2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #define UNCHECKED_BITSTREAM_READER 1
30 #include "libavutil/imgutils.h"
31 #include "libavutil/opt.h"
34 #include "cabac_functions.h"
36 #include "error_resilience.h"
38 #include "mpegvideo.h"
41 #include "h264chroma.h"
42 #include "h264_mvpred.h"
45 #include "rectangle.h"
48 #include "vdpau_internal.h"
49 #include "libavutil/avassert.h"
54 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
56 static const uint8_t rem6[QP_MAX_NUM + 1] = {
57 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
58 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
59 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
60 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
64 static const uint8_t div6[QP_MAX_NUM + 1] = {
65 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
66 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
67 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
68 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
72 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
73 #if CONFIG_H264_DXVA2_HWACCEL
76 #if CONFIG_H264_VAAPI_HWACCEL
79 #if CONFIG_H264_VDA_HWACCEL
82 #if CONFIG_H264_VDPAU_HWACCEL
89 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
90 #if CONFIG_H264_DXVA2_HWACCEL
93 #if CONFIG_H264_VAAPI_HWACCEL
96 #if CONFIG_H264_VDA_HWACCEL
99 #if CONFIG_H264_VDPAU_HWACCEL
106 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
108 H264Context *h = avctx->priv_data;
109 return h ? h->sps.num_reorder_frames : 0;
112 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
114 int mb_x, int mb_y, int mb_intra, int mb_skipped)
116 H264Context *h = opaque;
120 h->mb_xy = mb_x + mb_y * h->mb_stride;
121 memset(h->non_zero_count_cache, 0, sizeof(h->non_zero_count_cache));
122 av_assert1(ref >= 0);
123 /* FIXME: It is possible albeit uncommon that slice references
124 * differ between slices. We take the easy approach and ignore
125 * it for now. If this turns out to have any relevance in
126 * practice then correct remapping should be added. */
127 if (ref >= h->ref_count[0])
129 if (!h->ref_list[0][ref].f.data[0]) {
130 av_log(h->avctx, AV_LOG_DEBUG, "Reference not available for error concealing\n");
133 if ((h->ref_list[0][ref].reference&3) != 3) {
134 av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n");
137 fill_rectangle(&h->cur_pic.ref_index[0][4 * h->mb_xy],
139 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
140 fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8,
141 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
143 h->mb_field_decoding_flag = 0;
144 ff_h264_hl_decode_mb(h);
147 void ff_h264_draw_horiz_band(H264Context *h, int y, int height)
149 AVCodecContext *avctx = h->avctx;
150 Picture *cur = &h->cur_pic;
151 Picture *last = h->ref_list[0][0].f.data[0] ? &h->ref_list[0][0] : NULL;
152 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
153 int vshift = desc->log2_chroma_h;
154 const int field_pic = h->picture_structure != PICT_FRAME;
160 height = FFMIN(height, avctx->height - y);
162 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
165 if (avctx->draw_horiz_band) {
167 int offset[AV_NUM_DATA_POINTERS];
170 if (cur->f.pict_type == AV_PICTURE_TYPE_B || h->low_delay ||
171 (avctx->slice_flags & SLICE_FLAG_CODED_ORDER))
178 offset[0] = y * src->linesize[0];
180 offset[2] = (y >> vshift) * src->linesize[1];
181 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
186 avctx->draw_horiz_band(avctx, src, offset,
187 y, h->picture_structure, height);
191 static void unref_picture(H264Context *h, Picture *pic)
193 int off = offsetof(Picture, tf) + sizeof(pic->tf);
199 ff_thread_release_buffer(h->avctx, &pic->tf);
200 av_buffer_unref(&pic->hwaccel_priv_buf);
202 av_buffer_unref(&pic->qscale_table_buf);
203 av_buffer_unref(&pic->mb_type_buf);
204 for (i = 0; i < 2; i++) {
205 av_buffer_unref(&pic->motion_val_buf[i]);
206 av_buffer_unref(&pic->ref_index_buf[i]);
209 memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
212 static void release_unused_pictures(H264Context *h, int remove_current)
216 /* release non reference frames */
217 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
218 if (h->DPB[i].f.data[0] && !h->DPB[i].reference &&
219 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
220 unref_picture(h, &h->DPB[i]);
225 static int ref_picture(H264Context *h, Picture *dst, Picture *src)
229 av_assert0(!dst->f.buf[0]);
230 av_assert0(src->f.buf[0]);
234 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
239 dst->qscale_table_buf = av_buffer_ref(src->qscale_table_buf);
240 dst->mb_type_buf = av_buffer_ref(src->mb_type_buf);
241 if (!dst->qscale_table_buf || !dst->mb_type_buf)
243 dst->qscale_table = src->qscale_table;
244 dst->mb_type = src->mb_type;
246 for (i = 0; i < 2; i ++) {
247 dst->motion_val_buf[i] = av_buffer_ref(src->motion_val_buf[i]);
248 dst->ref_index_buf[i] = av_buffer_ref(src->ref_index_buf[i]);
249 if (!dst->motion_val_buf[i] || !dst->ref_index_buf[i])
251 dst->motion_val[i] = src->motion_val[i];
252 dst->ref_index[i] = src->ref_index[i];
255 if (src->hwaccel_picture_private) {
256 dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
257 if (!dst->hwaccel_priv_buf)
259 dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
262 for (i = 0; i < 2; i++)
263 dst->field_poc[i] = src->field_poc[i];
265 memcpy(dst->ref_poc, src->ref_poc, sizeof(src->ref_poc));
266 memcpy(dst->ref_count, src->ref_count, sizeof(src->ref_count));
269 dst->frame_num = src->frame_num;
270 dst->mmco_reset = src->mmco_reset;
271 dst->pic_id = src->pic_id;
272 dst->long_ref = src->long_ref;
273 dst->mbaff = src->mbaff;
274 dst->field_picture = src->field_picture;
275 dst->needs_realloc = src->needs_realloc;
276 dst->reference = src->reference;
277 dst->sync = src->sync;
281 unref_picture(h, dst);
286 static int alloc_scratch_buffers(H264Context *h, int linesize)
288 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
290 if (h->bipred_scratchpad)
293 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
294 // edge emu needs blocksize + filter length - 1
295 // (= 21x21 for h264)
296 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
297 h->me.scratchpad = av_mallocz(alloc_size * 2 * 16 * 2);
299 if (!h->bipred_scratchpad || !h->edge_emu_buffer || !h->me.scratchpad) {
300 av_freep(&h->bipred_scratchpad);
301 av_freep(&h->edge_emu_buffer);
302 av_freep(&h->me.scratchpad);
303 return AVERROR(ENOMEM);
306 h->me.temp = h->me.scratchpad;
311 static int init_table_pools(H264Context *h)
313 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
314 const int mb_array_size = h->mb_stride * h->mb_height;
315 const int b4_stride = h->mb_width * 4 + 1;
316 const int b4_array_size = b4_stride * h->mb_height * 4;
318 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
320 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
321 sizeof(uint32_t), av_buffer_allocz);
322 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
323 sizeof(int16_t), av_buffer_allocz);
324 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
326 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
327 !h->ref_index_pool) {
328 av_buffer_pool_uninit(&h->qscale_table_pool);
329 av_buffer_pool_uninit(&h->mb_type_pool);
330 av_buffer_pool_uninit(&h->motion_val_pool);
331 av_buffer_pool_uninit(&h->ref_index_pool);
332 return AVERROR(ENOMEM);
338 static int alloc_picture(H264Context *h, Picture *pic)
342 av_assert0(!pic->f.data[0]);
345 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
346 AV_GET_BUFFER_FLAG_REF : 0);
350 h->linesize = pic->f.linesize[0];
351 h->uvlinesize = pic->f.linesize[1];
353 if (h->avctx->hwaccel) {
354 const AVHWAccel *hwaccel = h->avctx->hwaccel;
355 av_assert0(!pic->hwaccel_picture_private);
356 if (hwaccel->priv_data_size) {
357 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size);
358 if (!pic->hwaccel_priv_buf)
359 return AVERROR(ENOMEM);
360 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
364 if (!h->qscale_table_pool) {
365 ret = init_table_pools(h);
370 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
371 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
372 if (!pic->qscale_table_buf || !pic->mb_type_buf)
375 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
376 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
378 for (i = 0; i < 2; i++) {
379 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
380 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
381 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
384 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
385 pic->ref_index[i] = pic->ref_index_buf[i]->data;
390 unref_picture(h, pic);
391 return (ret < 0) ? ret : AVERROR(ENOMEM);
394 static inline int pic_is_unused(H264Context *h, Picture *pic)
396 if (pic->f.data[0] == NULL)
398 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
403 static int find_unused_picture(H264Context *h)
407 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
408 if (pic_is_unused(h, &h->DPB[i]))
411 if (i == MAX_PICTURE_COUNT)
412 return AVERROR_INVALIDDATA;
414 if (h->DPB[i].needs_realloc) {
415 h->DPB[i].needs_realloc = 0;
416 unref_picture(h, &h->DPB[i]);
423 * Check if the top & left blocks are available if needed and
424 * change the dc mode so it only uses the available blocks.
426 int ff_h264_check_intra4x4_pred_mode(H264Context *h)
428 static const int8_t top[12] = {
429 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
431 static const int8_t left[12] = {
432 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
436 if (!(h->top_samples_available & 0x8000)) {
437 for (i = 0; i < 4; i++) {
438 int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
440 av_log(h->avctx, AV_LOG_ERROR,
441 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
442 status, h->mb_x, h->mb_y);
445 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
450 if ((h->left_samples_available & 0x8888) != 0x8888) {
451 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
452 for (i = 0; i < 4; i++)
453 if (!(h->left_samples_available & mask[i])) {
454 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
456 av_log(h->avctx, AV_LOG_ERROR,
457 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
458 status, h->mb_x, h->mb_y);
461 h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
467 } // FIXME cleanup like ff_h264_check_intra_pred_mode
470 * Check if the top & left blocks are available if needed and
471 * change the dc mode so it only uses the available blocks.
473 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
475 static const int8_t top[7] = { LEFT_DC_PRED8x8, 1, -1, -1 };
476 static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
479 av_log(h->avctx, AV_LOG_ERROR,
480 "out of range intra chroma pred mode at %d %d\n",
485 if (!(h->top_samples_available & 0x8000)) {
488 av_log(h->avctx, AV_LOG_ERROR,
489 "top block unavailable for requested intra mode at %d %d\n",
495 if ((h->left_samples_available & 0x8080) != 0x8080) {
497 if (is_chroma && (h->left_samples_available & 0x8080)) {
498 // mad cow disease mode, aka MBAFF + constrained_intra_pred
499 mode = ALZHEIMER_DC_L0T_PRED8x8 +
500 (!(h->left_samples_available & 0x8000)) +
501 2 * (mode == DC_128_PRED8x8);
504 av_log(h->avctx, AV_LOG_ERROR,
505 "left block unavailable for requested intra mode at %d %d\n",
514 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
515 int *dst_length, int *consumed, int length)
521 // src[0]&0x80; // forbidden bit
522 h->nal_ref_idc = src[0] >> 5;
523 h->nal_unit_type = src[0] & 0x1F;
528 #define STARTCODE_TEST \
529 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
530 if (src[i + 2] != 3) { \
531 /* startcode, so we must be past the end */ \
536 #if HAVE_FAST_UNALIGNED
537 #define FIND_FIRST_ZERO \
538 if (i > 0 && !src[i]) \
543 for (i = 0; i + 1 < length; i += 9) {
544 if (!((~AV_RN64A(src + i) &
545 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
546 0x8000800080008080ULL))
553 for (i = 0; i + 1 < length; i += 5) {
554 if (!((~AV_RN32A(src + i) &
555 (AV_RN32A(src + i) - 0x01000101U)) &
564 for (i = 0; i + 1 < length; i += 2) {
567 if (i > 0 && src[i - 1] == 0)
573 // use second escape buffer for inter data
574 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
576 si = h->rbsp_buffer_size[bufidx];
577 av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
578 dst = h->rbsp_buffer[bufidx];
583 if(i>=length-1){ //no escaped 0
585 *consumed= length+1; //+1 for the header
586 if(h->avctx->flags2 & CODEC_FLAG2_FAST){
589 memcpy(dst, src, length);
596 while (si + 2 < length) {
597 // remove escapes (very rare 1:2^22)
598 if (src[si + 2] > 3) {
599 dst[di++] = src[si++];
600 dst[di++] = src[si++];
601 } else if (src[si] == 0 && src[si + 1] == 0) {
602 if (src[si + 2] == 3) { // escape
607 } else // next start code
611 dst[di++] = src[si++];
614 dst[di++] = src[si++];
617 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
620 *consumed = si + 1; // +1 for the header
621 /* FIXME store exact number of bits in the getbitcontext
622 * (it is needed for decoding) */
627 * Identify the exact end of the bitstream
628 * @return the length of the trailing, or 0 if damaged
630 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
635 tprintf(h->avctx, "rbsp trailing %X\n", v);
637 for (r = 1; r < 9; r++) {
645 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
646 int height, int y_offset, int list)
648 int raw_my = h->mv_cache[list][scan8[n]][1];
649 int filter_height_down = (raw_my & 3) ? 3 : 0;
650 int full_my = (raw_my >> 2) + y_offset;
651 int bottom = full_my + filter_height_down + height;
653 av_assert2(height >= 0);
655 return FFMAX(0, bottom);
658 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
659 int height, int y_offset, int list0,
660 int list1, int *nrefs)
664 y_offset += 16 * (h->mb_y >> MB_FIELD(h));
667 int ref_n = h->ref_cache[0][scan8[n]];
668 Picture *ref = &h->ref_list[0][ref_n];
670 // Error resilience puts the current picture in the ref list.
671 // Don't try to wait on these as it will cause a deadlock.
672 // Fields can wait on each other, though.
673 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
674 (ref->reference & 3) != h->picture_structure) {
675 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
676 if (refs[0][ref_n] < 0)
678 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
683 int ref_n = h->ref_cache[1][scan8[n]];
684 Picture *ref = &h->ref_list[1][ref_n];
686 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
687 (ref->reference & 3) != h->picture_structure) {
688 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
689 if (refs[1][ref_n] < 0)
691 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
697 * Wait until all reference frames are available for MC operations.
