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;
278 dst->crop = src->crop;
279 dst->crop_left = src->crop_left;
280 dst->crop_top = src->crop_top;
284 unref_picture(h, dst);
289 static int alloc_scratch_buffers(H264Context *h, int linesize)
291 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
293 if (h->bipred_scratchpad)
296 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
297 // edge emu needs blocksize + filter length - 1
298 // (= 21x21 for h264)
299 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
300 h->me.scratchpad = av_mallocz(alloc_size * 2 * 16 * 2);
302 if (!h->bipred_scratchpad || !h->edge_emu_buffer || !h->me.scratchpad) {
303 av_freep(&h->bipred_scratchpad);
304 av_freep(&h->edge_emu_buffer);
305 av_freep(&h->me.scratchpad);
306 return AVERROR(ENOMEM);
309 h->me.temp = h->me.scratchpad;
314 static int init_table_pools(H264Context *h)
316 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
317 const int mb_array_size = h->mb_stride * h->mb_height;
318 const int b4_stride = h->mb_width * 4 + 1;
319 const int b4_array_size = b4_stride * h->mb_height * 4;
321 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
323 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
324 sizeof(uint32_t), av_buffer_allocz);
325 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
326 sizeof(int16_t), av_buffer_allocz);
327 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
329 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
330 !h->ref_index_pool) {
331 av_buffer_pool_uninit(&h->qscale_table_pool);
332 av_buffer_pool_uninit(&h->mb_type_pool);
333 av_buffer_pool_uninit(&h->motion_val_pool);
334 av_buffer_pool_uninit(&h->ref_index_pool);
335 return AVERROR(ENOMEM);
341 static int alloc_picture(H264Context *h, Picture *pic)
345 av_assert0(!pic->f.data[0]);
348 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
349 AV_GET_BUFFER_FLAG_REF : 0);
353 h->linesize = pic->f.linesize[0];
354 h->uvlinesize = pic->f.linesize[1];
355 pic->crop = h->sps.crop;
356 pic->crop_top = h->sps.crop_top;
357 pic->crop_left= h->sps.crop_left;
359 if (h->avctx->hwaccel) {
360 const AVHWAccel *hwaccel = h->avctx->hwaccel;
361 av_assert0(!pic->hwaccel_picture_private);
362 if (hwaccel->priv_data_size) {
363 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size);
364 if (!pic->hwaccel_priv_buf)
365 return AVERROR(ENOMEM);
366 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
370 if (!h->qscale_table_pool) {
371 ret = init_table_pools(h);
376 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
377 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
378 if (!pic->qscale_table_buf || !pic->mb_type_buf)
381 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
382 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
384 for (i = 0; i < 2; i++) {
385 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
386 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
387 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
390 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
391 pic->ref_index[i] = pic->ref_index_buf[i]->data;
396 unref_picture(h, pic);
397 return (ret < 0) ? ret : AVERROR(ENOMEM);
400 static inline int pic_is_unused(H264Context *h, Picture *pic)
402 if (pic->f.data[0] == NULL)
404 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
409 static int find_unused_picture(H264Context *h)
413 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
414 if (pic_is_unused(h, &h->DPB[i]))
417 if (i == MAX_PICTURE_COUNT)
418 return AVERROR_INVALIDDATA;
420 if (h->DPB[i].needs_realloc) {
421 h->DPB[i].needs_realloc = 0;
422 unref_picture(h, &h->DPB[i]);
429 * Check if the top & left blocks are available if needed and
430 * change the dc mode so it only uses the available blocks.
432 int ff_h264_check_intra4x4_pred_mode(H264Context *h)
434 static const int8_t top[12] = {
435 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
437 static const int8_t left[12] = {
438 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
442 if (!(h->top_samples_available & 0x8000)) {
443 for (i = 0; i < 4; i++) {
444 int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
446 av_log(h->avctx, AV_LOG_ERROR,
447 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
448 status, h->mb_x, h->mb_y);
451 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
456 if ((h->left_samples_available & 0x8888) != 0x8888) {
457 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
458 for (i = 0; i < 4; i++)
459 if (!(h->left_samples_available & mask[i])) {
460 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
462 av_log(h->avctx, AV_LOG_ERROR,
463 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
464 status, h->mb_x, h->mb_y);
467 h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
473 } // FIXME cleanup like ff_h264_check_intra_pred_mode
476 * Check if the top & left blocks are available if needed and
477 * change the dc mode so it only uses the available blocks.
479 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
481 static const int8_t top[7] = { LEFT_DC_PRED8x8, 1, -1, -1 };
482 static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
485 av_log(h->avctx, AV_LOG_ERROR,
486 "out of range intra chroma pred mode at %d %d\n",
491 if (!(h->top_samples_available & 0x8000)) {
494 av_log(h->avctx, AV_LOG_ERROR,
495 "top block unavailable for requested intra mode at %d %d\n",
501 if ((h->left_samples_available & 0x8080) != 0x8080) {
503 if (is_chroma && (h->left_samples_available & 0x8080)) {
504 // mad cow disease mode, aka MBAFF + constrained_intra_pred
505 mode = ALZHEIMER_DC_L0T_PRED8x8 +
506 (!(h->left_samples_available & 0x8000)) +
507 2 * (mode == DC_128_PRED8x8);
510 av_log(h->avctx, AV_LOG_ERROR,
511 "left block unavailable for requested intra mode at %d %d\n",
520 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
521 int *dst_length, int *consumed, int length)
527 // src[0]&0x80; // forbidden bit
528 h->nal_ref_idc = src[0] >> 5;
529 h->nal_unit_type = src[0] & 0x1F;
534 #define STARTCODE_TEST \
535 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
536 if (src[i + 2] != 3) { \
537 /* startcode, so we must be past the end */ \
542 #if HAVE_FAST_UNALIGNED
543 #define FIND_FIRST_ZERO \
544 if (i > 0 && !src[i]) \
549 for (i = 0; i + 1 < length; i += 9) {
550 if (!((~AV_RN64A(src + i) &
551 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
552 0x8000800080008080ULL))
559 for (i = 0; i + 1 < length; i += 5) {
560 if (!((~AV_RN32A(src + i) &
561 (AV_RN32A(src + i) - 0x01000101U)) &
570 for (i = 0; i + 1 < length; i += 2) {
573 if (i > 0 && src[i - 1] == 0)
579 // use second escape buffer for inter data
580 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
582 si = h->rbsp_buffer_size[bufidx];
583 av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
584 dst = h->rbsp_buffer[bufidx];
589 if(i>=length-1){ //no escaped 0
591 *consumed= length+1; //+1 for the header
592 if(h->avctx->flags2 & CODEC_FLAG2_FAST){
595 memcpy(dst, src, length);
602 while (si + 2 < length) {
603 // remove escapes (very rare 1:2^22)
604 if (src[si + 2] > 3) {
605 dst[di++] = src[si++];
606 dst[di++] = src[si++];
607 } else if (src[si] == 0 && src[si + 1] == 0) {
608 if (src[si + 2] == 3) { // escape
613 } else // next start code
617 dst[di++] = src[si++];
620 dst[di++] = src[si++];
623 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
626 *consumed = si + 1; // +1 for the header
627 /* FIXME store exact number of bits in the getbitcontext
628 * (it is needed for decoding) */
633 * Identify the exact end of the bitstream
634 * @return the length of the trailing, or 0 if damaged
636 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
641 tprintf(h->avctx, "rbsp trailing %X\n", v);
643 for (r = 1; r < 9; r++) {
651 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
652 int height, int y_offset, int list)
654 int raw_my = h->mv_cache[list][scan8[n]][1];
655 int filter_height_down = (raw_my & 3) ? 3 : 0;
656 int full_my = (raw_my >> 2) + y_offset;
657 int bottom = full_my + filter_height_down + height;
659 av_assert2(height >= 0);
661 return FFMAX(0, bottom);
664 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
665 int height, int y_offset, int list0,
666 int list1, int *nrefs)
670 y_offset += 16 * (h->mb_y >> MB_FIELD(h));
673 int ref_n = h->ref_cache[0][scan8[n]];
674 Picture *ref = &h->ref_list[0][ref_n];
676 // Error resilience puts the current picture in the ref list.
677 // Don't try to wait on these as it will cause a deadlock.
678 // Fields can wait on each other, though.
679 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
680 (ref->reference & 3) != h->picture_structure) {
681 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
682 if (refs[0][ref_n] < 0)
684 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
689 int ref_n = h->ref_cache[1][scan8[n]];
690 Picture *ref = &h->ref_list[1][ref_n];
692 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
693 (ref->reference & 3) != h->picture_structure) {
694 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
695 if (refs[1][ref_n] < 0)
697 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
703 * Wait until all reference frames are available for MC operations.
705 * @param h the H264 context
707 static void await_references(H264Context *h)
709 const int mb_xy = h->mb_xy;
710 const int mb_type = h->cur_pic.mb_type[mb_xy];
712 int nrefs[2] = { 0 };
715 memset(refs, -1, sizeof(refs));
717 if (IS_16X16(mb_type)) {
718 get_lowest_part_y(h, refs, 0, 16, 0,
719 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
720 } else if (IS_16X8(mb_type)) {
721 get_lowest_part_y(h, refs, 0, 8, 0,
722 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
723 get_lowest_part_y(h, refs, 8, 8, 8,
724 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
725 } else if (IS_8X16(mb_type)) {
726 get_lowest_part_y(h, refs, 0, 16, 0,
727 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
728 get_lowest_part_y(h, refs, 4, 16, 0,
729 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
733 av_assert2(IS_8X8(mb_type));
735 for (i = 0; i < 4; i++) {
736 const int sub_mb_type = h->sub_mb_type[i];
738 int y_offset = (i & 2) << 2;
740 if (IS_SUB_8X8(sub_mb_type)) {
741 get_lowest_part_y(h, refs, n, 8, y_offset,
742 IS_DIR(sub_mb_type, 0, 0),
743 IS_DIR(sub_mb_type, 0, 1),
745 } else if (IS_SUB_8X4(sub_mb_type)) {
746 get_lowest_part_y(h, refs, n, 4, y_offset,
747 IS_DIR(sub_mb_type, 0, 0),
748 IS_DIR(sub_mb_type, 0, 1),
750 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
751 IS_DIR(sub_mb_type, 0, 0),
752 IS_DIR(sub_mb_type, 0, 1),
754 } else if (IS_SUB_4X8(sub_mb_type)) {
755 get_lowest_part_y(h, refs, n, 8, y_offset,
756 IS_DIR(sub_mb_type, 0, 0),
757 IS_DIR(sub_mb_type, 0, 1),
759 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
760 IS_DIR(sub_mb_type, 0, 0),
761 IS_DIR(sub_mb_type, 0, 1),
765 av_assert2(IS_SUB_4X4(sub_mb_type));
766 for (j = 0; j < 4; j++) {
767 int sub_y_offset = y_offset + 2 * (j & 2);
768 get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
769 IS_DIR(sub_mb_type, 0, 0),
770 IS_DIR(sub_mb_type, 0, 1),
777 for (list = h->list_count - 1; list >= 0; list--)
778 for (ref = 0; ref < 48 && nrefs[list]; ref++) {
779 int row = refs[list][ref];
781 Picture *ref_pic = &h->ref_list[list][ref];
782 int ref_field = ref_pic->reference - 1;
783 int ref_field_picture = ref_pic->field_picture;
784 int pic_height = 16 * h->mb_height >> ref_field_picture;
789 if (!FIELD_PICTURE(h) && ref_field_picture) { // frame referencing two fields
790 ff_thread_await_progress(&ref_pic->tf,
791 FFMIN((row >> 1) - !(row & 1),
794 ff_thread_await_progress(&ref_pic->tf,
795 FFMIN((row >> 1), pic_height - 1),
797 } else if (FIELD_PICTURE(h) && !ref_field_picture) { // field referencing one field of a frame
798 ff_thread_await_progress(&ref_pic->tf,
799 FFMIN(row * 2 + ref_field,
802 } else if (FIELD_PICTURE(h)) {
803 ff_thread_await_progress(&ref_pic->tf,
804 FFMIN(row, pic_height - 1),
807 ff_thread_await_progress(&ref_pic->tf,
808 FFMIN(row, pic_height - 1),
815 static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
816 int n, int square, int height,
818 uint8_t *dest_y, uint8_t *dest_cb,
820 int src_x_offset, int src_y_offset,
821 qpel_mc_func *qpix_op,
822 h264_chroma_mc_func chroma_op,
823 int pixel_shift, int chroma_idc)
825 const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
826 int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
827 const int luma_xy = (mx & 3) + ((my & 3) << 2);
828 int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
829 uint8_t *src_y = pic->f.data[0] + offset;
830 uint8_t *src_cb, *src_cr;
832 int extra_height = 0;
834 const int full_mx = mx >> 2;
835 const int full_my = my >> 2;
836 const int pic_width = 16 * h->mb_width;
837 const int pic_height = 16 * h->mb_height >> MB_FIELD(h);
845 if (full_mx < 0 - extra_width ||
846 full_my < 0 - extra_height ||
847 full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
848 full_my + 16 /*FIXME*/ > pic_height + extra_height) {
849 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
850 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
852 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
853 full_my - 2, pic_width, pic_height);
854 src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
858 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
860 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
862 if (CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY)
865 if (chroma_idc == 3 /* yuv444 */) {
866 src_cb = pic->f.data[1] + offset;
868 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
869 src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
871 16 + 5, 16 + 5 /*FIXME*/,
872 full_mx - 2, full_my - 2,
873 pic_width, pic_height);
874 src_cb = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
876 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
878 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
880 src_cr = pic->f.data[2] + offset;
882 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
883 src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
885 16 + 5, 16 + 5 /*FIXME*/,
886 full_mx - 2, full_my - 2,
887 pic_width, pic_height);
