FFmpeg  4.3.6
svq1enc.c
Go to the documentation of this file.
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 "avcodec.h"
30 #include "hpeldsp.h"
31 #include "me_cmp.h"
32 #include "mpegvideo.h"
33 #include "h263.h"
34 #include "internal.h"
35 #include "mpegutils.h"
36 #include "packet_internal.h"
37 #include "svq1.h"
38 #include "svq1enc.h"
39 #include "svq1enc_cb.h"
40 #include "libavutil/avassert.h"
41 
42 
44 {
45  int i;
46 
47  /* frame code */
48  put_bits(&s->pb, 22, 0x20);
49 
50  /* temporal reference (sure hope this is a "don't care") */
51  put_bits(&s->pb, 8, 0x00);
52 
53  /* frame type */
54  put_bits(&s->pb, 2, frame_type - 1);
55 
56  if (frame_type == AV_PICTURE_TYPE_I) {
57  /* no checksum since frame code is 0x20 */
58  /* no embedded string either */
59  /* output 5 unknown bits (2 + 2 + 1) */
60  put_bits(&s->pb, 5, 2); /* 2 needed by quicktime decoder */
61 
63  FF_ARRAY_ELEMS(ff_svq1_frame_size_table),
64  s->frame_width, s->frame_height);
65  put_bits(&s->pb, 3, i);
66 
67  if (i == 7) {
68  put_bits(&s->pb, 12, s->frame_width);
69  put_bits(&s->pb, 12, s->frame_height);
70  }
71  }
72 
73  /* no checksum or extra data (next 2 bits get 0) */
74  put_bits(&s->pb, 2, 0);
75 }
76 
77 #define QUALITY_THRESHOLD 100
78 #define THRESHOLD_MULTIPLIER 0.6
79 
80 static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2,
81  intptr_t size)
82 {
83  int score = 0, i;
84 
85  for (i = 0; i < size; i++)
86  score += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
87  return score;
88 }
89 
91  uint8_t *decoded, int stride, int level,
92  int threshold, int lambda, int intra)
93 {
94  int count, y, x, i, j, split, best_mean, best_score, best_count;
95  int best_vector[6];
96  int block_sum[7] = { 0, 0, 0, 0, 0, 0 };
97  int w = 2 << (level + 2 >> 1);
98  int h = 2 << (level + 1 >> 1);
99  int size = w * h;
100  int16_t (*block)[256] = s->encoded_block_levels[level];
101  const int8_t *codebook_sum, *codebook;
102  const uint16_t(*mean_vlc)[2];
103  const uint8_t(*multistage_vlc)[2];
104 
105  best_score = 0;
106  // FIXME: Optimize, this does not need to be done multiple times.
107  if (intra) {
108  // level is 5 when encode_block is called from svq1_encode_plane
109  // and always < 4 when called recursively from this function.
110  codebook_sum = level < 4 ? svq1_intra_codebook_sum[level] : NULL;
111  codebook = ff_svq1_intra_codebooks[level];
112  mean_vlc = ff_svq1_intra_mean_vlc;
113  multistage_vlc = ff_svq1_intra_multistage_vlc[level];
114  for (y = 0; y < h; y++) {
115  for (x = 0; x < w; x++) {
116  int v = src[x + y * stride];
117  block[0][x + w * y] = v;
118  best_score += v * v;
119  block_sum[0] += v;
120  }
121  }
122  } else {
123  // level is 5 or < 4, see above for details.
124  codebook_sum = level < 4 ? svq1_inter_codebook_sum[level] : NULL;
125  codebook = ff_svq1_inter_codebooks[level];
126  mean_vlc = ff_svq1_inter_mean_vlc + 256;
127  multistage_vlc = ff_svq1_inter_multistage_vlc[level];
128  for (y = 0; y < h; y++) {
129  for (x = 0; x < w; x++) {
130  int v = src[x + y * stride] - ref[x + y * stride];
131  block[0][x + w * y] = v;
132  best_score += v * v;
133  block_sum[0] += v;
134  }
135  }
136  }
137 
138  best_count = 0;
139  best_score -= (int)((unsigned)block_sum[0] * block_sum[0] >> (level + 3));
140  best_mean = block_sum[0] + (size >> 1) >> (level + 3);
141 
142  if (level < 4) {
143  for (count = 1; count < 7; count++) {
144  int best_vector_score = INT_MAX;
145  int best_vector_sum = -999, best_vector_mean = -999;
146  const int stage = count - 1;
147  const int8_t *vector;
148 
149  for (i = 0; i < 16; i++) {
150  int sum = codebook_sum[stage * 16 + i];
151  int sqr, diff, score;
152 
153  vector = codebook + stage * size * 16 + i * size;
154  sqr = s->ssd_int8_vs_int16(vector, block[stage], size);
155  diff = block_sum[stage] - sum;
156  score = sqr - (diff * (int64_t)diff >> (level + 3)); // FIXME: 64 bits slooow
157  if (score < best_vector_score) {
158  int mean = diff + (size >> 1) >> (level + 3);
159  av_assert2(mean > -300 && mean < 300);
