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/avassert.h"
31 #include "libavutil/imgutils.h"
32 #include "libavutil/opt.h"
35 #include "cabac_functions.h"
37 #include "error_resilience.h"
39 #include "mpegvideo.h"
42 #include "h264chroma.h"
43 #include "h264_mvpred.h"
46 #include "rectangle.h"
49 #include "vdpau_internal.h"
53 static void flush_change(H264Context *h);
55 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
57 static const uint8_t rem6[QP_MAX_NUM + 1] = {
58 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
59 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
60 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
61 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
65 static const uint8_t div6[QP_MAX_NUM + 1] = {
66 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
67 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
68 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
69 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
73 static const uint8_t field_scan[16+1] = {
74 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
75 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
76 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
77 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
80 static const uint8_t field_scan8x8[64+1] = {
81 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
82 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
83 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
84 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
85 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
86 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
87 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
88 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
89 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
90 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
91 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
92 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
93 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
94 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
95 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
96 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
99 static const uint8_t field_scan8x8_cavlc[64+1] = {
100 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
101 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
102 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
103 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
104 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
105 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
106 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
107 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
108 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
109 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
110 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
111 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
112 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
113 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
114 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
115 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
118 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
119 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
120 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
121 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
122 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
123 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
124 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
125 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
126 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
127 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
128 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
129 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
130 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
131 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
132 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
133 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
134 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
135 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
138 static const uint8_t dequant4_coeff_init[6][3] = {
147 static const uint8_t dequant8_coeff_init_scan[16] = {
148 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
151 static const uint8_t dequant8_coeff_init[6][6] = {
152 { 20, 18, 32, 19, 25, 24 },
153 { 22, 19, 35, 21, 28, 26 },
154 { 26, 23, 42, 24, 33, 31 },
155 { 28, 25, 45, 26, 35, 33 },
156 { 32, 28, 51, 30, 40, 38 },
157 { 36, 32, 58, 34, 46, 43 },
160 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
161 #if CONFIG_H264_DXVA2_HWACCEL
162 AV_PIX_FMT_DXVA2_VLD,
164 #if CONFIG_H264_VAAPI_HWACCEL
165 AV_PIX_FMT_VAAPI_VLD,
167 #if CONFIG_H264_VDA_HWACCEL
170 #if CONFIG_H264_VDPAU_HWACCEL
177 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
178 #if CONFIG_H264_DXVA2_HWACCEL
179 AV_PIX_FMT_DXVA2_VLD,
181 #if CONFIG_H264_VAAPI_HWACCEL
182 AV_PIX_FMT_VAAPI_VLD,
184 #if CONFIG_H264_VDA_HWACCEL
187 #if CONFIG_H264_VDPAU_HWACCEL
194 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
196 H264Context *h = avctx->priv_data;
197 return h ? h->sps.num_reorder_frames : 0;
200 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
202 int mb_x, int mb_y, int mb_intra, int mb_skipped)
204 H264Context *h = opaque;
208 h->mb_xy = mb_x + mb_y * h->mb_stride;
209 memset(h->non_zero_count_cache, 0, sizeof(h->non_zero_count_cache));
210 av_assert1(ref >= 0);
211 /* FIXME: It is possible albeit uncommon that slice references
212 * differ between slices. We take the easy approach and ignore
213 * it for now. If this turns out to have any relevance in
214 * practice then correct remapping should be added. */
215 if (ref >= h->ref_count[0])
217 if (!h->ref_list[0][ref].f.data[0]) {
218 av_log(h->avctx, AV_LOG_DEBUG, "Reference not available for error concealing\n");
221 if ((h->ref_list[0][ref].reference&3) != 3) {
222 av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n");
225 fill_rectangle(&h->cur_pic.ref_index[0][4 * h->mb_xy],
227 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
228 fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8,
229 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
231 h->mb_field_decoding_flag = 0;
232 ff_h264_hl_decode_mb(h);
235 void ff_h264_draw_horiz_band(H264Context *h, int y, int height)
237 AVCodecContext *avctx = h->avctx;
238 Picture *cur = &h->cur_pic;
239 Picture *last = h->ref_list[0][0].f.data[0] ? &h->ref_list[0][0] : NULL;
240 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
241 int vshift = desc->log2_chroma_h;
242 const int field_pic = h->picture_structure != PICT_FRAME;
248 height = FFMIN(height, avctx->height - y);
250 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
