3 * Copyright (c) 2000,2001 Fabrice Bellard
4 * Copyright (c) 2002-2010 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "libavutil/attributes.h"
24 #include "libavutil/log.h"
25 #include "libavutil/opt.h"
26 #include "mpegutils.h"
27 #include "mpegvideo.h"
29 #include "mpeg4video.h"
31 /* The uni_DCtab_* tables below contain unified bits+length tables to encode DC
32 * differences in MPEG-4. Unified in the sense that the specification specifies
33 * this encoding in several steps. */
34 static uint8_t uni_DCtab_lum_len[512];
35 static uint8_t uni_DCtab_chrom_len[512];
36 static uint16_t uni_DCtab_lum_bits[512];
37 static uint16_t uni_DCtab_chrom_bits[512];
39 /* Unified encoding tables for run length encoding of coefficients.
40 * Unified in the sense that the specification specifies the encoding in several steps. */
41 static uint32_t uni_mpeg4_intra_rl_bits[64 * 64 * 2 * 2];
42 static uint8_t uni_mpeg4_intra_rl_len[64 * 64 * 2 * 2];
43 static uint32_t uni_mpeg4_inter_rl_bits[64 * 64 * 2 * 2];
44 static uint8_t uni_mpeg4_inter_rl_len[64 * 64 * 2 * 2];
46 //#define UNI_MPEG4_ENC_INDEX(last, run, level) ((last) * 128 + (run) * 256 + (level))
47 //#define UNI_MPEG4_ENC_INDEX(last, run, level) ((last) * 128 * 64 + (run) + (level) * 64)
48 #define UNI_MPEG4_ENC_INDEX(last, run, level) ((last) * 128 * 64 + (run) * 128 + (level))
61 * Return the number of bits that encoding the 8x8 block in block would need.
62 * @param[in] block_last_index last index in scantable order that refers to a non zero element in block.
64 static inline int get_block_rate(MpegEncContext *s, int16_t block[64],
65 int block_last_index, uint8_t scantable[64])
71 for (j = 1; j <= block_last_index; j++) {
72 const int index = scantable[j];
73 int level = block[index];
76 if ((level & (~127)) == 0) {
77 if (j < block_last_index)
78 rate += s->intra_ac_vlc_length[UNI_AC_ENC_INDEX(j - last - 1, level)];
80 rate += s->intra_ac_vlc_last_length[UNI_AC_ENC_INDEX(j - last - 1, level)];
82 rate += s->ac_esc_length;
92 * Restore the ac coefficients in block that have been changed by decide_ac_pred().
93 * This function also restores s->block_last_index.
94 * @param[in,out] block MB coefficients, these will be restored
95 * @param[in] dir ac prediction direction for each 8x8 block
96 * @param[out] st scantable for each 8x8 block
97 * @param[in] zigzag_last_index index referring to the last non zero coefficient in zigzag order
99 static inline void restore_ac_coeffs(MpegEncContext *s, int16_t block[6][64],
100 const int dir[6], uint8_t *st[6],
101 const int zigzag_last_index[6])
104 memcpy(s->block_last_index, zigzag_last_index, sizeof(int) * 6);
106 for (n = 0; n < 6; n++) {
107 int16_t *ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
109 st[n] = s->intra_scantable.permutated;
112 for (i = 1; i < 8; i++)
113 block[n][s->idsp.idct_permutation[i]] = ac_val[i + 8];
115 /* left prediction */
116 for (i = 1; i < 8; i++)
117 block[n][s->idsp.idct_permutation[i << 3]] = ac_val[i];
123 * Return the optimal value (0 or 1) for the ac_pred element for the given MB in MPEG-4.
124 * This function will also update s->block_last_index and s->ac_val.
125 * @param[in,out] block MB coefficients, these will be updated if 1 is returned
126 * @param[in] dir ac prediction direction for each 8x8 block
127 * @param[out] st scantable for each 8x8 block
128 * @param[out] zigzag_last_index index referring to the last non zero coefficient in zigzag order
130 static inline int decide_ac_pred(MpegEncContext *s, int16_t block[6][64],
131 const int dir[6], uint8_t *st[6],
132 int zigzag_last_index[6])
136 int8_t *const qscale_table = s->current_picture.qscale_table;
138 memcpy(zigzag_last_index, s->block_last_index, sizeof(int) * 6);
140 for (n = 0; n < 6; n++) {
141 int16_t *ac_val, *ac_val1;
143 score -= get_block_rate(s, block[n], s->block_last_index[n],
144 s->intra_scantable.permutated);
146 ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
149 const int xy = s->mb_x + s->mb_y * s->mb_stride - s->mb_stride;
151 ac_val -= s->block_wrap[n] * 16;
152 if (s->mb_y == 0 || s->qscale == qscale_table[xy] || n == 2 || n == 3) {
154 for (i = 1; i < 8; i++) {
155 const int level = block[n][s->idsp.idct_permutation[i]];
156 block[n][s->idsp.idct_permutation[i]] = level - ac_val[i + 8];
