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 int ff_mpeg_framesize_alloc(AVCodecContext *avctx, MotionEstContext *me,
365 ScratchpadContext *sc, int linesize)
367 int alloc_size = FFALIGN(FFABS(linesize) + 64, 32);
369 if (avctx->hwaccel || avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
373 av_log(avctx, AV_LOG_ERROR, "Image too small, temporary buffers cannot function\n");
374 return AVERROR_PATCHWELCOME;
377 // edge emu needs blocksize + filter length - 1
378 // (= 17x17 for halfpel / 21x21 for h264)
379 // VC1 computes luma and chroma simultaneously and needs 19X19 + 9x9
380 // at uvlinesize. It supports only YUV420 so 24x24 is enough
381 // linesize * interlaced * MBsize
382 // we also use this buffer for encoding in encode_mb_internal() needig an additional 32 lines
383 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, sc->edge_emu_buffer, alloc_size, 4 * 68,
386 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, me->scratchpad, alloc_size, 4 * 16 * 2,
388 me->temp = me->scratchpad;
389 sc->rd_scratchpad = me->scratchpad;
390 sc->b_scratchpad = me->scratchpad;
391 sc->obmc_scratchpad = me->scratchpad + 16;
395 av_freep(&sc->edge_emu_buffer);
396 return AVERROR(ENOMEM);
400 * Allocate a frame buffer
402 static int alloc_frame_buffer(AVCodecContext *avctx, Picture *pic,
403 MotionEstContext *me, ScratchpadContext *sc,
404 int chroma_x_shift, int chroma_y_shift,
405 int linesize, int uvlinesize)
407 int edges_needed = av_codec_is_encoder(avctx->codec);
411 if (avctx->codec_id != AV_CODEC_ID_WMV3IMAGE &&
412 avctx->codec_id != AV_CODEC_ID_VC1IMAGE &&
413 avctx->codec_id != AV_CODEC_ID_MSS2) {
415 pic->f->width = avctx->width + 2 * EDGE_WIDTH;
416 pic->f->height = avctx->height + 2 * EDGE_WIDTH;
419 r = ff_thread_get_buffer(avctx, &pic->tf,
420 pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);
422 pic->f->width = avctx->width;
423 pic->f->height = avctx->height;
424 pic->f->format = avctx->pix_fmt;
425 r = avcodec_default_get_buffer2(avctx, pic->f, 0);
428 if (r < 0 || !pic->f->buf[0]) {
429 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed (%d %p)\n",
436 for (i = 0; pic->f->data[i]; i++) {
437 int offset = (EDGE_WIDTH >> (i ? chroma_y_shift : 0)) *
438 pic->f->linesize[i] +
439 (EDGE_WIDTH >> (i ? chroma_x_shift : 0));
440 pic->f->data[i] += offset;
442 pic->f->width = avctx->width;
443 pic->f->height = avctx->height;
446 if (avctx->hwaccel) {
447 assert(!pic->hwaccel_picture_private);
448 if (avctx->hwaccel->frame_priv_data_size) {
449 pic->hwaccel_priv_buf = av_buffer_allocz(avctx->hwaccel->frame_priv_data_size);
450 if (!pic->hwaccel_priv_buf) {
451 av_log(avctx, AV_LOG_ERROR, "alloc_frame_buffer() failed (hwaccel private data allocation)\n");
454 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
458 if (linesize && (linesize != pic->f->linesize[0] ||
459 uvlinesize != pic->f->linesize[1])) {
460 av_log(avctx, AV_LOG_ERROR,
461 "get_buffer() failed (stride changed)\n");
462 ff_mpeg_unref_picture(avctx, pic);
466 if (pic->f->linesize[1] != pic->f->linesize[2]) {
467 av_log(avctx, AV_LOG_ERROR,
468 "get_buffer() failed (uv stride mismatch)\n");
469 ff_mpeg_unref_picture(avctx, pic);
473 if (!sc->edge_emu_buffer &&
474 (ret = ff_mpeg_framesize_alloc(avctx, me, sc,
475 pic->f->linesize[0])) < 0) {
476 av_log(avctx, AV_LOG_ERROR,
477 "get_buffer() failed to allocate context scratch buffers.\n");
478 ff_mpeg_unref_picture(avctx, pic);
485 void ff_free_picture_tables(Picture *pic)
489 pic->alloc_mb_width =
490 pic->alloc_mb_height = 0;
492 av_buffer_unref(&pic->mb_var_buf);
493 av_buffer_unref(&pic->mc_mb_var_buf);
494 av_buffer_unref(&pic->mb_mean_buf);
495 av_buffer_unref(&pic->mbskip_table_buf);
496 av_buffer_unref(&pic->qscale_table_buf);
497 av_buffer_unref(&pic->mb_type_buf);
499 for (i = 0; i < 2; i++) {
500 av_buffer_unref(&pic->motion_val_buf[i]);
501 av_buffer_unref(&pic->ref_index_buf[i]);
505 static int alloc_picture_tables(AVCodecContext *avctx, Picture *pic, int encoding, int out_format,
506 int mb_stride, int mb_width, int mb_height, int b8_stride)
508 const int big_mb_num = mb_stride * (mb_height + 1) + 1;
509 const int mb_array_size = mb_stride * mb_height;
510 const int b8_array_size = b8_stride * mb_height * 2;
514 pic->mbskip_table_buf = av_buffer_allocz(mb_array_size + 2);
515 pic->qscale_table_buf = av_buffer_allocz(big_mb_num + mb_stride);
516 pic->mb_type_buf = av_buffer_allocz((big_mb_num + mb_stride) *
518 if (!pic->mbskip_table_buf || !pic->qscale_table_buf || !pic->mb_type_buf)
519 return AVERROR(ENOMEM);
522 pic->mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t));
523 pic->mc_mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t));
524 pic->mb_mean_buf = av_buffer_allocz(mb_array_size);
525 if (!pic->mb_var_buf || !pic->mc_mb_var_buf || !pic->mb_mean_buf)
526 return AVERROR(ENOMEM);
529 if (out_format == FMT_H263 || encoding || avctx->debug_mv ||
530 (avctx->flags2 & CODEC_FLAG2_EXPORT_MVS)) {
531 int mv_size = 2 * (b8_array_size + 4) * sizeof(int16_t);
532 int ref_index_size = 4 * mb_array_size;
534 for (i = 0; mv_size && i < 2; i++) {
535 pic->motion_val_buf[i] = av_buffer_allocz(mv_size);
536 pic->ref_index_buf[i] = av_buffer_allocz(ref_index_size);
537 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
538 return AVERROR(ENOMEM);
542 pic->alloc_mb_width = mb_width;
543 pic->alloc_mb_height = mb_height;
548 static int make_tables_writable(Picture *pic)
551 #define MAKE_WRITABLE(table) \
554 (ret = av_buffer_make_writable(&pic->table)) < 0)\
558 MAKE_WRITABLE(mb_var_buf);
559 MAKE_WRITABLE(mc_mb_var_buf);
560 MAKE_WRITABLE(mb_mean_buf);
561 MAKE_WRITABLE(mbskip_table_buf);
562 MAKE_WRITABLE(qscale_table_buf);
563 MAKE_WRITABLE(mb_type_buf);
565 for (i = 0; i < 2; i++) {
566 MAKE_WRITABLE(motion_val_buf[i]);
567 MAKE_WRITABLE(ref_index_buf[i]);
573 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared)
575 return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, shared, 0,
576 s->chroma_x_shift, s->chroma_y_shift, s->out_format,
577 s->mb_stride, s->mb_width, s->mb_height, s->b8_stride,
578 &s->linesize, &s->uvlinesize);
582 * Allocate a Picture.
583 * The pixels are allocated/set by calling get_buffer() if shared = 0
585 int ff_alloc_picture(AVCodecContext *avctx, Picture *pic, MotionEstContext *me,
586 ScratchpadContext *sc, int shared, int encoding,
587 int chroma_x_shift, int chroma_y_shift, int out_format,
588 int mb_stride, int mb_width, int mb_height, int b8_stride,
589 ptrdiff_t *linesize, ptrdiff_t *uvlinesize)
593 if (pic->qscale_table_buf)
594 if ( pic->alloc_mb_width != mb_width
595 || pic->alloc_mb_height != mb_height)
596 ff_free_picture_tables(pic);
599 av_assert0(pic->f->data[0]);
602 av_assert0(!pic->f->buf[0]);
603 if (alloc_frame_buffer(avctx, pic, me, sc,
604 chroma_x_shift, chroma_y_shift,
605 *linesize, *uvlinesize) < 0)
608 *linesize = pic->f->linesize[0];
609 *uvlinesize = pic->f->linesize[1];
612 if (!pic->qscale_table_buf)
613 ret = alloc_picture_tables(avctx, pic, encoding, out_format,
614 mb_stride, mb_width, mb_height, b8_stride);
616 ret = make_tables_writable(pic);
621 pic->mb_var = (uint16_t*)pic->mb_var_buf->data;
622 pic->mc_mb_var = (uint16_t*)pic->mc_mb_var_buf->data;
623 pic->mb_mean = pic->mb_mean_buf->data;
626 pic->mbskip_table = pic->mbskip_table_buf->data;
627 pic->qscale_table = pic->qscale_table_buf->data + 2 * mb_stride + 1;
628 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * mb_stride + 1;
630 if (pic->motion_val_buf[0]) {
631 for (i = 0; i < 2; i++) {
632 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
633 pic->ref_index[i] = pic->ref_index_buf[i]->data;
639 av_log(avctx, AV_LOG_ERROR, "Error allocating a picture.\n");
640 ff_mpeg_unref_picture(avctx, pic);
641 ff_free_picture_tables(pic);
642 return AVERROR(ENOMEM);
646 * Deallocate a picture.