253 if (avctx->draw_horiz_band) {
255 int offset[AV_NUM_DATA_POINTERS];
258 if (cur->f.pict_type == AV_PICTURE_TYPE_B || h->low_delay ||
259 (avctx->slice_flags & SLICE_FLAG_CODED_ORDER))
266 offset[0] = y * src->linesize[0];
268 offset[2] = (y >> vshift) * src->linesize[1];
269 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
274 avctx->draw_horiz_band(avctx, src, offset,
275 y, h->picture_structure, height);
279 static void unref_picture(H264Context *h, Picture *pic)
281 int off = offsetof(Picture, tf) + sizeof(pic->tf);
287 ff_thread_release_buffer(h->avctx, &pic->tf);
288 av_buffer_unref(&pic->hwaccel_priv_buf);
290 av_buffer_unref(&pic->qscale_table_buf);
291 av_buffer_unref(&pic->mb_type_buf);
292 for (i = 0; i < 2; i++) {
293 av_buffer_unref(&pic->motion_val_buf[i]);
294 av_buffer_unref(&pic->ref_index_buf[i]);
297 memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
300 static void release_unused_pictures(H264Context *h, int remove_current)
304 /* release non reference frames */
305 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
306 if (h->DPB[i].f.data[0] && !h->DPB[i].reference &&
307 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
308 unref_picture(h, &h->DPB[i]);
313 static int ref_picture(H264Context *h, Picture *dst, Picture *src)
317 av_assert0(!dst->f.buf[0]);
318 av_assert0(src->f.buf[0]);
322 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
326 dst->qscale_table_buf = av_buffer_ref(src->qscale_table_buf);
327 dst->mb_type_buf = av_buffer_ref(src->mb_type_buf);
328 if (!dst->qscale_table_buf || !dst->mb_type_buf)
330 dst->qscale_table = src->qscale_table;
331 dst->mb_type = src->mb_type;
333 for (i = 0; i < 2; i++) {
334 dst->motion_val_buf[i] = av_buffer_ref(src->motion_val_buf[i]);
335 dst->ref_index_buf[i] = av_buffer_ref(src->ref_index_buf[i]);
336 if (!dst->motion_val_buf[i] || !dst->ref_index_buf[i])
338 dst->motion_val[i] = src->motion_val[i];
339 dst->ref_index[i] = src->ref_index[i];
342 if (src->hwaccel_picture_private) {
343 dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
344 if (!dst->hwaccel_priv_buf)
346 dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
349 for (i = 0; i < 2; i++)
350 dst->field_poc[i] = src->field_poc[i];
352 memcpy(dst->ref_poc, src->ref_poc, sizeof(src->ref_poc));
353 memcpy(dst->ref_count, src->ref_count, sizeof(src->ref_count));
356 dst->frame_num = src->frame_num;
357 dst->mmco_reset = src->mmco_reset;
358 dst->pic_id = src->pic_id;
359 dst->long_ref = src->long_ref;
360 dst->mbaff = src->mbaff;
361 dst->field_picture = src->field_picture;
362 dst->needs_realloc = src->needs_realloc;
363 dst->reference = src->reference;
364 dst->sync = src->sync;
365 dst->crop = src->crop;
366 dst->crop_left = src->crop_left;
367 dst->crop_top = src->crop_top;
371 unref_picture(h, dst);
375 static int alloc_scratch_buffers(H264Context *h, int linesize)
377 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
379 if (h->bipred_scratchpad)
382 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
383 // edge emu needs blocksize + filter length - 1
384 // (= 21x21 for h264)
385 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
386 h->me.scratchpad = av_mallocz(alloc_size * 2 * 16 * 2);
388 if (!h->bipred_scratchpad || !h->edge_emu_buffer || !h->me.scratchpad) {
389 av_freep(&h->bipred_scratchpad);
390 av_freep(&h->edge_emu_buffer);
391 av_freep(&h->me.scratchpad);
392 return AVERROR(ENOMEM);
395 h->me.temp = h->me.scratchpad;
400 static int init_table_pools(H264Context *h)
402 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
403 const int mb_array_size = h->mb_stride * h->mb_height;
404 const int b4_stride = h->mb_width * 4 + 1;
405 const int b4_array_size = b4_stride * h->mb_height * 4;
407 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
409 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
410 sizeof(uint32_t), av_buffer_allocz);
411 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
412 sizeof(int16_t), av_buffer_allocz);
413 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
415 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
416 !h->ref_index_pool) {
417 av_buffer_pool_uninit(&h->qscale_table_pool);
418 av_buffer_pool_uninit(&h->mb_type_pool);
419 av_buffer_pool_uninit(&h->motion_val_pool);
420 av_buffer_pool_uninit(&h->ref_index_pool);
421 return AVERROR(ENOMEM);
427 static int alloc_picture(H264Context *h, Picture *pic)
431 av_assert0(!pic->f.data[0]);
434 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
435 AV_GET_BUFFER_FLAG_REF : 0);
439 h->linesize = pic->f.linesize[0];
440 h->uvlinesize = pic->f.linesize[1];
441 pic->crop = h->sps.crop;
442 pic->crop_top = h->sps.crop_top;
443 pic->crop_left= h->sps.crop_left;
445 if (h->avctx->hwaccel) {
446 const AVHWAccel *hwaccel = h->avctx->hwaccel;
447 av_assert0(!pic->hwaccel_picture_private);
448 if (hwaccel->priv_data_size) {
449 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size);
450 if (!pic->hwaccel_priv_buf)
451 return AVERROR(ENOMEM);
452 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
455 if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
456 int h_chroma_shift, v_chroma_shift;
457 av_pix_fmt_get_chroma_sub_sample(pic->f.format,
458 &h_chroma_shift, &v_chroma_shift);
460 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
461 memset(pic->f.data[1] + pic->f.linesize[1]*i,
462 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
463 memset(pic->f.data[2] + pic->f.linesize[2]*i,
464 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
468 if (!h->qscale_table_pool) {
469 ret = init_table_pools(h);
474 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
475 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
476 if (!pic->qscale_table_buf || !pic->mb_type_buf)
479 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
480 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
482 for (i = 0; i < 2; i++) {
483 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
484 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
485 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
488 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
489 pic->ref_index[i] = pic->ref_index_buf[i]->data;
494 unref_picture(h, pic);
495 return (ret < 0) ? ret : AVERROR(ENOMEM);
498 static inline int pic_is_unused(H264Context *h, Picture *pic)
500 if (pic->f.data[0] == NULL)
502 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
507 static int find_unused_picture(H264Context *h)
511 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
512 if (pic_is_unused(h, &h->DPB[i]))
515 if (i == MAX_PICTURE_COUNT)
516 return AVERROR_INVALIDDATA;
518 if (h->DPB[i].needs_realloc) {
519 h->DPB[i].needs_realloc = 0;
520 unref_picture(h, &h->DPB[i]);
527 * Check if the top & left blocks are available if needed and
528 * change the dc mode so it only uses the available blocks.
