FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavcodec/svq1enc.c
Date: 2025-06-23 20:06:14
Exec Total Coverage
Lines: 347 361 96.1%
Functions: 8 8 100.0%
Branches: 160 184 87.0%
Line Branch Exec Source
1 /*
2 * SVQ1 Encoder
3 * Copyright (C) 2004 Mike Melanson <melanson@pcisys.net>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * Sorenson Vector Quantizer #1 (SVQ1) video codec.
25 * For more information of the SVQ1 algorithm, visit:
26 * http://www.pcisys.net/~melanson/codecs/
27 */
28
29 #include "libavutil/emms.h"
30 #include "libavutil/mem.h"
31 #include "avcodec.h"
32 #include "codec_internal.h"
33 #include "encode.h"
34 #include "hpeldsp.h"
35 #include "me_cmp.h"
36 #include "mpegvideo.h"
37 #include "h263.h"
38 #include "h263enc.h"
39 #include "internal.h"
40 #include "mpegutils.h"
41 #include "packet_internal.h"
42 #include "put_bits.h"
43 #include "svq1.h"
44 #include "svq1encdsp.h"
45 #include "svq1enc_cb.h"
46 #include "version.h"
47
48 #include "libavutil/avassert.h"
49 #include "libavutil/frame.h"
50 #include "libavutil/mem_internal.h"
51
52 // Workaround for GCC bug 102513
53 #if AV_GCC_VERSION_AT_LEAST(10, 0) && AV_GCC_VERSION_AT_MOST(12, 0) \
54 && !defined(__clang__) && !defined(__INTEL_COMPILER)
55 #pragma GCC optimize ("no-ipa-cp-clone")
56 #endif
57
58 typedef struct SVQ1EncContext {
59 /* FIXME: Needed for motion estimation, should not be used for anything
60 * else, the idea is to make the motion estimation eventually independent
61 * of MPVEncContext, so this will be removed then. */
62 MPVEncContext m;
63 AVCodecContext *avctx;
64 MECmpContext mecc;
65 AVFrame *current_picture;
66 AVFrame *last_picture;
67
68 /* Some compression statistics */
69 enum AVPictureType pict_type;
70 int quality;
71
72 /* why ooh why this sick breadth first order,
73 * everything is slower and more complex */
74 PutBitContext reorder_pb[6];
75
76 int frame_width;
77 int frame_height;
78
79 /* Y plane block dimensions */
80 int y_block_width;
81 int y_block_height;
82
83 DECLARE_ALIGNED(16, int16_t, encoded_block_levels)[6][7][256];
84
85 uint16_t *mb_type;
86 uint32_t *dummy;
87 int16_t (*motion_val8[3])[2];
88 int16_t (*motion_val16[3])[2];
89
90 int64_t rd_total;
91
92 uint8_t *scratchbuf;
93
94 SVQ1EncDSPContext svq1encdsp;
95 } SVQ1EncContext;
96
97 200 static void svq1_write_header(SVQ1EncContext *s, PutBitContext *pb, int frame_type)
98 {
99 int i;
100
101 /* frame code */
102 200 put_bits(pb, 22, 0x20);
103
104 /* temporal reference (sure hope this is a "don't care") */
105 200 put_bits(pb, 8, 0x00);
106
107 /* frame type */
108 200 put_bits(pb, 2, frame_type - 1);
109
110
2/2
✓ Branch 0 taken 20 times.
✓ Branch 1 taken 180 times.
 200 if (frame_type == AV_PICTURE_TYPE_I) {
111 /* no checksum since frame code is 0x20 */
112 /* no embedded string either */
113 /* output 5 unknown bits (2 + 2 + 1) */
114 20 put_bits(pb, 5, 2); /* 2 needed by quicktime decoder */
115
116 20 i = ff_match_2uint16(ff_svq1_frame_size_table,
117 FF_ARRAY_ELEMS(ff_svq1_frame_size_table),
118 s->frame_width, s->frame_height);
119 20 put_bits(pb, 3, i);
120
121
2/2
✓ Branch 0 taken 5 times.
✓ Branch 1 taken 15 times.
 20 if (i == 7) {
122 5 put_bits(pb, 12, s->frame_width);
123 5 put_bits(pb, 12, s->frame_height);
124 }
125 }
126
127 /* no checksum or extra data (next 2 bits get 0) */
128 200 put_bits(pb, 2, 0);
129 200 }
130
131 #define QUALITY_THRESHOLD 100
132 #define THRESHOLD_MULTIPLIER 0.6
133
134 4302522 static int encode_block(SVQ1EncContext *s, uint8_t *src, uint8_t *ref,
135 uint8_t *decoded, int stride, unsigned level,
136 int threshold, int lambda, int intra)
137 {
138 int count, y, x, i, j, split, best_mean, best_score, best_count;
139 int best_vector[6];
140 4302522 int block_sum[7] = { 0, 0, 0, 0, 0, 0 };
141 4302522 int w = 2 << (level + 2 >> 1);
142 4302522 int h = 2 << (level + 1 >> 1);
143 4302522 int size = w * h;
144 4302522 int16_t (*block)[256] = s->encoded_block_levels[level];
145 const int8_t *codebook_sum, *codebook;
146 const uint16_t(*mean_vlc)[2];
147 const uint8_t(*multistage_vlc)[2];
148
149 4302522 best_score = 0;
150 // FIXME: Optimize, this does not need to be done multiple times.