160  mean = av_clip(mean, intra ? 0 : -256, 255);
161  best_vector_score = score;
162  best_vector[stage] = i;
163  best_vector_sum = sum;
164  best_vector_mean = mean;
165  }
166  }
167  av_assert0(best_vector_mean != -999);
168  vector = codebook + stage * size * 16 + best_vector[stage] * size;
169  for (j = 0; j < size; j++)
170  block[stage + 1][j] = block[stage][j] - vector[j];
171  block_sum[stage + 1] = block_sum[stage] - best_vector_sum;
172  best_vector_score += lambda *
173  (+1 + 4 * count +
174  multistage_vlc[1 + count][1]
175  + mean_vlc[best_vector_mean][1]);
176 
177  if (best_vector_score < best_score) {
178  best_score = best_vector_score;
179  best_count = count;
180  best_mean = best_vector_mean;
181  }
182  }
183  }
184 
185  split = 0;
186  if (best_score > threshold && level) {
187  int score = 0;
188  int offset = level & 1 ? stride * h / 2 : w / 2;
189  PutBitContext backup[6];
190 
191  for (i = level - 1; i >= 0; i--)
192  backup[i] = s->reorder_pb[i];
193  score += encode_block(s, src, ref, decoded, stride, level - 1,
194  threshold >> 1, lambda, intra);
195  score += encode_block(s, src + offset, ref + offset, decoded + offset,
196  stride, level - 1, threshold >> 1, lambda, intra);
197  score += lambda;
198 
199  if (score < best_score) {
200  best_score = score;
201  split = 1;
202  } else {
203  for (i = level - 1; i >= 0; i--)
204  s->reorder_pb[i] = backup[i];
205  }
206  }
207  if (level > 0)
208  put_bits(&s->reorder_pb[level], 1, split);
209 
210  if (!split) {
211  av_assert1(best_mean >= 0 && best_mean < 256 || !intra);
212  av_assert1(best_mean >= -256 && best_mean < 256);
213  av_assert1(best_count >= 0 && best_count < 7);
214  av_assert1(level < 4 || best_count == 0);
215 
216  /* output the encoding */
217  put_bits(&s->reorder_pb[level],
218  multistage_vlc[1 + best_count][1],
219  multistage_vlc[1 + best_count][0]);
220  put_bits(&s->reorder_pb[level], mean_vlc[best_mean][1],
221  mean_vlc[best_mean][0]);
222 
223  for (i = 0; i < best_count; i++) {
224  av_assert2(best_vector[i] >= 0 && best_vector[i] < 16);
225  put_bits(&s->reorder_pb[level], 4, best_vector[i]);
226  }
227 
228  for (y = 0; y < h; y++)
229  for (x = 0; x < w; x++)
230  decoded[x + y * stride] = src[x + y * stride] -
231  block[best_count][x + w * y] +
232  best_mean;
233  }
234 
235  return best_score;
236 }
237 
239  s->block_index[0]= s->b8_stride*(s->mb_y*2 ) + s->mb_x*2;
240  s->block_index[1]= s->b8_stride*(s->mb_y*2 ) + 1 + s->mb_x*2;
241  s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) + s->mb_x*2;
242  s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) + 1 + s->mb_x*2;
243  s->block_index[4]= s->mb_stride*(s->mb_y + 1) + s->b8_stride*s->mb_height*2 + s->mb_x;
244  s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x;
245 }
246 
247 static int svq1_encode_plane(SVQ1EncContext *s, int plane,
248  unsigned char *src_plane,
249  unsigned char *ref_plane,
250  unsigned char *decoded_plane,
251  int width, int height, int src_stride, int stride)
252 {
253  int x, y;
254  int i;
255  int block_width, block_height;
256  int level;
257  int threshold[6];
258  uint8_t *src = s->scratchbuf + stride * 32;
259  const int lambda = (s->quality * s->quality) >>
260  (2 * FF_LAMBDA_SHIFT);
261 
262  /* figure out the acceptable level thresholds in advance */
263  threshold[5] = QUALITY_THRESHOLD;
264  for (level = 4; level >= 0; level--)
265  threshold[level] = threshold[level + 1] * THRESHOLD_MULTIPLIER;
266 
267  block_width = (width + 15) / 16;
268  block_height = (height + 15) / 16;
269 
270  if (s->pict_type == AV_PICTURE_TYPE_P) {
271  s->m.avctx = s->avctx;
273  s->m.last_picture_ptr = &s->m.last_picture;
274  s->m.last_picture.f->data[0] = ref_plane;
275  s->m.linesize =
276  s->m.last_picture.f->linesize[0] =
277  s->m.new_picture.f->linesize[0] =
278  s->m.current_picture.f->linesize[0] = stride;
279  s->m.width = width;
280  s->m.height = height;
281  s->m.mb_width = block_width;
282  s->m.mb_height = block_height;
283  s->m.mb_stride = s->m.mb_width + 1;