157 ac_val1[i] = block[n][s->idsp.idct_permutation[i << 3]];
158 ac_val1[i + 8] = level;
161 /* different qscale, we must rescale */
162 for (i = 1; i < 8; i++) {
163 const int level = block[n][s->idsp.idct_permutation[i]];
164 block[n][s->idsp.idct_permutation[i]] = level - ROUNDED_DIV(ac_val[i + 8] * qscale_table[xy], s->qscale);
165 ac_val1[i] = block[n][s->idsp.idct_permutation[i << 3]];
166 ac_val1[i + 8] = level;
169 st[n] = s->intra_h_scantable.permutated;
171 const int xy = s->mb_x - 1 + s->mb_y * s->mb_stride;
172 /* left prediction */
174 if (s->mb_x == 0 || s->qscale == qscale_table[xy] || n == 1 || n == 3) {
176 for (i = 1; i < 8; i++) {
177 const int level = block[n][s->idsp.idct_permutation[i << 3]];
178 block[n][s->idsp.idct_permutation[i << 3]] = level - ac_val[i];
180 ac_val1[i + 8] = block[n][s->idsp.idct_permutation[i]];
183 /* different qscale, we must rescale */
184 for (i = 1; i < 8; i++) {
185 const int level = block[n][s->idsp.idct_permutation[i << 3]];
186 block[n][s->idsp.idct_permutation[i << 3]] = level - ROUNDED_DIV(ac_val[i] * qscale_table[xy], s->qscale);
188 ac_val1[i + 8] = block[n][s->idsp.idct_permutation[i]];
191 st[n] = s->intra_v_scantable.permutated;
194 for (i = 63; i > 0; i--) // FIXME optimize
195 if (block[n][st[n][i]])
197 s->block_last_index[n] = i;
199 score += get_block_rate(s, block[n], s->block_last_index[n], st[n]);
205 restore_ac_coeffs(s, block, dir, st, zigzag_last_index);
211 * modify mb_type & qscale so that encoding is actually possible in MPEG-4
213 void ff_clean_mpeg4_qscales(MpegEncContext *s)
216 int8_t *const qscale_table = s->current_picture.qscale_table;
218 ff_clean_h263_qscales(s);
220 if (s->pict_type == AV_PICTURE_TYPE_B) {
222 /* ok, come on, this isn't funny anymore, there's more code for
223 * handling this MPEG-4 mess than for the actual adaptive quantization */
225 for (i = 0; i < s->mb_num; i++) {
226 int mb_xy = s->mb_index2xy[i];
227 odd += qscale_table[mb_xy] & 1;
230 if (2 * odd > s->mb_num)
235 for (i = 0; i < s->mb_num; i++) {
236 int mb_xy = s->mb_index2xy[i];
237 if ((qscale_table[mb_xy] & 1) != odd)
238 qscale_table[mb_xy]++;
239 if (qscale_table[mb_xy] > 31)
240 qscale_table[mb_xy] = 31;
243 for (i = 1; i < s->mb_num; i++) {
244 int mb_xy = s->mb_index2xy[i];
245 if (qscale_table[mb_xy] != qscale_table[s->mb_index2xy[i - 1]] &&
246 (s->mb_type[mb_xy] & CANDIDATE_MB_TYPE_DIRECT)) {
247 s->mb_type[mb_xy] |= CANDIDATE_MB_TYPE_BIDIR;
254 * Encode the dc value.
255 * @param n block index (0-3 are luma, 4-5 are chroma)
257 static inline void mpeg4_encode_dc(PutBitContext *s, int level, int n)
260 /* DC will overflow if level is outside the [-255,255] range. */
264 put_bits(s, uni_DCtab_lum_len[level], uni_DCtab_lum_bits[level]);
267 put_bits(s, uni_DCtab_chrom_len[level], uni_DCtab_chrom_bits[level]);
271 /* find number of bits */
281 put_bits(s, ff_mpeg4_DCtab_lum[size][1], ff_mpeg4_DCtab_lum[size][0]);
284 put_bits(s, ff_mpeg4_DCtab_chrom[size][1], ff_mpeg4_DCtab_chrom[size][0]);
287 /* encode remaining bits */
290 level = (-level) ^ ((1 << size) - 1);
291 put_bits(s, size, level);
298 static inline int mpeg4_get_dc_length(int level, int n)
301 return uni_DCtab_lum_len[level + 256];
303 return uni_DCtab_chrom_len[level + 256];
307 * Encode an 8x8 block.
308 * @param n block index (0-3 are luma, 4-5 are chroma)
310 static inline void mpeg4_encode_block(MpegEncContext *s,
311 int16_t *block, int n, int intra_dc,
312 uint8_t *scan_table, PutBitContext *dc_pb,
313 PutBitContext *ac_pb)
315 int i, last_non_zero;
318 const int last_index = s->block_last_index[n];
320 if (s->mb_intra) { // Note gcc (3.2.1 at least) will optimize this away
321 /* MPEG-4 based DC predictor */
322 mpeg4_encode_dc(dc_pb, intra_dc, n);
326 bits_tab = uni_mpeg4_intra_rl_bits;
327 len_tab = uni_mpeg4_intra_rl_len;
332 bits_tab = uni_mpeg4_inter_rl_bits;
333 len_tab = uni_mpeg4_inter_rl_len;
337 last_non_zero = i - 1;
338 for (; i < last_index; i++) {
339 int level = block[scan_table[i]];
341 int run = i - last_non_zero - 1;
343 if ((level & (~127)) == 0) {
344 const int index = UNI_MPEG4_ENC_INDEX(0, run, level);
345 put_bits(ac_pb, len_tab[index], bits_tab[index]);
348 7 + 2 + 1 + 6 + 1 + 12 + 1,
349 (3 << 23) + (3 << 21) + (0 << 20) + (run << 14) +
