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/motion_vector.h"
35 #include "libavutil/timer.h"
38 #include "h264chroma.h"
42 #include "mpegutils.h"
43 #include "mpegvideo.h"
44 #include "mpegvideodata.h"
52 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
53 int16_t *block, int n, int qscale)
55 int i, level, nCoeffs;
56 const uint16_t *quant_matrix;
58 nCoeffs= s->block_last_index[n];
60 block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
62 quant_matrix = s->intra_matrix;
63 for(i=1;i<=nCoeffs;i++) {
64 int j= s->intra_scantable.permutated[i];
69 level = (int)(level * qscale * quant_matrix[j]) >> 3;
70 level = (level - 1) | 1;
73 level = (int)(level * qscale * quant_matrix[j]) >> 3;
74 level = (level - 1) | 1;
81 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
82 int16_t *block, int n, int qscale)
84 int i, level, nCoeffs;
85 const uint16_t *quant_matrix;
87 nCoeffs= s->block_last_index[n];
89 quant_matrix = s->inter_matrix;
90 for(i=0; i<=nCoeffs; i++) {
91 int j= s->intra_scantable.permutated[i];
96 level = (((level << 1) + 1) * qscale *
97 ((int) (quant_matrix[j]))) >> 4;
98 level = (level - 1) | 1;
101 level = (((level << 1) + 1) * qscale *
102 ((int) (quant_matrix[j]))) >> 4;
103 level = (level - 1) | 1;
110 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
111 int16_t *block, int n, int qscale)
113 int i, level, nCoeffs;
114 const uint16_t *quant_matrix;
116 if(s->alternate_scan) nCoeffs= 63;
117 else nCoeffs= s->block_last_index[n];
119 block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
120 quant_matrix = s->intra_matrix;
121 for(i=1;i<=nCoeffs;i++) {
122 int j= s->intra_scantable.permutated[i];
127 level = (int)(level * qscale * quant_matrix[j]) >> 3;
130 level = (int)(level * qscale * quant_matrix[j]) >> 3;
137 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
138 int16_t *block, int n, int qscale)
140 int i, level, nCoeffs;
141 const uint16_t *quant_matrix;
144 if(s->alternate_scan) nCoeffs= 63;
145 else nCoeffs= s->block_last_index[n];
147 block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
149 quant_matrix = s->intra_matrix;
150 for(i=1;i<=nCoeffs;i++) {
151 int j= s->intra_scantable.permutated[i];
156 level = (int)(level * qscale * quant_matrix[j]) >> 3;
159 level = (int)(level * qscale * quant_matrix[j]) >> 3;
168 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
169 int16_t *block, int n, int qscale)
171 int i, level, nCoeffs;
172 const uint16_t *quant_matrix;
175 if(s->alternate_scan) nCoeffs= 63;
176 else nCoeffs= s->block_last_index[n];
178 quant_matrix = s->inter_matrix;
179 for(i=0; i<=nCoeffs; i++) {
180 int j= s->intra_scantable.permutated[i];
185 level = (((level << 1) + 1) * qscale *
186 ((int) (quant_matrix[j]))) >> 4;
189 level = (((level << 1) + 1) * qscale *
190 ((int) (quant_matrix[j]))) >> 4;
199 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
200 int16_t *block, int n, int qscale)
202 int i, level, qmul, qadd;
205 av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
210 block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
211 qadd = (qscale - 1) | 1;
218 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
220 for(i=1; i<=nCoeffs; i++) {
224 level = level * qmul - qadd;
226 level = level * qmul + qadd;
233 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
234 int16_t *block, int n, int qscale)
236 int i, level, qmul, qadd;
239 av_assert2(s->block_last_index[n]>=0);
241 qadd = (qscale - 1) | 1;
244 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
246 for(i=0; i<=nCoeffs; i++) {
250 level = level * qmul - qadd;
252 level = level * qmul + qadd;
259 static void mpeg_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
261 int mb_x, int mb_y, int mb_intra, int mb_skipped)
263 MpegEncContext *s = opaque;
266 s->mv_type = mv_type;
267 s->mb_intra = mb_intra;
268 s->mb_skipped = mb_skipped;
271 memcpy(s->mv, mv, sizeof(*mv));
273 ff_init_block_index(s);
274 ff_update_block_index(s);
276 s->bdsp.clear_blocks(s->block[0]);
278 s->dest[0] = s->current_picture.f->data[0] + (s->mb_y * 16 * s->linesize) + s->mb_x * 16;
279 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);
280 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);
283 av_log(s->avctx, AV_LOG_DEBUG,
284 "Interlaced error concealment is not fully implemented\n");
285 ff_mpv_decode_mb(s, s->block);
288 static void gray16(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
291 memset(dst + h*linesize, 128, 16);
294 static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
297 memset(dst + h*linesize, 128, 8);
300 /* init common dct for both encoder and decoder */
301 static av_cold int dct_init(MpegEncContext *s)
303 ff_blockdsp_init(&s->bdsp, s->avctx);
304 ff_h264chroma_init(&s->h264chroma, 8); //for lowres
305 ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
306 ff_mpegvideodsp_init(&s->mdsp);
307 ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
309 if (s->avctx->debug & FF_DEBUG_NOMC) {
311 for (i=0; i<4; i++) {
312 s->hdsp.avg_pixels_tab[0][i] = gray16;
313 s->hdsp.put_pixels_tab[0][i] = gray16;
314 s->hdsp.put_no_rnd_pixels_tab[0][i] = gray16;
316 s->hdsp.avg_pixels_tab[1][i] = gray8;
317 s->hdsp.put_pixels_tab[1][i] = gray8;
318 s->hdsp.put_no_rnd_pixels_tab[1][i] = gray8;
322 s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
323 s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
324 s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
325 s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
326 s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
327 if (s->avctx->flags & CODEC_FLAG_BITEXACT)
328 s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
329 s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
331 if (HAVE_INTRINSICS_NEON)
332 ff_mpv_common_init_neon(s);
335 ff_mpv_common_init_axp(s);
337 ff_mpv_common_init_arm(s);
339 ff_mpv_common_init_ppc(s);
341 ff_mpv_common_init_x86(s);
346 av_cold void ff_mpv_idct_init(MpegEncContext *s)
348 ff_idctdsp_init(&s->idsp, s->avctx);
350 /* load & permutate scantables
351 * note: only wmv uses different ones
353 if (s->alternate_scan) {
354 ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_alternate_vertical_scan);
355 ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_alternate_vertical_scan);
357 ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_zigzag_direct);
358 ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_zigzag_direct);
360 ff_init_scantable(s->idsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
361 ff_init_scantable(s->idsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
364 static int frame_size_alloc(MpegEncContext *s, int linesize)
366 int alloc_size = FFALIGN(FFABS(linesize) + 64, 32);
368 if (s->avctx->hwaccel || s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
372 av_log(s->avctx, AV_LOG_ERROR, "Image too small, temporary buffers cannot function\n");
373 return AVERROR_PATCHWELCOME;
376 // edge emu needs blocksize + filter length - 1
377 // (= 17x17 for halfpel / 21x21 for h264)
378 // VC1 computes luma and chroma simultaneously and needs 19X19 + 9x9
379 // at uvlinesize. It supports only YUV420 so 24x24 is enough
380 // linesize * interlaced * MBsize
381 // we also use this buffer for encoding in encode_mb_internal() needig an additional 32 lines
382 FF_ALLOCZ_ARRAY_OR_GOTO(s->avctx, s->edge_emu_buffer, alloc_size, 4 * 68,
385 FF_ALLOCZ_ARRAY_OR_GOTO(s->avctx, s->me.scratchpad, alloc_size, 4 * 16 * 2,
387 s->me.temp = s->me.scratchpad;
388 s->rd_scratchpad = s->me.scratchpad;
389 s->b_scratchpad = s->me.scratchpad;
390 s->obmc_scratchpad = s->me.scratchpad + 16;
394 av_freep(&s->edge_emu_buffer);
395 return AVERROR(ENOMEM);
399 * Allocate a frame buffer
401 static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
403 int edges_needed = av_codec_is_encoder(s->avctx->codec);
407 if (s->codec_id != AV_CODEC_ID_WMV3IMAGE &&
408 s->codec_id != AV_CODEC_ID_VC1IMAGE &&
409 s->codec_id != AV_CODEC_ID_MSS2) {
411 pic->f->width = s->avctx->width + 2 * EDGE_WIDTH;
412 pic->f->height = s->avctx->height + 2 * EDGE_WIDTH;
415 r = ff_thread_get_buffer(s->avctx, &pic->tf,
416 pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);
418 pic->f->width = s->avctx->width;
419 pic->f->height = s->avctx->height;
420 pic->f->format = s->avctx->pix_fmt;
421 r = avcodec_default_get_buffer2(s->avctx, pic->f, 0);
424 if (r < 0 || !pic->f->buf[0]) {
425 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %p)\n",
432 for (i = 0; pic->f->data[i]; i++) {
433 int offset = (EDGE_WIDTH >> (i ? s->chroma_y_shift : 0)) *
434 pic->f->linesize[i] +
435 (EDGE_WIDTH >> (i ? s->chroma_x_shift : 0));
436 pic->f->data[i] += offset;
438 pic->f->width = s->avctx->width;
439 pic->f->height = s->avctx->height;
442 if (s->avctx->hwaccel) {
443 assert(!pic->hwaccel_picture_private);
444 if (s->avctx->hwaccel->frame_priv_data_size) {
445 pic->hwaccel_priv_buf = av_buffer_allocz(s->avctx->hwaccel->frame_priv_data_size);
446 if (!pic->hwaccel_priv_buf) {
447 av_log(s->avctx, AV_LOG_ERROR, "alloc_frame_buffer() failed (hwaccel private data allocation)\n");
450 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
454 if (s->linesize && (s->linesize != pic->f->linesize[0] ||
455 s->uvlinesize != pic->f->linesize[1])) {
456 av_log(s->avctx, AV_LOG_ERROR,
457 "get_buffer() failed (stride changed)\n");
458 ff_mpeg_unref_picture(s->avctx, pic);
462 if (pic->f->linesize[1] != pic->f->linesize[2]) {
463 av_log(s->avctx, AV_LOG_ERROR,
464 "get_buffer() failed (uv stride mismatch)\n");
465 ff_mpeg_unref_picture(s->avctx, pic);
469 if (!s->edge_emu_buffer &&
470 (ret = frame_size_alloc(s, pic->f->linesize[0])) < 0) {
471 av_log(s->avctx, AV_LOG_ERROR,
472 "get_buffer() failed to allocate context scratch buffers.\n");
473 ff_mpeg_unref_picture(s->avctx, pic);
480 void ff_free_picture_tables(Picture *pic)
484 pic->alloc_mb_width =
485 pic->alloc_mb_height = 0;
487 av_buffer_unref(&pic->mb_var_buf);
488 av_buffer_unref(&pic->mc_mb_var_buf);
489 av_buffer_unref(&pic->mb_mean_buf);
490 av_buffer_unref(&pic->mbskip_table_buf);
491 av_buffer_unref(&pic->qscale_table_buf);
492 av_buffer_unref(&pic->mb_type_buf);
494 for (i = 0; i < 2; i++) {
495 av_buffer_unref(&pic->motion_val_buf[i]);
496 av_buffer_unref(&pic->ref_index_buf[i]);
500 static int alloc_picture_tables(MpegEncContext *s, Picture *pic)
502 const int big_mb_num = s->mb_stride * (s->mb_height + 1) + 1;
503 const int mb_array_size = s->mb_stride * s->mb_height;
504 const int b8_array_size = s->b8_stride * s->mb_height * 2;
508 pic->mbskip_table_buf = av_buffer_allocz(mb_array_size + 2);
509 pic->qscale_table_buf = av_buffer_allocz(big_mb_num + s->mb_stride);
510 pic->mb_type_buf = av_buffer_allocz((big_mb_num + s->mb_stride) *
512 if (!pic->mbskip_table_buf || !pic->qscale_table_buf || !pic->mb_type_buf)
513 return AVERROR(ENOMEM);
516 pic->mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t));
517 pic->mc_mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t));
518 pic->mb_mean_buf = av_buffer_allocz(mb_array_size);
519 if (!pic->mb_var_buf || !pic->mc_mb_var_buf || !pic->mb_mean_buf)
520 return AVERROR(ENOMEM);
523 if (s->out_format == FMT_H263 || s->encoding || s->avctx->debug_mv ||
524 (s->avctx->flags2 & CODEC_FLAG2_EXPORT_MVS)) {
525 int mv_size = 2 * (b8_array_size + 4) * sizeof(int16_t);
526 int ref_index_size = 4 * mb_array_size;
528 for (i = 0; mv_size && i < 2; i++) {
529 pic->motion_val_buf[i] = av_buffer_allocz(mv_size);
530 pic->ref_index_buf[i] = av_buffer_allocz(ref_index_size);
531 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
532 return AVERROR(ENOMEM);
536 pic->alloc_mb_width = s->mb_width;
537 pic->alloc_mb_height = s->mb_height;
542 static int make_tables_writable(Picture *pic)
545 #define MAKE_WRITABLE(table) \
548 (ret = av_buffer_make_writable(&pic->table)) < 0)\
552 MAKE_WRITABLE(mb_var_buf);
553 MAKE_WRITABLE(mc_mb_var_buf);
554 MAKE_WRITABLE(mb_mean_buf);
555 MAKE_WRITABLE(mbskip_table_buf);
556 MAKE_WRITABLE(qscale_table_buf);
557 MAKE_WRITABLE(mb_type_buf);
559 for (i = 0; i < 2; i++) {
560 MAKE_WRITABLE(motion_val_buf[i]);
561 MAKE_WRITABLE(ref_index_buf[i]);
568 * Allocate a Picture.