648 void ff_mpeg_unref_picture(AVCodecContext *avctx, Picture *pic)
650 int off = offsetof(Picture, mb_mean) + sizeof(pic->mb_mean);
653 /* WM Image / Screen codecs allocate internal buffers with different
654 * dimensions / colorspaces; ignore user-defined callbacks for these. */
655 if (avctx->codec->id != AV_CODEC_ID_WMV3IMAGE &&
656 avctx->codec->id != AV_CODEC_ID_VC1IMAGE &&
657 avctx->codec->id != AV_CODEC_ID_MSS2)
658 ff_thread_release_buffer(avctx, &pic->tf);
660 av_frame_unref(pic->f);
662 av_buffer_unref(&pic->hwaccel_priv_buf);
664 if (pic->needs_realloc)
665 ff_free_picture_tables(pic);
667 memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
670 static int update_picture_tables(Picture *dst, Picture *src)
674 #define UPDATE_TABLE(table)\
677 (!dst->table || dst->table->buffer != src->table->buffer)) {\
678 av_buffer_unref(&dst->table);\
679 dst->table = av_buffer_ref(src->table);\
681 ff_free_picture_tables(dst);\
682 return AVERROR(ENOMEM);\
687 UPDATE_TABLE(mb_var_buf);
688 UPDATE_TABLE(mc_mb_var_buf);
689 UPDATE_TABLE(mb_mean_buf);
690 UPDATE_TABLE(mbskip_table_buf);
691 UPDATE_TABLE(qscale_table_buf);
692 UPDATE_TABLE(mb_type_buf);
693 for (i = 0; i < 2; i++) {
694 UPDATE_TABLE(motion_val_buf[i]);
695 UPDATE_TABLE(ref_index_buf[i]);
698 dst->mb_var = src->mb_var;
699 dst->mc_mb_var = src->mc_mb_var;
700 dst->mb_mean = src->mb_mean;
701 dst->mbskip_table = src->mbskip_table;
702 dst->qscale_table = src->qscale_table;
703 dst->mb_type = src->mb_type;
704 for (i = 0; i < 2; i++) {
705 dst->motion_val[i] = src->motion_val[i];
706 dst->ref_index[i] = src->ref_index[i];
709 dst->alloc_mb_width = src->alloc_mb_width;
710 dst->alloc_mb_height = src->alloc_mb_height;
715 int ff_mpeg_ref_picture(AVCodecContext *avctx, Picture *dst, Picture *src)
719 av_assert0(!dst->f->buf[0]);
720 av_assert0(src->f->buf[0]);
724 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
728 ret = update_picture_tables(dst, src);
732 if (src->hwaccel_picture_private) {
733 dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
734 if (!dst->hwaccel_priv_buf)
736 dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
739 dst->field_picture = src->field_picture;
740 dst->mb_var_sum = src->mb_var_sum;
741 dst->mc_mb_var_sum = src->mc_mb_var_sum;
742 dst->b_frame_score = src->b_frame_score;
743 dst->needs_realloc = src->needs_realloc;
744 dst->reference = src->reference;
745 dst->shared = src->shared;
749 ff_mpeg_unref_picture(avctx, dst);
753 static int init_duplicate_context(MpegEncContext *s)
755 int y_size = s->b8_stride * (2 * s->mb_height + 1);
756 int c_size = s->mb_stride * (s->mb_height + 1);
757 int yc_size = y_size + 2 * c_size;
760 if (s->mb_height & 1)
761 yc_size += 2*s->b8_stride + 2*s->mb_stride;
763 s->sc.edge_emu_buffer =
766 s->sc.rd_scratchpad =
768 s->sc.obmc_scratchpad = NULL;
771 FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map,
772 ME_MAP_SIZE * sizeof(uint32_t), fail)
773 FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map,
774 ME_MAP_SIZE * sizeof(uint32_t), fail)
775 if (s->avctx->noise_reduction) {
776 FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum,
777 2 * 64 * sizeof(int), fail)
780 FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail)
781 s->block = s->blocks[0];
783 for (i = 0; i < 12; i++) {
784 s->pblocks[i] = &s->block[i];
786 if (s->avctx->codec_tag == AV_RL32("VCR2")) {
788 FFSWAP(void *, s->pblocks[4], s->pblocks[5]);
791 if (s->out_format == FMT_H263) {
793 FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base,
794 yc_size * sizeof(int16_t) * 16, fail);
795 s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
796 s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
797 s->ac_val[2] = s->ac_val[1] + c_size;
802 return -1; // free() through ff_mpv_common_end()
805 static void free_duplicate_context(MpegEncContext *s)
810 av_freep(&s->sc.edge_emu_buffer);
811 av_freep(&s->me.scratchpad);
813 s->sc.rd_scratchpad =
815 s->sc.obmc_scratchpad = NULL;
817 av_freep(&s->dct_error_sum);
818 av_freep(&s->me.map);
819 av_freep(&s->me.score_map);
820 av_freep(&s->blocks);
821 av_freep(&s->ac_val_base);
825 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
827 #define COPY(a) bak->a = src->a
828 COPY(sc.edge_emu_buffer);
831 COPY(sc.rd_scratchpad);
832 COPY(sc.b_scratchpad);
833 COPY(sc.obmc_scratchpad);
840 COPY(me.map_generation);
852 int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
856 // FIXME copy only needed parts
858 backup_duplicate_context(&bak, dst);
859 memcpy(dst, src, sizeof(MpegEncContext));
860 backup_duplicate_context(dst, &bak);
861 for (i = 0; i < 12; i++) {
862 dst->pblocks[i] = &dst->block[i];
864 if (dst->avctx->codec_tag == AV_RL32("VCR2")) {
866 FFSWAP(void *, dst->pblocks[4], dst->pblocks[5]);
868 if (!dst->sc.edge_emu_buffer &&
869 (ret = ff_mpeg_framesize_alloc(dst->avctx, &dst->me,
870 &dst->sc, 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) {
895 memcpy(s, s1, sizeof(MpegEncContext));
898 s->bitstream_buffer = NULL;
899 s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
901 if (s1->context_initialized){
902 // s->picture_range_start += MAX_PICTURE_COUNT;
903 // s->picture_range_end += MAX_PICTURE_COUNT;
905 if((err = ff_mpv_common_init(s)) < 0){
906 memset(s, 0, sizeof(MpegEncContext));
913 if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
914 s->context_reinit = 0;
915 s->height = s1->height;
916 s->width = s1->width;
917 if ((ret = ff_mpv_common_frame_size_change(s)) < 0)
921 s->avctx->coded_height = s1->avctx->coded_height;
922 s->avctx->coded_width = s1->avctx->coded_width;
923 s->avctx->width = s1->avctx->width;
924 s->avctx->height = s1->avctx->height;
926 s->coded_picture_number = s1->coded_picture_number;
927 s->picture_number = s1->picture_number;
929 av_assert0(!s->picture || s->picture != s1->picture);
931 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
932 ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
933 if (s1->picture[i].f->buf[0] &&
934 (ret = ff_mpeg_ref_picture(s->avctx, &s->picture[i], &s1->picture[i])) < 0)
938 #define UPDATE_PICTURE(pic)\
940 ff_mpeg_unref_picture(s->avctx, &s->pic);\
941 if (s1->pic.f && s1->pic.f->buf[0])\
942 ret = ff_mpeg_ref_picture(s->avctx, &s->pic, &s1->pic);\
944 ret = update_picture_tables(&s->pic, &s1->pic);\
949 UPDATE_PICTURE(current_picture);
950 UPDATE_PICTURE(last_picture);
951 UPDATE_PICTURE(next_picture);
953 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
954 ((pic && pic >= old_ctx->picture && \
955 pic < old_ctx->picture + MAX_PICTURE_COUNT) ? \
956 &new_ctx->picture[pic - old_ctx->picture] : NULL)
958 s->last_picture_ptr = REBASE_PICTURE(s1->last_picture_ptr, s, s1);
959 s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
960 s->next_picture_ptr = REBASE_PICTURE(s1->next_picture_ptr, s, s1);
962 // Error/bug resilience
963 s->next_p_frame_damaged = s1->next_p_frame_damaged;
964 s->workaround_bugs = s1->workaround_bugs;
965 s->padding_bug_score = s1->padding_bug_score;
968 memcpy(&s->last_time_base, &s1->last_time_base,
969 (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) -
970 (char *) &s1->last_time_base);
973 s->max_b_frames = s1->max_b_frames;
974 s->low_delay = s1->low_delay;
975 s->droppable = s1->droppable;
977 // DivX handling (doesn't work)
978 s->divx_packed = s1->divx_packed;
980 if (s1->bitstream_buffer) {
981 if (s1->bitstream_buffer_size +
982 FF_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size) {
983 av_fast_malloc(&s->bitstream_buffer,
984 &s->allocated_bitstream_buffer_size,
985 s1->allocated_bitstream_buffer_size);
986 if (!s->bitstream_buffer) {
987 s->bitstream_buffer_size = 0;
988 return AVERROR(ENOMEM);
991 s->bitstream_buffer_size = s1->bitstream_buffer_size;
992 memcpy(s->bitstream_buffer, s1->bitstream_buffer,
993 s1->bitstream_buffer_size);
994 memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
995 FF_INPUT_BUFFER_PADDING_SIZE);
998 // linesize dependend scratch buffer allocation
999 if (!s->sc.edge_emu_buffer)
1001 if (ff_mpeg_framesize_alloc(s->avctx, &s->me,
1002 &s->sc, s1->linesize) < 0) {
1003 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
1004 "scratch buffers.\n");
1005 return AVERROR(ENOMEM);
1008 av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
1009 "be allocated due to unknown size.\n");
1012 // MPEG2/interlacing info
1013 memcpy(&s->progressive_sequence, &s1->progressive_sequence,
1014 (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
1016 if (!s1->first_field) {
1017 s->last_pict_type = s1->pict_type;
1018 if (s1->current_picture_ptr)
1019 s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f->quality;
1026 * Set the given MpegEncContext to common defaults
1027 * (same for encoding and decoding).
1028 * The changed fields will not depend upon the
1029 * prior state of the MpegEncContext.
1031 void ff_mpv_common_defaults(MpegEncContext *s)
1033 s->y_dc_scale_table =
1034 s->c_dc_scale_table = ff_mpeg1_dc_scale_table;
1035 s->chroma_qscale_table = ff_default_chroma_qscale_table;
1036 s->progressive_frame = 1;
1037 s->progressive_sequence = 1;
1038 s->picture_structure = PICT_FRAME;
1040 s->coded_picture_number = 0;
1041 s->picture_number = 0;
1046 s->slice_context_count = 1;
1050 * Set the given MpegEncContext to defaults for decoding.
1051 * the changed fields will not depend upon
1052 * the prior state of the MpegEncContext.