350 (1 << 13) + (((level - 64) & 0xfff) << 1) + 1);
355 /* if (i <= last_index) */ {
356 int level = block[scan_table[i]];
357 int run = i - last_non_zero - 1;
359 if ((level & (~127)) == 0) {
360 const int index = UNI_MPEG4_ENC_INDEX(1, run, level);
361 put_bits(ac_pb, len_tab[index], bits_tab[index]);
364 7 + 2 + 1 + 6 + 1 + 12 + 1,
365 (3 << 23) + (3 << 21) + (1 << 20) + (run << 14) +
366 (1 << 13) + (((level - 64) & 0xfff) << 1) + 1);
371 static int mpeg4_get_block_length(MpegEncContext *s,
372 int16_t *block, int n,
373 int intra_dc, uint8_t *scan_table)
375 int i, last_non_zero;
377 const int last_index = s->block_last_index[n];
380 if (s->mb_intra) { // Note gcc (3.2.1 at least) will optimize this away
381 /* MPEG-4 based DC predictor */
382 len += mpeg4_get_dc_length(intra_dc, n);
386 len_tab = uni_mpeg4_intra_rl_len;
391 len_tab = uni_mpeg4_inter_rl_len;
395 last_non_zero = i - 1;
396 for (; i < last_index; i++) {
397 int level = block[scan_table[i]];
399 int run = i - last_non_zero - 1;
401 if ((level & (~127)) == 0) {
402 const int index = UNI_MPEG4_ENC_INDEX(0, run, level);
403 len += len_tab[index];
405 len += 7 + 2 + 1 + 6 + 1 + 12 + 1;
410 /* if (i <= last_index) */ {
411 int level = block[scan_table[i]];
412 int run = i - last_non_zero - 1;
414 if ((level & (~127)) == 0) {
415 const int index = UNI_MPEG4_ENC_INDEX(1, run, level);
416 len += len_tab[index];
418 len += 7 + 2 + 1 + 6 + 1 + 12 + 1;
425 static inline void mpeg4_encode_blocks(MpegEncContext *s, int16_t block[6][64],
426 int intra_dc[6], uint8_t **scan_table,
427 PutBitContext *dc_pb,
428 PutBitContext *ac_pb)
433 if (s->avctx->flags2 & AV_CODEC_FLAG2_NO_OUTPUT) {
434 for (i = 0; i < 6; i++)
435 skip_put_bits(&s->pb,
436 mpeg4_get_block_length(s, block[i], i,
437 intra_dc[i], scan_table[i]));
439 /* encode each block */
440 for (i = 0; i < 6; i++)
441 mpeg4_encode_block(s, block[i], i,
442 intra_dc[i], scan_table[i], dc_pb, ac_pb);
445 if (s->avctx->flags2 & AV_CODEC_FLAG2_NO_OUTPUT) {
446 for (i = 0; i < 6; i++)
447 skip_put_bits(&s->pb,
448 mpeg4_get_block_length(s, block[i], i, 0,
449 s->intra_scantable.permutated));
451 /* encode each block */
452 for (i = 0; i < 6; i++)
453 mpeg4_encode_block(s, block[i], i, 0,
454 s->intra_scantable.permutated, dc_pb, ac_pb);
459 static inline int get_b_cbp(MpegEncContext *s, int16_t block[6][64],
460 int motion_x, int motion_y, int mb_type)
464 if (s->mpv_flags & FF_MPV_FLAG_CBP_RD) {
466 const int lambda = s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
468 for (i = 0; i < 6; i++) {
469 if (s->coded_score[i] < 0) {
470 score += s->coded_score[i];
477 if ((motion_x | motion_y | s->dquant | mb_type) == 0)
478 zero_score -= 4; // 2 * MV + mb_type + cbp bit
480 zero_score *= lambda;
481 if (zero_score <= score)
485 for (i = 0; i < 6; i++) {
486 if (s->block_last_index[i] >= 0 && ((cbp >> (5 - i)) & 1) == 0) {
487 s->block_last_index[i] = -1;
488 s->bdsp.clear_block(s->block[i]);
492 for (i = 0; i < 6; i++) {
493 if (s->block_last_index[i] >= 0)
500 // FIXME this is duplicated to h263.c
501 static const int dquant_code[5] = { 1, 0, 9, 2, 3 };
503 void ff_mpeg4_encode_mb(MpegEncContext *s, int16_t block[6][64],
504 int motion_x, int motion_y)
506 int cbpc, cbpy, pred_x, pred_y;
507 PutBitContext *const pb2 = s->data_partitioning ? &s->pb2 : &s->pb;
508 PutBitContext *const tex_pb = s->data_partitioning && s->pict_type != AV_PICTURE_TYPE_B ? &s->tex_pb : &s->pb;
509 PutBitContext *const dc_pb = s->data_partitioning && s->pict_type != AV_PICTURE_TYPE_I ? &s->pb2 : &s->pb;
510 const int interleaved_stats = (s->avctx->flags & AV_CODEC_FLAG_PASS1) && !s->data_partitioning ? 1 : 0;
515 if (s->pict_type == AV_PICTURE_TYPE_B) {
516 /* convert from mv_dir to type */
517 static const int mb_type_table[8] = { -1, 3, 2, 1, -1, -1, -1, 0 };
518 int mb_type = mb_type_table[s->mv_dir];
521 for (i = 0; i < 2; i++)
522 s->last_mv[i][0][0] =
523 s->last_mv[i][0][1] =
524 s->last_mv[i][1][0] =
525 s->last_mv[i][1][1] = 0;
528 av_assert2(s->dquant >= -2 && s->dquant <= 2);
529 av_assert2((s->dquant & 1) == 0);
530 av_assert2(mb_type >= 0);
532 /* nothing to do if this MB was skipped in the next P-frame */
533 if (s->next_picture.mbskip_table[s->mb_y * s->mb_stride + s->mb_x]) { // FIXME avoid DCT & ...