569 * The pixels are allocated/set by calling get_buffer() if shared = 0
571 int ff_alloc_picture(MpegEncContext *s, Picture *pic, int shared)
575 if (pic->qscale_table_buf)
576 if ( pic->alloc_mb_width != s->mb_width
577 || pic->alloc_mb_height != s->mb_height)
578 ff_free_picture_tables(pic);
581 av_assert0(pic->f->data[0]);
584 av_assert0(!pic->f->buf[0]);
586 if (alloc_frame_buffer(s, pic) < 0)
589 s->linesize = pic->f->linesize[0];
590 s->uvlinesize = pic->f->linesize[1];
593 if (!pic->qscale_table_buf)
594 ret = alloc_picture_tables(s, pic);
596 ret = make_tables_writable(pic);
601 pic->mb_var = (uint16_t*)pic->mb_var_buf->data;
602 pic->mc_mb_var = (uint16_t*)pic->mc_mb_var_buf->data;
603 pic->mb_mean = pic->mb_mean_buf->data;
606 pic->mbskip_table = pic->mbskip_table_buf->data;
607 pic->qscale_table = pic->qscale_table_buf->data + 2 * s->mb_stride + 1;
608 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * s->mb_stride + 1;
610 if (pic->motion_val_buf[0]) {
611 for (i = 0; i < 2; i++) {
612 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
613 pic->ref_index[i] = pic->ref_index_buf[i]->data;
619 av_log(s->avctx, AV_LOG_ERROR, "Error allocating a picture.\n");
620 ff_mpeg_unref_picture(s->avctx, pic);
621 ff_free_picture_tables(pic);
622 return AVERROR(ENOMEM);
626 * Deallocate a picture.
628 void ff_mpeg_unref_picture(AVCodecContext *avctx, Picture *pic)
630 int off = offsetof(Picture, mb_mean) + sizeof(pic->mb_mean);
633 /* WM Image / Screen codecs allocate internal buffers with different
634 * dimensions / colorspaces; ignore user-defined callbacks for these. */
635 if (avctx->codec->id != AV_CODEC_ID_WMV3IMAGE &&
636 avctx->codec->id != AV_CODEC_ID_VC1IMAGE &&
637 avctx->codec->id != AV_CODEC_ID_MSS2)
638 ff_thread_release_buffer(avctx, &pic->tf);
640 av_frame_unref(pic->f);
642 av_buffer_unref(&pic->hwaccel_priv_buf);
644 if (pic->needs_realloc)
645 ff_free_picture_tables(pic);
647 memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
650 static int update_picture_tables(Picture *dst, Picture *src)
654 #define UPDATE_TABLE(table)\
657 (!dst->table || dst->table->buffer != src->table->buffer)) {\
658 av_buffer_unref(&dst->table);\
659 dst->table = av_buffer_ref(src->table);\
661 ff_free_picture_tables(dst);\
662 return AVERROR(ENOMEM);\
667 UPDATE_TABLE(mb_var_buf);
668 UPDATE_TABLE(mc_mb_var_buf);
669 UPDATE_TABLE(mb_mean_buf);
670 UPDATE_TABLE(mbskip_table_buf);
671 UPDATE_TABLE(qscale_table_buf);
672 UPDATE_TABLE(mb_type_buf);
673 for (i = 0; i < 2; i++) {
674 UPDATE_TABLE(motion_val_buf[i]);
675 UPDATE_TABLE(ref_index_buf[i]);
678 dst->mb_var = src->mb_var;
679 dst->mc_mb_var = src->mc_mb_var;
680 dst->mb_mean = src->mb_mean;
681 dst->mbskip_table = src->mbskip_table;
682 dst->qscale_table = src->qscale_table;
683 dst->mb_type = src->mb_type;
684 for (i = 0; i < 2; i++) {
685 dst->motion_val[i] = src->motion_val[i];
686 dst->ref_index[i] = src->ref_index[i];
689 dst->alloc_mb_width = src->alloc_mb_width;
690 dst->alloc_mb_height = src->alloc_mb_height;
695 int ff_mpeg_ref_picture(AVCodecContext *avctx, Picture *dst, Picture *src)
699 av_assert0(!dst->f->buf[0]);
700 av_assert0(src->f->buf[0]);
704 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
708 ret = update_picture_tables(dst, src);
712 if (src->hwaccel_picture_private) {
713 dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
714 if (!dst->hwaccel_priv_buf)
716 dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
719 dst->field_picture = src->field_picture;
720 dst->mb_var_sum = src->mb_var_sum;
721 dst->mc_mb_var_sum = src->mc_mb_var_sum;
722 dst->b_frame_score = src->b_frame_score;
723 dst->needs_realloc = src->needs_realloc;
724 dst->reference = src->reference;
725 dst->shared = src->shared;
729 ff_mpeg_unref_picture(avctx, dst);
733 static int init_duplicate_context(MpegEncContext *s)
735 int y_size = s->b8_stride * (2 * s->mb_height + 1);
736 int c_size = s->mb_stride * (s->mb_height + 1);
737 int yc_size = y_size + 2 * c_size;
740 if (s->mb_height & 1)
741 yc_size += 2*s->b8_stride + 2*s->mb_stride;
748 s->obmc_scratchpad = NULL;
751 FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map,
752 ME_MAP_SIZE * sizeof(uint32_t), fail)
753 FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map,
754 ME_MAP_SIZE * sizeof(uint32_t), fail)
755 if (s->avctx->noise_reduction) {
756 FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum,
757 2 * 64 * sizeof(int), fail)
760 FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail)
761 s->block = s->blocks[0];
763 for (i = 0; i < 12; i++) {
764 s->pblocks[i] = &s->block[i];
766 if (s->avctx->codec_tag == AV_RL32("VCR2")) {
768 FFSWAP(void *, s->pblocks[4], s->pblocks[5]);
771 if (s->out_format == FMT_H263) {
773 FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base,
774 yc_size * sizeof(int16_t) * 16, fail);
775 s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
776 s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
777 s->ac_val[2] = s->ac_val[1] + c_size;
782 return -1; // free() through ff_mpv_common_end()
785 static void free_duplicate_context(MpegEncContext *s)
790 av_freep(&s->edge_emu_buffer);
791 av_freep(&s->me.scratchpad);
795 s->obmc_scratchpad = NULL;
797 av_freep(&s->dct_error_sum);
798 av_freep(&s->me.map);
799 av_freep(&s->me.score_map);
800 av_freep(&s->blocks);
801 av_freep(&s->ac_val_base);
805 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
807 #define COPY(a) bak->a = src->a
808 COPY(edge_emu_buffer);
813 COPY(obmc_scratchpad);
820 COPY(me.map_generation);
832 int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
836 // FIXME copy only needed parts
838 backup_duplicate_context(&bak, dst);
839 memcpy(dst, src, sizeof(MpegEncContext));
840 backup_duplicate_context(dst, &bak);
841 for (i = 0; i < 12; i++) {
842 dst->pblocks[i] = &dst->block[i];
844 if (dst->avctx->codec_tag == AV_RL32("VCR2")) {
846 FFSWAP(void *, dst->pblocks[4], dst->pblocks[5]);
848 if (!dst->edge_emu_buffer &&
849 (ret = frame_size_alloc(dst, dst->linesize)) < 0) {
850 av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context "
851 "scratch buffers.\n");
854 // STOP_TIMER("update_duplicate_context")
855 // about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads
859 int ff_mpeg_update_thread_context(AVCodecContext *dst,
860 const AVCodecContext *src)
863 MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
870 // FIXME can parameters change on I-frames?