699 * @param h the H264 context
701 static void await_references(H264Context *h)
703 const int mb_xy = h->mb_xy;
704 const int mb_type = h->cur_pic.mb_type[mb_xy];
706 int nrefs[2] = { 0 };
709 memset(refs, -1, sizeof(refs));
711 if (IS_16X16(mb_type)) {
712 get_lowest_part_y(h, refs, 0, 16, 0,
713 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
714 } else if (IS_16X8(mb_type)) {
715 get_lowest_part_y(h, refs, 0, 8, 0,
716 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
717 get_lowest_part_y(h, refs, 8, 8, 8,
718 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
719 } else if (IS_8X16(mb_type)) {
720 get_lowest_part_y(h, refs, 0, 16, 0,
721 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
722 get_lowest_part_y(h, refs, 4, 16, 0,
723 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
727 av_assert2(IS_8X8(mb_type));
729 for (i = 0; i < 4; i++) {
730 const int sub_mb_type = h->sub_mb_type[i];
732 int y_offset = (i & 2) << 2;
734 if (IS_SUB_8X8(sub_mb_type)) {
735 get_lowest_part_y(h, refs, n, 8, y_offset,
736 IS_DIR(sub_mb_type, 0, 0),
737 IS_DIR(sub_mb_type, 0, 1),
739 } else if (IS_SUB_8X4(sub_mb_type)) {
740 get_lowest_part_y(h, refs, n, 4, y_offset,
741 IS_DIR(sub_mb_type, 0, 0),
742 IS_DIR(sub_mb_type, 0, 1),
744 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
745 IS_DIR(sub_mb_type, 0, 0),
746 IS_DIR(sub_mb_type, 0, 1),
748 } else if (IS_SUB_4X8(sub_mb_type)) {
749 get_lowest_part_y(h, refs, n, 8, y_offset,
750 IS_DIR(sub_mb_type, 0, 0),
751 IS_DIR(sub_mb_type, 0, 1),
753 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
754 IS_DIR(sub_mb_type, 0, 0),
755 IS_DIR(sub_mb_type, 0, 1),
759 av_assert2(IS_SUB_4X4(sub_mb_type));
760 for (j = 0; j < 4; j++) {
761 int sub_y_offset = y_offset + 2 * (j & 2);
762 get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
763 IS_DIR(sub_mb_type, 0, 0),
764 IS_DIR(sub_mb_type, 0, 1),
771 for (list = h->list_count - 1; list >= 0; list--)
772 for (ref = 0; ref < 48 && nrefs[list]; ref++) {
773 int row = refs[list][ref];
775 Picture *ref_pic = &h->ref_list[list][ref];
776 int ref_field = ref_pic->reference - 1;
777 int ref_field_picture = ref_pic->field_picture;
778 int pic_height = 16 * h->mb_height >> ref_field_picture;
783 if (!FIELD_PICTURE(h) && ref_field_picture) { // frame referencing two fields
784 ff_thread_await_progress(&ref_pic->tf,
785 FFMIN((row >> 1) - !(row & 1),
788 ff_thread_await_progress(&ref_pic->tf,
789 FFMIN((row >> 1), pic_height - 1),
791 } else if (FIELD_PICTURE(h) && !ref_field_picture) { // field referencing one field of a frame
792 ff_thread_await_progress(&ref_pic->tf,
793 FFMIN(row * 2 + ref_field,
796 } else if (FIELD_PICTURE(h)) {
797 ff_thread_await_progress(&ref_pic->tf,
798 FFMIN(row, pic_height - 1),
801 ff_thread_await_progress(&ref_pic->tf,
802 FFMIN(row, pic_height - 1),
809 static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
810 int n, int square, int height,
812 uint8_t *dest_y, uint8_t *dest_cb,
814 int src_x_offset, int src_y_offset,
815 qpel_mc_func *qpix_op,
816 h264_chroma_mc_func chroma_op,
817 int pixel_shift, int chroma_idc)
819 const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
820 int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
821 const int luma_xy = (mx & 3) + ((my & 3) << 2);
822 int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
823 uint8_t *src_y = pic->f.data[0] + offset;
824 uint8_t *src_cb, *src_cr;
826 int extra_height = 0;
828 const int full_mx = mx >> 2;
829 const int full_my = my >> 2;
830 const int pic_width = 16 * h->mb_width;
831 const int pic_height = 16 * h->mb_height >> MB_FIELD(h);
839 if (full_mx < 0 - extra_width ||
840 full_my < 0 - extra_height ||
841 full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
842 full_my + 16 /*FIXME*/ > pic_height + extra_height) {
843 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
844 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
846 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
847 full_my - 2, pic_width, pic_height);
848 src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
852 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
854 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
856 if (CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY)
859 if (chroma_idc == 3 /* yuv444 */) {
860 src_cb = pic->f.data[1] + offset;
862 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
863 src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
865 16 + 5, 16 + 5 /*FIXME*/,
866 full_mx - 2, full_my - 2,
867 pic_width, pic_height);
868 src_cb = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
870 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
872 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
874 src_cr = pic->f.data[2] + offset;
876 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
877 src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
879 16 + 5, 16 + 5 /*FIXME*/,
880 full_mx - 2, full_my - 2,
881 pic_width, pic_height);
882 src_cr = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
884 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
886 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
890 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
891 if (chroma_idc == 1 /* yuv420 */ && MB_FIELD(h)) {
892 // chroma offset when predicting from a field of opposite parity
893 my += 2 * ((h->mb_y & 1) - (pic->reference - 1));
894 emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
897 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
898 (my >> ysh) * h->mb_uvlinesize;
899 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
900 (my >> ysh) * h->mb_uvlinesize;
903 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->mb_uvlinesize,
904 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
905 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
906 src_cb = h->edge_emu_buffer;
908 chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
909 height >> (chroma_idc == 1 /* yuv420 */),
910 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
913 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->mb_uvlinesize,
914 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
915 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
916 src_cr = h->edge_emu_buffer;
918 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
919 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
922 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
923 int height, int delta,
924 uint8_t *dest_y, uint8_t *dest_cb,
926 int x_offset, int y_offset,
927 qpel_mc_func *qpix_put,
928 h264_chroma_mc_func chroma_put,
929 qpel_mc_func *qpix_avg,
930 h264_chroma_mc_func chroma_avg,
931 int list0, int list1,
932 int pixel_shift, int chroma_idc)
934 qpel_mc_func *qpix_op = qpix_put;
935 h264_chroma_mc_func chroma_op = chroma_put;
937 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
938 if (chroma_idc == 3 /* yuv444 */) {
939 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
940 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
941 } else if (chroma_idc == 2 /* yuv422 */) {
942 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
943 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
944 } else { /* yuv420 */
945 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
946 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
948 x_offset += 8 * h->mb_x;
949 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
952 Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
953 mc_dir_part(h, ref, n, square, height, delta, 0,
954 dest_y, dest_cb, dest_cr, x_offset, y_offset,
955 qpix_op, chroma_op, pixel_shift, chroma_idc);
958 chroma_op = chroma_avg;
962 Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
963 mc_dir_part(h, ref, n, square, height, delta, 1,
964 dest_y, dest_cb, dest_cr, x_offset, y_offset,
965 qpix_op, chroma_op, pixel_shift, chroma_idc);
969 static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
970 int height, int delta,
971 uint8_t *dest_y, uint8_t *dest_cb,
973 int x_offset, int y_offset,
974 qpel_mc_func *qpix_put,
975 h264_chroma_mc_func chroma_put,
976 h264_weight_func luma_weight_op,
977 h264_weight_func chroma_weight_op,
978 h264_biweight_func luma_weight_avg,
979 h264_biweight_func chroma_weight_avg,
980 int list0, int list1,
981 int pixel_shift, int chroma_idc)
985 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
986 if (chroma_idc == 3 /* yuv444 */) {
987 chroma_height = height;
988 chroma_weight_avg = luma_weight_avg;
989 chroma_weight_op = luma_weight_op;
990 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
991 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
992 } else if (chroma_idc == 2 /* yuv422 */) {
993 chroma_height = height;
994 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
995 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
996 } else { /* yuv420 */
997 chroma_height = height >> 1;
998 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
999 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1001 x_offset += 8 * h->mb_x;
1002 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
1004 if (list0 && list1) {
1005 /* don't optimize for luma-only case, since B-frames usually
1006 * use implicit weights => chroma too. */
1007 uint8_t *tmp_cb = h->bipred_scratchpad;
1008 uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
1009 uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
1010 int refn0 = h->ref_cache[0][scan8[n]];
1011 int refn1 = h->ref_cache[1][scan8[n]];
1013 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
1014 dest_y, dest_cb, dest_cr,
1015 x_offset, y_offset, qpix_put, chroma_put,
1016 pixel_shift, chroma_idc);
1017 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
1018 tmp_y, tmp_cb, tmp_cr,
1019 x_offset, y_offset, qpix_put, chroma_put,
1020 pixel_shift, chroma_idc);
1022 if (h->use_weight == 2) {
1023 int weight0 = h->implicit_weight[refn0][refn1][h->mb_y & 1];
1024 int weight1 = 64 - weight0;
1025 luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
1026 height, 5, weight0, weight1, 0);
1027 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
1028 chroma_height, 5, weight0, weight1, 0);
1029 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
1030 chroma_height, 5, weight0, weight1, 0);
1032 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
1033 h->luma_log2_weight_denom,
1034 h->luma_weight[refn0][0][0],
1035 h->luma_weight[refn1][1][0],
1036 h->luma_weight[refn0][0][1] +
1037 h->luma_weight[refn1][1][1]);
1038 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
1039 h->chroma_log2_weight_denom,
1040 h->chroma_weight[refn0][0][0][0],
1041 h->chroma_weight[refn1][1][0][0],
1042 h->chroma_weight[refn0][0][0][1] +
1043 h->chroma_weight[refn1][1][0][1]);
1044 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
1045 h->chroma_log2_weight_denom,
1046 h->chroma_weight[refn0][0][1][0],
1047 h->chroma_weight[refn1][1][1][0],
1048 h->chroma_weight[refn0][0][1][1] +
1049 h->chroma_weight[refn1][1][1][1]);
1052 int list = list1 ? 1 : 0;
1053 int refn = h->ref_cache[list][scan8[n]];
1054 Picture *ref = &h->ref_list[list][refn];
1055 mc_dir_part(h, ref, n, square, height, delta, list,
1056 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1057 qpix_put, chroma_put, pixel_shift, chroma_idc);
1059 luma_weight_op(dest_y, h->mb_linesize, height,
1060 h->luma_log2_weight_denom,
1061 h->luma_weight[refn][list][0],
1062 h->luma_weight[refn][list][1]);
1063 if (h->use_weight_chroma) {
1064 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
1065 h->chroma_log2_weight_denom,
1066 h->chroma_weight[refn][list][0][0],
1067 h->chroma_weight[refn][list][0][1]);
1068 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
1069 h->chroma_log2_weight_denom,
1070 h->chroma_weight[refn][list][1][0],
1071 h->chroma_weight[refn][list][1][1]);
1076 static av_always_inline void prefetch_motion(H264Context *h, int list,
1077 int pixel_shift, int chroma_idc)
1079 /* fetch pixels for estimated mv 4 macroblocks ahead
1080 * optimized for 64byte cache lines */
1081 const int refn = h->ref_cache[list][scan8[0]];
1083 const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * h->mb_x + 8;
1084 const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * h->mb_y;
1085 uint8_t **src = h->ref_list[list][refn].f.data;
1086 int off = (mx << pixel_shift) +
1087 (my + (h->mb_x & 3) * 4) * h->mb_linesize +
1088 (64 << pixel_shift);
1089 h->vdsp.prefetch(src[0] + off, h->linesize, 4);
1090 if (chroma_idc == 3 /* yuv444 */) {
1091 h->vdsp.prefetch(src[1] + off, h->linesize, 4);
1092 h->vdsp.prefetch(src[2] + off, h->linesize, 4);
1094 off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (h->mb_x&7))*h->uvlinesize;
1095 h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
1100 static void free_tables(H264Context *h, int free_rbsp)
1105 av_freep(&h->intra4x4_pred_mode);
1106 av_freep(&h->chroma_pred_mode_table);
1107 av_freep(&h->cbp_table);
1108 av_freep(&h->mvd_table[0]);
1109 av_freep(&h->mvd_table[1]);
1110 av_freep(&h->direct_table);
1111 av_freep(&h->non_zero_count);
1112 av_freep(&h->slice_table_base);
1113 h->slice_table = NULL;
1114 av_freep(&h->list_counts);
1116 av_freep(&h->mb2b_xy);
1117 av_freep(&h->mb2br_xy);
1119 for (i = 0; i < 3; i++)
1120 av_freep(&h->visualization_buffer[i]);
1122 av_buffer_pool_uninit(&h->qscale_table_pool);
1123 av_buffer_pool_uninit(&h->mb_type_pool);
1124 av_buffer_pool_uninit(&h->motion_val_pool);
1125 av_buffer_pool_uninit(&h->ref_index_pool);
1127 if (free_rbsp && h->DPB) {
1128 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1129 unref_picture(h, &h->DPB[i]);
1131 } else if (h->DPB) {
1132 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1133 h->DPB[i].needs_realloc = 1;
1136 h->cur_pic_ptr = NULL;
1138 for (i = 0; i < MAX_THREADS; i++) {
1139 hx = h->thread_context[i];
1142 av_freep(&hx->top_borders[1]);
1143 av_freep(&hx->top_borders[0]);
1144 av_freep(&hx->bipred_scratchpad);
1145 av_freep(&hx->edge_emu_buffer);
1146 av_freep(&hx->dc_val_base);
1147 av_freep(&hx->me.scratchpad);
1148 av_freep(&hx->er.mb_index2xy);
1149 av_freep(&hx->er.error_status_table);
1150 av_freep(&hx->er.er_temp_buffer);
1151 av_freep(&hx->er.mbintra_table);
1152 av_freep(&hx->er.mbskip_table);
1155 av_freep(&hx->rbsp_buffer[1]);
1156 av_freep(&hx->rbsp_buffer[0]);
1157 hx->rbsp_buffer_size[0] = 0;
1158 hx->rbsp_buffer_size[1] = 0;
1161 av_freep(&h->thread_context[i]);
1165 static void init_dequant8_coeff_table(H264Context *h)
1168 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1170 for (i = 0; i < 6; i++) {
1171 h->dequant8_coeff[i] = h->dequant8_buffer[i];
1172 for (j = 0; j < i; j++)
1173 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
1174 64 * sizeof(uint8_t))) {
1175 h->dequant8_coeff[i] = h->dequant8_buffer[j];
1181 for (q = 0; q < max_qp + 1; q++) {
1182 int shift = div6[q];
1184 for (x = 0; x < 64; x++)
1185 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
1186 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
1187 h->pps.scaling_matrix8[i][x]) << shift;
1192 static void init_dequant4_coeff_table(H264Context *h)
1195 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1196 for (i = 0; i < 6; i++) {
1197 h->dequant4_coeff[i] = h->dequant4_buffer[i];
1198 for (j = 0; j < i; j++)
1199 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
1200 16 * sizeof(uint8_t))) {
1201 h->dequant4_coeff[i] = h->dequant4_buffer[j];
1207 for (q = 0; q < max_qp + 1; q++) {
1208 int shift = div6[q] + 2;
1210 for (x = 0; x < 16; x++)
1211 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
1212 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
1213 h->pps.scaling_matrix4[i][x]) << shift;
1218 static void init_dequant_tables(H264Context *h)
1221 init_dequant4_coeff_table(h);
1222 if (h->pps.transform_8x8_mode)
1223 init_dequant8_coeff_table(h);
1224 if (h->sps.transform_bypass) {
1225 for (i = 0; i < 6; i++)
1226 for (x = 0; x < 16; x++)
1227 h->dequant4_coeff[i][0][x] = 1 << 6;
1228 if (h->pps.transform_8x8_mode)
1229 for (i = 0; i < 6; i++)
1230 for (x = 0; x < 64; x++)
1231 h->dequant8_coeff[i][0][x] = 1 << 6;
1235 int ff_h264_alloc_tables(H264Context *h)
1237 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
1238 const int row_mb_num = 2*h->mb_stride*FFMAX(h->avctx->thread_count, 1);
1241 FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
1242 row_mb_num * 8 * sizeof(uint8_t), fail)
1243 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
1244 big_mb_num * 48 * sizeof(uint8_t), fail)
1245 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
1246 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
1247 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
1248 big_mb_num * sizeof(uint16_t), fail)
1249 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
1250 big_mb_num * sizeof(uint8_t), fail)
1251 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0],
1252 16 * row_mb_num * sizeof(uint8_t), fail);
1253 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1],
1254 16 * row_mb_num * sizeof(uint8_t), fail);
1255 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
1256 4 * big_mb_num * sizeof(uint8_t), fail);
1257 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
1258 big_mb_num * sizeof(uint8_t), fail)
1260 memset(h->slice_table_base, -1,
1261 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
1262 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
1264 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
1265 big_mb_num * sizeof(uint32_t), fail);
1266 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
1267 big_mb_num * sizeof(uint32_t), fail);
1268 for (y = 0; y < h->mb_height; y++)
1269 for (x = 0; x < h->mb_width; x++) {
1270 const int mb_xy = x + y * h->mb_stride;
1271 const int b_xy = 4 * x + 4 * y * h->b_stride;
1273 h->mb2b_xy[mb_xy] = b_xy;
1274 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
1277 if (!h->dequant4_coeff[0])
1278 init_dequant_tables(h);
1281 h->DPB = av_mallocz_array(MAX_PICTURE_COUNT, sizeof(*h->DPB));
1283 return AVERROR(ENOMEM);
1284 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1285 avcodec_get_frame_defaults(&h->DPB[i].f);
1286 avcodec_get_frame_defaults(&h->cur_pic.f);
1297 * Mimic alloc_tables(), but for every context thread.
1299 static void clone_tables(H264Context *dst, H264Context *src, int i)
1301 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
1302 dst->non_zero_count = src->non_zero_count;
1303 dst->slice_table = src->slice_table;
1304 dst->cbp_table = src->cbp_table;
1305 dst->mb2b_xy = src->mb2b_xy;
1306 dst->mb2br_xy = src->mb2br_xy;
1307 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
1308 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
1309 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
1310 dst->direct_table = src->direct_table;
1311 dst->list_counts = src->list_counts;
1312 dst->DPB = src->DPB;
1313 dst->cur_pic_ptr = src->cur_pic_ptr;
1314 dst->cur_pic = src->cur_pic;
1315 dst->bipred_scratchpad = NULL;
1316 dst->edge_emu_buffer = NULL;
1317 dst->me.scratchpad = NULL;
1318 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
1319 src->sps.chroma_format_idc);
1324 * Allocate buffers which are not shared amongst multiple threads.
