 |
383 void WriteHuffmanTree(const uint8_t* depth, | 403 void BrotliWriteHuffmanTree(const uint8_t* depth, |
384 size_t length, | 404 size_t length, |
385 size_t* tree_size, | 405 size_t* tree_size, |
386 uint8_t* tree, | 406 uint8_t* tree, |
387 uint8_t* extra_bits_data) { | 407 uint8_t* extra_bits_data) { |
388 uint8_t previous_value = 8; | 408 uint8_t previous_value = BROTLI_INITIAL_REPEATED_CODE_LENGTH; |
| 409 size_t i; |
| 410 BROTLI_BOOL use_rle_for_non_zero = BROTLI_FALSE; |
| 411 BROTLI_BOOL use_rle_for_zero = BROTLI_FALSE; |
389 | 412 |
390 // Throw away trailing zeros. | 413 /* Throw away trailing zeros. */ |
391 size_t new_length = length; | 414 size_t new_length = length; |
392 for (size_t i = 0; i < length; ++i) { | 415 for (i = 0; i < length; ++i) { |
393 if (depth[length - i - 1] == 0) { | 416 if (depth[length - i - 1] == 0) { |
394 --new_length; | 417 --new_length; |
395 } else { | 418 } else { |
396 break; | 419 break; |
397 } | 420 } |
398 } | 421 } |
399 | 422 |
400 // First gather statistics on if it is a good idea to do rle. | 423 /* First gather statistics on if it is a good idea to do RLE. */ |
401 bool use_rle_for_non_zero = false; | |
402 bool use_rle_for_zero = false; | |
403 if (length > 50) { | 424 if (length > 50) { |
404 // Find rle coding for longer codes. | 425 /* Find RLE coding for longer codes. |
405 // Shorter codes seem not to benefit from rle. | 426 Shorter codes seem not to benefit from RLE. */ |
406 DecideOverRleUse(depth, new_length, | 427 DecideOverRleUse(depth, new_length, |
407 &use_rle_for_non_zero, &use_rle_for_zero); | 428 &use_rle_for_non_zero, &use_rle_for_zero); |
408 } | 429 } |
409 | 430 |
410 // Actual rle coding. | 431 /* Actual RLE coding. */ |
411 for (size_t i = 0; i < new_length;) { | 432 for (i = 0; i < new_length;) { |
412 const uint8_t value = depth[i]; | 433 const uint8_t value = depth[i]; |
413 size_t reps = 1; | 434 size_t reps = 1; |
414 if ((value != 0 && use_rle_for_non_zero) || | 435 if ((value != 0 && use_rle_for_non_zero) || |
415 (value == 0 && use_rle_for_zero)) { | 436 (value == 0 && use_rle_for_zero)) { |
416 for (size_t k = i + 1; k < new_length && depth[k] == value; ++k) { | 437 size_t k; |
| 438 for (k = i + 1; k < new_length && depth[k] == value; ++k) { |
417 ++reps; | 439 ++reps; |
418 } | 440 } |
419 } | 441 } |
420 if (value == 0) { | 442 if (value == 0) { |
421 WriteHuffmanTreeRepetitionsZeros(reps, tree_size, tree, extra_bits_data); | 443 BrotliWriteHuffmanTreeRepetitionsZeros( |
| 444 reps, tree_size, tree, extra_bits_data); |
422 } else { | 445 } else { |
423 WriteHuffmanTreeRepetitions(previous_value, | 446 BrotliWriteHuffmanTreeRepetitions(previous_value, |
424 value, reps, tree_size, | 447 value, reps, tree_size, |
425 tree, extra_bits_data); | 448 tree, extra_bits_data); |
426 previous_value = value; | 449 previous_value = value; |
427 } | 450 } |
428 i += reps; | 451 i += reps; |
429 } | 452 } |