284  s->m.b8_stride = 2 * s->m.mb_width + 1;
285  s->m.f_code = 1;
286  s->m.pict_type = s->pict_type;
287  s->m.motion_est = s->motion_est;
288  s->m.me.scene_change_score = 0;
289  // s->m.out_format = FMT_H263;
290  // s->m.unrestricted_mv = 1;
291  s->m.lambda = s->quality;
292  s->m.qscale = s->m.lambda * 139 +
293  FF_LAMBDA_SCALE * 64 >>
294  FF_LAMBDA_SHIFT + 7;
295  s->m.lambda2 = s->m.lambda * s->m.lambda +
296  FF_LAMBDA_SCALE / 2 >>
298 
299  if (!s->motion_val8[plane]) {
300  s->motion_val8[plane] = av_mallocz((s->m.b8_stride *
301  block_height * 2 + 2) *
302  2 * sizeof(int16_t));
303  s->motion_val16[plane] = av_mallocz((s->m.mb_stride *
304  (block_height + 2) + 1) *
305  2 * sizeof(int16_t));
306  if (!s->motion_val8[plane] || !s->motion_val16[plane])
307  return AVERROR(ENOMEM);
308  }
309 
310  s->m.mb_type = s->mb_type;
311 
312  // dummies, to avoid segfaults
314  s->m.current_picture.mb_var = (uint16_t *)s->dummy;
315  s->m.current_picture.mc_mb_var = (uint16_t *)s->dummy;
316  s->m.current_picture.mb_type = s->dummy;
317 
318  s->m.current_picture.motion_val[0] = s->motion_val8[plane] + 2;
319  s->m.p_mv_table = s->motion_val16[plane] +
320  s->m.mb_stride + 1;
321  s->m.mecc = s->mecc; // move
322  ff_init_me(&s->m);
323 
324  s->m.me.dia_size = s->avctx->dia_size;
325  s->m.first_slice_line = 1;
326  for (y = 0; y < block_height; y++) {
327  s->m.new_picture.f->data[0] = src - y * 16 * stride; // ugly
328  s->m.mb_y = y;
329 
330  for (i = 0; i < 16 && i + 16 * y < height; i++) {
331  memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
332  width);
333  for (x = width; x < 16 * block_width; x++)
334  src[i * stride + x] = src[i * stride + x - 1];
335  }
336  for (; i < 16 && i + 16 * y < 16 * block_height; i++)
337  memcpy(&src[i * stride], &src[(i - 1) * stride],
338  16 * block_width);
339 
340  for (x = 0; x < block_width; x++) {
341  s->m.mb_x = x;
342  init_block_index(&s->m);
343 
344  ff_estimate_p_frame_motion(&s->m, x, y);
345  }
346  s->m.first_slice_line = 0;
347  }
348 
350  ff_fix_long_mvs(&s->m, NULL, 0, s->m.p_mv_table, s->m.f_code,
352  }
353 
354  s->m.first_slice_line = 1;
355  for (y = 0; y < block_height; y++) {
356  for (i = 0; i < 16 && i + 16 * y < height; i++) {
357  memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
358  width);
359  for (x = width; x < 16 * block_width; x++)
360  src[i * stride + x] = src[i * stride + x - 1];
361  }
362  for (; i < 16 && i + 16 * y < 16 * block_height; i++)
363  memcpy(&src[i * stride], &src[(i - 1) * stride], 16 * block_width);
364 
365  s->m.mb_y = y;
366  for (x = 0; x < block_width; x++) {
367  uint8_t reorder_buffer[2][6][7 * 32];
368  int count[2][6];
369  int offset = y * 16 * stride + x * 16;
370  uint8_t *decoded = decoded_plane + offset;
371  uint8_t *ref = ref_plane + offset;
372  int score[4] = { 0, 0, 0, 0 }, best;
373  uint8_t *temp = s->scratchbuf;
374 
375  if (s->pb.buf_end - s->pb.buf -
376  (put_bits_count(&s->pb) >> 3) < 3000) { // FIXME: check size
377  av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
378  return -1;
379  }
380 
381  s->m.mb_x = x;
382  init_block_index(&s->m);
383 
384  if (s->pict_type == AV_PICTURE_TYPE_I ||
385  (s->m.mb_type[x + y * s->m.mb_stride] &
387  for (i = 0; i < 6; i++)
388  init_put_bits(&s->reorder_pb[i], reorder_buffer[0][i],
389  7 * 32);
390  if (s->pict_type == AV_PICTURE_TYPE_P) {
392  put_bits(&s->reorder_pb[5], vlc[1], vlc[0]);
393  score[0] = vlc[1] * lambda;
394  }
395  score[0] += encode_block(s, src + 16 * x, NULL, temp, stride,
396  5, 64, lambda, 1);
397  for (i = 0; i < 6; i++) {
398  count[0][i] = put_bits_count(&s->reorder_pb[i]);
399  flush_put_bits(&s->reorder_pb[i]);
400  }
401  } else
402  score[0] = INT_MAX;
403 
404  best = 0;
405 
406  if (s->pict_type == AV_PICTURE_TYPE_P) {
408  int mx, my, pred_x, pred_y, dxy;
409  int16_t *motion_ptr;
410 
411  motion_ptr = ff_h263_pred_motion(&s->m, 0, 0, &pred_x, &pred_y);
412  if (s->m.mb_type[x + y * s->m.mb_stride] &
414  for (i = 0; i < 6; i++)