151
2/2
✓ Branch 0 taken 643715 times.
✓ Branch 1 taken 3658807 times.
 4302522 if (intra) {
152 // level is 5 when encode_block is called from svq1_encode_plane
153 // and always < 4 when called recursively from this function.
154
2/2
✓ Branch 0 taken 612686 times.
✓ Branch 1 taken 31029 times.
 643715 codebook_sum = level < 4 ? svq1_intra_codebook_sum[level] : NULL;
155 643715 codebook = ff_svq1_intra_codebooks[level];
156 643715 mean_vlc = ff_svq1_intra_mean_vlc;
157 643715 multistage_vlc = ff_svq1_intra_multistage_vlc[level];
158
2/2
✓ Branch 0 taken 2296676 times.
✓ Branch 1 taken 643715 times.
 2940391 for (y = 0; y < h; y++) {
159
2/2
✓ Branch 0 taken 15798160 times.
✓ Branch 1 taken 2296676 times.
 18094836 for (x = 0; x < w; x++) {
160 15798160 int v = src[x + y * stride];
161 15798160 block[0][x + w * y] = v;
162 15798160 best_score += v * v;
163 15798160 block_sum[0] += v;
164 }
165 }
166 } else {
167 // level is 5 or < 4, see above for details.
168
2/2
✓ Branch 0 taken 3483018 times.
✓ Branch 1 taken 175789 times.
 3658807 codebook_sum = level < 4 ? svq1_inter_codebook_sum[level] : NULL;
169 3658807 codebook = ff_svq1_inter_codebooks[level];
170 3658807 mean_vlc = ff_svq1_inter_mean_vlc + 256;
171 3658807 multistage_vlc = ff_svq1_inter_multistage_vlc[level];
172
2/2
✓ Branch 0 taken 13044696 times.
✓ Branch 1 taken 3658807 times.
 16703503 for (y = 0; y < h; y++) {
173
2/2
✓ Branch 0 taken 89655008 times.
✓ Branch 1 taken 13044696 times.
 102699704 for (x = 0; x < w; x++) {
174 89655008 int v = src[x + y * stride] - ref[x + y * stride];
175 89655008 block[0][x + w * y] = v;
176 89655008 best_score += v * v;
177 89655008 block_sum[0] += v;
178 }
179 }
180 }
181
182 4302522 best_count = 0;
183 4302522 best_score -= (int)((unsigned)block_sum[0] * block_sum[0] >> (level + 3));
184 4302522 best_mean = block_sum[0] + (size >> 1) >> (level + 3);
185
186
2/2
✓ Branch 0 taken 4095704 times.
✓ Branch 1 taken 206818 times.
 4302522 if (level < 4) {
187
2/2
✓ Branch 0 taken 24574224 times.
✓ Branch 1 taken 4095704 times.
 28669928 for (count = 1; count < 7; count++) {
188 24574224 int best_vector_score = INT_MAX;
189 24574224 int best_vector_sum = -999, best_vector_mean = -999;
190 24574224 const int stage = count - 1;
191 const int8_t *vector;
192
193
2/2
✓ Branch 0 taken 393187584 times.
✓ Branch 1 taken 24574224 times.
 417761808 for (i = 0; i < 16; i++) {
194 393187584 int sum = codebook_sum[stage * 16 + i];
195 int sqr, diff, score;
196
197 393187584 vector = codebook + stage * size * 16 + i * size;
198 393187584 sqr = s->svq1encdsp.ssd_int8_vs_int16(vector, block[stage], size);
199 393187584 diff = block_sum[stage] - sum;
200 393187584 score = sqr - (diff * (int64_t)diff >> (level + 3)); // FIXME: 64 bits slooow
201
2/2
✓ Branch 0 taken 65426753 times.
✓ Branch 1 taken 327760831 times.
 393187584 if (score < best_vector_score) {
202 65426753 int mean = diff + (size >> 1) >> (level + 3);
203 av_assert2(mean > -300 && mean < 300);
204
2/2
✓ Branch 0 taken 9912177 times.
✓ Branch 1 taken 55514576 times.
 65426753 mean = av_clip(mean, intra ? 0 : -256, 255);
205 65426753 best_vector_score = score;
206 65426753 best_vector[stage] = i;
207 65426753 best_vector_sum = sum;
208 65426753 best_vector_mean = mean;
209 }
210 }
211
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 24574224 times.