539 s->mv_dir = MV_DIR_FORWARD; // doesn't matter
540 s->qscale -= s->dquant;
541 // s->mb_skipped = 1;
546 cbp = get_b_cbp(s, block, motion_x, motion_y, mb_type);
548 if ((cbp | motion_x | motion_y | mb_type) == 0) {
549 /* direct MB with MV={0,0} */
550 av_assert2(s->dquant == 0);
552 put_bits(&s->pb, 1, 1); /* mb not coded modb1=1 */
554 if (interleaved_stats) {
562 put_bits(&s->pb, 1, 0); /* mb coded modb1=0 */
563 put_bits(&s->pb, 1, cbp ? 0 : 1); /* modb2 */ // FIXME merge
564 put_bits(&s->pb, mb_type + 1, 1); // this table is so simple that we don't need it :)
566 put_bits(&s->pb, 6, cbp);
568 if (cbp && mb_type) {
570 put_bits(&s->pb, 2, (s->dquant >> 2) + 3);
572 put_bits(&s->pb, 1, 0);
574 s->qscale -= s->dquant;
576 if (!s->progressive_sequence) {
578 put_bits(&s->pb, 1, s->interlaced_dct);
579 if (mb_type) // not direct mode
580 put_bits(&s->pb, 1, s->mv_type == MV_TYPE_FIELD);
583 if (interleaved_stats)
584 s->misc_bits += get_bits_diff(s);
587 av_assert2(s->mv_dir & MV_DIRECT);
588 ff_h263_encode_motion_vector(s, motion_x, motion_y, 1);
592 av_assert2(mb_type > 0 && mb_type < 4);
593 if (s->mv_type != MV_TYPE_FIELD) {
594 if (s->mv_dir & MV_DIR_FORWARD) {
595 ff_h263_encode_motion_vector(s,
596 s->mv[0][0][0] - s->last_mv[0][0][0],
597 s->mv[0][0][1] - s->last_mv[0][0][1],
599 s->last_mv[0][0][0] =
600 s->last_mv[0][1][0] = s->mv[0][0][0];
601 s->last_mv[0][0][1] =
602 s->last_mv[0][1][1] = s->mv[0][0][1];
605 if (s->mv_dir & MV_DIR_BACKWARD) {
606 ff_h263_encode_motion_vector(s,
607 s->mv[1][0][0] - s->last_mv[1][0][0],
608 s->mv[1][0][1] - s->last_mv[1][0][1],
610 s->last_mv[1][0][0] =
611 s->last_mv[1][1][0] = s->mv[1][0][0];
612 s->last_mv[1][0][1] =
613 s->last_mv[1][1][1] = s->mv[1][0][1];
617 if (s->mv_dir & MV_DIR_FORWARD) {
618 put_bits(&s->pb, 1, s->field_select[0][0]);
619 put_bits(&s->pb, 1, s->field_select[0][1]);
621 if (s->mv_dir & MV_DIR_BACKWARD) {
622 put_bits(&s->pb, 1, s->field_select[1][0]);
623 put_bits(&s->pb, 1, s->field_select[1][1]);
625 if (s->mv_dir & MV_DIR_FORWARD) {
626 for (i = 0; i < 2; i++) {
627 ff_h263_encode_motion_vector(s,
628 s->mv[0][i][0] - s->last_mv[0][i][0],
629 s->mv[0][i][1] - s->last_mv[0][i][1] / 2,
631 s->last_mv[0][i][0] = s->mv[0][i][0];
632 s->last_mv[0][i][1] = s->mv[0][i][1] * 2;
636 if (s->mv_dir & MV_DIR_BACKWARD) {
637 for (i = 0; i < 2; i++) {
638 ff_h263_encode_motion_vector(s,
639 s->mv[1][i][0] - s->last_mv[1][i][0],
640 s->mv[1][i][1] - s->last_mv[1][i][1] / 2,
642 s->last_mv[1][i][0] = s->mv[1][i][0];
643 s->last_mv[1][i][1] = s->mv[1][i][1] * 2;
650 if (interleaved_stats)
651 s->mv_bits += get_bits_diff(s);
653 mpeg4_encode_blocks(s, block, NULL, NULL, NULL, &s->pb);
655 if (interleaved_stats)
656 s->p_tex_bits += get_bits_diff(s);
657 } else { /* s->pict_type==AV_PICTURE_TYPE_B */
658 cbp = get_p_cbp(s, block, motion_x, motion_y);
660 if ((cbp | motion_x | motion_y | s->dquant) == 0 &&
661 s->mv_type == MV_TYPE_16X16) {
662 /* Check if the B-frames can skip it too, as we must skip it
663 * if we skip here why didn't they just compress
664 * the skip-mb bits instead of reusing them ?! */
665 if (s->max_b_frames > 0) {
673 offset = x + y * s->linesize;
674 p_pic = s->new_picture.f->data[0] + offset;
677 for (i = 0; i < s->max_b_frames; i++) {
680 Picture *pic = s->reordered_input_picture[i + 1];
682 if (!pic || pic->f->pict_type != AV_PICTURE_TYPE_B)
685 b_pic = pic->f->data[0] + offset;
687 b_pic += INPLACE_OFFSET;
689 if (x + 16 > s->width || y + 16 > s->height) {
691 int xe = FFMIN(16, s->width - x);
692 int ye = FFMIN(16, s->height - y);
694 for (y1 = 0; y1 < ye; y1++) {
695 for (x1 = 0; x1 < xe; x1++) {
696 diff += FFABS(p_pic[x1 + y1 * s->linesize] - b_pic[x1 + y1 * s->linesize]);
699 diff = diff * 256 / (xe * ye);
701 diff = s->mecc.sad[0](NULL, p_pic, b_pic, s->linesize, 16);
703 if (diff > s->qscale * 70) { // FIXME check that 70 is optimal
711 if (s->mb_skipped == 1) {
712 /* skip macroblock */
713 put_bits(&s->pb, 1, 1);
715 if (interleaved_stats) {
725 put_bits(&s->pb, 1, 0); /* mb coded */
729 if (s->mv_type == MV_TYPE_16X16) {
733 ff_h263_inter_MCBPC_bits[cbpc],
734 ff_h263_inter_MCBPC_code[cbpc]);
736 put_bits(pb2, ff_h263_cbpy_tab[cbpy][1], ff_h263_cbpy_tab[cbpy][0]);
738 put_bits(pb2, 2, dquant_code[s->dquant + 2]);
740 if (!s->progressive_sequence) {
742 put_bits(pb2, 1, s->interlaced_dct);
746 if (interleaved_stats)
747 s->misc_bits += get_bits_diff(s);
