2 * The simplest mpeg encoder (well, it was the simplest!)
3 * Copyright (c) 2000,2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
8 * This file is part of FFmpeg.
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 * The simplest mpeg encoder (well, it was the simplest!).
30 #include "libavutil/attributes.h"
31 #include "libavutil/avassert.h"
32 #include "libavutil/imgutils.h"
33 #include "libavutil/internal.h"
34 #include "libavutil/timer.h"
37 #include "h264chroma.h"
40 #include "mpegutils.h"
41 #include "mpegvideo.h"
47 static const uint8_t ff_default_chroma_qscale_table[32] = {
48 // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
49 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
50 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
53 const uint8_t ff_mpeg1_dc_scale_table[128] = {
54 // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
55 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
56 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
57 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
58 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
59 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
60 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
61 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
62 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
65 static const uint8_t mpeg2_dc_scale_table1[128] = {
66 // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
67 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
68 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
69 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
70 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
71 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
72 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
73 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
74 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
77 static const uint8_t mpeg2_dc_scale_table2[128] = {
78 // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
79 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
80 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
81 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
82 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
83 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
84 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
85 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
86 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
89 static const uint8_t mpeg2_dc_scale_table3[128] = {
90 // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
91 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
92 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
93 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
94 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
95 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
96 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
97 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
98 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
101 const uint8_t *const ff_mpeg2_dc_scale_table[4] = {
102 ff_mpeg1_dc_scale_table,
103 mpeg2_dc_scale_table1,
104 mpeg2_dc_scale_table2,
105 mpeg2_dc_scale_table3,
108 const enum AVPixelFormat ff_pixfmt_list_420[] = {
113 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
114 int16_t *block, int n, int qscale)
116 int i, level, nCoeffs;
117 const uint16_t *quant_matrix;
119 nCoeffs= s->block_last_index[n];
121 block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
122 /* XXX: only mpeg1 */
123 quant_matrix = s->intra_matrix;
124 for(i=1;i<=nCoeffs;i++) {
125 int j= s->intra_scantable.permutated[i];
130 level = (int)(level * qscale * quant_matrix[j]) >> 3;
131 level = (level - 1) | 1;
134 level = (int)(level * qscale * quant_matrix[j]) >> 3;
135 level = (level - 1) | 1;
142 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
143 int16_t *block, int n, int qscale)
145 int i, level, nCoeffs;
146 const uint16_t *quant_matrix;
148 nCoeffs= s->block_last_index[n];
150 quant_matrix = s->inter_matrix;
151 for(i=0; i<=nCoeffs; i++) {
152 int j= s->intra_scantable.permutated[i];
157 level = (((level << 1) + 1) * qscale *
158 ((int) (quant_matrix[j]))) >> 4;
159 level = (level - 1) | 1;
162 level = (((level << 1) + 1) * qscale *
163 ((int) (quant_matrix[j]))) >> 4;
164 level = (level - 1) | 1;
171 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
172 int16_t *block, int n, int qscale)
174 int i, level, nCoeffs;
175 const uint16_t *quant_matrix;
177 if(s->alternate_scan) nCoeffs= 63;
178 else nCoeffs= s->block_last_index[n];
180 block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
181 quant_matrix = s->intra_matrix;
182 for(i=1;i<=nCoeffs;i++) {
183 int j= s->intra_scantable.permutated[i];
188 level = (int)(level * qscale * quant_matrix[j]) >> 3;
191 level = (int)(level * qscale * quant_matrix[j]) >> 3;
198 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
199 int16_t *block, int n, int qscale)
201 int i, level, nCoeffs;
202 const uint16_t *quant_matrix;
205 if(s->alternate_scan) nCoeffs= 63;
206 else nCoeffs= s->block_last_index[n];
208 block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
210 quant_matrix = s->intra_matrix;
211 for(i=1;i<=nCoeffs;i++) {
212 int j= s->intra_scantable.permutated[i];
217 level = (int)(level * qscale * quant_matrix[j]) >> 3;
220 level = (int)(level * qscale * quant_matrix[j]) >> 3;
229 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
230 int16_t *block, int n, int qscale)
232 int i, level, nCoeffs;
233 const uint16_t *quant_matrix;
236 if(s->alternate_scan) nCoeffs= 63;
237 else nCoeffs= s->block_last_index[n];
239 quant_matrix = s->inter_matrix;
240 for(i=0; i<=nCoeffs; i++) {
241 int j= s->intra_scantable.permutated[i];
246 level = (((level << 1) + 1) * qscale *
247 ((int) (quant_matrix[j]))) >> 4;
250 level = (((level << 1) + 1) * qscale *
251 ((int) (quant_matrix[j]))) >> 4;
260 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
261 int16_t *block, int n, int qscale)
263 int i, level, qmul, qadd;
266 av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
271 block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
272 qadd = (qscale - 1) | 1;
279 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
281 for(i=1; i<=nCoeffs; i++) {
285 level = level * qmul - qadd;
287 level = level * qmul + qadd;
294 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
295 int16_t *block, int n, int qscale)
297 int i, level, qmul, qadd;
300 av_assert2(s->block_last_index[n]>=0);
302 qadd = (qscale - 1) | 1;
305 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
307 for(i=0; i<=nCoeffs; i++) {
311 level = level * qmul - qadd;
313 level = level * qmul + qadd;
320 static void mpeg_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
322 int mb_x, int mb_y, int mb_intra, int mb_skipped)
324 MpegEncContext *s = opaque;
327 s->mv_type = mv_type;
328 s->mb_intra = mb_intra;
329 s->mb_skipped = mb_skipped;
332 memcpy(s->mv, mv, sizeof(*mv));
334 ff_init_block_index(s);
335 ff_update_block_index(s);
337 s->dsp.clear_blocks(s->block[0]);
339 s->dest[0] = s->current_picture.f.data[0] + (s->mb_y * 16 * s->linesize) + s->mb_x * 16;
340 s->dest[1] = s->current_picture.f.data[1] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift);
341 s->dest[2] = s->current_picture.f.data[2] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift);
344 av_log(s->avctx, AV_LOG_DEBUG, "Interlaced error concealment is not fully implemented\n");
345 ff_MPV_decode_mb(s, s->block);
348 /* init common dct for both encoder and decoder */
349 av_cold int ff_dct_common_init(MpegEncContext *s)
351 ff_dsputil_init(&s->dsp, s->avctx);
352 ff_h264chroma_init(&s->h264chroma, 8); //for lowres
353 ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
354 ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
356 s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
357 s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
358 s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
359 s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
360 s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
361 if (s->flags & CODEC_FLAG_BITEXACT)
362 s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
363 s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
366 ff_MPV_common_init_axp(s);
368 ff_MPV_common_init_arm(s);
370 ff_MPV_common_init_ppc(s);
372 ff_MPV_common_init_x86(s);
374 /* load & permutate scantables
375 * note: only wmv uses different ones
377 if (s->alternate_scan) {
378 ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable , ff_alternate_vertical_scan);
379 ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable , ff_alternate_vertical_scan);
381 ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable , ff_zigzag_direct);
382 ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable , ff_zigzag_direct);
384 ff_init_scantable(s->dsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
385 ff_init_scantable(s->dsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
390 static int frame_size_alloc(MpegEncContext *s, int linesize)
392 int alloc_size = FFALIGN(FFABS(linesize) + 64, 32);
394 // edge emu needs blocksize + filter length - 1
395 // (= 17x17 for halfpel / 21x21 for h264)
396 // VC1 computes luma and chroma simultaneously and needs 19X19 + 9x9
397 // at uvlinesize. It supports only YUV420 so 24x24 is enough
398 // linesize * interlaced * MBsize
399 FF_ALLOCZ_OR_GOTO(s->avctx, s->edge_emu_buffer, alloc_size * 4 * 24,
402 FF_ALLOCZ_OR_GOTO(s->avctx, s->me.scratchpad, alloc_size * 4 * 16 * 2,
404 s->me.temp = s->me.scratchpad;
405 s->rd_scratchpad = s->me.scratchpad;
406 s->b_scratchpad = s->me.scratchpad;
407 s->obmc_scratchpad = s->me.scratchpad + 16;
411 av_freep(&s->edge_emu_buffer);
412 return AVERROR(ENOMEM);
416 * Allocate a frame buffer
418 static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
420 int edges_needed = av_codec_is_encoder(s->avctx->codec);
424 if (s->codec_id != AV_CODEC_ID_WMV3IMAGE &&
425 s->codec_id != AV_CODEC_ID_VC1IMAGE &&
426 s->codec_id != AV_CODEC_ID_MSS2) {
428 pic->f.width = s->avctx->width + 2 * EDGE_WIDTH;
429 pic->f.height = s->avctx->height + 2 * EDGE_WIDTH;
432 r = ff_thread_get_buffer(s->avctx, &pic->tf,
433 pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);
435 pic->f.width = s->avctx->width;
436 pic->f.height = s->avctx->height;
437 pic->f.format = s->avctx->pix_fmt;
438 r = avcodec_default_get_buffer2(s->avctx, &pic->f, 0);
441 if (r < 0 || !pic->f.buf[0]) {
442 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %p)\n",
449 for (i = 0; pic->f.data[i]; i++) {
450 int offset = (EDGE_WIDTH >> (i ? s->chroma_y_shift : 0)) *
452 (EDGE_WIDTH >> (i ? s->chroma_x_shift : 0));
453 pic->f.data[i] += offset;
455 pic->f.width = s->avctx->width;
456 pic->f.height = s->avctx->height;
459 if (s->avctx->hwaccel) {
460 assert(!pic->hwaccel_picture_private);
461 if (s->avctx->hwaccel->priv_data_size) {
462 pic->hwaccel_priv_buf = av_buffer_allocz(s->avctx->hwaccel->priv_data_size);
463 if (!pic->hwaccel_priv_buf) {
464 av_log(s->avctx, AV_LOG_ERROR, "alloc_frame_buffer() failed (hwaccel private data allocation)\n");
467 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
471 if (s->linesize && (s->linesize != pic->f.linesize[0] ||
472 s->uvlinesize != pic->f.linesize[1])) {
473 av_log(s->avctx, AV_LOG_ERROR,
474 "get_buffer() failed (stride changed)\n");
475 ff_mpeg_unref_picture(s, pic);
479 if (pic->f.linesize[1] != pic->f.linesize[2]) {
480 av_log(s->avctx, AV_LOG_ERROR,
481 "get_buffer() failed (uv stride mismatch)\n");
482 ff_mpeg_unref_picture(s, pic);
486 if (!s->edge_emu_buffer &&
487 (ret = frame_size_alloc(s, pic->f.linesize[0])) < 0) {
488 av_log(s->avctx, AV_LOG_ERROR,
489 "get_buffer() failed to allocate context scratch buffers.\n");
490 ff_mpeg_unref_picture(s, pic);
497 void ff_free_picture_tables(Picture *pic)
501 pic->alloc_mb_width =
502 pic->alloc_mb_height = 0;
504 av_buffer_unref(&pic->mb_var_buf);
505 av_buffer_unref(&pic->mc_mb_var_buf);
506 av_buffer_unref(&pic->mb_mean_buf);
507 av_buffer_unref(&pic->mbskip_table_buf);
508 av_buffer_unref(&pic->qscale_table_buf);
509 av_buffer_unref(&pic->mb_type_buf);
511 for (i = 0; i < 2; i++) {
512 av_buffer_unref(&pic->motion_val_buf[i]);
513 av_buffer_unref(&pic->ref_index_buf[i]);
517 static int alloc_picture_tables(MpegEncContext *s, Picture *pic)
519 const int big_mb_num = s->mb_stride * (s->mb_height + 1) + 1;
520 const int mb_array_size = s->mb_stride * s->mb_height;
521 const int b8_array_size = s->b8_stride * s->mb_height * 2;
525 pic->mbskip_table_buf = av_buffer_allocz(mb_array_size + 2);
526 pic->qscale_table_buf = av_buffer_allocz(big_mb_num + s->mb_stride);
527 pic->mb_type_buf = av_buffer_allocz((big_mb_num + s->mb_stride) *
529 if (!pic->mbskip_table_buf || !pic->qscale_table_buf || !pic->mb_type_buf)
530 return AVERROR(ENOMEM);
533 pic->mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t));
534 pic->mc_mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t));
535 pic->mb_mean_buf = av_buffer_allocz(mb_array_size);
536 if (!pic->mb_var_buf || !pic->mc_mb_var_buf || !pic->mb_mean_buf)
537 return AVERROR(ENOMEM);
540 if (s->out_format == FMT_H263 || s->encoding || s->avctx->debug_mv) {
541 int mv_size = 2 * (b8_array_size + 4) * sizeof(int16_t);
542 int ref_index_size = 4 * mb_array_size;
544 for (i = 0; mv_size && i < 2; i++) {
545 pic->motion_val_buf[i] = av_buffer_allocz(mv_size);
546 pic->ref_index_buf[i] = av_buffer_allocz(ref_index_size);
547 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
548 return AVERROR(ENOMEM);
552 pic->alloc_mb_width = s->mb_width;
553 pic->alloc_mb_height = s->mb_height;
558 static int make_tables_writable(Picture *pic)
561 #define MAKE_WRITABLE(table) \
564 (ret = av_buffer_make_writable(&pic->table)) < 0)\
568 MAKE_WRITABLE(mb_var_buf);
569 MAKE_WRITABLE(mc_mb_var_buf);
570 MAKE_WRITABLE(mb_mean_buf);
571 MAKE_WRITABLE(mbskip_table_buf);
572 MAKE_WRITABLE(qscale_table_buf);
573 MAKE_WRITABLE(mb_type_buf);
575 for (i = 0; i < 2; i++) {
576 MAKE_WRITABLE(motion_val_buf[i]);
577 MAKE_WRITABLE(ref_index_buf[i]);
584 * Allocate a Picture.
