Line | Branch | Exec | Source |
---|---|---|---|
1 | /* | ||
2 | * Copyright (C) 2007 Vitor Sessak <vitor1001@gmail.com> | ||
3 | * | ||
4 | * This file is part of FFmpeg. | ||
5 | * | ||
6 | * FFmpeg is free software; you can redistribute it and/or | ||
7 | * modify it under the terms of the GNU Lesser General Public | ||
8 | * License as published by the Free Software Foundation; either | ||
9 | * version 2.1 of the License, or (at your option) any later version. | ||
10 | * | ||
11 | * FFmpeg is distributed in the hope that it will be useful, | ||
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
14 | * Lesser General Public License for more details. | ||
15 | * | ||
16 | * You should have received a copy of the GNU Lesser General Public | ||
17 | * License along with FFmpeg; if not, write to the Free Software | ||
18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
19 | */ | ||
20 | |||
21 | /** | ||
22 | * @file | ||
23 | * Codebook Generator using the ELBG algorithm | ||
24 | */ | ||
25 | |||
26 | #include <string.h> | ||
27 | |||
28 | #include "libavutil/avassert.h" | ||
29 | #include "libavutil/common.h" | ||
30 | #include "libavutil/lfg.h" | ||
31 | #include "libavutil/mem.h" | ||
32 | #include "elbg.h" | ||
33 | |||
34 | #define DELTA_ERR_MAX 0.1 ///< Precision of the ELBG algorithm (as percentage error) | ||
35 | |||
36 | /** | ||
37 | * In the ELBG jargon, a cell is the set of points that are closest to a | ||
38 | * codebook entry. Not to be confused with a RoQ Video cell. */ | ||
39 | typedef struct cell_s { | ||
40 | int index; | ||
41 | struct cell_s *next; | ||
42 | } cell; | ||
43 | |||
44 | /** | ||
45 | * ELBG internal data | ||
46 | */ | ||
47 | typedef struct ELBGContext { | ||
48 | int error; | ||
49 | int dim; | ||
50 | int num_cb; | ||
51 | int *codebook; | ||
52 | cell **cells; | ||
53 | int *utility; | ||
54 | int *utility_inc; | ||
55 | int *nearest_cb; | ||
56 | int *points; | ||
57 | int *temp_points; | ||
58 | int *size_part; | ||
59 | AVLFG *rand_state; | ||
60 | int *scratchbuf; | ||
61 | cell *cell_buffer; | ||
62 | |||
63 | /* Sizes for the buffers above. Pointers without such a field | ||
64 | * are not allocated by us and only valid for the duration | ||
65 | * of a single call to avpriv_elbg_do(). */ | ||
66 | unsigned utility_allocated; | ||
67 | unsigned utility_inc_allocated; | ||
68 | unsigned size_part_allocated; | ||
69 | unsigned cells_allocated; | ||
70 | unsigned scratchbuf_allocated; | ||
71 | unsigned cell_buffer_allocated; | ||
72 | unsigned temp_points_allocated; | ||
73 | } ELBGContext; | ||
74 | |||
75 | 2466940895 | static inline int distance_limited(int *a, int *b, int dim, int limit) | |
76 | { | ||
77 | 2466940895 | int i, dist=0; | |
78 |
2/2✓ Branch 0 taken 4854074942 times.
✓ Branch 1 taken 274545884 times.
|
5128620826 | for (i=0; i<dim; i++) { |
79 | 4854074942 | int64_t distance = a[i] - b[i]; | |
80 | |||
81 | 4854074942 | distance *= distance; | |
82 |
2/2✓ Branch 0 taken 2192395011 times.
✓ Branch 1 taken 2661679931 times.
|
4854074942 | if (dist >= limit - distance) |
83 | 2192395011 | return limit; | |
84 | 2661679931 | dist += distance; | |
85 | } | ||
86 | |||
87 | 274545884 | return dist; | |
88 | } | ||
89 | |||
90 | 17717679 | static inline void vect_division(int *res, int *vect, int div, int dim) | |
91 | { | ||
92 | int i; | ||
93 |
2/2✓ Branch 0 taken 12934060 times.
✓ Branch 1 taken 4783619 times.
|
17717679 | if (div > 1) |
94 |
2/2✓ Branch 0 taken 55987062 times.
✓ Branch 1 taken 12934060 times.
|
68921122 | for (i=0; i<dim; i++) |
95 |
1/2✓ Branch 0 taken 55987062 times.