749 /* motion vectors: 16x16 mode */
750 ff_h263_pred_motion(s, 0, 0, &pred_x, &pred_y);
752 ff_h263_encode_motion_vector(s,
756 } else if (s->mv_type == MV_TYPE_FIELD) {
760 ff_h263_inter_MCBPC_bits[cbpc],
761 ff_h263_inter_MCBPC_code[cbpc]);
763 put_bits(pb2, ff_h263_cbpy_tab[cbpy][1], ff_h263_cbpy_tab[cbpy][0]);
765 put_bits(pb2, 2, dquant_code[s->dquant + 2]);
767 av_assert2(!s->progressive_sequence);
769 put_bits(pb2, 1, s->interlaced_dct);
772 if (interleaved_stats)
773 s->misc_bits += get_bits_diff(s);
775 /* motion vectors: 16x8 interlaced mode */
776 ff_h263_pred_motion(s, 0, 0, &pred_x, &pred_y);
779 put_bits(&s->pb, 1, s->field_select[0][0]);
780 put_bits(&s->pb, 1, s->field_select[0][1]);
782 ff_h263_encode_motion_vector(s,
783 s->mv[0][0][0] - pred_x,
784 s->mv[0][0][1] - pred_y,
786 ff_h263_encode_motion_vector(s,
787 s->mv[0][1][0] - pred_x,
788 s->mv[0][1][1] - pred_y,
791 av_assert2(s->mv_type == MV_TYPE_8X8);
793 ff_h263_inter_MCBPC_bits[cbpc + 16],
794 ff_h263_inter_MCBPC_code[cbpc + 16]);
795 put_bits(pb2, ff_h263_cbpy_tab[cbpy][1], ff_h263_cbpy_tab[cbpy][0]);
797 if (!s->progressive_sequence && cbp)
798 put_bits(pb2, 1, s->interlaced_dct);
800 if (interleaved_stats)
801 s->misc_bits += get_bits_diff(s);
803 for (i = 0; i < 4; i++) {
804 /* motion vectors: 8x8 mode*/
805 ff_h263_pred_motion(s, i, 0, &pred_x, &pred_y);
807 ff_h263_encode_motion_vector(s,
808 s->current_picture.motion_val[0][s->block_index[i]][0] - pred_x,
809 s->current_picture.motion_val[0][s->block_index[i]][1] - pred_y,
814 if (interleaved_stats)
815 s->mv_bits += get_bits_diff(s);
817 mpeg4_encode_blocks(s, block, NULL, NULL, NULL, tex_pb);
819 if (interleaved_stats)
820 s->p_tex_bits += get_bits_diff(s);
826 int dc_diff[6]; // dc values with the dc prediction subtracted
827 int dir[6]; // prediction direction
828 int zigzag_last_index[6];
829 uint8_t *scan_table[6];
832 for (i = 0; i < 6; i++)
833 dc_diff[i] = ff_mpeg4_pred_dc(s, i, block[i][0], &dir[i], 1);
835 if (s->avctx->flags & AV_CODEC_FLAG_AC_PRED) {
836 s->ac_pred = decide_ac_pred(s, block, dir, scan_table, zigzag_last_index);
838 for (i = 0; i < 6; i++)
839 scan_table[i] = s->intra_scantable.permutated;
844 for (i = 0; i < 6; i++)
845 if (s->block_last_index[i] >= 1)
849 if (s->pict_type == AV_PICTURE_TYPE_I) {
853 ff_h263_intra_MCBPC_bits[cbpc],
854 ff_h263_intra_MCBPC_code[cbpc]);
858 put_bits(&s->pb, 1, 0); /* mb coded */
860 ff_h263_inter_MCBPC_bits[cbpc + 4],
861 ff_h263_inter_MCBPC_code[cbpc + 4]);
863 put_bits(pb2, 1, s->ac_pred);
865 put_bits(pb2, ff_h263_cbpy_tab[cbpy][1], ff_h263_cbpy_tab[cbpy][0]);
867 put_bits(dc_pb, 2, dquant_code[s->dquant + 2]);
869 if (!s->progressive_sequence)
870 put_bits(dc_pb, 1, s->interlaced_dct);
872 if (interleaved_stats)
873 s->misc_bits += get_bits_diff(s);
875 mpeg4_encode_blocks(s, block, dc_diff, scan_table, dc_pb, tex_pb);
877 if (interleaved_stats)
878 s->i_tex_bits += get_bits_diff(s);
881 /* restore ac coeffs & last_index stuff
882 * if we messed them up with the prediction */
884 restore_ac_coeffs(s, block, dir, scan_table, zigzag_last_index);
889 * add MPEG-4 stuffing bits (01...1)
891 void ff_mpeg4_stuffing(PutBitContext *pbc)
895 length = (-put_bits_count(pbc)) & 7;
897 put_bits(pbc, length, (1 << length) - 1);
900 /* must be called before writing the header */
901 void ff_set_mpeg4_time(MpegEncContext *s)
903 if (s->pict_type == AV_PICTURE_TYPE_B) {
904 ff_mpeg4_init_direct_mv(s);
906 s->last_time_base = s->time_base;
907 s->time_base = FFUDIV(s->time, s->avctx->time_base.den);
911 static void mpeg4_encode_gop_header(MpegEncContext *s)
913 int hours, minutes, seconds;
916 put_bits(&s->pb, 16, 0);
917 put_bits(&s->pb, 16, GOP_STARTCODE);
919 time = s->current_picture_ptr->f->pts;
920 if (s->reordered_input_picture[1])
921 time = FFMIN(time, s->reordered_input_picture[1]->f->pts);
922 time = time * s->avctx->time_base.num;
923 s->last_time_base = FFUDIV(time, s->avctx->time_base.den);
925 seconds = FFUDIV(time, s->avctx->time_base.den);
926 minutes = FFUDIV(seconds, 60); seconds = FFUMOD(seconds, 60);
927 hours = FFUDIV(minutes, 60); minutes = FFUMOD(minutes, 60);
928 hours = FFUMOD(hours , 24);
930 put_bits(&s->pb, 5, hours);
931 put_bits(&s->pb, 6, minutes);
932 put_bits(&s->pb, 1, 1);
933 put_bits(&s->pb, 6, seconds);
935 put_bits(&s->pb, 1, !!(s->avctx->flags & AV_CODEC_FLAG_CLOSED_GOP));
936 put_bits(&s->pb, 1, 0); // broken link == NO
938 ff_mpeg4_stuffing(&s->pb);