530 int ff_h264_check_intra4x4_pred_mode(H264Context *h)
532 static const int8_t top[12] = {
533 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
535 static const int8_t left[12] = {
536 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
540 if (!(h->top_samples_available & 0x8000)) {
541 for (i = 0; i < 4; i++) {
542 int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
544 av_log(h->avctx, AV_LOG_ERROR,
545 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
546 status, h->mb_x, h->mb_y);
547 return AVERROR_INVALIDDATA;
549 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
554 if ((h->left_samples_available & 0x8888) != 0x8888) {
555 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
556 for (i = 0; i < 4; i++)
557 if (!(h->left_samples_available & mask[i])) {
558 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
560 av_log(h->avctx, AV_LOG_ERROR,
561 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
562 status, h->mb_x, h->mb_y);
563 return AVERROR_INVALIDDATA;
565 h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
571 } // FIXME cleanup like ff_h264_check_intra_pred_mode
574 * Check if the top & left blocks are available if needed and
575 * change the dc mode so it only uses the available blocks.
577 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
579 static const int8_t top[4] = { LEFT_DC_PRED8x8, 1, -1, -1 };
580 static const int8_t left[5] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
583 av_log(h->avctx, AV_LOG_ERROR,
584 "out of range intra chroma pred mode at %d %d\n",
586 return AVERROR_INVALIDDATA;
589 if (!(h->top_samples_available & 0x8000)) {
592 av_log(h->avctx, AV_LOG_ERROR,
593 "top block unavailable for requested intra mode at %d %d\n",
595 return AVERROR_INVALIDDATA;
599 if ((h->left_samples_available & 0x8080) != 0x8080) {
602 av_log(h->avctx, AV_LOG_ERROR,
603 "left block unavailable for requested intra mode at %d %d\n",
605 return AVERROR_INVALIDDATA;
607 if (is_chroma && (h->left_samples_available & 0x8080)) {
608 // mad cow disease mode, aka MBAFF + constrained_intra_pred
609 mode = ALZHEIMER_DC_L0T_PRED8x8 +
610 (!(h->left_samples_available & 0x8000)) +
611 2 * (mode == DC_128_PRED8x8);
618 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
619 int *dst_length, int *consumed, int length)
625 // src[0]&0x80; // forbidden bit
626 h->nal_ref_idc = src[0] >> 5;
627 h->nal_unit_type = src[0] & 0x1F;
632 #define STARTCODE_TEST \
633 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
634 if (src[i + 2] != 3 && src[i + 2] != 0) { \
635 /* startcode, so we must be past the end */ \
641 #if HAVE_FAST_UNALIGNED
642 #define FIND_FIRST_ZERO \
643 if (i > 0 && !src[i]) \
649 for (i = 0; i + 1 < length; i += 9) {
650 if (!((~AV_RN64A(src + i) &
651 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
652 0x8000800080008080ULL))
659 for (i = 0; i + 1 < length; i += 5) {
660 if (!((~AV_RN32A(src + i) &
661 (AV_RN32A(src + i) - 0x01000101U)) &
670 for (i = 0; i + 1 < length; i += 2) {
673 if (i > 0 && src[i - 1] == 0)
679 // use second escape buffer for inter data
680 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
682 si = h->rbsp_buffer_size[bufidx];
683 av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
684 dst = h->rbsp_buffer[bufidx];
689 if(i>=length-1){ //no escaped 0
691 *consumed= length+1; //+1 for the header
692 if(h->avctx->flags2 & CODEC_FLAG2_FAST){
695 memcpy(dst, src, length);
702 while (si + 2 < length) {
703 // remove escapes (very rare 1:2^22)
704 if (src[si + 2] > 3) {
705 dst[di++] = src[si++];
706 dst[di++] = src[si++];
707 } else if (src[si] == 0 && src[si + 1] == 0 && src[si + 2] != 0) {
708 if (src[si + 2] == 3) { // escape
713 } else // next start code
717 dst[di++] = src[si++];
720 dst[di++] = src[si++];
723 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
726 *consumed = si + 1; // +1 for the header
727 /* FIXME store exact number of bits in the getbitcontext
728 * (it is needed for decoding) */
733 * Identify the exact end of the bitstream
734 * @return the length of the trailing, or 0 if damaged
736 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
741 tprintf(h->avctx, "rbsp trailing %X\n", v);
743 for (r = 1; r < 9; r++) {
751 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
752 int height, int y_offset, int list)
754 int raw_my = h->mv_cache[list][scan8[n]][1];
755 int filter_height_down = (raw_my & 3) ? 3 : 0;
756 int full_my = (raw_my >> 2) + y_offset;
757 int bottom = full_my + filter_height_down + height;
759 av_assert2(height >= 0);
761 return FFMAX(0, bottom);
764 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
765 int height, int y_offset, int list0,
766 int list1, int *nrefs)
770 y_offset += 16 * (h->mb_y >> MB_FIELD(h));
773 int ref_n = h->ref_cache[0][scan8[n]];
774 Picture *ref = &h->ref_list[0][ref_n];
776 // Error resilience puts the current picture in the ref list.