415  init_put_bits(&s->reorder_pb[i], reorder_buffer[1][i],
416  7 * 32);
417 
418  put_bits(&s->reorder_pb[5], vlc[1], vlc[0]);
419 
420  s->m.pb = s->reorder_pb[5];
421  mx = motion_ptr[0];
422  my = motion_ptr[1];
423  av_assert1(mx >= -32 && mx <= 31);
424  av_assert1(my >= -32 && my <= 31);
425  av_assert1(pred_x >= -32 && pred_x <= 31);
426  av_assert1(pred_y >= -32 && pred_y <= 31);
427  ff_h263_encode_motion(&s->m.pb, mx - pred_x, 1);
428  ff_h263_encode_motion(&s->m.pb, my - pred_y, 1);
429  s->reorder_pb[5] = s->m.pb;
430  score[1] += lambda * put_bits_count(&s->reorder_pb[5]);
431 
432  dxy = (mx & 1) + 2 * (my & 1);
433 
434  s->hdsp.put_pixels_tab[0][dxy](temp + 16*stride,
435  ref + (mx >> 1) +
436  stride * (my >> 1),
437  stride, 16);
438 
439  score[1] += encode_block(s, src + 16 * x, temp + 16*stride,
440  decoded, stride, 5, 64, lambda, 0);
441  best = score[1] <= score[0];
442 
444  score[2] = s->mecc.sse[0](NULL, src + 16 * x, ref,
445  stride, 16);
446  score[2] += vlc[1] * lambda;
447  if (score[2] < score[best] && mx == 0 && my == 0) {
448  best = 2;
449  s->hdsp.put_pixels_tab[0][0](decoded, ref, stride, 16);
450  put_bits(&s->pb, vlc[1], vlc[0]);
451  }
452  }
453 
454  if (best == 1) {
455  for (i = 0; i < 6; i++) {
456  count[1][i] = put_bits_count(&s->reorder_pb[i]);
457  flush_put_bits(&s->reorder_pb[i]);
458  }
459  } else {
460  motion_ptr[0] =
461  motion_ptr[1] =
462  motion_ptr[2] =
463  motion_ptr[3] =
464  motion_ptr[0 + 2 * s->m.b8_stride] =
465  motion_ptr[1 + 2 * s->m.b8_stride] =
466  motion_ptr[2 + 2 * s->m.b8_stride] =
467  motion_ptr[3 + 2 * s->m.b8_stride] = 0;
468  }
469  }
470 
471  s->rd_total += score[best];
472 
473  if (best != 2)
474  for (i = 5; i >= 0; i--)
475  avpriv_copy_bits(&s->pb, reorder_buffer[best][i],
476  count[best][i]);
477  if (best == 0)
478  s->hdsp.put_pixels_tab[0][0](decoded, temp, stride, 16);
479  }
480  s->m.first_slice_line = 0;
481  }
482  return 0;
483 }
484 
486 {
487  SVQ1EncContext *const s = avctx->priv_data;
488  int i;
489 
490  if (avctx->frame_number)
491  av_log(avctx, AV_LOG_DEBUG, "RD: %f\n",
492  s->rd_total / (double)(avctx->width * avctx->height *
493  avctx->frame_number));
494 
495  s->m.mb_type = NULL;
496  ff_mpv_common_end(&s->m);
497 
498  av_freep(&s->m.me.scratchpad);
499  av_freep(&s->m.me.map);
500  av_freep(&s->m.me.score_map);
501  av_freep(&s->mb_type);
502  av_freep(&s->dummy);
503  av_freep(&s->scratchbuf);
504 
505  for (i = 0; i < 3; i++) {
506  av_freep(&s->motion_val8[i]);
507  av_freep(&s->motion_val16[i]);
508  }
509 
512 
513  return 0;
514 }
515 
517 {
518  SVQ1EncContext *const s = avctx->priv_data;
519  int ret;
520 
521  if (avctx->width >= 4096 || avctx->height >= 4096) {
522  av_log(avctx, AV_LOG_ERROR, "Dimensions too large, maximum is 4095x4095\n");
523  return AVERROR(EINVAL);
524  }
525 
526  ff_hpeldsp_init(&s->hdsp, avctx->flags);
527  ff_me_cmp_init(&s->mecc, avctx);
528  ff_mpegvideoencdsp_init(&s->m.mpvencdsp, avctx);
529 
532  if (!s->current_picture || !s->last_picture) {
533  svq1_encode_end(avctx);
534  return AVERROR(ENOMEM);
535  }
536 
537  s->frame_width = avctx->width;
538  s->frame_height = avctx->height;
539 
540  s->y_block_width = (s->frame_width + 15) / 16;
541  s->y_block_height = (s->frame_height + 15) / 16;
542 
543  s->c_block_width = (s->frame_width / 4 + 15) / 16;
544  s->c_block_height = (s->frame_height / 4 + 15) / 16;
545 
546  s->avctx = avctx;
547  s->m.avctx = avctx;
548 
549  if ((ret = ff_mpv_common_init(&s->m)) < 0) {
550  svq1_encode_end(avctx);
551  return ret;
552  }
553 
555  s->m.me.temp =
556  s->m.me.scratchpad = av_mallocz((avctx->width + 64) *
557  2 * 16 * 2 * sizeof(uint8_t));
558  s->m.me.map = av_mallocz(ME_MAP_SIZE * sizeof(uint32_t));
559  s->m.me.score_map = av_mallocz(ME_MAP_SIZE * sizeof(uint32_t));
560  s->mb_type = av_mallocz((s->y_block_width + 1) *
561  s->y_block_height * sizeof(int16_t));
562  s->dummy = av_mallocz((s->y_block_width + 1) *
563  s->y_block_height * sizeof(int32_t));
565 
566  if (!s->m.me.temp || !s->m.me.scratchpad || !s->m.me.map ||