941 static void mpeg4_encode_visual_object_header(MpegEncContext *s)
943 int profile_and_level_indication;
946 if (s->avctx->profile != FF_PROFILE_UNKNOWN) {
947 profile_and_level_indication = s->avctx->profile << 4;
948 } else if (s->max_b_frames || s->quarter_sample) {
949 profile_and_level_indication = 0xF0; // adv simple
951 profile_and_level_indication = 0x00; // simple
954 if (s->avctx->level != FF_LEVEL_UNKNOWN)
955 profile_and_level_indication |= s->avctx->level;
957 profile_and_level_indication |= 1; // level 1
959 if (profile_and_level_indication >> 4 == 0xF)
966 put_bits(&s->pb, 16, 0);
967 put_bits(&s->pb, 16, VOS_STARTCODE);
969 put_bits(&s->pb, 8, profile_and_level_indication);
971 put_bits(&s->pb, 16, 0);
972 put_bits(&s->pb, 16, VISUAL_OBJ_STARTCODE);
974 put_bits(&s->pb, 1, 1);
975 put_bits(&s->pb, 4, vo_ver_id);
976 put_bits(&s->pb, 3, 1); // priority
978 put_bits(&s->pb, 4, 1); // visual obj type== video obj
980 put_bits(&s->pb, 1, 0); // video signal type == no clue // FIXME
982 ff_mpeg4_stuffing(&s->pb);
985 static void mpeg4_encode_vol_header(MpegEncContext *s,
991 if (!CONFIG_MPEG4_ENCODER)
994 if (s->max_b_frames || s->quarter_sample) {
996 s->vo_type = ADV_SIMPLE_VO_TYPE;
999 s->vo_type = SIMPLE_VO_TYPE;
1002 put_bits(&s->pb, 16, 0);
1003 put_bits(&s->pb, 16, 0x100 + vo_number); /* video obj */
1004 put_bits(&s->pb, 16, 0);
1005 put_bits(&s->pb, 16, 0x120 + vol_number); /* video obj layer */
1007 put_bits(&s->pb, 1, 0); /* random access vol */
1008 put_bits(&s->pb, 8, s->vo_type); /* video obj type indication */
1009 if (s->workaround_bugs & FF_BUG_MS) {
1010 put_bits(&s->pb, 1, 0); /* is obj layer id= no */
1012 put_bits(&s->pb, 1, 1); /* is obj layer id= yes */
1013 put_bits(&s->pb, 4, vo_ver_id); /* is obj layer ver id */
1014 put_bits(&s->pb, 3, 1); /* is obj layer priority */
1017 s->aspect_ratio_info = ff_h263_aspect_to_info(s->avctx->sample_aspect_ratio);
1019 put_bits(&s->pb, 4, s->aspect_ratio_info); /* aspect ratio info */
1020 if (s->aspect_ratio_info == FF_ASPECT_EXTENDED) {
1021 av_reduce(&s->avctx->sample_aspect_ratio.num, &s->avctx->sample_aspect_ratio.den,
1022 s->avctx->sample_aspect_ratio.num, s->avctx->sample_aspect_ratio.den, 255);
1023 put_bits(&s->pb, 8, s->avctx->sample_aspect_ratio.num);
1024 put_bits(&s->pb, 8, s->avctx->sample_aspect_ratio.den);
1027 if (s->workaround_bugs & FF_BUG_MS) {
1028 put_bits(&s->pb, 1, 0); /* vol control parameters= no @@@ */
1030 put_bits(&s->pb, 1, 1); /* vol control parameters= yes */
1031 put_bits(&s->pb, 2, 1); /* chroma format YUV 420/YV12 */
1032 put_bits(&s->pb, 1, s->low_delay);
1033 put_bits(&s->pb, 1, 0); /* vbv parameters= no */
1036 put_bits(&s->pb, 2, RECT_SHAPE); /* vol shape= rectangle */
1037 put_bits(&s->pb, 1, 1); /* marker bit */
1039 put_bits(&s->pb, 16, s->avctx->time_base.den);
1040 if (s->time_increment_bits < 1)
1041 s->time_increment_bits = 1;
1042 put_bits(&s->pb, 1, 1); /* marker bit */
1043 put_bits(&s->pb, 1, 0); /* fixed vop rate=no */
1044 put_bits(&s->pb, 1, 1); /* marker bit */
1045 put_bits(&s->pb, 13, s->width); /* vol width */
1046 put_bits(&s->pb, 1, 1); /* marker bit */
1047 put_bits(&s->pb, 13, s->height); /* vol height */
1048 put_bits(&s->pb, 1, 1); /* marker bit */
1049 put_bits(&s->pb, 1, s->progressive_sequence ? 0 : 1);
1050 put_bits(&s->pb, 1, 1); /* obmc disable */
1052 put_bits(&s->pb, 1, 0); /* sprite enable */
1054 put_bits(&s->pb, 2, 0); /* sprite enable */
1056 put_bits(&s->pb, 1, 0); /* not 8 bit == false */
1057 put_bits(&s->pb, 1, s->mpeg_quant); /* quant type = (0 = H.263 style) */
1059 if (s->mpeg_quant) {
1060 ff_write_quant_matrix(&s->pb, s->avctx->intra_matrix);
1061 ff_write_quant_matrix(&s->pb, s->avctx->inter_matrix);
1065 put_bits(&s->pb, 1, s->quarter_sample);
1066 put_bits(&s->pb, 1, 1); /* complexity estimation disable */
1067 put_bits(&s->pb, 1, s->rtp_mode ? 0 : 1); /* resync marker disable */
1068 put_bits(&s->pb, 1, s->data_partitioning ? 1 : 0);
1069 if (s->data_partitioning)
1070 put_bits(&s->pb, 1, 0); /* no rvlc */
1072 if (vo_ver_id != 1) {
1073 put_bits(&s->pb, 1, 0); /* newpred */
1074 put_bits(&s->pb, 1, 0); /* reduced res vop */
1076 put_bits(&s->pb, 1, 0); /* scalability */
1078 ff_mpeg4_stuffing(&s->pb);
1081 if (!(s->avctx->flags & AV_CODEC_FLAG_BITEXACT)) {
1082 put_bits(&s->pb, 16, 0);
1083 put_bits(&s->pb, 16, 0x1B2); /* user_data */
1084 avpriv_put_string(&s->pb, LIBAVCODEC_IDENT, 0);