430 } | 453 } |
431 | 454 |
432 namespace { | 455 static uint16_t BrotliReverseBits(size_t num_bits, uint16_t bits) { |
433 | 456 static const size_t kLut[16] = { /* Pre-reversed 4-bit values. */ |
434 uint16_t ReverseBits(int num_bits, uint16_t bits) { | |
435 static const size_t kLut[16] = { // Pre-reversed 4-bit values. | |
436 0x0, 0x8, 0x4, 0xc, 0x2, 0xa, 0x6, 0xe, | 457 0x0, 0x8, 0x4, 0xc, 0x2, 0xa, 0x6, 0xe, |
437 0x1, 0x9, 0x5, 0xd, 0x3, 0xb, 0x7, 0xf | 458 0x1, 0x9, 0x5, 0xd, 0x3, 0xb, 0x7, 0xf |
438 }; | 459 }; |
439 size_t retval = kLut[bits & 0xf]; | 460 size_t retval = kLut[bits & 0xf]; |
440 for (int i = 4; i < num_bits; i += 4) { | 461 size_t i; |
| 462 for (i = 4; i < num_bits; i += 4) { |
441 retval <<= 4; | 463 retval <<= 4; |
442 bits = static_cast<uint16_t>(bits >> 4); | 464 bits = (uint16_t)(bits >> 4); |
443 retval |= kLut[bits & 0xf]; | 465 retval |= kLut[bits & 0xf]; |
444 } | 466 } |
445 retval >>= (-num_bits & 0x3); | 467 retval >>= ((0 - num_bits) & 0x3); |
446 return static_cast<uint16_t>(retval); | 468 return (uint16_t)retval; |
447 } | 469 } |
448 | 470 |
449 } // namespace | 471 /* 0..15 are values for bits */ |
| 472 #define MAX_HUFFMAN_BITS 16 |
450 | 473 |
451 void ConvertBitDepthsToSymbols(const uint8_t *depth, | 474 void BrotliConvertBitDepthsToSymbols(const uint8_t *depth, |
452 size_t len, | 475 size_t len, |
453 uint16_t *bits) { | 476 uint16_t *bits) { |
454 // In Brotli, all bit depths are [1..15] | 477 /* In Brotli, all bit depths are [1..15] |
455 // 0 bit depth means that the symbol does not exist. | 478 0 bit depth means that the symbol does not exist. */ |
456 const int kMaxBits = 16; // 0..15 are values for bits | 479 uint16_t bl_count[MAX_HUFFMAN_BITS] = { 0 }; |
457 uint16_t bl_count[kMaxBits] = { 0 }; | 480 uint16_t next_code[MAX_HUFFMAN_BITS]; |
458 { | 481 size_t i; |
459 for (size_t i = 0; i < len; ++i) { | 482 int code = 0; |
460 ++bl_count[depth[i]]; | 483 for (i = 0; i < len; ++i) { |
461 } | 484 ++bl_count[depth[i]]; |
462 bl_count[0] = 0; | |
463 } | 485 } |
464 uint16_t next_code[kMaxBits]; | 486 bl_count[0] = 0; |
465 next_code[0] = 0; | 487 next_code[0] = 0; |
466 { | 488 for (i = 1; i < MAX_HUFFMAN_BITS; ++i) { |
467 int code = 0; | 489 code = (code + bl_count[i - 1]) << 1; |
468 for (int bits = 1; bits < kMaxBits; ++bits) { | 490 next_code[i] = (uint16_t)code; |
469 code = (code + bl_count[bits - 1]) << 1; | |
470 next_code[bits] = static_cast<uint16_t>(code); | |
471 } | |
472 } | 491 } |
473 for (size_t i = 0; i < len; ++i) { | 492 for (i = 0; i < len; ++i) { |
474 if (depth[i]) { | 493 if (depth[i]) { |
475 bits[i] = ReverseBits(depth[i], next_code[depth[i]]++); | 494 bits[i] = BrotliReverseBits(depth[i], next_code[depth[i]]++); |
476 } | 495 } |
477 } | 496 } |
478 } | 497 } |
479 | 498 |
480 } // namespace brotli | 499 #if defined(__cplusplus) || defined(c_plusplus) |
| 500 } /* extern "C" */ |
| 501 #endif |
OLD | NEW |