/src/brunsli/c/common/context.h
Line | Count | Source |
1 | | // Copyright (c) Google LLC 2019 |
2 | | // |
3 | | // Use of this source code is governed by an MIT-style |
4 | | // license that can be found in the LICENSE file or at |
5 | | // https://opensource.org/licenses/MIT. |
6 | | |
7 | | #ifndef BRUNSLI_COMMON_CONTEXT_H_ |
8 | | #define BRUNSLI_COMMON_CONTEXT_H_ |
9 | | |
10 | | #include <vector> |
11 | | |
12 | | #include "./distributions.h" |
13 | | #include <brunsli/jpeg_data.h> |
14 | | #include "./platform.h" |
15 | | #include <brunsli/types.h> |
16 | | |
17 | | namespace brunsli { |
18 | | |
19 | | static const size_t kMaxAverageContext = 8; |
20 | | static const size_t kNumAvrgContexts = kMaxAverageContext + 1u; |
21 | | // 6 bits allow encoding values 0..63; this range represents the possible |
22 | | // quantities of non-zero AC coefficients in the DCT block. |
23 | | static const size_t kNumNonZeroBits = 6u; |
24 | | /** |
25 | | * "number of non-zeros" value is decoded as a series of bits, |
26 | | * highest to lowest. |
27 | | * |
28 | | * Partially decoded value is used as a context for reading the next bit. |
29 | | * Contexts are organized in a binary tree. There are 64 final values, thus |
30 | | * there are 1-less non-leaf nodes. |
31 | | * Also, this constant also denotes the maximal value that could be encoded. |
32 | | * |
33 | | * static_assert(kNumNonZeroTreeSize == kDCTBlockSize - 1u) |
34 | | */ |
35 | | static const size_t kNumNonZeroTreeSize = (1u << kNumNonZeroBits) - 1u; |
36 | | static const size_t kNumNonZeroQuant = 2u; |
37 | | static const size_t kNumNonZeroContextMax = |
38 | | kNumNonZeroTreeSize / kNumNonZeroQuant; |
39 | | static const size_t kNumNonZeroContextCount = kNumNonZeroContextMax + 1u; |
40 | | |
41 | | static const uint8_t kNonzeroBuckets[64] = { |
42 | | 0, 1, 2, 3, 4, 4, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, |
43 | | 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
44 | | 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 10, 10, 10, |
45 | | 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, |
46 | | }; |
47 | | // kNonzeroBuckets[i] < kNumNonzeroBuckets |
48 | | static const uint8_t kNumNonzeroBuckets = 11; |
49 | | |
50 | | static const int kNumSchemes = 7; |
51 | | |
52 | | static const uint8_t kFreqContext[kNumSchemes][64] = { |
53 | | { |
54 | | 0, |
55 | | }, |
56 | | |
57 | | { |
58 | | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
59 | | 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, |
60 | | 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, |
61 | | }, |
62 | | |
63 | | { |
64 | | 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
65 | | 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, |
66 | | 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 1, 1, 1, |
67 | | }, |
68 | | |
69 | | { |
70 | | 0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, |
71 | | 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, |
72 | | 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 2, 2, 2, |
73 | | }, |
74 | | |
75 | | { |
76 | | 0, 1, 2, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 8, 8, |
77 | | 9, 9, 9, 9, 10, 10, 10, 10, 11, 11, 11, 11, 12, 12, 12, 12, |
78 | | 13, 13, 13, 13, 13, 13, 13, 13, 14, 14, 14, 14, 14, 14, 14, 14, |
79 | | 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, |
80 | | }, |
81 | | |
82 | | { |
83 | | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, |
84 | | 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 23, 23, |