1326 static int context_init(H264Context *h)
1328 ERContext *er = &h->er;
1329 int mb_array_size = h->mb_height * h->mb_stride;
1330 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
1331 int c_size = h->mb_stride * (h->mb_height + 1);
1332 int yc_size = y_size + 2 * c_size;
1335 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[0],
1336 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1337 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[1],
1338 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1340 h->ref_cache[0][scan8[5] + 1] =
1341 h->ref_cache[0][scan8[7] + 1] =
1342 h->ref_cache[0][scan8[13] + 1] =
1343 h->ref_cache[1][scan8[5] + 1] =
1344 h->ref_cache[1][scan8[7] + 1] =
1345 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
1347 if (CONFIG_ERROR_RESILIENCE) {
1349 er->avctx = h->avctx;
1351 er->decode_mb = h264_er_decode_mb;
1353 er->quarter_sample = 1;
1355 er->mb_num = h->mb_num;
1356 er->mb_width = h->mb_width;
1357 er->mb_height = h->mb_height;
1358 er->mb_stride = h->mb_stride;
1359 er->b8_stride = h->mb_width * 2 + 1;
1361 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy, (h->mb_num + 1) * sizeof(int),
1362 fail); // error ressilience code looks cleaner with this
1363 for (y = 0; y < h->mb_height; y++)
1364 for (x = 0; x < h->mb_width; x++)
1365 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
1367 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
1368 h->mb_stride + h->mb_width;
1370 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
1371 mb_array_size * sizeof(uint8_t), fail);
1373 FF_ALLOC_OR_GOTO(h->avctx, er->mbintra_table, mb_array_size, fail);
1374 memset(er->mbintra_table, 1, mb_array_size);
1376 FF_ALLOCZ_OR_GOTO(h->avctx, er->mbskip_table, mb_array_size + 2, fail);
1378 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer, h->mb_height * h->mb_stride,
1381 FF_ALLOCZ_OR_GOTO(h->avctx, h->dc_val_base, yc_size * sizeof(int16_t), fail);
1382 er->dc_val[0] = h->dc_val_base + h->mb_width * 2 + 2;
1383 er->dc_val[1] = h->dc_val_base + y_size + h->mb_stride + 1;
1384 er->dc_val[2] = er->dc_val[1] + c_size;
1385 for (i = 0; i < yc_size; i++)
1386 h->dc_val_base[i] = 1024;
1392 return -1; // free_tables will clean up for us
1395 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1396 int parse_extradata);
1398 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1400 AVCodecContext *avctx = h->avctx;
1402 if (!buf || size <= 0)
1406 int i, cnt, nalsize;
1407 const unsigned char *p = buf;
1412 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1415 /* sps and pps in the avcC always have length coded with 2 bytes,
1416 * so put a fake nal_length_size = 2 while parsing them */
1417 h->nal_length_size = 2;
1418 // Decode sps from avcC
1419 cnt = *(p + 5) & 0x1f; // Number of sps
1421 for (i = 0; i < cnt; i++) {
1422 nalsize = AV_RB16(p) + 2;
1423 if(nalsize > size - (p-buf))
1425 if (decode_nal_units(h, p, nalsize, 1) < 0) {
1426 av_log(avctx, AV_LOG_ERROR,
1427 "Decoding sps %d from avcC failed\n", i);
1432 // Decode pps from avcC
1433 cnt = *(p++); // Number of pps
1434 for (i = 0; i < cnt; i++) {
1435 nalsize = AV_RB16(p) + 2;
1436 if(nalsize > size - (p-buf))
1438 if (decode_nal_units(h, p, nalsize, 1) < 0) {
1439 av_log(avctx, AV_LOG_ERROR,
1440 "Decoding pps %d from avcC failed\n", i);
1445 // Now store right nal length size, that will be used to parse all other nals
1446 h->nal_length_size = (buf[4] & 0x03) + 1;
1449 if (decode_nal_units(h, buf, size, 1) < 0)
1455 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1457 H264Context *h = avctx->priv_data;
1462 h->bit_depth_luma = 8;
1463 h->chroma_format_idc = 1;
1465 h->avctx->bits_per_raw_sample = 8;
1466 h->cur_chroma_format_idc = 1;
1468 ff_h264dsp_init(&h->h264dsp, 8, 1);
1469 av_assert0(h->sps.bit_depth_chroma == 0);
1470 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1471 ff_h264qpel_init(&h->h264qpel, 8);
1472 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, 8, 1);
1474 h->dequant_coeff_pps = -1;
1476 /* needed so that IDCT permutation is known early */
1477 if (CONFIG_ERROR_RESILIENCE)
1478 ff_dsputil_init(&h->dsp, h->avctx);
1479 ff_videodsp_init(&h->vdsp, 8);
1481 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1482 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1484 h->picture_structure = PICT_FRAME;
1485 h->slice_context_count = 1;
1486 h->workaround_bugs = avctx->workaround_bugs;
1487 h->flags = avctx->flags;
1490 // s->decode_mb = ff_h263_decode_mb;
1491 if (!avctx->has_b_frames)
1494 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1496 ff_h264_decode_init_vlc();
1499 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1501 h->thread_context[0] = h;
1502 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1503 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1504 h->last_pocs[i] = INT_MIN;
1505 h->prev_poc_msb = 1 << 16;
1506 h->prev_frame_num = -1;
1508 ff_h264_reset_sei(h);
1509 if (avctx->codec_id == AV_CODEC_ID_H264) {
1510 if (avctx->ticks_per_frame == 1) {
1511 if(h->avctx->time_base.den < INT_MAX/2) {
1512 h->avctx->time_base.den *= 2;
1514 h->avctx->time_base.num /= 2;
1516 avctx->ticks_per_frame = 2;
1519 if (avctx->extradata_size > 0 && avctx->extradata &&
1520 ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) {
1521 ff_h264_free_context(h);
1525 if (h->sps.bitstream_restriction_flag &&
1526 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
1527 h->avctx->has_b_frames = h->sps.num_reorder_frames;
1531 ff_init_cabac_states();
1532 avctx->internal->allocate_progress = 1;
1537 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1538 #undef REBASE_PICTURE
1539 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
1540 ((pic && pic >= old_ctx->DPB && \
1541 pic < old_ctx->DPB + MAX_PICTURE_COUNT) ? \
1542 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
1544 static void copy_picture_range(Picture **to, Picture **from, int count,
1545 H264Context *new_base,
1546 H264Context *old_base)
1550 for (i = 0; i < count; i++) {
1551 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1552 IN_RANGE(from[i], old_base->DPB,
1553 sizeof(Picture) * MAX_PICTURE_COUNT) ||
1555 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1559 static void copy_parameter_set(void **to, void **from, int count, int size)
1563 for (i = 0; i < count; i++) {
1564 if (to[i] && !from[i])
1566 else if (from[i] && !to[i])
1567 to[i] = av_malloc(size);
1570 memcpy(to[i], from[i], size);
1574 static int decode_init_thread_copy(AVCodecContext *avctx)
1576 H264Context *h = avctx->priv_data;
1578 if (!avctx->internal->is_copy)
1580 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1581 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1583 h->context_initialized = 0;
1588 #define copy_fields(to, from, start_field, end_field) \
1589 memcpy(&to->start_field, &from->start_field, \
1590 (char *)&to->end_field - (char *)&to->start_field)
1592 static int h264_slice_header_init(H264Context *, int);
1594 static int h264_set_parameter_from_sps(H264Context *h);
1596 static int decode_update_thread_context(AVCodecContext *dst,
1597 const AVCodecContext *src)
1599 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1600 int inited = h->context_initialized, err = 0;
1601 int context_reinitialized = 0;
1608 (h->width != h1->width ||
1609 h->height != h1->height ||
1610 h->mb_width != h1->mb_width ||
1611 h->mb_height != h1->mb_height ||
1612 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
1613 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
1614 h->sps.colorspace != h1->sps.colorspace)) {
1616 /* set bits_per_raw_sample to the previous value. the check for changed
1617 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
1618 * the current value */
1619 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1621 av_freep(&h->bipred_scratchpad);
1623 h->width = h1->width;
1624 h->height = h1->height;
1625 h->mb_height = h1->mb_height;
1626 h->mb_width = h1->mb_width;
1627 h->mb_num = h1->mb_num;
1628 h->mb_stride = h1->mb_stride;
1629 h->b_stride = h1->b_stride;
1631 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1632 MAX_SPS_COUNT, sizeof(SPS));
1634 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1635 MAX_PPS_COUNT, sizeof(PPS));
1638 if ((err = h264_slice_header_init(h, 1)) < 0) {
1639 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
1642 context_reinitialized = 1;
1645 h264_set_parameter_from_sps(h);
1646 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
1647 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
1650 /* update linesize on resize for h264. The h264 decoder doesn't
1651 * necessarily call ff_MPV_frame_start in the new thread */
1652 h->linesize = h1->linesize;
1653 h->uvlinesize = h1->uvlinesize;
1655 /* copy block_offset since frame_start may not be called */
1656 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
1659 for (i = 0; i < MAX_SPS_COUNT; i++)
1660 av_freep(h->sps_buffers + i);
1662 for (i = 0; i < MAX_PPS_COUNT; i++)
1663 av_freep(h->pps_buffers + i);
1665 memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
1666 memcpy(&h->cabac, &h1->cabac,
1667 sizeof(H264Context) - offsetof(H264Context, cabac));
1668 av_assert0((void*)&h->cabac == &h->mb_padding + 1);
1670 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1671 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1673 memset(&h->er, 0, sizeof(h->er));
1674 memset(&h->me, 0, sizeof(h->me));
1675 memset(&h->mb, 0, sizeof(h->mb));
1676 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
1677 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
1681 h->qscale_table_pool = NULL;
1682 h->mb_type_pool = NULL;
1683 h->ref_index_pool = NULL;
1684 h->motion_val_pool = NULL;
1686 if (h1->context_initialized) {
1687 h->context_initialized = 0;
1689 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
1690 avcodec_get_frame_defaults(&h->cur_pic.f);
1691 h->cur_pic.tf.f = &h->cur_pic.f;
1693 if (ff_h264_alloc_tables(h) < 0) {
1694 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1695 return AVERROR(ENOMEM);
1700 for (i = 0; i < 2; i++) {
1701 h->rbsp_buffer[i] = NULL;
1702 h->rbsp_buffer_size[i] = 0;
1704 h->bipred_scratchpad = NULL;
1705 h->edge_emu_buffer = NULL;
1707 h->thread_context[0] = h;
1708 h->context_initialized = h1->context_initialized;
1711 h->avctx->coded_height = h1->avctx->coded_height;
1712 h->avctx->coded_width = h1->avctx->coded_width;
1713 h->avctx->width = h1->avctx->width;
1714 h->avctx->height = h1->avctx->height;
1715 h->coded_picture_number = h1->coded_picture_number;
1716 h->first_field = h1->first_field;
1717 h->picture_structure = h1->picture_structure;
1718 h->qscale = h1->qscale;
1719 h->droppable = h1->droppable;
1720 h->data_partitioning = h1->data_partitioning;
1721 h->low_delay = h1->low_delay;
1723 for (i = 0; h->DPB && i < MAX_PICTURE_COUNT; i++) {
1724 unref_picture(h, &h->DPB[i]);
1725 if (h1->DPB[i].f.data[0] &&
1726 (ret = ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
1730 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
1731 unref_picture(h, &h->cur_pic);
1732 if (h1->cur_pic.f.buf[0] && (ret = ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
1735 h->workaround_bugs = h1->workaround_bugs;
1736 h->low_delay = h1->low_delay;
1737 h->droppable = h1->droppable;
1739 // extradata/NAL handling
1740 h->is_avc = h1->is_avc;
1743 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1744 MAX_SPS_COUNT, sizeof(SPS));
1746 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1747 MAX_PPS_COUNT, sizeof(PPS));
1750 // Dequantization matrices
1751 // FIXME these are big - can they be only copied when PPS changes?
1752 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1754 for (i = 0; i < 6; i++)
1755 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1756 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1758 for (i = 0; i < 6; i++)
1759 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1760 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1762 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1765 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1768 copy_fields(h, h1, short_ref, cabac_init_idc);
1770 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
1771 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
1772 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1773 MAX_DELAYED_PIC_COUNT + 2, h, h1);
1777 if (context_reinitialized)
1778 h264_set_parameter_from_sps(h);
1780 if (!h->cur_pic_ptr)
1783 if (!h->droppable) {
1784 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1785 h->prev_poc_msb = h->poc_msb;
1786 h->prev_poc_lsb = h->poc_lsb;
1788 h->prev_frame_num_offset = h->frame_num_offset;
1789 h->prev_frame_num = h->frame_num;
1790 h->outputed_poc = h->next_outputed_poc;
1795 static int h264_frame_start(H264Context *h)
1799 const int pixel_shift = h->pixel_shift;
1801 1<<(h->sps.bit_depth_luma-1),
1802 1<<(h->sps.bit_depth_chroma-1),
1803 1<<(h->sps.bit_depth_chroma-1),
1807 if (!ff_thread_can_start_frame(h->avctx)) {
1808 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1812 release_unused_pictures(h, 1);
1813 h->cur_pic_ptr = NULL;
1815 i = find_unused_picture(h);
1817 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1822 pic->reference = h->droppable ? 0 : h->picture_structure;
1823 pic->f.coded_picture_number = h->coded_picture_number++;
1824 pic->field_picture = h->picture_structure != PICT_FRAME;
1827 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
1829 * See decode_nal_units().
1831 pic->f.key_frame = 0;
1833 pic->mmco_reset = 0;
1835 if ((ret = alloc_picture(h, pic)) < 0)
1837 if(!h->sync && !h->avctx->hwaccel &&
1838 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
1839 avpriv_color_frame(&pic->f, c);
1841 h->cur_pic_ptr = pic;
1842 unref_picture(h, &h->cur_pic);
1843 if ((ret = ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
1846 if (CONFIG_ERROR_RESILIENCE) {
1847 ff_er_frame_start(&h->er);
1849 h->er.next_pic = NULL;
1852 assert(h->linesize && h->uvlinesize);
1854 for (i = 0; i < 16; i++) {
1855 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1856 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1858 for (i = 0; i < 16; i++) {
1859 h->block_offset[16 + i] =
1860 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1861 h->block_offset[48 + 16 + i] =
1862 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1865 // s->decode = (h->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1866 // h->cur_pic.reference /* || h->contains_intra */ || 1;
1868 /* We mark the current picture as non-reference after allocating it, so
1869 * that if we break out due to an error it can be released automatically
1870 * in the next ff_MPV_frame_start().
1872 h->cur_pic_ptr->reference = 0;
1874 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
1876 h->next_output_pic = NULL;
1878 assert(h->cur_pic_ptr->long_ref == 0);
1884 * Run setup operations that must be run after slice header decoding.
1885 * This includes finding the next displayed frame.
1887 * @param h h264 master context
1888 * @param setup_finished enough NALs have been read that we can call
1889 * ff_thread_finish_setup()
1891 static void decode_postinit(H264Context *h, int setup_finished)
1893 Picture *out = h->cur_pic_ptr;
1894 Picture *cur = h->cur_pic_ptr;
1895 int i, pics, out_of_order, out_idx;
1897 h->cur_pic_ptr->f.pict_type = h->pict_type;
1899 if (h->next_output_pic)
1902 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1903 /* FIXME: if we have two PAFF fields in one packet, we can't start
1904 * the next thread here. If we have one field per packet, we can.