871 // in that case dst may need a reinit
872 if (!s->context_initialized) {
874 memcpy(s, s1, sizeof(MpegEncContext));
877 s->bitstream_buffer = NULL;
878 s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
880 if (s1->context_initialized){
881 // s->picture_range_start += MAX_PICTURE_COUNT;
882 // s->picture_range_end += MAX_PICTURE_COUNT;
884 if((err = ff_mpv_common_init(s)) < 0){
885 memset(s, 0, sizeof(MpegEncContext));
892 if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
893 s->context_reinit = 0;
894 s->height = s1->height;
895 s->width = s1->width;
896 if ((ret = ff_mpv_common_frame_size_change(s)) < 0)
900 s->avctx->coded_height = s1->avctx->coded_height;
901 s->avctx->coded_width = s1->avctx->coded_width;
902 s->avctx->width = s1->avctx->width;
903 s->avctx->height = s1->avctx->height;
905 s->coded_picture_number = s1->coded_picture_number;
906 s->picture_number = s1->picture_number;
908 av_assert0(!s->picture || s->picture != s1->picture);
910 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
911 ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
912 if (s1->picture[i].f->buf[0] &&
913 (ret = ff_mpeg_ref_picture(s->avctx, &s->picture[i], &s1->picture[i])) < 0)
917 #define UPDATE_PICTURE(pic)\
919 ff_mpeg_unref_picture(s->avctx, &s->pic);\
920 if (s1->pic.f && s1->pic.f->buf[0])\
921 ret = ff_mpeg_ref_picture(s->avctx, &s->pic, &s1->pic);\
923 ret = update_picture_tables(&s->pic, &s1->pic);\
928 UPDATE_PICTURE(current_picture);
929 UPDATE_PICTURE(last_picture);
930 UPDATE_PICTURE(next_picture);
932 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
933 ((pic && pic >= old_ctx->picture && \
934 pic < old_ctx->picture + MAX_PICTURE_COUNT) ? \
935 &new_ctx->picture[pic - old_ctx->picture] : NULL)
937 s->last_picture_ptr = REBASE_PICTURE(s1->last_picture_ptr, s, s1);
938 s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
939 s->next_picture_ptr = REBASE_PICTURE(s1->next_picture_ptr, s, s1);
941 // Error/bug resilience
942 s->next_p_frame_damaged = s1->next_p_frame_damaged;
943 s->workaround_bugs = s1->workaround_bugs;
944 s->padding_bug_score = s1->padding_bug_score;
947 memcpy(&s->last_time_base, &s1->last_time_base,
948 (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) -
949 (char *) &s1->last_time_base);
952 s->max_b_frames = s1->max_b_frames;
953 s->low_delay = s1->low_delay;
954 s->droppable = s1->droppable;
956 // DivX handling (doesn't work)
957 s->divx_packed = s1->divx_packed;
959 if (s1->bitstream_buffer) {
960 if (s1->bitstream_buffer_size +
961 FF_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size) {
962 av_fast_malloc(&s->bitstream_buffer,
963 &s->allocated_bitstream_buffer_size,
964 s1->allocated_bitstream_buffer_size);
965 if (!s->bitstream_buffer) {
966 s->bitstream_buffer_size = 0;
967 return AVERROR(ENOMEM);
970 s->bitstream_buffer_size = s1->bitstream_buffer_size;
971 memcpy(s->bitstream_buffer, s1->bitstream_buffer,
972 s1->bitstream_buffer_size);
973 memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
974 FF_INPUT_BUFFER_PADDING_SIZE);
977 // linesize dependend scratch buffer allocation
978 if (!s->edge_emu_buffer)
980 if (frame_size_alloc(s, s1->linesize) < 0) {
981 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
982 "scratch buffers.\n");
983 return AVERROR(ENOMEM);
986 av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
987 "be allocated due to unknown size.\n");
990 // MPEG2/interlacing info
991 memcpy(&s->progressive_sequence, &s1->progressive_sequence,
992 (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
994 if (!s1->first_field) {
995 s->last_pict_type = s1->pict_type;
996 if (s1->current_picture_ptr)
997 s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f->quality;
1004 * Set the given MpegEncContext to common defaults
1005 * (same for encoding and decoding).
1006 * The changed fields will not depend upon the
1007 * prior state of the MpegEncContext.
1009 void ff_mpv_common_defaults(MpegEncContext *s)
1011 s->y_dc_scale_table =
1012 s->c_dc_scale_table = ff_mpeg1_dc_scale_table;
1013 s->chroma_qscale_table = ff_default_chroma_qscale_table;
1014 s->progressive_frame = 1;
1015 s->progressive_sequence = 1;
1016 s->picture_structure = PICT_FRAME;
1018 s->coded_picture_number = 0;
1019 s->picture_number = 0;
1024 s->slice_context_count = 1;
1028 * Set the given MpegEncContext to defaults for decoding.
1029 * the changed fields will not depend upon
1030 * the prior state of the MpegEncContext.
1032 void ff_mpv_decode_defaults(MpegEncContext *s)
1034 ff_mpv_common_defaults(s);
1037 void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx)
1040 s->width = avctx->coded_width;
1041 s->height = avctx->coded_height;
1042 s->codec_id = avctx->codec->id;
1043 s->workaround_bugs = avctx->workaround_bugs;
1045 /* convert fourcc to upper case */
1046 s->codec_tag = avpriv_toupper4(avctx->codec_tag);
1049 static int init_er(MpegEncContext *s)
1051 ERContext *er = &s->er;
1052 int mb_array_size = s->mb_height * s->mb_stride;
1055 er->avctx = s->avctx;
1057 er->mb_index2xy = s->mb_index2xy;
1058 er->mb_num = s->mb_num;
1059 er->mb_width = s->mb_width;
1060 er->mb_height = s->mb_height;
1061 er->mb_stride = s->mb_stride;
1062 er->b8_stride = s->b8_stride;
1064 er->er_temp_buffer = av_malloc(s->mb_height * s->mb_stride);
1065 er->error_status_table = av_mallocz(mb_array_size);
1066 if (!er->er_temp_buffer || !er->error_status_table)
1069 er->mbskip_table = s->mbskip_table;
1070 er->mbintra_table = s->mbintra_table;
1072 for (i = 0; i < FF_ARRAY_ELEMS(s->dc_val); i++)
1073 er->dc_val[i] = s->dc_val[i];
1075 er->decode_mb = mpeg_er_decode_mb;
1080 av_freep(&er->er_temp_buffer);
1081 av_freep(&er->error_status_table);
1082 return AVERROR(ENOMEM);
1086 * Initialize and allocates MpegEncContext fields dependent on the resolution.
1088 static int init_context_frame(MpegEncContext *s)
1090 int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
1092 s->mb_width = (s->width + 15) / 16;
1093 s->mb_stride = s->mb_width + 1;
1094 s->b8_stride = s->mb_width * 2 + 1;
1095 mb_array_size = s->mb_height * s->mb_stride;
1096 mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
1098 /* set default edge pos, will be overridden
1099 * in decode_header if needed */
1100 s->h_edge_pos = s->mb_width * 16;
1101 s->v_edge_pos = s->mb_height * 16;
1103 s->mb_num = s->mb_width * s->mb_height;
1108 s->block_wrap[3] = s->b8_stride;
1110 s->block_wrap[5] = s->mb_stride;
1112 y_size = s->b8_stride * (2 * s->mb_height + 1);
1113 c_size = s->mb_stride * (s->mb_height + 1);
1114 yc_size = y_size + 2 * c_size;
1116 if (s->mb_height & 1)
1117 yc_size += 2*s->b8_stride + 2*s->mb_stride;
1119 FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int), fail); // error ressilience code looks cleaner with this
1120 for (y = 0; y < s->mb_height; y++)
1121 for (x = 0; x < s->mb_width; x++)
1122 s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
1124 s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
1127 /* Allocate MV tables */
1128 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1129 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1130 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1131 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1132 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1133 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1134 s->p_mv_table = s->p_mv_table_base + s->mb_stride + 1;
1135 s->b_forw_mv_table = s->b_forw_mv_table_base + s->mb_stride + 1;
1136 s->b_back_mv_table = s->b_back_mv_table_base + s->mb_stride + 1;
1137 s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
1138 s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
1139 s->b_direct_mv_table = s->b_direct_mv_table_base + s->mb_stride + 1;
1141 /* Allocate MB type table */
1142 FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding
1144 FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
1146 FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
1147 mb_array_size * sizeof(float), fail);
1148 FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab,
1149 mb_array_size * sizeof(float), fail);
1153 if (s->codec_id == AV_CODEC_ID_MPEG4 ||
1154 (s->avctx->flags & CODEC_FLAG_INTERLACED_ME)) {
1155 /* interlaced direct mode decoding tables */
1156 for (i = 0; i < 2; i++) {
1158 for (j = 0; j < 2; j++) {
1159 for (k = 0; k < 2; k++) {
1160 FF_ALLOCZ_OR_GOTO(s->avctx,
1161 s->b_field_mv_table_base[i][j][k],
1162 mv_table_size * 2 * sizeof(int16_t),
1164 s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
1167 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
1168 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
1169 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
1171 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
1174 if (s->out_format == FMT_H263) {
1176 FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride, fail);
1177 s->coded_block = s->coded_block_base + s->b8_stride + 1;
1179 /* cbp, ac_pred, pred_dir */
1180 FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table , mb_array_size * sizeof(uint8_t), fail);
1181 FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail);
1184 if (s->h263_pred || s->h263_plus || !s->encoding) {
1186 // MN: we need these for error resilience of intra-frames
1187 FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
1188 s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
1189 s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
1190 s->dc_val[2] = s->dc_val[1] + c_size;
1191 for (i = 0; i < yc_size; i++)
1192 s->dc_val_base[i] = 1024;
1195 /* which mb is a intra block */
1196 FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
1197 memset(s->mbintra_table, 1, mb_array_size);
1199 /* init macroblock skip table */
1200 FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
1201 // Note the + 1 is for a quicker mpeg4 slice_end detection
1205 return AVERROR(ENOMEM);
1209 * init common structure for both encoder and decoder.
1210 * this assumes that some variables like width/height are already set
1212 av_cold int ff_mpv_common_init(MpegEncContext *s)
1215 int nb_slices = (HAVE_THREADS &&
1216 s->avctx->active_thread_type & FF_THREAD_SLICE) ?
1217 s->avctx->thread_count : 1;
1219 if (s->encoding && s->avctx->slices)
1220 nb_slices = s->avctx->slices;
1222 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1223 s->mb_height = (s->height + 31) / 32 * 2;
1225 s->mb_height = (s->height + 15) / 16;
1227 if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
1228 av_log(s->avctx, AV_LOG_ERROR,
1229 "decoding to AV_PIX_FMT_NONE is not supported.\n");
1233 if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
1236 max_slices = FFMIN(MAX_THREADS, s->mb_height);
1238 max_slices = MAX_THREADS;
1239 av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
1240 " reducing to %d\n", nb_slices, max_slices);
1241 nb_slices = max_slices;
1244 if ((s->width || s->height) &&
1245 av_image_check_size(s->width, s->height, 0, s->avctx))
1250 /* set chroma shifts */
1251 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,
1253 &s->chroma_y_shift);
1256 FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
1257 MAX_PICTURE_COUNT * sizeof(Picture), fail);
1258 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1259 s->picture[i].f = av_frame_alloc();
1260 if (!s->picture[i].f)
1263 memset(&s->next_picture, 0, sizeof(s->next_picture));
1264 memset(&s->last_picture, 0, sizeof(s->last_picture));
1265 memset(&s->current_picture, 0, sizeof(s->current_picture));
1266 memset(&s->new_picture, 0, sizeof(s->new_picture));
1267 s->next_picture.f = av_frame_alloc();
1268 if (!s->next_picture.f)
1270 s->last_picture.f = av_frame_alloc();
1271 if (!s->last_picture.f)
1273 s->current_picture.f = av_frame_alloc();
1274 if (!s->current_picture.f)
1276 s->new_picture.f = av_frame_alloc();
1277 if (!s->new_picture.f)
1280 if (init_context_frame(s))
1283 s->parse_context.state = -1;
1285 s->context_initialized = 1;
1286 memset(s->thread_context, 0, sizeof(s->thread_context));
1287 s->thread_context[0] = s;
1289 // if (s->width && s->height) {
1290 if (nb_slices > 1) {
1291 for (i = 0; i < nb_slices; i++) {
1293 s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
1294 memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
1296 if (init_duplicate_context(s->thread_context[i]) < 0)
1298 s->thread_context[i]->start_mb_y =
1299 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1300 s->thread_context[i]->end_mb_y =
1301 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1304 if (init_duplicate_context(s) < 0)
1307 s->end_mb_y = s->mb_height;
1309 s->slice_context_count = nb_slices;
1314 ff_mpv_common_end(s);
1319 * Frees and resets MpegEncContext fields depending on the resolution.