888 src_cr = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
890 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
892 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
896 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
897 if (chroma_idc == 1 /* yuv420 */ && MB_FIELD(h)) {
898 // chroma offset when predicting from a field of opposite parity
899 my += 2 * ((h->mb_y & 1) - (pic->reference - 1));
900 emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
903 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
904 (my >> ysh) * h->mb_uvlinesize;
905 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
906 (my >> ysh) * h->mb_uvlinesize;
909 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->mb_uvlinesize,
910 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
911 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
912 src_cb = h->edge_emu_buffer;
914 chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
915 height >> (chroma_idc == 1 /* yuv420 */),
916 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
919 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->mb_uvlinesize,
920 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
921 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
922 src_cr = h->edge_emu_buffer;
924 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
925 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
928 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
929 int height, int delta,
930 uint8_t *dest_y, uint8_t *dest_cb,
932 int x_offset, int y_offset,
933 qpel_mc_func *qpix_put,
934 h264_chroma_mc_func chroma_put,
935 qpel_mc_func *qpix_avg,
936 h264_chroma_mc_func chroma_avg,
937 int list0, int list1,
938 int pixel_shift, int chroma_idc)
940 qpel_mc_func *qpix_op = qpix_put;
941 h264_chroma_mc_func chroma_op = chroma_put;
943 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
944 if (chroma_idc == 3 /* yuv444 */) {
945 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
946 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
947 } else if (chroma_idc == 2 /* yuv422 */) {
948 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
949 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
950 } else { /* yuv420 */
951 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
952 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
954 x_offset += 8 * h->mb_x;
955 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
958 Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
959 mc_dir_part(h, ref, n, square, height, delta, 0,
960 dest_y, dest_cb, dest_cr, x_offset, y_offset,
961 qpix_op, chroma_op, pixel_shift, chroma_idc);
964 chroma_op = chroma_avg;
968 Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
969 mc_dir_part(h, ref, n, square, height, delta, 1,
970 dest_y, dest_cb, dest_cr, x_offset, y_offset,
971 qpix_op, chroma_op, pixel_shift, chroma_idc);
975 static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
976 int height, int delta,
977 uint8_t *dest_y, uint8_t *dest_cb,
979 int x_offset, int y_offset,
980 qpel_mc_func *qpix_put,
981 h264_chroma_mc_func chroma_put,
982 h264_weight_func luma_weight_op,
983 h264_weight_func chroma_weight_op,
984 h264_biweight_func luma_weight_avg,
985 h264_biweight_func chroma_weight_avg,
986 int list0, int list1,
987 int pixel_shift, int chroma_idc)
991 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
992 if (chroma_idc == 3 /* yuv444 */) {
993 chroma_height = height;
994 chroma_weight_avg = luma_weight_avg;
995 chroma_weight_op = luma_weight_op;
996 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
997 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
998 } else if (chroma_idc == 2 /* yuv422 */) {
999 chroma_height = height;
1000 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1001 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1002 } else { /* yuv420 */
1003 chroma_height = height >> 1;
1004 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1005 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1007 x_offset += 8 * h->mb_x;
1008 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
1010 if (list0 && list1) {
1011 /* don't optimize for luma-only case, since B-frames usually
1012 * use implicit weights => chroma too. */
1013 uint8_t *tmp_cb = h->bipred_scratchpad;
1014 uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
1015 uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
1016 int refn0 = h->ref_cache[0][scan8[n]];
1017 int refn1 = h->ref_cache[1][scan8[n]];
1019 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
1020 dest_y, dest_cb, dest_cr,
1021 x_offset, y_offset, qpix_put, chroma_put,
1022 pixel_shift, chroma_idc);
1023 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
1024 tmp_y, tmp_cb, tmp_cr,
1025 x_offset, y_offset, qpix_put, chroma_put,
1026 pixel_shift, chroma_idc);
1028 if (h->use_weight == 2) {
1029 int weight0 = h->implicit_weight[refn0][refn1][h->mb_y & 1];
1030 int weight1 = 64 - weight0;
1031 luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
1032 height, 5, weight0, weight1, 0);
1033 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
1034 chroma_height, 5, weight0, weight1, 0);
1035 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
1036 chroma_height, 5, weight0, weight1, 0);
1038 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
1039 h->luma_log2_weight_denom,
1040 h->luma_weight[refn0][0][0],
1041 h->luma_weight[refn1][1][0],
1042 h->luma_weight[refn0][0][1] +
1043 h->luma_weight[refn1][1][1]);
1044 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
1045 h->chroma_log2_weight_denom,
1046 h->chroma_weight[refn0][0][0][0],
1047 h->chroma_weight[refn1][1][0][0],
1048 h->chroma_weight[refn0][0][0][1] +
1049 h->chroma_weight[refn1][1][0][1]);
1050 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
1051 h->chroma_log2_weight_denom,
1052 h->chroma_weight[refn0][0][1][0],
1053 h->chroma_weight[refn1][1][1][0],
1054 h->chroma_weight[refn0][0][1][1] +
1055 h->chroma_weight[refn1][1][1][1]);
1058 int list = list1 ? 1 : 0;
1059 int refn = h->ref_cache[list][scan8[n]];
1060 Picture *ref = &h->ref_list[list][refn];
1061 mc_dir_part(h, ref, n, square, height, delta, list,
1062 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1063 qpix_put, chroma_put, pixel_shift, chroma_idc);
1065 luma_weight_op(dest_y, h->mb_linesize, height,
1066 h->luma_log2_weight_denom,
1067 h->luma_weight[refn][list][0],
1068 h->luma_weight[refn][list][1]);
1069 if (h->use_weight_chroma) {
1070 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
1071 h->chroma_log2_weight_denom,
1072 h->chroma_weight[refn][list][0][0],
1073 h->chroma_weight[refn][list][0][1]);
1074 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
1075 h->chroma_log2_weight_denom,
1076 h->chroma_weight[refn][list][1][0],
1077 h->chroma_weight[refn][list][1][1]);
1082 static av_always_inline void prefetch_motion(H264Context *h, int list,
1083 int pixel_shift, int chroma_idc)
1085 /* fetch pixels for estimated mv 4 macroblocks ahead
1086 * optimized for 64byte cache lines */
1087 const int refn = h->ref_cache[list][scan8[0]];
1089 const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * h->mb_x + 8;
1090 const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * h->mb_y;
1091 uint8_t **src = h->ref_list[list][refn].f.data;
1092 int off = (mx << pixel_shift) +
1093 (my + (h->mb_x & 3) * 4) * h->mb_linesize +
1094 (64 << pixel_shift);
1095 h->vdsp.prefetch(src[0] + off, h->linesize, 4);
1096 if (chroma_idc == 3 /* yuv444 */) {
1097 h->vdsp.prefetch(src[1] + off, h->linesize, 4);
1098 h->vdsp.prefetch(src[2] + off, h->linesize, 4);
1100 off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (h->mb_x&7))*h->uvlinesize;
1101 h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
1106 static void free_tables(H264Context *h, int free_rbsp)
1111 av_freep(&h->intra4x4_pred_mode);
1112 av_freep(&h->chroma_pred_mode_table);
1113 av_freep(&h->cbp_table);
1114 av_freep(&h->mvd_table[0]);
1115 av_freep(&h->mvd_table[1]);
1116 av_freep(&h->direct_table);
1117 av_freep(&h->non_zero_count);
1118 av_freep(&h->slice_table_base);
1119 h->slice_table = NULL;
1120 av_freep(&h->list_counts);
1122 av_freep(&h->mb2b_xy);
1123 av_freep(&h->mb2br_xy);
1125 for (i = 0; i < 3; i++)
1126 av_freep(&h->visualization_buffer[i]);
1128 av_buffer_pool_uninit(&h->qscale_table_pool);
1129 av_buffer_pool_uninit(&h->mb_type_pool);
1130 av_buffer_pool_uninit(&h->motion_val_pool);
1131 av_buffer_pool_uninit(&h->ref_index_pool);
1133 if (free_rbsp && h->DPB) {
1134 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1135 unref_picture(h, &h->DPB[i]);
1137 } else if (h->DPB) {
1138 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1139 h->DPB[i].needs_realloc = 1;
1142 h->cur_pic_ptr = NULL;
1144 for (i = 0; i < MAX_THREADS; i++) {
1145 hx = h->thread_context[i];
1148 av_freep(&hx->top_borders[1]);
1149 av_freep(&hx->top_borders[0]);
1150 av_freep(&hx->bipred_scratchpad);
1151 av_freep(&hx->edge_emu_buffer);
1152 av_freep(&hx->dc_val_base);
1153 av_freep(&hx->me.scratchpad);
1154 av_freep(&hx->er.mb_index2xy);
1155 av_freep(&hx->er.error_status_table);
1156 av_freep(&hx->er.er_temp_buffer);
1157 av_freep(&hx->er.mbintra_table);
1158 av_freep(&hx->er.mbskip_table);
1161 av_freep(&hx->rbsp_buffer[1]);
1162 av_freep(&hx->rbsp_buffer[0]);
1163 hx->rbsp_buffer_size[0] = 0;
1164 hx->rbsp_buffer_size[1] = 0;
1167 av_freep(&h->thread_context[i]);
1171 static void init_dequant8_coeff_table(H264Context *h)
1174 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1176 for (i = 0; i < 6; i++) {
1177 h->dequant8_coeff[i] = h->dequant8_buffer[i];
1178 for (j = 0; j < i; j++)
1179 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
1180 64 * sizeof(uint8_t))) {
1181 h->dequant8_coeff[i] = h->dequant8_buffer[j];
1187 for (q = 0; q < max_qp + 1; q++) {
1188 int shift = div6[q];
1190 for (x = 0; x < 64; x++)
1191 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
1192 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
1193 h->pps.scaling_matrix8[i][x]) << shift;
1198 static void init_dequant4_coeff_table(H264Context *h)
1201 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1202 for (i = 0; i < 6; i++) {
1203 h->dequant4_coeff[i] = h->dequant4_buffer[i];
1204 for (j = 0; j < i; j++)
1205 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
1206 16 * sizeof(uint8_t))) {
1207 h->dequant4_coeff[i] = h->dequant4_buffer[j];
1213 for (q = 0; q < max_qp + 1; q++) {
1214 int shift = div6[q] + 2;
1216 for (x = 0; x < 16; x++)
1217 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
1218 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
1219 h->pps.scaling_matrix4[i][x]) << shift;
1224 static void init_dequant_tables(H264Context *h)
1227 init_dequant4_coeff_table(h);
1228 if (h->pps.transform_8x8_mode)
1229 init_dequant8_coeff_table(h);
1230 if (h->sps.transform_bypass) {
1231 for (i = 0; i < 6; i++)
1232 for (x = 0; x < 16; x++)
1233 h->dequant4_coeff[i][0][x] = 1 << 6;
1234 if (h->pps.transform_8x8_mode)
1235 for (i = 0; i < 6; i++)
1236 for (x = 0; x < 64; x++)
1237 h->dequant8_coeff[i][0][x] = 1 << 6;
1241 int ff_h264_alloc_tables(H264Context *h)
1243 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
1244 const int row_mb_num = 2*h->mb_stride*FFMAX(h->avctx->thread_count, 1);
1247 FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
1248 row_mb_num * 8 * sizeof(uint8_t), fail)
1249 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
1250 big_mb_num * 48 * sizeof(uint8_t), fail)
1251 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
1252 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
1253 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
1254 big_mb_num * sizeof(uint16_t), fail)
1255 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
1256 big_mb_num * sizeof(uint8_t), fail)
1257 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0],
1258 16 * row_mb_num * sizeof(uint8_t), fail);
1259 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1],
1260 16 * row_mb_num * sizeof(uint8_t), fail);
1261 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
1262 4 * big_mb_num * sizeof(uint8_t), fail);
1263 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
1264 big_mb_num * sizeof(uint8_t), fail)
1266 memset(h->slice_table_base, -1,
1267 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
1268 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
1270 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
1271 big_mb_num * sizeof(uint32_t), fail);
1272 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
1273 big_mb_num * sizeof(uint32_t), fail);
1274 for (y = 0; y < h->mb_height; y++)
1275 for (x = 0; x < h->mb_width; x++) {
1276 const int mb_xy = x + y * h->mb_stride;
1277 const int b_xy = 4 * x + 4 * y * h->b_stride;
1279 h->mb2b_xy[mb_xy] = b_xy;
1280 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
1283 if (!h->dequant4_coeff[0])
1284 init_dequant_tables(h);
1287 h->DPB = av_mallocz_array(MAX_PICTURE_COUNT, sizeof(*h->DPB));
1289 return AVERROR(ENOMEM);
1290 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1291 avcodec_get_frame_defaults(&h->DPB[i].f);
1292 avcodec_get_frame_defaults(&h->cur_pic.f);
1303 * Mimic alloc_tables(), but for every context thread.
1305 static void clone_tables(H264Context *dst, H264Context *src, int i)
1307 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
1308 dst->non_zero_count = src->non_zero_count;
1309 dst->slice_table = src->slice_table;
1310 dst->cbp_table = src->cbp_table;
1311 dst->mb2b_xy = src->mb2b_xy;
1312 dst->mb2br_xy = src->mb2br_xy;
1313 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
1314 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
1315 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
1316 dst->direct_table = src->direct_table;
1317 dst->list_counts = src->list_counts;
1318 dst->DPB = src->DPB;
1319 dst->cur_pic_ptr = src->cur_pic_ptr;
1320 dst->cur_pic = src->cur_pic;
1321 dst->bipred_scratchpad = NULL;
1322 dst->edge_emu_buffer = NULL;
1323 dst->me.scratchpad = NULL;
1324 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
1325 src->sps.chroma_format_idc);
1330 * Allocate buffers which are not shared amongst multiple threads.