 24574224 av_assert0(best_vector_mean != -999);
212 24574224 vector = codebook + stage * size * 16 + best_vector[stage] * size;
213
2/2
✓ Branch 0 taken 420929184 times.
✓ Branch 1 taken 24574224 times.
 445503408 for (j = 0; j < size; j++)
214 420929184 block[stage + 1][j] = block[stage][j] - vector[j];
215 24574224 block_sum[stage + 1] = block_sum[stage] - best_vector_sum;
216 24574224 best_vector_score += lambda *
217 24574224 (+1 + 4 * count +
218 24574224 multistage_vlc[1 + count][1]
219 24574224 + mean_vlc[best_vector_mean][1]);
220
221
2/2
✓ Branch 0 taken 5416629 times.
✓ Branch 1 taken 19157595 times.
 24574224 if (best_vector_score < best_score) {
222 5416629 best_score = best_vector_score;
223 5416629 best_count = count;
224 5416629 best_mean = best_vector_mean;
225 }
226 }
227 }
228
229
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 4302510 times.
 4302522 if (best_mean == -128)
230 12 best_mean = -127;
231
2/2
✓ Branch 0 taken 6210 times.
✓ Branch 1 taken 4296300 times.
 4302510 else if (best_mean == 128)
232 6210 best_mean = 127;
233
234 4302522 split = 0;
235
4/4
✓ Branch 0 taken 4243230 times.
✓ Branch 1 taken 59292 times.
✓ Branch 2 taken 2116786 times.
✓ Branch 3 taken 2126444 times.
 4302522 if (best_score > threshold && level) {
236 2116786 int score = 0;
237
2/2
✓ Branch 0 taken 1431267 times.
✓ Branch 1 taken 685519 times.
 2116786 int offset = level & 1 ? stride * h / 2 : w / 2;
238 PutBitContext backup[6];
239
240
2/2
✓ Branch 0 taken 3903357 times.
✓ Branch 1 taken 2116786 times.
 6020143 for (i = level - 1; i >= 0; i--)
241 3903357 backup[i] = s->reorder_pb[i];
242 2116786 score += encode_block(s, src, ref, decoded, stride, level - 1,
243 threshold >> 1, lambda, intra);
244 2116786 score += encode_block(s, src + offset, ref + offset, decoded + offset,
245 stride, level - 1, threshold >> 1, lambda, intra);
246 2116786 score += lambda;
247
248
2/2
✓ Branch 0 taken 1359243 times.
✓ Branch 1 taken 757543 times.
 2116786 if (score < best_score) {
249 1359243 best_score = score;
250 1359243 split = 1;
251 } else {
252
2/2
✓ Branch 0 taken 1046147 times.
✓ Branch 1 taken 757543 times.
 1803690 for (i = level - 1; i >= 0; i--)
253 1046147 s->reorder_pb[i] = backup[i];
254 }
255 }
256
2/2
✓ Branch 0 taken 2127616 times.
✓ Branch 1 taken 2174906 times.
 4302522 if (level > 0)
257 2127616 put_bits(&s->reorder_pb[level], 1, split);
258
259
2/2
✓ Branch 0 taken 2943279 times.
✓ Branch 1 taken 1359243 times.
 4302522 if (!split) {
260 av_assert1(best_mean >= 0 && best_mean < 256 || !intra);
261 av_assert1(best_mean >= -256 && best_mean < 256);
262 av_assert1(best_count >= 0 && best_count < 7);
263 av_assert1(level < 4 || best_count == 0);
264
265 /* output the encoding */
266 2943279 put_bits(&s->reorder_pb[level],
267 2943279 multistage_vlc[1 + best_count][1],
268 2943279 multistage_vlc[1 + best_count][0]);
269 2943279 put_bits(&s->reorder_pb[level], mean_vlc[best_mean][1],
270 2943279 mean_vlc[best_mean][0]);
271
272
2/2
✓ Branch 0 taken 3649044 times.
✓ Branch 1 taken 2943279 times.
 6592323 for (i = 0; i < best_count; i++) {
273 av_assert2(best_vector[i] >= 0 && best_vector[i] < 16);
274 3649044 put_bits(&s->reorder_pb[level], 4, best_vector[i]);
275 }
276
277
2/2
✓ Branch 0 taken 7680728 times.
✓ Branch 1 taken 2943279 times.
 10624007 for (y = 0; y < h; y++)
278
2/2
✓ Branch 0 taken 36196848 times.
✓ Branch 1 taken 7680728 times.