585 * The pixels are allocated/set by calling get_buffer() if shared = 0
587 int ff_alloc_picture(MpegEncContext *s, Picture *pic, int shared)
591 if (pic->qscale_table_buf)
592 if ( pic->alloc_mb_width != s->mb_width
593 || pic->alloc_mb_height != s->mb_height)
594 ff_free_picture_tables(pic);
597 av_assert0(pic->f.data[0]);
600 av_assert0(!pic->f.buf[0]);
602 if (alloc_frame_buffer(s, pic) < 0)
605 s->linesize = pic->f.linesize[0];
606 s->uvlinesize = pic->f.linesize[1];
609 if (!pic->qscale_table_buf)
610 ret = alloc_picture_tables(s, pic);
612 ret = make_tables_writable(pic);
617 pic->mb_var = (uint16_t*)pic->mb_var_buf->data;
618 pic->mc_mb_var = (uint16_t*)pic->mc_mb_var_buf->data;
619 pic->mb_mean = pic->mb_mean_buf->data;
622 pic->mbskip_table = pic->mbskip_table_buf->data;
623 pic->qscale_table = pic->qscale_table_buf->data + 2 * s->mb_stride + 1;
624 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * s->mb_stride + 1;
626 if (pic->motion_val_buf[0]) {
627 for (i = 0; i < 2; i++) {
628 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
629 pic->ref_index[i] = pic->ref_index_buf[i]->data;
635 av_log(s->avctx, AV_LOG_ERROR, "Error allocating a picture.\n");
636 ff_mpeg_unref_picture(s, pic);
637 ff_free_picture_tables(pic);
638 return AVERROR(ENOMEM);
642 * Deallocate a picture.
644 void ff_mpeg_unref_picture(MpegEncContext *s, Picture *pic)
646 int off = offsetof(Picture, mb_mean) + sizeof(pic->mb_mean);
649 /* WM Image / Screen codecs allocate internal buffers with different
650 * dimensions / colorspaces; ignore user-defined callbacks for these. */
651 if (s->codec_id != AV_CODEC_ID_WMV3IMAGE &&
652 s->codec_id != AV_CODEC_ID_VC1IMAGE &&
653 s->codec_id != AV_CODEC_ID_MSS2)
654 ff_thread_release_buffer(s->avctx, &pic->tf);
656 av_frame_unref(&pic->f);
658 av_buffer_unref(&pic->hwaccel_priv_buf);
660 if (pic->needs_realloc)
661 ff_free_picture_tables(pic);
663 memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
666 static int update_picture_tables(Picture *dst, Picture *src)
670 #define UPDATE_TABLE(table)\
673 (!dst->table || dst->table->buffer != src->table->buffer)) {\
674 av_buffer_unref(&dst->table);\
675 dst->table = av_buffer_ref(src->table);\
677 ff_free_picture_tables(dst);\
678 return AVERROR(ENOMEM);\
683 UPDATE_TABLE(mb_var_buf);
684 UPDATE_TABLE(mc_mb_var_buf);
685 UPDATE_TABLE(mb_mean_buf);
686 UPDATE_TABLE(mbskip_table_buf);
687 UPDATE_TABLE(qscale_table_buf);
688 UPDATE_TABLE(mb_type_buf);
689 for (i = 0; i < 2; i++) {
690 UPDATE_TABLE(motion_val_buf[i]);
691 UPDATE_TABLE(ref_index_buf[i]);
694 dst->mb_var = src->mb_var;
695 dst->mc_mb_var = src->mc_mb_var;
696 dst->mb_mean = src->mb_mean;
697 dst->mbskip_table = src->mbskip_table;
698 dst->qscale_table = src->qscale_table;
699 dst->mb_type = src->mb_type;
700 for (i = 0; i < 2; i++) {
701 dst->motion_val[i] = src->motion_val[i];
702 dst->ref_index[i] = src->ref_index[i];
705 dst->alloc_mb_width = src->alloc_mb_width;
706 dst->alloc_mb_height = src->alloc_mb_height;
711 int ff_mpeg_ref_picture(MpegEncContext *s, Picture *dst, Picture *src)
715 av_assert0(!dst->f.buf[0]);
716 av_assert0(src->f.buf[0]);
720 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
724 ret = update_picture_tables(dst, src);
728 if (src->hwaccel_picture_private) {
729 dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
730 if (!dst->hwaccel_priv_buf)
732 dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
735 dst->field_picture = src->field_picture;
736 dst->mb_var_sum = src->mb_var_sum;
737 dst->mc_mb_var_sum = src->mc_mb_var_sum;
738 dst->b_frame_score = src->b_frame_score;
739 dst->needs_realloc = src->needs_realloc;
740 dst->reference = src->reference;
741 dst->shared = src->shared;
745 ff_mpeg_unref_picture(s, dst);
749 static void exchange_uv(MpegEncContext *s)
754 s->pblocks[4] = s->pblocks[5];
758 static int init_duplicate_context(MpegEncContext *s)
760 int y_size = s->b8_stride * (2 * s->mb_height + 1);
761 int c_size = s->mb_stride * (s->mb_height + 1);
762 int yc_size = y_size + 2 * c_size;
765 if (s->mb_height & 1)
766 yc_size += 2*s->b8_stride + 2*s->mb_stride;
773 s->obmc_scratchpad = NULL;
776 FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map,
777 ME_MAP_SIZE * sizeof(uint32_t), fail)
778 FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map,
779 ME_MAP_SIZE * sizeof(uint32_t), fail)
780 if (s->avctx->noise_reduction) {
781 FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum,
782 2 * 64 * sizeof(int), fail)
785 FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail)
786 s->block = s->blocks[0];
788 for (i = 0; i < 12; i++) {
789 s->pblocks[i] = &s->block[i];
791 if (s->avctx->codec_tag == AV_RL32("VCR2"))
794 if (s->out_format == FMT_H263) {
796 FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base,
797 yc_size * sizeof(int16_t) * 16, fail);
798 s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
799 s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
800 s->ac_val[2] = s->ac_val[1] + c_size;
805 return -1; // free() through ff_MPV_common_end()
808 static void free_duplicate_context(MpegEncContext *s)
813 av_freep(&s->edge_emu_buffer);
814 av_freep(&s->me.scratchpad);
818 s->obmc_scratchpad = NULL;
820 av_freep(&s->dct_error_sum);
821 av_freep(&s->me.map);
822 av_freep(&s->me.score_map);
823 av_freep(&s->blocks);
824 av_freep(&s->ac_val_base);
828 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
830 #define COPY(a) bak->a = src->a
831 COPY(edge_emu_buffer);
836 COPY(obmc_scratchpad);
843 COPY(me.map_generation);
855 int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
859 // FIXME copy only needed parts
861 backup_duplicate_context(&bak, dst);
862 memcpy(dst, src, sizeof(MpegEncContext));
863 backup_duplicate_context(dst, &bak);
864 for (i = 0; i < 12; i++) {
865 dst->pblocks[i] = &dst->block[i];
867 if (dst->avctx->codec_tag == AV_RL32("VCR2"))
869 if (!dst->edge_emu_buffer &&
870 (ret = frame_size_alloc(dst, dst->linesize)) < 0) {
871 av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context "
872 "scratch buffers.\n");
875 // STOP_TIMER("update_duplicate_context")
876 // about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads
880 int ff_mpeg_update_thread_context(AVCodecContext *dst,
881 const AVCodecContext *src)
884 MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
891 // FIXME can parameters change on I-frames?