1332 static int context_init(H264Context *h)
1334 ERContext *er = &h->er;
1335 int mb_array_size = h->mb_height * h->mb_stride;
1336 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
1337 int c_size = h->mb_stride * (h->mb_height + 1);
1338 int yc_size = y_size + 2 * c_size;
1341 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[0],
1342 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1343 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[1],
1344 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1346 h->ref_cache[0][scan8[5] + 1] =
1347 h->ref_cache[0][scan8[7] + 1] =
1348 h->ref_cache[0][scan8[13] + 1] =
1349 h->ref_cache[1][scan8[5] + 1] =
1350 h->ref_cache[1][scan8[7] + 1] =
1351 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
1353 if (CONFIG_ERROR_RESILIENCE) {
1355 er->avctx = h->avctx;
1357 er->decode_mb = h264_er_decode_mb;
1359 er->quarter_sample = 1;
1361 er->mb_num = h->mb_num;
1362 er->mb_width = h->mb_width;
1363 er->mb_height = h->mb_height;
1364 er->mb_stride = h->mb_stride;
1365 er->b8_stride = h->mb_width * 2 + 1;
1367 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy, (h->mb_num + 1) * sizeof(int),
1368 fail); // error ressilience code looks cleaner with this
1369 for (y = 0; y < h->mb_height; y++)
1370 for (x = 0; x < h->mb_width; x++)
1371 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
1373 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
1374 h->mb_stride + h->mb_width;
1376 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
1377 mb_array_size * sizeof(uint8_t), fail);
1379 FF_ALLOC_OR_GOTO(h->avctx, er->mbintra_table, mb_array_size, fail);
1380 memset(er->mbintra_table, 1, mb_array_size);
1382 FF_ALLOCZ_OR_GOTO(h->avctx, er->mbskip_table, mb_array_size + 2, fail);
1384 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer, h->mb_height * h->mb_stride,
1387 FF_ALLOCZ_OR_GOTO(h->avctx, h->dc_val_base, yc_size * sizeof(int16_t), fail);
1388 er->dc_val[0] = h->dc_val_base + h->mb_width * 2 + 2;
1389 er->dc_val[1] = h->dc_val_base + y_size + h->mb_stride + 1;
1390 er->dc_val[2] = er->dc_val[1] + c_size;
1391 for (i = 0; i < yc_size; i++)
1392 h->dc_val_base[i] = 1024;
1398 return -1; // free_tables will clean up for us
1401 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1402 int parse_extradata);
1404 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1406 AVCodecContext *avctx = h->avctx;
1408 if (!buf || size <= 0)
1412 int i, cnt, nalsize;
1413 const unsigned char *p = buf;
1418 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1421 /* sps and pps in the avcC always have length coded with 2 bytes,
1422 * so put a fake nal_length_size = 2 while parsing them */
1423 h->nal_length_size = 2;
1424 // Decode sps from avcC
1425 cnt = *(p + 5) & 0x1f; // Number of sps
1427 for (i = 0; i < cnt; i++) {
1428 nalsize = AV_RB16(p) + 2;
1429 if(nalsize > size - (p-buf))
1431 if (decode_nal_units(h, p, nalsize, 1) < 0) {
1432 av_log(avctx, AV_LOG_ERROR,
1433 "Decoding sps %d from avcC failed\n", i);
1438 // Decode pps from avcC
1439 cnt = *(p++); // Number of pps
1440 for (i = 0; i < cnt; i++) {
1441 nalsize = AV_RB16(p) + 2;
1442 if(nalsize > size - (p-buf))
1444 if (decode_nal_units(h, p, nalsize, 1) < 0) {
1445 av_log(avctx, AV_LOG_ERROR,
1446 "Decoding pps %d from avcC failed\n", i);
1451 // Now store right nal length size, that will be used to parse all other nals
1452 h->nal_length_size = (buf[4] & 0x03) + 1;
1455 if (decode_nal_units(h, buf, size, 1) < 0)
1461 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1463 H264Context *h = avctx->priv_data;
1468 h->bit_depth_luma = 8;
1469 h->chroma_format_idc = 1;
1471 h->avctx->bits_per_raw_sample = 8;
1472 h->cur_chroma_format_idc = 1;
1474 ff_h264dsp_init(&h->h264dsp, 8, 1);
1475 av_assert0(h->sps.bit_depth_chroma == 0);
1476 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1477 ff_h264qpel_init(&h->h264qpel, 8);
1478 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, 8, 1);
1480 h->dequant_coeff_pps = -1;
1481 h->current_sps_id = -1;
1483 /* needed so that IDCT permutation is known early */
1484 if (CONFIG_ERROR_RESILIENCE)
1485 ff_dsputil_init(&h->dsp, h->avctx);
1486 ff_videodsp_init(&h->vdsp, 8);
1488 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1489 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1491 h->picture_structure = PICT_FRAME;
1492 h->slice_context_count = 1;
1493 h->workaround_bugs = avctx->workaround_bugs;
1494 h->flags = avctx->flags;
1497 // s->decode_mb = ff_h263_decode_mb;
1498 if (!avctx->has_b_frames)
1501 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1503 ff_h264_decode_init_vlc();
1506 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1508 h->thread_context[0] = h;
1509 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1510 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1511 h->last_pocs[i] = INT_MIN;
1512 h->prev_poc_msb = 1 << 16;
1513 h->prev_frame_num = -1;
1515 ff_h264_reset_sei(h);
1516 if (avctx->codec_id == AV_CODEC_ID_H264) {
1517 if (avctx->ticks_per_frame == 1) {
1518 if(h->avctx->time_base.den < INT_MAX/2) {
1519 h->avctx->time_base.den *= 2;
1521 h->avctx->time_base.num /= 2;
1523 avctx->ticks_per_frame = 2;
1526 if (avctx->extradata_size > 0 && avctx->extradata &&
1527 ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) {
1528 ff_h264_free_context(h);
1532 if (h->sps.bitstream_restriction_flag &&
1533 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
1534 h->avctx->has_b_frames = h->sps.num_reorder_frames;
1538 ff_init_cabac_states();
1539 avctx->internal->allocate_progress = 1;
1544 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1545 #undef REBASE_PICTURE
1546 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
1547 ((pic && pic >= old_ctx->DPB && \
1548 pic < old_ctx->DPB + MAX_PICTURE_COUNT) ? \
1549 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
1551 static void copy_picture_range(Picture **to, Picture **from, int count,
1552 H264Context *new_base,
1553 H264Context *old_base)
1557 for (i = 0; i < count; i++) {
1558 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1559 IN_RANGE(from[i], old_base->DPB,
1560 sizeof(Picture) * MAX_PICTURE_COUNT) ||
1562 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1566 static void copy_parameter_set(void **to, void **from, int count, int size)
1570 for (i = 0; i < count; i++) {
1571 if (to[i] && !from[i])
1573 else if (from[i] && !to[i])
1574 to[i] = av_malloc(size);
1577 memcpy(to[i], from[i], size);
1581 static int decode_init_thread_copy(AVCodecContext *avctx)
1583 H264Context *h = avctx->priv_data;
1585 if (!avctx->internal->is_copy)
1587 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1588 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1590 h->context_initialized = 0;
1595 #define copy_fields(to, from, start_field, end_field) \
1596 memcpy(&to->start_field, &from->start_field, \
1597 (char *)&to->end_field - (char *)&to->start_field)
1599 static int h264_slice_header_init(H264Context *, int);
1601 static int h264_set_parameter_from_sps(H264Context *h);
1603 static int decode_update_thread_context(AVCodecContext *dst,
1604 const AVCodecContext *src)
1606 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1607 int inited = h->context_initialized, err = 0;
1608 int context_reinitialized = 0;
1615 (h->width != h1->width ||
1616 h->height != h1->height ||
1617 h->mb_width != h1->mb_width ||
1618 h->mb_height != h1->mb_height ||
1619 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
1620 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
1621 h->sps.colorspace != h1->sps.colorspace)) {
1623 /* set bits_per_raw_sample to the previous value. the check for changed
1624 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
1625 * the current value */
1626 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1628 av_freep(&h->bipred_scratchpad);
1630 h->width = h1->width;
1631 h->height = h1->height;
1632 h->mb_height = h1->mb_height;
1633 h->mb_width = h1->mb_width;
1634 h->mb_num = h1->mb_num;
1635 h->mb_stride = h1->mb_stride;
1636 h->b_stride = h1->b_stride;
1638 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1639 MAX_SPS_COUNT, sizeof(SPS));
1641 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1642 MAX_PPS_COUNT, sizeof(PPS));
1645 if ((err = h264_slice_header_init(h, 1)) < 0) {
1646 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
1649 context_reinitialized = 1;
1652 h264_set_parameter_from_sps(h);
1653 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
1654 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
1657 /* update linesize on resize for h264. The h264 decoder doesn't
1658 * necessarily call ff_MPV_frame_start in the new thread */
1659 h->linesize = h1->linesize;
1660 h->uvlinesize = h1->uvlinesize;
1662 /* copy block_offset since frame_start may not be called */
1663 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
1666 for (i = 0; i < MAX_SPS_COUNT; i++)
1667 av_freep(h->sps_buffers + i);
1669 for (i = 0; i < MAX_PPS_COUNT; i++)
1670 av_freep(h->pps_buffers + i);
1672 memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
1673 memcpy(&h->cabac, &h1->cabac,
1674 sizeof(H264Context) - offsetof(H264Context, cabac));
1675 av_assert0((void*)&h->cabac == &h->mb_padding + 1);
1677 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1678 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1680 memset(&h->er, 0, sizeof(h->er));
1681 memset(&h->me, 0, sizeof(h->me));
1682 memset(&h->mb, 0, sizeof(h->mb));
1683 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
1684 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
1688 h->qscale_table_pool = NULL;
1689 h->mb_type_pool = NULL;
1690 h->ref_index_pool = NULL;
1691 h->motion_val_pool = NULL;
1693 if (h1->context_initialized) {
1694 h->context_initialized = 0;
1696 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
1697 avcodec_get_frame_defaults(&h->cur_pic.f);
1698 h->cur_pic.tf.f = &h->cur_pic.f;
1700 if (ff_h264_alloc_tables(h) < 0) {
1701 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1702 return AVERROR(ENOMEM);
1707 for (i = 0; i < 2; i++) {
1708 h->rbsp_buffer[i] = NULL;
1709 h->rbsp_buffer_size[i] = 0;
1711 h->bipred_scratchpad = NULL;
1712 h->edge_emu_buffer = NULL;
1714 h->thread_context[0] = h;
1715 h->context_initialized = h1->context_initialized;
1718 h->avctx->coded_height = h1->avctx->coded_height;
1719 h->avctx->coded_width = h1->avctx->coded_width;
1720 h->avctx->width = h1->avctx->width;
1721 h->avctx->height = h1->avctx->height;
1722 h->coded_picture_number = h1->coded_picture_number;
1723 h->first_field = h1->first_field;
1724 h->picture_structure = h1->picture_structure;
1725 h->qscale = h1->qscale;
1726 h->droppable = h1->droppable;
1727 h->data_partitioning = h1->data_partitioning;
1728 h->low_delay = h1->low_delay;
1730 for (i = 0; h->DPB && i < MAX_PICTURE_COUNT; i++) {
1731 unref_picture(h, &h->DPB[i]);
1732 if (h1->DPB[i].f.data[0] &&
1733 (ret = ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
1737 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
1738 unref_picture(h, &h->cur_pic);
1739 if (h1->cur_pic.f.buf[0] && (ret = ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
1742 h->workaround_bugs = h1->workaround_bugs;
1743 h->low_delay = h1->low_delay;
1744 h->droppable = h1->droppable;
1746 // extradata/NAL handling
1747 h->is_avc = h1->is_avc;
1750 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1751 MAX_SPS_COUNT, sizeof(SPS));
1753 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1754 MAX_PPS_COUNT, sizeof(PPS));
1757 // Dequantization matrices
1758 // FIXME these are big - can they be only copied when PPS changes?
1759 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1761 for (i = 0; i < 6; i++)
1762 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1763 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1765 for (i = 0; i < 6; i++)
1766 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1767 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1769 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1772 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1775 copy_fields(h, h1, short_ref, cabac_init_idc);
1777 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
1778 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
1779 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1780 MAX_DELAYED_PIC_COUNT + 2, h, h1);
1784 if (context_reinitialized)
1785 h264_set_parameter_from_sps(h);
1787 if (!h->cur_pic_ptr)
1790 if (!h->droppable) {
1791 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1792 h->prev_poc_msb = h->poc_msb;
1793 h->prev_poc_lsb = h->poc_lsb;
1795 h->prev_frame_num_offset = h->frame_num_offset;
1796 h->prev_frame_num = h->frame_num;
1797 h->outputed_poc = h->next_outputed_poc;
1802 static int h264_frame_start(H264Context *h)
1806 const int pixel_shift = h->pixel_shift;
1808 1<<(h->sps.bit_depth_luma-1),
1809 1<<(h->sps.bit_depth_chroma-1),
1810 1<<(h->sps.bit_depth_chroma-1),
1814 if (!ff_thread_can_start_frame(h->avctx)) {
1815 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1819 release_unused_pictures(h, 1);
1820 h->cur_pic_ptr = NULL;
1822 i = find_unused_picture(h);
1824 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1829 pic->reference = h->droppable ? 0 : h->picture_structure;
1830 pic->f.coded_picture_number = h->coded_picture_number++;
1831 pic->field_picture = h->picture_structure != PICT_FRAME;
1834 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
1836 * See decode_nal_units().
1838 pic->f.key_frame = 0;
1840 pic->mmco_reset = 0;
1842 if ((ret = alloc_picture(h, pic)) < 0)
1844 if(!h->sync && !h->avctx->hwaccel &&
1845 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
1846 avpriv_color_frame(&pic->f, c);
1848 h->cur_pic_ptr = pic;
1849 unref_picture(h, &h->cur_pic);
1850 if ((ret = ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
1853 if (CONFIG_ERROR_RESILIENCE) {
1854 ff_er_frame_start(&h->er);
1856 h->er.next_pic = NULL;
1859 assert(h->linesize && h->uvlinesize);
1861 for (i = 0; i < 16; i++) {
1862 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1863 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1865 for (i = 0; i < 16; i++) {
1866 h->block_offset[16 + i] =
1867 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1868 h->block_offset[48 + 16 + i] =
1869 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1872 // s->decode = (h->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1873 // h->cur_pic.reference /* || h->contains_intra */ || 1;
1875 /* We mark the current picture as non-reference after allocating it, so
1876 * that if we break out due to an error it can be released automatically
1877 * in the next ff_MPV_frame_start().
1879 h->cur_pic_ptr->reference = 0;
1881 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
1883 h->next_output_pic = NULL;
1885 assert(h->cur_pic_ptr->long_ref == 0);
1891 * Run setup operations that must be run after slice header decoding.
1892 * This includes finding the next displayed frame.
1894 * @param h h264 master context
1895 * @param setup_finished enough NALs have been read that we can call
1896 * ff_thread_finish_setup()
1898 static void decode_postinit(H264Context *h, int setup_finished)
1900 Picture *out = h->cur_pic_ptr;
1901 Picture *cur = h->cur_pic_ptr;
1902 int i, pics, out_of_order, out_idx;
1904 h->cur_pic_ptr->f.pict_type = h->pict_type;
1906 if (h->next_output_pic)
1909 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1910 /* FIXME: if we have two PAFF fields in one packet, we can't start
1911 * the next thread here. If we have one field per packet, we can.