✗ Branch 1 not taken.
|
55987062 | res[i] = ROUNDED_DIV(vect[i],div); |
96 |
2/2✓ Branch 0 taken 733096 times.
✓ Branch 1 taken 4050523 times.
|
4783619 | else if (res != vect) |
97 | 733096 | memcpy(res, vect, dim*sizeof(int)); | |
98 | |||
99 | 17717679 | } | |
100 | |||
101 | 2884774 | static int eval_error_cell(ELBGContext *elbg, int *centroid, cell *cells) | |
102 | { | ||
103 | 2884774 | int error=0; | |
104 |
2/2✓ Branch 0 taken 25169204 times.
✓ Branch 1 taken 2884774 times.
|
28053978 | for (; cells; cells=cells->next) { |
105 | 25169204 | int distance = distance_limited(centroid, elbg->points + cells->index*elbg->dim, elbg->dim, INT_MAX); | |
106 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 25169204 times.
|
25169204 | if (error >= INT_MAX - distance) |
107 | ✗ | return INT_MAX; | |
108 | 25169204 | error += distance; | |
109 | } | ||
110 | |||
111 | 2884774 | return error; | |
112 | } | ||
113 | |||
114 | 5872699 | static int get_closest_codebook(ELBGContext *elbg, int index) | |
115 | { | ||
116 | 5872699 | int pick = 0; | |
117 |
2/2✓ Branch 0 taken 227277060 times.
✓ Branch 1 taken 5872699 times.
|
233149759 | for (int i = 0, diff_min = INT_MAX; i < elbg->num_cb; i++) |
118 |
2/2✓ Branch 0 taken 221404361 times.
✓ Branch 1 taken 5872699 times.
|
227277060 | if (i != index) { |
119 | int diff; | ||
120 | 221404361 | diff = distance_limited(elbg->codebook + i*elbg->dim, elbg->codebook + index*elbg->dim, elbg->dim, diff_min); | |
121 |
2/2✓ Branch 0 taken 11934164 times.
✓ Branch 1 taken 209470197 times.
|
221404361 | if (diff < diff_min) { |
122 | 11934164 | pick = i; | |
123 | 11934164 | diff_min = diff; | |
124 | } | ||
125 | } | ||
126 | 5872699 | return pick; | |
127 | } | ||
128 | |||
129 | 5872699 | static int get_high_utility_cell(ELBGContext *elbg) | |
130 | { | ||
131 | 5872699 | int i=0; | |
132 | /* Using linear search, do binary if it ever turns to be speed critical */ | ||
133 | uint64_t r; | ||
134 | |||
135 |
1/2✓ Branch 0 taken 5872699 times.
✗ Branch 1 not taken.
|
5872699 | if (elbg->utility_inc[elbg->num_cb - 1] < INT_MAX) { |
136 | 5872699 | r = av_lfg_get(elbg->rand_state) % (unsigned int)elbg->utility_inc[elbg->num_cb - 1] + 1; | |
137 | } else { | ||
138 | ✗ | r = av_lfg_get(elbg->rand_state); | |
139 | ✗ | r = (av_lfg_get(elbg->rand_state) + (r<<32)) % elbg->utility_inc[elbg->num_cb - 1] + 1; | |
140 | } | ||
141 | |||
142 |
2/2✓ Branch 0 taken 109370285 times.
✓ Branch 1 taken 5872699 times.
|
115242984 | while (elbg->utility_inc[i] < r) { |
143 | 109370285 | i++; | |
144 | } | ||
145 | |||
146 | av_assert2(elbg->cells[i]); | ||
147 | |||
148 | 5872699 | return i; | |
149 | } | ||
150 | |||
151 | /** | ||
152 | * Implementation of the simple LBG algorithm for just two codebooks | ||
153 | */ | ||
154 | 1442387 | static int simple_lbg(ELBGContext *elbg, | |
155 | int dim, | ||
156 | int *centroid[3], | ||
157 | int newutility[3], | ||
158 | int *points, | ||
159 | cell *cells) | ||
160 | { | ||
161 | int i, idx; | ||
162 | 1442387 | int numpoints[2] = {0,0}; | |
163 | 1442387 | int *newcentroid[2] = { | |
164 | 1442387 | elbg->scratchbuf + 3*dim, | |
165 | 1442387 | elbg->scratchbuf + 4*dim | |
166 | }; | ||
167 | cell *tempcell; | ||
168 | |||
169 | 1442387 | memset(newcentroid[0], 0, 2 * dim * sizeof(*newcentroid[0])); | |
170 | |||
171 | 1442387 | newutility[0] = | |
172 | 1442387 | newutility[1] = 0; | |
173 | |||
174 |
2/2✓ Branch 0 taken 19693397 times.