1905 * The check in decode_nal_units() is not good enough to find this
1906 * yet, so we assume the worst for now. */
1907 // if (setup_finished)
1908 // ff_thread_finish_setup(h->avctx);
1912 cur->f.interlaced_frame = 0;
1913 cur->f.repeat_pict = 0;
1915 /* Signal interlacing information externally. */
1916 /* Prioritize picture timing SEI information over used
1917 * decoding process if it exists. */
1919 if (h->sps.pic_struct_present_flag) {
1920 switch (h->sei_pic_struct) {
1921 case SEI_PIC_STRUCT_FRAME:
1923 case SEI_PIC_STRUCT_TOP_FIELD:
1924 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1925 cur->f.interlaced_frame = 1;
1927 case SEI_PIC_STRUCT_TOP_BOTTOM:
1928 case SEI_PIC_STRUCT_BOTTOM_TOP:
1929 if (FIELD_OR_MBAFF_PICTURE(h))
1930 cur->f.interlaced_frame = 1;
1932 // try to flag soft telecine progressive
1933 cur->f.interlaced_frame = h->prev_interlaced_frame;
1935 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1936 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1937 /* Signal the possibility of telecined film externally
1938 * (pic_struct 5,6). From these hints, let the applications
1939 * decide if they apply deinterlacing. */
1940 cur->f.repeat_pict = 1;
1942 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1943 cur->f.repeat_pict = 2;
1945 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1946 cur->f.repeat_pict = 4;
1950 if ((h->sei_ct_type & 3) &&
1951 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1952 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1954 /* Derive interlacing flag from used decoding process. */
1955 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
1957 h->prev_interlaced_frame = cur->f.interlaced_frame;
1959 if (cur->field_poc[0] != cur->field_poc[1]) {
1960 /* Derive top_field_first from field pocs. */
1961 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1963 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1964 /* Use picture timing SEI information. Even if it is a
1965 * information of a past frame, better than nothing. */
1966 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1967 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1968 cur->f.top_field_first = 1;
1970 cur->f.top_field_first = 0;
1972 /* Most likely progressive */
1973 cur->f.top_field_first = 0;
1977 cur->mmco_reset = h->mmco_reset;
1979 // FIXME do something with unavailable reference frames
1981 /* Sort B-frames into display order */
1983 if (h->sps.bitstream_restriction_flag &&
1984 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
1985 h->avctx->has_b_frames = h->sps.num_reorder_frames;
1989 if (h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
1990 !h->sps.bitstream_restriction_flag) {
1991 h->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1995 for (i = 0; 1; i++) {
1996 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1998 h->last_pocs[i-1] = cur->poc;
2001 h->last_pocs[i-1]= h->last_pocs[i];
2004 out_of_order = MAX_DELAYED_PIC_COUNT - i;
2005 if( cur->f.pict_type == AV_PICTURE_TYPE_B
2006 || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
2007 out_of_order = FFMAX(out_of_order, 1);
2008 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
2009 av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
2010 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
2011 h->last_pocs[i] = INT_MIN;
2012 h->last_pocs[0] = cur->poc;
2013 cur->mmco_reset = 1;
2014 } else if(h->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
2015 av_log(h->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
2016 h->avctx->has_b_frames = out_of_order;
2021 while (h->delayed_pic[pics])
2024 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
2026 h->delayed_pic[pics++] = cur;
2027 if (cur->reference == 0)
2028 cur->reference = DELAYED_PIC_REF;
2030 out = h->delayed_pic[0];
2032 for (i = 1; h->delayed_pic[i] &&
2033 !h->delayed_pic[i]->f.key_frame &&
2034 !h->delayed_pic[i]->mmco_reset;
2036 if (h->delayed_pic[i]->poc < out->poc) {
2037 out = h->delayed_pic[i];
2040 if (h->avctx->has_b_frames == 0 &&
2041 (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
2042 h->next_outputed_poc = INT_MIN;
2043 out_of_order = out->poc < h->next_outputed_poc;
2045 if (out_of_order || pics > h->avctx->has_b_frames) {
2046 out->reference &= ~DELAYED_PIC_REF;
2047 // for frame threading, the owner must be the second field's thread or
2048 // else the first thread can release the picture and reuse it unsafely
2049 for (i = out_idx; h->delayed_pic[i]; i++)
2050 h->delayed_pic[i] = h->delayed_pic[i + 1];
2052 if (!out_of_order && pics > h->avctx->has_b_frames) {
2053 h->next_output_pic = out;
2054 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
2055 h->next_outputed_poc = INT_MIN;
2057 h->next_outputed_poc = out->poc;
2059 av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
2062 if (h->next_output_pic && h->next_output_pic->sync) {
2066 if (setup_finished && !h->avctx->hwaccel)
2067 ff_thread_finish_setup(h->avctx);
2070 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
2071 uint8_t *src_cb, uint8_t *src_cr,
2072 int linesize, int uvlinesize,
2075 uint8_t *top_border;
2077 const int pixel_shift = h->pixel_shift;
2078 int chroma444 = CHROMA444(h);
2079 int chroma422 = CHROMA422(h);
2082 src_cb -= uvlinesize;
2083 src_cr -= uvlinesize;
2085 if (!simple && FRAME_MBAFF(h)) {
2088 top_border = h->top_borders[0][h->mb_x];
2089 AV_COPY128(top_border, src_y + 15 * linesize);
2091 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
2092 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2095 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2096 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
2097 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
2098 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
2100 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
2101 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
2103 } else if (chroma422) {
2105 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2106 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
2108 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
2109 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
2113 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
2114 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
2116 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
2117 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
2122 } else if (MB_MBAFF(h)) {
2128 top_border = h->top_borders[top_idx][h->mb_x];
2129 /* There are two lines saved, the line above the top macroblock
2130 * of a pair, and the line above the bottom macroblock. */
2131 AV_COPY128(top_border, src_y + 16 * linesize);
2133 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
2135 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2138 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
2139 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
2140 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
2141 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
2143 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
2144 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
2146 } else if (chroma422) {
2148 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
2149 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
2151 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
2152 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
2156 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
2157 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
2159 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
2160 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
2166 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
2167 uint8_t *src_cb, uint8_t *src_cr,
2168 int linesize, int uvlinesize,
2169 int xchg, int chroma444,
2170 int simple, int pixel_shift)
2172 int deblock_topleft;
2175 uint8_t *top_border_m1;
2176 uint8_t *top_border;
2178 if (!simple && FRAME_MBAFF(h)) {
2183 top_idx = MB_MBAFF(h) ? 0 : 1;
2187 if (h->deblocking_filter == 2) {
2188 deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num;
2189 deblock_top = h->top_type;
2191 deblock_topleft = (h->mb_x > 0);
2192 deblock_top = (h->mb_y > !!MB_FIELD(h));
2195 src_y -= linesize + 1 + pixel_shift;
2196 src_cb -= uvlinesize + 1 + pixel_shift;
2197 src_cr -= uvlinesize + 1 + pixel_shift;
2199 top_border_m1 = h->top_borders[top_idx][h->mb_x - 1];
2200 top_border = h->top_borders[top_idx][h->mb_x];
2202 #define XCHG(a, b, xchg) \
2203 if (pixel_shift) { \
2205 AV_SWAP64(b + 0, a + 0); \
2206 AV_SWAP64(b + 8, a + 8); \
2216 if (deblock_topleft) {
2217 XCHG(top_border_m1 + (8 << pixel_shift),
2218 src_y - (7 << pixel_shift), 1);
2220 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
2221 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
2222 if (h->mb_x + 1 < h->mb_width) {
2223 XCHG(h->top_borders[top_idx][h->mb_x + 1],
2224 src_y + (17 << pixel_shift), 1);
2227 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2229 if (deblock_topleft) {
2230 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2231 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2233 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
2234 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
2235 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
2236 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
2237 if (h->mb_x + 1 < h->mb_width) {
2238 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
2239 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
2243 if (deblock_topleft) {
2244 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2245 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2247 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
2248 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
2254 static av_always_inline int dctcoef_get(int16_t *mb, int high_bit_depth,
2257 if (high_bit_depth) {
2258 return AV_RN32A(((int32_t *)mb) + index);
2260 return AV_RN16A(mb + index);
2263 static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth,
2264 int index, int value)
2266 if (high_bit_depth) {
2267 AV_WN32A(((int32_t *)mb) + index, value);
2269 AV_WN16A(mb + index, value);
2272 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
2273 int mb_type, int is_h264,
2275 int transform_bypass,
2279 uint8_t *dest_y, int p)
2281 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2282 void (*idct_dc_add)(uint8_t *dst, int16_t *block, int stride);
2284 int qscale = p == 0 ? h->qscale : h->chroma_qp[p - 1];
2285 block_offset += 16 * p;
2286 if (IS_INTRA4x4(mb_type)) {
2287 if (IS_8x8DCT(mb_type)) {
2288 if (transform_bypass) {
2290 idct_add = h->h264dsp.h264_add_pixels8_clear;
2292 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
2293 idct_add = h->h264dsp.h264_idct8_add;
2295 for (i = 0; i < 16; i += 4) {
2296 uint8_t *const ptr = dest_y + block_offset[i];
2297 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2298 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2299 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2301 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2302 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
2303 (h->topright_samples_available << i) & 0x4000, linesize);
2305 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2306 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2308 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2313 if (transform_bypass) {
2315 idct_add = h->h264dsp.h264_add_pixels4_clear;
2317 idct_dc_add = h->h264dsp.h264_idct_dc_add;
2318 idct_add = h->h264dsp.h264_idct_add;
2320 for (i = 0; i < 16; i++) {
2321 uint8_t *const ptr = dest_y + block_offset[i];
2322 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2324 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2325 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2330 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
2331 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
2332 av_assert2(h->mb_y || linesize <= block_offset[i]);
2333 if (!topright_avail) {
2335 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
2336 topright = (uint8_t *)&tr_high;
2338 tr = ptr[3 - linesize] * 0x01010101u;
2339 topright = (uint8_t *)&tr;
2342 topright = ptr + (4 << pixel_shift) - linesize;
2346 h->hpc.pred4x4[dir](ptr, topright, linesize);
2347 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2350 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2351 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2353 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2354 } else if (CONFIG_SVQ3_DECODER)
2355 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
2361 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
2363 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
2364 if (!transform_bypass)
2365 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
2367 h->dequant4_coeff[p][qscale][0]);
2369 static const uint8_t dc_mapping[16] = {
2370 0 * 16, 1 * 16, 4 * 16, 5 * 16,
2371 2 * 16, 3 * 16, 6 * 16, 7 * 16,
2372 8 * 16, 9 * 16, 12 * 16, 13 * 16,
2373 10 * 16, 11 * 16, 14 * 16, 15 * 16 };
2374 for (i = 0; i < 16; i++)
2375 dctcoef_set(h->mb + (p * 256 << pixel_shift),
2376 pixel_shift, dc_mapping[i],
2377 dctcoef_get(h->mb_luma_dc[p],
2381 } else if (CONFIG_SVQ3_DECODER)
2382 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
2383 h->mb_luma_dc[p], qscale);
2387 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
2388 int is_h264, int simple,
2389 int transform_bypass,
2393 uint8_t *dest_y, int p)
2395 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2397 block_offset += 16 * p;
2398 if (!IS_INTRA4x4(mb_type)) {
2400 if (IS_INTRA16x16(mb_type)) {
2401 if (transform_bypass) {
2402 if (h->sps.profile_idc == 244 &&
2403 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
2404 h->intra16x16_pred_mode == HOR_PRED8x8)) {
2405 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
2406 h->mb + (p * 256 << pixel_shift),
2409 for (i = 0; i < 16; i++)
2410 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
2411 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2412 h->h264dsp.h264_add_pixels4_clear(dest_y + block_offset[i],
2413 h->mb + (i * 16 + p * 256 << pixel_shift),
2417 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
2418 h->mb + (p * 256 << pixel_shift),
2420 h->non_zero_count_cache + p * 5 * 8);
2422 } else if (h->cbp & 15) {
2423 if (transform_bypass) {
2424 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2425 idct_add = IS_8x8DCT(mb_type) ? h->h264dsp.h264_add_pixels8_clear
2426 : h->h264dsp.h264_add_pixels4_clear;
2427 for (i = 0; i < 16; i += di)
2428 if (h->non_zero_count_cache[scan8[i + p * 16]])
2429 idct_add(dest_y + block_offset[i],
2430 h->mb + (i * 16 + p * 256 << pixel_shift),
2433 if (IS_8x8DCT(mb_type))
2434 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
2435 h->mb + (p * 256 << pixel_shift),
2437 h->non_zero_count_cache + p * 5 * 8);
2439 h->h264dsp.h264_idct_add16(dest_y, block_offset,
2440 h->mb + (p * 256 << pixel_shift),
2442 h->non_zero_count_cache + p * 5 * 8);
2445 } else if (CONFIG_SVQ3_DECODER) {
2446 for (i = 0; i < 16; i++)
2447 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
2448 // FIXME benchmark weird rule, & below
2449 uint8_t *const ptr = dest_y + block_offset[i];
2450 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
2451 h->qscale, IS_INTRA(mb_type) ? 1 : 0);
2459 #include "h264_mb_template.c"
2463 #include "h264_mb_template.c"
2467 #include "h264_mb_template.c"
2469 void ff_h264_hl_decode_mb(H264Context *h)
2471 const int mb_xy = h->mb_xy;
2472 const int mb_type = h->cur_pic.mb_type[mb_xy];
2473 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || h->qscale == 0;
2476 if (is_complex || h->pixel_shift)
2477 hl_decode_mb_444_complex(h);
2479 hl_decode_mb_444_simple_8(h);
2480 } else if (is_complex) {
2481 hl_decode_mb_complex(h);
2482 } else if (h->pixel_shift) {
2483 hl_decode_mb_simple_16(h);
2485 hl_decode_mb_simple_8(h);
2488 static int pred_weight_table(H264Context *h)
2491 int luma_def, chroma_def;
2494 h->use_weight_chroma = 0;
2495 h->luma_log2_weight_denom = get_ue_golomb(&h->gb);
2496 if (h->sps.chroma_format_idc)
2497 h->chroma_log2_weight_denom = get_ue_golomb(&h->gb);
2498 luma_def = 1 << h->luma_log2_weight_denom;
2499 chroma_def = 1 << h->chroma_log2_weight_denom;
2501 for (list = 0; list < 2; list++) {
2502 h->luma_weight_flag[list] = 0;
2503 h->chroma_weight_flag[list] = 0;
2504 for (i = 0; i < h->ref_count[list]; i++) {
2505 int luma_weight_flag, chroma_weight_flag;
2507 luma_weight_flag = get_bits1(&h->gb);
2508 if (luma_weight_flag) {
2509 h->luma_weight[i][list][0] = get_se_golomb(&h->gb);
2510 h->luma_weight[i][list][1] = get_se_golomb(&h->gb);
2511 if (h->luma_weight[i][list][0] != luma_def ||
2512 h->luma_weight[i][list][1] != 0) {
2514 h->luma_weight_flag[list] = 1;
2517 h->luma_weight[i][list][0] = luma_def;
2518 h->luma_weight[i][list][1] = 0;
2521 if (h->sps.chroma_format_idc) {
2522 chroma_weight_flag = get_bits1(&h->gb);
2523 if (chroma_weight_flag) {
2525 for (j = 0; j < 2; j++) {
2526 h->chroma_weight[i][list][j][0] = get_se_golomb(&h->gb);
2527 h->chroma_weight[i][list][j][1] = get_se_golomb(&h->gb);
2528 if (h->chroma_weight[i][list][j][0] != chroma_def ||
2529 h->chroma_weight[i][list][j][1] != 0) {
2530 h->use_weight_chroma = 1;
2531 h->chroma_weight_flag[list] = 1;
2536 for (j = 0; j < 2; j++) {
2537 h->chroma_weight[i][list][j][0] = chroma_def;
2538 h->chroma_weight[i][list][j][1] = 0;
2543 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2546 h->use_weight = h->use_weight || h->use_weight_chroma;
2551 * Initialize implicit_weight table.
2552 * @param field 0/1 initialize the weight for interlaced MBAFF
2553 * -1 initializes the rest
2555 static void implicit_weight_table(H264Context *h, int field)
2557 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2559 for (i = 0; i < 2; i++) {
2560 h->luma_weight_flag[i] = 0;
2561 h->chroma_weight_flag[i] = 0;
2565 if (h->picture_structure == PICT_FRAME) {
2566 cur_poc = h->cur_pic_ptr->poc;
2568 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
2570 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
2571 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2573 h->use_weight_chroma = 0;
2577 ref_count0 = h->ref_count[0];
2578 ref_count1 = h->ref_count[1];
2580 cur_poc = h->cur_pic_ptr->field_poc[field];
2582 ref_count0 = 16 + 2 * h->ref_count[0];
2583 ref_count1 = 16 + 2 * h->ref_count[1];
2587 h->use_weight_chroma = 2;
2588 h->luma_log2_weight_denom = 5;
2589 h->chroma_log2_weight_denom = 5;
2591 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2592 int poc0 = h->ref_list[0][ref0].poc;
2593 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2595 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2596 int poc1 = h->ref_list[1][ref1].poc;
2597 int td = av_clip(poc1 - poc0, -128, 127);
2599 int tb = av_clip(cur_poc - poc0, -128, 127);
2600 int tx = (16384 + (FFABS(td) >> 1)) / td;
2601 int dist_scale_factor = (tb * tx + 32) >> 8;
2602 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2603 w = 64 - dist_scale_factor;
2607 h->implicit_weight[ref0][ref1][0] =
2608 h->implicit_weight[ref0][ref1][1] = w;
2610 h->implicit_weight[ref0][ref1][field] = w;
2617 * instantaneous decoder refresh.