1912 * The check in decode_nal_units() is not good enough to find this
1913 * yet, so we assume the worst for now. */
1914 // if (setup_finished)
1915 // ff_thread_finish_setup(h->avctx);
1919 cur->f.interlaced_frame = 0;
1920 cur->f.repeat_pict = 0;
1922 /* Signal interlacing information externally. */
1923 /* Prioritize picture timing SEI information over used
1924 * decoding process if it exists. */
1926 if (h->sps.pic_struct_present_flag) {
1927 switch (h->sei_pic_struct) {
1928 case SEI_PIC_STRUCT_FRAME:
1930 case SEI_PIC_STRUCT_TOP_FIELD:
1931 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1932 cur->f.interlaced_frame = 1;
1934 case SEI_PIC_STRUCT_TOP_BOTTOM:
1935 case SEI_PIC_STRUCT_BOTTOM_TOP:
1936 if (FIELD_OR_MBAFF_PICTURE(h))
1937 cur->f.interlaced_frame = 1;
1939 // try to flag soft telecine progressive
1940 cur->f.interlaced_frame = h->prev_interlaced_frame;
1942 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1943 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1944 /* Signal the possibility of telecined film externally
1945 * (pic_struct 5,6). From these hints, let the applications
1946 * decide if they apply deinterlacing. */
1947 cur->f.repeat_pict = 1;
1949 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1950 cur->f.repeat_pict = 2;
1952 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1953 cur->f.repeat_pict = 4;
1957 if ((h->sei_ct_type & 3) &&
1958 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1959 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1961 /* Derive interlacing flag from used decoding process. */
1962 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
1964 h->prev_interlaced_frame = cur->f.interlaced_frame;
1966 if (cur->field_poc[0] != cur->field_poc[1]) {
1967 /* Derive top_field_first from field pocs. */
1968 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1970 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1971 /* Use picture timing SEI information. Even if it is a
1972 * information of a past frame, better than nothing. */
1973 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1974 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1975 cur->f.top_field_first = 1;
1977 cur->f.top_field_first = 0;
1979 /* Most likely progressive */
1980 cur->f.top_field_first = 0;
1984 cur->mmco_reset = h->mmco_reset;
1986 // FIXME do something with unavailable reference frames
1988 /* Sort B-frames into display order */
1990 if (h->sps.bitstream_restriction_flag &&
1991 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
1992 h->avctx->has_b_frames = h->sps.num_reorder_frames;
1996 if (h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
1997 !h->sps.bitstream_restriction_flag) {
1998 h->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
2002 for (i = 0; 1; i++) {
2003 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
2005 h->last_pocs[i-1] = cur->poc;
2008 h->last_pocs[i-1]= h->last_pocs[i];
2011 out_of_order = MAX_DELAYED_PIC_COUNT - i;
2012 if( cur->f.pict_type == AV_PICTURE_TYPE_B
2013 || (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))
2014 out_of_order = FFMAX(out_of_order, 1);
2015 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
2016 av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
2017 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
2018 h->last_pocs[i] = INT_MIN;
2019 h->last_pocs[0] = cur->poc;
2020 cur->mmco_reset = 1;
2021 } else if(h->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
2022 av_log(h->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
2023 h->avctx->has_b_frames = out_of_order;
2028 while (h->delayed_pic[pics])
2031 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
2033 h->delayed_pic[pics++] = cur;
2034 if (cur->reference == 0)
2035 cur->reference = DELAYED_PIC_REF;
2037 out = h->delayed_pic[0];
2039 for (i = 1; h->delayed_pic[i] &&
2040 !h->delayed_pic[i]->f.key_frame &&
2041 !h->delayed_pic[i]->mmco_reset;
2043 if (h->delayed_pic[i]->poc < out->poc) {
2044 out = h->delayed_pic[i];
2047 if (h->avctx->has_b_frames == 0 &&
2048 (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
2049 h->next_outputed_poc = INT_MIN;
2050 out_of_order = out->poc < h->next_outputed_poc;
2052 if (out_of_order || pics > h->avctx->has_b_frames) {
2053 out->reference &= ~DELAYED_PIC_REF;
2054 // for frame threading, the owner must be the second field's thread or
2055 // else the first thread can release the picture and reuse it unsafely
2056 for (i = out_idx; h->delayed_pic[i]; i++)
2057 h->delayed_pic[i] = h->delayed_pic[i + 1];
2059 if (!out_of_order && pics > h->avctx->has_b_frames) {
2060 h->next_output_pic = out;
2061 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
2062 h->next_outputed_poc = INT_MIN;
2064 h->next_outputed_poc = out->poc;
2066 av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
2069 if (h->next_output_pic && h->next_output_pic->sync) {
2073 if (setup_finished && !h->avctx->hwaccel)
2074 ff_thread_finish_setup(h->avctx);
2077 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
2078 uint8_t *src_cb, uint8_t *src_cr,
2079 int linesize, int uvlinesize,
2082 uint8_t *top_border;
2084 const int pixel_shift = h->pixel_shift;
2085 int chroma444 = CHROMA444(h);
2086 int chroma422 = CHROMA422(h);
2089 src_cb -= uvlinesize;
2090 src_cr -= uvlinesize;
2092 if (!simple && FRAME_MBAFF(h)) {
2095 top_border = h->top_borders[0][h->mb_x];
2096 AV_COPY128(top_border, src_y + 15 * linesize);
2098 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
2099 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2102 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2103 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
2104 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
2105 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
2107 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
2108 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
2110 } else if (chroma422) {
2112 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2113 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
2115 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
2116 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
2120 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
2121 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
2123 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
2124 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
2129 } else if (MB_MBAFF(h)) {
2135 top_border = h->top_borders[top_idx][h->mb_x];
2136 /* There are two lines saved, the line above the top macroblock
2137 * of a pair, and the line above the bottom macroblock. */
2138 AV_COPY128(top_border, src_y + 16 * linesize);
2140 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
2142 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2145 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
2146 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
2147 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
2148 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
2150 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
2151 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
2153 } else if (chroma422) {
2155 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
2156 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
2158 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
2159 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
2163 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
2164 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
2166 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
2167 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
2173 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
2174 uint8_t *src_cb, uint8_t *src_cr,
2175 int linesize, int uvlinesize,
2176 int xchg, int chroma444,
2177 int simple, int pixel_shift)
2179 int deblock_topleft;
2182 uint8_t *top_border_m1;
2183 uint8_t *top_border;
2185 if (!simple && FRAME_MBAFF(h)) {
2190 top_idx = MB_MBAFF(h) ? 0 : 1;
2194 if (h->deblocking_filter == 2) {
2195 deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num;
2196 deblock_top = h->top_type;
2198 deblock_topleft = (h->mb_x > 0);
2199 deblock_top = (h->mb_y > !!MB_FIELD(h));
2202 src_y -= linesize + 1 + pixel_shift;
2203 src_cb -= uvlinesize + 1 + pixel_shift;
2204 src_cr -= uvlinesize + 1 + pixel_shift;
2206 top_border_m1 = h->top_borders[top_idx][h->mb_x - 1];
2207 top_border = h->top_borders[top_idx][h->mb_x];
2209 #define XCHG(a, b, xchg) \
2210 if (pixel_shift) { \
2212 AV_SWAP64(b + 0, a + 0); \
2213 AV_SWAP64(b + 8, a + 8); \
2223 if (deblock_topleft) {
2224 XCHG(top_border_m1 + (8 << pixel_shift),
2225 src_y - (7 << pixel_shift), 1);
2227 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
2228 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
2229 if (h->mb_x + 1 < h->mb_width) {
2230 XCHG(h->top_borders[top_idx][h->mb_x + 1],
2231 src_y + (17 << pixel_shift), 1);
2234 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2236 if (deblock_topleft) {
2237 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2238 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2240 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
2241 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
2242 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
2243 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
2244 if (h->mb_x + 1 < h->mb_width) {
2245 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
2246 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
2250 if (deblock_topleft) {
2251 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2252 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2254 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
2255 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
2261 static av_always_inline int dctcoef_get(int16_t *mb, int high_bit_depth,
2264 if (high_bit_depth) {
2265 return AV_RN32A(((int32_t *)mb) + index);
2267 return AV_RN16A(mb + index);
2270 static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth,
2271 int index, int value)
2273 if (high_bit_depth) {
2274 AV_WN32A(((int32_t *)mb) + index, value);
2276 AV_WN16A(mb + index, value);
2279 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
2280 int mb_type, int is_h264,
2282 int transform_bypass,
2286 uint8_t *dest_y, int p)
2288 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2289 void (*idct_dc_add)(uint8_t *dst, int16_t *block, int stride);
2291 int qscale = p == 0 ? h->qscale : h->chroma_qp[p - 1];
2292 block_offset += 16 * p;
2293 if (IS_INTRA4x4(mb_type)) {
2294 if (IS_8x8DCT(mb_type)) {
2295 if (transform_bypass) {
2297 idct_add = h->h264dsp.h264_add_pixels8_clear;
2299 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
2300 idct_add = h->h264dsp.h264_idct8_add;
2302 for (i = 0; i < 16; i += 4) {
2303 uint8_t *const ptr = dest_y + block_offset[i];
2304 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2305 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2306 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2308 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2309 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
2310 (h->topright_samples_available << i) & 0x4000, linesize);
2312 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2313 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2315 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2320 if (transform_bypass) {
2322 idct_add = h->h264dsp.h264_add_pixels4_clear;
2324 idct_dc_add = h->h264dsp.h264_idct_dc_add;
2325 idct_add = h->h264dsp.h264_idct_add;
2327 for (i = 0; i < 16; i++) {
2328 uint8_t *const ptr = dest_y + block_offset[i];
2329 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2331 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2332 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2337 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
2338 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
2339 av_assert2(h->mb_y || linesize <= block_offset[i]);
2340 if (!topright_avail) {
2342 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
2343 topright = (uint8_t *)&tr_high;
2345 tr = ptr[3 - linesize] * 0x01010101u;
2346 topright = (uint8_t *)&tr;
2349 topright = ptr + (4 << pixel_shift) - linesize;
2353 h->hpc.pred4x4[dir](ptr, topright, linesize);
2354 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2357 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2358 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2360 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2361 } else if (CONFIG_SVQ3_DECODER)
2362 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
2368 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
2370 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
2371 if (!transform_bypass)
2372 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
2374 h->dequant4_coeff[p][qscale][0]);
2376 static const uint8_t dc_mapping[16] = {
2377 0 * 16, 1 * 16, 4 * 16, 5 * 16,
2378 2 * 16, 3 * 16, 6 * 16, 7 * 16,
2379 8 * 16, 9 * 16, 12 * 16, 13 * 16,
2380 10 * 16, 11 * 16, 14 * 16, 15 * 16 };
2381 for (i = 0; i < 16; i++)
2382 dctcoef_set(h->mb + (p * 256 << pixel_shift),
2383 pixel_shift, dc_mapping[i],
2384 dctcoef_get(h->mb_luma_dc[p],
2388 } else if (CONFIG_SVQ3_DECODER)
2389 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
2390 h->mb_luma_dc[p], qscale);
2394 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
2395 int is_h264, int simple,
2396 int transform_bypass,
2400 uint8_t *dest_y, int p)
2402 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2404 block_offset += 16 * p;
2405 if (!IS_INTRA4x4(mb_type)) {
2407 if (IS_INTRA16x16(mb_type)) {
2408 if (transform_bypass) {
2409 if (h->sps.profile_idc == 244 &&
2410 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
2411 h->intra16x16_pred_mode == HOR_PRED8x8)) {
2412 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
2413 h->mb + (p * 256 << pixel_shift),
2416 for (i = 0; i < 16; i++)
2417 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
2418 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2419 h->h264dsp.h264_add_pixels4_clear(dest_y + block_offset[i],
2420 h->mb + (i * 16 + p * 256 << pixel_shift),
2424 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
2425 h->mb + (p * 256 << pixel_shift),
2427 h->non_zero_count_cache + p * 5 * 8);
2429 } else if (h->cbp & 15) {
2430 if (transform_bypass) {
2431 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2432 idct_add = IS_8x8DCT(mb_type) ? h->h264dsp.h264_add_pixels8_clear
2433 : h->h264dsp.h264_add_pixels4_clear;
2434 for (i = 0; i < 16; i += di)
2435 if (h->non_zero_count_cache[scan8[i + p * 16]])
2436 idct_add(dest_y + block_offset[i],
2437 h->mb + (i * 16 + p * 256 << pixel_shift),
2440 if (IS_8x8DCT(mb_type))
2441 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
2442 h->mb + (p * 256 << pixel_shift),
2444 h->non_zero_count_cache + p * 5 * 8);
2446 h->h264dsp.h264_idct_add16(dest_y, block_offset,
2447 h->mb + (p * 256 << pixel_shift),
2449 h->non_zero_count_cache + p * 5 * 8);
2452 } else if (CONFIG_SVQ3_DECODER) {
2453 for (i = 0; i < 16; i++)
2454 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
2455 // FIXME benchmark weird rule, & below
2456 uint8_t *const ptr = dest_y + block_offset[i];
2457 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
2458 h->qscale, IS_INTRA(mb_type) ? 1 : 0);
2466 #include "h264_mb_template.c"
2470 #include "h264_mb_template.c"
2474 #include "h264_mb_template.c"
2476 void ff_h264_hl_decode_mb(H264Context *h)
2478 const int mb_xy = h->mb_xy;
2479 const int mb_type = h->cur_pic.mb_type[mb_xy];
2480 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || h->qscale == 0;
2483 if (is_complex || h->pixel_shift)
2484 hl_decode_mb_444_complex(h);
2486 hl_decode_mb_444_simple_8(h);
2487 } else if (is_complex) {
2488 hl_decode_mb_complex(h);
2489 } else if (h->pixel_shift) {
2490 hl_decode_mb_simple_16(h);
2492 hl_decode_mb_simple_8(h);
2495 static int pred_weight_table(H264Context *h)
2498 int luma_def, chroma_def;
2501 h->use_weight_chroma = 0;
2502 h->luma_log2_weight_denom = get_ue_golomb(&h->gb);
2503 if (h->sps.chroma_format_idc)
2504 h->chroma_log2_weight_denom = get_ue_golomb(&h->gb);
2505 luma_def = 1 << h->luma_log2_weight_denom;
2506 chroma_def = 1 << h->chroma_log2_weight_denom;
2508 for (list = 0; list < 2; list++) {
2509 h->luma_weight_flag[list] = 0;
2510 h->chroma_weight_flag[list] = 0;
2511 for (i = 0; i < h->ref_count[list]; i++) {
2512 int luma_weight_flag, chroma_weight_flag;
2514 luma_weight_flag = get_bits1(&h->gb);
2515 if (luma_weight_flag) {
2516 h->luma_weight[i][list][0] = get_se_golomb(&h->gb);
2517 h->luma_weight[i][list][1] = get_se_golomb(&h->gb);
2518 if (h->luma_weight[i][list][0] != luma_def ||
2519 h->luma_weight[i][list][1] != 0) {
2521 h->luma_weight_flag[list] = 1;
2524 h->luma_weight[i][list][0] = luma_def;
2525 h->luma_weight[i][list][1] = 0;
2528 if (h->sps.chroma_format_idc) {
2529 chroma_weight_flag = get_bits1(&h->gb);
2530 if (chroma_weight_flag) {
2532 for (j = 0; j < 2; j++) {
2533 h->chroma_weight[i][list][j][0] = get_se_golomb(&h->gb);
2534 h->chroma_weight[i][list][j][1] = get_se_golomb(&h->gb);
2535 if (h->chroma_weight[i][list][j][0] != chroma_def ||
2536 h->chroma_weight[i][list][j][1] != 0) {
2537 h->use_weight_chroma = 1;
2538 h->chroma_weight_flag[list] = 1;
2543 for (j = 0; j < 2; j++) {
2544 h->chroma_weight[i][list][j][0] = chroma_def;
2545 h->chroma_weight[i][list][j][1] = 0;
2550 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2553 h->use_weight = h->use_weight || h->use_weight_chroma;
2558 * Initialize implicit_weight table.
2559 * @param field 0/1 initialize the weight for interlaced MBAFF
2560 * -1 initializes the rest
2562 static void implicit_weight_table(H264Context *h, int field)
2564 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2566 for (i = 0; i < 2; i++) {
2567 h->luma_weight_flag[i] = 0;
2568 h->chroma_weight_flag[i] = 0;
2572 if (h->picture_structure == PICT_FRAME) {
2573 cur_poc = h->cur_pic_ptr->poc;
2575 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
2577 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
2578 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2580 h->use_weight_chroma = 0;
2584 ref_count0 = h->ref_count[0];
2585 ref_count1 = h->ref_count[1];
2587 cur_poc = h->cur_pic_ptr->field_poc[field];
2589 ref_count0 = 16 + 2 * h->ref_count[0];
2590 ref_count1 = 16 + 2 * h->ref_count[1];
2594 h->use_weight_chroma = 2;
2595 h->luma_log2_weight_denom = 5;
2596 h->chroma_log2_weight_denom = 5;
2598 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2599 int poc0 = h->ref_list[0][ref0].poc;
2600 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2602 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2603 int poc1 = h->ref_list[1][ref1].poc;
2604 int td = av_clip(poc1 - poc0, -128, 127);
2606 int tb = av_clip(cur_poc - poc0, -128, 127);
2607 int tx = (16384 + (FFABS(td) >> 1)) / td;
2608 int dist_scale_factor = (tb * tx + 32) >> 8;
2609 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2610 w = 64 - dist_scale_factor;
2614 h->implicit_weight[ref0][ref1][0] =
2615 h->implicit_weight[ref0][ref1][1] = w;
2617 h->implicit_weight[ref0][ref1][field] = w;
2624 * instantaneous decoder refresh.