✓ Branch 1 taken 1442387 times.
|
21135784 | for (tempcell = cells; tempcell; tempcell=tempcell->next) { |
175 | 19693397 | idx = distance_limited(centroid[0], points + tempcell->index*dim, dim, INT_MAX)>= | |
176 | 19693397 | distance_limited(centroid[1], points + tempcell->index*dim, dim, INT_MAX); | |
177 | 19693397 | numpoints[idx]++; | |
178 |
2/2✓ Branch 0 taken 121291626 times.
✓ Branch 1 taken 19693397 times.
|
140985023 | for (i=0; i<dim; i++) |
179 | 121291626 | newcentroid[idx][i] += points[tempcell->index*dim + i]; | |
180 | } | ||
181 | |||
182 | 1442387 | vect_division(centroid[0], newcentroid[0], numpoints[0], dim); | |
183 | 1442387 | vect_division(centroid[1], newcentroid[1], numpoints[1], dim); | |
184 | |||
185 |
2/2✓ Branch 0 taken 19693397 times.
✓ Branch 1 taken 1442387 times.
|
21135784 | for (tempcell = cells; tempcell; tempcell=tempcell->next) { |
186 | 19693397 | int dist[2] = {distance_limited(centroid[0], points + tempcell->index*dim, dim, INT_MAX), | |
187 | 19693397 | distance_limited(centroid[1], points + tempcell->index*dim, dim, INT_MAX)}; | |
188 | 19693397 | int idx = dist[0] > dist[1]; | |
189 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 19693397 times.
|
19693397 | if (newutility[idx] >= INT_MAX - dist[idx]) |
190 | ✗ | newutility[idx] = INT_MAX; | |
191 | else | ||
192 | 19693397 | newutility[idx] += dist[idx]; | |
193 | } | ||
194 | |||
195 |
1/2✓ Branch 0 taken 1442387 times.
✗ Branch 1 not taken.
|
1442387 | return (newutility[0] >= INT_MAX - newutility[1]) ? INT_MAX : newutility[0] + newutility[1]; |
196 | } | ||
197 | |||
198 | 1442387 | static void get_new_centroids(ELBGContext *elbg, int huc, int *newcentroid_i, | |
199 | int *newcentroid_p) | ||
200 | { | ||
201 | cell *tempcell; | ||
202 | 1442387 | int *min = newcentroid_i; | |
203 | 1442387 | int *max = newcentroid_p; | |
204 | int i; | ||
205 | |||
206 |
2/2✓ Branch 0 taken 8844582 times.
✓ Branch 1 taken 1442387 times.
|
10286969 | for (i=0; i< elbg->dim; i++) { |
207 | 8844582 | min[i]=INT_MAX; | |
208 | 8844582 | max[i]=0; | |
209 | } | ||
210 | |||
211 |
2/2✓ Branch 0 taken 19693397 times.
✓ Branch 1 taken 1442387 times.
|
21135784 | for (tempcell = elbg->cells[huc]; tempcell; tempcell = tempcell->next) |
212 |
2/2✓ Branch 0 taken 121291626 times.
✓ Branch 1 taken 19693397 times.
|
140985023 | for(i=0; i<elbg->dim; i++) { |
213 | 121291626 | min[i]=FFMIN(min[i], elbg->points[tempcell->index*elbg->dim + i]); | |
214 | 121291626 | max[i]=FFMAX(max[i], elbg->points[tempcell->index*elbg->dim + i]); | |
215 | } | ||
216 | |||
217 |
2/2✓ Branch 0 taken 8844582 times.