2619 static void idr(H264Context *h)
2622 ff_h264_remove_all_refs(h);
2623 h->prev_frame_num = 0;
2624 h->prev_frame_num_offset = 0;
2625 h->prev_poc_msb = 1<<16;
2626 h->prev_poc_lsb = 0;
2627 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2628 h->last_pocs[i] = INT_MIN;
2631 /* forget old pics after a seek */
2632 static void flush_change(H264Context *h)
2636 h->outputed_poc = h->next_outputed_poc = INT_MIN;
2637 h->prev_interlaced_frame = 1;
2640 h->prev_frame_num = -1;
2641 if (h->cur_pic_ptr) {
2642 h->cur_pic_ptr->reference = 0;
2643 for (j=i=0; h->delayed_pic[i]; i++)
2644 if (h->delayed_pic[i] != h->cur_pic_ptr)
2645 h->delayed_pic[j++] = h->delayed_pic[i];
2646 h->delayed_pic[j] = NULL;
2649 memset(h->ref_list[0], 0, sizeof(h->ref_list[0]));
2650 memset(h->ref_list[1], 0, sizeof(h->ref_list[1]));
2651 memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0]));
2652 memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1]));
2653 ff_h264_reset_sei(h);
2654 h->recovery_frame= -1;
2657 h->current_slice = 0;
2660 /* forget old pics after a seek */
2661 static void flush_dpb(AVCodecContext *avctx)
2663 H264Context *h = avctx->priv_data;
2666 for (i = 0; i <= MAX_DELAYED_PIC_COUNT; i++) {
2667 if (h->delayed_pic[i])
2668 h->delayed_pic[i]->reference = 0;
2669 h->delayed_pic[i] = NULL;
2675 for (i = 0; i < MAX_PICTURE_COUNT; i++)
2676 unref_picture(h, &h->DPB[i]);
2677 h->cur_pic_ptr = NULL;
2678 unref_picture(h, &h->cur_pic);
2680 h->mb_x = h->mb_y = 0;
2682 h->parse_context.state = -1;
2683 h->parse_context.frame_start_found = 0;
2684 h->parse_context.overread = 0;
2685 h->parse_context.overread_index = 0;
2686 h->parse_context.index = 0;
2687 h->parse_context.last_index = 0;
2690 static int init_poc(H264Context *h)
2692 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2694 Picture *cur = h->cur_pic_ptr;
2696 h->frame_num_offset = h->prev_frame_num_offset;
2697 if (h->frame_num < h->prev_frame_num)
2698 h->frame_num_offset += max_frame_num;
2700 if (h->sps.poc_type == 0) {
2701 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2703 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2704 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2705 else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2706 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2708 h->poc_msb = h->prev_poc_msb;
2710 field_poc[1] = h->poc_msb + h->poc_lsb;
2711 if (h->picture_structure == PICT_FRAME)
2712 field_poc[1] += h->delta_poc_bottom;
2713 } else if (h->sps.poc_type == 1) {
2714 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2717 if (h->sps.poc_cycle_length != 0)
2718 abs_frame_num = h->frame_num_offset + h->frame_num;
2722 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2725 expected_delta_per_poc_cycle = 0;
2726 for (i = 0; i < h->sps.poc_cycle_length; i++)
2727 // FIXME integrate during sps parse
2728 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2730 if (abs_frame_num > 0) {
2731 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2732 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2734 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2735 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2736 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2740 if (h->nal_ref_idc == 0)
2741 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2743 field_poc[0] = expectedpoc + h->delta_poc[0];
2744 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2746 if (h->picture_structure == PICT_FRAME)
2747 field_poc[1] += h->delta_poc[1];
2749 int poc = 2 * (h->frame_num_offset + h->frame_num);
2751 if (!h->nal_ref_idc)
2758 if (h->picture_structure != PICT_BOTTOM_FIELD)
2759 h->cur_pic_ptr->field_poc[0] = field_poc[0];
2760 if (h->picture_structure != PICT_TOP_FIELD)
2761 h->cur_pic_ptr->field_poc[1] = field_poc[1];
2762 cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]);
2768 * initialize scan tables
2770 static void init_scan_tables(H264Context *h)
2773 for (i = 0; i < 16; i++) {
2774 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2775 h->zigzag_scan[i] = T(zigzag_scan[i]);
2776 h->field_scan[i] = T(field_scan[i]);
2779 for (i = 0; i < 64; i++) {
2780 #define T(x) (x >> 3) | ((x & 7) << 3)
2781 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2782 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2783 h->field_scan8x8[i] = T(field_scan8x8[i]);
2784 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2787 if (h->sps.transform_bypass) { // FIXME same ugly
2788 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2789 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
2790 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2791 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
2792 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2793 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2795 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2796 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
2797 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2798 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
2799 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2800 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2804 static int field_end(H264Context *h, int in_setup)
2806 AVCodecContext *const avctx = h->avctx;
2810 if (CONFIG_H264_VDPAU_DECODER &&
2811 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2812 ff_vdpau_h264_set_reference_frames(h);
2814 if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2815 if (!h->droppable) {
2816 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2817 h->prev_poc_msb = h->poc_msb;
2818 h->prev_poc_lsb = h->poc_lsb;
2820 h->prev_frame_num_offset = h->frame_num_offset;
2821 h->prev_frame_num = h->frame_num;
2822 h->outputed_poc = h->next_outputed_poc;
2825 if (avctx->hwaccel) {
2826 if (avctx->hwaccel->end_frame(avctx) < 0)
2827 av_log(avctx, AV_LOG_ERROR,
2828 "hardware accelerator failed to decode picture\n");
2831 if (CONFIG_H264_VDPAU_DECODER &&
2832 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2833 ff_vdpau_h264_picture_complete(h);
2836 * FIXME: Error handling code does not seem to support interlaced
2837 * when slices span multiple rows
2838 * The ff_er_add_slice calls don't work right for bottom
2839 * fields; they cause massive erroneous error concealing
2840 * Error marking covers both fields (top and bottom).
2841 * This causes a mismatched s->error_count
2842 * and a bad error table. Further, the error count goes to
2843 * INT_MAX when called for bottom field, because mb_y is
2844 * past end by one (callers fault) and resync_mb_y != 0
2845 * causes problems for the first MB line, too.
2847 if (CONFIG_ERROR_RESILIENCE &&
2848 !FIELD_PICTURE(h) && h->current_slice && !h->sps.new) {
2849 h->er.cur_pic = h->cur_pic_ptr;
2850 ff_er_frame_end(&h->er);
2852 if (!in_setup && !h->droppable)
2853 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
2854 h->picture_structure == PICT_BOTTOM_FIELD);
2857 h->current_slice = 0;
2863 * Replicate H264 "master" context to thread contexts.
2865 static int clone_slice(H264Context *dst, H264Context *src)
2867 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2868 dst->cur_pic_ptr = src->cur_pic_ptr;
2869 dst->cur_pic = src->cur_pic;
2870 dst->linesize = src->linesize;
2871 dst->uvlinesize = src->uvlinesize;
2872 dst->first_field = src->first_field;
2874 dst->prev_poc_msb = src->prev_poc_msb;
2875 dst->prev_poc_lsb = src->prev_poc_lsb;
2876 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2877 dst->prev_frame_num = src->prev_frame_num;
2878 dst->short_ref_count = src->short_ref_count;
2880 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2881 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2882 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2884 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2885 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2891 * Compute profile from profile_idc and constraint_set?_flags.
2895 * @return profile as defined by FF_PROFILE_H264_*
2897 int ff_h264_get_profile(SPS *sps)
2899 int profile = sps->profile_idc;
2901 switch (sps->profile_idc) {
2902 case FF_PROFILE_H264_BASELINE:
2903 // constraint_set1_flag set to 1
2904 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2906 case FF_PROFILE_H264_HIGH_10:
2907 case FF_PROFILE_H264_HIGH_422:
2908 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2909 // constraint_set3_flag set to 1
2910 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
2917 static int h264_set_parameter_from_sps(H264Context *h)
2919 if (h->flags & CODEC_FLAG_LOW_DELAY ||
2920 (h->sps.bitstream_restriction_flag &&
2921 !h->sps.num_reorder_frames)) {
2922 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
2923 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
2924 "Reenabling low delay requires a codec flush.\n");
2929 if (h->avctx->has_b_frames < 2)
2930 h->avctx->has_b_frames = !h->low_delay;
2932 if (h->sps.bit_depth_luma != h->sps.bit_depth_chroma) {
2933 avpriv_request_sample(h->avctx,
2934 "Different chroma and luma bit depth");
2935 return AVERROR_PATCHWELCOME;
2938 if (h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2939 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
2940 if (h->avctx->codec &&
2941 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU &&
2942 (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
2943 av_log(h->avctx, AV_LOG_ERROR,
2944 "VDPAU decoding does not support video colorspace.\n");
2945 return AVERROR_INVALIDDATA;
2947 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 &&
2948 h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13) {
2949 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
2950 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
2951 h->pixel_shift = h->sps.bit_depth_luma > 8;
2953 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
2954 h->sps.chroma_format_idc);
2955 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
2956 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
2957 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
2958 h->sps.chroma_format_idc);
2960 if (CONFIG_ERROR_RESILIENCE)
2961 ff_dsputil_init(&h->dsp, h->avctx);
2962 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
2964 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
2965 h->sps.bit_depth_luma);
2966 return AVERROR_INVALIDDATA;
2972 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
2974 switch (h->sps.bit_depth_luma) {
2977 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
2978 return AV_PIX_FMT_GBRP9;
2980 return AV_PIX_FMT_YUV444P9;
2981 } else if (CHROMA422(h))
2982 return AV_PIX_FMT_YUV422P9;
2984 return AV_PIX_FMT_YUV420P9;
2988 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
2989 return AV_PIX_FMT_GBRP10;
2991 return AV_PIX_FMT_YUV444P10;
2992 } else if (CHROMA422(h))
2993 return AV_PIX_FMT_YUV422P10;
2995 return AV_PIX_FMT_YUV420P10;
2999 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3000 return AV_PIX_FMT_GBRP12;
3002 return AV_PIX_FMT_YUV444P12;
3003 } else if (CHROMA422(h))
3004 return AV_PIX_FMT_YUV422P12;
3006 return AV_PIX_FMT_YUV420P12;
3010 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3011 return AV_PIX_FMT_GBRP14;
3013 return AV_PIX_FMT_YUV444P14;
3014 } else if (CHROMA422(h))
3015 return AV_PIX_FMT_YUV422P14;
3017 return AV_PIX_FMT_YUV420P14;
3021 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3022 av_log(h->avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
3023 return AV_PIX_FMT_GBR24P;
3024 } else if (h->avctx->colorspace == AVCOL_SPC_YCGCO) {
3025 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
3027 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
3028 : AV_PIX_FMT_YUV444P;
3029 } else if (CHROMA422(h)) {
3030 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
3031 : AV_PIX_FMT_YUV422P;
3034 const enum AVPixelFormat * fmt = h->avctx->codec->pix_fmts ?
3035 h->avctx->codec->pix_fmts :
3036 h->avctx->color_range == AVCOL_RANGE_JPEG ?
3037 h264_hwaccel_pixfmt_list_jpeg_420 :
3038 h264_hwaccel_pixfmt_list_420;
3040 for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++)
3041 if (fmt[i] == h->avctx->pix_fmt && !force_callback)
3043 return ff_thread_get_format(h->avctx, fmt);
3047 av_log(h->avctx, AV_LOG_ERROR,
3048 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3049 return AVERROR_INVALIDDATA;
3053 /* export coded and cropped frame dimensions to AVCodecContext */
3054 static int init_dimensions(H264Context *h)
3056 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
3057 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
3058 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
3059 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
3061 /* handle container cropping */
3063 FFALIGN(h->avctx->width, 16) == h->width &&
3064 FFALIGN(h->avctx->height, 16) == h->height) {
3065 width = h->avctx->width;
3066 height = h->avctx->height;
3069 if (width <= 0 || height <= 0) {
3070 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
3072 if (h->avctx->err_recognition & AV_EF_EXPLODE)
3073 return AVERROR_INVALIDDATA;
3075 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
3076 h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
3083 h->avctx->coded_width = h->width;
3084 h->avctx->coded_height = h->height;
3085 h->avctx->width = width;
3086 h->avctx->height = height;
3091 static int h264_slice_header_init(H264Context *h, int reinit)
3093 int nb_slices = (HAVE_THREADS &&
3094 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
3095 h->avctx->thread_count : 1;
3098 h->avctx->sample_aspect_ratio = h->sps.sar;
3099 av_assert0(h->avctx->sample_aspect_ratio.den);
3100 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
3101 &h->chroma_x_shift, &h->chroma_y_shift);
3103 if (h->sps.timing_info_present_flag) {
3104 int64_t den = h->sps.time_scale;
3105 if (h->x264_build < 44U)
3107 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
3108 h->sps.num_units_in_tick, den, 1 << 30);
3111 h->avctx->hwaccel = ff_find_hwaccel(h->avctx->codec->id, h->avctx->pix_fmt);
3116 h->prev_interlaced_frame = 1;
3118 init_scan_tables(h);
3119 if (ff_h264_alloc_tables(h) < 0) {
3120 av_log(h->avctx, AV_LOG_ERROR,
3121 "Could not allocate memory for h264\n");
3122 return AVERROR(ENOMEM);
3125 if (nb_slices > MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
3128 max_slices = FFMIN(MAX_THREADS, h->mb_height);
3130 max_slices = MAX_THREADS;
3131 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
3132 " reducing to %d\n", nb_slices, max_slices);
3133 nb_slices = max_slices;
3135 h->slice_context_count = nb_slices;
3137 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
3138 if (context_init(h) < 0) {
3139 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
3143 for (i = 1; i < h->slice_context_count; i++) {
3145 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
3146 c->avctx = h->avctx;
3147 if (CONFIG_ERROR_RESILIENCE) {
3151 c->h264dsp = h->h264dsp;
3152 c->h264qpel = h->h264qpel;
3153 c->h264chroma = h->h264chroma;
3156 c->pixel_shift = h->pixel_shift;
3157 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
3158 c->width = h->width;
3159 c->height = h->height;
3160 c->linesize = h->linesize;
3161 c->uvlinesize = h->uvlinesize;
3162 c->chroma_x_shift = h->chroma_x_shift;
3163 c->chroma_y_shift = h->chroma_y_shift;
3164 c->qscale = h->qscale;
3165 c->droppable = h->droppable;
3166 c->data_partitioning = h->data_partitioning;
3167 c->low_delay = h->low_delay;
3168 c->mb_width = h->mb_width;
3169 c->mb_height = h->mb_height;
3170 c->mb_stride = h->mb_stride;
3171 c->mb_num = h->mb_num;
3172 c->flags = h->flags;
3173 c->workaround_bugs = h->workaround_bugs;
3174 c->pict_type = h->pict_type;
3176 init_scan_tables(c);
3177 clone_tables(c, h, i);
3178 c->context_initialized = 1;
3181 for (i = 0; i < h->slice_context_count; i++)
3182 if (context_init(h->thread_context[i]) < 0) {
3183 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
3188 h->context_initialized = 1;
3194 * Decode a slice header.
3195 * This will also call ff_MPV_common_init() and frame_start() as needed.