85 | | 24, 24, 24, 24, 25, 25, 25, 25, 26, 26, 26, 26, 27, 27, 27, 27, |
86 | | 28, 28, 28, 28, 29, 29, 29, 29, 30, 30, 30, 30, 31, 31, 31, 31, |
87 | | }, |
88 | | |
89 | | { |
90 | | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, |
91 | | 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, |
92 | | 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, |
93 | | 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, |
94 | | }, |
95 | | }; |
96 | | |
97 | | static const uint16_t kNumNonzeroContext[kNumSchemes][64] = { |
98 | | {0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, |
99 | | 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, |
100 | | 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7}, |
101 | | {0, 2, 2, 4, 4, 4, 6, 6, 6, 6, 8, 8, 8, 8, 8, 8, |
102 | | 10, 10, 10, 10, 10, 10, 10, 10, 12, 12, 12, 12, 12, 12, 12, 12, |
103 | | 12, 12, 12, 12, 12, 12, 12, 12, 14, 14, 14, 14, 14, 14, 14, 14, |
104 | | 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14}, |
105 | | {0, 4, 4, 8, 8, 8, 12, 12, 12, 12, 16, 16, 16, 16, 16, 16, |
106 | | 20, 20, 20, 20, 20, 20, 20, 20, 24, 24, 24, 24, 24, 24, 24, 24, |
107 | | 24, 24, 24, 24, 24, 24, 24, 24, 28, 28, 28, 28, 28, 28, 28, 28, |
108 | | 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28}, |
109 | | {0, 8, 8, 16, 16, 16, 24, 24, 24, 24, 32, 32, 32, 32, 32, 32, |
110 | | 40, 40, 40, 40, 40, 40, 40, 40, 48, 48, 48, 48, 48, 48, 48, 48, |
111 | | 48, 48, 48, 48, 48, 48, 48, 48, 55, 55, 55, 55, 55, 55, 55, 55, |
112 | | 55, 55, 55, 55, 55, 55, 55, 55, 55, 55, 55, 55, 55, 55, 55, 55}, |
113 | | {0, 16, 16, 32, 32, 32, 48, 48, 48, 48, 64, 64, 64, |
114 | | 64, 64, 64, 80, 80, 80, 80, 80, 80, 80, 80, 95, 95, |
115 | | 95, 95, 95, 95, 95, 95, 95, 95, 95, 95, 95, 95, 95, |
116 | | 95, 109, 109, 109, 109, 109, 109, 109, 109, 109, 109, 109, 109, |
117 | | 109, 109, 109, 109, 109, 109, 109, 109, 109, 109, 109, 109}, |
118 | | {0, 32, 32, 64, 64, 64, 96, 96, 96, 96, 127, 127, 127, |
119 | | 127, 127, 127, 157, 157, 157, 157, 157, 157, 157, 157, 185, 185, |
120 | | 185, 185, 185, 185, 185, 185, 185, 185, 185, 185, 185, 185, 185, |
121 | | 185, 211, 211, 211, 211, 211, 211, 211, 211, 211, 211, 211, 211, |
122 | | 211, 211, 211, 211, 211, 211, 211, 211, 211, 211, 211, 211}, |
123 | | {0, 64, 64, 127, 127, 127, 188, 188, 188, 188, 246, 246, 246, |
124 | | 246, 246, 246, 300, 300, 300, 300, 300, 300, 300, 300, 348, 348, |
125 | | 348, 348, 348, 348, 348, 348, 348, 348, 348, 348, 348, 348, 348, |
126 | | 348, 388, 388, 388, 388, 388, 388, 388, 388, 388, 388, 388, 388, |
127 | | 388, 388, 388, 388, 388, 388, 388, 388, 388, 388, 388, 388}}; |
128 | | |
129 | | static const uint16_t kNumNonzeroContextSkip[kNumSchemes] = {8, 15, 31, 61, |
130 | | 120, 231, 412}; |
131 | | |
132 | | /** |
133 | | * Table that specifies, how context is calculated. |
134 | | * |
135 | | * Each value corresponds to DCT coefficient and is a sum of flags: |
136 | | * - 1: context should be calculated using ACPredictContextRow |
137 | | * - 2: context should be calculated using ACPredictContextCol |
138 | | */ |
139 | | static const uint8_t kContextAlgorithm[128] = { |
140 | | // JPEG XL layout |
141 | | 0, 1, 1, 1, 1, 0, 0, 0, // |
142 | | 2, 3, 1, 1, 1, 0, 0, 0, // |
143 | | 2, 2, 0, 0, 0, 0, 0, 0, // |
144 | | 2, 2, 0, 0, 0, 0, 0, 0, // |
145 | | 2, 2, 0, 0, 0, 0, 0, 0, // |
146 | | 0, 0, 0, 0, 0, 0, 0, 0, // |
147 | | 0, 0, 0, 0, 0, 0, 0, 0, // |
148 | | 0, 0, 0, 0, 0, 0, 0, 0, |