567  !s->m.me.score_map || !s->mb_type || !s->dummy) {
568  svq1_encode_end(avctx);
569  return AVERROR(ENOMEM);
570  }
571 
572  if (ARCH_PPC)
574  if (ARCH_X86)
576 
577  ff_h263_encode_init(&s->m); // mv_penalty
578 
579  return 0;
580 }
581 
583  const AVFrame *pict, int *got_packet)
584 {
585  SVQ1EncContext *const s = avctx->priv_data;
586  int i, ret;
587 
588  if ((ret = ff_alloc_packet2(avctx, pkt, s->y_block_width * s->y_block_height *
590  return ret;
591 
592  if (avctx->pix_fmt != AV_PIX_FMT_YUV410P) {
593  av_log(avctx, AV_LOG_ERROR, "unsupported pixel format\n");
594  return -1;
595  }
596 
597  if (!s->current_picture->data[0]) {
598  if ((ret = ff_get_buffer(avctx, s->current_picture, 0)) < 0) {
599  return ret;
600  }
601  }
602  if (!s->last_picture->data[0]) {
603  ret = ff_get_buffer(avctx, s->last_picture, 0);
604  if (ret < 0)
605  return ret;
606  }
607  if (!s->scratchbuf) {
608  s->scratchbuf = av_malloc_array(s->current_picture->linesize[0], 16 * 3);
609  if (!s->scratchbuf)
610  return AVERROR(ENOMEM);
611  }
612 
614 
615  init_put_bits(&s->pb, pkt->data, pkt->size);
616 
617  if (avctx->gop_size && (avctx->frame_number % avctx->gop_size))
619  else
621  s->quality = pict->quality;
622 
623 #if FF_API_CODED_FRAME
625  avctx->coded_frame->pict_type = s->pict_type;
628 #endif
629 
631 
633  for (i = 0; i < 3; i++) {
634  int ret = svq1_encode_plane(s, i,
635  pict->data[i],
636  s->last_picture->data[i],
637  s->current_picture->data[i],
638  s->frame_width / (i ? 4 : 1),
639  s->frame_height / (i ? 4 : 1),
640  pict->linesize[i],
641  s->current_picture->linesize[i]);
642  emms_c();
643  if (ret < 0) {
644  int j;
645  for (j = 0; j < i; j++) {
646  av_freep(&s->motion_val8[j]);
647  av_freep(&s->motion_val16[j]);
648  }
649  av_freep(&s->scratchbuf);
650  return -1;
651  }
652  }
653 
654  // avpriv_align_put_bits(&s->pb);
655  while (put_bits_count(&s->pb) & 31)
656  put_bits(&s->pb, 1, 0);
657 
658  flush_put_bits(&s->pb);
659 
660  pkt->size = put_bits_count(&s->pb) / 8;
661  if (s->pict_type == AV_PICTURE_TYPE_I)
662  pkt->flags |= AV_PKT_FLAG_KEY;
663  *got_packet = 1;
664 
665  return 0;
666 }
667 
668 #define OFFSET(x) offsetof(struct SVQ1EncContext, x)
669 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
670 static const AVOption options[] = {
671  { "motion-est", "Motion estimation algorithm", OFFSET(motion_est), AV_OPT_TYPE_INT, { .i64 = FF_ME_EPZS }, FF_ME_ZERO, FF_ME_XONE, VE, "motion-est"},
672  { "zero", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_ZERO }, 0, 0, FF_MPV_OPT_FLAGS, "motion-est" },
673  { "epzs", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_EPZS }, 0, 0, FF_MPV_OPT_FLAGS, "motion-est" },
674  { "xone", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_XONE }, 0, 0, FF_MPV_OPT_FLAGS, "motion-est" },
675 
676  { NULL },
677 };
678 
679 static const AVClass svq1enc_class = {
680  .class_name = "svq1enc",
681  .item_name = av_default_item_name,
682  .option = options,
683  .version = LIBAVUTIL_VERSION_INT,
684 };
685 
687  .name = "svq1",
688  .long_name = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 1 / Sorenson Video 1 / SVQ1"),
689  .type = AVMEDIA_TYPE_VIDEO,
690  .id = AV_CODEC_ID_SVQ1,
691  .priv_data_size = sizeof(SVQ1EncContext),
692  .priv_class = &svq1enc_class,
694  .encode2 = svq1_encode_frame,
695  .close = svq1_encode_end,
696  .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUV410P,
697  AV_PIX_FMT_NONE },
698 };
uint8_t * scratchpad
data area for the ME algo, so that the ME does not need to malloc/free.
Definition: motion_est.h:52
av_cold void ff_me_cmp_init(MECmpContext *c, AVCodecContext *avctx)
Definition: me_cmp.c:1035
#define NULL
Definition: coverity.c:32
static const AVClass svq1enc_class
Definition: svq1enc.c:679
int size
This structure describes decoded (raw) audio or video data.
Definition: frame.h:300
uint16_t * mb_type
Definition: svq1enc.h:67
int16_t(* p_mv_table)[2]
MV table (1MV per MB) P-frame encoding.
Definition: mpegvideo.h:248
AVOption.