3197 * @param h h264context
3198 * @param h0 h264 master context (differs from 'h' when doing sliced based
3199 * parallel decoding)
3201 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3203 static int decode_slice_header(H264Context *h, H264Context *h0)
3205 unsigned int first_mb_in_slice;
3206 unsigned int pps_id;
3207 int num_ref_idx_active_override_flag, ret;
3208 unsigned int slice_type, tmp, i, j;
3209 int last_pic_structure, last_pic_droppable;
3211 int needs_reinit = 0;
3213 h->me.qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
3214 h->me.qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
3216 first_mb_in_slice = get_ue_golomb_long(&h->gb);
3218 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
3219 if (h0->current_slice && FIELD_PICTURE(h)) {
3223 h0->current_slice = 0;
3224 if (!h0->first_field) {
3225 if (h->cur_pic_ptr && !h->droppable) {
3226 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
3227 h->picture_structure == PICT_BOTTOM_FIELD);
3229 h->cur_pic_ptr = NULL;
3233 slice_type = get_ue_golomb_31(&h->gb);
3234 if (slice_type > 9) {
3235 av_log(h->avctx, AV_LOG_ERROR,
3236 "slice type too large (%d) at %d %d\n",
3237 slice_type, h->mb_x, h->mb_y);
3240 if (slice_type > 4) {
3242 h->slice_type_fixed = 1;
3244 h->slice_type_fixed = 0;
3246 slice_type = golomb_to_pict_type[slice_type];
3247 h->slice_type = slice_type;
3248 h->slice_type_nos = slice_type & 3;
3250 // to make a few old functions happy, it's wrong though
3251 h->pict_type = h->slice_type;
3253 pps_id = get_ue_golomb(&h->gb);
3254 if (pps_id >= MAX_PPS_COUNT) {
3255 av_log(h->avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
3258 if (!h0->pps_buffers[pps_id]) {
3259 av_log(h->avctx, AV_LOG_ERROR,
3260 "non-existing PPS %u referenced\n",
3264 h->pps = *h0->pps_buffers[pps_id];
3266 if (!h0->sps_buffers[h->pps.sps_id]) {
3267 av_log(h->avctx, AV_LOG_ERROR,
3268 "non-existing SPS %u referenced\n",
3273 if (h->pps.sps_id != h->current_sps_id ||
3274 h0->sps_buffers[h->pps.sps_id]->new) {
3275 h0->sps_buffers[h->pps.sps_id]->new = 0;
3277 h->current_sps_id = h->pps.sps_id;
3278 h->sps = *h0->sps_buffers[h->pps.sps_id];
3280 if (h->mb_width != h->sps.mb_width ||
3281 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
3282 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3283 h->cur_chroma_format_idc != h->sps.chroma_format_idc
3287 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
3288 h->chroma_format_idc != h->sps.chroma_format_idc) {
3289 h->bit_depth_luma = h->sps.bit_depth_luma;
3290 h->chroma_format_idc = h->sps.chroma_format_idc;
3293 if ((ret = h264_set_parameter_from_sps(h)) < 0)
3297 h->avctx->profile = ff_h264_get_profile(&h->sps);
3298 h->avctx->level = h->sps.level_idc;
3299 h->avctx->refs = h->sps.ref_frame_count;
3301 must_reinit = (h->context_initialized &&
3302 ( 16*h->sps.mb_width != h->avctx->coded_width
3303 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
3304 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
3305 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
3306 || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)));
3307 if (h0->avctx->pix_fmt != get_pixel_format(h0, 0))
3310 h->mb_width = h->sps.mb_width;
3311 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3312 h->mb_num = h->mb_width * h->mb_height;
3313 h->mb_stride = h->mb_width + 1;
3315 h->b_stride = h->mb_width * 4;
3317 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
3319 h->width = 16 * h->mb_width;
3320 h->height = 16 * h->mb_height;
3322 ret = init_dimensions(h);
3326 if (h->sps.video_signal_type_present_flag) {
3327 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
3329 if (h->sps.colour_description_present_flag) {
3330 if (h->avctx->colorspace != h->sps.colorspace)
3332 h->avctx->color_primaries = h->sps.color_primaries;
3333 h->avctx->color_trc = h->sps.color_trc;
3334 h->avctx->colorspace = h->sps.colorspace;
3338 if (h->context_initialized &&
3339 (h->width != h->avctx->coded_width ||
3340 h->height != h->avctx->coded_height ||
3345 av_log(h->avctx, AV_LOG_ERROR, "changing width/height on "
3346 "slice %d\n", h0->current_slice + 1);
3347 return AVERROR_INVALIDDATA;
3352 if ((ret = get_pixel_format(h, 1)) < 0)
3354 h->avctx->pix_fmt = ret;
3356 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
3357 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
3359 if ((ret = h264_slice_header_init(h, 1)) < 0) {
3360 av_log(h->avctx, AV_LOG_ERROR,
3361 "h264_slice_header_init() failed\n");
3365 if (!h->context_initialized) {
3367 av_log(h->avctx, AV_LOG_ERROR,
3368 "Cannot (re-)initialize context during parallel decoding.\n");
3372 if ((ret = get_pixel_format(h, 1)) < 0)
3374 h->avctx->pix_fmt = ret;
3376 if ((ret = h264_slice_header_init(h, 0)) < 0) {
3377 av_log(h->avctx, AV_LOG_ERROR,
3378 "h264_slice_header_init() failed\n");
3383 if (h == h0 && h->dequant_coeff_pps != pps_id) {
3384 h->dequant_coeff_pps = pps_id;
3385 init_dequant_tables(h);
3388 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
3391 h->mb_aff_frame = 0;
3392 last_pic_structure = h0->picture_structure;
3393 last_pic_droppable = h0->droppable;
3394 h->droppable = h->nal_ref_idc == 0;
3395 if (h->sps.frame_mbs_only_flag) {
3396 h->picture_structure = PICT_FRAME;
3398 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
3399 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
3402 if (get_bits1(&h->gb)) { // field_pic_flag
3403 h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
3405 h->picture_structure = PICT_FRAME;
3406 h->mb_aff_frame = h->sps.mb_aff;
3409 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
3411 if (h0->current_slice != 0) {
3412 if (last_pic_structure != h->picture_structure ||
3413 last_pic_droppable != h->droppable) {
3414 av_log(h->avctx, AV_LOG_ERROR,
3415 "Changing field mode (%d -> %d) between slices is not allowed\n",
3416 last_pic_structure, h->picture_structure);
3417 h->picture_structure = last_pic_structure;
3418 h->droppable = last_pic_droppable;
3419 return AVERROR_INVALIDDATA;
3420 } else if (!h0->cur_pic_ptr) {
3421 av_log(h->avctx, AV_LOG_ERROR,
3422 "unset cur_pic_ptr on %d. slice\n",
3423 h0->current_slice + 1);
3424 return AVERROR_INVALIDDATA;
3427 /* Shorten frame num gaps so we don't have to allocate reference
3428 * frames just to throw them away */
3429 if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
3430 int unwrap_prev_frame_num = h->prev_frame_num;
3431 int max_frame_num = 1 << h->sps.log2_max_frame_num;
3433 if (unwrap_prev_frame_num > h->frame_num)
3434 unwrap_prev_frame_num -= max_frame_num;
3436 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
3437 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
3438 if (unwrap_prev_frame_num < 0)
3439 unwrap_prev_frame_num += max_frame_num;
3441 h->prev_frame_num = unwrap_prev_frame_num;
3445 /* See if we have a decoded first field looking for a pair...
3446 * Here, we're using that to see if we should mark previously
3447 * decode frames as "finished".
3448 * We have to do that before the "dummy" in-between frame allocation,
3449 * since that can modify h->cur_pic_ptr. */
3450 if (h0->first_field) {
3451 assert(h0->cur_pic_ptr);
3452 assert(h0->cur_pic_ptr->f.data[0]);
3453 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
3455 /* Mark old field/frame as completed */
3456 if (!last_pic_droppable && h0->cur_pic_ptr->tf.owner == h0->avctx) {
3457 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3458 last_pic_structure == PICT_BOTTOM_FIELD);
3461 /* figure out if we have a complementary field pair */
3462 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
3463 /* Previous field is unmatched. Don't display it, but let it
3464 * remain for reference if marked as such. */
3465 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
3466 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3467 last_pic_structure == PICT_TOP_FIELD);
3470 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
3471 /* This and previous field were reference, but had
3472 * different frame_nums. Consider this field first in
3473 * pair. Throw away previous field except for reference
3475 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
3476 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3477 last_pic_structure == PICT_TOP_FIELD);
3480 /* Second field in complementary pair */
3481 if (!((last_pic_structure == PICT_TOP_FIELD &&
3482 h->picture_structure == PICT_BOTTOM_FIELD) ||
3483 (last_pic_structure == PICT_BOTTOM_FIELD &&
3484 h->picture_structure == PICT_TOP_FIELD))) {
3485 av_log(h->avctx, AV_LOG_ERROR,
3486 "Invalid field mode combination %d/%d\n",
3487 last_pic_structure, h->picture_structure);
3488 h->picture_structure = last_pic_structure;
3489 h->droppable = last_pic_droppable;
3490 return AVERROR_INVALIDDATA;
3491 } else if (last_pic_droppable != h->droppable) {
3492 avpriv_request_sample(h->avctx,
3493 "Found reference and non-reference fields in the same frame, which");
3494 h->picture_structure = last_pic_structure;
3495 h->droppable = last_pic_droppable;
3496 return AVERROR_PATCHWELCOME;
3502 while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 && !h0->first_field &&
3503 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
3504 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
3505 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
3506 h->frame_num, h->prev_frame_num);
3507 if (!h->sps.gaps_in_frame_num_allowed_flag)
3508 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
3509 h->last_pocs[i] = INT_MIN;
3510 if (h264_frame_start(h) < 0)
3512 h->prev_frame_num++;
3513 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
3514 h->cur_pic_ptr->frame_num = h->prev_frame_num;
3515 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
3516 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
3517 if ((ret = ff_generate_sliding_window_mmcos(h, 1)) < 0 &&
3518 h->avctx->err_recognition & AV_EF_EXPLODE)
3520 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
3521 (h->avctx->err_recognition & AV_EF_EXPLODE))
3522 return AVERROR_INVALIDDATA;
3523 /* Error concealment: if a ref is missing, copy the previous ref in its place.
3524 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
3525 * about there being no actual duplicates.
3526 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
3527 * concealing a lost frame, this probably isn't noticeable by comparison, but it should
3529 if (h->short_ref_count) {
3531 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
3532 (const uint8_t **)prev->f.data, prev->f.linesize,
3533 h->avctx->pix_fmt, h->mb_width * 16, h->mb_height * 16);
3534 h->short_ref[0]->poc = prev->poc + 2;
3536 h->short_ref[0]->frame_num = h->prev_frame_num;
3540 /* See if we have a decoded first field looking for a pair...
3541 * We're using that to see whether to continue decoding in that
3542 * frame, or to allocate a new one. */
3543 if (h0->first_field) {
3544 assert(h0->cur_pic_ptr);
3545 assert(h0->cur_pic_ptr->f.data[0]);
3546 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
3548 /* figure out if we have a complementary field pair */
3549 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
3550 /* Previous field is unmatched. Don't display it, but let it
3551 * remain for reference if marked as such. */
3552 h0->cur_pic_ptr = NULL;
3553 h0->first_field = FIELD_PICTURE(h);
3555 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
3556 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3557 h0->picture_structure==PICT_BOTTOM_FIELD);
3558 /* This and the previous field had different frame_nums.
3559 * Consider this field first in pair. Throw away previous
3560 * one except for reference purposes. */
3561 h0->first_field = 1;
3562 h0->cur_pic_ptr = NULL;
3564 /* Second field in complementary pair */
3565 h0->first_field = 0;
3569 /* Frame or first field in a potentially complementary pair */
3570 h0->first_field = FIELD_PICTURE(h);
3573 if (!FIELD_PICTURE(h) || h0->first_field) {
3574 if (h264_frame_start(h) < 0) {
3575 h0->first_field = 0;
3579 release_unused_pictures(h, 0);
3581 /* Some macroblocks can be accessed before they're available in case
3582 * of lost slices, MBAFF or threading. */
3583 if (FIELD_PICTURE(h)) {
3584 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
3585 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
3587 memset(h->slice_table, -1,
3588 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
3590 h0->last_slice_type = -1;
3592 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
3595 /* can't be in alloc_tables because linesize isn't known there.
3596 * FIXME: redo bipred weight to not require extra buffer? */
3597 for (i = 0; i < h->slice_context_count; i++)
3598 if (h->thread_context[i]) {
3599 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
3604 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
3606 av_assert1(h->mb_num == h->mb_width * h->mb_height);
3607 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
3608 first_mb_in_slice >= h->mb_num) {
3609 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3612 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
3613 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) << FIELD_OR_MBAFF_PICTURE(h);
3614 if (h->picture_structure == PICT_BOTTOM_FIELD)
3615 h->resync_mb_y = h->mb_y = h->mb_y + 1;
3616 av_assert1(h->mb_y < h->mb_height);
3618 if (h->picture_structure == PICT_FRAME) {
3619 h->curr_pic_num = h->frame_num;
3620 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
3622 h->curr_pic_num = 2 * h->frame_num + 1;
3623 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
3626 if (h->nal_unit_type == NAL_IDR_SLICE)
3627 get_ue_golomb(&h->gb); /* idr_pic_id */
3629 if (h->sps.poc_type == 0) {
3630 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
3632 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
3633 h->delta_poc_bottom = get_se_golomb(&h->gb);
3636 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
3637 h->delta_poc[0] = get_se_golomb(&h->gb);
3639 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
3640 h->delta_poc[1] = get_se_golomb(&h->gb);
3645 if (h->pps.redundant_pic_cnt_present)
3646 h->redundant_pic_count = get_ue_golomb(&h->gb);
3648 // set defaults, might be overridden a few lines later
3649 h->ref_count[0] = h->pps.ref_count[0];
3650 h->ref_count[1] = h->pps.ref_count[1];
3652 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3654 max[0] = max[1] = h->picture_structure == PICT_FRAME ? 15 : 31;
3656 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3657 h->direct_spatial_mv_pred = get_bits1(&h->gb);
3658 num_ref_idx_active_override_flag = get_bits1(&h->gb);
3660 if (num_ref_idx_active_override_flag) {
3661 h->ref_count[0] = get_ue_golomb(&h->gb) + 1;
3662 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3663 h->ref_count[1] = get_ue_golomb(&h->gb) + 1;
3665 // full range is spec-ok in this case, even for frames
3666 h->ref_count[1] = 1;
3669 if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
3670 av_log(h->avctx, AV_LOG_ERROR, "reference overflow %u > %u or %u > %u\n", h->ref_count[0]-1, max[0], h->ref_count[1]-1, max[1]);
3671 h->ref_count[0] = h->ref_count[1] = 0;
3672 return AVERROR_INVALIDDATA;
3675 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3681 h->ref_count[0] = h->ref_count[1] = 0;
3683 if (slice_type != AV_PICTURE_TYPE_I &&
3684 (h0->current_slice == 0 ||
3685 slice_type != h0->last_slice_type ||
3686 memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
3687 ff_h264_fill_default_ref_list(h);
3690 if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
3691 ff_h264_decode_ref_pic_list_reordering(h) < 0) {
3692 h->ref_count[1] = h->ref_count[0] = 0;
3696 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
3697 (h->pps.weighted_bipred_idc == 1 &&
3698 h->slice_type_nos == AV_PICTURE_TYPE_B))
3699 pred_weight_table(h);
3700 else if (h->pps.weighted_bipred_idc == 2 &&
3701 h->slice_type_nos == AV_PICTURE_TYPE_B) {
3702 implicit_weight_table(h, -1);
3705 for (i = 0; i < 2; i++) {
3706 h->luma_weight_flag[i] = 0;
3707 h->chroma_weight_flag[i] = 0;
3711 // If frame-mt is enabled, only update mmco tables for the first slice
3712 // in a field. Subsequent slices can temporarily clobber h->mmco_index
3713 // or h->mmco, which will cause ref list mix-ups and decoding errors
3714 // further down the line. This may break decoding if the first slice is
3715 // corrupt, thus we only do this if frame-mt is enabled.