777 // Don't try to wait on these as it will cause a deadlock.
778 // Fields can wait on each other, though.
779 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
780 (ref->reference & 3) != h->picture_structure) {
781 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
782 if (refs[0][ref_n] < 0)
784 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
789 int ref_n = h->ref_cache[1][scan8[n]];
790 Picture *ref = &h->ref_list[1][ref_n];
792 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
793 (ref->reference & 3) != h->picture_structure) {
794 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
795 if (refs[1][ref_n] < 0)
797 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
803 * Wait until all reference frames are available for MC operations.
805 * @param h the H264 context
807 static void await_references(H264Context *h)
809 const int mb_xy = h->mb_xy;
810 const int mb_type = h->cur_pic.mb_type[mb_xy];
812 int nrefs[2] = { 0 };
815 memset(refs, -1, sizeof(refs));
817 if (IS_16X16(mb_type)) {
818 get_lowest_part_y(h, refs, 0, 16, 0,
819 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
820 } else if (IS_16X8(mb_type)) {
821 get_lowest_part_y(h, refs, 0, 8, 0,
822 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
823 get_lowest_part_y(h, refs, 8, 8, 8,
824 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
825 } else if (IS_8X16(mb_type)) {
826 get_lowest_part_y(h, refs, 0, 16, 0,
827 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
828 get_lowest_part_y(h, refs, 4, 16, 0,
829 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
833 av_assert2(IS_8X8(mb_type));
835 for (i = 0; i < 4; i++) {
836 const int sub_mb_type = h->sub_mb_type[i];
838 int y_offset = (i & 2) << 2;
840 if (IS_SUB_8X8(sub_mb_type)) {
841 get_lowest_part_y(h, refs, n, 8, y_offset,
842 IS_DIR(sub_mb_type, 0, 0),
843 IS_DIR(sub_mb_type, 0, 1),
845 } else if (IS_SUB_8X4(sub_mb_type)) {
846 get_lowest_part_y(h, refs, n, 4, y_offset,
847 IS_DIR(sub_mb_type, 0, 0),
848 IS_DIR(sub_mb_type, 0, 1),
850 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
851 IS_DIR(sub_mb_type, 0, 0),
852 IS_DIR(sub_mb_type, 0, 1),
854 } else if (IS_SUB_4X8(sub_mb_type)) {
855 get_lowest_part_y(h, refs, n, 8, y_offset,
856 IS_DIR(sub_mb_type, 0, 0),
857 IS_DIR(sub_mb_type, 0, 1),
859 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
860 IS_DIR(sub_mb_type, 0, 0),
861 IS_DIR(sub_mb_type, 0, 1),
865 av_assert2(IS_SUB_4X4(sub_mb_type));
866 for (j = 0; j < 4; j++) {
867 int sub_y_offset = y_offset + 2 * (j & 2);
868 get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
869 IS_DIR(sub_mb_type, 0, 0),
870 IS_DIR(sub_mb_type, 0, 1),
877 for (list = h->list_count - 1; list >= 0; list--)
878 for (ref = 0; ref < 48 && nrefs[list]; ref++) {
879 int row = refs[list][ref];
881 Picture *ref_pic = &h->ref_list[list][ref];
882 int ref_field = ref_pic->reference - 1;
883 int ref_field_picture = ref_pic->field_picture;
884 int pic_height = 16 * h->mb_height >> ref_field_picture;
889 if (!FIELD_PICTURE(h) && ref_field_picture) { // frame referencing two fields
890 ff_thread_await_progress(&ref_pic->tf,
891 FFMIN((row >> 1) - !(row & 1),
894 ff_thread_await_progress(&ref_pic->tf,
895 FFMIN((row >> 1), pic_height - 1),
897 } else if (FIELD_PICTURE(h) && !ref_field_picture) { // field referencing one field of a frame
898 ff_thread_await_progress(&ref_pic->tf,
899 FFMIN(row * 2 + ref_field,
902 } else if (FIELD_PICTURE(h)) {
903 ff_thread_await_progress(&ref_pic->tf,
904 FFMIN(row, pic_height - 1),
907 ff_thread_await_progress(&ref_pic->tf,
908 FFMIN(row, pic_height - 1),
915 static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
916 int n, int square, int height,
918 uint8_t *dest_y, uint8_t *dest_cb,
920 int src_x_offset, int src_y_offset,
921 qpel_mc_func *qpix_op,
922 h264_chroma_mc_func chroma_op,
923 int pixel_shift, int chroma_idc)
925 const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
926 int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
927 const int luma_xy = (mx & 3) + ((my & 3) << 2);
928 ptrdiff_t offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
929 uint8_t *src_y = pic->f.data[0] + offset;
930 uint8_t *src_cb, *src_cr;
932 int extra_height = 0;
934 const int full_mx = mx >> 2;
935 const int full_my = my >> 2;
936 const int pic_width = 16 * h->mb_width;
937 const int pic_height = 16 * h->mb_height >> MB_FIELD(h);
945 if (full_mx < 0 - extra_width ||
946 full_my < 0 - extra_height ||
947 full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
948 full_my + 16 /*FIXME*/ > pic_height + extra_height) {
949 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_linesize,
950 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
952 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