1054 void ff_mpv_decode_defaults(MpegEncContext *s)
1056 ff_mpv_common_defaults(s);
1059 void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx)
1062 s->width = avctx->coded_width;
1063 s->height = avctx->coded_height;
1064 s->codec_id = avctx->codec->id;
1065 s->workaround_bugs = avctx->workaround_bugs;
1067 /* convert fourcc to upper case */
1068 s->codec_tag = avpriv_toupper4(avctx->codec_tag);
1071 static int init_er(MpegEncContext *s)
1073 ERContext *er = &s->er;
1074 int mb_array_size = s->mb_height * s->mb_stride;
1077 er->avctx = s->avctx;
1079 er->mb_index2xy = s->mb_index2xy;
1080 er->mb_num = s->mb_num;
1081 er->mb_width = s->mb_width;
1082 er->mb_height = s->mb_height;
1083 er->mb_stride = s->mb_stride;
1084 er->b8_stride = s->b8_stride;
1086 er->er_temp_buffer = av_malloc(s->mb_height * s->mb_stride);
1087 er->error_status_table = av_mallocz(mb_array_size);
1088 if (!er->er_temp_buffer || !er->error_status_table)
1091 er->mbskip_table = s->mbskip_table;
1092 er->mbintra_table = s->mbintra_table;
1094 for (i = 0; i < FF_ARRAY_ELEMS(s->dc_val); i++)
1095 er->dc_val[i] = s->dc_val[i];
1097 er->decode_mb = mpeg_er_decode_mb;
1102 av_freep(&er->er_temp_buffer);
1103 av_freep(&er->error_status_table);
1104 return AVERROR(ENOMEM);
1108 * Initialize and allocates MpegEncContext fields dependent on the resolution.
1110 static int init_context_frame(MpegEncContext *s)
1112 int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
1114 s->mb_width = (s->width + 15) / 16;
1115 s->mb_stride = s->mb_width + 1;
1116 s->b8_stride = s->mb_width * 2 + 1;
1117 mb_array_size = s->mb_height * s->mb_stride;
1118 mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
1120 /* set default edge pos, will be overridden
1121 * in decode_header if needed */
1122 s->h_edge_pos = s->mb_width * 16;
1123 s->v_edge_pos = s->mb_height * 16;
1125 s->mb_num = s->mb_width * s->mb_height;
1130 s->block_wrap[3] = s->b8_stride;
1132 s->block_wrap[5] = s->mb_stride;
1134 y_size = s->b8_stride * (2 * s->mb_height + 1);
1135 c_size = s->mb_stride * (s->mb_height + 1);
1136 yc_size = y_size + 2 * c_size;
1138 if (s->mb_height & 1)
1139 yc_size += 2*s->b8_stride + 2*s->mb_stride;
1141 FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int), fail); // error ressilience code looks cleaner with this
1142 for (y = 0; y < s->mb_height; y++)
1143 for (x = 0; x < s->mb_width; x++)
1144 s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
1146 s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
1149 /* Allocate MV tables */
1150 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1151 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1152 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1153 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1154 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1155 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
1156 s->p_mv_table = s->p_mv_table_base + s->mb_stride + 1;
1157 s->b_forw_mv_table = s->b_forw_mv_table_base + s->mb_stride + 1;
1158 s->b_back_mv_table = s->b_back_mv_table_base + s->mb_stride + 1;
1159 s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
1160 s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
1161 s->b_direct_mv_table = s->b_direct_mv_table_base + s->mb_stride + 1;
1163 /* Allocate MB type table */
1164 FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding
1166 FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
1168 FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
1169 mb_array_size * sizeof(float), fail);
1170 FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab,
1171 mb_array_size * sizeof(float), fail);
1175 if (s->codec_id == AV_CODEC_ID_MPEG4 ||
1176 (s->avctx->flags & CODEC_FLAG_INTERLACED_ME)) {
1177 /* interlaced direct mode decoding tables */
1178 for (i = 0; i < 2; i++) {
1180 for (j = 0; j < 2; j++) {
1181 for (k = 0; k < 2; k++) {
1182 FF_ALLOCZ_OR_GOTO(s->avctx,
1183 s->b_field_mv_table_base[i][j][k],
1184 mv_table_size * 2 * sizeof(int16_t),
1186 s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
1189 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
1190 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
1191 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
1193 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
1196 if (s->out_format == FMT_H263) {
1198 FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride, fail);
1199 s->coded_block = s->coded_block_base + s->b8_stride + 1;
1201 /* cbp, ac_pred, pred_dir */
1202 FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table , mb_array_size * sizeof(uint8_t), fail);
1203 FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail);
1206 if (s->h263_pred || s->h263_plus || !s->encoding) {
1208 // MN: we need these for error resilience of intra-frames
1209 FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
1210 s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
1211 s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
1212 s->dc_val[2] = s->dc_val[1] + c_size;
1213 for (i = 0; i < yc_size; i++)
1214 s->dc_val_base[i] = 1024;
1217 /* which mb is a intra block */
1218 FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
1219 memset(s->mbintra_table, 1, mb_array_size);
1221 /* init macroblock skip table */
1222 FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
1223 // Note the + 1 is for a quicker mpeg4 slice_end detection
1227 return AVERROR(ENOMEM);
1231 * init common structure for both encoder and decoder.
1232 * this assumes that some variables like width/height are already set
1234 av_cold int ff_mpv_common_init(MpegEncContext *s)
1237 int nb_slices = (HAVE_THREADS &&
1238 s->avctx->active_thread_type & FF_THREAD_SLICE) ?
1239 s->avctx->thread_count : 1;
1241 if (s->encoding && s->avctx->slices)
1242 nb_slices = s->avctx->slices;
1244 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1245 s->mb_height = (s->height + 31) / 32 * 2;
1247 s->mb_height = (s->height + 15) / 16;
1249 if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
1250 av_log(s->avctx, AV_LOG_ERROR,
1251 "decoding to AV_PIX_FMT_NONE is not supported.\n");
1255 if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
1258 max_slices = FFMIN(MAX_THREADS, s->mb_height);
1260 max_slices = MAX_THREADS;
1261 av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
1262 " reducing to %d\n", nb_slices, max_slices);
1263 nb_slices = max_slices;
1266 if ((s->width || s->height) &&
1267 av_image_check_size(s->width, s->height, 0, s->avctx))
1272 /* set chroma shifts */
1273 avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,
1275 &s->chroma_y_shift);
1278 FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
1279 MAX_PICTURE_COUNT * sizeof(Picture), fail);
1280 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1281 s->picture[i].f = av_frame_alloc();
1282 if (!s->picture[i].f)
1285 memset(&s->next_picture, 0, sizeof(s->next_picture));
1286 memset(&s->last_picture, 0, sizeof(s->last_picture));
1287 memset(&s->current_picture, 0, sizeof(s->current_picture));
1288 memset(&s->new_picture, 0, sizeof(s->new_picture));
1289 s->next_picture.f = av_frame_alloc();
1290 if (!s->next_picture.f)
1292 s->last_picture.f = av_frame_alloc();
1293 if (!s->last_picture.f)
1295 s->current_picture.f = av_frame_alloc();
1296 if (!s->current_picture.f)
1298 s->new_picture.f = av_frame_alloc();
1299 if (!s->new_picture.f)
1302 if (init_context_frame(s))
1305 s->parse_context.state = -1;
1307 s->context_initialized = 1;
1308 memset(s->thread_context, 0, sizeof(s->thread_context));
1309 s->thread_context[0] = s;
1311 // if (s->width && s->height) {
1312 if (nb_slices > 1) {
1313 for (i = 0; i < nb_slices; i++) {
1315 s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
1316 if (!s->thread_context[i])
1319 if (init_duplicate_context(s->thread_context[i]) < 0)
1321 s->thread_context[i]->start_mb_y =
1322 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1323 s->thread_context[i]->end_mb_y =
1324 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1327 if (init_duplicate_context(s) < 0)
1330 s->end_mb_y = s->mb_height;
1332 s->slice_context_count = nb_slices;
1337 ff_mpv_common_end(s);
1342 * Frees and resets MpegEncContext fields depending on the resolution.