1088 /* write MPEG-4 VOP header */
1089 int ff_mpeg4_encode_picture_header(MpegEncContext *s, int picture_number)
1092 int64_t time_div, time_mod;
1094 if (s->pict_type == AV_PICTURE_TYPE_I) {
1095 if (!(s->avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER)) {
1096 if (s->strict_std_compliance < FF_COMPLIANCE_VERY_STRICT) // HACK, the reference sw is buggy
1097 mpeg4_encode_visual_object_header(s);
1098 if (s->strict_std_compliance < FF_COMPLIANCE_VERY_STRICT || picture_number == 0) // HACK, the reference sw is buggy
1099 mpeg4_encode_vol_header(s, 0, 0);
1101 if (!(s->workaround_bugs & FF_BUG_MS))
1102 mpeg4_encode_gop_header(s);
1105 s->partitioned_frame = s->data_partitioning && s->pict_type != AV_PICTURE_TYPE_B;
1107 put_bits(&s->pb, 16, 0); /* vop header */
1108 put_bits(&s->pb, 16, VOP_STARTCODE); /* vop header */
1109 put_bits(&s->pb, 2, s->pict_type - 1); /* pict type: I = 0 , P = 1 */
1111 time_div = FFUDIV(s->time, s->avctx->time_base.den);
1112 time_mod = FFUMOD(s->time, s->avctx->time_base.den);
1113 time_incr = time_div - s->last_time_base;
1115 // This limits the frame duration to max 1 hour
1116 if (time_incr > 3600) {
1117 av_log(s->avctx, AV_LOG_ERROR, "time_incr %"PRIu64" too large\n", time_incr);
1118 return AVERROR(EINVAL);
1121 put_bits(&s->pb, 1, 1);
1123 put_bits(&s->pb, 1, 0);
1125 put_bits(&s->pb, 1, 1); /* marker */
1126 put_bits(&s->pb, s->time_increment_bits, time_mod); /* time increment */
1127 put_bits(&s->pb, 1, 1); /* marker */
1128 put_bits(&s->pb, 1, 1); /* vop coded */
1129 if (s->pict_type == AV_PICTURE_TYPE_P) {
1130 put_bits(&s->pb, 1, s->no_rounding); /* rounding type */
1132 put_bits(&s->pb, 3, 0); /* intra dc VLC threshold */
1133 if (!s->progressive_sequence) {
1134 put_bits(&s->pb, 1, s->current_picture_ptr->f->top_field_first);
1135 put_bits(&s->pb, 1, s->alternate_scan);
1137 // FIXME sprite stuff
1139 put_bits(&s->pb, 5, s->qscale);
1141 if (s->pict_type != AV_PICTURE_TYPE_I)
1142 put_bits(&s->pb, 3, s->f_code); /* fcode_for */
1143 if (s->pict_type == AV_PICTURE_TYPE_B)
1144 put_bits(&s->pb, 3, s->b_code); /* fcode_back */
1149 static av_cold void init_uni_dc_tab(void)
1151 int level, uni_code, uni_len;
1153 for (level = -256; level < 256; level++) {
1155 /* find number of bits */
1164 l = (-level) ^ ((1 << size) - 1);
1169 uni_code = ff_mpeg4_DCtab_lum[size][0];
1170 uni_len = ff_mpeg4_DCtab_lum[size][1];
1182 uni_DCtab_lum_bits[level + 256] = uni_code;
1183 uni_DCtab_lum_len[level + 256] = uni_len;
1186 uni_code = ff_mpeg4_DCtab_chrom[size][0];
1187 uni_len = ff_mpeg4_DCtab_chrom[size][1];
1199 uni_DCtab_chrom_bits[level + 256] = uni_code;
1200 uni_DCtab_chrom_len[level + 256] = uni_len;
1204 static av_cold void init_uni_mpeg4_rl_tab(RLTable *rl, uint32_t *bits_tab,
1207 int slevel, run, last;
1209 av_assert0(MAX_LEVEL >= 64);
1210 av_assert0(MAX_RUN >= 63);
1212 for (slevel = -64; slevel < 64; slevel++) {
1215 for (run = 0; run < 64; run++) {
1216 for (last = 0; last <= 1; last++) {
1217 const int index = UNI_MPEG4_ENC_INDEX(last, run, slevel + 64);
1218 int level = slevel < 0 ? -slevel : slevel;
1219 int sign = slevel < 0 ? 1 : 0;
1220 int bits, len, code;
1223 len_tab[index] = 100;
1226 code = get_rl_index(rl, last, run, level);
1227 bits = rl->table_vlc[code][0];
1228 len = rl->table_vlc[code][1];
1229 bits = bits * 2 + sign;
1232 if (code != rl->n && len < len_tab[index]) {
1233 bits_tab[index] = bits;
1234 len_tab[index] = len;
1237 bits = rl->table_vlc[rl->n][0];
1238 len = rl->table_vlc[rl->n][1];
1241 level1 = level - rl->max_level[last][run];
1243 code = get_rl_index(rl, last, run, level1);
1244 bits <<= rl->table_vlc[code][1];
1245 len += rl->table_vlc[code][1];
1246 bits += rl->table_vlc[code][0];
1247 bits = bits * 2 + sign;
1250 if (code != rl->n && len < len_tab[index]) {
1251 bits_tab[index] = bits;
1252 len_tab[index] = len;
1256 bits = rl->table_vlc[rl->n][0];
1257 len = rl->table_vlc[rl->n][1];
1258 bits = bits * 4 + 2;
1260 run1 = run - rl->max_run[last][level] - 1;
1262 code = get_rl_index(rl, last, run1, level);
1263 bits <<= rl->table_vlc[code][1];
1264 len += rl->table_vlc[code][1];
1265 bits += rl->table_vlc[code][0];
1266 bits = bits * 2 + sign;
1269 if (code != rl->n && len < len_tab[index]) {
1270 bits_tab[index] = bits;
1271 len_tab[index] = len;
1275 bits = rl->table_vlc[rl->n][0];
1276 len = rl->table_vlc[rl->n][1];
1277 bits = bits * 4 + 3;
1279 bits = bits * 2 + last;