953 full_my - 2, pic_width, pic_height);
954 src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
958 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
960 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
962 if (CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY)
965 if (chroma_idc == 3 /* yuv444 */) {
966 src_cb = pic->f.data[1] + offset;
968 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_linesize,
969 src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
971 16 + 5, 16 + 5 /*FIXME*/,
972 full_mx - 2, full_my - 2,
973 pic_width, pic_height);
974 src_cb = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
976 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
978 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
980 src_cr = pic->f.data[2] + offset;
982 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_linesize,
983 src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
985 16 + 5, 16 + 5 /*FIXME*/,
986 full_mx - 2, full_my - 2,
987 pic_width, pic_height);
988 src_cr = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
990 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
992 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
996 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
997 if (chroma_idc == 1 /* yuv420 */ && MB_FIELD(h)) {
998 // chroma offset when predicting from a field of opposite parity
999 my += 2 * ((h->mb_y & 1) - (pic->reference - 1));
1000 emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
1003 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
1004 (my >> ysh) * h->mb_uvlinesize;
1005 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
1006 (my >> ysh) * h->mb_uvlinesize;
1009 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_uvlinesize, src_cb, h->mb_uvlinesize,
1010 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
1011 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
1012 src_cb = h->edge_emu_buffer;
1014 chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
1015 height >> (chroma_idc == 1 /* yuv420 */),
1016 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
1019 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_uvlinesize, src_cr, h->mb_uvlinesize,
1020 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
1021 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
1022 src_cr = h->edge_emu_buffer;
1024 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
1025 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
1028 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
1029 int height, int delta,
1030 uint8_t *dest_y, uint8_t *dest_cb,
1032 int x_offset, int y_offset,
1033 qpel_mc_func *qpix_put,
1034 h264_chroma_mc_func chroma_put,
1035 qpel_mc_func *qpix_avg,
1036 h264_chroma_mc_func chroma_avg,
1037 int list0, int list1,
1038 int pixel_shift, int chroma_idc)
1040 qpel_mc_func *qpix_op = qpix_put;
1041 h264_chroma_mc_func chroma_op = chroma_put;
1043 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1044 if (chroma_idc == 3 /* yuv444 */) {
1045 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1046 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1047 } else if (chroma_idc == 2 /* yuv422 */) {
1048 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1049 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1050 } else { /* yuv420 */
1051 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1052 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1054 x_offset += 8 * h->mb_x;
1055 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
1058 Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
1059 mc_dir_part(h, ref, n, square, height, delta, 0,
1060 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1061 qpix_op, chroma_op, pixel_shift, chroma_idc);
1064 chroma_op = chroma_avg;
1068 Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
1069 mc_dir_part(h, ref, n, square, height, delta, 1,
1070 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1071 qpix_op, chroma_op, pixel_shift, chroma_idc);
1075 static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
1076 int height, int delta,
1077 uint8_t *dest_y, uint8_t *dest_cb,
1079 int x_offset, int y_offset,
1080 qpel_mc_func *qpix_put,
1081 h264_chroma_mc_func chroma_put,
1082 h264_weight_func luma_weight_op,
1083 h264_weight_func chroma_weight_op,
1084 h264_biweight_func luma_weight_avg,
1085 h264_biweight_func chroma_weight_avg,
1086 int list0, int list1,
1087 int pixel_shift, int chroma_idc)
1091 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1092 if (chroma_idc == 3 /* yuv444 */) {
1093 chroma_height = height;
1094 chroma_weight_avg = luma_weight_avg;
1095 chroma_weight_op = luma_weight_op;
1096 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1097 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1098 } else if (chroma_idc == 2 /* yuv422 */) {