1343 * Is used during resolution changes to avoid a full reinitialization of the
1346 static void free_context_frame(MpegEncContext *s)
1350 av_freep(&s->mb_type);
1351 av_freep(&s->p_mv_table_base);
1352 av_freep(&s->b_forw_mv_table_base);
1353 av_freep(&s->b_back_mv_table_base);
1354 av_freep(&s->b_bidir_forw_mv_table_base);
1355 av_freep(&s->b_bidir_back_mv_table_base);
1356 av_freep(&s->b_direct_mv_table_base);
1357 s->p_mv_table = NULL;
1358 s->b_forw_mv_table = NULL;
1359 s->b_back_mv_table = NULL;
1360 s->b_bidir_forw_mv_table = NULL;
1361 s->b_bidir_back_mv_table = NULL;
1362 s->b_direct_mv_table = NULL;
1363 for (i = 0; i < 2; i++) {
1364 for (j = 0; j < 2; j++) {
1365 for (k = 0; k < 2; k++) {
1366 av_freep(&s->b_field_mv_table_base[i][j][k]);
1367 s->b_field_mv_table[i][j][k] = NULL;
1369 av_freep(&s->b_field_select_table[i][j]);
1370 av_freep(&s->p_field_mv_table_base[i][j]);
1371 s->p_field_mv_table[i][j] = NULL;
1373 av_freep(&s->p_field_select_table[i]);
1376 av_freep(&s->dc_val_base);
1377 av_freep(&s->coded_block_base);
1378 av_freep(&s->mbintra_table);
1379 av_freep(&s->cbp_table);
1380 av_freep(&s->pred_dir_table);
1382 av_freep(&s->mbskip_table);
1384 av_freep(&s->er.error_status_table);
1385 av_freep(&s->er.er_temp_buffer);
1386 av_freep(&s->mb_index2xy);
1387 av_freep(&s->lambda_table);
1389 av_freep(&s->cplx_tab);
1390 av_freep(&s->bits_tab);
1392 s->linesize = s->uvlinesize = 0;
1395 int ff_mpv_common_frame_size_change(MpegEncContext *s)
1399 if (!s->context_initialized)
1400 return AVERROR(EINVAL);
1402 if (s->slice_context_count > 1) {
1403 for (i = 0; i < s->slice_context_count; i++) {
1404 free_duplicate_context(s->thread_context[i]);
1406 for (i = 1; i < s->slice_context_count; i++) {
1407 av_freep(&s->thread_context[i]);
1410 free_duplicate_context(s);
1412 free_context_frame(s);
1415 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1416 s->picture[i].needs_realloc = 1;
1419 s->last_picture_ptr =
1420 s->next_picture_ptr =
1421 s->current_picture_ptr = NULL;
1424 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1425 s->mb_height = (s->height + 31) / 32 * 2;
1427 s->mb_height = (s->height + 15) / 16;
1429 if ((s->width || s->height) &&
1430 (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0)
1433 if ((err = init_context_frame(s)))
1436 memset(s->thread_context, 0, sizeof(s->thread_context));
1437 s->thread_context[0] = s;
1439 if (s->width && s->height) {
1440 int nb_slices = s->slice_context_count;
1441 if (nb_slices > 1) {
1442 for (i = 0; i < nb_slices; i++) {
1444 s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
1445 if (!s->thread_context[i]) {
1446 err = AVERROR(ENOMEM);
1450 if ((err = init_duplicate_context(s->thread_context[i])) < 0)
1452 s->thread_context[i]->start_mb_y =
1453 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1454 s->thread_context[i]->end_mb_y =
1455 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1458 err = init_duplicate_context(s);
1462 s->end_mb_y = s->mb_height;
1464 s->slice_context_count = nb_slices;
1469 ff_mpv_common_end(s);
1473 /* init common structure for both encoder and decoder */
1474 void ff_mpv_common_end(MpegEncContext *s)
1478 if (s->slice_context_count > 1) {
1479 for (i = 0; i < s->slice_context_count; i++) {
1480 free_duplicate_context(s->thread_context[i]);
1482 for (i = 1; i < s->slice_context_count; i++) {
1483 av_freep(&s->thread_context[i]);
1485 s->slice_context_count = 1;
1486 } else free_duplicate_context(s);
1488 av_freep(&s->parse_context.buffer);
1489 s->parse_context.buffer_size = 0;
1491 av_freep(&s->bitstream_buffer);
1492 s->allocated_bitstream_buffer_size = 0;
1495 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1496 ff_free_picture_tables(&s->picture[i]);
1497 ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1498 av_frame_free(&s->picture[i].f);
1501 av_freep(&s->picture);
1502 ff_free_picture_tables(&s->last_picture);
1503 ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1504 av_frame_free(&s->last_picture.f);
1505 ff_free_picture_tables(&s->current_picture);
1506 ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1507 av_frame_free(&s->current_picture.f);
1508 ff_free_picture_tables(&s->next_picture);
1509 ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1510 av_frame_free(&s->next_picture.f);
1511 ff_free_picture_tables(&s->new_picture);
1512 ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1513 av_frame_free(&s->new_picture.f);
1515 free_context_frame(s);
1517 s->context_initialized = 0;
1518 s->last_picture_ptr =
1519 s->next_picture_ptr =
1520 s->current_picture_ptr = NULL;
1521 s->linesize = s->uvlinesize = 0;
1524 static void release_unused_pictures(AVCodecContext *avctx, Picture *picture)
1528 /* release non reference frames */
1529 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1530 if (!picture[i].reference)
1531 ff_mpeg_unref_picture(avctx, &picture[i]);
1535 static inline int pic_is_unused(Picture *pic)
1537 if (!pic->f->buf[0])
1539 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
1544 static int find_unused_picture(AVCodecContext *avctx, Picture *picture, int shared)
1549 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1550 if (!picture[i].f->buf[0])
1554 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1555 if (pic_is_unused(&picture[i]))
1560 av_log(avctx, AV_LOG_FATAL,
1561 "Internal error, picture buffer overflow\n");
1562 /* We could return -1, but the codec would crash trying to draw into a
1563 * non-existing frame anyway. This is safer than waiting for a random crash.
1564 * Also the return of this is never useful, an encoder must only allocate
1565 * as much as allowed in the specification. This has no relationship to how
1566 * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
1567 * enough for such valid streams).
1568 * Plus, a decoder has to check stream validity and remove frames if too
1569 * many reference frames are around. Waiting for "OOM" is not correct at
1570 * all. Similarly, missing reference frames have to be replaced by
1571 * interpolated/MC frames, anything else is a bug in the codec ...
1577 int ff_find_unused_picture(AVCodecContext *avctx, Picture *picture, int shared)
1579 int ret = find_unused_picture(avctx, picture, shared);
1581 if (ret >= 0 && ret < MAX_PICTURE_COUNT) {
1582 if (picture[ret].needs_realloc) {
1583 picture[ret].needs_realloc = 0;
1584 ff_free_picture_tables(&picture[ret]);
1585 ff_mpeg_unref_picture(avctx, &picture[ret]);
1591 static void gray_frame(AVFrame *frame)
1593 int i, h_chroma_shift, v_chroma_shift;
1595 av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift);
1597 for(i=0; i<frame->height; i++)
1598 memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width);
1599 for(i=0; i<FF_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) {
1600 memset(frame->data[1] + frame->linesize[1]*i,
1601 0x80, FF_CEIL_RSHIFT(frame->width, h_chroma_shift));
1602 memset(frame->data[2] + frame->linesize[2]*i,
1603 0x80, FF_CEIL_RSHIFT(frame->width, h_chroma_shift));
1608 * generic function called after decoding
1609 * the header and before a frame is decoded.
1611 int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1617 if (!ff_thread_can_start_frame(avctx)) {
1618 av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1622 /* mark & release old frames */
1623 if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1624 s->last_picture_ptr != s->next_picture_ptr &&
1625 s->last_picture_ptr->f->buf[0]) {
1626 ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1629 /* release forgotten pictures */
1630 /* if (mpeg124/h263) */
1631 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1632 if (&s->picture[i] != s->last_picture_ptr &&
1633 &s->picture[i] != s->next_picture_ptr &&
1634 s->picture[i].reference && !s->picture[i].needs_realloc) {
1635 if (!(avctx->active_thread_type & FF_THREAD_FRAME))
1636 av_log(avctx, AV_LOG_ERROR,
1637 "releasing zombie picture\n");
1638 ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1642 ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1644 release_unused_pictures(s->avctx, s->picture);
1646 if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) {
1647 // we already have a unused image
1648 // (maybe it was set before reading the header)
1649 pic = s->current_picture_ptr;
1651 i = ff_find_unused_picture(s->avctx, s->picture, 0);
1653 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1656 pic = &s->picture[i];
1660 if (!s->droppable) {
1661 if (s->pict_type != AV_PICTURE_TYPE_B)
1665 pic->f->coded_picture_number = s->coded_picture_number++;
1667 if (alloc_picture(s, pic, 0) < 0)
1670 s->current_picture_ptr = pic;
1671 // FIXME use only the vars from current_pic
1672 s->current_picture_ptr->f->top_field_first = s->top_field_first;
1673 if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
1674 s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1675 if (s->picture_structure != PICT_FRAME)
1676 s->current_picture_ptr->f->top_field_first =
1677 (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
1679 s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame &&
1680 !s->progressive_sequence;
1681 s->current_picture_ptr->field_picture = s->picture_structure != PICT_FRAME;
1683 s->current_picture_ptr->f->pict_type = s->pict_type;
1684 // if (s->avctx->flags && CODEC_FLAG_QSCALE)
1685 // s->current_picture_ptr->quality = s->new_picture_ptr->quality;
1686 s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1688 if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1689 s->current_picture_ptr)) < 0)
1692 if (s->pict_type != AV_PICTURE_TYPE_B) {
1693 s->last_picture_ptr = s->next_picture_ptr;
1695 s->next_picture_ptr = s->current_picture_ptr;
1697 ff_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
1698 s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1699 s->last_picture_ptr ? s->last_picture_ptr->f->data[0] : NULL,
1700 s->next_picture_ptr ? s->next_picture_ptr->f->data[0] : NULL,
1701 s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL,
1702 s->pict_type, s->droppable);
1704 if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) &&
1705 (s->pict_type != AV_PICTURE_TYPE_I ||
1706 s->picture_structure != PICT_FRAME)) {
1707 int h_chroma_shift, v_chroma_shift;
1708 av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1709 &h_chroma_shift, &v_chroma_shift);
1710 if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
1711 av_log(avctx, AV_LOG_DEBUG,
1712 "allocating dummy last picture for B frame\n");
1713 else if (s->pict_type != AV_PICTURE_TYPE_I)
1714 av_log(avctx, AV_LOG_ERROR,
1715 "warning: first frame is no keyframe\n");
1716 else if (s->picture_structure != PICT_FRAME)
1717 av_log(avctx, AV_LOG_DEBUG,
1718 "allocate dummy last picture for field based first keyframe\n");
1720 /* Allocate a dummy frame */
1721 i = ff_find_unused_picture(s->avctx, s->picture, 0);
1723 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1726 s->last_picture_ptr = &s->picture[i];
1728 s->last_picture_ptr->reference = 3;
1729 s->last_picture_ptr->f->key_frame = 0;
1730 s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1732 if (alloc_picture(s, s->last_picture_ptr, 0) < 0) {
1733 s->last_picture_ptr = NULL;
1737 if (!avctx->hwaccel && !(avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)) {
1738 for(i=0; i<avctx->height; i++)
1739 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
1740 0x80, avctx->width);
1741 if (s->last_picture_ptr->f->data[2]) {
1742 for(i=0; i<FF_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
1743 memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
1744 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1745 memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
1746 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1750 if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
1751 for(i=0; i<avctx->height; i++)
1752 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
1756 ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
1757 ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
1759 if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) &&
1760 s->pict_type == AV_PICTURE_TYPE_B) {
1761 /* Allocate a dummy frame */
1762 i = ff_find_unused_picture(s->avctx, s->picture, 0);
1764 av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1767 s->next_picture_ptr = &s->picture[i];
1769 s->next_picture_ptr->reference = 3;
1770 s->next_picture_ptr->f->key_frame = 0;
1771 s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1773 if (alloc_picture(s, s->next_picture_ptr, 0) < 0) {
1774 s->next_picture_ptr = NULL;
1777 ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
1778 ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
1781 #if 0 // BUFREF-FIXME
1782 memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
1783 memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
1785 if (s->last_picture_ptr) {
1786 ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1787 if (s->last_picture_ptr->f->buf[0] &&
1788 (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1789 s->last_picture_ptr)) < 0)
1792 if (s->next_picture_ptr) {
1793 ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1794 if (s->next_picture_ptr->f->buf[0] &&
1795 (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1796 s->next_picture_ptr)) < 0)
1800 av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
1801 s->last_picture_ptr->f->buf[0]));
1803 if (s->picture_structure!= PICT_FRAME) {
1805 for (i = 0; i < 4; i++) {
1806 if (s->picture_structure == PICT_BOTTOM_FIELD) {
1807 s->current_picture.f->data[i] +=
1808 s->current_picture.f->linesize[i];
1810 s->current_picture.f->linesize[i] *= 2;
1811 s->last_picture.f->linesize[i] *= 2;
1812 s->next_picture.f->linesize[i] *= 2;
1816 /* set dequantizer, we can't do it during init as
1817 * it might change for mpeg4 and we can't do it in the header
1818 * decode as init is not called for mpeg4 there yet */
1819 if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1820 s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1821 s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1822 } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1823 s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1824 s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1826 s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1827 s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1830 if (s->avctx->debug & FF_DEBUG_NOMC) {
1831 gray_frame(s->current_picture_ptr->f);
1837 /* called after a frame has been decoded. */
1838 void ff_mpv_frame_end(MpegEncContext *s)
1842 if (s->current_picture.reference)
1843 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1848 static int clip_line(int *sx, int *sy, int *ex, int *ey, int maxx)
1851 return clip_line(ex, ey, sx, sy, maxx);
1856 *sy = *ey + (*sy - *ey) * (int64_t)*ex / (*ex - *sx);
1863 *ey = *sy + (*ey - *sy) * (int64_t)(maxx - *sx) / (*ex - *sx);
1871 * Draw a line from (ex, ey) -> (sx, sy).
