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
svq1_write_header(SVQ1EncContext * s,int frame_type)43 static void svq1_write_header(SVQ1EncContext *s, int frame_type)
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
62 i = ff_match_2uint16((void*)ff_svq1_frame_size_table,
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
ssd_int8_vs_int16_c(const int8_t * pix1,const int16_t * pix2,intptr_t size)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
encode_block(SVQ1EncContext * s,uint8_t * src,uint8_t * ref,uint8_t * decoded,int stride,int level,int threshold,int lambda,int intra)90 static int encode_block(SVQ1EncContext *s, uint8_t *src, uint8_t *ref,
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
init_block_index(MpegEncContext * s)238 static void init_block_index(MpegEncContext *s){
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
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)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;
272 s->m.current_picture_ptr = &s->m.current_picture;
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 >>
297 FF_LAMBDA_SHIFT;
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
313 s->m.current_picture.mb_mean = (uint8_t *)s->dummy;
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
349 ff_fix_long_p_mvs(&s->m, CANDIDATE_MB_TYPE_INTRA);
350 ff_fix_long_mvs(&s->m, NULL, 0, s->m.p_mv_table, s->m.f_code,
351 CANDIDATE_MB_TYPE_INTER, 0);
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] &
386 CANDIDATE_MB_TYPE_INTRA)) {
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) {
391 const uint8_t *vlc = ff_svq1_block_type_vlc[SVQ1_BLOCK_INTRA];
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) {
407 const uint8_t *vlc = ff_svq1_block_type_vlc[SVQ1_BLOCK_INTER];
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] &
413 CANDIDATE_MB_TYPE_INTER) {
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
443 vlc = ff_svq1_block_type_vlc[SVQ1_BLOCK_SKIP];
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 ff_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
svq1_encode_end(AVCodecContext * avctx)485 static av_cold int svq1_encode_end(AVCodecContext *avctx)
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
510 av_frame_free(&s->current_picture);
511 av_frame_free(&s->last_picture);
512
513 return 0;
514 }
515
svq1_encode_init(AVCodecContext * avctx)516 static av_cold int svq1_encode_init(AVCodecContext *avctx)
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
530 s->current_picture = av_frame_alloc();
531 s->last_picture = av_frame_alloc();
532 if (!s->current_picture || !s->last_picture) {
533 return AVERROR(ENOMEM);
534 }
535
536 s->frame_width = avctx->width;
537 s->frame_height = avctx->height;
538
539 s->y_block_width = (s->frame_width + 15) / 16;
540 s->y_block_height = (s->frame_height + 15) / 16;
541
542 s->c_block_width = (s->frame_width / 4 + 15) / 16;
543 s->c_block_height = (s->frame_height / 4 + 15) / 16;
544
545 s->avctx = avctx;
546 s->m.avctx = avctx;
547
548 if ((ret = ff_mpv_common_init(&s->m)) < 0) {
549 return ret;
550 }
551
552 s->m.picture_structure = PICT_FRAME;
553 s->m.me.temp =
554 s->m.me.scratchpad = av_mallocz((avctx->width + 64) *
555 2 * 16 * 2 * sizeof(uint8_t));
556 s->m.me.map = av_mallocz(ME_MAP_SIZE * sizeof(uint32_t));
557 s->m.me.score_map = av_mallocz(ME_MAP_SIZE * sizeof(uint32_t));
558 s->mb_type = av_mallocz((s->y_block_width + 1) *
559 s->y_block_height * sizeof(int16_t));
560 s->dummy = av_mallocz((s->y_block_width + 1) *
561 s->y_block_height * sizeof(int32_t));
562 s->ssd_int8_vs_int16 = ssd_int8_vs_int16_c;
563
564 if (!s->m.me.temp || !s->m.me.scratchpad || !s->m.me.map ||
565 !s->m.me.score_map || !s->mb_type || !s->dummy) {
566 return AVERROR(ENOMEM);
567 }
568
569 if (ARCH_PPC)
570 ff_svq1enc_init_ppc(s);
571 if (ARCH_X86)
572 ff_svq1enc_init_x86(s);
573