1320 * Is used during resolution changes to avoid a full reinitialization of the
1323 static void free_context_frame(MpegEncContext *s)
1327 av_freep(&s->mb_type);
1328 av_freep(&s->p_mv_table_base);
1329 av_freep(&s->b_forw_mv_table_base);
1330 av_freep(&s->b_back_mv_table_base);
1331 av_freep(&s->b_bidir_forw_mv_table_base);
1332 av_freep(&s->b_bidir_back_mv_table_base);
1333 av_freep(&s->b_direct_mv_table_base);
1334 s->p_mv_table = NULL;
1335 s->b_forw_mv_table = NULL;
1336 s->b_back_mv_table = NULL;
1337 s->b_bidir_forw_mv_table = NULL;
1338 s->b_bidir_back_mv_table = NULL;
1339 s->b_direct_mv_table = NULL;
1340 for (i = 0; i < 2; i++) {
1341 for (j = 0; j < 2; j++) {
1342 for (k = 0; k < 2; k++) {
1343 av_freep(&s->b_field_mv_table_base[i][j][k]);
1344 s->b_field_mv_table[i][j][k] = NULL;
1346 av_freep(&s->b_field_select_table[i][j]);
1347 av_freep(&s->p_field_mv_table_base[i][j]);
1348 s->p_field_mv_table[i][j] = NULL;
1350 av_freep(&s->p_field_select_table[i]);
1353 av_freep(&s->dc_val_base);
1354 av_freep(&s->coded_block_base);
1355 av_freep(&s->mbintra_table);
1356 av_freep(&s->cbp_table);
1357 av_freep(&s->pred_dir_table);
1359 av_freep(&s->mbskip_table);
1361 av_freep(&s->er.error_status_table);
1362 av_freep(&s->er.er_temp_buffer);
1363 av_freep(&s->mb_index2xy);
1364 av_freep(&s->lambda_table);
1366 av_freep(&s->cplx_tab);
1367 av_freep(&s->bits_tab);
1369 s->linesize = s->uvlinesize = 0;
1372 int ff_mpv_common_frame_size_change(MpegEncContext *s)
1376 if (!s->context_initialized)
1377 return AVERROR(EINVAL);
1379 if (s->slice_context_count > 1) {
1380 for (i = 0; i < s->slice_context_count; i++) {
1381 free_duplicate_context(s->thread_context[i]);
1383 for (i = 1; i < s->slice_context_count; i++) {
1384 av_freep(&s->thread_context[i]);
1387 free_duplicate_context(s);
1389 free_context_frame(s);
1392 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1393 s->picture[i].needs_realloc = 1;
1396 s->last_picture_ptr =
1397 s->next_picture_ptr =
1398 s->current_picture_ptr = NULL;
1401 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1402 s->mb_height = (s->height + 31) / 32 * 2;
1404 s->mb_height = (s->height + 15) / 16;
1406 if ((s->width || s->height) &&
1407 (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0)
1410 if ((err = init_context_frame(s)))
1413 memset(s->thread_context, 0, sizeof(s->thread_context));
1414 s->thread_context[0] = s;
1416 if (s->width && s->height) {
1417 int nb_slices = s->slice_context_count;
1418 if (nb_slices > 1) {
1419 for (i = 0; i < nb_slices; i++) {
1421 s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
1422 memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
1424 if ((err = init_duplicate_context(s->thread_context[i])) < 0)
1426 s->thread_context[i]->start_mb_y =
1427 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1428 s->thread_context[i]->end_mb_y =
1429 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1432 err = init_duplicate_context(s);
1436 s->end_mb_y = s->mb_height;
1438 s->slice_context_count = nb_slices;
1443 ff_mpv_common_end(s);
1447 /* init common structure for both encoder and decoder */
1448 void ff_mpv_common_end(MpegEncContext *s)
1452 if (s->slice_context_count > 1) {
1453 for (i = 0; i < s->slice_context_count; i++) {
1454 free_duplicate_context(s->thread_context[i]);
1456 for (i = 1; i < s->slice_context_count; i++) {
1457 av_freep(&s->thread_context[i]);
1459 s->slice_context_count = 1;
1460 } else free_duplicate_context(s);
1462 av_freep(&s->parse_context.buffer);
1463 s->parse_context.buffer_size = 0;
1465 av_freep(&s->bitstream_buffer);
1466 s->allocated_bitstream_buffer_size = 0;
1469 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1470 ff_free_picture_tables(&s->picture[i]);
1471 ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1472 av_frame_free(&s->picture[i].f);
1475 av_freep(&s->picture);
1476 ff_free_picture_tables(&s->last_picture);
1477 ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1478 av_frame_free(&s->last_picture.f);
1479 ff_free_picture_tables(&s->current_picture);
1480 ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1481 av_frame_free(&s->current_picture.f);
1482 ff_free_picture_tables(&s->next_picture);
1483 ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1484 av_frame_free(&s->next_picture.f);
1485 ff_free_picture_tables(&s->new_picture);
1486 ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1487 av_frame_free(&s->new_picture.f);
1489 free_context_frame(s);
1491 s->context_initialized = 0;
1492 s->last_picture_ptr =
1493 s->next_picture_ptr =
1494 s->current_picture_ptr = NULL;
1495 s->linesize = s->uvlinesize = 0;
1498 static void release_unused_pictures(AVCodecContext *avctx, Picture *picture)
1502 /* release non reference frames */
1503 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1504 if (!picture[i].reference)
1505 ff_mpeg_unref_picture(avctx, &picture[i]);
1509 static inline int pic_is_unused(Picture *pic)
1511 if (!pic->f->buf[0])
1513 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
1518 static int find_unused_picture(AVCodecContext *avctx, Picture *picture, int shared)
1523 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1524 if (!picture[i].f->buf[0])
1528 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1529 if (pic_is_unused(&picture[i]))
1534 av_log(avctx, AV_LOG_FATAL,
1535 "Internal error, picture buffer overflow\n");
1536 /* We could return -1, but the codec would crash trying to draw into a
1537 * non-existing frame anyway. This is safer than waiting for a random crash.
1538 * Also the return of this is never useful, an encoder must only allocate
1539 * as much as allowed in the specification. This has no relationship to how
1540 * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
1541 * enough for such valid streams).
1542 * Plus, a decoder has to check stream validity and remove frames if too
1543 * many reference frames are around. Waiting for "OOM" is not correct at
1544 * all. Similarly, missing reference frames have to be replaced by
1545 * interpolated/MC frames, anything else is a bug in the codec ...
1551 int ff_find_unused_picture(AVCodecContext *avctx, Picture *picture, int shared)
1553 int ret = find_unused_picture(avctx, picture, shared);
1555 if (ret >= 0 && ret < MAX_PICTURE_COUNT) {
1556 if (picture[ret].needs_realloc) {
1557 picture[ret].needs_realloc = 0;
1558 ff_free_picture_tables(&picture[ret]);
1559 ff_mpeg_unref_picture(avctx, &picture[ret]);
1565 static void gray_frame(AVFrame *frame)
1567 int i, h_chroma_shift, v_chroma_shift;
1569 av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift);
1571 for(i=0; i<frame->height; i++)
1572 memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width);
1573 for(i=0; i<FF_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) {
1574 memset(frame->data[1] + frame->linesize[1]*i,
1575 0x80, FF_CEIL_RSHIFT(frame->width, h_chroma_shift));
1576 memset(frame->data[2] + frame->linesize[2]*i,
1577 0x80, FF_CEIL_RSHIFT(frame->width, h_chroma_shift));
1582 * generic function called after decoding
1583 * the header and before a frame is decoded.