1281 bits = bits * 64 + run;
1283 bits = bits * 2 + 1;
1285 bits = bits * 4096 + (slevel & 0xfff);
1287 bits = bits * 2 + 1;
1290 if (len < len_tab[index]) {
1291 bits_tab[index] = bits;
1292 len_tab[index] = len;
1299 static av_cold int encode_init(AVCodecContext *avctx)
1301 MpegEncContext *s = avctx->priv_data;
1303 static int done = 0;
1305 if (avctx->width >= (1<<13) || avctx->height >= (1<<13)) {
1306 av_log(avctx, AV_LOG_ERROR, "dimensions too large for MPEG-4\n");
1307 return AVERROR(EINVAL);
1310 if ((ret = ff_mpv_encode_init(avctx)) < 0)
1318 ff_rl_init(&ff_mpeg4_rl_intra, ff_mpeg4_static_rl_table_store[0]);
1320 init_uni_mpeg4_rl_tab(&ff_mpeg4_rl_intra, uni_mpeg4_intra_rl_bits, uni_mpeg4_intra_rl_len);
1321 init_uni_mpeg4_rl_tab(&ff_h263_rl_inter, uni_mpeg4_inter_rl_bits, uni_mpeg4_inter_rl_len);
1324 s->min_qcoeff = -2048;
1325 s->max_qcoeff = 2047;
1326 s->intra_ac_vlc_length = uni_mpeg4_intra_rl_len;
1327 s->intra_ac_vlc_last_length = uni_mpeg4_intra_rl_len + 128 * 64;
1328 s->inter_ac_vlc_length = uni_mpeg4_inter_rl_len;
1329 s->inter_ac_vlc_last_length = uni_mpeg4_inter_rl_len + 128 * 64;
1330 s->luma_dc_vlc_length = uni_DCtab_lum_len;
1331 s->ac_esc_length = 7 + 2 + 1 + 6 + 1 + 12 + 1;
1332 s->y_dc_scale_table = ff_mpeg4_y_dc_scale_table;
1333 s->c_dc_scale_table = ff_mpeg4_c_dc_scale_table;
1335 if (s->avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
1336 s->avctx->extradata = av_malloc(1024);
1337 init_put_bits(&s->pb, s->avctx->extradata, 1024);
1339 if (!(s->workaround_bugs & FF_BUG_MS))
1340 mpeg4_encode_visual_object_header(s);
1341 mpeg4_encode_vol_header(s, 0, 0);
1343 // ff_mpeg4_stuffing(&s->pb); ?
1344 flush_put_bits(&s->pb);
1345 s->avctx->extradata_size = (put_bits_count(&s->pb) + 7) >> 3;
1350 void ff_mpeg4_init_partitions(MpegEncContext *s)
1352 uint8_t *start = put_bits_ptr(&s->pb);
1353 uint8_t *end = s->pb.buf_end;
1354 int size = end - start;
1355 int pb_size = (((intptr_t)start + size / 3) & (~3)) - (intptr_t)start;
1356 int tex_size = (size - 2 * pb_size) & (~3);
1358 set_put_bits_buffer_size(&s->pb, pb_size);
1359 init_put_bits(&s->tex_pb, start + pb_size, tex_size);
1360 init_put_bits(&s->pb2, start + pb_size + tex_size, pb_size);
1363 void ff_mpeg4_merge_partitions(MpegEncContext *s)
1365 const int pb2_len = put_bits_count(&s->pb2);
1366 const int tex_pb_len = put_bits_count(&s->tex_pb);
1367 const int bits = put_bits_count(&s->pb);
1369 if (s->pict_type == AV_PICTURE_TYPE_I) {
1370 put_bits(&s->pb, 19, DC_MARKER);
1371 s->misc_bits += 19 + pb2_len + bits - s->last_bits;
1372 s->i_tex_bits += tex_pb_len;
1374 put_bits(&s->pb, 17, MOTION_MARKER);
1375 s->misc_bits += 17 + pb2_len;
1376 s->mv_bits += bits - s->last_bits;
1377 s->p_tex_bits += tex_pb_len;
1380 flush_put_bits(&s->pb2);
1381 flush_put_bits(&s->tex_pb);
1383 set_put_bits_buffer_size(&s->pb, s->pb2.buf_end - s->pb.buf);
1384 avpriv_copy_bits(&s->pb, s->pb2.buf, pb2_len);
1385 avpriv_copy_bits(&s->pb, s->tex_pb.buf, tex_pb_len);
1386 s->last_bits = put_bits_count(&s->pb);
1389 void ff_mpeg4_encode_video_packet_header(MpegEncContext *s)
1391 int mb_num_bits = av_log2(s->mb_num - 1) + 1;
1393 put_bits(&s->pb, ff_mpeg4_get_video_packet_prefix_length(s), 0);
1394 put_bits(&s->pb, 1, 1);
1396 put_bits(&s->pb, mb_num_bits, s->mb_x + s->mb_y * s->mb_width);
1397 put_bits(&s->pb, s->quant_precision, s->qscale);
1398 put_bits(&s->pb, 1, 0); /* no HEC */
1401 #define OFFSET(x) offsetof(MpegEncContext, x)
1402 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
1403 static const AVOption options[] = {
1404 { "data_partitioning", "Use data partitioning.", OFFSET(data_partitioning), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
1405 { "alternate_scan", "Enable alternate scantable.", OFFSET(alternate_scan), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
1410 static const AVClass mpeg4enc_class = {
1411 .class_name = "MPEG4 encoder",
1412 .item_name = av_default_item_name,
1414 .version = LIBAVUTIL_VERSION_INT,
1417 AVCodec ff_mpeg4_encoder = {
1419 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 part 2"),
1420 .type = AVMEDIA_TYPE_VIDEO,
1421 .id = AV_CODEC_ID_MPEG4,
1422 .priv_data_size = sizeof(MpegEncContext),
1423 .init = encode_init,
1424 .encode2 = ff_mpv_encode_picture,
1425 .close = ff_mpv_encode_end,
1426 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
1427 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS,
1428 .priv_class = &mpeg4enc_class,