149 | | // Legacy layout |
150 | | 0, 1, 1, 1, 1, 1, 1, 1, // |
151 | | 2, 0, 0, 0, 0, 0, 0, 0, // |
152 | | 2, 0, 0, 0, 0, 0, 0, 0, // |
153 | | 2, 0, 0, 0, 0, 0, 0, 0, // |
154 | | 2, 0, 0, 0, 0, 0, 0, 0, // |
155 | | 2, 0, 0, 0, 0, 0, 0, 0, // |
156 | | 2, 0, 0, 0, 0, 0, 0, 0, // |
157 | | 2, 0, 0, 0, 0, 0, 0, 0, |
158 | | }; |
159 | | |
160 | | inline uint16_t ZeroDensityContext(size_t nonzeros_left, size_t k, |
161 | 6.54M | size_t bits) { |
162 | 6.54M | return kNumNonzeroContext[bits][nonzeros_left] + kFreqContext[bits][k]; |
163 | 6.54M | } |
164 | | |
165 | | // Returns the context for the absolute value of the prediction error of |
166 | | // the next DC coefficient in column x, using the one row size ringbuffer of |
167 | | // previous absolute prediction errors in vals. |
168 | 4.42M | inline int WeightedAverageContextDC(const int* vals, int x) { |
169 | | // Since vals is a ringbuffer, vals[x] and vals[x + 1] refer to the |
170 | | // previous row. |
171 | 4.42M | int sum = 1 + vals[x - 2] + vals[x - 1] + vals[x] + vals[x + 1]; |
172 | 4.42M | if ((sum >> kMaxAverageContext) != 0) { |
173 | 268k | return kMaxAverageContext; |
174 | 268k | } |
175 | 4.15M | return Log2FloorNonZero(sum); |
176 | 4.42M | } |
177 | | |
178 | | /** |
179 | | * Calculates the context on the base of average of already decoded |
180 | | * neighbour values. |
181 | | * |
182 | | * It is considered that vals[0] represents the value 2 rows above the current, |
183 | | * while the (locally) previous elements represent the current row. If y < 2, |
184 | | * then vals[0] should be 0. |
185 | | * Elements (locally) around vals[prev_row_delta] correspond to the row above |
186 | | * currnent one. |
187 | | * |
188 | | * Values are summed up with the following weights: |
189 | | * |
190 | | * 0|0|1|0 |
191 | | * -+-+-+- |
192 | | * 0|1|2|1 |
193 | | * -+-+-+- |
194 | | * 1|2|*| |
195 | | * ^ |
196 | | * current position |
197 | | * |
198 | | * This method should not be invoked on the 0-th row or 0-th column. |
199 | | * It is also considered, that there are 2 extra fence columns before the 0-th |
200 | | * column and 1 fence column to the right of the last column, |
201 | | * all initialized with zeroes. |
202 | | */ |
203 | 3.74M | inline int WeightedAverageContext(const int* vals, int prev_row_delta) { |
204 | 3.74M | int sum = 4 + vals[0] + (vals[-kDCTBlockSize] + vals[prev_row_delta]) * 2 + |
205 | 3.74M | vals[-2 * kDCTBlockSize] + vals[prev_row_delta - kDCTBlockSize] + |
206 | 3.74M | vals[prev_row_delta + kDCTBlockSize]; |
207 | 3.74M | if ((sum >> (kMaxAverageContext + 2)) != 0) { |
208 | 1.04M | return kMaxAverageContext; |
209 | 1.04M | } |
210 | 2.70M | return Log2FloorNonZero(sum) - 2; |
211 | 3.74M | } |
212 | | |
213 | | static const int kACPredictPrecisionBits = 13; |
214 | | static const int kACPredictPrecision = 1 << kACPredictPrecisionBits; |
215 | | |
216 | | void ComputeACPredictMultipliers(const int* quant, int* mult_row, |
217 | | int* mult_col); |
218 | | |
219 | | // Computes average and sign context from the AC prediction. |
220 | 2.68M | inline void ACPredictContext(int64_t p, size_t* avg_ctx, size_t* sgn) { |
221 | 2.68M | int multiplier; |
222 | 2.68M | if (p >= 0) { |
223 | 1.61M | multiplier = 1; |
224 | 1.61M | } else { |
225 | 1.06M | multiplier = -1; |
226 | 1.06M | p = -p; |
227 | 1.06M | } |
228 | 2.68M | size_t ctx; |
229 | 2.68M | if (p >= (1u << kMaxAverageContext)) { |
230 | 1.68M | ctx = kMaxAverageContext; |