1585 int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1591 if (!ff_thread_can_start_frame(avctx)) {
1592 av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1596 /* mark & release old frames */
1597 if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1598 s->last_picture_ptr != s->next_picture_ptr &&
1599 s->last_picture_ptr->f->buf[0]) {
1600 ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1603 /* release forgotten pictures */
1604 /* if (mpeg124/h263) */
1605 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1606 if (&s->picture[i] != s->last_picture_ptr &&
1607 &s->picture[i] != s->next_picture_ptr &&
1608 s->picture[i].reference && !s->picture[i].needs_realloc) {
1609 if (!(avctx->active_thread_type & FF_THREAD_FRAME))
1610 av_log(avctx, AV_LOG_ERROR,
1611 "releasing zombie picture\n");
1612 ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1616 ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1618 release_unused_pictures(s->avctx, s->picture);
1620 if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) {
1621 // we already have a unused image
1622 // (maybe it was set before reading the header)
1623 pic = s->current_picture_ptr;
1625 i = ff_find_unused_picture(s->avctx, s->picture, 0);
1627 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1630 pic = &s->picture[i];
1634 if (!s->droppable) {
1635 if (s->pict_type != AV_PICTURE_TYPE_B)
1639 pic->f->coded_picture_number = s->coded_picture_number++;
1641 if (ff_alloc_picture(s, pic, 0) < 0)
1644 s->current_picture_ptr = pic;
1645 // FIXME use only the vars from current_pic
1646 s->current_picture_ptr->f->top_field_first = s->top_field_first;
1647 if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
1648 s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1649 if (s->picture_structure != PICT_FRAME)
1650 s->current_picture_ptr->f->top_field_first =
1651 (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
1653 s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame &&
1654 !s->progressive_sequence;
1655 s->current_picture_ptr->field_picture = s->picture_structure != PICT_FRAME;
1657 s->current_picture_ptr->f->pict_type = s->pict_type;
1658 // if (s->avctx->flags && CODEC_FLAG_QSCALE)
1659 // s->current_picture_ptr->quality = s->new_picture_ptr->quality;
1660 s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1662 if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1663 s->current_picture_ptr)) < 0)
1666 if (s->pict_type != AV_PICTURE_TYPE_B) {
1667 s->last_picture_ptr = s->next_picture_ptr;
1669 s->next_picture_ptr = s->current_picture_ptr;
1671 ff_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
1672 s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1673 s->last_picture_ptr ? s->last_picture_ptr->f->data[0] : NULL,
1674 s->next_picture_ptr ? s->next_picture_ptr->f->data[0] : NULL,
1675 s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL,
1676 s->pict_type, s->droppable);
1678 if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) &&
1679 (s->pict_type != AV_PICTURE_TYPE_I ||
1680 s->picture_structure != PICT_FRAME)) {
1681 int h_chroma_shift, v_chroma_shift;
1682 av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1683 &h_chroma_shift, &v_chroma_shift);
1684 if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
1685 av_log(avctx, AV_LOG_DEBUG,
1686 "allocating dummy last picture for B frame\n");
1687 else if (s->pict_type != AV_PICTURE_TYPE_I)
1688 av_log(avctx, AV_LOG_ERROR,
1689 "warning: first frame is no keyframe\n");
1690 else if (s->picture_structure != PICT_FRAME)
1691 av_log(avctx, AV_LOG_DEBUG,
1692 "allocate dummy last picture for field based first keyframe\n");
1694 /* Allocate a dummy frame */
1695 i = ff_find_unused_picture(s->avctx, s->picture, 0);
1697 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1700 s->last_picture_ptr = &s->picture[i];
1702 s->last_picture_ptr->reference = 3;
1703 s->last_picture_ptr->f->key_frame = 0;
1704 s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1706 if (ff_alloc_picture(s, s->last_picture_ptr, 0) < 0) {
1707 s->last_picture_ptr = NULL;
1711 if (!avctx->hwaccel && !(avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)) {
1712 for(i=0; i<avctx->height; i++)
1713 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
1714 0x80, avctx->width);
1715 if (s->last_picture_ptr->f->data[2]) {
1716 for(i=0; i<FF_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
1717 memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
1718 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1719 memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
1720 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1724 if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
1725 for(i=0; i<avctx->height; i++)
1726 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
1730 ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
1731 ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
1733 if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) &&
1734 s->pict_type == AV_PICTURE_TYPE_B) {
1735 /* Allocate a dummy frame */
1736 i = ff_find_unused_picture(s->avctx, s->picture, 0);
1738 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1741 s->next_picture_ptr = &s->picture[i];
1743 s->next_picture_ptr->reference = 3;
1744 s->next_picture_ptr->f->key_frame = 0;
1745 s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1747 if (ff_alloc_picture(s, s->next_picture_ptr, 0) < 0) {
1748 s->next_picture_ptr = NULL;
1751 ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
1752 ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
1755 #if 0 // BUFREF-FIXME
1756 memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
1757 memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
1759 if (s->last_picture_ptr) {
1760 ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1761 if (s->last_picture_ptr->f->buf[0] &&
1762 (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1763 s->last_picture_ptr)) < 0)
1766 if (s->next_picture_ptr) {
1767 ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1768 if (s->next_picture_ptr->f->buf[0] &&
1769 (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1770 s->next_picture_ptr)) < 0)
1774 av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
1775 s->last_picture_ptr->f->buf[0]));
1777 if (s->picture_structure!= PICT_FRAME) {
1779 for (i = 0; i < 4; i++) {
1780 if (s->picture_structure == PICT_BOTTOM_FIELD) {
1781 s->current_picture.f->data[i] +=
1782 s->current_picture.f->linesize[i];
1784 s->current_picture.f->linesize[i] *= 2;
1785 s->last_picture.f->linesize[i] *= 2;
1786 s->next_picture.f->linesize[i] *= 2;
1790 /* set dequantizer, we can't do it during init as
1791 * it might change for mpeg4 and we can't do it in the header
1792 * decode as init is not called for mpeg4 there yet */
1793 if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1794 s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1795 s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1796 } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1797 s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1798 s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1800 s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1801 s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1804 if (s->avctx->debug & FF_DEBUG_NOMC) {
1805 gray_frame(s->current_picture_ptr->f);
1811 /* called after a frame has been decoded. */
1812 void ff_mpv_frame_end(MpegEncContext *s)
1816 if (s->current_picture.reference)
1817 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1822 static int clip_line(int *sx, int *sy, int *ex, int *ey, int maxx)
1825 return clip_line(ex, ey, sx, sy, maxx);
1830 *sy = *ey + (*sy - *ey) * (int64_t)*ex / (*ex - *sx);
1837 *ey = *sy + (*ey - *sy) * (int64_t)(maxx - *sx) / (*ex - *sx);
1845 * Draw a line from (ex, ey) -> (sx, sy).
1846 * @param w width of the image
1847 * @param h height of the image
1848 * @param stride stride/linesize of the image
1849 * @param color color of the arrow
1851 static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey,
1852 int w, int h, int stride, int color)
1856 if (clip_line(&sx, &sy, &ex, &ey, w - 1))
1858 if (clip_line(&sy, &sx, &ey, &ex, h - 1))
1861 sx = av_clip(sx, 0, w - 1);
1862 sy = av_clip(sy, 0, h - 1);
1863 ex = av_clip(ex, 0, w - 1);
1864 ey = av_clip(ey, 0, h - 1);
1866 buf[sy * stride + sx] += color;
1868 if (FFABS(ex - sx) > FFABS(ey - sy)) {
1870 FFSWAP(int, sx, ex);
1871 FFSWAP(int, sy, ey);
1873 buf += sx + sy * stride;
1875 f = ((ey - sy) << 16) / ex;
1876 for (x = 0; x <= ex; x++) {
1878 fr = (x * f) & 0xFFFF;
1879 buf[y * stride + x] += (color * (0x10000 - fr)) >> 16;
1880 if(fr) buf[(y + 1) * stride + x] += (color * fr ) >> 16;
1884 FFSWAP(int, sx, ex);
1885 FFSWAP(int, sy, ey);
1887 buf += sx + sy * stride;
1890 f = ((ex - sx) << 16) / ey;
1893 for(y= 0; y <= ey; y++){
1895 fr = (y*f) & 0xFFFF;
1896 buf[y * stride + x] += (color * (0x10000 - fr)) >> 16;
1897 if(fr) buf[y * stride + x + 1] += (color * fr ) >> 16;
1903 * Draw an arrow from (ex, ey) -> (sx, sy).
1904 * @param w width of the image
1905 * @param h height of the image
1906 * @param stride stride/linesize of the image
1907 * @param color color of the arrow
1909 static void draw_arrow(uint8_t *buf, int sx, int sy, int ex,
1910 int ey, int w, int h, int stride, int color, int tail, int direction)
1915 FFSWAP(int, sx, ex);
1916 FFSWAP(int, sy, ey);
1919 sx = av_clip(sx, -100, w + 100);
1920 sy = av_clip(sy, -100, h + 100);
1921 ex = av_clip(ex, -100, w + 100);
1922 ey = av_clip(ey, -100, h + 100);
1927 if (dx * dx + dy * dy > 3 * 3) {
1930 int length = ff_sqrt((rx * rx + ry * ry) << 8);
1932 // FIXME subpixel accuracy
1933 rx = ROUNDED_DIV(rx * 3 << 4, length);
1934 ry = ROUNDED_DIV(ry * 3 << 4, length);
1941 draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
1942 draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
1944 draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
1948 static int add_mb(AVMotionVector *mb, uint32_t mb_type,
1949 int dst_x, int dst_y,
1950 int src_x, int src_y,
1953 mb->w = IS_8X8(mb_type) || IS_8X16(mb_type) ? 8 : 16;
1954 mb->h = IS_8X8(mb_type) || IS_16X8(mb_type) ? 8 : 16;
1959 mb->source = direction ? 1 : -1;
1960 mb->flags = 0; // XXX: does mb_type contain extra information that could be exported here?
1965 * Print debugging info for the given picture.
1967 void ff_print_debug_info2(AVCodecContext *avctx, AVFrame *pict, uint8_t *mbskip_table,
1968 uint32_t *mbtype_table, int8_t *qscale_table, int16_t (*motion_val[2])[2],
1970 int mb_width, int mb_height, int mb_stride, int quarter_sample)
1972 if ((avctx->flags2 & CODEC_FLAG2_EXPORT_MVS) && mbtype_table && motion_val[0]) {
1973 const int shift = 1 + quarter_sample;
1974 const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1;
1975 const int mv_stride = (mb_width << mv_sample_log2) +
1976 (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1);
1977 int mb_x, mb_y, mbcount = 0;
1979 /* size is width * height * 2 * 4 where 2 is for directions and 4 is
1980 * for the maximum number of MB (4 MB in case of IS_8x8) */
1981 AVMotionVector *mvs = av_malloc_array(mb_width * mb_height, 2 * 4 * sizeof(AVMotionVector));