 43877576 for (x = 0; x < w; x++)
279 36196848 decoded[x + y * stride] = src[x + y * stride] -
280 36196848 block[best_count][x + w * y] +
281 best_mean;
282 }
283
284 4302522 return best_score;
285 }
286
287 131005 static void init_block_index(MpegEncContext *const s)
288 {
289 131005 s->block_index[0]= s->b8_stride*(s->mb_y*2 ) + s->mb_x*2;
290 131005 s->block_index[1]= s->b8_stride*(s->mb_y*2 ) + 1 + s->mb_x*2;
291 131005 s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) + s->mb_x*2;
292 131005 s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) + 1 + s->mb_x*2;
293 131005 }
294
295 600 static int svq1_encode_plane(SVQ1EncContext *s, int plane,
296 PutBitContext *pb,
297 const unsigned char *src_plane,
298 unsigned char *ref_plane,
299 unsigned char *decoded_plane,
300 int width, int height, int src_stride, int stride)
301 {
302 600 MpegEncContext *const s2 = &s->m.c;
303 int x, y;
304 int i;
305 int block_width, block_height;
306 int level;
307 int threshold[6];
308 600 uint8_t *src = s->scratchbuf + stride * 32;
309 600 const int lambda = (s->quality * s->quality) >>
310 (2 * FF_LAMBDA_SHIFT);
311
312 /* figure out the acceptable level thresholds in advance */
313 600 threshold[5] = QUALITY_THRESHOLD;
314
2/2
✓ Branch 0 taken 3000 times.
✓ Branch 1 taken 600 times.
 3600 for (level = 4; level >= 0; level--)
315 3000 threshold[level] = threshold[level + 1] * THRESHOLD_MULTIPLIER;
316
317 600 block_width = (width + 15) / 16;
318 600 block_height = (height + 15) / 16;
319
320
2/2
✓ Branch 0 taken 540 times.
✓ Branch 1 taken 60 times.
 600 if (s->pict_type == AV_PICTURE_TYPE_P) {
321 540 s2->last_pic.data[0] = ref_plane;
322 540 s2->linesize =
323 540 s2->last_pic.linesize[0] =
324 540 s->m.new_pic->linesize[0] =
325 540 s2->cur_pic.linesize[0] = stride;
326 540 s2->width = width;
327 540 s2->height = height;
328 540 s2->mb_width = block_width;
329 540 s2->mb_height = block_height;
330 540 s2->mb_stride = s2->mb_width + 1;
331 540 s2->b8_stride = 2 * s2->mb_width + 1;
332 540 s->m.f_code = 1;
333 540 s2->pict_type = s->pict_type;
334 540 s->m.me.scene_change_score = 0;
335 // s2->out_format = FMT_H263;
336 // s2->unrestricted_mv = 1;
337 540 s->m.lambda = s->quality;
338 540 s2->qscale = s->m.lambda * 139 +
339 540 FF_LAMBDA_SCALE * 64 >>
340 FF_LAMBDA_SHIFT + 7;
341 540 s->m.lambda2 = s->m.lambda * s->m.lambda +
342 540 FF_LAMBDA_SCALE / 2 >>
343 FF_LAMBDA_SHIFT;
344
345 540 s->m.mb_type = s->mb_type;
346
347 // dummies, to avoid segfaults
348 540 s->m.mb_mean = (uint8_t *)s->dummy;
349 540 s->m.mb_var = (uint16_t *)s->dummy;
350 540 s->m.mc_mb_var = (uint16_t *)s->dummy;
351 540 s2->cur_pic.mb_type = s->dummy;
352
353 540 s2->cur_pic.motion_val[0] = s->motion_val8[plane] + 2;
354 540 s->m.p_mv_table = s->motion_val16[plane] +
355 540 s2->mb_stride + 1;
356 540 ff_me_init_pic(&s->m);
357
358 540 s->m.me.dia_size = s->avctx->dia_size;
359 540 s2->first_slice_line = 1;
360
2/2
✓ Branch 0 taken 4005 times.
✓ Branch 1 taken 540 times.
 4545 for (y = 0; y < block_height; y++) {
361 4005 s->m.new_pic->data[0] = src - y * 16 * stride; // ugly
362 4005 s2->mb_y = y;
363
364
4/4
✓ Branch 0 taken 60975 times.
✓ Branch 1 taken 3600 times.
✓ Branch 2 taken 60570 times.
✓ Branch 3 taken 405 times.
 64575 for (i = 0; i < 16 && i + 16 * y < height; i++) {
365 60570 memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
366 width);
367
2/2
✓ Branch 0 taken 182700 times.
✓ Branch 1 taken 60570 times.
 243270 for (x = width; x < 16 * block_width; x++)
368 182700 src[i * stride + x] = src[i * stride + x - 1];
369 }
370
3/4
✓ Branch 0 taken 3510 times.
✓ Branch 1 taken 4005 times.
✓ Branch 2 taken 3510 times.
✗ Branch 3 not taken.
 7515 for (; i < 16 && i + 16 * y < 16 * block_height; i++)
371 3510 memcpy(&src[i * stride], &src[(i - 1) * stride],
372 3510 16 * block_width);
373
374
2/2
✓ Branch 0 taken 62055 times.