231 | 1.68M | } else { |
232 | | // 0 -> 0, 1 -> 1, 2..3 -> 2, 4..7 -> 3, etc. |
233 | 997k | ctx = Log2FloorNonZero(2 * static_cast<uint32_t>(p)+ 1); |
234 | 997k | } |
235 | 2.68M | *avg_ctx = ctx; |
236 | 2.68M | *sgn = kMaxAverageContext + multiplier * ctx; |
237 | 2.68M | } |
238 | | |
239 | | inline void ACPredictContextCol(const coeff_t* prev, const coeff_t* cur, |
240 | 1.42M | const int* mult, size_t* avg_ctx, size_t* sgn) { |
241 | 1.42M | coeff_t terms[8]; |
242 | 1.42M | terms[0] = 0; |
243 | 1.42M | terms[1] = cur[1] + prev[1]; |
244 | 1.42M | terms[2] = cur[2] - prev[2]; |
245 | 1.42M | terms[3] = cur[3] + prev[3]; |
246 | 1.42M | terms[4] = cur[4] - prev[4]; |
247 | 1.42M | terms[5] = cur[5] + prev[5]; |
248 | 1.42M | terms[6] = cur[6] - prev[6]; |
249 | 1.42M | terms[7] = cur[7] + prev[7]; |
250 | 1.42M | int64_t delta = terms[0] * static_cast<int64_t>(mult[0]) + |
251 | 1.42M | terms[1] * static_cast<int64_t>(mult[1]) + |
252 | 1.42M | terms[2] * static_cast<int64_t>(mult[2]) + |
253 | 1.42M | terms[3] * static_cast<int64_t>(mult[3]) + |
254 | 1.42M | terms[4] * static_cast<int64_t>(mult[4]) + |
255 | 1.42M | terms[5] * static_cast<int64_t>(mult[5]) + |
256 | 1.42M | terms[6] * static_cast<int64_t>(mult[6]) + |
257 | 1.42M | terms[7] * static_cast<int64_t>(mult[7]); |
258 | 1.42M | ACPredictContext(prev[0] - delta / kACPredictPrecision, avg_ctx, sgn); |
259 | 1.42M | } |
260 | | |
261 | | inline void ACPredictContextRow(const coeff_t* prev, const coeff_t* cur, |
262 | 1.26M | const int* mult, size_t* avg_ctx, size_t* sgn) { |
263 | 1.26M | coeff_t terms[8]; |
264 | 1.26M | terms[0] = 0; |
265 | 1.26M | terms[1] = cur[8] + prev[8]; |
266 | 1.26M | terms[2] = cur[16] - prev[16]; |
267 | 1.26M | terms[3] = cur[24] + prev[24]; |
268 | 1.26M | terms[4] = cur[32] - prev[32]; |
269 | 1.26M | terms[5] = cur[40] + prev[40]; |
270 | 1.26M | terms[6] = cur[48] - prev[48]; |
271 | 1.26M | terms[7] = cur[56] + prev[56]; |
272 | 1.26M | int64_t delta = terms[0] * static_cast<int64_t>(mult[0]) + |
273 | 1.26M | terms[1] * static_cast<int64_t>(mult[1]) + |
274 | 1.26M | terms[2] * static_cast<int64_t>(mult[2]) + |
275 | 1.26M | terms[3] * static_cast<int64_t>(mult[3]) + |
276 | 1.26M | terms[4] * static_cast<int64_t>(mult[4]) + |
277 | 1.26M | terms[5] * static_cast<int64_t>(mult[5]) + |
278 | 1.26M | terms[6] * static_cast<int64_t>(mult[6]) + |
279 | 1.26M | terms[7] * static_cast<int64_t>(mult[7]); |
280 | 1.26M | ACPredictContext(prev[0] - delta / kACPredictPrecision, avg_ctx, sgn); |
281 | 1.26M | } |
282 | | |
283 | | /** |
284 | | * PRECONDITION: 0 <= prev[i] <= 63 |
285 | | * PRECONDITION: elements of prev at and after x correspond to previous |
286 | | * row; elements before x correspond to current row |
287 | | */ |
288 | 3.17M | inline uint8_t NumNonzerosContext(const uint8_t* prev, int x, int y) { |
289 | 3.17M | size_t prediction; |
290 | 3.17M | if (y == 0) { |
291 | 237k | if (x == 0) { |
292 | | // Special case: top-left block. |
293 | 5.90k | prediction = 0; |
294 | 231k | } else { |
295 | | // No row above; use block at left. |
296 | 231k | prediction = prev[x - 1]; |
297 | 231k | } |
298 | 2.93M | } else if (x == 0) { |
299 | | // No column to the left; use block above. |
300 | 64.5k | prediction = prev[x]; |
301 | 2.86M | } else { |
302 | | // Average of left and above blocks. |
303 | 2.86M | prediction = (prev[x - 1] + prev[x] + 1) / 2; |
304 | 2.86M | } |
305 | 3.17M | BRUNSLI_DCHECK(prediction <= kNumNonZeroTreeSize); |