892 // in that case dst may need a reinit
893 if (!s->context_initialized) {
894 memcpy(s, s1, sizeof(MpegEncContext));
897 s->bitstream_buffer = NULL;
898 s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
900 if (s1->context_initialized){
901 // s->picture_range_start += MAX_PICTURE_COUNT;
902 // s->picture_range_end += MAX_PICTURE_COUNT;
903 if((ret = ff_MPV_common_init(s)) < 0){
904 memset(s, 0, sizeof(MpegEncContext));
911 if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
912 s->context_reinit = 0;
913 s->height = s1->height;
914 s->width = s1->width;
915 if ((ret = ff_MPV_common_frame_size_change(s)) < 0)
919 s->avctx->coded_height = s1->avctx->coded_height;
920 s->avctx->coded_width = s1->avctx->coded_width;
921 s->avctx->width = s1->avctx->width;
922 s->avctx->height = s1->avctx->height;
924 s->coded_picture_number = s1->coded_picture_number;
925 s->picture_number = s1->picture_number;
927 av_assert0(!s->picture || s->picture != s1->picture);
929 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
930 ff_mpeg_unref_picture(s, &s->picture[i]);
931 if (s1->picture[i].f.buf[0] &&
932 (ret = ff_mpeg_ref_picture(s, &s->picture[i], &s1->picture[i])) < 0)
936 #define UPDATE_PICTURE(pic)\
938 ff_mpeg_unref_picture(s, &s->pic);\
939 if (s1->pic.f.buf[0])\
940 ret = ff_mpeg_ref_picture(s, &s->pic, &s1->pic);\
942 ret = update_picture_tables(&s->pic, &s1->pic);\
947 UPDATE_PICTURE(current_picture);
948 UPDATE_PICTURE(last_picture);
949 UPDATE_PICTURE(next_picture);
951 s->last_picture_ptr = REBASE_PICTURE(s1->last_picture_ptr, s, s1);
952 s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
953 s->next_picture_ptr = REBASE_PICTURE(s1->next_picture_ptr, s, s1);
955 // Error/bug resilience
956 s->next_p_frame_damaged = s1->next_p_frame_damaged;
957 s->workaround_bugs = s1->workaround_bugs;
958 s->padding_bug_score = s1->padding_bug_score;
961 memcpy(&s->last_time_base, &s1->last_time_base,
962 (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) -
963 (char *) &s1->last_time_base);
966 s->max_b_frames = s1->max_b_frames;
967 s->low_delay = s1->low_delay;
968 s->droppable = s1->droppable;
970 // DivX handling (doesn't work)
971 s->divx_packed = s1->divx_packed;
973 if (s1->bitstream_buffer) {
974 if (s1->bitstream_buffer_size +
975 FF_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size)
976 av_fast_malloc(&s->bitstream_buffer,
977 &s->allocated_bitstream_buffer_size,
978 s1->allocated_bitstream_buffer_size);
979 s->bitstream_buffer_size = s1->bitstream_buffer_size;
980 memcpy(s->bitstream_buffer, s1->bitstream_buffer,
981 s1->bitstream_buffer_size);
982 memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
983 FF_INPUT_BUFFER_PADDING_SIZE);
986 // linesize dependend scratch buffer allocation
987 if (!s->edge_emu_buffer)
989 if (frame_size_alloc(s, s1->linesize) < 0) {
990 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
991 "scratch buffers.\n");
992 return AVERROR(ENOMEM);
995 av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
996 "be allocated due to unknown size.\n");
999 // MPEG2/interlacing info
1000 memcpy(&s->progressive_sequence, &s1->progressive_sequence,
1001 (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
1003 if (!s1->first_field) {
1004 s->last_pict_type = s1->pict_type;
1005 if (s1->current_picture_ptr)
1006 s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f.quality;
1013 * Set the given MpegEncContext to common defaults
1014 * (same for encoding and decoding).
1015 * The changed fields will not depend upon the
1016 * prior state of the MpegEncContext.
1018 void ff_MPV_common_defaults(MpegEncContext *s)
1020 s->y_dc_scale_table =
1021 s->c_dc_scale_table = ff_mpeg1_dc_scale_table;
1022 s->chroma_qscale_table = ff_default_chroma_qscale_table;
1023 s->progressive_frame = 1;
1024 s->progressive_sequence = 1;
1025 s->picture_structure = PICT_FRAME;
1027 s->coded_picture_number = 0;
1028 s->picture_number = 0;
1033 s->slice_context_count = 1;
1037 * Set the given MpegEncContext to defaults for decoding.
1038 * the changed fields will not depend upon
1039 * the prior state of the MpegEncContext.
1041 void ff_MPV_decode_defaults(MpegEncContext *s)
1043 ff_MPV_common_defaults(s);
1046 static int init_er(MpegEncContext *s)
1048 ERContext *er = &s->er;
1049 int mb_array_size = s->mb_height * s->mb_stride;
1052 er->avctx = s->avctx;
1055 er->mb_index2xy = s->mb_index2xy;
1056 er->mb_num = s->mb_num;
1057 er->mb_width = s->mb_width;
1058 er->mb_height = s->mb_height;
1059 er->mb_stride = s->mb_stride;
1060 er->b8_stride = s->b8_stride;
1062 er->er_temp_buffer = av_malloc(s->mb_height * s->mb_stride);
1063 er->error_status_table = av_mallocz(mb_array_size);
1064 if (!er->er_temp_buffer || !er->error_status_table)
1067 er->mbskip_table = s->mbskip_table;
1068 er->mbintra_table = s->mbintra_table;
1070 for (i = 0; i < FF_ARRAY_ELEMS(s->dc_val); i++)
1071 er->dc_val[i] = s->dc_val[i];
1073 er->decode_mb = mpeg_er_decode_mb;
1078 av_freep(&er->er_temp_buffer);
1079 av_freep(&er->error_status_table);
1080 return AVERROR(ENOMEM);
1084 * Initialize and allocates MpegEncContext fields dependent on the resolution.
1086 static int init_context_frame(MpegEncContext *s)
1088 int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
1090 s->mb_width = (s->width + 15) / 16;
1091 s->mb_stride = s->mb_width + 1;
1092 s->b8_stride = s->mb_width * 2 + 1;
1093 s->b4_stride = s->mb_width * 4 + 1;
1094 mb_array_size = s->mb_height * s->mb_stride;
1095 mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
1097 /* set default edge pos, will be overriden
1098 * in decode_header if needed */
1099 s->h_edge_pos = s->mb_width * 16;
1100 s->v_edge_pos = s->mb_height * 16;
1102 s->mb_num = s->mb_width * s->mb_height;
1107 s->block_wrap[3] = s->b8_stride;
1109 s->block_wrap[5] = s->mb_stride;
1111 y_size = s->b8_stride * (2 * s->mb_height + 1);
1112 c_size = s->mb_stride * (s->mb_height + 1);
1113 yc_size = y_size + 2 * c_size;
1115 if (s->mb_height & 1)
1116 yc_size += 2*s->b8_stride + 2*s->mb_stride;
1118 FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int), fail); // error ressilience code looks cleaner with this
1119 for (y = 0; y < s->mb_height; y++)
1120 for (x = 0; x < s->mb_width; x++)
1121 s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
1123 s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
1126 /* Allocate MV tables */
1127 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1128 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1129 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1130 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1131 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1132 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1133 s->p_mv_table = s->p_mv_table_base + s->mb_stride + 1;
1134 s->b_forw_mv_table = s->b_forw_mv_table_base + s->mb_stride + 1;
1135 s->b_back_mv_table = s->b_back_mv_table_base + s->mb_stride + 1;
1136 s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
1137 s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
1138 s->b_direct_mv_table = s->b_direct_mv_table_base + s->mb_stride + 1;
1140 /* Allocate MB type table */
1141 FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding
1143 FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
1145 FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
1146 mb_array_size * sizeof(float), fail);
1147 FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab,
1148 mb_array_size * sizeof(float), fail);
1152 if (s->codec_id == AV_CODEC_ID_MPEG4 ||
1153 (s->flags & CODEC_FLAG_INTERLACED_ME)) {
1154 /* interlaced direct mode decoding tables */
1155 for (i = 0; i < 2; i++) {
1157 for (j = 0; j < 2; j++) {
1158 for (k = 0; k < 2; k++) {
1159 FF_ALLOCZ_OR_GOTO(s->avctx,
1160 s->b_field_mv_table_base[i][j][k],
1161 mv_table_size * 2 * sizeof(int16_t),
1163 s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
1166 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
1167 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
1168 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
1170 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
1173 if (s->out_format == FMT_H263) {
1175 FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride, fail);
1176 s->coded_block = s->coded_block_base + s->b8_stride + 1;
1178 /* cbp, ac_pred, pred_dir */
1179 FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table , mb_array_size * sizeof(uint8_t), fail);
1180 FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail);
1183 if (s->h263_pred || s->h263_plus || !s->encoding) {
1185 // MN: we need these for error resilience of intra-frames
1186 FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
1187 s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
1188 s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
1189 s->dc_val[2] = s->dc_val[1] + c_size;
1190 for (i = 0; i < yc_size; i++)
1191 s->dc_val_base[i] = 1024;
1194 /* which mb is a intra block */
1195 FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
1196 memset(s->mbintra_table, 1, mb_array_size);
1198 /* init macroblock skip table */
1199 FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
1200 // Note the + 1 is for a quicker mpeg4 slice_end detection
1204 return AVERROR(ENOMEM);
1208 * init common structure for both encoder and decoder.
1209 * this assumes that some variables like width/height are already set
1211 av_cold int ff_MPV_common_init(MpegEncContext *s)
1214 int nb_slices = (HAVE_THREADS &&
1215 s->avctx->active_thread_type & FF_THREAD_SLICE) ?
1216 s->avctx->thread_count : 1;
1218 if (s->encoding && s->avctx->slices)
1219 nb_slices = s->avctx->slices;
1221 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1222 s->mb_height = (s->height + 31) / 32 * 2;
1224 s->mb_height = (s->height + 15) / 16;
1226 if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
1227 av_log(s->avctx, AV_LOG_ERROR,
1228 "decoding to AV_PIX_FMT_NONE is not supported.\n");
1232 if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
1235 max_slices = FFMIN(MAX_THREADS, s->mb_height);
1237 max_slices = MAX_THREADS;
1238 av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
1239 " reducing to %d\n", nb_slices, max_slices);
1240 nb_slices = max_slices;
1243 if ((s->width || s->height) &&
1244 av_image_check_size(s->width, s->height, 0, s->avctx))
1247 ff_dct_common_init(s);
1249 s->flags = s->avctx->flags;
1250 s->flags2 = s->avctx->flags2;
1252 /* set chroma shifts */
1253 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,
1255 &s->chroma_y_shift);
1257 /* convert fourcc to upper case */
1258 s->codec_tag = avpriv_toupper4(s->avctx->codec_tag);
1260 s->stream_codec_tag = avpriv_toupper4(s->avctx->stream_codec_tag);
1262 FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
1263 MAX_PICTURE_COUNT * sizeof(Picture), fail);
1264 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1265 av_frame_unref(&s->picture[i].f);
1267 memset(&s->next_picture, 0, sizeof(s->next_picture));
1268 memset(&s->last_picture, 0, sizeof(s->last_picture));
1269 memset(&s->current_picture, 0, sizeof(s->current_picture));
1270 av_frame_unref(&s->next_picture.f);
1271 av_frame_unref(&s->last_picture.f);
1272 av_frame_unref(&s->current_picture.f);
1274 if (init_context_frame(s))
1277 s->parse_context.state = -1;
1279 s->context_initialized = 1;
1280 s->thread_context[0] = s;
1282 // if (s->width && s->height) {
1283 if (nb_slices > 1) {
1284 for (i = 1; i < nb_slices; i++) {
1285 s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
1286 memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
1289 for (i = 0; i < nb_slices; i++) {
1290 if (init_duplicate_context(s->thread_context[i]) < 0)
1292 s->thread_context[i]->start_mb_y =
1293 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1294 s->thread_context[i]->end_mb_y =
1295 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1298 if (init_duplicate_context(s) < 0)
1301 s->end_mb_y = s->mb_height;
1303 s->slice_context_count = nb_slices;
1308 ff_MPV_common_end(s);
1313 * Frees and resets MpegEncContext fields depending on the resolution.