✓ Branch 1 taken 1442387 times.
|
10286969 | for (i=0; i<elbg->dim; i++) { |
218 | 8844582 | int ni = min[i] + (max[i] - min[i])/3; | |
219 | 8844582 | int np = min[i] + (2*(max[i] - min[i]))/3; | |
220 | 8844582 | newcentroid_i[i] = ni; | |
221 | 8844582 | newcentroid_p[i] = np; | |
222 | } | ||
223 | 1442387 | } | |
224 | |||
225 | /** | ||
226 | * Add the points in the low utility cell to its closest cell. Split the high | ||
227 | * utility cell, putting the separated points in the (now empty) low utility | ||
228 | * cell. | ||
229 | * | ||
230 | * @param elbg Internal elbg data | ||
231 | * @param indexes {luc, huc, cluc} | ||
232 | * @param newcentroid A vector with the position of the new centroids | ||
233 | */ | ||
234 | 740259 | static void shift_codebook(ELBGContext *elbg, int *indexes, | |
235 | int *newcentroid[3]) | ||
236 | { | ||
237 | cell *tempdata; | ||
238 | 740259 | cell **pp = &elbg->cells[indexes[2]]; | |
239 | |||
240 |
2/2✓ Branch 0 taken 5906853 times.
✓ Branch 1 taken 740259 times.
|
6647112 | while(*pp) |
241 | 5906853 | pp= &(*pp)->next; | |
242 | |||
243 | 740259 | *pp = elbg->cells[indexes[0]]; | |
244 | |||
245 | 740259 | elbg->cells[indexes[0]] = NULL; | |
246 | 740259 | tempdata = elbg->cells[indexes[1]]; | |
247 | 740259 | elbg->cells[indexes[1]] = NULL; | |
248 | |||
249 |
2/2✓ Branch 0 taken 8410723 times.
✓ Branch 1 taken 740259 times.
|
9150982 | while(tempdata) { |
250 | 8410723 | cell *tempcell2 = tempdata->next; | |
251 | 8410723 | int idx = distance_limited(elbg->points + tempdata->index*elbg->dim, | |
252 | 8410723 | newcentroid[0], elbg->dim, INT_MAX) > | |
253 | 8410723 | distance_limited(elbg->points + tempdata->index*elbg->dim, | |
254 | 8410723 | newcentroid[1], elbg->dim, INT_MAX); | |
255 | |||
256 | 8410723 | tempdata->next = elbg->cells[indexes[idx]]; | |
257 | 8410723 | elbg->cells[indexes[idx]] = tempdata; | |
258 | 8410723 | tempdata = tempcell2; | |
259 | } | ||
260 | 740259 | } | |
261 | |||
262 | 6568199 | static void evaluate_utility_inc(ELBGContext *elbg) | |
263 | { | ||
264 | 6568199 | int64_t inc=0; | |
265 | |||
266 |
2/2✓ Branch 0 taken 128838782 times.
✓ Branch 1 taken 6568199 times.
|
135406981 | for (int i = 0; i < elbg->num_cb; i++) { |
267 |
2/2✓ Branch 0 taken 44577128 times.
✓ Branch 1 taken 84261654 times.
|
128838782 | if (elbg->num_cb * (int64_t)elbg->utility[i] > elbg->error) |
268 | 44577128 | inc += elbg->utility[i]; | |
269 | 128838782 | elbg->utility_inc[i] = FFMIN(inc, INT_MAX); | |
270 | } | ||
271 | 6568199 | } | |
272 | |||
273 | |||
274 | 2220777 | static void update_utility_and_n_cb(ELBGContext *elbg, int idx, int newutility) | |
275 | { | ||
276 | cell *tempcell; | ||
277 | |||
278 | 2220777 | elbg->utility[idx] = newutility; | |
279 |
2/2✓ Branch 0 taken 18424544 times.
✓ Branch 1 taken 2220777 times.
|
20645321 | for (tempcell=elbg->cells[idx]; tempcell; tempcell=tempcell->next) |
280 | 18424544 | elbg->nearest_cb[tempcell->index] = idx; | |
281 | 2220777 | } | |
282 | |||
283 | /** | ||
284 | * Evaluate if a shift lower the error. If it does, call shift_codebooks | ||
285 | * and update elbg->error, elbg->utility and elbg->nearest_cb. | ||
286 | * | ||
287 | * @param elbg Internal elbg data | ||
288 | * @param idx {luc (low utility cell, huc (high utility cell), cluc (closest cell to low utility cell)} | ||
289 | */ | ||
290 | 1442387 | static void try_shift_candidate(ELBGContext *elbg, int idx[3]) | |
291 | { | ||
292 | 1442387 | int j, k, cont=0, tmp; | |
293 | 1442387 | int64_t olderror=0, newerror; | |
294 | int newutility[3]; | ||
295 | 1442387 | int *newcentroid[3] = { | |
296 | 1442387 | elbg->scratchbuf, | |
297 | 1442387 | elbg->scratchbuf + elbg->dim, | |
298 | 1442387 | elbg->scratchbuf + 2*elbg->dim | |
299 | }; | ||
300 | cell *tempcell; | ||
301 | |||
302 |
2/2✓ Branch 0 taken 4327161 times.