2626 static void idr(H264Context *h)
2629 ff_h264_remove_all_refs(h);
2630 h->prev_frame_num = 0;
2631 h->prev_frame_num_offset = 0;
2632 h->prev_poc_msb = 1<<16;
2633 h->prev_poc_lsb = 0;
2634 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2635 h->last_pocs[i] = INT_MIN;
2638 /* forget old pics after a seek */
2639 static void flush_change(H264Context *h)
2643 h->outputed_poc = h->next_outputed_poc = INT_MIN;
2644 h->prev_interlaced_frame = 1;
2647 h->prev_frame_num = -1;
2648 if (h->cur_pic_ptr) {
2649 h->cur_pic_ptr->reference = 0;
2650 for (j=i=0; h->delayed_pic[i]; i++)
2651 if (h->delayed_pic[i] != h->cur_pic_ptr)
2652 h->delayed_pic[j++] = h->delayed_pic[i];
2653 h->delayed_pic[j] = NULL;
2656 memset(h->ref_list[0], 0, sizeof(h->ref_list[0]));
2657 memset(h->ref_list[1], 0, sizeof(h->ref_list[1]));
2658 memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0]));
2659 memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1]));
2660 ff_h264_reset_sei(h);
2661 h->recovery_frame= -1;
2664 h->current_slice = 0;
2667 /* forget old pics after a seek */
2668 static void flush_dpb(AVCodecContext *avctx)
2670 H264Context *h = avctx->priv_data;
2673 for (i = 0; i <= MAX_DELAYED_PIC_COUNT; i++) {
2674 if (h->delayed_pic[i])
2675 h->delayed_pic[i]->reference = 0;
2676 h->delayed_pic[i] = NULL;
2682 for (i = 0; i < MAX_PICTURE_COUNT; i++)
2683 unref_picture(h, &h->DPB[i]);
2684 h->cur_pic_ptr = NULL;
2685 unref_picture(h, &h->cur_pic);
2687 h->mb_x = h->mb_y = 0;
2689 h->parse_context.state = -1;
2690 h->parse_context.frame_start_found = 0;
2691 h->parse_context.overread = 0;
2692 h->parse_context.overread_index = 0;
2693 h->parse_context.index = 0;
2694 h->parse_context.last_index = 0;
2697 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
2699 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2702 h->frame_num_offset = h->prev_frame_num_offset;
2703 if (h->frame_num < h->prev_frame_num)
2704 h->frame_num_offset += max_frame_num;
2706 if (h->sps.poc_type == 0) {
2707 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2709 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2710 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2711 else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2712 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2714 h->poc_msb = h->prev_poc_msb;
2716 field_poc[1] = h->poc_msb + h->poc_lsb;
2717 if (h->picture_structure == PICT_FRAME)
2718 field_poc[1] += h->delta_poc_bottom;
2719 } else if (h->sps.poc_type == 1) {
2720 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2723 if (h->sps.poc_cycle_length != 0)
2724 abs_frame_num = h->frame_num_offset + h->frame_num;
2728 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2731 expected_delta_per_poc_cycle = 0;
2732 for (i = 0; i < h->sps.poc_cycle_length; i++)
2733 // FIXME integrate during sps parse
2734 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2736 if (abs_frame_num > 0) {
2737 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2738 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2740 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2741 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2742 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2746 if (h->nal_ref_idc == 0)
2747 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2749 field_poc[0] = expectedpoc + h->delta_poc[0];
2750 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2752 if (h->picture_structure == PICT_FRAME)
2753 field_poc[1] += h->delta_poc[1];
2755 int poc = 2 * (h->frame_num_offset + h->frame_num);
2757 if (!h->nal_ref_idc)
2764 if (h->picture_structure != PICT_BOTTOM_FIELD)
2765 pic_field_poc[0] = field_poc[0];
2766 if (h->picture_structure != PICT_TOP_FIELD)
2767 pic_field_poc[1] = field_poc[1];
2769 *pic_poc = FFMIN(pic_field_poc[0], pic_field_poc[1]);
2775 * initialize scan tables
2777 static void init_scan_tables(H264Context *h)
2780 for (i = 0; i < 16; i++) {
2781 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2782 h->zigzag_scan[i] = T(zigzag_scan[i]);
2783 h->field_scan[i] = T(field_scan[i]);
2786 for (i = 0; i < 64; i++) {
2787 #define T(x) (x >> 3) | ((x & 7) << 3)
2788 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2789 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2790 h->field_scan8x8[i] = T(field_scan8x8[i]);
2791 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2794 if (h->sps.transform_bypass) { // FIXME same ugly
2795 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2796 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
2797 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2798 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
2799 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2800 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2802 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2803 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
2804 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2805 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
2806 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2807 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2811 static int field_end(H264Context *h, int in_setup)
2813 AVCodecContext *const avctx = h->avctx;
2817 if (CONFIG_H264_VDPAU_DECODER &&
2818 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2819 ff_vdpau_h264_set_reference_frames(h);
2821 if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2822 if (!h->droppable) {
2823 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2824 h->prev_poc_msb = h->poc_msb;
2825 h->prev_poc_lsb = h->poc_lsb;
2827 h->prev_frame_num_offset = h->frame_num_offset;
2828 h->prev_frame_num = h->frame_num;
2829 h->outputed_poc = h->next_outputed_poc;
2832 if (avctx->hwaccel) {
2833 if (avctx->hwaccel->end_frame(avctx) < 0)
2834 av_log(avctx, AV_LOG_ERROR,
2835 "hardware accelerator failed to decode picture\n");
2838 if (CONFIG_H264_VDPAU_DECODER &&
2839 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2840 ff_vdpau_h264_picture_complete(h);
2843 * FIXME: Error handling code does not seem to support interlaced
2844 * when slices span multiple rows
2845 * The ff_er_add_slice calls don't work right for bottom
2846 * fields; they cause massive erroneous error concealing
2847 * Error marking covers both fields (top and bottom).
2848 * This causes a mismatched s->error_count
2849 * and a bad error table. Further, the error count goes to
2850 * INT_MAX when called for bottom field, because mb_y is
2851 * past end by one (callers fault) and resync_mb_y != 0
2852 * causes problems for the first MB line, too.
2854 if (CONFIG_ERROR_RESILIENCE &&
2855 !FIELD_PICTURE(h) && h->current_slice && !h->sps.new) {
2856 h->er.cur_pic = h->cur_pic_ptr;
2857 ff_er_frame_end(&h->er);
2859 if (!in_setup && !h->droppable)
2860 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
2861 h->picture_structure == PICT_BOTTOM_FIELD);
2864 h->current_slice = 0;
2870 * Replicate H264 "master" context to thread contexts.
2872 static int clone_slice(H264Context *dst, H264Context *src)
2874 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2875 dst->cur_pic_ptr = src->cur_pic_ptr;
2876 dst->cur_pic = src->cur_pic;
2877 dst->linesize = src->linesize;
2878 dst->uvlinesize = src->uvlinesize;
2879 dst->first_field = src->first_field;
2881 dst->prev_poc_msb = src->prev_poc_msb;
2882 dst->prev_poc_lsb = src->prev_poc_lsb;
2883 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2884 dst->prev_frame_num = src->prev_frame_num;
2885 dst->short_ref_count = src->short_ref_count;
2887 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2888 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2889 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2891 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2892 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2898 * Compute profile from profile_idc and constraint_set?_flags.
2902 * @return profile as defined by FF_PROFILE_H264_*
2904 int ff_h264_get_profile(SPS *sps)
2906 int profile = sps->profile_idc;
2908 switch (sps->profile_idc) {
2909 case FF_PROFILE_H264_BASELINE:
2910 // constraint_set1_flag set to 1
2911 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2913 case FF_PROFILE_H264_HIGH_10:
2914 case FF_PROFILE_H264_HIGH_422:
2915 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2916 // constraint_set3_flag set to 1
2917 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
2924 static int h264_set_parameter_from_sps(H264Context *h)
2926 if (h->flags & CODEC_FLAG_LOW_DELAY ||
2927 (h->sps.bitstream_restriction_flag &&
2928 !h->sps.num_reorder_frames)) {
2929 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
2930 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
2931 "Reenabling low delay requires a codec flush.\n");
2936 if (h->avctx->has_b_frames < 2)
2937 h->avctx->has_b_frames = !h->low_delay;
2939 if (h->sps.bit_depth_luma != h->sps.bit_depth_chroma) {
2940 avpriv_request_sample(h->avctx,
2941 "Different chroma and luma bit depth");
2942 return AVERROR_PATCHWELCOME;
2945 if (h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2946 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
2947 if (h->avctx->codec &&
2948 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU &&
2949 (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
2950 av_log(h->avctx, AV_LOG_ERROR,
2951 "VDPAU decoding does not support video colorspace.\n");
2952 return AVERROR_INVALIDDATA;
2954 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 &&
2955 h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13) {
2956 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
2957 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
2958 h->pixel_shift = h->sps.bit_depth_luma > 8;
2960 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
2961 h->sps.chroma_format_idc);
2962 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
2963 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
2964 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
2965 h->sps.chroma_format_idc);
2967 if (CONFIG_ERROR_RESILIENCE)
2968 ff_dsputil_init(&h->dsp, h->avctx);
2969 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
2971 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
2972 h->sps.bit_depth_luma);
2973 return AVERROR_INVALIDDATA;
2979 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
2981 switch (h->sps.bit_depth_luma) {
2984 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
2985 return AV_PIX_FMT_GBRP9;
2987 return AV_PIX_FMT_YUV444P9;
2988 } else if (CHROMA422(h))
2989 return AV_PIX_FMT_YUV422P9;
2991 return AV_PIX_FMT_YUV420P9;
2995 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
2996 return AV_PIX_FMT_GBRP10;
2998 return AV_PIX_FMT_YUV444P10;
2999 } else if (CHROMA422(h))
3000 return AV_PIX_FMT_YUV422P10;
3002 return AV_PIX_FMT_YUV420P10;
3006 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3007 return AV_PIX_FMT_GBRP12;
3009 return AV_PIX_FMT_YUV444P12;
3010 } else if (CHROMA422(h))
3011 return AV_PIX_FMT_YUV422P12;
3013 return AV_PIX_FMT_YUV420P12;
3017 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3018 return AV_PIX_FMT_GBRP14;
3020 return AV_PIX_FMT_YUV444P14;
3021 } else if (CHROMA422(h))
3022 return AV_PIX_FMT_YUV422P14;
3024 return AV_PIX_FMT_YUV420P14;
3028 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3029 av_log(h->avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
3030 return AV_PIX_FMT_GBR24P;
3031 } else if (h->avctx->colorspace == AVCOL_SPC_YCGCO) {
3032 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
3034 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
3035 : AV_PIX_FMT_YUV444P;
3036 } else if (CHROMA422(h)) {
3037 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
3038 : AV_PIX_FMT_YUV422P;
3041 const enum AVPixelFormat * fmt = h->avctx->codec->pix_fmts ?
3042 h->avctx->codec->pix_fmts :
3043 h->avctx->color_range == AVCOL_RANGE_JPEG ?
3044 h264_hwaccel_pixfmt_list_jpeg_420 :
3045 h264_hwaccel_pixfmt_list_420;
3047 for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++)
3048 if (fmt[i] == h->avctx->pix_fmt && !force_callback)
3050 return ff_thread_get_format(h->avctx, fmt);
3054 av_log(h->avctx, AV_LOG_ERROR,
3055 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3056 return AVERROR_INVALIDDATA;
3060 /* export coded and cropped frame dimensions to AVCodecContext */
3061 static int init_dimensions(H264Context *h)
3063 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
3064 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
3065 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
3066 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
3068 /* handle container cropping */
3070 FFALIGN(h->avctx->width, 16) == h->width &&
3071 FFALIGN(h->avctx->height, 16) == h->height) {
3072 width = h->avctx->width;
3073 height = h->avctx->height;
3076 if (width <= 0 || height <= 0) {
3077 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
3079 if (h->avctx->err_recognition & AV_EF_EXPLODE)
3080 return AVERROR_INVALIDDATA;
3082 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
3083 h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
3090 h->avctx->coded_width = h->width;
3091 h->avctx->coded_height = h->height;
3092 h->avctx->width = width;
3093 h->avctx->height = height;
3098 static int h264_slice_header_init(H264Context *h, int reinit)
3100 int nb_slices = (HAVE_THREADS &&
3101 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
3102 h->avctx->thread_count : 1;
3105 h->avctx->sample_aspect_ratio = h->sps.sar;
3106 av_assert0(h->avctx->sample_aspect_ratio.den);
3107 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
3108 &h->chroma_x_shift, &h->chroma_y_shift);
3110 if (h->sps.timing_info_present_flag) {
3111 int64_t den = h->sps.time_scale;
3112 if (h->x264_build < 44U)
3114 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
3115 h->sps.num_units_in_tick, den, 1 << 30);
3118 h->avctx->hwaccel = ff_find_hwaccel(h->avctx->codec->id, h->avctx->pix_fmt);
3123 h->prev_interlaced_frame = 1;
3125 init_scan_tables(h);
3126 if (ff_h264_alloc_tables(h) < 0) {
3127 av_log(h->avctx, AV_LOG_ERROR,
3128 "Could not allocate memory for h264\n");
3129 return AVERROR(ENOMEM);
3132 if (nb_slices > MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
3135 max_slices = FFMIN(MAX_THREADS, h->mb_height);
3137 max_slices = MAX_THREADS;
3138 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
3139 " reducing to %d\n", nb_slices, max_slices);
3140 nb_slices = max_slices;
3142 h->slice_context_count = nb_slices;
3144 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
3145 if (context_init(h) < 0) {
3146 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
3150 for (i = 1; i < h->slice_context_count; i++) {
3152 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
3153 c->avctx = h->avctx;
3154 if (CONFIG_ERROR_RESILIENCE) {
3158 c->h264dsp = h->h264dsp;
3159 c->h264qpel = h->h264qpel;
3160 c->h264chroma = h->h264chroma;
3163 c->pixel_shift = h->pixel_shift;
3164 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
3165 c->width = h->width;
3166 c->height = h->height;
3167 c->linesize = h->linesize;
3168 c->uvlinesize = h->uvlinesize;
3169 c->chroma_x_shift = h->chroma_x_shift;
3170 c->chroma_y_shift = h->chroma_y_shift;
3171 c->qscale = h->qscale;
3172 c->droppable = h->droppable;
3173 c->data_partitioning = h->data_partitioning;
3174 c->low_delay = h->low_delay;
3175 c->mb_width = h->mb_width;
3176 c->mb_height = h->mb_height;
3177 c->mb_stride = h->mb_stride;
3178 c->mb_num = h->mb_num;
3179 c->flags = h->flags;
3180 c->workaround_bugs = h->workaround_bugs;
3181 c->pict_type = h->pict_type;
3183 init_scan_tables(c);
3184 clone_tables(c, h, i);
3185 c->context_initialized = 1;
3188 for (i = 0; i < h->slice_context_count; i++)
3189 if (context_init(h->thread_context[i]) < 0) {
3190 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
3195 h->context_initialized = 1;
3201 * Decode a slice header.
3202 * This will also call ff_MPV_common_init() and frame_start() as needed.