1314 * Is used during resolution changes to avoid a full reinitialization of the
1317 static int free_context_frame(MpegEncContext *s)
1321 av_freep(&s->mb_type);
1322 av_freep(&s->p_mv_table_base);
1323 av_freep(&s->b_forw_mv_table_base);
1324 av_freep(&s->b_back_mv_table_base);
1325 av_freep(&s->b_bidir_forw_mv_table_base);
1326 av_freep(&s->b_bidir_back_mv_table_base);
1327 av_freep(&s->b_direct_mv_table_base);
1328 s->p_mv_table = NULL;
1329 s->b_forw_mv_table = NULL;
1330 s->b_back_mv_table = NULL;
1331 s->b_bidir_forw_mv_table = NULL;
1332 s->b_bidir_back_mv_table = NULL;
1333 s->b_direct_mv_table = NULL;
1334 for (i = 0; i < 2; i++) {
1335 for (j = 0; j < 2; j++) {
1336 for (k = 0; k < 2; k++) {
1337 av_freep(&s->b_field_mv_table_base[i][j][k]);
1338 s->b_field_mv_table[i][j][k] = NULL;
1340 av_freep(&s->b_field_select_table[i][j]);
1341 av_freep(&s->p_field_mv_table_base[i][j]);
1342 s->p_field_mv_table[i][j] = NULL;
1344 av_freep(&s->p_field_select_table[i]);
1347 av_freep(&s->dc_val_base);
1348 av_freep(&s->coded_block_base);
1349 av_freep(&s->mbintra_table);
1350 av_freep(&s->cbp_table);
1351 av_freep(&s->pred_dir_table);
1353 av_freep(&s->mbskip_table);
1355 av_freep(&s->er.error_status_table);
1356 av_freep(&s->er.er_temp_buffer);
1357 av_freep(&s->mb_index2xy);
1358 av_freep(&s->lambda_table);
1360 av_freep(&s->cplx_tab);
1361 av_freep(&s->bits_tab);
1363 s->linesize = s->uvlinesize = 0;
1368 int ff_MPV_common_frame_size_change(MpegEncContext *s)
1372 if (s->slice_context_count > 1) {
1373 for (i = 0; i < s->slice_context_count; i++) {
1374 free_duplicate_context(s->thread_context[i]);
1376 for (i = 1; i < s->slice_context_count; i++) {
1377 av_freep(&s->thread_context[i]);
1380 free_duplicate_context(s);
1382 if ((err = free_context_frame(s)) < 0)
1386 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1387 s->picture[i].needs_realloc = 1;
1390 s->last_picture_ptr =
1391 s->next_picture_ptr =
1392 s->current_picture_ptr = NULL;
1395 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1396 s->mb_height = (s->height + 31) / 32 * 2;
1398 s->mb_height = (s->height + 15) / 16;
1400 if ((s->width || s->height) &&
1401 av_image_check_size(s->width, s->height, 0, s->avctx))
1402 return AVERROR_INVALIDDATA;
1404 if ((err = init_context_frame(s)))
1407 s->thread_context[0] = s;
1409 if (s->width && s->height) {
1410 int nb_slices = s->slice_context_count;
1411 if (nb_slices > 1) {
1412 for (i = 1; i < nb_slices; i++) {
1413 s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
1414 memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
1417 for (i = 0; i < nb_slices; i++) {
1418 if (init_duplicate_context(s->thread_context[i]) < 0)
1420 s->thread_context[i]->start_mb_y =
1421 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1422 s->thread_context[i]->end_mb_y =
1423 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1426 err = init_duplicate_context(s);
1430 s->end_mb_y = s->mb_height;
1432 s->slice_context_count = nb_slices;
1437 ff_MPV_common_end(s);
1441 /* init common structure for both encoder and decoder */
1442 void ff_MPV_common_end(MpegEncContext *s)
1446 if (s->slice_context_count > 1) {
1447 for (i = 0; i < s->slice_context_count; i++) {
1448 free_duplicate_context(s->thread_context[i]);
1450 for (i = 1; i < s->slice_context_count; i++) {
1451 av_freep(&s->thread_context[i]);
1453 s->slice_context_count = 1;
1454 } else free_duplicate_context(s);
1456 av_freep(&s->parse_context.buffer);
1457 s->parse_context.buffer_size = 0;
1459 av_freep(&s->bitstream_buffer);
1460 s->allocated_bitstream_buffer_size = 0;
1463 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1464 ff_free_picture_tables(&s->picture[i]);
1465 ff_mpeg_unref_picture(s, &s->picture[i]);
1468 av_freep(&s->picture);
1469 ff_free_picture_tables(&s->last_picture);
1470 ff_mpeg_unref_picture(s, &s->last_picture);
1471 ff_free_picture_tables(&s->current_picture);
1472 ff_mpeg_unref_picture(s, &s->current_picture);
1473 ff_free_picture_tables(&s->next_picture);
1474 ff_mpeg_unref_picture(s, &s->next_picture);
1476 free_context_frame(s);
1478 s->context_initialized = 0;
1479 s->last_picture_ptr =
1480 s->next_picture_ptr =
1481 s->current_picture_ptr = NULL;
1482 s->linesize = s->uvlinesize = 0;
1485 av_cold void ff_init_rl(RLTable *rl,
1486 uint8_t static_store[2][2 * MAX_RUN + MAX_LEVEL + 3])
1488 int8_t max_level[MAX_RUN + 1], max_run[MAX_LEVEL + 1];
1489 uint8_t index_run[MAX_RUN + 1];
1490 int last, run, level, start, end, i;
1492 /* If table is static, we can quit if rl->max_level[0] is not NULL */
1493 if (static_store && rl->max_level[0])
1496 /* compute max_level[], max_run[] and index_run[] */
1497 for (last = 0; last < 2; last++) {
1506 memset(max_level, 0, MAX_RUN + 1);
1507 memset(max_run, 0, MAX_LEVEL + 1);
1508 memset(index_run, rl->n, MAX_RUN + 1);
1509 for (i = start; i < end; i++) {
1510 run = rl->table_run[i];
1511 level = rl->table_level[i];
1512 if (index_run[run] == rl->n)
1514 if (level > max_level[run])
1515 max_level[run] = level;
1516 if (run > max_run[level])
1517 max_run[level] = run;
1520 rl->max_level[last] = static_store[last];
1522 rl->max_level[last] = av_malloc(MAX_RUN + 1);
1523 memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
1525 rl->max_run[last] = static_store[last] + MAX_RUN + 1;
1527 rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
1528 memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
1530 rl->index_run[last] = static_store[last] + MAX_RUN + MAX_LEVEL + 2;
1532 rl->index_run[last] = av_malloc(MAX_RUN + 1);
1533 memcpy(rl->index_run[last], index_run, MAX_RUN + 1);
1537 av_cold void ff_init_vlc_rl(RLTable *rl)
1541 for (q = 0; q < 32; q++) {
1543 int qadd = (q - 1) | 1;
1549 for (i = 0; i < rl->vlc.table_size; i++) {
1550 int code = rl->vlc.table[i][0];
1551 int len = rl->vlc.table[i][1];
1554 if (len == 0) { // illegal code
1557 } else if (len < 0) { // more bits needed
1561 if (code == rl->n) { // esc
1565 run = rl->table_run[code] + 1;
1566 level = rl->table_level[code] * qmul + qadd;
1567 if (code >= rl->last) run += 192;
1570 rl->rl_vlc[q][i].len = len;
1571 rl->rl_vlc[q][i].level = level;
1572 rl->rl_vlc[q][i].run = run;
1577 static void release_unused_pictures(MpegEncContext *s)
1581 /* release non reference frames */
1582 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1583 if (!s->picture[i].reference)
1584 ff_mpeg_unref_picture(s, &s->picture[i]);
1588 static inline int pic_is_unused(MpegEncContext *s, Picture *pic)
1590 if (pic == s->last_picture_ptr)
1592 if (pic->f.buf[0] == NULL)
1594 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
1599 static int find_unused_picture(MpegEncContext *s, int shared)
1604 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1605 if (s->picture[i].f.buf[0] == NULL && &s->picture[i] != s->last_picture_ptr)
1609 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1610 if (pic_is_unused(s, &s->picture[i]))
1615 av_log(s->avctx, AV_LOG_FATAL,
1616 "Internal error, picture buffer overflow\n");
1617 /* We could return -1, but the codec would crash trying to draw into a
1618 * non-existing frame anyway. This is safer than waiting for a random crash.
1619 * Also the return of this is never useful, an encoder must only allocate
1620 * as much as allowed in the specification. This has no relationship to how
1621 * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
1622 * enough for such valid streams).
1623 * Plus, a decoder has to check stream validity and remove frames if too
1624 * many reference frames are around. Waiting for "OOM" is not correct at
1625 * all. Similarly, missing reference frames have to be replaced by
1626 * interpolated/MC frames, anything else is a bug in the codec ...
1632 int ff_find_unused_picture(MpegEncContext *s, int shared)
1634 int ret = find_unused_picture(s, shared);
1636 if (ret >= 0 && ret < MAX_PICTURE_COUNT) {
1637 if (s->picture[ret].needs_realloc) {
1638 s->picture[ret].needs_realloc = 0;
1639 ff_free_picture_tables(&s->picture[ret]);
1640 ff_mpeg_unref_picture(s, &s->picture[ret]);
1647 * generic function called after decoding
1648 * the header and before a frame is decoded.