✓ Branch 1 taken 1442387 times.
|
5769548 | for (j=0; j<3; j++) |
303 | 4327161 | olderror += elbg->utility[idx[j]]; | |
304 | |||
305 | 1442387 | memset(newcentroid[2], 0, elbg->dim*sizeof(int)); | |
306 | |||
307 |
2/2✓ Branch 0 taken 2884774 times.
✓ Branch 1 taken 1442387 times.
|
4327161 | for (k=0; k<2; k++) |
308 |
2/2✓ Branch 0 taken 25169204 times.
✓ Branch 1 taken 2884774 times.
|
28053978 | for (tempcell=elbg->cells[idx[2*k]]; tempcell; tempcell=tempcell->next) { |
309 | 25169204 | cont++; | |
310 |
2/2✓ Branch 0 taken 158109438 times.
✓ Branch 1 taken 25169204 times.
|
183278642 | for (j=0; j<elbg->dim; j++) |
311 | 158109438 | newcentroid[2][j] += elbg->points[tempcell->index*elbg->dim + j]; | |
312 | } | ||
313 | |||
314 | 1442387 | vect_division(newcentroid[2], newcentroid[2], cont, elbg->dim); | |
315 | |||
316 | 1442387 | get_new_centroids(elbg, idx[1], newcentroid[0], newcentroid[1]); | |
317 | |||
318 | 1442387 | newutility[2] = eval_error_cell(elbg, newcentroid[2], elbg->cells[idx[0]]); | |
319 | 1442387 | tmp = eval_error_cell(elbg, newcentroid[2], elbg->cells[idx[2]]); | |
320 |
1/2✓ Branch 0 taken 1442387 times.
✗ Branch 1 not taken.
|
1442387 | newutility[2] = (tmp >= INT_MAX - newutility[2]) ? INT_MAX : newutility[2] + tmp; |
321 | |||
322 | 1442387 | newerror = newutility[2]; | |
323 | |||
324 | 1442387 | tmp = simple_lbg(elbg, elbg->dim, newcentroid, newutility, elbg->points, | |
325 | 1442387 | elbg->cells[idx[1]]); | |
326 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1442387 times.
|
1442387 | if (tmp >= INT_MAX - newerror) |
327 | ✗ | newerror = INT_MAX; | |
328 | else | ||
329 | 1442387 | newerror += tmp; | |
330 | |||
331 |
2/2✓ Branch 0 taken 740259 times.
✓ Branch 1 taken 702128 times.
|
1442387 | if (olderror > newerror) { |
332 | 740259 | shift_codebook(elbg, idx, newcentroid); | |
333 | |||
334 | 740259 | elbg->error += newerror - olderror; | |
335 | |||
336 |
2/2✓ Branch 0 taken 2220777 times.
✓ Branch 1 taken 740259 times.
|
2961036 | for (j=0; j<3; j++) |
337 | 2220777 | update_utility_and_n_cb(elbg, idx[j], newutility[j]); | |
338 | |||
339 | 740259 | evaluate_utility_inc(elbg); | |
340 | } | ||
341 | 1442387 | } | |
342 | |||
343 | /** | ||
344 | * Implementation of the ELBG block | ||
345 | */ | ||
346 | 5827940 | static void do_shiftings(ELBGContext *elbg) | |
347 | { | ||
348 | int idx[3]; | ||
349 | |||
350 | 5827940 | evaluate_utility_inc(elbg); | |
351 | |||
352 |
2/2✓ Branch 0 taken 13390518 times.
✓ Branch 1 taken 5827940 times.
|
19218458 | for (idx[0]=0; idx[0] < elbg->num_cb; idx[0]++) |
353 |
2/2✓ Branch 0 taken 5872699 times.
✓ Branch 1 taken 7517819 times.
|
13390518 | if (elbg->num_cb * (int64_t)elbg->utility[idx[0]] < elbg->error) { |
354 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 5872699 times.
|
5872699 | if (elbg->utility_inc[elbg->num_cb - 1] == 0) |
355 | ✗ | return; | |
356 | |||
357 | 5872699 | idx[1] = get_high_utility_cell(elbg); | |
358 | 5872699 | idx[2] = get_closest_codebook(elbg, idx[0]); | |
359 | |||
360 |
3/4✓ Branch 0 taken 5872699 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 1442387 times.