✓ Branch 1 taken 4005 times.
 66060 for (x = 0; x < block_width; x++) {
375 62055 s2->mb_x = x;
376 62055 init_block_index(s2);
377
378 62055 ff_estimate_p_frame_motion(&s->m, x, y);
379 }
380 4005 s2->first_slice_line = 0;
381 }
382
383 540 ff_fix_long_p_mvs(&s->m, CANDIDATE_MB_TYPE_INTRA);
384 540 ff_fix_long_mvs(&s->m, NULL, 0, s->m.p_mv_table, s->m.f_code,
385 CANDIDATE_MB_TYPE_INTER, 0);
386 }
387
388 600 s2->first_slice_line = 1;
389
2/2
✓ Branch 0 taken 4450 times.
✓ Branch 1 taken 600 times.
 5050 for (y = 0; y < block_height; y++) {
390
4/4
✓ Branch 0 taken 67750 times.
✓ Branch 1 taken 4000 times.
✓ Branch 2 taken 67300 times.
✓ Branch 3 taken 450 times.
 71750 for (i = 0; i < 16 && i + 16 * y < height; i++) {
391 67300 memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
392 width);
393
2/2
✓ Branch 0 taken 203000 times.
✓ Branch 1 taken 67300 times.
 270300 for (x = width; x < 16 * block_width; x++)
394 203000 src[i * stride + x] = src[i * stride + x - 1];
395 }
396
3/4
✓ Branch 0 taken 3900 times.
✓ Branch 1 taken 4450 times.
✓ Branch 2 taken 3900 times.
✗ Branch 3 not taken.
 8350 for (; i < 16 && i + 16 * y < 16 * block_height; i++)
397 3900 memcpy(&src[i * stride], &src[(i - 1) * stride], 16 * block_width);
398
399 4450 s2->mb_y = y;
400
2/2
✓ Branch 0 taken 68950 times.
✓ Branch 1 taken 4450 times.
 73400 for (x = 0; x < block_width; x++) {
401 uint8_t reorder_buffer[2][6][7 * 32];
402 int count[2][6];
403 68950 int offset = y * 16 * stride + x * 16;
404 68950 uint8_t *decoded = decoded_plane + offset;
405 68950 const uint8_t *ref = ref_plane + offset;
406 68950 int score[4] = { 0, 0, 0, 0 }, best;
407 68950 uint8_t *temp = s->scratchbuf;
408
409
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 68950 times.
 68950 if (put_bytes_left(pb, 0) < 3000) { // FIXME: check size
410 ✗ av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
411 ✗ return -1;
412 }
413
414 68950 s2->mb_x = x;
415 68950 init_block_index(s2);
416
417
2/2
✓ Branch 0 taken 62055 times.
✓ Branch 1 taken 6895 times.
 68950 if (s->pict_type == AV_PICTURE_TYPE_I ||
418
2/2
✓ Branch 0 taken 3450 times.
✓ Branch 1 taken 58605 times.
 62055 (s->m.mb_type[x + y * s2->mb_stride] &
419 CANDIDATE_MB_TYPE_INTRA)) {
420
2/2
✓ Branch 0 taken 62070 times.
✓ Branch 1 taken 10345 times.
 72415 for (i = 0; i < 6; i++)
421 62070 init_put_bits(&s->reorder_pb[i], reorder_buffer[0][i],
422 7 * 32);
423
2/2
✓ Branch 0 taken 3450 times.
✓ Branch 1 taken 6895 times.
 10345 if (s->pict_type == AV_PICTURE_TYPE_P) {
424 3450 put_bits(&s->reorder_pb[5], SVQ1_BLOCK_INTRA_LEN, SVQ1_BLOCK_INTRA_CODE);
425 3450 score[0] = SVQ1_BLOCK_INTRA_LEN * lambda;
426 }
427 10345 score[0] += encode_block(s, src + 16 * x, src + 16 * x /* unused */,
428 temp, stride, 5, 64, lambda, 1);
429
2/2
✓ Branch 0 taken 62070 times.
✓ Branch 1 taken 10345 times.
 72415 for (i = 0; i < 6; i++) {
430 62070 count[0][i] = put_bits_count(&s->reorder_pb[i]);
431 62070 flush_put_bits(&s->reorder_pb[i]);
432 }
433 } else
434 58605 score[0] = INT_MAX;
435
436 68950 best = 0;
437
438
2/2
✓ Branch 0 taken 62055 times.
✓ Branch 1 taken 6895 times.
 68950 if (s->pict_type == AV_PICTURE_TYPE_P) {
439 int mx, my, pred_x, pred_y, dxy;
440 int16_t *motion_ptr;
441
442 62055 motion_ptr = ff_h263_pred_motion(s2, 0, 0, &pred_x, &pred_y);
443
2/2
✓ Branch 0 taken 58605 times.
✓ Branch 1 taken 3450 times.