1985 for (mb_y = 0; mb_y < mb_height; mb_y++) {
1986 for (mb_x = 0; mb_x < mb_width; mb_x++) {
1987 int i, direction, mb_type = mbtype_table[mb_x + mb_y * mb_stride];
1988 for (direction = 0; direction < 2; direction++) {
1989 if (!USES_LIST(mb_type, direction))
1991 if (IS_8X8(mb_type)) {
1992 for (i = 0; i < 4; i++) {
1993 int sx = mb_x * 16 + 4 + 8 * (i & 1);
1994 int sy = mb_y * 16 + 4 + 8 * (i >> 1);
1995 int xy = (mb_x * 2 + (i & 1) +
1996 (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
1997 int mx = (motion_val[direction][xy][0] >> shift) + sx;
1998 int my = (motion_val[direction][xy][1] >> shift) + sy;
1999 mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx, my, direction);
2001 } else if (IS_16X8(mb_type)) {
2002 for (i = 0; i < 2; i++) {
2003 int sx = mb_x * 16 + 8;
2004 int sy = mb_y * 16 + 4 + 8 * i;
2005 int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
2006 int mx = (motion_val[direction][xy][0] >> shift);
2007 int my = (motion_val[direction][xy][1] >> shift);
2009 if (IS_INTERLACED(mb_type))
2012 mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx + sx, my + sy, direction);
2014 } else if (IS_8X16(mb_type)) {
2015 for (i = 0; i < 2; i++) {
2016 int sx = mb_x * 16 + 4 + 8 * i;
2017 int sy = mb_y * 16 + 8;
2018 int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
2019 int mx = motion_val[direction][xy][0] >> shift;
2020 int my = motion_val[direction][xy][1] >> shift;
2022 if (IS_INTERLACED(mb_type))
2025 mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx + sx, my + sy, direction);
2028 int sx = mb_x * 16 + 8;
2029 int sy = mb_y * 16 + 8;
2030 int xy = (mb_x + mb_y * mv_stride) << mv_sample_log2;
2031 int mx = (motion_val[direction][xy][0]>>shift) + sx;
2032 int my = (motion_val[direction][xy][1]>>shift) + sy;
2033 mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx, my, direction);
2040 AVFrameSideData *sd;
2042 av_log(avctx, AV_LOG_DEBUG, "Adding %d MVs info to frame %d\n", mbcount, avctx->frame_number);
2043 sd = av_frame_new_side_data(pict, AV_FRAME_DATA_MOTION_VECTORS, mbcount * sizeof(AVMotionVector));
2048 memcpy(sd->data, mvs, mbcount * sizeof(AVMotionVector));
2054 /* TODO: export all the following to make them accessible for users (and filters) */
2055 if (avctx->hwaccel || !mbtype_table
2056 || (avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU))
2060 if (avctx->debug & (FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)) {
2063 av_log(avctx, AV_LOG_DEBUG, "New frame, type: %c\n",
2064 av_get_picture_type_char(pict->pict_type));
2065 for (y = 0; y < mb_height; y++) {
2066 for (x = 0; x < mb_width; x++) {
2067 if (avctx->debug & FF_DEBUG_SKIP) {
2068 int count = mbskip_table ? mbskip_table[x + y * mb_stride] : 0;
2071 av_log(avctx, AV_LOG_DEBUG, "%1d", count);
2073 if (avctx->debug & FF_DEBUG_QP) {
2074 av_log(avctx, AV_LOG_DEBUG, "%2d",
2075 qscale_table[x + y * mb_stride]);
2077 if (avctx->debug & FF_DEBUG_MB_TYPE) {
2078 int mb_type = mbtype_table[x + y * mb_stride];
2079 // Type & MV direction
2080 if (IS_PCM(mb_type))
2081 av_log(avctx, AV_LOG_DEBUG, "P");
2082 else if (IS_INTRA(mb_type) && IS_ACPRED(mb_type))
2083 av_log(avctx, AV_LOG_DEBUG, "A");
2084 else if (IS_INTRA4x4(mb_type))
2085 av_log(avctx, AV_LOG_DEBUG, "i");
2086 else if (IS_INTRA16x16(mb_type))
2087 av_log(avctx, AV_LOG_DEBUG, "I");
2088 else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type))
2089 av_log(avctx, AV_LOG_DEBUG, "d");
2090 else if (IS_DIRECT(mb_type))
2091 av_log(avctx, AV_LOG_DEBUG, "D");
2092 else if (IS_GMC(mb_type) && IS_SKIP(mb_type))
2093 av_log(avctx, AV_LOG_DEBUG, "g");
2094 else if (IS_GMC(mb_type))
2095 av_log(avctx, AV_LOG_DEBUG, "G");
2096 else if (IS_SKIP(mb_type))
2097 av_log(avctx, AV_LOG_DEBUG, "S");
2098 else if (!USES_LIST(mb_type, 1))
2099 av_log(avctx, AV_LOG_DEBUG, ">");
2100 else if (!USES_LIST(mb_type, 0))
2101 av_log(avctx, AV_LOG_DEBUG, "<");
2103 av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
2104 av_log(avctx, AV_LOG_DEBUG, "X");
2108 if (IS_8X8(mb_type))
2109 av_log(avctx, AV_LOG_DEBUG, "+");
2110 else if (IS_16X8(mb_type))
2111 av_log(avctx, AV_LOG_DEBUG, "-");
2112 else if (IS_8X16(mb_type))
2113 av_log(avctx, AV_LOG_DEBUG, "|");
2114 else if (IS_INTRA(mb_type) || IS_16X16(mb_type))
2115 av_log(avctx, AV_LOG_DEBUG, " ");
2117 av_log(avctx, AV_LOG_DEBUG, "?");
2120 if (IS_INTERLACED(mb_type))
2121 av_log(avctx, AV_LOG_DEBUG, "=");
2123 av_log(avctx, AV_LOG_DEBUG, " ");
2126 av_log(avctx, AV_LOG_DEBUG, "\n");
2130 if ((avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
2131 (avctx->debug_mv)) {
2134 int h_chroma_shift, v_chroma_shift, block_height;
2136 const int shift = 1 + quarter_sample;
2138 const int width = avctx->width;
2139 const int height = avctx->height;
2141 const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1;
2142 const int mv_stride = (mb_width << mv_sample_log2) +
2143 (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1);
2145 *low_delay = 0; // needed to see the vectors without trashing the buffers
2147 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
2149 av_frame_make_writable(pict);
2151 pict->opaque = NULL;
2153 ptr = pict->data[0];
2155 block_height = 16 >> v_chroma_shift;
2157 for (mb_y = 0; mb_y < mb_height; mb_y++) {
2159 for (mb_x = 0; mb_x < mb_width; mb_x++) {
2160 const int mb_index = mb_x + mb_y * mb_stride;
2162 if ((avctx->debug_mv) && motion_val[0]) {
2164 for (type = 0; type < 3; type++) {
2168 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_P_FOR)) ||
2169 (pict->pict_type!= AV_PICTURE_TYPE_P))
2174 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_FOR)) ||
2175 (pict->pict_type!= AV_PICTURE_TYPE_B))
2180 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_BACK)) ||
2181 (pict->pict_type!= AV_PICTURE_TYPE_B))
2186 if (!USES_LIST(mbtype_table[mb_index], direction))
2189 if (IS_8X8(mbtype_table[mb_index])) {
2191 for (i = 0; i < 4; i++) {
2192 int sx = mb_x * 16 + 4 + 8 * (i & 1);
2193 int sy = mb_y * 16 + 4 + 8 * (i >> 1);
2194 int xy = (mb_x * 2 + (i & 1) +
2195 (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
2196 int mx = (motion_val[direction][xy][0] >> shift) + sx;
2197 int my = (motion_val[direction][xy][1] >> shift) + sy;
2198 draw_arrow(ptr, sx, sy, mx, my, width,
2199 height, pict->linesize[0], 100, 0, direction);
2201 } else if (IS_16X8(mbtype_table[mb_index])) {
2203 for (i = 0; i < 2; i++) {
2204 int sx = mb_x * 16 + 8;
2205 int sy = mb_y * 16 + 4 + 8 * i;
2206 int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
2207 int mx = (motion_val[direction][xy][0] >> shift);
2208 int my = (motion_val[direction][xy][1] >> shift);
2210 if (IS_INTERLACED(mbtype_table[mb_index]))
2213 draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
2214 height, pict->linesize[0], 100, 0, direction);
2216 } else if (IS_8X16(mbtype_table[mb_index])) {
2218 for (i = 0; i < 2; i++) {
2219 int sx = mb_x * 16 + 4 + 8 * i;
2220 int sy = mb_y * 16 + 8;
2221 int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
2222 int mx = motion_val[direction][xy][0] >> shift;
2223 int my = motion_val[direction][xy][1] >> shift;
2225 if (IS_INTERLACED(mbtype_table[mb_index]))
2228 draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
2229 height, pict->linesize[0], 100, 0, direction);
2232 int sx= mb_x * 16 + 8;
2233 int sy= mb_y * 16 + 8;
2234 int xy= (mb_x + mb_y * mv_stride) << mv_sample_log2;
2235 int mx= (motion_val[direction][xy][0]>>shift) + sx;
2236 int my= (motion_val[direction][xy][1]>>shift) + sy;
2237 draw_arrow(ptr, sx, sy, mx, my, width, height, pict->linesize[0], 100, 0, direction);
2242 if ((avctx->debug & FF_DEBUG_VIS_QP)) {
2243 uint64_t c = (qscale_table[mb_index] * 128 / 31) *
2244 0x0101010101010101ULL;
2246 for (y = 0; y < block_height; y++) {
2247 *(uint64_t *)(pict->data[1] + 8 * mb_x +
2248 (block_height * mb_y + y) *
2249 pict->linesize[1]) = c;
2250 *(uint64_t *)(pict->data[2] + 8 * mb_x +
2251 (block_height * mb_y + y) *
2252 pict->linesize[2]) = c;
2255 if ((avctx->debug & FF_DEBUG_VIS_MB_TYPE) &&
2257 int mb_type = mbtype_table[mb_index];
2260 #define COLOR(theta, r) \
2261 u = (int)(128 + r * cos(theta * 3.141592 / 180)); \
2262 v = (int)(128 + r * sin(theta * 3.141592 / 180));
2266 if (IS_PCM(mb_type)) {
2268 } else if ((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) ||
2269 IS_INTRA16x16(mb_type)) {
2271 } else if (IS_INTRA4x4(mb_type)) {
2273 } else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type)) {
2275 } else if (IS_DIRECT(mb_type)) {
2277 } else if (IS_GMC(mb_type) && IS_SKIP(mb_type)) {
2279 } else if (IS_GMC(mb_type)) {
2281 } else if (IS_SKIP(mb_type)) {
2283 } else if (!USES_LIST(mb_type, 1)) {
2285 } else if (!USES_LIST(mb_type, 0)) {
2288 av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
2292 u *= 0x0101010101010101ULL;
2293 v *= 0x0101010101010101ULL;
2294 for (y = 0; y < block_height; y++) {
2295 *(uint64_t *)(pict->data[1] + 8 * mb_x +
2296 (block_height * mb_y + y) * pict->linesize[1]) = u;
2297 *(uint64_t *)(pict->data[2] + 8 * mb_x +
2298 (block_height * mb_y + y) * pict->linesize[2]) = v;
2302 if (IS_8X8(mb_type) || IS_16X8(mb_type)) {
2303 *(uint64_t *)(pict->data[0] + 16 * mb_x + 0 +
2304 (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
2305 *(uint64_t *)(pict->data[0] + 16 * mb_x + 8 +
2306 (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
2308 if (IS_8X8(mb_type) || IS_8X16(mb_type)) {
2309 for (y = 0; y < 16; y++)
2310 pict->data[0][16 * mb_x + 8 + (16 * mb_y + y) *
2311 pict->linesize[0]] ^= 0x80;
2313 if (IS_8X8(mb_type) && mv_sample_log2 >= 2) {
2314 int dm = 1 << (mv_sample_log2 - 2);
2315 for (i = 0; i < 4; i++) {
2316 int sx = mb_x * 16 + 8 * (i & 1);
2317 int sy = mb_y * 16 + 8 * (i >> 1);
2318 int xy = (mb_x * 2 + (i & 1) +
2319 (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
2321 int32_t *mv = (int32_t *) &motion_val[0][xy];
2322 if (mv[0] != mv[dm] ||
2323 mv[dm * mv_stride] != mv[dm * (mv_stride + 1)])
2324 for (y = 0; y < 8; y++)
2325 pict->data[0][sx + 4 + (sy + y) * pict->linesize[0]] ^= 0x80;
2326 if (mv[0] != mv[dm * mv_stride] || mv[dm] != mv[dm * (mv_stride + 1)])
2327 *(uint64_t *)(pict->data[0] + sx + (sy + 4) *
2328 pict->linesize[0]) ^= 0x8080808080808080ULL;
2332 if (IS_INTERLACED(mb_type) &&
2333 avctx->codec->id == AV_CODEC_ID_H264) {
2338 mbskip_table[mb_index] = 0;
2344 void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
2346 ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
2347 p->qscale_table, p->motion_val, &s->low_delay,
2348 s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
2351 int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
2353 AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
2354 int offset = 2*s->mb_stride + 1;
2356 return AVERROR(ENOMEM);
2357 av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
2358 ref->size -= offset;
2359 ref->data += offset;
2360 return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
2363 static inline int hpel_motion_lowres(MpegEncContext *s,
2364 uint8_t *dest, uint8_t *src,
2365 int field_based, int field_select,
2366 int src_x, int src_y,
2367 int width, int height, ptrdiff_t stride,