✓ Branch 3 taken 4430312 times.
|
5872699 | if (idx[1] != idx[0] && idx[1] != idx[2]) |
361 | 1442387 | try_shift_candidate(elbg, idx); | |
362 | } | ||
363 | } | ||
364 | |||
365 | 5790190 | static void do_elbg(ELBGContext *restrict elbg, int *points, int numpoints, | |
366 | int max_steps) | ||
367 | { | ||
368 | 5790190 | int *const size_part = elbg->size_part; | |
369 | 5790190 | int i, j, steps = 0; | |
370 | 5790190 | int best_idx = 0; | |
371 | int last_error; | ||
372 | |||
373 | 5790190 | elbg->error = INT_MAX; | |
374 | 5790190 | elbg->points = points; | |
375 | |||
376 | do { | ||
377 | 5827940 | cell *free_cells = elbg->cell_buffer; | |
378 | 5827940 | last_error = elbg->error; | |
379 | 5827940 | steps++; | |
380 | 5827940 | memset(elbg->utility, 0, elbg->num_cb * sizeof(*elbg->utility)); | |
381 | 5827940 | memset(elbg->cells, 0, elbg->num_cb * sizeof(*elbg->cells)); | |
382 | |||
383 | 5827940 | elbg->error = 0; | |
384 | |||
385 | /* This loop evaluate the actual Voronoi partition. It is the most | ||
386 | costly part of the algorithm. */ | ||
387 |
2/2✓ Branch 0 taken 76741726 times.
✓ Branch 1 taken 5827940 times.
|
82569666 | for (i=0; i < numpoints; i++) { |
388 | 76741726 | int best_dist = distance_limited(elbg->points + i * elbg->dim, | |
389 | 76741726 | elbg->codebook + best_idx * elbg->dim, | |
390 | elbg->dim, INT_MAX); | ||
391 |
2/2✓ Branch 0 taken 2048030570 times.
✓ Branch 1 taken 76741726 times.
|
2124772296 | for (int k = 0; k < elbg->num_cb; k++) { |
392 | 2048030570 | int dist = distance_limited(elbg->points + i * elbg->dim, | |
393 | 2048030570 | elbg->codebook + k * elbg->dim, | |
394 | elbg->dim, best_dist); | ||
395 |
2/2✓ Branch 0 taken 65105756 times.
✓ Branch 1 taken 1982924814 times.
|
2048030570 | if (dist < best_dist) { |
396 | 65105756 | best_dist = dist; | |
397 | 65105756 | best_idx = k; | |
398 | } | ||
399 | } | ||
400 | 76741726 | elbg->nearest_cb[i] = best_idx; | |
401 |
1/2✓ Branch 0 taken 76741726 times.
✗ Branch 1 not taken.
|
76741726 | elbg->error = (elbg->error >= INT_MAX - best_dist) ? INT_MAX : elbg->error + best_dist; |
402 | 153483452 | elbg->utility[elbg->nearest_cb[i]] = (elbg->utility[elbg->nearest_cb[i]] >= INT_MAX - best_dist) ? | |
403 |
1/2✓ Branch 0 taken 76741726 times.
✗ Branch 1 not taken.
|
76741726 | INT_MAX : elbg->utility[elbg->nearest_cb[i]] + best_dist; |
404 | 76741726 | free_cells->index = i; | |
405 | 76741726 | free_cells->next = elbg->cells[elbg->nearest_cb[i]]; | |
406 | 76741726 | elbg->cells[elbg->nearest_cb[i]] = free_cells; | |
407 | 76741726 | free_cells++; | |
408 | } | ||
409 | |||
410 | 5827940 | do_shiftings(elbg); | |
411 | |||
412 | 5827940 | memset(size_part, 0, elbg->num_cb * sizeof(*size_part)); | |
413 | |||
414 | 5827940 | memset(elbg->codebook, 0, elbg->num_cb * elbg->dim * sizeof(*elbg->codebook)); | |
415 | |||
416 |
2/2✓ Branch 0 taken 76741726 times.
✓ Branch 1 taken 5827940 times.
|
82569666 | for (i=0; i < numpoints; i++) { |
417 | 76741726 | size_part[elbg->nearest_cb[i]]++; | |
418 |
2/2✓ Branch 0 taken 327942996 times.