Definition: opt.h:246
int ff_side_data_set_encoder_stats(AVPacket *pkt, int quality, int64_t *error, int error_count, int pict_type)
Definition: avpacket.c:728
void ff_estimate_p_frame_motion(MpegEncContext *s, int mb_x, int mb_y)
Definition: motion_est.c:885
uint8_t * mb_mean
Table for MB luminance.
Definition: mpegpicture.h:74
#define THRESHOLD_MULTIPLIER
Definition: svq1enc.c:78
int frame_height
Definition: svq1enc.h:55
#define ARCH_PPC
Definition: config.h:29
MpegEncContext m
Definition: svq1enc.h:38
static void put_bits(Jpeg2000EncoderContext *s, int val, int n)
put n times val bit
Definition: j2kenc.c:208
#define LIBAVUTIL_VERSION_INT
Definition: version.h:85
else temp
Definition: vf_mcdeint.c:256
void ff_h263_encode_init(MpegEncContext *s)
Definition: ituh263enc.c:761
static av_cold int init(AVCodecContext *avctx)
Definition: avrndec.c:35
uint16_t * mb_var
Table for MB variances.
Definition: mpegpicture.h:65
AVFrame * current_picture
Definition: svq1enc.h:42
int size
Definition: packet.h:356
const char * av_default_item_name(void *ptr)
Return the context name.
Definition: log.c:235
void avpriv_copy_bits(PutBitContext *pb, const uint8_t *src, int length)
Copy the content of src to the bitstream.
Definition: bitstream.c:64
#define FF_MPV_OPT_FLAGS
Definition: mpegvideo.h:615
static int svq1_encode_plane(SVQ1EncContext *s, int plane, unsigned char *src_plane, unsigned char *ref_plane, unsigned char *decoded_plane, int width, int height, int src_stride, int stride)
Definition: svq1enc.c:247
enum AVPixelFormat pix_fmt
Pixel format, see AV_PIX_FMT_xxx.
Definition: avcodec.h:736
int c_block_height
Definition: svq1enc.h:63
uint32_t * score_map
map to store the scores
Definition: motion_est.h:59
mpegvideo header.
int scene_change_score
Definition: motion_est.h:87
static AVPacket pkt
#define SVQ1_BLOCK_INTRA
Definition: svq1.h:43
#define FF_LAMBDA_SHIFT
Definition: avutil.h:225
int stride
Definition: mace.c:144
AVCodec.
Definition: codec.h:190
int qscale
QP.
Definition: mpegvideo.h:204
int16_t * ff_h263_pred_motion(MpegEncContext *s, int block, int dir, int *px, int *py)
Definition: h263.c:307
void ff_h263_encode_motion(PutBitContext *pb, int val, int f_code)
Definition: ituh263enc.c:646
svq1 code books.
#define FF_ME_ZERO
Definition: motion_est.h:40
static const int8_t svq1_intra_codebook_sum[4][16 *6]
Definition: svq1enc_cb.h:59
frame_type
static void init_block_index(MpegEncContext *s)
Definition: svq1enc.c:238
static int16_t block[64]
Definition: dct.c:115
const uint16_t ff_svq1_frame_size_table[7][2]
Definition: svq1.c:40
const char * class_name
The name of the class; usually it is the same name as the context structure type to which the AVClass...
Definition: log.h:72
#define SVQ1_BLOCK_SKIP
Definition: svq1.h:40
#define CANDIDATE_MB_TYPE_INTER
Definition: mpegutils.h:105
#define av_assert0(cond)
assert() equivalent, that is always enabled.
Definition: avassert.h:37
int ff_alloc_packet2(AVCodecContext *avctx, AVPacket *avpkt, int64_t size, int64_t min_size)
Check AVPacket size and/or allocate data.
Definition: encode.c:32
static av_cold int svq1_encode_init(AVCodecContext *avctx)
Definition: svq1enc.c:516
int y_block_height
Definition: svq1enc.h:59
av_cold void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c, AVCodecContext *avctx)
uint8_t
#define ME_MAP_SIZE
Definition: motion_est.h:38
#define av_cold
Definition: attributes.h:88
AVFrame * av_frame_alloc(void)
Allocate an AVFrame and set its fields to default values.
Definition: frame.c:190
#define av_assert2(cond)
assert() equivalent, that does lie in speed critical code.
Definition: avassert.h:64
int frame_width
Definition: svq1enc.h:54
const int8_t *const ff_svq1_inter_codebooks[6]
Definition: svq1_cb.h:776
#define emms_c()
Definition: internal.h:55
int16_t encoded_block_levels[6][7][256]
Definition: svq1enc.h:65
enum AVPictureType pict_type
Definition: svq1enc.h:47
Picture current_picture
copy of the current picture structure.
Definition: mpegvideo.h:180
#define height
uint8_t * data
Definition: packet.h:355
int mb_height
number of MBs horizontally & vertically
Definition: mpegvideo.h:129
#define AV_INPUT_BUFFER_MIN_SIZE
minimum encoding buffer size Used to avoid some checks during header writing.
Definition: avcodec.h:222
#define av_log(a,...)
#define AV_PKT_FLAG_KEY
The packet contains a keyframe.
Definition: packet.h:388
int c_block_width
Definition: svq1enc.h:62
static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2, intptr_t size)
Definition: svq1enc.c:80
#define src
Definition: vp8dsp.c:254
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:269
int64_t rd_total
Definition: svq1enc.h:72
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:176
#define MAX_MB_BYTES
Definition: mpegutils.h:47
#define ARCH_X86
Definition: config.h:38
#define AVERROR(e)
Definition: error.h:43
uint8_t * scratchbuf
Definition: svq1enc.h:74
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:203
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
Definition: internal.h:188
#define AV_LOG_DEBUG
Stuff which is only useful for libav* developers.