1872 * @param w width of the image
1873 * @param h height of the image
1874 * @param stride stride/linesize of the image
1875 * @param color color of the arrow
1877 static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey,
1878 int w, int h, int stride, int color)
1882 if (clip_line(&sx, &sy, &ex, &ey, w - 1))
1884 if (clip_line(&sy, &sx, &ey, &ex, h - 1))
1887 sx = av_clip(sx, 0, w - 1);
1888 sy = av_clip(sy, 0, h - 1);
1889 ex = av_clip(ex, 0, w - 1);
1890 ey = av_clip(ey, 0, h - 1);
1892 buf[sy * stride + sx] += color;
1894 if (FFABS(ex - sx) > FFABS(ey - sy)) {
1896 FFSWAP(int, sx, ex);
1897 FFSWAP(int, sy, ey);
1899 buf += sx + sy * stride;
1901 f = ((ey - sy) << 16) / ex;
1902 for (x = 0; x <= ex; x++) {
1904 fr = (x * f) & 0xFFFF;
1905 buf[y * stride + x] += (color * (0x10000 - fr)) >> 16;
1906 if(fr) buf[(y + 1) * stride + x] += (color * fr ) >> 16;
1910 FFSWAP(int, sx, ex);
1911 FFSWAP(int, sy, ey);
1913 buf += sx + sy * stride;
1916 f = ((ex - sx) << 16) / ey;
1919 for(y= 0; y <= ey; y++){
1921 fr = (y*f) & 0xFFFF;
1922 buf[y * stride + x] += (color * (0x10000 - fr)) >> 16;
1923 if(fr) buf[y * stride + x + 1] += (color * fr ) >> 16;
1929 * Draw an arrow from (ex, ey) -> (sx, sy).
1930 * @param w width of the image
1931 * @param h height of the image
1932 * @param stride stride/linesize of the image
1933 * @param color color of the arrow
1935 static void draw_arrow(uint8_t *buf, int sx, int sy, int ex,
1936 int ey, int w, int h, int stride, int color, int tail, int direction)
1941 FFSWAP(int, sx, ex);
1942 FFSWAP(int, sy, ey);
1945 sx = av_clip(sx, -100, w + 100);
1946 sy = av_clip(sy, -100, h + 100);
1947 ex = av_clip(ex, -100, w + 100);
1948 ey = av_clip(ey, -100, h + 100);
1953 if (dx * dx + dy * dy > 3 * 3) {
1956 int length = ff_sqrt((rx * rx + ry * ry) << 8);
1958 // FIXME subpixel accuracy
1959 rx = ROUNDED_DIV(rx * 3 << 4, length);
1960 ry = ROUNDED_DIV(ry * 3 << 4, length);
1967 draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
1968 draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
1970 draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
1974 static int add_mb(AVMotionVector *mb, uint32_t mb_type,
1975 int dst_x, int dst_y,
1976 int src_x, int src_y,
1979 mb->w = IS_8X8(mb_type) || IS_8X16(mb_type) ? 8 : 16;
1980 mb->h = IS_8X8(mb_type) || IS_16X8(mb_type) ? 8 : 16;
1985 mb->source = direction ? 1 : -1;
1986 mb->flags = 0; // XXX: does mb_type contain extra information that could be exported here?
1991 * Print debugging info for the given picture.
1993 void ff_print_debug_info2(AVCodecContext *avctx, AVFrame *pict, uint8_t *mbskip_table,
1994 uint32_t *mbtype_table, int8_t *qscale_table, int16_t (*motion_val[2])[2],
1996 int mb_width, int mb_height, int mb_stride, int quarter_sample)
1998 if ((avctx->flags2 & CODEC_FLAG2_EXPORT_MVS) && mbtype_table && motion_val[0]) {
1999 const int shift = 1 + quarter_sample;
2000 const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1;
2001 const int mv_stride = (mb_width << mv_sample_log2) +
2002 (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1);
2003 int mb_x, mb_y, mbcount = 0;
2005 /* size is width * height * 2 * 4 where 2 is for directions and 4 is
2006 * for the maximum number of MB (4 MB in case of IS_8x8) */
2007 AVMotionVector *mvs = av_malloc_array(mb_width * mb_height, 2 * 4 * sizeof(AVMotionVector));
2011 for (mb_y = 0; mb_y < mb_height; mb_y++) {
2012 for (mb_x = 0; mb_x < mb_width; mb_x++) {
2013 int i, direction, mb_type = mbtype_table[mb_x + mb_y * mb_stride];
2014 for (direction = 0; direction < 2; direction++) {
2015 if (!USES_LIST(mb_type, direction))
2017 if (IS_8X8(mb_type)) {
2018 for (i = 0; i < 4; i++) {
2019 int sx = mb_x * 16 + 4 + 8 * (i & 1);
2020 int sy = mb_y * 16 + 4 + 8 * (i >> 1);
2021 int xy = (mb_x * 2 + (i & 1) +
2022 (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
2023 int mx = (motion_val[direction][xy][0] >> shift) + sx;
2024 int my = (motion_val[direction][xy][1] >> shift) + sy;
2025 mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx, my, direction);
2027 } else if (IS_16X8(mb_type)) {
2028 for (i = 0; i < 2; i++) {
2029 int sx = mb_x * 16 + 8;
2030 int sy = mb_y * 16 + 4 + 8 * i;
2031 int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
2032 int mx = (motion_val[direction][xy][0] >> shift);
2033 int my = (motion_val[direction][xy][1] >> shift);
2035 if (IS_INTERLACED(mb_type))
2038 mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx + sx, my + sy, direction);
2040 } else if (IS_8X16(mb_type)) {
2041 for (i = 0; i < 2; i++) {
2042 int sx = mb_x * 16 + 4 + 8 * i;
2043 int sy = mb_y * 16 + 8;
2044 int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
2045 int mx = motion_val[direction][xy][0] >> shift;
2046 int my = motion_val[direction][xy][1] >> shift;
2048 if (IS_INTERLACED(mb_type))
2051 mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx + sx, my + sy, direction);
2054 int sx = mb_x * 16 + 8;
2055 int sy = mb_y * 16 + 8;
2056 int xy = (mb_x + mb_y * mv_stride) << mv_sample_log2;
2057 int mx = (motion_val[direction][xy][0]>>shift) + sx;
2058 int my = (motion_val[direction][xy][1]>>shift) + sy;
2059 mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx, my, direction);
2066 AVFrameSideData *sd;
2068 av_log(avctx, AV_LOG_DEBUG, "Adding %d MVs info to frame %d\n", mbcount, avctx->frame_number);
2069 sd = av_frame_new_side_data(pict, AV_FRAME_DATA_MOTION_VECTORS, mbcount * sizeof(AVMotionVector));
2074 memcpy(sd->data, mvs, mbcount * sizeof(AVMotionVector));
2080 /* TODO: export all the following to make them accessible for users (and filters) */
2081 if (avctx->hwaccel || !mbtype_table
2082 || (avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU))
2086 if (avctx->debug & (FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)) {
2089 av_log(avctx, AV_LOG_DEBUG, "New frame, type: %c\n",
2090 av_get_picture_type_char(pict->pict_type));
2091 for (y = 0; y < mb_height; y++) {
2092 for (x = 0; x < mb_width; x++) {
2093 if (avctx->debug & FF_DEBUG_SKIP) {
2094 int count = mbskip_table ? mbskip_table[x + y * mb_stride] : 0;
2097 av_log(avctx, AV_LOG_DEBUG, "%1d", count);
2099 if (avctx->debug & FF_DEBUG_QP) {
2100 av_log(avctx, AV_LOG_DEBUG, "%2d",
2101 qscale_table[x + y * mb_stride]);
2103 if (avctx->debug & FF_DEBUG_MB_TYPE) {
2104 int mb_type = mbtype_table[x + y * mb_stride];
2105 // Type & MV direction
2106 if (IS_PCM(mb_type))
2107 av_log(avctx, AV_LOG_DEBUG, "P");
2108 else if (IS_INTRA(mb_type) && IS_ACPRED(mb_type))
2109 av_log(avctx, AV_LOG_DEBUG, "A");
2110 else if (IS_INTRA4x4(mb_type))
2111 av_log(avctx, AV_LOG_DEBUG, "i");
2112 else if (IS_INTRA16x16(mb_type))
2113 av_log(avctx, AV_LOG_DEBUG, "I");
2114 else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type))
2115 av_log(avctx, AV_LOG_DEBUG, "d");
2116 else if (IS_DIRECT(mb_type))
2117 av_log(avctx, AV_LOG_DEBUG, "D");
2118 else if (IS_GMC(mb_type) && IS_SKIP(mb_type))
2119 av_log(avctx, AV_LOG_DEBUG, "g");
2120 else if (IS_GMC(mb_type))
2121 av_log(avctx, AV_LOG_DEBUG, "G");
2122 else if (IS_SKIP(mb_type))
2123 av_log(avctx, AV_LOG_DEBUG, "S");
2124 else if (!USES_LIST(mb_type, 1))
2125 av_log(avctx, AV_LOG_DEBUG, ">");
2126 else if (!USES_LIST(mb_type, 0))
2127 av_log(avctx, AV_LOG_DEBUG, "<");
2129 av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
2130 av_log(avctx, AV_LOG_DEBUG, "X");
2134 if (IS_8X8(mb_type))
2135 av_log(avctx, AV_LOG_DEBUG, "+");
2136 else if (IS_16X8(mb_type))
2137 av_log(avctx, AV_LOG_DEBUG, "-");
2138 else if (IS_8X16(mb_type))
2139 av_log(avctx, AV_LOG_DEBUG, "|");
2140 else if (IS_INTRA(mb_type) || IS_16X16(mb_type))
2141 av_log(avctx, AV_LOG_DEBUG, " ");
2143 av_log(avctx, AV_LOG_DEBUG, "?");
2146 if (IS_INTERLACED(mb_type))
2147 av_log(avctx, AV_LOG_DEBUG, "=");
2149 av_log(avctx, AV_LOG_DEBUG, " ");
2152 av_log(avctx, AV_LOG_DEBUG, "\n");
2156 if ((avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
2157 (avctx->debug_mv)) {
2160 int h_chroma_shift, v_chroma_shift, block_height;
2162 const int shift = 1 + quarter_sample;
2164 const int width = avctx->width;
2165 const int height = avctx->height;
2167 const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1;
2168 const int mv_stride = (mb_width << mv_sample_log2) +
2169 (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1);
2171 *low_delay = 0; // needed to see the vectors without trashing the buffers
2173 avcodec_get_chroma_sub_sample(avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
2175 av_frame_make_writable(pict);
2177 pict->opaque = NULL;
2179 ptr = pict->data[0];
2181 block_height = 16 >> v_chroma_shift;
2183 for (mb_y = 0; mb_y < mb_height; mb_y++) {
2185 for (mb_x = 0; mb_x < mb_width; mb_x++) {
2186 const int mb_index = mb_x + mb_y * mb_stride;
2188 if ((avctx->debug_mv) && motion_val[0]) {
2190 for (type = 0; type < 3; type++) {
2194 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_P_FOR)) ||