 62055 if (s->m.mb_type[x + y * s2->mb_stride] &
444 CANDIDATE_MB_TYPE_INTER) {
445
2/2
✓ Branch 0 taken 351630 times.
✓ Branch 1 taken 58605 times.
 410235 for (i = 0; i < 6; i++)
446 351630 init_put_bits(&s->reorder_pb[i], reorder_buffer[1][i],
447 7 * 32);
448
449 58605 put_bits(&s->reorder_pb[5], SVQ1_BLOCK_INTER_LEN, SVQ1_BLOCK_INTER_CODE);
450
451 58605 mx = motion_ptr[0];
452 58605 my = motion_ptr[1];
453 av_assert1(mx >= -32 && mx <= 31);
454 av_assert1(my >= -32 && my <= 31);
455 av_assert1(pred_x >= -32 && pred_x <= 31);
456 av_assert1(pred_y >= -32 && pred_y <= 31);
457 58605 ff_h263_encode_motion(&s->reorder_pb[5], mx - pred_x, 1);
458 58605 ff_h263_encode_motion(&s->reorder_pb[5], my - pred_y, 1);
459 58605 score[1] += lambda * put_bits_count(&s->reorder_pb[5]);
460
461 58605 dxy = (mx & 1) + 2 * (my & 1);
462
463 58605 s2->hdsp.put_pixels_tab[0][dxy](temp + 16*stride,
464 58605 ref + (mx >> 1) +
465 58605 stride * (my >> 1),
466 stride, 16);
467
468 58605 score[1] += encode_block(s, src + 16 * x, temp + 16*stride,
469 decoded, stride, 5, 64, lambda, 0);
470 58605 best = score[1] <= score[0];
471
472 58605 score[2] = s->mecc.sse[0](NULL, src + 16 * x, ref,
473 stride, 16);
474 58605 score[2] += SVQ1_BLOCK_SKIP_LEN * lambda;
475
6/6
✓ Branch 0 taken 119 times.
✓ Branch 1 taken 58486 times.
✓ Branch 2 taken 88 times.
✓ Branch 3 taken 31 times.
✓ Branch 4 taken 83 times.
✓ Branch 5 taken 5 times.
 58605 if (score[2] < score[best] && mx == 0 && my == 0) {
476 83 best = 2;
477 83 s2->hdsp.put_pixels_tab[0][0](decoded, ref, stride, 16);
478 83 put_bits(pb, SVQ1_BLOCK_SKIP_LEN, SVQ1_BLOCK_SKIP_CODE);
479 }
480 }
481
482
2/2
✓ Branch 0 taken 58522 times.
✓ Branch 1 taken 3533 times.
 62055 if (best == 1) {
483
2/2
✓ Branch 0 taken 351132 times.
✓ Branch 1 taken 58522 times.
 409654 for (i = 0; i < 6; i++) {
484 351132 count[1][i] = put_bits_count(&s->reorder_pb[i]);
485 351132 flush_put_bits(&s->reorder_pb[i]);
486 }
487 } else {
488 3533 motion_ptr[0] =
489 3533 motion_ptr[1] =
490 3533 motion_ptr[2] =
491 3533 motion_ptr[3] =
492 3533 motion_ptr[0 + 2 * s2->b8_stride] =
493 3533 motion_ptr[1 + 2 * s2->b8_stride] =
494 3533 motion_ptr[2 + 2 * s2->b8_stride] =
495 3533 motion_ptr[3 + 2 * s2->b8_stride] = 0;
496 }
497 }
498
499 68950 s->rd_total += score[best];
500
501
2/2
✓ Branch 0 taken 68867 times.
✓ Branch 1 taken 83 times.
 68950 if (best != 2)
502
2/2
✓ Branch 0 taken 413202 times.
✓ Branch 1 taken 68867 times.
 482069 for (i = 5; i >= 0; i--)
503 413202 ff_copy_bits(pb, reorder_buffer[best][i],
504 count[best][i]);
505
2/2
✓ Branch 0 taken 10345 times.
✓ Branch 1 taken 58605 times.
 68950 if (best == 0)
506 10345 s2->hdsp.put_pixels_tab[0][0](decoded, temp, stride, 16);
507 }
508 4450 s2->first_slice_line = 0;
509 }
510 600 return 0;
511 }
512
513 4 static av_cold int svq1_encode_end(AVCodecContext *avctx)
514 {
515 4 SVQ1EncContext *const s = avctx->priv_data;
516 int i;
517
518
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
 4 if (avctx->frame_num)
519 4 av_log(avctx, AV_LOG_DEBUG, "RD: %f\n",
520 4 s->rd_total / (double)(avctx->width * avctx->height *
521 4 avctx->frame_num));
522
523 4 av_freep(&s->m.me.scratchpad);
524 4 av_freep(&s->mb_type);
525 4 av_freep(&s->dummy);
526 4 av_freep(&s->scratchbuf);
527
528
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 4 times.