574 ff_h263_encode_init(&s->m); // mv_penalty
575
576 return 0;
577 }
578
svq1_encode_frame(AVCodecContext * avctx,AVPacket * pkt,const AVFrame * pict,int * got_packet)579 static int svq1_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
580 const AVFrame *pict, int *got_packet)
581 {
582 SVQ1EncContext *const s = avctx->priv_data;
583 int i, ret;
584
585 if ((ret = ff_alloc_packet2(avctx, pkt, s->y_block_width * s->y_block_height *
586 MAX_MB_BYTES*3 + AV_INPUT_BUFFER_MIN_SIZE, 0)) < 0)
587 return ret;
588
589 if (avctx->pix_fmt != AV_PIX_FMT_YUV410P) {
590 av_log(avctx, AV_LOG_ERROR, "unsupported pixel format\n");
591 return -1;
592 }
593
594 if (!s->current_picture->data[0]) {
595 if ((ret = ff_get_buffer(avctx, s->current_picture, 0)) < 0) {
596 return ret;
597 }
598 }
599 if (!s->last_picture->data[0]) {
600 ret = ff_get_buffer(avctx, s->last_picture, 0);
601 if (ret < 0)
602 return ret;
603 }
604 if (!s->scratchbuf) {
605 s->scratchbuf = av_malloc_array(s->current_picture->linesize[0], 16 * 3);
606 if (!s->scratchbuf)
607 return AVERROR(ENOMEM);
608 }
609
610 FFSWAP(AVFrame*, s->current_picture, s->last_picture);
611
612 init_put_bits(&s->pb, pkt->data, pkt->size);
613
614 if (avctx->gop_size && (avctx->frame_number % avctx->gop_size))
615 s->pict_type = AV_PICTURE_TYPE_P;
616 else
617 s->pict_type = AV_PICTURE_TYPE_I;
618 s->quality = pict->quality;
619
620 #if FF_API_CODED_FRAME
621 FF_DISABLE_DEPRECATION_WARNINGS
622 avctx->coded_frame->pict_type = s->pict_type;
623 avctx->coded_frame->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
624 FF_ENABLE_DEPRECATION_WARNINGS
625 #endif
626
627 ff_side_data_set_encoder_stats(pkt, pict->quality, NULL, 0, s->pict_type);
628
629 svq1_write_header(s, s->pict_type);
630 for (i = 0; i < 3; i++) {
631 int ret = svq1_encode_plane(s, i,
632 pict->data[i],
633 s->last_picture->data[i],
634 s->current_picture->data[i],
635 s->frame_width / (i ? 4 : 1),
636 s->frame_height / (i ? 4 : 1),
637 pict->linesize[i],
638 s->current_picture->linesize[i]);
639 emms_c();
640 if (ret < 0) {
641 int j;
642 for (j = 0; j < i; j++) {
643 av_freep(&s->motion_val8[j]);
644 av_freep(&s->motion_val16[j]);
645 }
646 av_freep(&s->scratchbuf);
647 return -1;
648 }
649 }
650
651 // align_put_bits(&s->pb);
652 while (put_bits_count(&s->pb) & 31)
653 put_bits(&s->pb, 1, 0);
654
655 flush_put_bits(&s->pb);
656
657 pkt->size = put_bits_count(&s->pb) / 8;
658 if (s->pict_type == AV_PICTURE_TYPE_I)
659 pkt->flags |= AV_PKT_FLAG_KEY;
660 *got_packet = 1;
661
662 return 0;
663 }
664
665 #define OFFSET(x) offsetof(struct SVQ1EncContext, x)
666 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
667 static const AVOption options[] = {
668 { "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"},
669 { "zero", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_ZERO }, 0, 0, FF_MPV_OPT_FLAGS, "motion-est" },
670 { "epzs", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_EPZS }, 0, 0, FF_MPV_OPT_FLAGS, "motion-est" },
671 { "xone", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_XONE }, 0, 0, FF_MPV_OPT_FLAGS, "motion-est" },
672
673 { NULL },
674 };
675
676 static const AVClass svq1enc_class = {
677 .class_name = "svq1enc",
678 .item_name = av_default_item_name,
679 .option = options,
680 .version = LIBAVUTIL_VERSION_INT,
681 };
682
683 AVCodec ff_svq1_encoder = {
684 .name = "svq1",
685 .long_name = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 1 / Sorenson Video 1 / SVQ1"),
686 .type = AVMEDIA_TYPE_VIDEO,
687 .id = AV_CODEC_ID_SVQ1,
688 .priv_data_size = sizeof(SVQ1EncContext),
689 .priv_class = &svq1enc_class,
690 .init = svq1_encode_init,
691 .encode2 = svq1_encode_frame,
692 .close = svq1_encode_end,
693 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
694 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUV410P,
695 AV_PIX_FMT_NONE },
696 };
697