1650 int ff_MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1656 if (!ff_thread_can_start_frame(avctx)) {
1657 av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1661 /* mark & release old frames */
1662 if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1663 s->last_picture_ptr != s->next_picture_ptr &&
1664 s->last_picture_ptr->f.buf[0]) {
1665 ff_mpeg_unref_picture(s, s->last_picture_ptr);
1668 /* release forgotten pictures */
1669 /* if (mpeg124/h263) */
1670 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1671 if (&s->picture[i] != s->last_picture_ptr &&
1672 &s->picture[i] != s->next_picture_ptr &&
1673 s->picture[i].reference && !s->picture[i].needs_realloc) {
1674 if (!(avctx->active_thread_type & FF_THREAD_FRAME))
1675 av_log(avctx, AV_LOG_ERROR,
1676 "releasing zombie picture\n");
1677 ff_mpeg_unref_picture(s, &s->picture[i]);
1681 ff_mpeg_unref_picture(s, &s->current_picture);
1683 release_unused_pictures(s);
1685 if (s->current_picture_ptr &&
1686 s->current_picture_ptr->f.buf[0] == NULL) {
1687 // we already have a unused image
1688 // (maybe it was set before reading the header)
1689 pic = s->current_picture_ptr;
1691 i = ff_find_unused_picture(s, 0);
1693 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1696 pic = &s->picture[i];
1700 if (!s->droppable) {
1701 if (s->pict_type != AV_PICTURE_TYPE_B)
1705 pic->f.coded_picture_number = s->coded_picture_number++;
1707 if (ff_alloc_picture(s, pic, 0) < 0)
1710 s->current_picture_ptr = pic;
1711 // FIXME use only the vars from current_pic
1712 s->current_picture_ptr->f.top_field_first = s->top_field_first;
1713 if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
1714 s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1715 if (s->picture_structure != PICT_FRAME)
1716 s->current_picture_ptr->f.top_field_first =
1717 (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
1719 s->current_picture_ptr->f.interlaced_frame = !s->progressive_frame &&
1720 !s->progressive_sequence;
1721 s->current_picture_ptr->field_picture = s->picture_structure != PICT_FRAME;
1723 s->current_picture_ptr->f.pict_type = s->pict_type;
1724 // if (s->flags && CODEC_FLAG_QSCALE)
1725 // s->current_picture_ptr->quality = s->new_picture_ptr->quality;
1726 s->current_picture_ptr->f.key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1728 if ((ret = ff_mpeg_ref_picture(s, &s->current_picture,
1729 s->current_picture_ptr)) < 0)
1732 if (s->pict_type != AV_PICTURE_TYPE_B) {
1733 s->last_picture_ptr = s->next_picture_ptr;
1735 s->next_picture_ptr = s->current_picture_ptr;
1737 av_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
1738 s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1739 s->last_picture_ptr ? s->last_picture_ptr->f.data[0] : NULL,
1740 s->next_picture_ptr ? s->next_picture_ptr->f.data[0] : NULL,
1741 s->current_picture_ptr ? s->current_picture_ptr->f.data[0] : NULL,
1742 s->pict_type, s->droppable);
1744 if ((s->last_picture_ptr == NULL ||
1745 s->last_picture_ptr->f.buf[0] == NULL) &&
1746 (s->pict_type != AV_PICTURE_TYPE_I ||
1747 s->picture_structure != PICT_FRAME)) {
1748 int h_chroma_shift, v_chroma_shift;
1749 av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1750 &h_chroma_shift, &v_chroma_shift);
1751 if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f.buf[0])
1752 av_log(avctx, AV_LOG_DEBUG,
1753 "allocating dummy last picture for B frame\n");
1754 else if (s->pict_type != AV_PICTURE_TYPE_I)
1755 av_log(avctx, AV_LOG_ERROR,
1756 "warning: first frame is no keyframe\n");
1757 else if (s->picture_structure != PICT_FRAME)
1758 av_log(avctx, AV_LOG_DEBUG,
1759 "allocate dummy last picture for field based first keyframe\n");
1761 /* Allocate a dummy frame */
1762 i = ff_find_unused_picture(s, 0);
1764 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1767 s->last_picture_ptr = &s->picture[i];
1769 s->last_picture_ptr->reference = 3;
1770 s->last_picture_ptr->f.key_frame = 0;
1771 s->last_picture_ptr->f.pict_type = AV_PICTURE_TYPE_P;
1773 if (ff_alloc_picture(s, s->last_picture_ptr, 0) < 0) {
1774 s->last_picture_ptr = NULL;
1778 if (!avctx->hwaccel) {
1779 for(i=0; i<avctx->height; i++)
1780 memset(s->last_picture_ptr->f.data[0] + s->last_picture_ptr->f.linesize[0]*i,
1781 0x80, avctx->width);
1782 for(i=0; i<FF_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
1783 memset(s->last_picture_ptr->f.data[1] + s->last_picture_ptr->f.linesize[1]*i,
1784 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1785 memset(s->last_picture_ptr->f.data[2] + s->last_picture_ptr->f.linesize[2]*i,
1786 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1789 if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
1790 for(i=0; i<avctx->height; i++)
1791 memset(s->last_picture_ptr->f.data[0] + s->last_picture_ptr->f.linesize[0]*i, 16, avctx->width);
1795 ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
1796 ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
1798 if ((s->next_picture_ptr == NULL ||
1799 s->next_picture_ptr->f.buf[0] == NULL) &&
1800 s->pict_type == AV_PICTURE_TYPE_B) {
1801 /* Allocate a dummy frame */
1802 i = ff_find_unused_picture(s, 0);
1804 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1807 s->next_picture_ptr = &s->picture[i];
1809 s->next_picture_ptr->reference = 3;
1810 s->next_picture_ptr->f.key_frame = 0;
1811 s->next_picture_ptr->f.pict_type = AV_PICTURE_TYPE_P;
1813 if (ff_alloc_picture(s, s->next_picture_ptr, 0) < 0) {
1814 s->next_picture_ptr = NULL;
1817 ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
1818 ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
1821 #if 0 // BUFREF-FIXME
1822 memset(s->last_picture.f.data, 0, sizeof(s->last_picture.f.data));
1823 memset(s->next_picture.f.data, 0, sizeof(s->next_picture.f.data));
1825 if (s->last_picture_ptr) {
1826 ff_mpeg_unref_picture(s, &s->last_picture);
1827 if (s->last_picture_ptr->f.buf[0] &&
1828 (ret = ff_mpeg_ref_picture(s, &s->last_picture,
1829 s->last_picture_ptr)) < 0)
1832 if (s->next_picture_ptr) {
1833 ff_mpeg_unref_picture(s, &s->next_picture);
1834 if (s->next_picture_ptr->f.buf[0] &&
1835 (ret = ff_mpeg_ref_picture(s, &s->next_picture,
1836 s->next_picture_ptr)) < 0)
1840 av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
1841 s->last_picture_ptr->f.buf[0]));
1843 if (s->picture_structure!= PICT_FRAME) {
1845 for (i = 0; i < 4; i++) {
1846 if (s->picture_structure == PICT_BOTTOM_FIELD) {
1847 s->current_picture.f.data[i] +=
1848 s->current_picture.f.linesize[i];
1850 s->current_picture.f.linesize[i] *= 2;
1851 s->last_picture.f.linesize[i] *= 2;
1852 s->next_picture.f.linesize[i] *= 2;
1856 s->err_recognition = avctx->err_recognition;
1858 /* set dequantizer, we can't do it during init as
1859 * it might change for mpeg4 and we can't do it in the header
1860 * decode as init is not called for mpeg4 there yet */
1861 if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1862 s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1863 s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1864 } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1865 s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1866 s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1868 s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1869 s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1875 /* called after a frame has been decoded. */
1876 void ff_MPV_frame_end(MpegEncContext *s)
1880 if (s->current_picture.reference)
1881 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1885 * Draw a line from (ex, ey) -> (sx, sy).
1886 * @param w width of the image
1887 * @param h height of the image
1888 * @param stride stride/linesize of the image
1889 * @param color color of the arrow
1891 static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey,
1892 int w, int h, int stride, int color)
1896 sx = av_clip(sx, 0, w - 1);
1897 sy = av_clip(sy, 0, h - 1);
1898 ex = av_clip(ex, 0, w - 1);
1899 ey = av_clip(ey, 0, h - 1);
1901 buf[sy * stride + sx] += color;
1903 if (FFABS(ex - sx) > FFABS(ey - sy)) {
1905 FFSWAP(int, sx, ex);
1906 FFSWAP(int, sy, ey);
1908 buf += sx + sy * stride;
1910 f = ((ey - sy) << 16) / ex;
1911 for (x = 0; x <= ex; x++) {
1913 fr = (x * f) & 0xFFFF;
1914 buf[y * stride + x] += (color * (0x10000 - fr)) >> 16;
1915 if(fr) buf[(y + 1) * stride + x] += (color * fr ) >> 16;
1919 FFSWAP(int, sx, ex);
1920 FFSWAP(int, sy, ey);
1922 buf += sx + sy * stride;
1925 f = ((ex - sx) << 16) / ey;
1928 for(y= 0; y <= ey; y++){
1930 fr = (y*f) & 0xFFFF;
1931 buf[y * stride + x] += (color * (0x10000 - fr)) >> 16;
1932 if(fr) buf[y * stride + x + 1] += (color * fr ) >> 16;
1938 * Draw an arrow from (ex, ey) -> (sx, sy).
1939 * @param w width of the image
1940 * @param h height of the image
1941 * @param stride stride/linesize of the image
1942 * @param color color of the arrow
1944 static void draw_arrow(uint8_t *buf, int sx, int sy, int ex,
1945 int ey, int w, int h, int stride, int color)
1949 sx = av_clip(sx, -100, w + 100);
1950 sy = av_clip(sy, -100, h + 100);
1951 ex = av_clip(ex, -100, w + 100);
1952 ey = av_clip(ey, -100, h + 100);
1957 if (dx * dx + dy * dy > 3 * 3) {
1960 int length = ff_sqrt((rx * rx + ry * ry) << 8);
1962 // FIXME subpixel accuracy
1963 rx = ROUNDED_DIV(rx * 3 << 4, length);
1964 ry = ROUNDED_DIV(ry * 3 << 4, length);
1966 draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
1967 draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
1969 draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
1973 * Print debugging info for the given picture.