2195 (pict->pict_type!= AV_PICTURE_TYPE_P))
2200 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_FOR)) ||
2201 (pict->pict_type!= AV_PICTURE_TYPE_B))
2206 if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_BACK)) ||
2207 (pict->pict_type!= AV_PICTURE_TYPE_B))
2212 if (!USES_LIST(mbtype_table[mb_index], direction))
2215 if (IS_8X8(mbtype_table[mb_index])) {
2217 for (i = 0; i < 4; i++) {
2218 int sx = mb_x * 16 + 4 + 8 * (i & 1);
2219 int sy = mb_y * 16 + 4 + 8 * (i >> 1);
2220 int xy = (mb_x * 2 + (i & 1) +
2221 (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
2222 int mx = (motion_val[direction][xy][0] >> shift) + sx;
2223 int my = (motion_val[direction][xy][1] >> shift) + sy;
2224 draw_arrow(ptr, sx, sy, mx, my, width,
2225 height, pict->linesize[0], 100, 0, direction);
2227 } else if (IS_16X8(mbtype_table[mb_index])) {
2229 for (i = 0; i < 2; i++) {
2230 int sx = mb_x * 16 + 8;
2231 int sy = mb_y * 16 + 4 + 8 * i;
2232 int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
2233 int mx = (motion_val[direction][xy][0] >> shift);
2234 int my = (motion_val[direction][xy][1] >> shift);
2236 if (IS_INTERLACED(mbtype_table[mb_index]))
2239 draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
2240 height, pict->linesize[0], 100, 0, direction);
2242 } else if (IS_8X16(mbtype_table[mb_index])) {
2244 for (i = 0; i < 2; i++) {
2245 int sx = mb_x * 16 + 4 + 8 * i;
2246 int sy = mb_y * 16 + 8;
2247 int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
2248 int mx = motion_val[direction][xy][0] >> shift;
2249 int my = motion_val[direction][xy][1] >> shift;
2251 if (IS_INTERLACED(mbtype_table[mb_index]))
2254 draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
2255 height, pict->linesize[0], 100, 0, direction);
2258 int sx= mb_x * 16 + 8;
2259 int sy= mb_y * 16 + 8;
2260 int xy= (mb_x + mb_y * mv_stride) << mv_sample_log2;
2261 int mx= (motion_val[direction][xy][0]>>shift) + sx;
2262 int my= (motion_val[direction][xy][1]>>shift) + sy;
2263 draw_arrow(ptr, sx, sy, mx, my, width, height, pict->linesize[0], 100, 0, direction);
2268 if ((avctx->debug & FF_DEBUG_VIS_QP)) {
2269 uint64_t c = (qscale_table[mb_index] * 128 / 31) *
2270 0x0101010101010101ULL;
2272 for (y = 0; y < block_height; y++) {
2273 *(uint64_t *)(pict->data[1] + 8 * mb_x +
2274 (block_height * mb_y + y) *
2275 pict->linesize[1]) = c;
2276 *(uint64_t *)(pict->data[2] + 8 * mb_x +
2277 (block_height * mb_y + y) *
2278 pict->linesize[2]) = c;
2281 if ((avctx->debug & FF_DEBUG_VIS_MB_TYPE) &&
2283 int mb_type = mbtype_table[mb_index];
2286 #define COLOR(theta, r) \
2287 u = (int)(128 + r * cos(theta * 3.141592 / 180)); \
2288 v = (int)(128 + r * sin(theta * 3.141592 / 180));
2292 if (IS_PCM(mb_type)) {
2294 } else if ((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) ||
2295 IS_INTRA16x16(mb_type)) {
2297 } else if (IS_INTRA4x4(mb_type)) {
2299 } else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type)) {
2301 } else if (IS_DIRECT(mb_type)) {
2303 } else if (IS_GMC(mb_type) && IS_SKIP(mb_type)) {
2305 } else if (IS_GMC(mb_type)) {
2307 } else if (IS_SKIP(mb_type)) {
2309 } else if (!USES_LIST(mb_type, 1)) {
2311 } else if (!USES_LIST(mb_type, 0)) {
2314 av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
2318 u *= 0x0101010101010101ULL;
2319 v *= 0x0101010101010101ULL;
2320 for (y = 0; y < block_height; y++) {
2321 *(uint64_t *)(pict->data[1] + 8 * mb_x +
2322 (block_height * mb_y + y) * pict->linesize[1]) = u;
2323 *(uint64_t *)(pict->data[2] + 8 * mb_x +
2324 (block_height * mb_y + y) * pict->linesize[2]) = v;
2328 if (IS_8X8(mb_type) || IS_16X8(mb_type)) {
2329 *(uint64_t *)(pict->data[0] + 16 * mb_x + 0 +
2330 (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
2331 *(uint64_t *)(pict->data[0] + 16 * mb_x + 8 +
2332 (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
2334 if (IS_8X8(mb_type) || IS_8X16(mb_type)) {
2335 for (y = 0; y < 16; y++)
2336 pict->data[0][16 * mb_x + 8 + (16 * mb_y + y) *
2337 pict->linesize[0]] ^= 0x80;
2339 if (IS_8X8(mb_type) && mv_sample_log2 >= 2) {
2340 int dm = 1 << (mv_sample_log2 - 2);
2341 for (i = 0; i < 4; i++) {
2342 int sx = mb_x * 16 + 8 * (i & 1);
2343 int sy = mb_y * 16 + 8 * (i >> 1);
2344 int xy = (mb_x * 2 + (i & 1) +
2345 (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
2347 int32_t *mv = (int32_t *) &motion_val[0][xy];
2348 if (mv[0] != mv[dm] ||
2349 mv[dm * mv_stride] != mv[dm * (mv_stride + 1)])
2350 for (y = 0; y < 8; y++)
2351 pict->data[0][sx + 4 + (sy + y) * pict->linesize[0]] ^= 0x80;
2352 if (mv[0] != mv[dm * mv_stride] || mv[dm] != mv[dm * (mv_stride + 1)])
2353 *(uint64_t *)(pict->data[0] + sx + (sy + 4) *
2354 pict->linesize[0]) ^= 0x8080808080808080ULL;
2358 if (IS_INTERLACED(mb_type) &&
2359 avctx->codec->id == AV_CODEC_ID_H264) {
2364 mbskip_table[mb_index] = 0;
2370 void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
2372 ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
2373 p->qscale_table, p->motion_val, &s->low_delay,
2374 s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
2377 int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
2379 AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
2380 int offset = 2*s->mb_stride + 1;
2382 return AVERROR(ENOMEM);
2383 av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
2384 ref->size -= offset;
2385 ref->data += offset;
2386 return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
2389 static inline int hpel_motion_lowres(MpegEncContext *s,
2390 uint8_t *dest, uint8_t *src,
2391 int field_based, int field_select,
2392 int src_x, int src_y,
2393 int width, int height, ptrdiff_t stride,
2394 int h_edge_pos, int v_edge_pos,
2395 int w, int h, h264_chroma_mc_func *pix_op,
2396 int motion_x, int motion_y)
2398 const int lowres = s->avctx->lowres;
2399 const int op_index = FFMIN(lowres, 3);
2400 const int s_mask = (2 << lowres) - 1;
2404 if (s->quarter_sample) {
2409 sx = motion_x & s_mask;
2410 sy = motion_y & s_mask;
2411 src_x += motion_x >> lowres + 1;
2412 src_y += motion_y >> lowres + 1;
2414 src += src_y * stride + src_x;
2416 if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w, 0) ||
2417 (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
2418 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
2419 s->linesize, s->linesize,
2420 w + 1, (h + 1) << field_based,
2421 src_x, src_y << field_based,
2422 h_edge_pos, v_edge_pos);
2423 src = s->sc.edge_emu_buffer;
2427 sx = (sx << 2) >> lowres;
2428 sy = (sy << 2) >> lowres;
2431 pix_op[op_index](dest, src, stride, h, sx, sy);
2435 /* apply one mpeg motion vector to the three components */
2436 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
2443 uint8_t **ref_picture,
2444 h264_chroma_mc_func *pix_op,
2445 int motion_x, int motion_y,
2448 uint8_t *ptr_y, *ptr_cb, *ptr_cr;
2449 int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
2450 ptrdiff_t uvlinesize, linesize;
2451 const int lowres = s->avctx->lowres;
2452 const int op_index = FFMIN(lowres-1+s->chroma_x_shift, 3);
2453 const int block_s = 8>>lowres;
2454 const int s_mask = (2 << lowres) - 1;
2455 const int h_edge_pos = s->h_edge_pos >> lowres;
2456 const int v_edge_pos = s->v_edge_pos >> lowres;
2457 linesize = s->current_picture.f->linesize[0] << field_based;
2458 uvlinesize = s->current_picture.f->linesize[1] << field_based;
2460 // FIXME obviously not perfect but qpel will not work in lowres anyway
2461 if (s->quarter_sample) {
2467 motion_y += (bottom_field - field_select)*((1 << lowres)-1);
2470 sx = motion_x & s_mask;
2471 sy = motion_y & s_mask;
2472 src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
2473 src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
2475 if (s->out_format == FMT_H263) {
2476 uvsx = ((motion_x >> 1) & s_mask) | (sx & 1);
2477 uvsy = ((motion_y >> 1) & s_mask) | (sy & 1);
2478 uvsrc_x = src_x >> 1;
2479 uvsrc_y = src_y >> 1;
2480 } else if (s->out_format == FMT_H261) {
2481 // even chroma mv's are full pel in H261
2484 uvsx = (2 * mx) & s_mask;
2485 uvsy = (2 * my) & s_mask;
2486 uvsrc_x = s->mb_x * block_s + (mx >> lowres);
2487 uvsrc_y = mb_y * block_s + (my >> lowres);
2489 if(s->chroma_y_shift){
2494 uvsrc_x = s->mb_x * block_s + (mx >> lowres + 1);
2495 uvsrc_y = (mb_y * block_s >> field_based) + (my >> lowres + 1);
2497 if(s->chroma_x_shift){
2501 uvsy = motion_y & s_mask;
2503 uvsrc_x = s->mb_x*block_s + (mx >> (lowres+1));
2506 uvsx = motion_x & s_mask;
2507 uvsy = motion_y & s_mask;
2514 ptr_y = ref_picture[0] + src_y * linesize + src_x;
2515 ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
2516 ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
2518 if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s, 0) || uvsrc_y<0 ||
2519 (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
2520 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