3204 * @param h h264context
3205 * @param h0 h264 master context (differs from 'h' when doing sliced based
3206 * parallel decoding)
3208 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3210 static int decode_slice_header(H264Context *h, H264Context *h0)
3212 unsigned int first_mb_in_slice;
3213 unsigned int pps_id;
3214 int num_ref_idx_active_override_flag, ret;
3215 unsigned int slice_type, tmp, i, j;
3216 int last_pic_structure, last_pic_droppable;
3218 int needs_reinit = 0;
3220 h->me.qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
3221 h->me.qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
3223 first_mb_in_slice = get_ue_golomb_long(&h->gb);
3225 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
3226 if (h0->current_slice && FIELD_PICTURE(h)) {
3230 h0->current_slice = 0;
3231 if (!h0->first_field) {
3232 if (h->cur_pic_ptr && !h->droppable) {
3233 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
3234 h->picture_structure == PICT_BOTTOM_FIELD);
3236 h->cur_pic_ptr = NULL;
3240 slice_type = get_ue_golomb_31(&h->gb);
3241 if (slice_type > 9) {
3242 av_log(h->avctx, AV_LOG_ERROR,
3243 "slice type too large (%d) at %d %d\n",
3244 slice_type, h->mb_x, h->mb_y);
3247 if (slice_type > 4) {
3249 h->slice_type_fixed = 1;
3251 h->slice_type_fixed = 0;
3253 slice_type = golomb_to_pict_type[slice_type];
3254 h->slice_type = slice_type;
3255 h->slice_type_nos = slice_type & 3;
3257 // to make a few old functions happy, it's wrong though
3258 h->pict_type = h->slice_type;
3260 pps_id = get_ue_golomb(&h->gb);
3261 if (pps_id >= MAX_PPS_COUNT) {
3262 av_log(h->avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
3265 if (!h0->pps_buffers[pps_id]) {
3266 av_log(h->avctx, AV_LOG_ERROR,
3267 "non-existing PPS %u referenced\n",
3271 h->pps = *h0->pps_buffers[pps_id];
3273 if (!h0->sps_buffers[h->pps.sps_id]) {
3274 av_log(h->avctx, AV_LOG_ERROR,
3275 "non-existing SPS %u referenced\n",
3280 if (h->pps.sps_id != h->current_sps_id ||
3281 h0->sps_buffers[h->pps.sps_id]->new) {
3282 h0->sps_buffers[h->pps.sps_id]->new = 0;
3284 h->current_sps_id = h->pps.sps_id;
3285 h->sps = *h0->sps_buffers[h->pps.sps_id];
3287 if (h->mb_width != h->sps.mb_width ||
3288 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
3289 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3290 h->cur_chroma_format_idc != h->sps.chroma_format_idc
3294 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
3295 h->chroma_format_idc != h->sps.chroma_format_idc) {
3296 h->bit_depth_luma = h->sps.bit_depth_luma;
3297 h->chroma_format_idc = h->sps.chroma_format_idc;
3300 if ((ret = h264_set_parameter_from_sps(h)) < 0)
3304 h->avctx->profile = ff_h264_get_profile(&h->sps);
3305 h->avctx->level = h->sps.level_idc;
3306 h->avctx->refs = h->sps.ref_frame_count;
3308 must_reinit = (h->context_initialized &&
3309 ( 16*h->sps.mb_width != h->avctx->coded_width
3310 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
3311 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
3312 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
3313 || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)
3314 || h->mb_width != h->sps.mb_width
3315 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
3317 if (h0->avctx->pix_fmt != get_pixel_format(h0, 0))
3320 h->mb_width = h->sps.mb_width;
3321 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3322 h->mb_num = h->mb_width * h->mb_height;
3323 h->mb_stride = h->mb_width + 1;
3325 h->b_stride = h->mb_width * 4;
3327 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
3329 h->width = 16 * h->mb_width;
3330 h->height = 16 * h->mb_height;
3332 ret = init_dimensions(h);
3336 if (h->sps.video_signal_type_present_flag) {
3337 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
3339 if (h->sps.colour_description_present_flag) {
3340 if (h->avctx->colorspace != h->sps.colorspace)
3342 h->avctx->color_primaries = h->sps.color_primaries;
3343 h->avctx->color_trc = h->sps.color_trc;
3344 h->avctx->colorspace = h->sps.colorspace;
3348 if (h->context_initialized &&
3349 (h->width != h->avctx->coded_width ||
3350 h->height != h->avctx->coded_height ||
3355 av_log(h->avctx, AV_LOG_ERROR, "changing width/height on "
3356 "slice %d\n", h0->current_slice + 1);
3357 return AVERROR_INVALIDDATA;
3362 if ((ret = get_pixel_format(h, 1)) < 0)
3364 h->avctx->pix_fmt = ret;
3366 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
3367 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
3369 if ((ret = h264_slice_header_init(h, 1)) < 0) {
3370 av_log(h->avctx, AV_LOG_ERROR,
3371 "h264_slice_header_init() failed\n");
3375 if (!h->context_initialized) {
3377 av_log(h->avctx, AV_LOG_ERROR,
3378 "Cannot (re-)initialize context during parallel decoding.\n");
3382 if ((ret = get_pixel_format(h, 1)) < 0)
3384 h->avctx->pix_fmt = ret;
3386 if ((ret = h264_slice_header_init(h, 0)) < 0) {
3387 av_log(h->avctx, AV_LOG_ERROR,
3388 "h264_slice_header_init() failed\n");
3393 if (h == h0 && h->dequant_coeff_pps != pps_id) {
3394 h->dequant_coeff_pps = pps_id;
3395 init_dequant_tables(h);
3398 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
3401 h->mb_aff_frame = 0;
3402 last_pic_structure = h0->picture_structure;
3403 last_pic_droppable = h0->droppable;
3404 h->droppable = h->nal_ref_idc == 0;
3405 if (h->sps.frame_mbs_only_flag) {
3406 h->picture_structure = PICT_FRAME;
3408 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
3409 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
3412 if (get_bits1(&h->gb)) { // field_pic_flag
3413 h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
3415 h->picture_structure = PICT_FRAME;
3416 h->mb_aff_frame = h->sps.mb_aff;
3419 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
3421 if (h0->current_slice != 0) {
3422 if (last_pic_structure != h->picture_structure ||
3423 last_pic_droppable != h->droppable) {
3424 av_log(h->avctx, AV_LOG_ERROR,
3425 "Changing field mode (%d -> %d) between slices is not allowed\n",
3426 last_pic_structure, h->picture_structure);
3427 h->picture_structure = last_pic_structure;
3428 h->droppable = last_pic_droppable;
3429 return AVERROR_INVALIDDATA;
3430 } else if (!h0->cur_pic_ptr) {
3431 av_log(h->avctx, AV_LOG_ERROR,
3432 "unset cur_pic_ptr on %d. slice\n",
3433 h0->current_slice + 1);
3434 return AVERROR_INVALIDDATA;
3437 /* Shorten frame num gaps so we don't have to allocate reference
3438 * frames just to throw them away */
3439 if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
3440 int unwrap_prev_frame_num = h->prev_frame_num;
3441 int max_frame_num = 1 << h->sps.log2_max_frame_num;
3443 if (unwrap_prev_frame_num > h->frame_num)
3444 unwrap_prev_frame_num -= max_frame_num;
3446 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
3447 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
3448 if (unwrap_prev_frame_num < 0)
3449 unwrap_prev_frame_num += max_frame_num;
3451 h->prev_frame_num = unwrap_prev_frame_num;
3455 /* See if we have a decoded first field looking for a pair...
3456 * Here, we're using that to see if we should mark previously
3457 * decode frames as "finished".
3458 * We have to do that before the "dummy" in-between frame allocation,
3459 * since that can modify h->cur_pic_ptr. */
3460 if (h0->first_field) {
3461 assert(h0->cur_pic_ptr);
3462 assert(h0->cur_pic_ptr->f.data[0]);
3463 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
3465 /* Mark old field/frame as completed */
3466 if (!last_pic_droppable && h0->cur_pic_ptr->tf.owner == h0->avctx) {
3467 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3468 last_pic_structure == PICT_BOTTOM_FIELD);
3471 /* figure out if we have a complementary field pair */
3472 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
3473 /* Previous field is unmatched. Don't display it, but let it
3474 * remain for reference if marked as such. */
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 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
3481 /* This and previous field were reference, but had
3482 * different frame_nums. Consider this field first in
3483 * pair. Throw away previous field except for reference
3485 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
3486 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3487 last_pic_structure == PICT_TOP_FIELD);
3490 /* Second field in complementary pair */
3491 if (!((last_pic_structure == PICT_TOP_FIELD &&
3492 h->picture_structure == PICT_BOTTOM_FIELD) ||
3493 (last_pic_structure == PICT_BOTTOM_FIELD &&
3494 h->picture_structure == PICT_TOP_FIELD))) {
3495 av_log(h->avctx, AV_LOG_ERROR,
3496 "Invalid field mode combination %d/%d\n",
3497 last_pic_structure, h->picture_structure);
3498 h->picture_structure = last_pic_structure;
3499 h->droppable = last_pic_droppable;
3500 return AVERROR_INVALIDDATA;
3501 } else if (last_pic_droppable != h->droppable) {
3502 avpriv_request_sample(h->avctx,
3503 "Found reference and non-reference fields in the same frame, which");
3504 h->picture_structure = last_pic_structure;
3505 h->droppable = last_pic_droppable;
3506 return AVERROR_PATCHWELCOME;
3512 while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 && !h0->first_field &&
3513 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
3514 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
3515 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
3516 h->frame_num, h->prev_frame_num);
3517 if (!h->sps.gaps_in_frame_num_allowed_flag)
3518 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
3519 h->last_pocs[i] = INT_MIN;
3520 if (h264_frame_start(h) < 0)
3522 h->prev_frame_num++;
3523 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
3524 h->cur_pic_ptr->frame_num = h->prev_frame_num;
3525 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
3526 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
3527 if ((ret = ff_generate_sliding_window_mmcos(h, 1)) < 0 &&
3528 h->avctx->err_recognition & AV_EF_EXPLODE)
3530 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
3531 (h->avctx->err_recognition & AV_EF_EXPLODE))
3532 return AVERROR_INVALIDDATA;
3533 /* Error concealment: if a ref is missing, copy the previous ref in its place.
3534 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
3535 * about there being no actual duplicates.
3536 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
3537 * concealing a lost frame, this probably isn't noticeable by comparison, but it should
3539 if (h->short_ref_count) {
3541 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
3542 (const uint8_t **)prev->f.data, prev->f.linesize,
3543 h->avctx->pix_fmt, h->mb_width * 16, h->mb_height * 16);
3544 h->short_ref[0]->poc = prev->poc + 2;
3546 h->short_ref[0]->frame_num = h->prev_frame_num;
3550 /* See if we have a decoded first field looking for a pair...
3551 * We're using that to see whether to continue decoding in that
3552 * frame, or to allocate a new one. */
3553 if (h0->first_field) {
3554 assert(h0->cur_pic_ptr);
3555 assert(h0->cur_pic_ptr->f.data[0]);
3556 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
3558 /* figure out if we have a complementary field pair */
3559 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
3560 /* Previous field is unmatched. Don't display it, but let it
3561 * remain for reference if marked as such. */
3562 h0->cur_pic_ptr = NULL;
3563 h0->first_field = FIELD_PICTURE(h);
3565 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
3566 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3567 h0->picture_structure==PICT_BOTTOM_FIELD);
3568 /* This and the previous field had different frame_nums.
3569 * Consider this field first in pair. Throw away previous
3570 * one except for reference purposes. */
3571 h0->first_field = 1;
3572 h0->cur_pic_ptr = NULL;
3574 /* Second field in complementary pair */
3575 h0->first_field = 0;
3579 /* Frame or first field in a potentially complementary pair */
3580 h0->first_field = FIELD_PICTURE(h);
3583 if (!FIELD_PICTURE(h) || h0->first_field) {
3584 if (h264_frame_start(h) < 0) {
3585 h0->first_field = 0;
3589 release_unused_pictures(h, 0);
3591 /* Some macroblocks can be accessed before they're available in case
3592 * of lost slices, MBAFF or threading. */
3593 if (FIELD_PICTURE(h)) {
3594 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
3595 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
3597 memset(h->slice_table, -1,
3598 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
3600 h0->last_slice_type = -1;
3602 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
3605 /* can't be in alloc_tables because linesize isn't known there.
3606 * FIXME: redo bipred weight to not require extra buffer? */
3607 for (i = 0; i < h->slice_context_count; i++)
3608 if (h->thread_context[i]) {
3609 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
3614 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
3616 av_assert1(h->mb_num == h->mb_width * h->mb_height);
3617 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
3618 first_mb_in_slice >= h->mb_num) {
3619 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3622 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
3623 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) << FIELD_OR_MBAFF_PICTURE(h);
3624 if (h->picture_structure == PICT_BOTTOM_FIELD)
3625 h->resync_mb_y = h->mb_y = h->mb_y + 1;
3626 av_assert1(h->mb_y < h->mb_height);
3628 if (h->picture_structure == PICT_FRAME) {
3629 h->curr_pic_num = h->frame_num;
3630 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
3632 h->curr_pic_num = 2 * h->frame_num + 1;
3633 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
3636 if (h->nal_unit_type == NAL_IDR_SLICE)
3637 get_ue_golomb(&h->gb); /* idr_pic_id */
3639 if (h->sps.poc_type == 0) {
3640 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
3642 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
3643 h->delta_poc_bottom = get_se_golomb(&h->gb);
3646 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
3647 h->delta_poc[0] = get_se_golomb(&h->gb);
3649 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
3650 h->delta_poc[1] = get_se_golomb(&h->gb);
3653 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
3655 if (h->pps.redundant_pic_cnt_present)
3656 h->redundant_pic_count = get_ue_golomb(&h->gb);
3658 // set defaults, might be overridden a few lines later
3659 h->ref_count[0] = h->pps.ref_count[0];
3660 h->ref_count[1] = h->pps.ref_count[1];
3662 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3664 max[0] = max[1] = h->picture_structure == PICT_FRAME ? 15 : 31;
3666 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3667 h->direct_spatial_mv_pred = get_bits1(&h->gb);
3668 num_ref_idx_active_override_flag = get_bits1(&h->gb);
3670 if (num_ref_idx_active_override_flag) {
3671 h->ref_count[0] = get_ue_golomb(&h->gb) + 1;
3672 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3673 h->ref_count[1] = get_ue_golomb(&h->gb) + 1;
3675 // full range is spec-ok in this case, even for frames
3676 h->ref_count[1] = 1;
3679 if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
3680 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]);
3681 h->ref_count[0] = h->ref_count[1] = 0;
3682 return AVERROR_INVALIDDATA;
3685 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3691 h->ref_count[0] = h->ref_count[1] = 0;
3693 if (slice_type != AV_PICTURE_TYPE_I &&
3694 (h0->current_slice == 0 ||
3695 slice_type != h0->last_slice_type ||
3696 memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
3697 ff_h264_fill_default_ref_list(h);
3700 if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
3701 ff_h264_decode_ref_pic_list_reordering(h) < 0) {
3702 h->ref_count[1] = h->ref_count[0] = 0;
3706 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
3707 (h->pps.weighted_bipred_idc == 1 &&
3708 h->slice_type_nos == AV_PICTURE_TYPE_B))
3709 pred_weight_table(h);
3710 else if (h->pps.weighted_bipred_idc == 2 &&
3711 h->slice_type_nos == AV_PICTURE_TYPE_B) {
3712 implicit_weight_table(h, -1);
3715 for (i = 0; i < 2; i++) {
3716 h->luma_weight_flag[i] = 0;
3717 h->chroma_weight_flag[i] = 0;
3721 // If frame-mt is enabled, only update mmco tables for the first slice
3722 // in a field. Subsequent slices can temporarily clobber h->mmco_index
3723 // or h->mmco, which will cause ref list mix-ups and decoding errors
3724 // further down the line. This may break decoding if the first slice is
3725 // corrupt, thus we only do this if frame-mt is enabled.