1975 void ff_print_debug_info2(AVCodecContext *avctx, Picture *p, AVFrame *pict, uint8_t *mbskip_table,
1977 int mb_width, int mb_height, int mb_stride, int quarter_sample)
1979 if (avctx->hwaccel || !p || !p->mb_type
1980 || (avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU))
1984 if (avctx->debug & (FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)) {
1987 av_log(avctx, AV_LOG_DEBUG, "New frame, type: %c\n",
1988 av_get_picture_type_char(pict->pict_type));
1989 for (y = 0; y < mb_height; y++) {
1990 for (x = 0; x < mb_width; x++) {
1991 if (avctx->debug & FF_DEBUG_SKIP) {
1992 int count = mbskip_table[x + y * mb_stride];
1995 av_log(avctx, AV_LOG_DEBUG, "%1d", count);
1997 if (avctx->debug & FF_DEBUG_QP) {
1998 av_log(avctx, AV_LOG_DEBUG, "%2d",
1999 p->qscale_table[x + y * mb_stride]);
2001 if (avctx->debug & FF_DEBUG_MB_TYPE) {
2002 int mb_type = p->mb_type[x + y * mb_stride];
2003 // Type & MV direction
2004 if (IS_PCM(mb_type))
2005 av_log(avctx, AV_LOG_DEBUG, "P");
2006 else if (IS_INTRA(mb_type) && IS_ACPRED(mb_type))
2007 av_log(avctx, AV_LOG_DEBUG, "A");
2008 else if (IS_INTRA4x4(mb_type))
2009 av_log(avctx, AV_LOG_DEBUG, "i");
2010 else if (IS_INTRA16x16(mb_type))
2011 av_log(avctx, AV_LOG_DEBUG, "I");
2012 else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type))
2013 av_log(avctx, AV_LOG_DEBUG, "d");
2014 else if (IS_DIRECT(mb_type))
2015 av_log(avctx, AV_LOG_DEBUG, "D");
2016 else if (IS_GMC(mb_type) && IS_SKIP(mb_type))
2017 av_log(avctx, AV_LOG_DEBUG, "g");
2018 else if (IS_GMC(mb_type))
2019 av_log(avctx, AV_LOG_DEBUG, "G");
2020 else if (IS_SKIP(mb_type))
2021 av_log(avctx, AV_LOG_DEBUG, "S");
2022 else if (!USES_LIST(mb_type, 1))
2023 av_log(avctx, AV_LOG_DEBUG, ">");
2024 else if (!USES_LIST(mb_type, 0))
2025 av_log(avctx, AV_LOG_DEBUG, "<");
2027 av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
2028 av_log(avctx, AV_LOG_DEBUG, "X");
2032 if (IS_8X8(mb_type))
2033 av_log(avctx, AV_LOG_DEBUG, "+");
2034 else if (IS_16X8(mb_type))
2035 av_log(avctx, AV_LOG_DEBUG, "-");
2036 else if (IS_8X16(mb_type))
2037 av_log(avctx, AV_LOG_DEBUG, "|");
2038 else if (IS_INTRA(mb_type) || IS_16X16(mb_type))
2039 av_log(avctx, AV_LOG_DEBUG, " ");
2041 av_log(avctx, AV_LOG_DEBUG, "?");
2044 if (IS_INTERLACED(mb_type))
2045 av_log(avctx, AV_LOG_DEBUG, "=");
2047 av_log(avctx, AV_LOG_DEBUG, " ");
2050 av_log(avctx, AV_LOG_DEBUG, "\n");
2054 if ((avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
2055 (avctx->debug_mv)) {
2056 const int shift = 1 + quarter_sample;
2060 int h_chroma_shift, v_chroma_shift, block_height;
2061 const int width = avctx->width;
2062 const int height = avctx->height;
2063 const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1;
2064 const int mv_stride = (mb_width << mv_sample_log2) +
2065 (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1);
2067 *low_delay = 0; // needed to see the vectors without trashing the buffers
2069 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
2071 av_frame_make_writable(pict);
2073 pict->opaque = NULL;
2074 ptr = pict->data[0];
2075 block_height = 16 >> v_chroma_shift;
2077 for (mb_y = 0; mb_y < mb_height; mb_y++) {
2079 for (mb_x = 0; mb_x < mb_width; mb_x++) {
2080 const int mb_index = mb_x + mb_y * mb_stride;
2081 if ((avctx->debug_mv) && p->motion_val[0]) {
2083 for (type = 0; type < 3; type++) {
2087 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_P_FOR)) ||
2088 (pict->pict_type!= AV_PICTURE_TYPE_P))
2093 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_FOR)) ||
2094 (pict->pict_type!= AV_PICTURE_TYPE_B))
2099 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_BACK)) ||
2100 (pict->pict_type!= AV_PICTURE_TYPE_B))
2105 if (!USES_LIST(p->mb_type[mb_index], direction))
2108 if (IS_8X8(p->mb_type[mb_index])) {
2110 for (i = 0; i < 4; i++) {
2111 int sx = mb_x * 16 + 4 + 8 * (i & 1);
2112 int sy = mb_y * 16 + 4 + 8 * (i >> 1);
2113 int xy = (mb_x * 2 + (i & 1) +
2114 (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
2115 int mx = (p->motion_val[direction][xy][0] >> shift) + sx;
2116 int my = (p->motion_val[direction][xy][1] >> shift) + sy;
2117 draw_arrow(ptr, sx, sy, mx, my, width,
2118 height, pict->linesize[0], 100);
2120 } else if (IS_16X8(p->mb_type[mb_index])) {
2122 for (i = 0; i < 2; i++) {
2123 int sx = mb_x * 16 + 8;
2124 int sy = mb_y * 16 + 4 + 8 * i;
2125 int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
2126 int mx = (p->motion_val[direction][xy][0] >> shift);
2127 int my = (p->motion_val[direction][xy][1] >> shift);
2129 if (IS_INTERLACED(p->mb_type[mb_index]))
2132 draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
2133 height, pict->linesize[0], 100);
2135 } else if (IS_8X16(p->mb_type[mb_index])) {
2137 for (i = 0; i < 2; i++) {
2138 int sx = mb_x * 16 + 4 + 8 * i;
2139 int sy = mb_y * 16 + 8;
2140 int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
2141 int mx = p->motion_val[direction][xy][0] >> shift;
2142 int my = p->motion_val[direction][xy][1] >> shift;
2144 if (IS_INTERLACED(p->mb_type[mb_index]))
2147 draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
2148 height, pict->linesize[0], 100);
2151 int sx= mb_x * 16 + 8;
2152 int sy= mb_y * 16 + 8;
2153 int xy= (mb_x + mb_y * mv_stride) << mv_sample_log2;
2154 int mx= (p->motion_val[direction][xy][0]>>shift) + sx;
2155 int my= (p->motion_val[direction][xy][1]>>shift) + sy;
2156 draw_arrow(ptr, sx, sy, mx, my, width, height, pict->linesize[0], 100);
2160 if ((avctx->debug & FF_DEBUG_VIS_QP)) {
2161 uint64_t c = (p->qscale_table[mb_index] * 128 / 31) *
2162 0x0101010101010101ULL;
2164 for (y = 0; y < block_height; y++) {
2165 *(uint64_t *)(pict->data[1] + 8 * mb_x +
2166 (block_height * mb_y + y) *
2167 pict->linesize[1]) = c;
2168 *(uint64_t *)(pict->data[2] + 8 * mb_x +
2169 (block_height * mb_y + y) *
2170 pict->linesize[2]) = c;
2173 if ((avctx->debug & FF_DEBUG_VIS_MB_TYPE) &&
2175 int mb_type = p->mb_type[mb_index];
2178 #define COLOR(theta, r) \
2179 u = (int)(128 + r * cos(theta * 3.141592 / 180)); \
2180 v = (int)(128 + r * sin(theta * 3.141592 / 180));
2184 if (IS_PCM(mb_type)) {
2186 } else if ((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) ||
2187 IS_INTRA16x16(mb_type)) {
2189 } else if (IS_INTRA4x4(mb_type)) {
2191 } else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type)) {
2193 } else if (IS_DIRECT(mb_type)) {
2195 } else if (IS_GMC(mb_type) && IS_SKIP(mb_type)) {
2197 } else if (IS_GMC(mb_type)) {
2199 } else if (IS_SKIP(mb_type)) {
2201 } else if (!USES_LIST(mb_type, 1)) {
2203 } else if (!USES_LIST(mb_type, 0)) {
2206 av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
2210 u *= 0x0101010101010101ULL;
2211 v *= 0x0101010101010101ULL;
2212 for (y = 0; y < block_height; y++) {
2213 *(uint64_t *)(pict->data[1] + 8 * mb_x +
2214 (block_height * mb_y + y) * pict->linesize[1]) = u;
2215 *(uint64_t *)(pict->data[2] + 8 * mb_x +
2216 (block_height * mb_y + y) * pict->linesize[2]) = v;
2220 if (IS_8X8(mb_type) || IS_16X8(mb_type)) {
2221 *(uint64_t *)(pict->data[0] + 16 * mb_x + 0 +
2222 (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
2223 *(uint64_t *)(pict->data[0] + 16 * mb_x + 8 +
2224 (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
2226 if (IS_8X8(mb_type) || IS_8X16(mb_type)) {
2227 for (y = 0; y < 16; y++)
2228 pict->data[0][16 * mb_x + 8 + (16 * mb_y + y) *
2229 pict->linesize[0]] ^= 0x80;
2231 if (IS_8X8(mb_type) && mv_sample_log2 >= 2) {
2232 int dm = 1 << (mv_sample_log2 - 2);
2233 for (i = 0; i < 4; i++) {
2234 int sx = mb_x * 16 + 8 * (i & 1);
2235 int sy = mb_y * 16 + 8 * (i >> 1);
2236 int xy = (mb_x * 2 + (i & 1) +
2237 (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
2239 int32_t *mv = (int32_t *) &p->motion_val[0][xy];
2240 if (mv[0] != mv[dm] ||
2241 mv[dm * mv_stride] != mv[dm * (mv_stride + 1)])
2242 for (y = 0; y < 8; y++)
2243 pict->data[0][sx + 4 + (sy + y) * pict->linesize[0]] ^= 0x80;
2244 if (mv[0] != mv[dm * mv_stride] || mv[dm] != mv[dm * (mv_stride + 1)])
2245 *(uint64_t *)(pict->data[0] + sx + (sy + 4) *
2246 pict->linesize[0]) ^= 0x8080808080808080ULL;
2250 if (IS_INTERLACED(mb_type) &&
2251 avctx->codec->id == AV_CODEC_ID_H264) {
2255 mbskip_table[mb_index] = 0;
2261 void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
2263 ff_print_debug_info2(s->avctx, p, pict, s->mbskip_table, &s->low_delay,
2264 s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
2267 int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
2269 AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
2270 int offset = 2*s->mb_stride + 1;
2272 return AVERROR(ENOMEM);
2273 av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
2274 ref->size -= offset;
2275 ref->data += offset;
2276 return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
2279 static inline int hpel_motion_lowres(MpegEncContext *s,
2280 uint8_t *dest, uint8_t *src,
2281 int field_based, int field_select,
2282 int src_x, int src_y,
2283 int width, int height, ptrdiff_t stride,
2284 int h_edge_pos, int v_edge_pos,
2285 int w, int h, h264_chroma_mc_func *pix_op,
2286 int motion_x, int motion_y)
2288 const int lowres = s->avctx->lowres;
2289 const int op_index = FFMIN(lowres, 3);
2290 const int s_mask = (2 << lowres) - 1;
2294 if (s->quarter_sample) {
2299 sx = motion_x & s_mask;
2300 sy = motion_y & s_mask;
2301 src_x += motion_x >> lowres + 1;
2302 src_y += motion_y >> lowres + 1;
2304 src += src_y * stride + src_x;
2306 if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w, 0) ||
2307 (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
2308 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
2309 s->linesize, s->linesize,
2310 w + 1, (h + 1) << field_based,
2311 src_x, src_y << field_based,
2312 h_edge_pos, v_edge_pos);
2313 src = s->edge_emu_buffer;
2317 sx = (sx << 2) >> lowres;
2318 sy = (sy << 2) >> lowres;
2321 pix_op[op_index](dest, src, stride, h, sx, sy);
2325 /* apply one mpeg motion vector to the three components */
2326 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
2333 uint8_t **ref_picture,
2334 h264_chroma_mc_func *pix_op,
2335 int motion_x, int motion_y,
2338 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
2339 int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
2340 ptrdiff_t uvlinesize, linesize;
2341 const int lowres = s->avctx->lowres;
2342 const int op_index = FFMIN(lowres-1+s->chroma_x_shift, 3);
2343 const int block_s = 8>>lowres;
2344 const int s_mask = (2 << lowres) - 1;
2345 const int h_edge_pos = s->h_edge_pos >> lowres;