2521 linesize >> field_based, linesize >> field_based,
2522 17, 17 + field_based,
2523 src_x, src_y << field_based, h_edge_pos,
2525 ptr_y = s->sc.edge_emu_buffer;
2526 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
2527 uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
2528 uint8_t *vbuf =ubuf + 9 * s->uvlinesize;
2529 s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
2530 uvlinesize >> field_based, uvlinesize >> field_based,
2532 uvsrc_x, uvsrc_y << field_based,
2533 h_edge_pos >> 1, v_edge_pos >> 1);
2534 s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
2535 uvlinesize >> field_based,uvlinesize >> field_based,
2537 uvsrc_x, uvsrc_y << field_based,
2538 h_edge_pos >> 1, v_edge_pos >> 1);
2544 // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
2546 dest_y += s->linesize;
2547 dest_cb += s->uvlinesize;
2548 dest_cr += s->uvlinesize;
2552 ptr_y += s->linesize;
2553 ptr_cb += s->uvlinesize;
2554 ptr_cr += s->uvlinesize;
2557 sx = (sx << 2) >> lowres;
2558 sy = (sy << 2) >> lowres;
2559 pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
2561 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
2562 int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
2563 uvsx = (uvsx << 2) >> lowres;
2564 uvsy = (uvsy << 2) >> lowres;
2566 pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
2567 pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
2570 // FIXME h261 lowres loop filter
2573 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
2574 uint8_t *dest_cb, uint8_t *dest_cr,
2575 uint8_t **ref_picture,
2576 h264_chroma_mc_func * pix_op,
2579 const int lowres = s->avctx->lowres;
2580 const int op_index = FFMIN(lowres, 3);
2581 const int block_s = 8 >> lowres;
2582 const int s_mask = (2 << lowres) - 1;
2583 const int h_edge_pos = s->h_edge_pos >> lowres + 1;
2584 const int v_edge_pos = s->v_edge_pos >> lowres + 1;
2585 int emu = 0, src_x, src_y, sx, sy;
2589 if (s->quarter_sample) {
2594 /* In case of 8X8, we construct a single chroma motion vector
2595 with a special rounding */
2596 mx = ff_h263_round_chroma(mx);
2597 my = ff_h263_round_chroma(my);
2601 src_x = s->mb_x * block_s + (mx >> lowres + 1);
2602 src_y = s->mb_y * block_s + (my >> lowres + 1);
2604 offset = src_y * s->uvlinesize + src_x;
2605 ptr = ref_picture[1] + offset;
2606 if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
2607 (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
2608 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
2609 s->uvlinesize, s->uvlinesize,
2611 src_x, src_y, h_edge_pos, v_edge_pos);
2612 ptr = s->sc.edge_emu_buffer;
2615 sx = (sx << 2) >> lowres;
2616 sy = (sy << 2) >> lowres;
2617 pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
2619 ptr = ref_picture[2] + offset;
2621 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
2622 s->uvlinesize, s->uvlinesize,
2624 src_x, src_y, h_edge_pos, v_edge_pos);
2625 ptr = s->sc.edge_emu_buffer;
2627 pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
2631 * motion compensation of a single macroblock
2633 * @param dest_y luma destination pointer
2634 * @param dest_cb chroma cb/u destination pointer
2635 * @param dest_cr chroma cr/v destination pointer
2636 * @param dir direction (0->forward, 1->backward)
2637 * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
2638 * @param pix_op halfpel motion compensation function (average or put normally)
2639 * the motion vectors are taken from s->mv and the MV type from s->mv_type
2641 static inline void MPV_motion_lowres(MpegEncContext *s,
2642 uint8_t *dest_y, uint8_t *dest_cb,
2644 int dir, uint8_t **ref_picture,
2645 h264_chroma_mc_func *pix_op)
2649 const int lowres = s->avctx->lowres;
2650 const int block_s = 8 >>lowres;
2655 switch (s->mv_type) {
2657 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2659 ref_picture, pix_op,
2660 s->mv[dir][0][0], s->mv[dir][0][1],
2666 for (i = 0; i < 4; i++) {
2667 hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
2668 s->linesize) * block_s,
2669 ref_picture[0], 0, 0,
2670 (2 * mb_x + (i & 1)) * block_s,
2671 (2 * mb_y + (i >> 1)) * block_s,
2672 s->width, s->height, s->linesize,
2673 s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
2674 block_s, block_s, pix_op,
2675 s->mv[dir][i][0], s->mv[dir][i][1]);
2677 mx += s->mv[dir][i][0];
2678 my += s->mv[dir][i][1];
2681 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY))
2682 chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
2686 if (s->picture_structure == PICT_FRAME) {
2688 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2689 1, 0, s->field_select[dir][0],
2690 ref_picture, pix_op,
2691 s->mv[dir][0][0], s->mv[dir][0][1],
2694 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2695 1, 1, s->field_select[dir][1],
2696 ref_picture, pix_op,
2697 s->mv[dir][1][0], s->mv[dir][1][1],
2700 if (s->picture_structure != s->field_select[dir][0] + 1 &&
2701 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
2702 ref_picture = s->current_picture_ptr->f->data;
2705 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2706 0, 0, s->field_select[dir][0],
2707 ref_picture, pix_op,
2709 s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
2713 for (i = 0; i < 2; i++) {
2714 uint8_t **ref2picture;
2716 if (s->picture_structure == s->field_select[dir][i] + 1 ||
2717 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
2718 ref2picture = ref_picture;
2720 ref2picture = s->current_picture_ptr->f->data;
2723 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2724 0, 0, s->field_select[dir][i],
2725 ref2picture, pix_op,
2726 s->mv[dir][i][0], s->mv[dir][i][1] +
2727 2 * block_s * i, block_s, mb_y >> 1);
2729 dest_y += 2 * block_s * s->linesize;
2730 dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
2731 dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
2735 if (s->picture_structure == PICT_FRAME) {
2736 for (i = 0; i < 2; i++) {
2738 for (j = 0; j < 2; j++) {
2739 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2741 ref_picture, pix_op,
2742 s->mv[dir][2 * i + j][0],
2743 s->mv[dir][2 * i + j][1],
2746 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
2749 for (i = 0; i < 2; i++) {
2750 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
2751 0, 0, s->picture_structure != i + 1,
2752 ref_picture, pix_op,
2753 s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
2754 2 * block_s, mb_y >> 1);
2756 // after put we make avg of the same block
2757 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
2759 // opposite parity is always in the same
2760 // frame if this is second field
2761 if (!s->first_field) {
2762 ref_picture = s->current_picture_ptr->f->data;
2773 * find the lowest MB row referenced in the MVs
2775 int ff_mpv_lowest_referenced_row(MpegEncContext *s, int dir)
2777 int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
2778 int my, off, i, mvs;
2780 if (s->picture_structure != PICT_FRAME || s->mcsel)
2783 switch (s->mv_type) {
2797 for (i = 0; i < mvs; i++) {
2798 my = s->mv[dir][i][1];
2799 my_max = FFMAX(my_max, my);
2800 my_min = FFMIN(my_min, my);
2803 off = ((FFMAX(-my_min, my_max)<<qpel_shift) + 63) >> 6;
2805 return av_clip(s->mb_y + off, 0, s->mb_height - 1);
2807 return s->mb_height-1;
2810 /* put block[] to dest[] */
2811 static inline void put_dct(MpegEncContext *s,
2812 int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
2814 s->dct_unquantize_intra(s, block, i, qscale);
2815 s->idsp.idct_put(dest, line_size, block);
2818 /* add block[] to dest[] */
2819 static inline void add_dct(MpegEncContext *s,
2820 int16_t *block, int i, uint8_t *dest, int line_size)
2822 if (s->block_last_index[i] >= 0) {
2823 s->idsp.idct_add(dest, line_size, block);
2827 static inline void add_dequant_dct(MpegEncContext *s,
2828 int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
2830 if (s->block_last_index[i] >= 0) {
2831 s->dct_unquantize_inter(s, block, i, qscale);
2833 s->idsp.idct_add(dest, line_size, block);
2838 * Clean dc, ac, coded_block for the current non-intra MB.
2840 void ff_clean_intra_table_entries(MpegEncContext *s)
2842 int wrap = s->b8_stride;
2843 int xy = s->block_index[0];
2846 s->dc_val[0][xy + 1 ] =
2847 s->dc_val[0][xy + wrap] =
2848 s->dc_val[0][xy + 1 + wrap] = 1024;
2850 memset(s->ac_val[0][xy ], 0, 32 * sizeof(int16_t));
2851 memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
2852 if (s->msmpeg4_version>=3) {
2853 s->coded_block[xy ] =
2854 s->coded_block[xy + 1 ] =
2855 s->coded_block[xy + wrap] =
2856 s->coded_block[xy + 1 + wrap] = 0;
2859 wrap = s->mb_stride;
2860 xy = s->mb_x + s->mb_y * wrap;
2862 s->dc_val[2][xy] = 1024;
2864 memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
2865 memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
2867 s->mbintra_table[xy]= 0;
2870 /* generic function called after a macroblock has been parsed by the
2871 decoder or after it has been encoded by the encoder.