3726 if (h->nal_ref_idc &&
3727 ff_h264_decode_ref_pic_marking(h0, &h->gb,
3728 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
3729 h0->current_slice == 0) < 0 &&
3730 (h->avctx->err_recognition & AV_EF_EXPLODE))
3731 return AVERROR_INVALIDDATA;
3733 if (FRAME_MBAFF(h)) {
3734 ff_h264_fill_mbaff_ref_list(h);
3736 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
3737 implicit_weight_table(h, 0);
3738 implicit_weight_table(h, 1);
3742 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3743 ff_h264_direct_dist_scale_factor(h);
3744 ff_h264_direct_ref_list_init(h);
3746 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3747 tmp = get_ue_golomb_31(&h->gb);
3749 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3752 h->cabac_init_idc = tmp;
3755 h->last_qscale_diff = 0;
3756 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
3757 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3758 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3762 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
3763 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
3764 // FIXME qscale / qp ... stuff
3765 if (h->slice_type == AV_PICTURE_TYPE_SP)
3766 get_bits1(&h->gb); /* sp_for_switch_flag */
3767 if (h->slice_type == AV_PICTURE_TYPE_SP ||
3768 h->slice_type == AV_PICTURE_TYPE_SI)
3769 get_se_golomb(&h->gb); /* slice_qs_delta */
3771 h->deblocking_filter = 1;
3772 h->slice_alpha_c0_offset = 52;
3773 h->slice_beta_offset = 52;
3774 if (h->pps.deblocking_filter_parameters_present) {
3775 tmp = get_ue_golomb_31(&h->gb);
3777 av_log(h->avctx, AV_LOG_ERROR,
3778 "deblocking_filter_idc %u out of range\n", tmp);
3781 h->deblocking_filter = tmp;
3782 if (h->deblocking_filter < 2)
3783 h->deblocking_filter ^= 1; // 1<->0
3785 if (h->deblocking_filter) {
3786 h->slice_alpha_c0_offset += get_se_golomb(&h->gb) << 1;
3787 h->slice_beta_offset += get_se_golomb(&h->gb) << 1;
3788 if (h->slice_alpha_c0_offset > 104U ||
3789 h->slice_beta_offset > 104U) {
3790 av_log(h->avctx, AV_LOG_ERROR,
3791 "deblocking filter parameters %d %d out of range\n",
3792 h->slice_alpha_c0_offset, h->slice_beta_offset);
3798 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3799 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3800 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3801 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
3802 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3803 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3804 h->nal_ref_idc == 0))
3805 h->deblocking_filter = 0;
3807 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3808 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
3809 /* Cheat slightly for speed:
3810 * Do not bother to deblock across slices. */
3811 h->deblocking_filter = 2;
3813 h0->max_contexts = 1;
3814 if (!h0->single_decode_warning) {
3815 av_log(h->avctx, AV_LOG_INFO,
3816 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3817 h0->single_decode_warning = 1;
3820 av_log(h->avctx, AV_LOG_ERROR,
3821 "Deblocking switched inside frame.\n");
3826 h->qp_thresh = 15 + 52 -
3827 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3829 h->pps.chroma_qp_index_offset[0],
3830 h->pps.chroma_qp_index_offset[1]) +
3831 6 * (h->sps.bit_depth_luma - 8);
3833 h0->last_slice_type = slice_type;
3834 memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
3835 h->slice_num = ++h0->current_slice;
3838 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
3839 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
3840 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
3841 && h->slice_num >= MAX_SLICES) {
3842 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3843 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);
3846 for (j = 0; j < 2; j++) {
3848 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3849 for (i = 0; i < 16; i++) {
3851 if (j < h->list_count && i < h->ref_count[j] && h->ref_list[j][i].f.buf[0]) {
3853 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
3854 for (k = 0; k < h->short_ref_count; k++)
3855 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
3859 for (k = 0; k < h->long_ref_count; k++)
3860 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
3861 id_list[i] = h->short_ref_count + k;
3869 for (i = 0; i < 16; i++)
3870 ref2frm[i + 2] = 4 * id_list[i] +
3871 (h->ref_list[j][i].reference & 3);
3873 ref2frm[18 + 1] = -1;
3874 for (i = 16; i < 48; i++)
3875 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3876 (h->ref_list[j][i].reference & 3);
3879 if (h->ref_count[0]) h->er.last_pic = &h->ref_list[0][0];
3880 if (h->ref_count[1]) h->er.next_pic = &h->ref_list[1][0];
3882 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
3883 av_log(h->avctx, AV_LOG_DEBUG,
3884 "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",
3886 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
3888 av_get_picture_type_char(h->slice_type),
3889 h->slice_type_fixed ? " fix" : "",
3890 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3891 pps_id, h->frame_num,
3892 h->cur_pic_ptr->field_poc[0],
3893 h->cur_pic_ptr->field_poc[1],
3894 h->ref_count[0], h->ref_count[1],
3896 h->deblocking_filter,
3897 h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
3899 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3900 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
3906 int ff_h264_get_slice_type(const H264Context *h)
3908 switch (h->slice_type) {
3909 case AV_PICTURE_TYPE_P:
3911 case AV_PICTURE_TYPE_B:
3913 case AV_PICTURE_TYPE_I:
3915 case AV_PICTURE_TYPE_SP:
3917 case AV_PICTURE_TYPE_SI:
3924 static av_always_inline void fill_filter_caches_inter(H264Context *h,
3925 int mb_type, int top_xy,
3926 int left_xy[LEFT_MBS],
3928 int left_type[LEFT_MBS],
3929 int mb_xy, int list)
3931 int b_stride = h->b_stride;
3932 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3933 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3934 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3935 if (USES_LIST(top_type, list)) {
3936 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
3937 const int b8_xy = 4 * top_xy + 2;
3938 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
3939 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
3940 ref_cache[0 - 1 * 8] =
3941 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
3942 ref_cache[2 - 1 * 8] =
3943 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
3945 AV_ZERO128(mv_dst - 1 * 8);
3946 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3949 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3950 if (USES_LIST(left_type[LTOP], list)) {
3951 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
3952 const int b8_xy = 4 * left_xy[LTOP] + 1;
3953 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
3954 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
3955 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
3956 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
3957 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
3959 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
3960 ref_cache[-1 + 16] =
3961 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
3963 AV_ZERO32(mv_dst - 1 + 0);
3964 AV_ZERO32(mv_dst - 1 + 8);
3965 AV_ZERO32(mv_dst - 1 + 16);
3966 AV_ZERO32(mv_dst - 1 + 24);
3969 ref_cache[-1 + 16] =
3970 ref_cache[-1 + 24] = LIST_NOT_USED;
3975 if (!USES_LIST(mb_type, list)) {
3976 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
3977 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3978 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3979 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3980 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3985 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
3986 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
3987 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
3988 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
3989 AV_WN32A(&ref_cache[0 * 8], ref01);
3990 AV_WN32A(&ref_cache[1 * 8], ref01);
3991 AV_WN32A(&ref_cache[2 * 8], ref23);
3992 AV_WN32A(&ref_cache[3 * 8], ref23);
3996 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
3997 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
3998 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
3999 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
4000 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
4006 * @return non zero if the loop filter can be skipped
4008 static int fill_filter_caches(H264Context *h, int mb_type)
4010 const int mb_xy = h->mb_xy;
4011 int top_xy, left_xy[LEFT_MBS];
4012 int top_type, left_type[LEFT_MBS];
4016 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
4018 /* Wow, what a mess, why didn't they simplify the interlacing & intra
4019 * stuff, I can't imagine that these complex rules are worth it. */
4021 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
4022 if (FRAME_MBAFF(h)) {
4023 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
4024 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
4026 if (left_mb_field_flag != curr_mb_field_flag)
4027 left_xy[LTOP] -= h->mb_stride;
4029 if (curr_mb_field_flag)
4030 top_xy += h->mb_stride &
4031 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
4032 if (left_mb_field_flag != curr_mb_field_flag)
4033 left_xy[LBOT] += h->mb_stride;
4037 h->top_mb_xy = top_xy;
4038 h->left_mb_xy[LTOP] = left_xy[LTOP];
4039 h->left_mb_xy[LBOT] = left_xy[LBOT];
4041 /* For sufficiently low qp, filtering wouldn't do anything.
4042 * This is a conservative estimate: could also check beta_offset
4043 * and more accurate chroma_qp. */
4044 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
4045 int qp = h->cur_pic.qscale_table[mb_xy];
4046 if (qp <= qp_thresh &&
4047 (left_xy[LTOP] < 0 ||
4048 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
4050 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
4051 if (!FRAME_MBAFF(h))
4053 if ((left_xy[LTOP] < 0 ||
4054 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
4055 (top_xy < h->mb_stride ||
4056 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
4061 top_type = h->cur_pic.mb_type[top_xy];
4062 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
4063 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
4064 if (h->deblocking_filter == 2) {
4065 if (h->slice_table[top_xy] != h->slice_num)
4067 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
4068 left_type[LTOP] = left_type[LBOT] = 0;
4070 if (h->slice_table[top_xy] == 0xFFFF)
4072 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
4073 left_type[LTOP] = left_type[LBOT] = 0;
4075 h->top_type = top_type;
4076 h->left_type[LTOP] = left_type[LTOP];
4077 h->left_type[LBOT] = left_type[LBOT];
4079 if (IS_INTRA(mb_type))
4082 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
4083 top_type, left_type, mb_xy, 0);
4084 if (h->list_count == 2)
4085 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
4086 top_type, left_type, mb_xy, 1);
4088 nnz = h->non_zero_count[mb_xy];
4089 nnz_cache = h->non_zero_count_cache;
4090 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
4091 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
4092 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
4093 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
4094 h->cbp = h->cbp_table[mb_xy];
4097 nnz = h->non_zero_count[top_xy];
4098 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
4101 if (left_type[LTOP]) {
4102 nnz = h->non_zero_count[left_xy[LTOP]];
4103 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
4104 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
4105 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
4106 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
4109 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
4110 * from what the loop filter needs */
4111 if (!CABAC(h) && h->pps.transform_8x8_mode) {
4112 if (IS_8x8DCT(top_type)) {
4113 nnz_cache[4 + 8 * 0] =
4114 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
4115 nnz_cache[6 + 8 * 0] =
4116 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
4118 if (IS_8x8DCT(left_type[LTOP])) {
4119 nnz_cache[3 + 8 * 1] =
4120 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
4122 if (IS_8x8DCT(left_type[LBOT])) {
4123 nnz_cache[3 + 8 * 3] =
4124 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
4127 if (IS_8x8DCT(mb_type)) {
4128 nnz_cache[scan8[0]] =
4129 nnz_cache[scan8[1]] =
4130 nnz_cache[scan8[2]] =
4131 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
4133 nnz_cache[scan8[0 + 4]] =
4134 nnz_cache[scan8[1 + 4]] =
4135 nnz_cache[scan8[2 + 4]] =
4136 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
4138 nnz_cache[scan8[0 + 8]] =
4139 nnz_cache[scan8[1 + 8]] =
4140 nnz_cache[scan8[2 + 8]] =
4141 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
4143 nnz_cache[scan8[0 + 12]] =
4144 nnz_cache[scan8[1 + 12]] =
4145 nnz_cache[scan8[2 + 12]] =
4146 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
4153 static void loop_filter(H264Context *h, int start_x, int end_x)
4155 uint8_t *dest_y, *dest_cb, *dest_cr;
4156 int linesize, uvlinesize, mb_x, mb_y;
4157 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
4158 const int old_slice_type = h->slice_type;
4159 const int pixel_shift = h->pixel_shift;
4160 const int block_h = 16 >> h->chroma_y_shift;
4162 if (h->deblocking_filter) {
4163 for (mb_x = start_x; mb_x < end_x; mb_x++)
4164 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
4166 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
4167 h->slice_num = h->slice_table[mb_xy];
4168 mb_type = h->cur_pic.mb_type[mb_xy];
4169 h->list_count = h->list_counts[mb_xy];
4173 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
4177 dest_y = h->cur_pic.f.data[0] +
4178 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
4179 dest_cb = h->cur_pic.f.data[1] +
4180 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
4181 mb_y * h->uvlinesize * block_h;
4182 dest_cr = h->cur_pic.f.data[2] +
4183 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
4184 mb_y * h->uvlinesize * block_h;
4185 // FIXME simplify above
4188 linesize = h->mb_linesize = h->linesize * 2;
4189 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
4190 if (mb_y & 1) { // FIXME move out of this function?
4191 dest_y -= h->linesize * 15;
4192 dest_cb -= h->uvlinesize * (block_h - 1);
4193 dest_cr -= h->uvlinesize * (block_h - 1);
4196 linesize = h->mb_linesize = h->linesize;
4197 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
4199 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
4201 if (fill_filter_caches(h, mb_type))
4203 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
4204 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
4206 if (FRAME_MBAFF(h)) {
4207 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
4208 linesize, uvlinesize);
4210 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
4211 dest_cr, linesize, uvlinesize);
4215 h->slice_type = old_slice_type;
4217 h->mb_y = end_mb_y - FRAME_MBAFF(h);
4218 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
4219 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
4222 static void predict_field_decoding_flag(H264Context *h)
4224 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
4225 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
4226 h->cur_pic.mb_type[mb_xy - 1] :
4227 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
4228 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
4229 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4233 * Draw edges and report progress for the last MB row.
4235 static void decode_finish_row(H264Context *h)
4237 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
4238 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
4239 int height = 16 << FRAME_MBAFF(h);
4240 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
4242 if (h->deblocking_filter) {
4243 if ((top + height) >= pic_height)
4244 height += deblock_border;
4245 top -= deblock_border;
4248 if (top >= pic_height || (top + height) < 0)
4251 height = FFMIN(height, pic_height - top);
4253 height = top + height;
4257 ff_h264_draw_horiz_band(h, top, height);
4259 if (h->droppable || h->er.error_occurred)
4262 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
4263 h->picture_structure == PICT_BOTTOM_FIELD);
4266 static void er_add_slice(H264Context *h, int startx, int starty,
4267 int endx, int endy, int status)
4269 if (CONFIG_ERROR_RESILIENCE) {
4270 ERContext *er = &h->er;
4272 er->ref_count = h->ref_count[0];
4273 ff_er_add_slice(er, startx, starty, endx, endy, status);
4277 static int decode_slice(struct AVCodecContext *avctx, void *arg)
4279 H264Context *h = *(void **)arg;
4280 int lf_x_start = h->mb_x;
4282 h->mb_skip_run = -1;
4284 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
4286 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
4287 avctx->codec_id != AV_CODEC_ID_H264 ||
4288 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
4290 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME) {
4291 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
4293 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
4294 prev_status &= ~ VP_START;
4295 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
4296 h->er.error_occurred = 1;
4302 align_get_bits(&h->gb);
4305 ff_init_cabac_decoder(&h->cabac,
4306 h->gb.buffer + get_bits_count(&h->gb) / 8,
4307 (get_bits_left(&h->gb) + 7) / 8);
4309 ff_h264_init_cabac_states(h);
4313 int ret = ff_h264_decode_mb_cabac(h);
4315 // STOP_TIMER("decode_mb_cabac")
4318 ff_h264_hl_decode_mb(h);
4320 // FIXME optimal? or let mb_decode decode 16x32 ?
4321 if (ret >= 0 && FRAME_MBAFF(h)) {
4324 ret = ff_h264_decode_mb_cabac(h);
4327 ff_h264_hl_decode_mb(h);
4330 eos = get_cabac_terminate(&h->cabac);
4332 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
4333 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
4334 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
4335 h->mb_y, ER_MB_END);
4336 if (h->mb_x >= lf_x_start)
4337 loop_filter(h, lf_x_start, h->mb_x + 1);
4340 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
4341 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
4342 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
4343 av_log(h->avctx, AV_LOG_ERROR,
4344 "error while decoding MB %d %d, bytestream (%td)\n",
4346 h->cabac.bytestream_end - h->cabac.bytestream);
4347 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4348 h->mb_y, ER_MB_ERROR);
4352 if (++h->mb_x >= h->mb_width) {
4353 loop_filter(h, lf_x_start, h->mb_x);
4354 h->mb_x = lf_x_start = 0;
4355 decode_finish_row(h);
4357 if (FIELD_OR_MBAFF_PICTURE(h)) {
4359 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
4360 predict_field_decoding_flag(h);
4364 if (eos || h->mb_y >= h->mb_height) {
4365 tprintf(h->avctx, "slice end %d %d\n",
4366 get_bits_count(&h->gb), h->gb.size_in_bits);
4367 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
4368 h->mb_y, ER_MB_END);
4369 if (h->mb_x > lf_x_start)
4370 loop_filter(h, lf_x_start, h->mb_x);
4376 int ret = ff_h264_decode_mb_cavlc(h);
4379 ff_h264_hl_decode_mb(h);
4381 // FIXME optimal? or let mb_decode decode 16x32 ?
4382 if (ret >= 0 && FRAME_MBAFF(h)) {
4384 ret = ff_h264_decode_mb_cavlc(h);
4387 ff_h264_hl_decode_mb(h);
4392 av_log(h->avctx, AV_LOG_ERROR,
4393 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
4394 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4395 h->mb_y, ER_MB_ERROR);
4399 if (++h->mb_x >= h->mb_width) {
4400 loop_filter(h, lf_x_start, h->mb_x);
4401 h->mb_x = lf_x_start = 0;
4402 decode_finish_row(h);
4404 if (FIELD_OR_MBAFF_PICTURE(h)) {
4406 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
4407 predict_field_decoding_flag(h);
4409 if (h->mb_y >= h->mb_height) {
4410 tprintf(h->avctx, "slice end %d %d\n",
4411 get_bits_count(&h->gb), h->gb.size_in_bits);
4413 if ( get_bits_left(&h->gb) == 0
4414 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
4415 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4416 h->mb_x - 1, h->mb_y,
4421 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4430 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
4431 tprintf(h->avctx, "slice end %d %d\n",
4432 get_bits_count(&h->gb), h->gb.size_in_bits);
4433 if (get_bits_left(&h->gb) == 0) {
4434 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4435 h->mb_x - 1, h->mb_y,
4437 if (h->mb_x > lf_x_start)
4438 loop_filter(h, lf_x_start, h->mb_x);
4442 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4443 h->mb_y, ER_MB_ERROR);
4453 * Call decode_slice() for each context.