Definition: log.h:197
int y_block_width
Definition: svq1enc.h:58
int flags
AV_CODEC_FLAG_*.
Definition: avcodec.h:606
uint8_t * buf
Definition: put_bits.h:38
simple assert() macros that are a bit more flexible than ISO C assert().
void * av_mallocz(size_t size)
Allocate a memory block with alignment suitable for all memory accesses (including vectors if availab...
Definition: mem.c:237
MpegvideoEncDSPContext mpvencdsp
Definition: mpegvideo.h:233
const char * name
Name of the codec implementation.
Definition: codec.h:197
int16_t(*[3] motion_val16)[2]
Definition: svq1enc.h:70
uint16_t * mb_type
Table for candidate MB types for encoding (defines in mpegutils.h)
Definition: mpegvideo.h:291
#define SVQ1_BLOCK_INTER
Definition: svq1.h:41
static const uint8_t offset[127][2]
Definition: vf_spp.c:93
void ff_mpv_common_end(MpegEncContext *s)
Definition: mpegvideo.c:1138
static int encode_block(SVQ1EncContext *s, uint8_t *src, uint8_t *ref, uint8_t *decoded, int stride, int level, int threshold, int lambda, int intra)
Definition: svq1enc.c:90
int flags
A combination of AV_PKT_FLAG values.
Definition: packet.h:361
int motion_est
Definition: svq1enc.h:76
Sorenson Vector Quantizer #1 (SVQ1) video codec.
static int put_bits_count(PutBitContext *s)
Definition: put_bits.h:67
static char * split(char *message, char delim)
Definition: af_channelmap.c:81
av_cold void ff_hpeldsp_init(HpelDSPContext *c, int flags)
Definition: hpeldsp.c:338
PutBitContext reorder_pb[6]
Definition: svq1enc.h:52
enum AVPictureType pict_type
Picture type of the frame.
Definition: frame.h:383
#define av_assert1(cond)
assert() equivalent, that does not lie in speed critical code.
Definition: avassert.h:53
int(* ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, intptr_t size)
Definition: svq1enc.h:78
Picture new_picture
copy of the source picture structure for encoding.
Definition: mpegvideo.h:174
const uint8_t ff_svq1_block_type_vlc[4][2]
Definition: svq1_vlc.h:27
#define width
int width
picture width / height.
Definition: avcodec.h:699
#define VE
Definition: svq1enc.c:669
uint8_t w
Definition: llviddspenc.c:38
int16_t(*[2] motion_val)[2]
Definition: mpegpicture.h:53
Picture * current_picture_ptr
pointer to the current picture
Definition: mpegvideo.h:184
MECmpContext mecc
Definition: svq1enc.h:40
int32_t
AVCodec ff_svq1_encoder
Definition: svq1enc.c:686
#define s(width, name)
Definition: cbs_vp9.c:257
int quality
quality (between 1 (good) and FF_LAMBDA_MAX (bad))
Definition: frame.h:423
MotionEstContext me
Definition: mpegvideo.h:282
#define FF_ARRAY_ELEMS(a)
const uint8_t ff_svq1_inter_multistage_vlc[6][8][2]
Definition: svq1_vlc.h:50
int block_index[6]
index to current MB in block based arrays with edges
Definition: mpegvideo.h:293
static av_cold int svq1_encode_end(AVCodecContext *avctx)
Definition: svq1enc.c:485
PutBitContext pb
Definition: svq1enc.h:44
int first_slice_line
used in MPEG-4 too to handle resync markers
Definition: mpegvideo.h:436
uint16_t * mc_mb_var
Table for motion compensated MB variances.
Definition: mpegpicture.h:68
int16_t(*[3] motion_val8)[2]
Definition: svq1enc.h:69
Half-pel DSP functions.
#define FF_ME_XONE
Definition: motion_est.h:42
#define FF_LAMBDA_SCALE
Definition: avutil.h:226
unsigned int lambda2
(lambda*lambda) >> FF_LAMBDA_SHIFT
Definition: mpegvideo.h:207
Libavcodec external API header.
ptrdiff_t linesize
line size, in bytes, may be different from width
Definition: mpegvideo.h:134
static void svq1_write_header(SVQ1EncContext *s, int frame_type)
Definition: svq1enc.c:43
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
Definition: frame.h:331
av_cold void ff_svq1enc_init_ppc(SVQ1EncContext *c)
main external API structure.
Definition: avcodec.h:526
int height
picture size. must be a multiple of 16
Definition: mpegvideo.h:100
#define FF_ME_EPZS
Definition: motion_est.h:41
int ff_get_buffer(AVCodecContext *avctx, AVFrame *frame, int flags)
Get a buffer for a frame.
Definition: decode.c:1854
op_pixels_func put_pixels_tab[4][4]
Halfpel motion compensation with rounding (a+b+1)>>1.
Definition: hpeldsp.h:56
uint8_t * buf_end
Definition: put_bits.h:38
planar YUV 4:1:0, 9bpp, (1 Cr & Cb sample per 4x4 Y samples)
Definition: pixfmt.h:72
Describe the class of an AVClass context structure.