2873 Important variables used:
2874 s->mb_intra : true if intra macroblock
2875 s->mv_dir : motion vector direction
2876 s->mv_type : motion vector type
2877 s->mv : motion vector
2878 s->interlaced_dct : true if interlaced dct used (mpeg2)
2880 static av_always_inline
2881 void mpv_decode_mb_internal(MpegEncContext *s, int16_t block[12][64],
2882 int lowres_flag, int is_mpeg12)
2884 const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
2887 s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
2888 s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
2892 if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
2893 /* print DCT coefficients */
2895 av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
2897 for(j=0; j<64; j++){
2898 av_log(s->avctx, AV_LOG_DEBUG, "%5d",
2899 block[i][s->idsp.idct_permutation[j]]);
2901 av_log(s->avctx, AV_LOG_DEBUG, "\n");
2905 s->current_picture.qscale_table[mb_xy] = s->qscale;
2907 /* update DC predictors for P macroblocks */
2909 if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
2910 if(s->mbintra_table[mb_xy])
2911 ff_clean_intra_table_entries(s);
2915 s->last_dc[2] = 128 << s->intra_dc_precision;
2918 else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
2919 s->mbintra_table[mb_xy]=1;
2921 if ((s->avctx->flags & CODEC_FLAG_PSNR) || s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor ||
2922 !(s->encoding && (s->intra_only || s->pict_type == AV_PICTURE_TYPE_B) &&
2923 s->avctx->mb_decision != FF_MB_DECISION_RD)) { // FIXME precalc
2924 uint8_t *dest_y, *dest_cb, *dest_cr;
2925 int dct_linesize, dct_offset;
2926 op_pixels_func (*op_pix)[4];
2927 qpel_mc_func (*op_qpix)[16];
2928 const int linesize = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
2929 const int uvlinesize = s->current_picture.f->linesize[1];
2930 const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
2931 const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
2933 /* avoid copy if macroblock skipped in last frame too */
2934 /* skip only during decoding as we might trash the buffers during encoding a bit */
2936 uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
2938 if (s->mb_skipped) {
2940 av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
2942 } else if(!s->current_picture.reference) {
2945 *mbskip_ptr = 0; /* not skipped */
2949 dct_linesize = linesize << s->interlaced_dct;
2950 dct_offset = s->interlaced_dct ? linesize : linesize * block_size;
2954 dest_cb= s->dest[1];
2955 dest_cr= s->dest[2];
2957 dest_y = s->sc.b_scratchpad;
2958 dest_cb= s->sc.b_scratchpad+16*linesize;
2959 dest_cr= s->sc.b_scratchpad+32*linesize;
2963 /* motion handling */
2964 /* decoding or more than one mb_type (MC was already done otherwise) */
2967 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
2968 if (s->mv_dir & MV_DIR_FORWARD) {
2969 ff_thread_await_progress(&s->last_picture_ptr->tf,
2970 ff_mpv_lowest_referenced_row(s, 0),
2973 if (s->mv_dir & MV_DIR_BACKWARD) {
2974 ff_thread_await_progress(&s->next_picture_ptr->tf,
2975 ff_mpv_lowest_referenced_row(s, 1),
2981 h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
2983 if (s->mv_dir & MV_DIR_FORWARD) {
2984 MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
2985 op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
2987 if (s->mv_dir & MV_DIR_BACKWARD) {
2988 MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
2991 op_qpix = s->me.qpel_put;
2992 if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
2993 op_pix = s->hdsp.put_pixels_tab;
2995 op_pix = s->hdsp.put_no_rnd_pixels_tab;
2997 if (s->mv_dir & MV_DIR_FORWARD) {
2998 ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
2999 op_pix = s->hdsp.avg_pixels_tab;
3000 op_qpix= s->me.qpel_avg;
3002 if (s->mv_dir & MV_DIR_BACKWARD) {
3003 ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
3008 /* skip dequant / idct if we are really late ;) */
3009 if(s->avctx->skip_idct){
3010 if( (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
3011 ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
3012 || s->avctx->skip_idct >= AVDISCARD_ALL)
3016 /* add dct residue */
3017 if(s->encoding || !( s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
3018 || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
3019 add_dequant_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale);
3020 add_dequant_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale);
3021 add_dequant_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale);
3022 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
3024 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3025 if (s->chroma_y_shift){
3026 add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
3027 add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
3031 add_dequant_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale);
3032 add_dequant_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale);
3033 add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
3034 add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
3037 } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
3038 add_dct(s, block[0], 0, dest_y , dct_linesize);
3039 add_dct(s, block[1], 1, dest_y + block_size, dct_linesize);
3040 add_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize);
3041 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
3043 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3044 if(s->chroma_y_shift){//Chroma420
3045 add_dct(s, block[4], 4, dest_cb, uvlinesize);
3046 add_dct(s, block[5], 5, dest_cr, uvlinesize);
3049 dct_linesize = uvlinesize << s->interlaced_dct;
3050 dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
3052 add_dct(s, block[4], 4, dest_cb, dct_linesize);
3053 add_dct(s, block[5], 5, dest_cr, dct_linesize);
3054 add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
3055 add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
3056 if(!s->chroma_x_shift){//Chroma444
3057 add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
3058 add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
3059 add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
3060 add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
3065 else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
3066 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
3069 /* dct only in intra block */
3070 if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
3071 put_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale);
3072 put_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale);
3073 put_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale);
3074 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
3076 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3077 if(s->chroma_y_shift){
3078 put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
3079 put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
3083 put_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale);
3084 put_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale);
3085 put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
3086 put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
3090 s->idsp.idct_put(dest_y, dct_linesize, block[0]);
3091 s->idsp.idct_put(dest_y + block_size, dct_linesize, block[1]);
3092 s->idsp.idct_put(dest_y + dct_offset, dct_linesize, block[2]);
3093 s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
3095 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3096 if(s->chroma_y_shift){
3097 s->idsp.idct_put(dest_cb, uvlinesize, block[4]);
3098 s->idsp.idct_put(dest_cr, uvlinesize, block[5]);
3101 dct_linesize = uvlinesize << s->interlaced_dct;
3102 dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
3104 s->idsp.idct_put(dest_cb, dct_linesize, block[4]);
3105 s->idsp.idct_put(dest_cr, dct_linesize, block[5]);
3106 s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
3107 s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
3108 if(!s->chroma_x_shift){//Chroma444
3109 s->idsp.idct_put(dest_cb + block_size, dct_linesize, block[8]);
3110 s->idsp.idct_put(dest_cr + block_size, dct_linesize, block[9]);
3111 s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
3112 s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
3120 s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y , linesize,16);
3121 if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
3122 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
3123 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
3129 void ff_mpv_decode_mb(MpegEncContext *s, int16_t block[12][64])
3132 if(s->out_format == FMT_MPEG1) {
3133 if(s->avctx->lowres) mpv_decode_mb_internal(s, block, 1, 1);
3134 else mpv_decode_mb_internal(s, block, 0, 1);
3137 if(s->avctx->lowres) mpv_decode_mb_internal(s, block, 1, 0);
3138 else mpv_decode_mb_internal(s, block, 0, 0);
3141 void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
3143 ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
3144 s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
3145 s->first_field, s->low_delay);
3148 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
3149 const int linesize = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
3150 const int uvlinesize = s->current_picture.f->linesize[1];
3151 const int mb_size= 4 - s->avctx->lowres;
3153 s->block_index[0]= s->b8_stride*(s->mb_y*2 ) - 2 + s->mb_x*2;
3154 s->block_index[1]= s->b8_stride*(s->mb_y*2 ) - 1 + s->mb_x*2;
3155 s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
3156 s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
3157 s->block_index[4]= s->mb_stride*(s->mb_y + 1) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
3158 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;
3159 //block_index is not used by mpeg2, so it is not affected by chroma_format
3161 s->dest[0] = s->current_picture.f->data[0] + (int)((s->mb_x - 1U) << mb_size);
3162 s->dest[1] = s->current_picture.f->data[1] + (int)((s->mb_x - 1U) << (mb_size - s->chroma_x_shift));
3163 s->dest[2] = s->current_picture.f->data[2] + (int)((s->mb_x - 1U) << (mb_size - s->chroma_x_shift));
3165 if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
3167 if(s->picture_structure==PICT_FRAME){
3168 s->dest[0] += s->mb_y * linesize << mb_size;
3169 s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
3170 s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
3172 s->dest[0] += (s->mb_y>>1) * linesize << mb_size;
3173 s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
3174 s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
3175 av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
3181 * Permute an 8x8 block.
3182 * @param block the block which will be permuted according to the given permutation vector
3183 * @param permutation the permutation vector
3184 * @param last the last non zero coefficient in scantable order, used to speed the permutation up
3185 * @param scantable the used scantable, this is only used to speed the permutation up, the block is not
3186 * (inverse) permutated to scantable order!
3188 void ff_block_permute(int16_t *block, uint8_t *permutation, const uint8_t *scantable, int last)
3194 //if(permutation[1]==1) return; //FIXME it is ok but not clean and might fail for some permutations
3196 for(i=0; i<=last; i++){
3197 const int j= scantable[i];
3202 for(i=0; i<=last; i++){
3203 const int j= scantable[i];
3204 const int perm_j= permutation[j];
3205 block[perm_j]= temp[j];
3209 void ff_mpeg_flush(AVCodecContext *avctx){
3211 MpegEncContext *s = avctx->priv_data;
3213 if (!s || !s->picture)
3216 for (i = 0; i < MAX_PICTURE_COUNT; i++)
3217 ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
3218 s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
3220 ff_mpeg_unref_picture(s->avctx, &s->current_picture);
3221 ff_mpeg_unref_picture(s->avctx, &s->last_picture);
3222 ff_mpeg_unref_picture(s->avctx, &s->next_picture);
3224 s->mb_x= s->mb_y= 0;
3227 s->parse_context.state= -1;
3228 s->parse_context.frame_start_found= 0;
3229 s->parse_context.overread= 0;
3230 s->parse_context.overread_index= 0;
3231 s->parse_context.index= 0;
3232 s->parse_context.last_index= 0;
3233 s->bitstream_buffer_size=0;
3238 * set qscale and update qscale dependent variables.
3240 void ff_set_qscale(MpegEncContext * s, int qscale)
3244 else if (qscale > 31)
3248 s->chroma_qscale= s->chroma_qscale_table[qscale];
3250 s->y_dc_scale= s->y_dc_scale_table[ qscale ];
3251 s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
3254 void ff_mpv_report_decode_progress(MpegEncContext *s)
3256 if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
3257 ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);