4455 * @param h h264 master context
4456 * @param context_count number of contexts to execute
4458 static int execute_decode_slices(H264Context *h, int context_count)
4460 AVCodecContext *const avctx = h->avctx;
4464 if (h->avctx->hwaccel ||
4465 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4467 if (context_count == 1) {
4468 return decode_slice(avctx, &h);
4470 av_assert0(context_count > 0);
4471 for (i = 1; i < context_count; i++) {
4472 hx = h->thread_context[i];
4473 if (CONFIG_ERROR_RESILIENCE) {
4474 hx->er.error_count = 0;
4476 hx->x264_build = h->x264_build;
4479 avctx->execute(avctx, decode_slice, h->thread_context,
4480 NULL, context_count, sizeof(void *));
4482 /* pull back stuff from slices to master context */
4483 hx = h->thread_context[context_count - 1];
4486 h->droppable = hx->droppable;
4487 h->picture_structure = hx->picture_structure;
4488 if (CONFIG_ERROR_RESILIENCE) {
4489 for (i = 1; i < context_count; i++)
4490 h->er.error_count += h->thread_context[i]->er.error_count;
4497 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
4498 int parse_extradata)
4500 AVCodecContext *const avctx = h->avctx;
4501 H264Context *hx; ///< thread context
4505 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
4506 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
4509 int first_slice = 0;
4511 h->nal_unit_type= 0;
4513 if(!h->slice_context_count)
4514 h->slice_context_count= 1;
4515 h->max_contexts = h->slice_context_count;
4516 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS)) {
4517 h->current_slice = 0;
4518 if (!h->first_field)
4519 h->cur_pic_ptr = NULL;
4520 ff_h264_reset_sei(h);
4523 if (h->nal_length_size == 4) {
4524 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
4526 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
4530 for (; pass <= 1; pass++) {
4533 next_avc = h->is_avc ? 0 : buf_size;
4543 if (buf_index >= next_avc) {
4544 if (buf_index >= buf_size - h->nal_length_size)
4547 for (i = 0; i < h->nal_length_size; i++)
4548 nalsize = (nalsize << 8) | buf[buf_index++];
4549 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
4550 av_log(h->avctx, AV_LOG_ERROR,
4551 "AVC: nal size %d\n", nalsize);
4554 next_avc = buf_index + nalsize;
4556 // start code prefix search
4557 for (; buf_index + 3 < next_avc; buf_index++)
4558 // This should always succeed in the first iteration.
4559 if (buf[buf_index] == 0 &&
4560 buf[buf_index + 1] == 0 &&
4561 buf[buf_index + 2] == 1)
4564 if (buf_index + 3 >= buf_size) {
4565 buf_index = buf_size;
4570 if (buf_index >= next_avc)
4574 hx = h->thread_context[context_count];
4576 ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
4577 &consumed, next_avc - buf_index);
4578 if (ptr == NULL || dst_length < 0) {
4582 i = buf_index + consumed;
4583 if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
4584 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
4585 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
4586 h->workaround_bugs |= FF_BUG_TRUNCATED;
4588 if (!(h->workaround_bugs & FF_BUG_TRUNCATED))
4589 while(dst_length > 0 && ptr[dst_length - 1] == 0)
4591 bit_length = !dst_length ? 0
4593 decode_rbsp_trailing(h, ptr + dst_length - 1));
4595 if (h->avctx->debug & FF_DEBUG_STARTCODE)
4596 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);
4598 if (h->is_avc && (nalsize != consumed) && nalsize)
4599 av_log(h->avctx, AV_LOG_DEBUG,
4600 "AVC: Consumed only %d bytes instead of %d\n",
4603 buf_index += consumed;
4607 /* packets can sometimes contain multiple PPS/SPS,
4608 * e.g. two PAFF field pictures in one packet, or a demuxer
4609 * which splits NALs strangely if so, when frame threading we
4610 * can't start the next thread until we've read all of them */
4611 switch (hx->nal_unit_type) {
4614 nals_needed = nal_index;
4619 init_get_bits(&hx->gb, ptr, bit_length);
4620 if (!get_ue_golomb(&hx->gb) || !first_slice)
4621 nals_needed = nal_index;
4623 first_slice = hx->nal_unit_type;
4629 switch (hx->nal_unit_type) {
4633 first_slice = hx->nal_unit_type;
4636 // FIXME do not discard SEI id
4637 if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
4641 /* Ignore per frame NAL unit type during extradata
4642 * parsing. Decoding slices is not possible in codec init
4644 if (parse_extradata) {
4645 switch (hx->nal_unit_type) {
4651 case NAL_AUXILIARY_SLICE:
4652 av_log(h->avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header/extradata\n", hx->nal_unit_type);
4653 hx->nal_unit_type = NAL_FF_IGNORE;
4659 switch (hx->nal_unit_type) {
4661 if (first_slice != NAL_IDR_SLICE) {
4662 av_log(h->avctx, AV_LOG_ERROR,
4663 "Invalid mix of idr and non-idr slices\n");
4668 idr(h); // FIXME ensure we don't lose some frames if there is reordering
4671 init_get_bits(&hx->gb, ptr, bit_length);
4673 hx->inter_gb_ptr = &hx->gb;
4674 hx->data_partitioning = 0;
4676 if ((err = decode_slice_header(hx, h)))
4679 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
4680 h->valid_recovery_point = 1;
4682 if ( h->sei_recovery_frame_cnt >= 0
4683 && ( h->recovery_frame<0
4684 || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
4685 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
4686 (1 << h->sps.log2_max_frame_num);
4688 if (!h->valid_recovery_point)
4689 h->recovery_frame = h->frame_num;
4692 h->cur_pic_ptr->f.key_frame |=
4693 (hx->nal_unit_type == NAL_IDR_SLICE);
4695 if (h->recovery_frame == h->frame_num) {
4696 h->cur_pic_ptr->sync |= 1;
4697 h->recovery_frame = -1;
4700 h->sync |= !!h->cur_pic_ptr->f.key_frame;
4701 h->sync |= 3*!!(avctx->flags2 & CODEC_FLAG2_SHOW_ALL);
4702 h->cur_pic_ptr->sync |= h->sync;
4704 if (h->current_slice == 1) {
4705 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS))
4706 decode_postinit(h, nal_index >= nals_needed);
4708 if (h->avctx->hwaccel &&
4709 h->avctx->hwaccel->start_frame(h->avctx, NULL, 0) < 0)
4711 if (CONFIG_H264_VDPAU_DECODER &&
4712 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4713 ff_vdpau_h264_picture_start(h);
4716 if (hx->redundant_pic_count == 0 &&
4717 (avctx->skip_frame < AVDISCARD_NONREF ||
4719 (avctx->skip_frame < AVDISCARD_BIDIR ||
4720 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4721 (avctx->skip_frame < AVDISCARD_NONKEY ||
4722 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4723 avctx->skip_frame < AVDISCARD_ALL) {
4724 if (avctx->hwaccel) {
4725 if (avctx->hwaccel->decode_slice(avctx,
4726 &buf[buf_index - consumed],
4729 } else if (CONFIG_H264_VDPAU_DECODER &&
4730 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
4731 static const uint8_t start_code[] = {
4733 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0], start_code,
4734 sizeof(start_code));
4735 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0], &buf[buf_index - consumed],
4742 init_get_bits(&hx->gb, ptr, bit_length);
4744 hx->inter_gb_ptr = NULL;
4746 if ((err = decode_slice_header(hx, h)) < 0)
4749 hx->data_partitioning = 1;
4752 init_get_bits(&hx->intra_gb, ptr, bit_length);
4753 hx->intra_gb_ptr = &hx->intra_gb;
4756 init_get_bits(&hx->inter_gb, ptr, bit_length);
4757 hx->inter_gb_ptr = &hx->inter_gb;
4759 av_log(h->avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
4762 if (hx->redundant_pic_count == 0 &&
4764 hx->data_partitioning &&
4765 h->cur_pic_ptr && h->context_initialized &&
4766 (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4767 (avctx->skip_frame < AVDISCARD_BIDIR ||
4768 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4769 (avctx->skip_frame < AVDISCARD_NONKEY ||
4770 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4771 avctx->skip_frame < AVDISCARD_ALL)
4775 init_get_bits(&h->gb, ptr, bit_length);
4776 ff_h264_decode_sei(h);
4779 init_get_bits(&h->gb, ptr, bit_length);
4780 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) {
4781 av_log(h->avctx, AV_LOG_DEBUG,
4782 "SPS decoding failure, trying again with the complete NAL\n");
4784 av_assert0(next_avc - buf_index + consumed == nalsize);
4785 if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4787 init_get_bits(&h->gb, &buf[buf_index + 1 - consumed],
4788 8*(next_avc - buf_index + consumed - 1));
4789 ff_h264_decode_seq_parameter_set(h);
4794 init_get_bits(&h->gb, ptr, bit_length);
4795 ff_h264_decode_picture_parameter_set(h, bit_length);
4798 case NAL_END_SEQUENCE:
4799 case NAL_END_STREAM:
4800 case NAL_FILLER_DATA:
4802 case NAL_AUXILIARY_SLICE:
4807 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4808 hx->nal_unit_type, bit_length);
4811 if (context_count == h->max_contexts) {
4812 execute_decode_slices(h, context_count);
4817 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4818 else if (err == 1) {
4819 /* Slice could not be decoded in parallel mode, copy down
4820 * NAL unit stuff to context 0 and restart. Note that
4821 * rbsp_buffer is not transferred, but since we no longer
4822 * run in parallel mode this should not be an issue. */
4823 h->nal_unit_type = hx->nal_unit_type;
4824 h->nal_ref_idc = hx->nal_ref_idc;
4831 execute_decode_slices(h, context_count);
4835 if (h->cur_pic_ptr && !h->droppable) {
4836 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
4837 h->picture_structure == PICT_BOTTOM_FIELD);
4844 * Return the number of bytes consumed for building the current frame.
4846 static int get_consumed_bytes(int pos, int buf_size)
4849 pos = 1; // avoid infinite loops (i doubt that is needed but ...)
4850 if (pos + 10 > buf_size)
4851 pos = buf_size; // oops ;)
4856 static int output_frame(H264Context *h, AVFrame *dst, Picture *srcp)
4858 AVFrame *src = &srcp->f;
4860 int ret = av_frame_ref(dst, src);
4867 for (i = 0; i < 3; i++) {
4868 int hshift = (i > 0) ? h->chroma_x_shift : 0;
4869 int vshift = (i > 0) ? h->chroma_y_shift : 0;
4870 int off = ((srcp->crop_left >> hshift) << h->pixel_shift) +
4871 (srcp->crop_top >> vshift) * dst->linesize[i];
4872 dst->data[i] += off;
4877 static int decode_frame(AVCodecContext *avctx, void *data,
4878 int *got_frame, AVPacket *avpkt)
4880 const uint8_t *buf = avpkt->data;
4881 int buf_size = avpkt->size;
4882 H264Context *h = avctx->priv_data;
4883 AVFrame *pict = data;
4889 h->flags = avctx->flags;
4891 /* end of stream, output what is still in the buffers */
4892 if (buf_size == 0) {
4895 h->cur_pic_ptr = NULL;
4898 // FIXME factorize this with the output code below
4899 out = h->delayed_pic[0];
4902 h->delayed_pic[i] &&
4903 !h->delayed_pic[i]->f.key_frame &&
4904 !h->delayed_pic[i]->mmco_reset;
4906 if (h->delayed_pic[i]->poc < out->poc) {
4907 out = h->delayed_pic[i];
4911 for (i = out_idx; h->delayed_pic[i]; i++)
4912 h->delayed_pic[i] = h->delayed_pic[i + 1];
4915 out->reference &= ~DELAYED_PIC_REF;
4916 ret = output_frame(h, pict, out);
4924 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4925 int cnt= buf[5]&0x1f;
4926 const uint8_t *p= buf+6;
4928 int nalsize= AV_RB16(p) + 2;
4929 if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4937 int nalsize= AV_RB16(p) + 2;
4938 if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4943 return ff_h264_decode_extradata(h, buf, buf_size);
4947 buf_index = decode_nal_units(h, buf, buf_size, 0);
4951 if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4952 av_assert0(buf_index <= buf_size);
4956 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
4957 if (avctx->skip_frame >= AVDISCARD_NONREF ||
4958 buf_size >= 4 && !memcmp("Q264", buf, 4))
4960 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4964 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) ||
4965 (h->mb_y >= h->mb_height && h->mb_height)) {
4966 if (avctx->flags2 & CODEC_FLAG2_CHUNKS)
4967 decode_postinit(h, 1);
4971 /* Wait for second field. */
4973 if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4974 ret = output_frame(h, pict, h->next_output_pic);
4978 if (CONFIG_MPEGVIDEO) {
4979 ff_print_debug_info2(h->avctx, h->next_output_pic, pict, h->er.mbskip_table,
4981 h->mb_width, h->mb_height, h->mb_stride, 1);
4986 assert(pict->data[0] || !*got_frame);
4988 return get_consumed_bytes(buf_index, buf_size);
4991 av_cold void ff_h264_free_context(H264Context *h)
4995 free_tables(h, 1); // FIXME cleanup init stuff perhaps
4997 for (i = 0; i < MAX_SPS_COUNT; i++)
4998 av_freep(h->sps_buffers + i);
5000 for (i = 0; i < MAX_PPS_COUNT; i++)
5001 av_freep(h->pps_buffers + i);
5004 static av_cold int h264_decode_end(AVCodecContext *avctx)
5006 H264Context *h = avctx->priv_data;
5008 ff_h264_remove_all_refs(h);
5009 ff_h264_free_context(h);
5011 unref_picture(h, &h->cur_pic);
5016 static const AVProfile profiles[] = {
5017 { FF_PROFILE_H264_BASELINE, "Baseline" },
5018 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
5019 { FF_PROFILE_H264_MAIN, "Main" },
5020 { FF_PROFILE_H264_EXTENDED, "Extended" },
5021 { FF_PROFILE_H264_HIGH, "High" },
5022 { FF_PROFILE_H264_HIGH_10, "High 10" },
5023 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
5024 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
5025 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
5026 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
5027 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
5028 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
5029 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
5030 { FF_PROFILE_UNKNOWN },
5033 static const AVOption h264_options[] = {
5034 {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
5035 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
5039 static const AVClass h264_class = {
5040 .class_name = "H264 Decoder",
5041 .item_name = av_default_item_name,
5042 .option = h264_options,
5043 .version = LIBAVUTIL_VERSION_INT,
5046 static const AVClass h264_vdpau_class = {
5047 .class_name = "H264 VDPAU Decoder",
5048 .item_name = av_default_item_name,
5049 .option = h264_options,
5050 .version = LIBAVUTIL_VERSION_INT,
5053 AVCodec ff_h264_decoder = {
5055 .type = AVMEDIA_TYPE_VIDEO,
5056 .id = AV_CODEC_ID_H264,
5057 .priv_data_size = sizeof(H264Context),
5058 .init = ff_h264_decode_init,
5059 .close = h264_decode_end,
5060 .decode = decode_frame,
5061 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
5062 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
5063 CODEC_CAP_FRAME_THREADS,
5065 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
5066 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
5067 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
5068 .profiles = NULL_IF_CONFIG_SMALL(profiles),
5069 .priv_class = &h264_class,
5072 #if CONFIG_H264_VDPAU_DECODER
5073 AVCodec ff_h264_vdpau_decoder = {
5074 .name = "h264_vdpau",
5075 .type = AVMEDIA_TYPE_VIDEO,
5076 .id = AV_CODEC_ID_H264,
5077 .priv_data_size = sizeof(H264Context),
5078 .init = ff_h264_decode_init,
5079 .close = h264_decode_end,
5080 .decode = decode_frame,
5081 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
5083 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
5084 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
5086 .profiles = NULL_IF_CONFIG_SMALL(profiles),
5087 .priv_class = &h264_vdpau_class,