 16 for (i = 0; i < 3; i++) {
529 12 av_freep(&s->motion_val8[i]);
530 12 av_freep(&s->motion_val16[i]);
531 }
532
533 4 av_frame_free(&s->current_picture);
534 4 av_frame_free(&s->last_picture);
535 4 av_frame_free(&s->m.new_pic);
536
537 4 return 0;
538 }
539
540 4 static av_cold int write_ident(AVCodecContext *avctx, const char *ident)
541 {
542 4 int size = strlen(ident);
543 4 avctx->extradata = av_malloc(size + 8);
544
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (!avctx->extradata)
545 ✗ return AVERROR(ENOMEM);
546 4 AV_WB32(avctx->extradata, size + 8);
547 4 AV_WL32(avctx->extradata + 4, MKTAG('S', 'V', 'Q', '1'));
548 4 memcpy(avctx->extradata + 8, ident, size);
549 4 avctx->extradata_size = size + 8;
550 4 return 0;
551 }
552
553 4 static av_cold int svq1_encode_init(AVCodecContext *avctx)
554 {
555 4 SVQ1EncContext *const s = avctx->priv_data;
556 int ret;
557
558
2/4
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 4 times.
 4 if (avctx->width >= 4096 || avctx->height >= 4096) {
559 ✗ av_log(avctx, AV_LOG_ERROR, "Dimensions too large, maximum is 4095x4095\n");
560 ✗ return AVERROR(EINVAL);
561 }
562
563 4 ff_hpeldsp_init(&s->m.c.hdsp, avctx->flags);
564 4 ff_me_cmp_init(&s->mecc, avctx);
565 4 ret = ff_me_init(&s->m.me, avctx, &s->mecc, 0);
566
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (ret < 0)
567 ✗ return ret;
568 4 ff_mpegvideoencdsp_init(&s->m.mpvencdsp, avctx);
569
570 4 s->current_picture = av_frame_alloc();
571 4 s->last_picture = av_frame_alloc();
572
2/4
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 4 times.
 4 if (!s->current_picture || !s->last_picture) {
573 ✗ return AVERROR(ENOMEM);
574 }
575 4 ret = ff_encode_alloc_frame(avctx, s->current_picture);
576
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (ret < 0)
577 ✗ return ret;
578 4 ret = ff_encode_alloc_frame(avctx, s->last_picture);
579
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (ret < 0)
580 ✗ return ret;
581 4 s->scratchbuf = av_malloc_array(s->current_picture->linesize[0], 16 * 3);
582
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (!s->scratchbuf)
583 ✗ return AVERROR(ENOMEM);
584
585 4 s->frame_width = avctx->width;
586 4 s->frame_height = avctx->height;
587
588 4 s->y_block_width = (s->frame_width + 15) / 16;
589 4 s->y_block_height = (s->frame_height + 15) / 16;
590
591 4 s->avctx = avctx;
592 4 s->m.c.avctx = avctx;
593
594
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 4 times.
 16 for (size_t plane = 0; plane < FF_ARRAY_ELEMS(s->motion_val16); ++plane) {
595
2/2
✓ Branch 0 taken 8 times.
✓ Branch 1 taken 4 times.
 12 const int shift = plane ? 2 : 0;
596 12 unsigned block_height = ((s->frame_height >> shift) + 15U) / 16;
597 12 unsigned block_width = ((s->frame_width >> shift) + 15U) / 16;
598
599 12 s->motion_val8[plane] = av_calloc((2 * block_width + 1) * block_height * 2 + 2,
600 2 * sizeof(int16_t));
601 12 s->motion_val16[plane] = av_calloc((block_width + 1) * (block_height + 2) + 1,
602 2 * sizeof(int16_t));
603
2/4
✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 12 times.
 12 if (!s->motion_val8[plane] || !s->motion_val16[plane])
604 ✗ return AVERROR(ENOMEM);
605 }
606
607 4 s->m.c.picture_structure = PICT_FRAME;
608 4 s->m.me.temp =
609 8 s->m.me.scratchpad = av_mallocz((avctx->width + 64) *
610 4 2 * 16 * 2 * sizeof(uint8_t));
611 8 s->mb_type = av_mallocz((s->y_block_width + 1) *
612 4 s->y_block_height * sizeof(int16_t));
613 8 s->dummy = av_mallocz((s->y_block_width + 1) *
614 4 s->y_block_height * sizeof(int32_t));
615 4 s->m.new_pic = av_frame_alloc();
616
617
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
 4 if (!s->m.me.scratchpad ||
618
3/6
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✓ Branch 5 taken 4 times.