2368 int h_edge_pos, int v_edge_pos,
2369 int w, int h, h264_chroma_mc_func *pix_op,
2370 int motion_x, int motion_y)
2372 const int lowres = s->avctx->lowres;
2373 const int op_index = FFMIN(lowres, 3);
2374 const int s_mask = (2 << lowres) - 1;
2378 if (s->quarter_sample) {
2383 sx = motion_x & s_mask;
2384 sy = motion_y & s_mask;
2385 src_x += motion_x >> lowres + 1;
2386 src_y += motion_y >> lowres + 1;
2388 src += src_y * stride + src_x;
2390 if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w, 0) ||
2391 (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
2392 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
2393 s->linesize, s->linesize,
2394 w + 1, (h + 1) << field_based,
2395 src_x, src_y << field_based,
2396 h_edge_pos, v_edge_pos);
2397 src = s->edge_emu_buffer;
2401 sx = (sx << 2) >> lowres;
2402 sy = (sy << 2) >> lowres;
2405 pix_op[op_index](dest, src, stride, h, sx, sy);
2409 /* apply one mpeg motion vector to the three components */
2410 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
2417 uint8_t **ref_picture,
2418 h264_chroma_mc_func *pix_op,
2419 int motion_x, int motion_y,
2422 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
2423 int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
2424 ptrdiff_t uvlinesize, linesize;
2425 const int lowres = s->avctx->lowres;
2426 const int op_index = FFMIN(lowres-1+s->chroma_x_shift, 3);
2427 const int block_s = 8>>lowres;
2428 const int s_mask = (2 << lowres) - 1;
2429 const int h_edge_pos = s->h_edge_pos >> lowres;
2430 const int v_edge_pos = s->v_edge_pos >> lowres;
2431 linesize = s->current_picture.f->linesize[0] << field_based;
2432 uvlinesize = s->current_picture.f->linesize[1] << field_based;
2434 // FIXME obviously not perfect but qpel will not work in lowres anyway
2435 if (s->quarter_sample) {
2441 motion_y += (bottom_field - field_select)*((1 << lowres)-1);
2444 sx = motion_x & s_mask;
2445 sy = motion_y & s_mask;
2446 src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
2447 src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
2449 if (s->out_format == FMT_H263) {
2450 uvsx = ((motion_x >> 1) & s_mask) | (sx & 1);
2451 uvsy = ((motion_y >> 1) & s_mask) | (sy & 1);
2452 uvsrc_x = src_x >> 1;
2453 uvsrc_y = src_y >> 1;
2454 } else if (s->out_format == FMT_H261) {
2455 // even chroma mv's are full pel in H261
2458 uvsx = (2 * mx) & s_mask;
2459 uvsy = (2 * my) & s_mask;
2460 uvsrc_x = s->mb_x * block_s + (mx >> lowres);
2461 uvsrc_y = mb_y * block_s + (my >> lowres);
2463 if(s->chroma_y_shift){
2468 uvsrc_x = s->mb_x * block_s + (mx >> lowres + 1);
2469 uvsrc_y = (mb_y * block_s >> field_based) + (my >> lowres + 1);
2471 if(s->chroma_x_shift){
2475 uvsy = motion_y & s_mask;
2477 uvsrc_x = s->mb_x*block_s + (mx >> (lowres+1));
2480 uvsx = motion_x & s_mask;
2481 uvsy = motion_y & s_mask;
2488 ptr_y = ref_picture[0] + src_y * linesize + src_x;
2489 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
2490 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
2492 if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s, 0) || uvsrc_y<0 ||
2493 (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
2494 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
2495 linesize >> field_based, linesize >> field_based,
2496 17, 17 + field_based,
2497 src_x, src_y << field_based, h_edge_pos,
2499 ptr_y = s->edge_emu_buffer;
2500 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
2501 uint8_t *ubuf = s->edge_emu_buffer + 18 * s->linesize;
2502 uint8_t *vbuf =ubuf + 9 * s->uvlinesize;
2503 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
2504 uvlinesize >> field_based, uvlinesize >> field_based,
2506 uvsrc_x, uvsrc_y << field_based,
2507 h_edge_pos >> 1, v_edge_pos >> 1);
2508 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
2509 uvlinesize >> field_based,uvlinesize >> field_based,
2511 uvsrc_x, uvsrc_y << field_based,
2512 h_edge_pos >> 1, v_edge_pos >> 1);
2518 // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
2520 dest_y += s->linesize;
2521 dest_cb += s->uvlinesize;
2522 dest_cr += s->uvlinesize;
2526 ptr_y += s->linesize;
2527 ptr_cb += s->uvlinesize;
2528 ptr_cr += s->uvlinesize;
2531 sx = (sx << 2) >> lowres;
2532 sy = (sy << 2) >> lowres;
2533 pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
2535 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
2536 int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
2537 uvsx = (uvsx << 2) >> lowres;
2538 uvsy = (uvsy << 2) >> lowres;
2540 pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
2541 pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
2544 // FIXME h261 lowres loop filter
2547 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
2548 uint8_t *dest_cb, uint8_t *dest_cr,
2549 uint8_t **ref_picture,
2550 h264_chroma_mc_func * pix_op,
2553 const int lowres = s->avctx->lowres;
2554 const int op_index = FFMIN(lowres, 3);
2555 const int block_s = 8 >> lowres;
2556 const int s_mask = (2 << lowres) - 1;
2557 const int h_edge_pos = s->h_edge_pos >> lowres + 1;
2558 const int v_edge_pos = s->v_edge_pos >> lowres + 1;
2559 int emu = 0, src_x, src_y, sx, sy;
2563 if (s->quarter_sample) {
2568 /* In case of 8X8, we construct a single chroma motion vector
2569 with a special rounding */
2570 mx = ff_h263_round_chroma(mx);
2571 my = ff_h263_round_chroma(my);
2575 src_x = s->mb_x * block_s + (mx >> lowres + 1);
2576 src_y = s->mb_y * block_s + (my >> lowres + 1);
2578 offset = src_y * s->uvlinesize + src_x;
2579 ptr = ref_picture[1] + offset;
2580 if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
2581 (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
2582 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
2583 s->uvlinesize, s->uvlinesize,
2585 src_x, src_y, h_edge_pos, v_edge_pos);
2586 ptr = s->edge_emu_buffer;
2589 sx = (sx << 2) >> lowres;
2590 sy = (sy << 2) >> lowres;
2591 pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
2593 ptr = ref_picture[2] + offset;
2595 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
2596 s->uvlinesize, s->uvlinesize,
2598 src_x, src_y, h_edge_pos, v_edge_pos);
2599 ptr = s->edge_emu_buffer;
2601 pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
2605 * motion compensation of a single macroblock
2607 * @param dest_y luma destination pointer
2608 * @param dest_cb chroma cb/u destination pointer
2609 * @param dest_cr chroma cr/v destination pointer
2610 * @param dir direction (0->forward, 1->backward)
2611 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
2612 * @param pix_op halfpel motion compensation function (average or put normally)
2613 * the motion vectors are taken from s->mv and the MV type from s->mv_type
2615 static inline void MPV_motion_lowres(MpegEncContext *s,
2616 uint8_t *dest_y, uint8_t *dest_cb,
2618 int dir, uint8_t **ref_picture,
2619 h264_chroma_mc_func *pix_op)
2623 const int lowres = s->avctx->lowres;
2624 const int block_s = 8 >>lowres;
2629 switch (s->mv_type) {
2631 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2633 ref_picture, pix_op,
2634 s->mv[dir][0][0], s->mv[dir][0][1],
2640 for (i = 0; i < 4; i++) {
2641 hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
2642 s->linesize) * block_s,
2643 ref_picture[0], 0, 0,
2644 (2 * mb_x + (i & 1)) * block_s,
2645 (2 * mb_y + (i >> 1)) * block_s,
2646 s->width, s->height, s->linesize,
2647 s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
2648 block_s, block_s, pix_op,
2649 s->mv[dir][i][0], s->mv[dir][i][1]);
2651 mx += s->mv[dir][i][0];
2652 my += s->mv[dir][i][1];
2655 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY))
2656 chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
2660 if (s->picture_structure == PICT_FRAME) {
2662 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2663 1, 0, s->field_select[dir][0],
2664 ref_picture, pix_op,
2665 s->mv[dir][0][0], s->mv[dir][0][1],
2668 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2669 1, 1, s->field_select[dir][1],
2670 ref_picture, pix_op,
2671 s->mv[dir][1][0], s->mv[dir][1][1],
2674 if (s->picture_structure != s->field_select[dir][0] + 1 &&
2675 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
2676 ref_picture = s->current_picture_ptr->f->data;
2679 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2680 0, 0, s->field_select[dir][0],
2681 ref_picture, pix_op,
2683 s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
2687 for (i = 0; i < 2; i++) {
2688 uint8_t **ref2picture;
2690 if (s->picture_structure == s->field_select[dir][i] + 1 ||
2691 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
2692 ref2picture = ref_picture;
2694 ref2picture = s->current_picture_ptr->f->data;
2697 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2698 0, 0, s->field_select[dir][i],
2699 ref2picture, pix_op,
2700 s->mv[dir][i][0], s->mv[dir][i][1] +
2701 2 * block_s * i, block_s, mb_y >> 1);
2703 dest_y += 2 * block_s * s->linesize;
2704 dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
2705 dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
2709 if (s->picture_structure == PICT_FRAME) {
2710 for (i = 0; i < 2; i++) {
2712 for (j = 0; j < 2; j++) {
2713 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2715 ref_picture, pix_op,
2716 s->mv[dir][2 * i + j][0],
2717 s->mv[dir][2 * i + j][1],
2720 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
2723 for (i = 0; i < 2; i++) {
2724 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2725 0, 0, s->picture_structure != i + 1,
2726 ref_picture, pix_op,
2727 s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
2728 2 * block_s, mb_y >> 1);
2730 // after put we make avg of the same block
2731 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
2733 // opposite parity is always in the same
2734 // frame if this is second field
2735 if (!s->first_field) {
2736 ref_picture = s->current_picture_ptr->f->data;
2747 * find the lowest MB row referenced in the MVs
2749 int ff_mpv_lowest_referenced_row(MpegEncContext *s, int dir)
2751 int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
2752 int my, off, i, mvs;
2754 if (s->picture_structure != PICT_FRAME || s->mcsel)
2757 switch (s->mv_type) {
2771 for (i = 0; i < mvs; i++) {
2772 my = s->mv[dir][i][1];
2773 my_max = FFMAX(my_max, my);
2774 my_min = FFMIN(my_min, my);
2777 off = ((FFMAX(-my_min, my_max)<<qpel_shift) + 63) >> 6;
2779 return av_clip(s->mb_y + off, 0, s->mb_height - 1);
2781 return s->mb_height-1;
2784 /* put block[] to dest[] */
2785 static inline void put_dct(MpegEncContext *s,
2786 int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
2788 s->dct_unquantize_intra(s, block, i, qscale);
2789 s->idsp.idct_put(dest, line_size, block);
2792 /* add block[] to dest[] */
2793 static inline void add_dct(MpegEncContext *s,
2794 int16_t *block, int i, uint8_t *dest, int line_size)
2796 if (s->block_last_index[i] >= 0) {
2797 s->idsp.idct_add(dest, line_size, block);
2801 static inline void add_dequant_dct(MpegEncContext *s,
2802 int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
2804 if (s->block_last_index[i] >= 0) {
2805 s->dct_unquantize_inter(s, block, i, qscale);
2807 s->idsp.idct_add(dest, line_size, block);
2812 * Clean dc, ac, coded_block for the current non-intra MB.