1099 chroma_height = height;
1100 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1101 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1102 } else { /* yuv420 */
1103 chroma_height = height >> 1;
1104 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1105 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1107 x_offset += 8 * h->mb_x;
1108 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
1110 if (list0 && list1) {
1111 /* don't optimize for luma-only case, since B-frames usually
1112 * use implicit weights => chroma too. */
1113 uint8_t *tmp_cb = h->bipred_scratchpad;
1114 uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
1115 uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
1116 int refn0 = h->ref_cache[0][scan8[n]];
1117 int refn1 = h->ref_cache[1][scan8[n]];
1119 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
1120 dest_y, dest_cb, dest_cr,
1121 x_offset, y_offset, qpix_put, chroma_put,
1122 pixel_shift, chroma_idc);
1123 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
1124 tmp_y, tmp_cb, tmp_cr,
1125 x_offset, y_offset, qpix_put, chroma_put,
1126 pixel_shift, chroma_idc);
1128 if (h->use_weight == 2) {
1129 int weight0 = h->implicit_weight[refn0][refn1][h->mb_y & 1];
1130 int weight1 = 64 - weight0;
1131 luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
1132 height, 5, weight0, weight1, 0);
1133 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
1134 chroma_height, 5, weight0, weight1, 0);
1135 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
1136 chroma_height, 5, weight0, weight1, 0);
1138 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
1139 h->luma_log2_weight_denom,
1140 h->luma_weight[refn0][0][0],
1141 h->luma_weight[refn1][1][0],
1142 h->luma_weight[refn0][0][1] +
1143 h->luma_weight[refn1][1][1]);
1144 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
1145 h->chroma_log2_weight_denom,
1146 h->chroma_weight[refn0][0][0][0],
1147 h->chroma_weight[refn1][1][0][0],
1148 h->chroma_weight[refn0][0][0][1] +
1149 h->chroma_weight[refn1][1][0][1]);
1150 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
1151 h->chroma_log2_weight_denom,
1152 h->chroma_weight[refn0][0][1][0],
1153 h->chroma_weight[refn1][1][1][0],
1154 h->chroma_weight[refn0][0][1][1] +
1155 h->chroma_weight[refn1][1][1][1]);
1158 int list = list1 ? 1 : 0;
1159 int refn = h->ref_cache[list][scan8[n]];
1160 Picture *ref = &h->ref_list[list][refn];
1161 mc_dir_part(h, ref, n, square, height, delta, list,
1162 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1163 qpix_put, chroma_put, pixel_shift, chroma_idc);
1165 luma_weight_op(dest_y, h->mb_linesize, height,
1166 h->luma_log2_weight_denom,
1167 h->luma_weight[refn][list][0],
1168 h->luma_weight[refn][list][1]);
1169 if (h->use_weight_chroma) {
1170 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
1171 h->chroma_log2_weight_denom,
1172 h->chroma_weight[refn][list][0][0],
1173 h->chroma_weight[refn][list][0][1]);
1174 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
1175 h->chroma_log2_weight_denom,
1176 h->chroma_weight[refn][list][1][0],
1177 h->chroma_weight[refn][list][1][1]);
1182 static av_always_inline void prefetch_motion(H264Context *h, int list,
1183 int pixel_shift, int chroma_idc)
1185 /* fetch pixels for estimated mv 4 macroblocks ahead
1186 * optimized for 64byte cache lines */
1187 const int refn = h->ref_cache[list][scan8[0]];
1189 const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * h->mb_x + 8;
1190 const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * h->mb_y;
1191 uint8_t **src = h->ref_list[list][refn].f.data;
1192 int off = (mx << pixel_shift) +
1193 (my + (h->mb_x & 3) * 4) * h->mb_linesize +
1194 (64 << pixel_shift);
1195 h->vdsp.prefetch(src[0] + off, h->linesize, 4);
1196 if (chroma_idc == 3 /* yuv444 */) {
1197 h->vdsp.prefetch(src[1] + off, h->linesize, 4);
1198 h->vdsp.prefetch(src[2] + off, h->linesize, 4);
1200 off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (h->mb_x&7))*h->uvlinesize;
1201 h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
1206 static void free_tables(H264Context *h, int free_rbsp)
1211 av_freep(&h->intra4x4_pred_mode);
1212 av_freep(&h->chroma_pred_mode_table);
1213 av_freep(&h->cbp_table);
1214 av_freep(&h->mvd_table[0]);
1215 av_freep(&h->mvd_table[1]);
1216 av_freep(&h->direct_table);
1217 av_freep(&h->non_zero_count);
1218 av_freep(&h->slice_table_base);
1219 h->slice_table = NULL;
1220 av_freep(&h->list_counts);
1222 av_freep(&h->mb2b_xy);
1223 av_freep(&h->mb2br_xy);
1225 for (i = 0; i < 3; i++)
1226 av_freep(&h->visualization_buffer[i]);
1228 av_buffer_pool_uninit(&h->qscale_table_pool);
1229 av_buffer_pool_uninit(&h->mb_type_pool);
1230 av_buffer_pool_uninit(&h->motion_val_pool);
1231 av_buffer_pool_uninit(&h->ref_index_pool);