Definition: log.h:67
struct AVFrame * f
Definition: mpegpicture.h:46
int f_code
forward MV resolution
Definition: mpegvideo.h:238
int pict_type
AV_PICTURE_TYPE_I, AV_PICTURE_TYPE_P, AV_PICTURE_TYPE_B, ...
Definition: mpegvideo.h:212
static enum AVPixelFormat pix_fmts[]
Definition: libkvazaar.c:275
int motion_est
ME algorithm.
Definition: mpegvideo.h:258
void ff_svq1enc_init_x86(SVQ1EncContext *c)
Definition: svq1enc_init.c:32
AVFrame * last_picture
Definition: svq1enc.h:43
int ff_init_me(MpegEncContext *s)
Definition: motion_est.c:306
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:314
uint8_t level
Definition: svq3.c:210
int b8_stride
2*mb_width+1 used for some 8x8 block arrays to allow simple addressing
Definition: mpegvideo.h:131
me_cmp_func sse[6]
Definition: me_cmp.h:57
MpegEncContext.
Definition: mpegvideo.h:81
struct AVCodecContext * avctx
Definition: mpegvideo.h:98
int gop_size
the number of pictures in a group of pictures, or 0 for intra_only
Definition: avcodec.h:721
PutBitContext pb
bit output
Definition: mpegvideo.h:151
int
MECmpContext mecc
Definition: mpegvideo.h:231
#define FF_DISABLE_DEPRECATION_WARNINGS
Definition: internal.h:84
const int8_t *const ff_svq1_intra_codebooks[6]
Definition: svq1_cb.h:1519
common internal api header.
int mb_stride
mb_width+1 used for some arrays to allow simple addressing of left & top MBs without sig11 ...
Definition: mpegvideo.h:130
static void flush_put_bits(PutBitContext *s)
Pad the end of the output stream with zeros.
Definition: put_bits.h:101
AVCodecContext * avctx
Definition: svq1enc.h:39
static int ref[MAX_W *MAX_W]
Definition: jpeg2000dwt.c:107
const uint8_t ff_svq1_intra_multistage_vlc[6][8][2]
Definition: svq1_vlc.h:33
Picture last_picture
copy of the previous picture structure.
Definition: mpegvideo.h:162
Picture * last_picture_ptr
pointer to the previous picture.
Definition: mpegvideo.h:182
static const int8_t svq1_inter_codebook_sum[4][16 *6]
Definition: svq1enc_cb.h:32
attribute_deprecated AVFrame * coded_frame
the picture in the bitstream
Definition: avcodec.h:1776
const uint16_t ff_svq1_inter_mean_vlc[512][2]
Definition: svq1_vlc.h:136
#define OFFSET(x)
Definition: svq1enc.c:668
static void init_put_bits(PutBitContext *s, uint8_t *buffer, int buffer_size)
Initialize the PutBitContext s.
Definition: put_bits.h:48
uint32_t * map
map to avoid duplicate evaluations
Definition: motion_est.h:58
static const AVOption options[]
Definition: svq1enc.c:670
#define CANDIDATE_MB_TYPE_INTRA
Definition: mpegutils.h:104
void * priv_data
Definition: avcodec.h:553
#define PICT_FRAME
Definition: mpegutils.h:39
av_cold int ff_mpv_common_init(MpegEncContext *s)
init common structure for both encoder and decoder.
Definition: mpegvideo.c:890
static av_always_inline int diff(const uint32_t a, const uint32_t b)
int picture_structure
Definition: mpegvideo.h:461
int dia_size
ME diamond size & shape.
Definition: avcodec.h:954
#define FF_ENABLE_DEPRECATION_WARNINGS
Definition: internal.h:85
uint32_t * dummy
Definition: svq1enc.h:68
int key_frame
1 -> keyframe, 0-> not
Definition: frame.h:378
void ff_fix_long_mvs(MpegEncContext *s, uint8_t *field_select_table, int field_select, int16_t(*mv_table)[2], int f_code, int type, int truncate)
Definition: motion_est.c:1699
static int svq1_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *pict, int *got_packet)
Definition: svq1enc.c:582
HpelDSPContext hdsp
Definition: svq1enc.h:41
int frame_number
Frame counter, set by libavcodec.
Definition: avcodec.h:1217
void ff_fix_long_p_mvs(MpegEncContext *s, int type)
Definition: motion_est.c:1650
uint32_t * mb_type
types and macros are defined in mpegutils.h
Definition: mpegpicture.h:56
#define av_freep(p)
uint8_t * temp
Definition: motion_est.h:56
#define av_malloc_array(a, b)
#define FFSWAP(type, a, b)
Definition: common.h:99
#define QUALITY_THRESHOLD
Definition: svq1enc.c:77
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
This structure stores compressed data.
Definition: packet.h:332
const uint16_t ff_svq1_intra_mean_vlc[256][2]
Definition: svq1_vlc.h:67
int ff_match_2uint16(const uint16_t(*tab)[2], int size, int a, int b)
Return the index into tab at which {a,b} match elements {[0],[1]} of tab.
Definition: utils.c:1847
Predicted.
Definition: avutil.h:275
unsigned int lambda
Lagrange multiplier used in rate distortion.
Definition: mpegvideo.h:206