2814 void ff_clean_intra_table_entries(MpegEncContext *s)
2816 int wrap = s->b8_stride;
2817 int xy = s->block_index[0];
2820 s->dc_val[0][xy + 1 ] =
2821 s->dc_val[0][xy + wrap] =
2822 s->dc_val[0][xy + 1 + wrap] = 1024;
2824 memset(s->ac_val[0][xy ], 0, 32 * sizeof(int16_t));
2825 memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
2826 if (s->msmpeg4_version>=3) {
2827 s->coded_block[xy ] =
2828 s->coded_block[xy + 1 ] =
2829 s->coded_block[xy + wrap] =
2830 s->coded_block[xy + 1 + wrap] = 0;
2833 wrap = s->mb_stride;
2834 xy = s->mb_x + s->mb_y * wrap;
2836 s->dc_val[2][xy] = 1024;
2838 memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
2839 memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
2841 s->mbintra_table[xy]= 0;
2844 /* generic function called after a macroblock has been parsed by the
2845 decoder or after it has been encoded by the encoder.
2847 Important variables used:
2848 s->mb_intra : true if intra macroblock
2849 s->mv_dir : motion vector direction
2850 s->mv_type : motion vector type
2851 s->mv : motion vector
2852 s->interlaced_dct : true if interlaced dct used (mpeg2)
2854 static av_always_inline
2855 void mpv_decode_mb_internal(MpegEncContext *s, int16_t block[12][64],
2856 int lowres_flag, int is_mpeg12)
2858 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
2861 s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
2862 s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
2866 if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
2867 /* print DCT coefficients */
2869 av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
2871 for(j=0; j<64; j++){
2872 av_log(s->avctx, AV_LOG_DEBUG, "%5d",
2873 block[i][s->idsp.idct_permutation[j]]);
2875 av_log(s->avctx, AV_LOG_DEBUG, "\n");
2879 s->current_picture.qscale_table[mb_xy] = s->qscale;
2881 /* update DC predictors for P macroblocks */
2883 if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
2884 if(s->mbintra_table[mb_xy])
2885 ff_clean_intra_table_entries(s);
2889 s->last_dc[2] = 128 << s->intra_dc_precision;
2892 else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
2893 s->mbintra_table[mb_xy]=1;
2895 if ((s->avctx->flags & CODEC_FLAG_PSNR) || s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor ||
2896 !(s->encoding && (s->intra_only || s->pict_type == AV_PICTURE_TYPE_B) &&
2897 s->avctx->mb_decision != FF_MB_DECISION_RD)) { // FIXME precalc
2898 uint8_t *dest_y, *dest_cb, *dest_cr;
2899 int dct_linesize, dct_offset;
2900 op_pixels_func (*op_pix)[4];
2901 qpel_mc_func (*op_qpix)[16];
2902 const int linesize = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
2903 const int uvlinesize = s->current_picture.f->linesize[1];
2904 const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
2905 const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
2907 /* avoid copy if macroblock skipped in last frame too */
2908 /* skip only during decoding as we might trash the buffers during encoding a bit */
2910 uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
2912 if (s->mb_skipped) {
2914 av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
2916 } else if(!s->current_picture.reference) {
2919 *mbskip_ptr = 0; /* not skipped */
2923 dct_linesize = linesize << s->interlaced_dct;
2924 dct_offset = s->interlaced_dct ? linesize : linesize * block_size;
2928 dest_cb= s->dest[1];
2929 dest_cr= s->dest[2];
2931 dest_y = s->b_scratchpad;
2932 dest_cb= s->b_scratchpad+16*linesize;
2933 dest_cr= s->b_scratchpad+32*linesize;
2937 /* motion handling */
2938 /* decoding or more than one mb_type (MC was already done otherwise) */
2941 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
2942 if (s->mv_dir & MV_DIR_FORWARD) {
2943 ff_thread_await_progress(&s->last_picture_ptr->tf,
2944 ff_mpv_lowest_referenced_row(s, 0),
2947 if (s->mv_dir & MV_DIR_BACKWARD) {
2948 ff_thread_await_progress(&s->next_picture_ptr->tf,
2949 ff_mpv_lowest_referenced_row(s, 1),
2955 h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
2957 if (s->mv_dir & MV_DIR_FORWARD) {
2958 MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
2959 op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
2961 if (s->mv_dir & MV_DIR_BACKWARD) {
2962 MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
2965 op_qpix = s->me.qpel_put;
2966 if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
2967 op_pix = s->hdsp.put_pixels_tab;
2969 op_pix = s->hdsp.put_no_rnd_pixels_tab;
2971 if (s->mv_dir & MV_DIR_FORWARD) {
2972 ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
2973 op_pix = s->hdsp.avg_pixels_tab;
2974 op_qpix= s->me.qpel_avg;
2976 if (s->mv_dir & MV_DIR_BACKWARD) {
2977 ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
2982 /* skip dequant / idct if we are really late ;) */
2983 if(s->avctx->skip_idct){
2984 if( (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
2985 ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
2986 || s->avctx->skip_idct >= AVDISCARD_ALL)
2990 /* add dct residue */
2991 if(s->encoding || !( s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
2992 || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
2993 add_dequant_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale);
2994 add_dequant_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale);
2995 add_dequant_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale);
2996 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2998 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
2999 if (s->chroma_y_shift){
3000 add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
3001 add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
3005 add_dequant_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale);
3006 add_dequant_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale);
3007 add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
3008 add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
3011 } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
3012 add_dct(s, block[0], 0, dest_y , dct_linesize);
3013 add_dct(s, block[1], 1, dest_y + block_size, dct_linesize);
3014 add_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize);
3015 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
3017 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3018 if(s->chroma_y_shift){//Chroma420
3019 add_dct(s, block[4], 4, dest_cb, uvlinesize);
3020 add_dct(s, block[5], 5, dest_cr, uvlinesize);
3023 dct_linesize = uvlinesize << s->interlaced_dct;
3024 dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
3026 add_dct(s, block[4], 4, dest_cb, dct_linesize);
3027 add_dct(s, block[5], 5, dest_cr, dct_linesize);
3028 add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
3029 add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
3030 if(!s->chroma_x_shift){//Chroma444
3031 add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
3032 add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
3033 add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
3034 add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
3039 else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
3040 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
3043 /* dct only in intra block */
3044 if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
3045 put_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale);
3046 put_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale);
3047 put_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale);
3048 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
3050 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3051 if(s->chroma_y_shift){
3052 put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
3053 put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
3057 put_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale);
3058 put_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale);
3059 put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
3060 put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
3064 s->idsp.idct_put(dest_y, dct_linesize, block[0]);
3065 s->idsp.idct_put(dest_y + block_size, dct_linesize, block[1]);
3066 s->idsp.idct_put(dest_y + dct_offset, dct_linesize, block[2]);
3067 s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
3069 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3070 if(s->chroma_y_shift){
3071 s->idsp.idct_put(dest_cb, uvlinesize, block[4]);
3072 s->idsp.idct_put(dest_cr, uvlinesize, block[5]);
3075 dct_linesize = uvlinesize << s->interlaced_dct;
3076 dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
3078 s->idsp.idct_put(dest_cb, dct_linesize, block[4]);
3079 s->idsp.idct_put(dest_cr, dct_linesize, block[5]);
3080 s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
3081 s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
3082 if(!s->chroma_x_shift){//Chroma444
3083 s->idsp.idct_put(dest_cb + block_size, dct_linesize, block[8]);
3084 s->idsp.idct_put(dest_cr + block_size, dct_linesize, block[9]);
3085 s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
3086 s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
3094 s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y , linesize,16);
3095 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3096 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
3097 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
3103 void ff_mpv_decode_mb(MpegEncContext *s, int16_t block[12][64])
3106 if(s->out_format == FMT_MPEG1) {
3107 if(s->avctx->lowres) mpv_decode_mb_internal(s, block, 1, 1);
3108 else mpv_decode_mb_internal(s, block, 0, 1);
3111 if(s->avctx->lowres) mpv_decode_mb_internal(s, block, 1, 0);
3112 else mpv_decode_mb_internal(s, block, 0, 0);
3115 void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
3117 ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
3118 s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
3119 s->first_field, s->low_delay);
3122 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
3123 const int linesize = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
3124 const int uvlinesize = s->current_picture.f->linesize[1];
3125 const int mb_size= 4 - s->avctx->lowres;
3127 s->block_index[0]= s->b8_stride*(s->mb_y*2 ) - 2 + s->mb_x*2;
3128 s->block_index[1]= s->b8_stride*(s->mb_y*2 ) - 1 + s->mb_x*2;
3129 s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
3130 s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
3131 s->block_index[4]= s->mb_stride*(s->mb_y + 1) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
3132 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;
3133 //block_index is not used by mpeg2, so it is not affected by chroma_format
3135 s->dest[0] = s->current_picture.f->data[0] + (int)((s->mb_x - 1U) << mb_size);
3136 s->dest[1] = s->current_picture.f->data[1] + (int)((s->mb_x - 1U) << (mb_size - s->chroma_x_shift));
3137 s->dest[2] = s->current_picture.f->data[2] + (int)((s->mb_x - 1U) << (mb_size - s->chroma_x_shift));
3139 if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
3141 if(s->picture_structure==PICT_FRAME){
3142 s->dest[0] += s->mb_y * linesize << mb_size;
3143 s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
3144 s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
3146 s->dest[0] += (s->mb_y>>1) * linesize << mb_size;
3147 s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
3148 s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
3149 av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
3155 * Permute an 8x8 block.
3156 * @param block the block which will be permuted according to the given permutation vector
3157 * @param permutation the permutation vector
3158 * @param last the last non zero coefficient in scantable order, used to speed the permutation up
3159 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not
3160 * (inverse) permutated to scantable order!
3162 void ff_block_permute(int16_t *block, uint8_t *permutation, const uint8_t *scantable, int last)
3168 //if(permutation[1]==1) return; //FIXME it is ok but not clean and might fail for some permutations
3170 for(i=0; i<=last; i++){
3171 const int j= scantable[i];
3176 for(i=0; i<=last; i++){
3177 const int j= scantable[i];
3178 const int perm_j= permutation[j];
3179 block[perm_j]= temp[j];
3183 void ff_mpeg_flush(AVCodecContext *avctx){
3185 MpegEncContext *s = avctx->priv_data;
3187 if (!s || !s->picture)
3190 for (i = 0; i < MAX_PICTURE_COUNT; i++)
3191 ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
3192 s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
3194 ff_mpeg_unref_picture(s->avctx, &s->current_picture);
3195 ff_mpeg_unref_picture(s->avctx, &s->last_picture);
3196 ff_mpeg_unref_picture(s->avctx, &s->next_picture);
3198 s->mb_x= s->mb_y= 0;
3201 s->parse_context.state= -1;
3202 s->parse_context.frame_start_found= 0;
3203 s->parse_context.overread= 0;
3204 s->parse_context.overread_index= 0;
3205 s->parse_context.index= 0;
3206 s->parse_context.last_index= 0;
3207 s->bitstream_buffer_size=0;
3212 * set qscale and update qscale dependent variables.
3214 void ff_set_qscale(MpegEncContext * s, int qscale)
3218 else if (qscale > 31)
3222 s->chroma_qscale= s->chroma_qscale_table[qscale];
3224 s->y_dc_scale= s->y_dc_scale_table[ qscale ];
3225 s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
3228 void ff_mpv_report_decode_progress(MpegEncContext *s)
3230 if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
3231 ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);