1233 if (free_rbsp && h->DPB) {
1234 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1235 unref_picture(h, &h->DPB[i]);
1237 } else if (h->DPB) {
1238 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1239 h->DPB[i].needs_realloc = 1;
1242 h->cur_pic_ptr = NULL;
1244 for (i = 0; i < MAX_THREADS; i++) {
1245 hx = h->thread_context[i];
1248 av_freep(&hx->top_borders[1]);
1249 av_freep(&hx->top_borders[0]);
1250 av_freep(&hx->bipred_scratchpad);
1251 av_freep(&hx->edge_emu_buffer);
1252 av_freep(&hx->dc_val_base);
1253 av_freep(&hx->me.scratchpad);
1254 av_freep(&hx->er.mb_index2xy);
1255 av_freep(&hx->er.error_status_table);
1256 av_freep(&hx->er.er_temp_buffer);
1257 av_freep(&hx->er.mbintra_table);
1258 av_freep(&hx->er.mbskip_table);
1261 av_freep(&hx->rbsp_buffer[1]);
1262 av_freep(&hx->rbsp_buffer[0]);
1263 hx->rbsp_buffer_size[0] = 0;
1264 hx->rbsp_buffer_size[1] = 0;
1267 av_freep(&h->thread_context[i]);
1271 static void init_dequant8_coeff_table(H264Context *h)
1274 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1276 for (i = 0; i < 6; i++) {
1277 h->dequant8_coeff[i] = h->dequant8_buffer[i];
1278 for (j = 0; j < i; j++)
1279 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
1280 64 * sizeof(uint8_t))) {
1281 h->dequant8_coeff[i] = h->dequant8_buffer[j];
1287 for (q = 0; q < max_qp + 1; q++) {
1288 int shift = div6[q];
1290 for (x = 0; x < 64; x++)
1291 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
1292 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
1293 h->pps.scaling_matrix8[i][x]) << shift;
1298 static void init_dequant4_coeff_table(H264Context *h)
1301 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1302 for (i = 0; i < 6; i++) {
1303 h->dequant4_coeff[i] = h->dequant4_buffer[i];
1304 for (j = 0; j < i; j++)
1305 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
1306 16 * sizeof(uint8_t))) {
1307 h->dequant4_coeff[i] = h->dequant4_buffer[j];
1313 for (q = 0; q < max_qp + 1; q++) {
1314 int shift = div6[q] + 2;
1316 for (x = 0; x < 16; x++)
1317 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
1318 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
1319 h->pps.scaling_matrix4[i][x]) << shift;
1324 static void init_dequant_tables(H264Context *h)
1327 init_dequant4_coeff_table(h);
1328 if (h->pps.transform_8x8_mode)
1329 init_dequant8_coeff_table(h);
1330 if (h->sps.transform_bypass) {
1331 for (i = 0; i < 6; i++)
1332 for (x = 0; x < 16; x++)
1333 h->dequant4_coeff[i][0][x] = 1 << 6;
1334 if (h->pps.transform_8x8_mode)
1335 for (i = 0; i < 6; i++)
1336 for (x = 0; x < 64; x++)
1337 h->dequant8_coeff[i][0][x] = 1 << 6;
1341 int ff_h264_alloc_tables(H264Context *h)
1343 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
1344 const int row_mb_num = 2*h->mb_stride*FFMAX(h->avctx->thread_count, 1);
1347 FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
1348 row_mb_num * 8 * sizeof(uint8_t), fail)
1349 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
1350 big_mb_num * 48 * sizeof(uint8_t), fail)
1351 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
1352 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
1353 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
1354 big_mb_num * sizeof(uint16_t), fail)
1355 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
1356 big_mb_num * sizeof(uint8_t), fail)
1357 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0],
1358 16 * row_mb_num * sizeof(uint8_t), fail);
1359 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1],
1360 16 * row_mb_num * sizeof(uint8_t), fail);
1361 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
1362 4 * big_mb_num * sizeof(uint8_t), fail);
1363 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
1364 big_mb_num * sizeof(uint8_t), fail)
1366 memset(h->slice_table_base, -1,
1367 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
1368 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
1370 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
1371 big_mb_num * sizeof(uint32_t), fail);
1372 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
1373 big_mb_num * sizeof(uint32_t), fail);
1374 for (y = 0; y < h->mb_height; y++)
1375 for (x = 0; x < h->mb_width; x++) {
1376 const int mb_xy = x + y * h->mb_stride;
1377 const int b_xy = 4 * x + 4 * y * h->b_stride;
1379 h->mb2b_xy[mb_xy] = b_xy;
1380 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
1383 if (!h->dequant4_coeff[0])
1384 init_dequant_tables(h);
1387 h->DPB = av_mallocz_array(MAX_PICTURE_COUNT, sizeof(*h->DPB));
1389 return AVERROR(ENOMEM);
1390 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1391 avcodec_get_frame_defaults(&h->DPB[i].f);
1392 avcodec_get_frame_defaults(&h->cur_pic.f);
1399 return AVERROR(ENOMEM);
1403 * Mimic alloc_tables(), but for every context thread.