✓ Branch 1 taken 76741726 times.
|
404684722 | for (j=0; j < elbg->dim; j++) |
419 | 327942996 | elbg->codebook[elbg->nearest_cb[i]*elbg->dim + j] += | |
420 | 327942996 | elbg->points[i*elbg->dim + j]; | |
421 | } | ||
422 | |||
423 |
2/2✓ Branch 0 taken 13390518 times.
✓ Branch 1 taken 5827940 times.
|
19218458 | for (int i = 0; i < elbg->num_cb; i++) |
424 | 13390518 | vect_division(elbg->codebook + i*elbg->dim, | |
425 | 13390518 | elbg->codebook + i*elbg->dim, size_part[i], elbg->dim); | |
426 | |||
427 |
4/4✓ Branch 0 taken 5815243 times.
✓ Branch 1 taken 12697 times.
✓ Branch 2 taken 37750 times.
✓ Branch 3 taken 5777493 times.
|
5827940 | } while(((last_error - elbg->error) > DELTA_ERR_MAX*elbg->error) && |
428 | (steps < max_steps)); | ||
429 | 5790190 | } | |
430 | |||
431 | #define BIG_PRIME 433494437LL | ||
432 | |||
433 | /** | ||
434 | * Initialize the codebook vector for the elbg algorithm. | ||
435 | * If numpoints <= 24 * num_cb this function fills codebook with random numbers. | ||
436 | * If not, it calls do_elbg for a (smaller) random sample of the points in | ||
437 | * points. | ||
438 | */ | ||
439 | 5790190 | static void init_elbg(ELBGContext *restrict elbg, int *points, int *temp_points, | |
440 | int numpoints, int max_steps) | ||
441 | { | ||
442 | 5790190 | int dim = elbg->dim; | |
443 | |||
444 |
2/2✓ Branch 0 taken 30673 times.
✓ Branch 1 taken 5759517 times.
|
5790190 | if (numpoints > 24LL * elbg->num_cb) { |
445 | /* ELBG is very costly for a big number of points. So if we have a lot | ||
446 | of them, get a good initial codebook to save on iterations */ | ||
447 |
2/2✓ Branch 0 taken 1934047 times.
✓ Branch 1 taken 30673 times.
|
1964720 | for (int i = 0; i < numpoints / 8; i++) { |
448 | 1934047 | int k = (i*BIG_PRIME) % numpoints; | |
449 | 1934047 | memcpy(temp_points + i*dim, points + k*dim, dim * sizeof(*temp_points)); | |
450 | } | ||
451 | |||
452 | /* If anything is changed in the recursion parameters, | ||
453 | * the allocated size of temp_points will also need to be updated. */ | ||
454 | 30673 | init_elbg(elbg, temp_points, temp_points + numpoints / 8 * dim, | |
455 | numpoints / 8, 2 * max_steps); | ||
456 | 30673 | do_elbg(elbg, temp_points, numpoints / 8, 2 * max_steps); | |
457 | } else // If not, initialize the codebook with random positions | ||
458 |
2/2✓ Branch 0 taken 13040848 times.
✓ Branch 1 taken 5759517 times.
|
18800365 | for (int i = 0; i < elbg->num_cb; i++) |
459 | 13040848 | memcpy(elbg->codebook + i * dim, points + ((i*BIG_PRIME)%numpoints)*dim, | |
460 | dim * sizeof(*elbg->codebook)); | ||
461 | 5790190 | } | |
462 | |||
463 | 5759517 | int avpriv_elbg_do(ELBGContext **elbgp, int *points, int dim, int numpoints, | |
464 | int *codebook, int num_cb, int max_steps, | ||
465 | int *closest_cb, AVLFG *rand_state, uintptr_t flags) | ||
466 | { | ||
467 |
2/2✓ Branch 0 taken 5759507 times.