2346 const int v_edge_pos = s->v_edge_pos >> lowres;
2347 linesize = s->current_picture.f.linesize[0] << field_based;
2348 uvlinesize = s->current_picture.f.linesize[1] << field_based;
2350 // FIXME obviously not perfect but qpel will not work in lowres anyway
2351 if (s->quarter_sample) {
2357 motion_y += (bottom_field - field_select)*((1 << lowres)-1);
2360 sx = motion_x & s_mask;
2361 sy = motion_y & s_mask;
2362 src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
2363 src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
2365 if (s->out_format == FMT_H263) {
2366 uvsx = ((motion_x >> 1) & s_mask) | (sx & 1);
2367 uvsy = ((motion_y >> 1) & s_mask) | (sy & 1);
2368 uvsrc_x = src_x >> 1;
2369 uvsrc_y = src_y >> 1;
2370 } else if (s->out_format == FMT_H261) {
2371 // even chroma mv's are full pel in H261
2374 uvsx = (2 * mx) & s_mask;
2375 uvsy = (2 * my) & s_mask;
2376 uvsrc_x = s->mb_x * block_s + (mx >> lowres);
2377 uvsrc_y = mb_y * block_s + (my >> lowres);
2379 if(s->chroma_y_shift){
2384 uvsrc_x = s->mb_x * block_s + (mx >> lowres + 1);
2385 uvsrc_y = (mb_y * block_s >> field_based) + (my >> lowres + 1);
2387 if(s->chroma_x_shift){
2391 uvsy = motion_y & s_mask;
2393 uvsrc_x = s->mb_x*block_s + (mx >> (lowres+1));
2396 uvsx = motion_x & s_mask;
2397 uvsy = motion_y & s_mask;
2404 ptr_y = ref_picture[0] + src_y * linesize + src_x;
2405 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
2406 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
2408 if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s, 0) || uvsrc_y<0 ||
2409 (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
2410 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
2411 linesize >> field_based, linesize >> field_based,
2412 17, 17 + field_based,
2413 src_x, src_y << field_based, h_edge_pos,
2415 ptr_y = s->edge_emu_buffer;
2416 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
2417 uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
2418 s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
2419 uvlinesize >> field_based, uvlinesize >> field_based,
2421 uvsrc_x, uvsrc_y << field_based,
2422 h_edge_pos >> 1, v_edge_pos >> 1);
2423 s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
2424 uvlinesize >> field_based,uvlinesize >> field_based,
2426 uvsrc_x, uvsrc_y << field_based,
2427 h_edge_pos >> 1, v_edge_pos >> 1);
2429 ptr_cr = uvbuf + 16;
2433 // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f.data
2435 dest_y += s->linesize;
2436 dest_cb += s->uvlinesize;
2437 dest_cr += s->uvlinesize;
2441 ptr_y += s->linesize;
2442 ptr_cb += s->uvlinesize;
2443 ptr_cr += s->uvlinesize;
2446 sx = (sx << 2) >> lowres;
2447 sy = (sy << 2) >> lowres;
2448 pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
2450 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
2451 int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
2452 uvsx = (uvsx << 2) >> lowres;
2453 uvsy = (uvsy << 2) >> lowres;
2455 pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
2456 pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
2459 // FIXME h261 lowres loop filter
2462 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
2463 uint8_t *dest_cb, uint8_t *dest_cr,
2464 uint8_t **ref_picture,
2465 h264_chroma_mc_func * pix_op,
2468 const int lowres = s->avctx->lowres;
2469 const int op_index = FFMIN(lowres, 3);
2470 const int block_s = 8 >> lowres;
2471 const int s_mask = (2 << lowres) - 1;
2472 const int h_edge_pos = s->h_edge_pos >> lowres + 1;
2473 const int v_edge_pos = s->v_edge_pos >> lowres + 1;
2474 int emu = 0, src_x, src_y, sx, sy;
2478 if (s->quarter_sample) {
2483 /* In case of 8X8, we construct a single chroma motion vector
2484 with a special rounding */
2485 mx = ff_h263_round_chroma(mx);
2486 my = ff_h263_round_chroma(my);
2490 src_x = s->mb_x * block_s + (mx >> lowres + 1);
2491 src_y = s->mb_y * block_s + (my >> lowres + 1);
2493 offset = src_y * s->uvlinesize + src_x;
2494 ptr = ref_picture[1] + offset;
2495 if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
2496 (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
2497 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
2498 s->uvlinesize, s->uvlinesize,
2500 src_x, src_y, h_edge_pos, v_edge_pos);
2501 ptr = s->edge_emu_buffer;
2504 sx = (sx << 2) >> lowres;
2505 sy = (sy << 2) >> lowres;
2506 pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
2508 ptr = ref_picture[2] + offset;
2510 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
2511 s->uvlinesize, s->uvlinesize,
2513 src_x, src_y, h_edge_pos, v_edge_pos);
2514 ptr = s->edge_emu_buffer;
2516 pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
2520 * motion compensation of a single macroblock
2522 * @param dest_y luma destination pointer
2523 * @param dest_cb chroma cb/u destination pointer
2524 * @param dest_cr chroma cr/v destination pointer
2525 * @param dir direction (0->forward, 1->backward)
2526 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
2527 * @param pix_op halfpel motion compensation function (average or put normally)
2528 * the motion vectors are taken from s->mv and the MV type from s->mv_type
2530 static inline void MPV_motion_lowres(MpegEncContext *s,
2531 uint8_t *dest_y, uint8_t *dest_cb,
2533 int dir, uint8_t **ref_picture,
2534 h264_chroma_mc_func *pix_op)
2538 const int lowres = s->avctx->lowres;
2539 const int block_s = 8 >>lowres;
2544 switch (s->mv_type) {
2546 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2548 ref_picture, pix_op,
2549 s->mv[dir][0][0], s->mv[dir][0][1],
2555 for (i = 0; i < 4; i++) {
2556 hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
2557 s->linesize) * block_s,
2558 ref_picture[0], 0, 0,
2559 (2 * mb_x + (i & 1)) * block_s,
2560 (2 * mb_y + (i >> 1)) * block_s,
2561 s->width, s->height, s->linesize,
2562 s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
2563 block_s, block_s, pix_op,
2564 s->mv[dir][i][0], s->mv[dir][i][1]);
2566 mx += s->mv[dir][i][0];
2567 my += s->mv[dir][i][1];
2570 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
2571 chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
2575 if (s->picture_structure == PICT_FRAME) {
2577 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2578 1, 0, s->field_select[dir][0],
2579 ref_picture, pix_op,
2580 s->mv[dir][0][0], s->mv[dir][0][1],
2583 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2584 1, 1, s->field_select[dir][1],
2585 ref_picture, pix_op,
2586 s->mv[dir][1][0], s->mv[dir][1][1],
2589 if (s->picture_structure != s->field_select[dir][0] + 1 &&
2590 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
2591 ref_picture = s->current_picture_ptr->f.data;
2594 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2595 0, 0, s->field_select[dir][0],
2596 ref_picture, pix_op,
2598 s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
2602 for (i = 0; i < 2; i++) {
2603 uint8_t **ref2picture;
2605 if (s->picture_structure == s->field_select[dir][i] + 1 ||
2606 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
2607 ref2picture = ref_picture;
2609 ref2picture = s->current_picture_ptr->f.data;
2612 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2613 0, 0, s->field_select[dir][i],
2614 ref2picture, pix_op,
2615 s->mv[dir][i][0], s->mv[dir][i][1] +
2616 2 * block_s * i, block_s, mb_y >> 1);
2618 dest_y += 2 * block_s * s->linesize;
2619 dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
2620 dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
2624 if (s->picture_structure == PICT_FRAME) {
2625 for (i = 0; i < 2; i++) {
2627 for (j = 0; j < 2; j++) {
2628 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2630 ref_picture, pix_op,
2631 s->mv[dir][2 * i + j][0],
2632 s->mv[dir][2 * i + j][1],
2635 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
2638 for (i = 0; i < 2; i++) {
2639 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2640 0, 0, s->picture_structure != i + 1,
2641 ref_picture, pix_op,
2642 s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
2643 2 * block_s, mb_y >> 1);
2645 // after put we make avg of the same block
2646 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
2648 // opposite parity is always in the same
2649 // frame if this is second field
2650 if (!s->first_field) {
2651 ref_picture = s->current_picture_ptr->f.data;
2662 * find the lowest MB row referenced in the MVs
2664 int ff_MPV_lowest_referenced_row(MpegEncContext *s, int dir)
2666 int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
2667 int my, off, i, mvs;
2669 if (s->picture_structure != PICT_FRAME || s->mcsel)
2672 switch (s->mv_type) {
2686 for (i = 0; i < mvs; i++) {
2687 my = s->mv[dir][i][1]<<qpel_shift;
2688 my_max = FFMAX(my_max, my);
2689 my_min = FFMIN(my_min, my);
2692 off = (FFMAX(-my_min, my_max) + 63) >> 6;
2694 return FFMIN(FFMAX(s->mb_y + off, 0), s->mb_height-1);
2696 return s->mb_height-1;
2699 /* put block[] to dest[] */
2700 static inline void put_dct(MpegEncContext *s,
2701 int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
2703 s->dct_unquantize_intra(s, block, i, qscale);
2704 s->dsp.idct_put (dest, line_size, block);
2707 /* add block[] to dest[] */
2708 static inline void add_dct(MpegEncContext *s,
2709 int16_t *block, int i, uint8_t *dest, int line_size)
2711 if (s->block_last_index[i] >= 0) {
2712 s->dsp.idct_add (dest, line_size, block);
2716 static inline void add_dequant_dct(MpegEncContext *s,
2717 int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
2719 if (s->block_last_index[i] >= 0) {
2720 s->dct_unquantize_inter(s, block, i, qscale);
2722 s->dsp.idct_add (dest, line_size, block);
2727 * Clean dc, ac, coded_block for the current non-intra MB.
2729 void ff_clean_intra_table_entries(MpegEncContext *s)
2731 int wrap = s->b8_stride;
2732 int xy = s->block_index[0];
2735 s->dc_val[0][xy + 1 ] =
2736 s->dc_val[0][xy + wrap] =
2737 s->dc_val[0][xy + 1 + wrap] = 1024;
2739 memset(s->ac_val[0][xy ], 0, 32 * sizeof(int16_t));
2740 memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
2741 if (s->msmpeg4_version>=3) {
2742 s->coded_block[xy ] =
2743 s->coded_block[xy + 1 ] =
2744 s->coded_block[xy + wrap] =
2745 s->coded_block[xy + 1 + wrap] = 0;
2748 wrap = s->mb_stride;
2749 xy = s->mb_x + s->mb_y * wrap;
2751 s->dc_val[2][xy] = 1024;
2753 memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
2754 memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
2756 s->mbintra_table[xy]= 0;
2759 /* generic function called after a macroblock has been parsed by the
2760 decoder or after it has been encoded by the encoder.