 4 !s->mb_type || !s->dummy || !s->m.new_pic)
619 ✗ return AVERROR(ENOMEM);
620
621 4 ff_svq1enc_init(&s->svq1encdsp);
622
623 4 s->m.me.mv_penalty = ff_h263_get_mv_penalty();
624
625
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
 4 return write_ident(avctx, s->avctx->flags & AV_CODEC_FLAG_BITEXACT ? "Lavc" : LIBAVCODEC_IDENT);
626 }
627
628 200 static int svq1_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
629 const AVFrame *pict, int *got_packet)
630 {
631 200 SVQ1EncContext *const s = avctx->priv_data;
632 PutBitContext pb;
633 int i, ret;
634
635 200 ret = ff_alloc_packet(avctx, pkt, s->y_block_width * s->y_block_height *
636 200 MAX_MB_BYTES * 3 + FF_INPUT_BUFFER_MIN_SIZE);
637
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 200 times.
 200 if (ret < 0)
638 ✗ return ret;
639
640 200 FFSWAP(AVFrame*, s->current_picture, s->last_picture);
641
642
3/4
✓ Branch 0 taken 200 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 180 times.
✓ Branch 3 taken 20 times.
 200 if (avctx->gop_size && (avctx->frame_num % avctx->gop_size))
643 180 s->pict_type = AV_PICTURE_TYPE_P;
644 else
645 20 s->pict_type = AV_PICTURE_TYPE_I;
646 200 s->quality = pict->quality;
647
648 200 ff_side_data_set_encoder_stats(pkt, pict->quality, NULL, 0, s->pict_type);
649
650 200 init_put_bits(&pb, pkt->data, pkt->size);
651 200 svq1_write_header(s, &pb, s->pict_type);
652
2/2
✓ Branch 0 taken 600 times.
✓ Branch 1 taken 200 times.
 800 for (i = 0; i < 3; i++) {
653 1800 int ret = svq1_encode_plane(s, i, &pb,
654 600 pict->data[i],
655 600 s->last_picture->data[i],
656 600 s->current_picture->data[i],
657
2/2
✓ Branch 0 taken 400 times.
✓ Branch 1 taken 200 times.
 600 s->frame_width / (i ? 4 : 1),
658 600 s->frame_height / (i ? 4 : 1),
659 600 pict->linesize[i],
660
2/2
✓ Branch 0 taken 400 times.
✓ Branch 1 taken 200 times.
 600 s->current_picture->linesize[i]);
661 600 emms_c();
662
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 600 times.
 600 if (ret < 0)
663 ✗ return ret;
664 }
665
666 // align_put_bits(&pb);
667
2/2
✓ Branch 1 taken 3154 times.
✓ Branch 2 taken 200 times.
 3354 while (put_bits_count(&pb) & 31)
668 3154 put_bits(&pb, 1, 0);
669
670 200 flush_put_bits(&pb);
671
672 200 pkt->size = put_bytes_output(&pb);
673
2/2
✓ Branch 0 taken 20 times.
✓ Branch 1 taken 180 times.
 200 if (s->pict_type == AV_PICTURE_TYPE_I)
674 20 pkt->flags |= AV_PKT_FLAG_KEY;
675 200 *got_packet = 1;
676
677 200 return 0;
678 }
679
680 #define OFFSET(x) offsetof(struct SVQ1EncContext, x)
681 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
682 static const AVOption options[] = {
683 { "motion-est", "Motion estimation algorithm", OFFSET(m.me.motion_est), AV_OPT_TYPE_INT, { .i64 = FF_ME_EPZS }, FF_ME_ZERO, FF_ME_XONE, VE, .unit = "motion-est"},
684 { "zero", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_ZERO }, 0, 0, FF_MPV_OPT_FLAGS, .unit = "motion-est" },
685 { "epzs", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_EPZS }, 0, 0, FF_MPV_OPT_FLAGS, .unit = "motion-est" },
686 { "xone", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_XONE }, 0, 0, FF_MPV_OPT_FLAGS, .unit = "motion-est" },
687
688 { NULL },
689 };
690
691 static const AVClass svq1enc_class = {
692 .class_name = "svq1enc",
693 .item_name = av_default_item_name,
694 .option = options,
695 .version = LIBAVUTIL_VERSION_INT,
696 };
697
698 const FFCodec ff_svq1_encoder = {
699 .p.name = "svq1",
700 CODEC_LONG_NAME("Sorenson Vector Quantizer 1 / Sorenson Video 1 / SVQ1"),
701 .p.type = AVMEDIA_TYPE_VIDEO,
702 .p.id = AV_CODEC_ID_SVQ1,
703 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_ENCODER_REORDERED_OPAQUE,
704 .priv_data_size = sizeof(SVQ1EncContext),
705 .p.priv_class = &svq1enc_class,
706 .init = svq1_encode_init,
707 FF_CODEC_ENCODE_CB(svq1_encode_frame),
708 .close = svq1_encode_end,
709 CODEC_PIXFMTS(AV_PIX_FMT_YUV410P),
710 .color_ranges = AVCOL_RANGE_MPEG,
711 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
712 };
713