✓ Branch 1 taken 10 times.
|
5759517 | ELBGContext *const restrict elbg = *elbgp ? *elbgp : av_mallocz(sizeof(*elbg)); |
468 | |||
469 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 5759517 times.
|
5759517 | if (!elbg) |
470 | ✗ | return AVERROR(ENOMEM); | |
471 | 5759517 | *elbgp = elbg; | |
472 | |||
473 | 5759517 | elbg->nearest_cb = closest_cb; | |
474 | 5759517 | elbg->rand_state = rand_state; | |
475 | 5759517 | elbg->codebook = codebook; | |
476 | 5759517 | elbg->num_cb = num_cb; | |
477 | 5759517 | elbg->dim = dim; | |
478 | |||
479 | #define ALLOCATE_IF_NECESSARY(field, new_elements, multiplicator) \ | ||
480 | if (elbg->field ## _allocated < new_elements) { \ | ||
481 | av_freep(&elbg->field); \ | ||
482 | elbg->field = av_malloc_array(new_elements, \ | ||
483 | multiplicator * sizeof(*elbg->field)); \ | ||
484 | if (!elbg->field) { \ | ||
485 | elbg->field ## _allocated = 0; \ | ||
486 | return AVERROR(ENOMEM); \ | ||
487 | } \ | ||
488 | elbg->field ## _allocated = new_elements; \ | ||
489 | } | ||
490 | /* Allocating the buffers for do_elbg() here once relies | ||
491 | * on their size being always the same even when do_elbg() | ||
492 | * is called from init_elbg(). It also relies on do_elbg() | ||
493 | * never calling itself recursively. */ | ||
494 |
3/4✓ Branch 0 taken 29 times.
✓ Branch 1 taken 5759488 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 29 times.
|
5759517 | ALLOCATE_IF_NECESSARY(cells, num_cb, 1) |
495 |
3/4✓ Branch 0 taken 29 times.
✓ Branch 1 taken 5759488 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 29 times.
|
5759517 | ALLOCATE_IF_NECESSARY(utility, num_cb, 1) |
496 |
3/4✓ Branch 0 taken 29 times.
✓ Branch 1 taken 5759488 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 29 times.
|
5759517 | ALLOCATE_IF_NECESSARY(utility_inc, num_cb, 1) |
497 |
3/4✓ Branch 0 taken 29 times.
✓ Branch 1 taken 5759488 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 29 times.
|
5759517 | ALLOCATE_IF_NECESSARY(size_part, num_cb, 1) |
498 |
3/4✓ Branch 0 taken 17 times.
✓ Branch 1 taken 5759500 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 17 times.
|
5759517 | ALLOCATE_IF_NECESSARY(cell_buffer, numpoints, 1) |
499 |
3/4✓ Branch 0 taken 10 times.
✓ Branch 1 taken 5759507 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 10 times.
|
5759517 | ALLOCATE_IF_NECESSARY(scratchbuf, dim, 5) |
500 |
2/2✓ Branch 0 taken 23325 times.
✓ Branch 1 taken 5736192 times.
|
5759517 | if (numpoints > 24LL * elbg->num_cb) { |
501 | /* The first step in the recursion in init_elbg() needs a buffer with | ||
502 | * (numpoints / 8) * dim elements; the next step needs numpoints / 8 / 8 | ||
503 | * * dim elements etc. The geometric series leads to an upper bound of | ||
504 | * numpoints / 8 * 8 / 7 * dim elements. */ | ||
505 | 23325 | uint64_t prod = dim * (uint64_t)(numpoints / 7U); | |
506 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 23325 times.
|
23325 | if (prod > INT_MAX) |
507 | ✗ | return AVERROR(ERANGE); | |
508 |
3/4✓ Branch 0 taken 11 times.
✓ Branch 1 taken 23314 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 11 times.
|
23325 | ALLOCATE_IF_NECESSARY(temp_points, prod, 1) |
509 | } | ||
510 | |||
511 | 5759517 | init_elbg(elbg, points, elbg->temp_points, numpoints, max_steps); | |
512 | 5759517 | do_elbg (elbg, points, numpoints, max_steps); | |
513 | 5759517 | return 0; | |
514 | } | ||
515 | |||
516 | 10 | av_cold void avpriv_elbg_free(ELBGContext **elbgp) | |
517 | { | ||
518 | 10 | ELBGContext *elbg = *elbgp; | |
519 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
|
10 | if (!elbg) |
520 | ✗ | return; | |
521 | |||
522 | 10 | av_freep(&elbg->size_part); | |
523 | 10 | av_freep(&elbg->utility); | |
524 | 10 | av_freep(&elbg->cell_buffer); | |
525 | 10 | av_freep(&elbg->cells); | |
526 | 10 | av_freep(&elbg->utility_inc); | |
527 | 10 | av_freep(&elbg->scratchbuf); | |
528 | 10 | av_freep(&elbg->temp_points); | |
529 | |||
530 | 10 | av_freep(elbgp); | |
531 | } | ||
532 |