2762 Important variables used:
2763 s->mb_intra : true if intra macroblock
2764 s->mv_dir : motion vector direction
2765 s->mv_type : motion vector type
2766 s->mv : motion vector
2767 s->interlaced_dct : true if interlaced dct used (mpeg2)
2769 static av_always_inline
2770 void MPV_decode_mb_internal(MpegEncContext *s, int16_t block[12][64],
2771 int lowres_flag, int is_mpeg12)
2773 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
2776 s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
2777 s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
2781 if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
2782 /* print DCT coefficients */
2784 av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
2786 for(j=0; j<64; j++){
2787 av_log(s->avctx, AV_LOG_DEBUG, "%5d", block[i][s->dsp.idct_permutation[j]]);
2789 av_log(s->avctx, AV_LOG_DEBUG, "\n");
2793 s->current_picture.qscale_table[mb_xy] = s->qscale;
2795 /* update DC predictors for P macroblocks */
2797 if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
2798 if(s->mbintra_table[mb_xy])
2799 ff_clean_intra_table_entries(s);
2803 s->last_dc[2] = 128 << s->intra_dc_precision;
2806 else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
2807 s->mbintra_table[mb_xy]=1;
2809 if ( (s->flags&CODEC_FLAG_PSNR)
2810 || s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor
2811 || !(s->encoding && (s->intra_only || s->pict_type==AV_PICTURE_TYPE_B) && s->avctx->mb_decision != FF_MB_DECISION_RD)) { //FIXME precalc
2812 uint8_t *dest_y, *dest_cb, *dest_cr;
2813 int dct_linesize, dct_offset;
2814 op_pixels_func (*op_pix)[4];
2815 qpel_mc_func (*op_qpix)[16];
2816 const int linesize = s->current_picture.f.linesize[0]; //not s->linesize as this would be wrong for field pics
2817 const int uvlinesize = s->current_picture.f.linesize[1];
2818 const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
2819 const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
2821 /* avoid copy if macroblock skipped in last frame too */
2822 /* skip only during decoding as we might trash the buffers during encoding a bit */
2824 uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
2826 if (s->mb_skipped) {
2828 av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
2830 } else if(!s->current_picture.reference) {
2833 *mbskip_ptr = 0; /* not skipped */
2837 dct_linesize = linesize << s->interlaced_dct;
2838 dct_offset = s->interlaced_dct ? linesize : linesize * block_size;
2842 dest_cb= s->dest[1];
2843 dest_cr= s->dest[2];
2845 dest_y = s->b_scratchpad;
2846 dest_cb= s->b_scratchpad+16*linesize;
2847 dest_cr= s->b_scratchpad+32*linesize;
2851 /* motion handling */
2852 /* decoding or more than one mb_type (MC was already done otherwise) */
2855 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
2856 if (s->mv_dir & MV_DIR_FORWARD) {
2857 ff_thread_await_progress(&s->last_picture_ptr->tf,
2858 ff_MPV_lowest_referenced_row(s, 0),
2861 if (s->mv_dir & MV_DIR_BACKWARD) {
2862 ff_thread_await_progress(&s->next_picture_ptr->tf,
2863 ff_MPV_lowest_referenced_row(s, 1),
2869 h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
2871 if (s->mv_dir & MV_DIR_FORWARD) {
2872 MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix);
2873 op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
2875 if (s->mv_dir & MV_DIR_BACKWARD) {
2876 MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix);
2879 op_qpix = s->me.qpel_put;
2880 if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
2881 op_pix = s->hdsp.put_pixels_tab;
2883 op_pix = s->hdsp.put_no_rnd_pixels_tab;
2885 if (s->mv_dir & MV_DIR_FORWARD) {
2886 ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix, op_qpix);
2887 op_pix = s->hdsp.avg_pixels_tab;
2888 op_qpix= s->me.qpel_avg;
2890 if (s->mv_dir & MV_DIR_BACKWARD) {
2891 ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix, op_qpix);
2896 /* skip dequant / idct if we are really late ;) */
2897 if(s->avctx->skip_idct){
2898 if( (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
2899 ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
2900 || s->avctx->skip_idct >= AVDISCARD_ALL)
2904 /* add dct residue */
2905 if(s->encoding || !( s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
2906 || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
2907 add_dequant_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale);
2908 add_dequant_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale);
2909 add_dequant_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale);
2910 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2912 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2913 if (s->chroma_y_shift){
2914 add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2915 add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2919 add_dequant_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale);
2920 add_dequant_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale);
2921 add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2922 add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2925 } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
2926 add_dct(s, block[0], 0, dest_y , dct_linesize);
2927 add_dct(s, block[1], 1, dest_y + block_size, dct_linesize);
2928 add_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize);
2929 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
2931 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2932 if(s->chroma_y_shift){//Chroma420
2933 add_dct(s, block[4], 4, dest_cb, uvlinesize);
2934 add_dct(s, block[5], 5, dest_cr, uvlinesize);
2937 dct_linesize = uvlinesize << s->interlaced_dct;
2938 dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2940 add_dct(s, block[4], 4, dest_cb, dct_linesize);
2941 add_dct(s, block[5], 5, dest_cr, dct_linesize);
2942 add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
2943 add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
2944 if(!s->chroma_x_shift){//Chroma444
2945 add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
2946 add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
2947 add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
2948 add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
2953 else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2954 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2957 /* dct only in intra block */
2958 if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
2959 put_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale);
2960 put_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale);
2961 put_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale);
2962 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2964 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2965 if(s->chroma_y_shift){
2966 put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2967 put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2971 put_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale);
2972 put_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale);
2973 put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2974 put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2978 s->dsp.idct_put(dest_y , dct_linesize, block[0]);
2979 s->dsp.idct_put(dest_y + block_size, dct_linesize, block[1]);
2980 s->dsp.idct_put(dest_y + dct_offset , dct_linesize, block[2]);
2981 s->dsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2983 if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2984 if(s->chroma_y_shift){
2985 s->dsp.idct_put(dest_cb, uvlinesize, block[4]);
2986 s->dsp.idct_put(dest_cr, uvlinesize, block[5]);
2989 dct_linesize = uvlinesize << s->interlaced_dct;
2990 dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2992 s->dsp.idct_put(dest_cb, dct_linesize, block[4]);
2993 s->dsp.idct_put(dest_cr, dct_linesize, block[5]);
2994 s->dsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2995 s->dsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2996 if(!s->chroma_x_shift){//Chroma444
2997 s->dsp.idct_put(dest_cb + block_size, dct_linesize, block[8]);
2998 s->dsp.idct_put(dest_cr + block_size, dct_linesize, block[9]);
2999 s->dsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
3000 s->dsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
3008 s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y , linesize,16);
3009 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
3010 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
3015 void ff_MPV_decode_mb(MpegEncContext *s, int16_t block[12][64]){
3017 if(s->out_format == FMT_MPEG1) {
3018 if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 1);
3019 else MPV_decode_mb_internal(s, block, 0, 1);
3022 if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 0);
3023 else MPV_decode_mb_internal(s, block, 0, 0);
3026 void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
3028 ff_draw_horiz_band(s->avctx, &s->current_picture_ptr->f,
3029 &s->last_picture_ptr->f, y, h, s->picture_structure,
3030 s->first_field, s->low_delay);
3033 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
3034 const int linesize = s->current_picture.f.linesize[0]; //not s->linesize as this would be wrong for field pics
3035 const int uvlinesize = s->current_picture.f.linesize[1];
3036 const int mb_size= 4 - s->avctx->lowres;
3038 s->block_index[0]= s->b8_stride*(s->mb_y*2 ) - 2 + s->mb_x*2;
3039 s->block_index[1]= s->b8_stride*(s->mb_y*2 ) - 1 + s->mb_x*2;
3040 s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
3041 s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
3042 s->block_index[4]= s->mb_stride*(s->mb_y + 1) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
3043 s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
3044 //block_index is not used by mpeg2, so it is not affected by chroma_format
3046 s->dest[0] = s->current_picture.f.data[0] + ((s->mb_x - 1) << mb_size);
3047 s->dest[1] = s->current_picture.f.data[1] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
3048 s->dest[2] = s->current_picture.f.data[2] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
3050 if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
3052 if(s->picture_structure==PICT_FRAME){
3053 s->dest[0] += s->mb_y * linesize << mb_size;
3054 s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
3055 s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
3057 s->dest[0] += (s->mb_y>>1) * linesize << mb_size;
3058 s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
3059 s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
3060 av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
3066 * Permute an 8x8 block.
3067 * @param block the block which will be permuted according to the given permutation vector
3068 * @param permutation the permutation vector
3069 * @param last the last non zero coefficient in scantable order, used to speed the permutation up
3070 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not
3071 * (inverse) permutated to scantable order!
3073 void ff_block_permute(int16_t *block, uint8_t *permutation, const uint8_t *scantable, int last)
3079 //if(permutation[1]==1) return; //FIXME it is ok but not clean and might fail for some permutations
3081 for(i=0; i<=last; i++){
3082 const int j= scantable[i];
3087 for(i=0; i<=last; i++){
3088 const int j= scantable[i];
3089 const int perm_j= permutation[j];
3090 block[perm_j]= temp[j];
3094 void ff_mpeg_flush(AVCodecContext *avctx){
3096 MpegEncContext *s = avctx->priv_data;
3098 if(s==NULL || s->picture==NULL)
3101 for (i = 0; i < MAX_PICTURE_COUNT; i++)
3102 ff_mpeg_unref_picture(s, &s->picture[i]);
3103 s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
3105 ff_mpeg_unref_picture(s, &s->current_picture);
3106 ff_mpeg_unref_picture(s, &s->last_picture);
3107 ff_mpeg_unref_picture(s, &s->next_picture);
3109 s->mb_x= s->mb_y= 0;
3112 s->parse_context.state= -1;
3113 s->parse_context.frame_start_found= 0;
3114 s->parse_context.overread= 0;
3115 s->parse_context.overread_index= 0;
3116 s->parse_context.index= 0;
3117 s->parse_context.last_index= 0;
3118 s->bitstream_buffer_size=0;
3123 * set qscale and update qscale dependent variables.
3125 void ff_set_qscale(MpegEncContext * s, int qscale)
3129 else if (qscale > 31)
3133 s->chroma_qscale= s->chroma_qscale_table[qscale];
3135 s->y_dc_scale= s->y_dc_scale_table[ qscale ];
3136 s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
3139 void ff_MPV_report_decode_progress(MpegEncContext *s)
3141 if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
3142 ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);
3145 #if CONFIG_ERROR_RESILIENCE
3146 void ff_mpeg_set_erpic(ERPicture *dst, Picture *src)
3150 memset(dst, 0, sizeof(*dst));
3157 for (i = 0; i < 2; i++) {
3158 dst->motion_val[i] = src->motion_val[i];
3159 dst->ref_index[i] = src->ref_index[i];
3162 dst->mb_type = src->mb_type;
3163 dst->field_picture = src->field_picture;
3166 void ff_mpeg_er_frame_start(MpegEncContext *s)
3168 ERContext *er = &s->er;
3170 ff_mpeg_set_erpic(&er->cur_pic, s->current_picture_ptr);
3171 ff_mpeg_set_erpic(&er->next_pic, s->next_picture_ptr);
3172 ff_mpeg_set_erpic(&er->last_pic, s->last_picture_ptr);
3174 er->pp_time = s->pp_time;
3175 er->pb_time = s->pb_time;
3176 er->quarter_sample = s->quarter_sample;
3177 er->partitioned_frame = s->partitioned_frame;
3179 ff_er_frame_start(er);
3181 #endif /* CONFIG_ERROR_RESILIENCE */