3716 if (h->nal_ref_idc &&
3717 ff_h264_decode_ref_pic_marking(h0, &h->gb,
3718 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
3719 h0->current_slice == 0) < 0 &&
3720 (h->avctx->err_recognition & AV_EF_EXPLODE))
3721 return AVERROR_INVALIDDATA;
3723 if (FRAME_MBAFF(h)) {
3724 ff_h264_fill_mbaff_ref_list(h);
3726 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
3727 implicit_weight_table(h, 0);
3728 implicit_weight_table(h, 1);
3732 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3733 ff_h264_direct_dist_scale_factor(h);
3734 ff_h264_direct_ref_list_init(h);
3736 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3737 tmp = get_ue_golomb_31(&h->gb);
3739 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3742 h->cabac_init_idc = tmp;
3745 h->last_qscale_diff = 0;
3746 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
3747 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3748 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3752 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
3753 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
3754 // FIXME qscale / qp ... stuff
3755 if (h->slice_type == AV_PICTURE_TYPE_SP)
3756 get_bits1(&h->gb); /* sp_for_switch_flag */
3757 if (h->slice_type == AV_PICTURE_TYPE_SP ||
3758 h->slice_type == AV_PICTURE_TYPE_SI)
3759 get_se_golomb(&h->gb); /* slice_qs_delta */
3761 h->deblocking_filter = 1;
3762 h->slice_alpha_c0_offset = 52;
3763 h->slice_beta_offset = 52;
3764 if (h->pps.deblocking_filter_parameters_present) {
3765 tmp = get_ue_golomb_31(&h->gb);
3767 av_log(h->avctx, AV_LOG_ERROR,
3768 "deblocking_filter_idc %u out of range\n", tmp);
3771 h->deblocking_filter = tmp;
3772 if (h->deblocking_filter < 2)
3773 h->deblocking_filter ^= 1; // 1<->0
3775 if (h->deblocking_filter) {
3776 h->slice_alpha_c0_offset += get_se_golomb(&h->gb) << 1;
3777 h->slice_beta_offset += get_se_golomb(&h->gb) << 1;
3778 if (h->slice_alpha_c0_offset > 104U ||
3779 h->slice_beta_offset > 104U) {
3780 av_log(h->avctx, AV_LOG_ERROR,
3781 "deblocking filter parameters %d %d out of range\n",
3782 h->slice_alpha_c0_offset, h->slice_beta_offset);
3788 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3789 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3790 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3791 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
3792 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3793 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3794 h->nal_ref_idc == 0))
3795 h->deblocking_filter = 0;
3797 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3798 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
3799 /* Cheat slightly for speed:
3800 * Do not bother to deblock across slices. */
3801 h->deblocking_filter = 2;
3803 h0->max_contexts = 1;
3804 if (!h0->single_decode_warning) {
3805 av_log(h->avctx, AV_LOG_INFO,
3806 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3807 h0->single_decode_warning = 1;
3810 av_log(h->avctx, AV_LOG_ERROR,
3811 "Deblocking switched inside frame.\n");
3816 h->qp_thresh = 15 + 52 -
3817 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3819 h->pps.chroma_qp_index_offset[0],
3820 h->pps.chroma_qp_index_offset[1]) +
3821 6 * (h->sps.bit_depth_luma - 8);
3823 h0->last_slice_type = slice_type;
3824 memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
3825 h->slice_num = ++h0->current_slice;
3828 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
3829 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
3830 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
3831 && h->slice_num >= MAX_SLICES) {
3832 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3833 av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
3836 for (j = 0; j < 2; j++) {
3838 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3839 for (i = 0; i < 16; i++) {
3841 if (j < h->list_count && i < h->ref_count[j] && h->ref_list[j][i].f.buf[0]) {
3843 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
3844 for (k = 0; k < h->short_ref_count; k++)
3845 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
3849 for (k = 0; k < h->long_ref_count; k++)
3850 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
3851 id_list[i] = h->short_ref_count + k;
3859 for (i = 0; i < 16; i++)
3860 ref2frm[i + 2] = 4 * id_list[i] +
3861 (h->ref_list[j][i].reference & 3);
3863 ref2frm[18 + 1] = -1;
3864 for (i = 16; i < 48; i++)
3865 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3866 (h->ref_list[j][i].reference & 3);
3869 if (h->ref_count[0]) h->er.last_pic = &h->ref_list[0][0];
3870 if (h->ref_count[1]) h->er.next_pic = &h->ref_list[1][0];
3872 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
3873 av_log(h->avctx, AV_LOG_DEBUG,
3874 "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
3876 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
3878 av_get_picture_type_char(h->slice_type),
3879 h->slice_type_fixed ? " fix" : "",
3880 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3881 pps_id, h->frame_num,
3882 h->cur_pic_ptr->field_poc[0],
3883 h->cur_pic_ptr->field_poc[1],
3884 h->ref_count[0], h->ref_count[1],
3886 h->deblocking_filter,
3887 h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
3889 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3890 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
3896 int ff_h264_get_slice_type(const H264Context *h)
3898 switch (h->slice_type) {
3899 case AV_PICTURE_TYPE_P:
3901 case AV_PICTURE_TYPE_B:
3903 case AV_PICTURE_TYPE_I:
3905 case AV_PICTURE_TYPE_SP:
3907 case AV_PICTURE_TYPE_SI:
3914 static av_always_inline void fill_filter_caches_inter(H264Context *h,
3915 int mb_type, int top_xy,
3916 int left_xy[LEFT_MBS],
3918 int left_type[LEFT_MBS],
3919 int mb_xy, int list)
3921 int b_stride = h->b_stride;
3922 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3923 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3924 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3925 if (USES_LIST(top_type, list)) {
3926 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
3927 const int b8_xy = 4 * top_xy + 2;
3928 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
3929 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
3930 ref_cache[0 - 1 * 8] =
3931 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
3932 ref_cache[2 - 1 * 8] =
3933 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
3935 AV_ZERO128(mv_dst - 1 * 8);
3936 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3939 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3940 if (USES_LIST(left_type[LTOP], list)) {
3941 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
3942 const int b8_xy = 4 * left_xy[LTOP] + 1;
3943 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
3944 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
3945 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
3946 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
3947 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
3949 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
3950 ref_cache[-1 + 16] =
3951 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
3953 AV_ZERO32(mv_dst - 1 + 0);
3954 AV_ZERO32(mv_dst - 1 + 8);
3955 AV_ZERO32(mv_dst - 1 + 16);
3956 AV_ZERO32(mv_dst - 1 + 24);
3959 ref_cache[-1 + 16] =
3960 ref_cache[-1 + 24] = LIST_NOT_USED;
3965 if (!USES_LIST(mb_type, list)) {
3966 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
3967 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3968 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3969 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3970 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3975 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
3976 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
3977 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
3978 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
3979 AV_WN32A(&ref_cache[0 * 8], ref01);
3980 AV_WN32A(&ref_cache[1 * 8], ref01);
3981 AV_WN32A(&ref_cache[2 * 8], ref23);
3982 AV_WN32A(&ref_cache[3 * 8], ref23);
3986 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
3987 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
3988 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
3989 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
3990 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
3996 * @return non zero if the loop filter can be skipped
3998 static int fill_filter_caches(H264Context *h, int mb_type)
4000 const int mb_xy = h->mb_xy;
4001 int top_xy, left_xy[LEFT_MBS];
4002 int top_type, left_type[LEFT_MBS];
4006 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
4008 /* Wow, what a mess, why didn't they simplify the interlacing & intra
4009 * stuff, I can't imagine that these complex rules are worth it. */
4011 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
4012 if (FRAME_MBAFF(h)) {
4013 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
4014 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
4016 if (left_mb_field_flag != curr_mb_field_flag)
4017 left_xy[LTOP] -= h->mb_stride;
4019 if (curr_mb_field_flag)
4020 top_xy += h->mb_stride &
4021 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
4022 if (left_mb_field_flag != curr_mb_field_flag)
4023 left_xy[LBOT] += h->mb_stride;
4027 h->top_mb_xy = top_xy;
4028 h->left_mb_xy[LTOP] = left_xy[LTOP];
4029 h->left_mb_xy[LBOT] = left_xy[LBOT];
4031 /* For sufficiently low qp, filtering wouldn't do anything.
4032 * This is a conservative estimate: could also check beta_offset
4033 * and more accurate chroma_qp. */
4034 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
4035 int qp = h->cur_pic.qscale_table[mb_xy];
4036 if (qp <= qp_thresh &&
4037 (left_xy[LTOP] < 0 ||
4038 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
4040 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
4041 if (!FRAME_MBAFF(h))
4043 if ((left_xy[LTOP] < 0 ||
4044 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
4045 (top_xy < h->mb_stride ||
4046 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
4051 top_type = h->cur_pic.mb_type[top_xy];
4052 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
4053 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
4054 if (h->deblocking_filter == 2) {
4055 if (h->slice_table[top_xy] != h->slice_num)
4057 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
4058 left_type[LTOP] = left_type[LBOT] = 0;
4060 if (h->slice_table[top_xy] == 0xFFFF)
4062 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
4063 left_type[LTOP] = left_type[LBOT] = 0;
4065 h->top_type = top_type;
4066 h->left_type[LTOP] = left_type[LTOP];
4067 h->left_type[LBOT] = left_type[LBOT];
4069 if (IS_INTRA(mb_type))
4072 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
4073 top_type, left_type, mb_xy, 0);
4074 if (h->list_count == 2)
4075 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
4076 top_type, left_type, mb_xy, 1);
4078 nnz = h->non_zero_count[mb_xy];
4079 nnz_cache = h->non_zero_count_cache;
4080 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
4081 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
4082 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
4083 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
4084 h->cbp = h->cbp_table[mb_xy];
4087 nnz = h->non_zero_count[top_xy];
4088 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
4091 if (left_type[LTOP]) {
4092 nnz = h->non_zero_count[left_xy[LTOP]];
4093 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
4094 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
4095 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
4096 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
4099 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
4100 * from what the loop filter needs */
4101 if (!CABAC(h) && h->pps.transform_8x8_mode) {
4102 if (IS_8x8DCT(top_type)) {
4103 nnz_cache[4 + 8 * 0] =
4104 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
4105 nnz_cache[6 + 8 * 0] =
4106 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
4108 if (IS_8x8DCT(left_type[LTOP])) {
4109 nnz_cache[3 + 8 * 1] =
4110 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
4112 if (IS_8x8DCT(left_type[LBOT])) {
4113 nnz_cache[3 + 8 * 3] =
4114 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
4117 if (IS_8x8DCT(mb_type)) {
4118 nnz_cache[scan8[0]] =
4119 nnz_cache[scan8[1]] =
4120 nnz_cache[scan8[2]] =
4121 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
4123 nnz_cache[scan8[0 + 4]] =
4124 nnz_cache[scan8[1 + 4]] =
4125 nnz_cache[scan8[2 + 4]] =
4126 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
4128 nnz_cache[scan8[0 + 8]] =
4129 nnz_cache[scan8[1 + 8]] =
4130 nnz_cache[scan8[2 + 8]] =
4131 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
4133 nnz_cache[scan8[0 + 12]] =
4134 nnz_cache[scan8[1 + 12]] =
4135 nnz_cache[scan8[2 + 12]] =
4136 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
4143 static void loop_filter(H264Context *h, int start_x, int end_x)
4145 uint8_t *dest_y, *dest_cb, *dest_cr;
4146 int linesize, uvlinesize, mb_x, mb_y;
4147 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
4148 const int old_slice_type = h->slice_type;
4149 const int pixel_shift = h->pixel_shift;
4150 const int block_h = 16 >> h->chroma_y_shift;
4152 if (h->deblocking_filter) {
4153 for (mb_x = start_x; mb_x < end_x; mb_x++)
4154 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
4156 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
4157 h->slice_num = h->slice_table[mb_xy];
4158 mb_type = h->cur_pic.mb_type[mb_xy];
4159 h->list_count = h->list_counts[mb_xy];
4163 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
4167 dest_y = h->cur_pic.f.data[0] +
4168 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
4169 dest_cb = h->cur_pic.f.data[1] +
4170 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
4171 mb_y * h->uvlinesize * block_h;
4172 dest_cr = h->cur_pic.f.data[2] +
4173 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
4174 mb_y * h->uvlinesize * block_h;
4175 // FIXME simplify above
4178 linesize = h->mb_linesize = h->linesize * 2;
4179 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
4180 if (mb_y & 1) { // FIXME move out of this function?
4181 dest_y -= h->linesize * 15;
4182 dest_cb -= h->uvlinesize * (block_h - 1);
4183 dest_cr -= h->uvlinesize * (block_h - 1);
4186 linesize = h->mb_linesize = h->linesize;
4187 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
4189 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
4191 if (fill_filter_caches(h, mb_type))
4193 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
4194 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
4196 if (FRAME_MBAFF(h)) {
4197 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
4198 linesize, uvlinesize);
4200 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
4201 dest_cr, linesize, uvlinesize);
4205 h->slice_type = old_slice_type;
4207 h->mb_y = end_mb_y - FRAME_MBAFF(h);
4208 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
4209 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
4212 static void predict_field_decoding_flag(H264Context *h)
4214 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
4215 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
4216 h->cur_pic.mb_type[mb_xy - 1] :
4217 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
4218 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
4219 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4223 * Draw edges and report progress for the last MB row.
4225 static void decode_finish_row(H264Context *h)
4227 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
4228 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
4229 int height = 16 << FRAME_MBAFF(h);
4230 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
4232 if (h->deblocking_filter) {
4233 if ((top + height) >= pic_height)
4234 height += deblock_border;
4235 top -= deblock_border;
4238 if (top >= pic_height || (top + height) < 0)
4241 height = FFMIN(height, pic_height - top);
4243 height = top + height;
4247 ff_h264_draw_horiz_band(h, top, height);
4249 if (h->droppable || h->er.error_occurred)
4252 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
4253 h->picture_structure == PICT_BOTTOM_FIELD);
4256 static void er_add_slice(H264Context *h, int startx, int starty,
4257 int endx, int endy, int status)
4259 if (CONFIG_ERROR_RESILIENCE) {
4260 ERContext *er = &h->er;
4262 er->ref_count = h->ref_count[0];
4263 ff_er_add_slice(er, startx, starty, endx, endy, status);
4267 static int decode_slice(struct AVCodecContext *avctx, void *arg)
4269 H264Context *h = *(void **)arg;
4270 int lf_x_start = h->mb_x;
4272 h->mb_skip_run = -1;
4274 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
4276 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
4277 avctx->codec_id != AV_CODEC_ID_H264 ||
4278 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
4280 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME) {
4281 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
4283 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
4284 prev_status &= ~ VP_START;
4285 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
4286 h->er.error_occurred = 1;
4292 align_get_bits(&h->gb);
4295 ff_init_cabac_decoder(&h->cabac,
4296 h->gb.buffer + get_bits_count(&h->gb) / 8,
4297 (get_bits_left(&h->gb) + 7) / 8);
4299 ff_h264_init_cabac_states(h);
4303 int ret = ff_h264_decode_mb_cabac(h);
4305 // STOP_TIMER("decode_mb_cabac")
4308 ff_h264_hl_decode_mb(h);
4310 // FIXME optimal? or let mb_decode decode 16x32 ?
4311 if (ret >= 0 && FRAME_MBAFF(h)) {
4314 ret = ff_h264_decode_mb_cabac(h);
4317 ff_h264_hl_decode_mb(h);
4320 eos = get_cabac_terminate(&h->cabac);
4322 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
4323 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
4324 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
4325 h->mb_y, ER_MB_END);
4326 if (h->mb_x >= lf_x_start)
4327 loop_filter(h, lf_x_start, h->mb_x + 1);
4330 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
4331 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
4332 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
4333 av_log(h->avctx, AV_LOG_ERROR,
4334 "error while decoding MB %d %d, bytestream (%td)\n",
4336 h->cabac.bytestream_end - h->cabac.bytestream);
4337 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4338 h->mb_y, ER_MB_ERROR);
4342 if (++h->mb_x >= h->mb_width) {
4343 loop_filter(h, lf_x_start, h->mb_x);
4344 h->mb_x = lf_x_start = 0;
4345 decode_finish_row(h);
4347 if (FIELD_OR_MBAFF_PICTURE(h)) {
4349 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
4350 predict_field_decoding_flag(h);
4354 if (eos || h->mb_y >= h->mb_height) {
4355 tprintf(h->avctx, "slice end %d %d\n",
4356 get_bits_count(&h->gb), h->gb.size_in_bits);
4357 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
4358 h->mb_y, ER_MB_END);
4359 if (h->mb_x > lf_x_start)
4360 loop_filter(h, lf_x_start, h->mb_x);
4366 int ret = ff_h264_decode_mb_cavlc(h);
4369 ff_h264_hl_decode_mb(h);
4371 // FIXME optimal? or let mb_decode decode 16x32 ?
4372 if (ret >= 0 && FRAME_MBAFF(h)) {
4374 ret = ff_h264_decode_mb_cavlc(h);
4377 ff_h264_hl_decode_mb(h);
4382 av_log(h->avctx, AV_LOG_ERROR,
4383 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
4384 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4385 h->mb_y, ER_MB_ERROR);
4389 if (++h->mb_x >= h->mb_width) {
4390 loop_filter(h, lf_x_start, h->mb_x);
4391 h->mb_x = lf_x_start = 0;
4392 decode_finish_row(h);
4394 if (FIELD_OR_MBAFF_PICTURE(h)) {
4396 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
4397 predict_field_decoding_flag(h);
4399 if (h->mb_y >= h->mb_height) {
4400 tprintf(h->avctx, "slice end %d %d\n",
4401 get_bits_count(&h->gb), h->gb.size_in_bits);
4403 if ( get_bits_left(&h->gb) == 0
4404 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
4405 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4406 h->mb_x - 1, h->mb_y,
4411 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4420 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
4421 tprintf(h->avctx, "slice end %d %d\n",
4422 get_bits_count(&h->gb), h->gb.size_in_bits);
4423 if (get_bits_left(&h->gb) == 0) {
4424 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4425 h->mb_x - 1, h->mb_y,
4427 if (h->mb_x > lf_x_start)
4428 loop_filter(h, lf_x_start, h->mb_x);
4432 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4433 h->mb_y, ER_MB_ERROR);
4443 * Call decode_slice() for each context.