306 | 3.17M | return static_cast<uint8_t>(prediction / kNumNonZeroQuant); |
307 | 3.17M | } |
308 | | |
309 | | // Context for the emptyness of a block is the number of non-empty blocks in the |
310 | | // previous and up neighborhood (blocks beyond the border are assumed |
311 | | // non-empty). |
312 | | static const int kNumIsEmptyBlockContexts = 3; |
313 | 71.9M | inline int IsEmptyBlockContext(const int* prev, int x) { |
314 | 71.9M | return prev[x - 1] + prev[x]; |
315 | 71.9M | } |
316 | | |
317 | | // Holds all encoding/decoding state for an image component that is needed to |
318 | | // switch between components during interleaved encoding/decoding. |
319 | | struct ComponentStateDC { |
320 | | ComponentStateDC() |
321 | 12.4k | : width(0), |
322 | 12.4k | is_empty_block_prob(kNumIsEmptyBlockContexts), |
323 | 12.4k | sign_prob(9), |
324 | 12.4k | first_extra_bit_prob(10) { |
325 | 12.4k | InitAll(); |
326 | 12.4k | } |
327 | | |
328 | 12.4k | void SetWidth(int w) { |
329 | 12.4k | width = w; |
330 | 12.4k | prev_is_nonempty.resize(w + 1, 1); |
331 | 12.4k | prev_abs_coeff.resize(w + 3); |
332 | 12.4k | prev_sign.resize(w + 1); |
333 | 12.4k | } |
334 | | |
335 | | int width; |
336 | | Prob is_zero_prob; |
337 | | std::vector<Prob> is_empty_block_prob; |
338 | | std::vector<Prob> sign_prob; |
339 | | std::vector<Prob> first_extra_bit_prob; |
340 | | std::vector<int> prev_is_nonempty; |
341 | | std::vector<int> prev_abs_coeff; |
342 | | std::vector<int> prev_sign; |
343 | | |
344 | | protected: |
345 | | void InitAll(); |
346 | | }; |
347 | | |
348 | | struct ComponentState { |
349 | | ComponentState() |
350 | 11.2k | : width(0), |
351 | 11.2k | is_zero_prob(kNumNonzeroBuckets * kDCTBlockSize), |
352 | 11.2k | sign_prob((2 * kMaxAverageContext + 1) * kDCTBlockSize), |
353 | 11.2k | first_extra_bit_prob(10 * kDCTBlockSize) { |
354 | 11.2k | InitAll(); |
355 | 11.2k | } |
356 | | |
357 | 11.2k | void SetWidth(int w) { |
358 | 11.2k | width = w; |
359 | 11.2k | prev_is_nonempty.resize(w + 1, 1); |
360 | 11.2k | prev_num_nonzeros.resize(w); |
361 | 11.2k | prev_abs_coeff.resize(kDCTBlockSize * 2 * (w + 3)); |
362 | 11.2k | prev_sign.resize(kDCTBlockSize * (w + 1)); |
363 | 11.2k | } |
364 | | |
365 | | // Returns the size of the object after constructor and SetWidth(w). |
366 | | // Used in estimating peak heap memory usage of the brunsli codec. |
367 | 0 | static size_t SizeInBytes(int w) { |
368 | 0 | return (4 + (10 + 3 * w) * kDCTBlockSize + 2 * w) * sizeof(int) + |
369 | 0 | ((kNumNonzeroBuckets + 2 * kMaxAverageContext + 11) * kDCTBlockSize + |
370 | 0 | kNumNonZeroContextCount * kNumNonZeroTreeSize) * |
371 | 0 | sizeof(Prob); |
372 | 0 | } |
373 | | |
374 | | int width; |
375 | | int context_offset; |
376 | | uint32_t order[kDCTBlockSize]; |
377 | | int mult_row[kDCTBlockSize]; |
378 | | // mult_col is transposed for more effective ACPredictContextRow execution. |
379 | | int mult_col[kDCTBlockSize]; |
380 | | std::vector<Prob> is_zero_prob; |
381 | | std::vector<Prob> sign_prob; |
382 | | Prob num_nonzero_prob[kNumNonZeroContextCount * kNumNonZeroTreeSize]; |
383 | | std::vector<Prob> first_extra_bit_prob; |
384 | | std::vector<int> prev_is_nonempty; |
385 | | std::vector<uint8_t> prev_num_nonzeros; |
386 | | std::vector<int> prev_abs_coeff; |
387 | | std::vector<int> prev_sign; |
388 | | |
389 | | protected: |
390 | | void InitAll(); |
391 | | }; |
392 | | |
393 | | } // namespace brunsli |
394 | | |
395 | | #endif // BRUNSLI_COMMON_CONTEXT_H_ |