4445 * @param h h264 master context
4446 * @param context_count number of contexts to execute
4448 static int execute_decode_slices(H264Context *h, int context_count)
4450 AVCodecContext *const avctx = h->avctx;
4454 if (h->avctx->hwaccel ||
4455 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4457 if (context_count == 1) {
4458 return decode_slice(avctx, &h);
4460 av_assert0(context_count > 0);
4461 for (i = 1; i < context_count; i++) {
4462 hx = h->thread_context[i];
4463 if (CONFIG_ERROR_RESILIENCE) {
4464 hx->er.error_count = 0;
4466 hx->x264_build = h->x264_build;
4469 avctx->execute(avctx, decode_slice, h->thread_context,
4470 NULL, context_count, sizeof(void *));
4472 /* pull back stuff from slices to master context */
4473 hx = h->thread_context[context_count - 1];
4476 h->droppable = hx->droppable;
4477 h->picture_structure = hx->picture_structure;
4478 if (CONFIG_ERROR_RESILIENCE) {
4479 for (i = 1; i < context_count; i++)
4480 h->er.error_count += h->thread_context[i]->er.error_count;
4487 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
4488 int parse_extradata)
4490 AVCodecContext *const avctx = h->avctx;
4491 H264Context *hx; ///< thread context
4495 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
4496 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
4499 int first_slice = 0;
4501 h->nal_unit_type= 0;
4503 if(!h->slice_context_count)
4504 h->slice_context_count= 1;
4505 h->max_contexts = h->slice_context_count;
4506 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS)) {
4507 h->current_slice = 0;
4508 if (!h->first_field)
4509 h->cur_pic_ptr = NULL;
4510 ff_h264_reset_sei(h);
4513 if (h->nal_length_size == 4) {
4514 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
4516 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
4520 for (; pass <= 1; pass++) {
4523 next_avc = h->is_avc ? 0 : buf_size;
4533 if (buf_index >= next_avc) {
4534 if (buf_index >= buf_size - h->nal_length_size)
4537 for (i = 0; i < h->nal_length_size; i++)
4538 nalsize = (nalsize << 8) | buf[buf_index++];
4539 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
4540 av_log(h->avctx, AV_LOG_ERROR,
4541 "AVC: nal size %d\n", nalsize);
4544 next_avc = buf_index + nalsize;
4546 // start code prefix search
4547 for (; buf_index + 3 < next_avc; buf_index++)
4548 // This should always succeed in the first iteration.
4549 if (buf[buf_index] == 0 &&
4550 buf[buf_index + 1] == 0 &&
4551 buf[buf_index + 2] == 1)
4554 if (buf_index + 3 >= buf_size) {
4555 buf_index = buf_size;
4560 if (buf_index >= next_avc)
4564 hx = h->thread_context[context_count];
4566 ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
4567 &consumed, next_avc - buf_index);
4568 if (ptr == NULL || dst_length < 0) {
4572 i = buf_index + consumed;
4573 if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
4574 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
4575 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
4576 h->workaround_bugs |= FF_BUG_TRUNCATED;
4578 if (!(h->workaround_bugs & FF_BUG_TRUNCATED))
4579 while(dst_length > 0 && ptr[dst_length - 1] == 0)
4581 bit_length = !dst_length ? 0
4583 decode_rbsp_trailing(h, ptr + dst_length - 1));
4585 if (h->avctx->debug & FF_DEBUG_STARTCODE)
4586 av_log(h->avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d pass %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length, pass);
4588 if (h->is_avc && (nalsize != consumed) && nalsize)
4589 av_log(h->avctx, AV_LOG_DEBUG,
4590 "AVC: Consumed only %d bytes instead of %d\n",
4593 buf_index += consumed;
4597 /* packets can sometimes contain multiple PPS/SPS,
4598 * e.g. two PAFF field pictures in one packet, or a demuxer
4599 * which splits NALs strangely if so, when frame threading we
4600 * can't start the next thread until we've read all of them */
4601 switch (hx->nal_unit_type) {
4604 nals_needed = nal_index;
4609 init_get_bits(&hx->gb, ptr, bit_length);
4610 if (!get_ue_golomb(&hx->gb) || !first_slice)
4611 nals_needed = nal_index;
4613 first_slice = hx->nal_unit_type;
4619 switch (hx->nal_unit_type) {
4623 first_slice = hx->nal_unit_type;
4626 // FIXME do not discard SEI id
4627 if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
4631 /* Ignore per frame NAL unit type during extradata
4632 * parsing. Decoding slices is not possible in codec init
4634 if (parse_extradata) {
4635 switch (hx->nal_unit_type) {
4641 case NAL_AUXILIARY_SLICE:
4642 av_log(h->avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header/extradata\n", hx->nal_unit_type);
4643 hx->nal_unit_type = NAL_FF_IGNORE;
4649 switch (hx->nal_unit_type) {
4651 if (first_slice != NAL_IDR_SLICE) {
4652 av_log(h->avctx, AV_LOG_ERROR,
4653 "Invalid mix of idr and non-idr slices\n");
4658 idr(h); // FIXME ensure we don't lose some frames if there is reordering
4661 init_get_bits(&hx->gb, ptr, bit_length);
4663 hx->inter_gb_ptr = &hx->gb;
4664 hx->data_partitioning = 0;
4666 if ((err = decode_slice_header(hx, h)))
4669 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
4670 h->valid_recovery_point = 1;
4672 if ( h->sei_recovery_frame_cnt >= 0
4673 && ( h->recovery_frame<0
4674 || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
4675 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
4676 (1 << h->sps.log2_max_frame_num);
4678 if (!h->valid_recovery_point)
4679 h->recovery_frame = h->frame_num;
4682 h->cur_pic_ptr->f.key_frame |=
4683 (hx->nal_unit_type == NAL_IDR_SLICE);
4685 if (h->recovery_frame == h->frame_num) {
4686 h->cur_pic_ptr->sync |= 1;
4687 h->recovery_frame = -1;
4690 h->sync |= !!h->cur_pic_ptr->f.key_frame;
4691 h->sync |= 3*!!(avctx->flags2 & CODEC_FLAG2_SHOW_ALL);
4692 h->cur_pic_ptr->sync |= h->sync;
4694 if (h->current_slice == 1) {
4695 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS))
4696 decode_postinit(h, nal_index >= nals_needed);
4698 if (h->avctx->hwaccel &&
4699 h->avctx->hwaccel->start_frame(h->avctx, NULL, 0) < 0)
4701 if (CONFIG_H264_VDPAU_DECODER &&
4702 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4703 ff_vdpau_h264_picture_start(h);
4706 if (hx->redundant_pic_count == 0 &&
4707 (avctx->skip_frame < AVDISCARD_NONREF ||
4709 (avctx->skip_frame < AVDISCARD_BIDIR ||
4710 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4711 (avctx->skip_frame < AVDISCARD_NONKEY ||
4712 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4713 avctx->skip_frame < AVDISCARD_ALL) {
4714 if (avctx->hwaccel) {
4715 if (avctx->hwaccel->decode_slice(avctx,
4716 &buf[buf_index - consumed],
4719 } else if (CONFIG_H264_VDPAU_DECODER &&
4720 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
4721 static const uint8_t start_code[] = {
4723 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0], start_code,
4724 sizeof(start_code));
4725 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0], &buf[buf_index - consumed],
4732 init_get_bits(&hx->gb, ptr, bit_length);
4734 hx->inter_gb_ptr = NULL;
4736 if ((err = decode_slice_header(hx, h)) < 0)
4739 hx->data_partitioning = 1;
4742 init_get_bits(&hx->intra_gb, ptr, bit_length);
4743 hx->intra_gb_ptr = &hx->intra_gb;
4746 init_get_bits(&hx->inter_gb, ptr, bit_length);
4747 hx->inter_gb_ptr = &hx->inter_gb;
4749 av_log(h->avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
4752 if (hx->redundant_pic_count == 0 &&
4754 hx->data_partitioning &&
4755 h->cur_pic_ptr && h->context_initialized &&
4756 (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4757 (avctx->skip_frame < AVDISCARD_BIDIR ||
4758 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4759 (avctx->skip_frame < AVDISCARD_NONKEY ||
4760 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4761 avctx->skip_frame < AVDISCARD_ALL)
4765 init_get_bits(&h->gb, ptr, bit_length);
4766 ff_h264_decode_sei(h);
4769 init_get_bits(&h->gb, ptr, bit_length);
4770 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) {
4771 av_log(h->avctx, AV_LOG_DEBUG,
4772 "SPS decoding failure, trying again with the complete NAL\n");
4774 av_assert0(next_avc - buf_index + consumed == nalsize);
4775 if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4777 init_get_bits(&h->gb, &buf[buf_index + 1 - consumed],
4778 8*(next_avc - buf_index + consumed - 1));
4779 ff_h264_decode_seq_parameter_set(h);
4784 init_get_bits(&h->gb, ptr, bit_length);
4785 ff_h264_decode_picture_parameter_set(h, bit_length);
4788 case NAL_END_SEQUENCE:
4789 case NAL_END_STREAM:
4790 case NAL_FILLER_DATA:
4792 case NAL_AUXILIARY_SLICE:
4797 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4798 hx->nal_unit_type, bit_length);
4801 if (context_count == h->max_contexts) {
4802 execute_decode_slices(h, context_count);
4807 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4808 else if (err == 1) {
4809 /* Slice could not be decoded in parallel mode, copy down
4810 * NAL unit stuff to context 0 and restart. Note that
4811 * rbsp_buffer is not transferred, but since we no longer
4812 * run in parallel mode this should not be an issue. */
4813 h->nal_unit_type = hx->nal_unit_type;
4814 h->nal_ref_idc = hx->nal_ref_idc;
4821 execute_decode_slices(h, context_count);
4825 if (h->cur_pic_ptr && !h->droppable) {
4826 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
4827 h->picture_structure == PICT_BOTTOM_FIELD);
4834 * Return the number of bytes consumed for building the current frame.
4836 static int get_consumed_bytes(int pos, int buf_size)
4839 pos = 1; // avoid infinite loops (i doubt that is needed but ...)
4840 if (pos + 10 > buf_size)
4841 pos = buf_size; // oops ;)
4846 static int output_frame(H264Context *h, AVFrame *dst, AVFrame *src)
4849 int ret = av_frame_ref(dst, src);
4856 for (i = 0; i < 3; i++) {
4857 int hshift = (i > 0) ? h->chroma_x_shift : 0;
4858 int vshift = (i > 0) ? h->chroma_y_shift : 0;
4859 int off = ((h->sps.crop_left >> hshift) << h->pixel_shift) +
4860 (h->sps.crop_top >> vshift) * dst->linesize[i];
4861 dst->data[i] += off;
4866 static int decode_frame(AVCodecContext *avctx, void *data,
4867 int *got_frame, AVPacket *avpkt)
4869 const uint8_t *buf = avpkt->data;
4870 int buf_size = avpkt->size;
4871 H264Context *h = avctx->priv_data;
4872 AVFrame *pict = data;
4878 h->flags = avctx->flags;
4880 /* end of stream, output what is still in the buffers */
4881 if (buf_size == 0) {
4884 h->cur_pic_ptr = NULL;
4887 // FIXME factorize this with the output code below
4888 out = h->delayed_pic[0];
4891 h->delayed_pic[i] &&
4892 !h->delayed_pic[i]->f.key_frame &&
4893 !h->delayed_pic[i]->mmco_reset;
4895 if (h->delayed_pic[i]->poc < out->poc) {
4896 out = h->delayed_pic[i];
4900 for (i = out_idx; h->delayed_pic[i]; i++)
4901 h->delayed_pic[i] = h->delayed_pic[i + 1];
4904 out->reference &= ~DELAYED_PIC_REF;
4905 ret = output_frame(h, pict, &out->f);
4913 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4914 int cnt= buf[5]&0x1f;
4915 const uint8_t *p= buf+6;
4917 int nalsize= AV_RB16(p) + 2;
4918 if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4926 int nalsize= AV_RB16(p) + 2;
4927 if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4932 return ff_h264_decode_extradata(h, buf, buf_size);
4936 buf_index = decode_nal_units(h, buf, buf_size, 0);
4940 if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4941 av_assert0(buf_index <= buf_size);
4945 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
4946 if (avctx->skip_frame >= AVDISCARD_NONREF ||
4947 buf_size >= 4 && !memcmp("Q264", buf, 4))
4949 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4953 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) ||
4954 (h->mb_y >= h->mb_height && h->mb_height)) {
4955 if (avctx->flags2 & CODEC_FLAG2_CHUNKS)
4956 decode_postinit(h, 1);
4960 /* Wait for second field. */
4962 if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4963 ret = output_frame(h, pict, &h->next_output_pic->f);
4967 if (CONFIG_MPEGVIDEO) {
4968 ff_print_debug_info2(h->avctx, h->next_output_pic, pict, h->er.mbskip_table,
4970 h->mb_width, h->mb_height, h->mb_stride, 1);
4975 assert(pict->data[0] || !*got_frame);
4977 return get_consumed_bytes(buf_index, buf_size);
4980 av_cold void ff_h264_free_context(H264Context *h)
4984 free_tables(h, 1); // FIXME cleanup init stuff perhaps
4986 for (i = 0; i < MAX_SPS_COUNT; i++)
4987 av_freep(h->sps_buffers + i);
4989 for (i = 0; i < MAX_PPS_COUNT; i++)
4990 av_freep(h->pps_buffers + i);
4993 static av_cold int h264_decode_end(AVCodecContext *avctx)
4995 H264Context *h = avctx->priv_data;
4997 ff_h264_remove_all_refs(h);
4998 ff_h264_free_context(h);
5000 unref_picture(h, &h->cur_pic);
5005 static const AVProfile profiles[] = {
5006 { FF_PROFILE_H264_BASELINE, "Baseline" },
5007 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
5008 { FF_PROFILE_H264_MAIN, "Main" },
5009 { FF_PROFILE_H264_EXTENDED, "Extended" },
5010 { FF_PROFILE_H264_HIGH, "High" },
5011 { FF_PROFILE_H264_HIGH_10, "High 10" },
5012 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
5013 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
5014 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
5015 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
5016 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
5017 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
5018 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
5019 { FF_PROFILE_UNKNOWN },
5022 static const AVOption h264_options[] = {
5023 {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
5024 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
5028 static const AVClass h264_class = {
5029 .class_name = "H264 Decoder",
5030 .item_name = av_default_item_name,
5031 .option = h264_options,
5032 .version = LIBAVUTIL_VERSION_INT,
5035 static const AVClass h264_vdpau_class = {
5036 .class_name = "H264 VDPAU Decoder",
5037 .item_name = av_default_item_name,
5038 .option = h264_options,
5039 .version = LIBAVUTIL_VERSION_INT,
5042 AVCodec ff_h264_decoder = {
5044 .type = AVMEDIA_TYPE_VIDEO,
5045 .id = AV_CODEC_ID_H264,
5046 .priv_data_size = sizeof(H264Context),
5047 .init = ff_h264_decode_init,
5048 .close = h264_decode_end,
5049 .decode = decode_frame,
5050 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
5051 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
5052 CODEC_CAP_FRAME_THREADS,
5054 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
5055 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
5056 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
5057 .profiles = NULL_IF_CONFIG_SMALL(profiles),
5058 .priv_class = &h264_class,
5061 #if CONFIG_H264_VDPAU_DECODER
5062 AVCodec ff_h264_vdpau_decoder = {
5063 .name = "h264_vdpau",
5064 .type = AVMEDIA_TYPE_VIDEO,
5065 .id = AV_CODEC_ID_H264,
5066 .priv_data_size = sizeof(H264Context),
5067 .init = ff_h264_decode_init,
5068 .close = h264_decode_end,
5069 .decode = decode_frame,
5070 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
5072 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
5073 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
5075 .profiles = NULL_IF_CONFIG_SMALL(profiles),
5076 .priv_class = &h264_vdpau_class,