1405 static void clone_tables(H264Context *dst, H264Context *src, int i)
1407 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
1408 dst->non_zero_count = src->non_zero_count;
1409 dst->slice_table = src->slice_table;
1410 dst->cbp_table = src->cbp_table;
1411 dst->mb2b_xy = src->mb2b_xy;
1412 dst->mb2br_xy = src->mb2br_xy;
1413 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
1414 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
1415 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
1416 dst->direct_table = src->direct_table;
1417 dst->list_counts = src->list_counts;
1418 dst->DPB = src->DPB;
1419 dst->cur_pic_ptr = src->cur_pic_ptr;
1420 dst->cur_pic = src->cur_pic;
1421 dst->bipred_scratchpad = NULL;
1422 dst->edge_emu_buffer = NULL;
1423 dst->me.scratchpad = NULL;
1424 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
1425 src->sps.chroma_format_idc);
1430 * Allocate buffers which are not shared amongst multiple threads.
1432 static int context_init(H264Context *h)
1434 ERContext *er = &h->er;
1435 int mb_array_size = h->mb_height * h->mb_stride;
1436 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
1437 int c_size = h->mb_stride * (h->mb_height + 1);
1438 int yc_size = y_size + 2 * c_size;
1441 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[0],
1442 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1443 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[1],
1444 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1446 h->ref_cache[0][scan8[5] + 1] =
1447 h->ref_cache[0][scan8[7] + 1] =
1448 h->ref_cache[0][scan8[13] + 1] =
1449 h->ref_cache[1][scan8[5] + 1] =
1450 h->ref_cache[1][scan8[7] + 1] =
1451 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
1453 if (CONFIG_ERROR_RESILIENCE) {
1455 er->avctx = h->avctx;
1457 er->decode_mb = h264_er_decode_mb;
1459 er->quarter_sample = 1;
1461 er->mb_num = h->mb_num;
1462 er->mb_width = h->mb_width;
1463 er->mb_height = h->mb_height;
1464 er->mb_stride = h->mb_stride;
1465 er->b8_stride = h->mb_width * 2 + 1;
1467 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy, (h->mb_num + 1) * sizeof(int),
1468 fail); // error ressilience code looks cleaner with this
1469 for (y = 0; y < h->mb_height; y++)
1470 for (x = 0; x < h->mb_width; x++)
1471 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
1473 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
1474 h->mb_stride + h->mb_width;
1476 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
1477 mb_array_size * sizeof(uint8_t), fail);
1479 FF_ALLOC_OR_GOTO(h->avctx, er->mbintra_table, mb_array_size, fail);
1480 memset(er->mbintra_table, 1, mb_array_size);
1482 FF_ALLOCZ_OR_GOTO(h->avctx, er->mbskip_table, mb_array_size + 2, fail);
1484 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer, h->mb_height * h->mb_stride,
1487 FF_ALLOCZ_OR_GOTO(h->avctx, h->dc_val_base, yc_size * sizeof(int16_t), fail);
1488 er->dc_val[0] = h->dc_val_base + h->mb_width * 2 + 2;
1489 er->dc_val[1] = h->dc_val_base + y_size + h->mb_stride + 1;
1490 er->dc_val[2] = er->dc_val[1] + c_size;
1491 for (i = 0; i < yc_size; i++)
1492 h->dc_val_base[i] = 1024;
1498 return AVERROR(ENOMEM); // free_tables will clean up for us
1501 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1502 int parse_extradata);
1504 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1506 AVCodecContext *avctx = h->avctx;
1509 if (!buf || size <= 0)
1513 int i, cnt, nalsize;
1514 const unsigned char *p = buf;
1519 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1520 return AVERROR_INVALIDDATA;
1522 /* sps and pps in the avcC always have length coded with 2 bytes,
1523 * so put a fake nal_length_size = 2 while parsing them */
1524 h->nal_length_size = 2;
1525 // Decode sps from avcC
1526 cnt = *(p + 5) & 0x1f; // Number of sps
1528 for (i = 0; i < cnt; i++) {
1529 nalsize = AV_RB16(p) + 2;
1530 if(nalsize > size - (p-buf))
1531 return AVERROR_INVALIDDATA;
1532 ret = decode_nal_units(h, p, nalsize, 1);
1534 av_log(avctx, AV_LOG_ERROR,
1535 "Decoding sps %d from avcC failed\n", i);
1540 // Decode pps from avcC
1541 cnt = *(p++); // Number of pps
1542 for (i = 0; i < cnt; i++) {
1543 nalsize = AV_RB16(p) + 2;
1544 if(nalsize > size - (p-buf))
1545 return AVERROR_INVALIDDATA;
1546 ret = decode_nal_units(h, p, nalsize, 1);
1548 av_log(avctx, AV_LOG_ERROR,
1549 "Decoding pps %d from avcC failed\n", i);
1554 // Now store right nal length size, that will be used to parse all other nals
1555 h->nal_length_size = (buf[4] & 0x03) + 1;
1558 ret = decode_nal_units(h, buf, size, 1);
1565 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1567 H264Context *h = avctx->priv_data;
1573 h->bit_depth_luma = 8;
1574 h->chroma_format_idc = 1;
1576 h->avctx->bits_per_raw_sample = 8;
1577 h->cur_chroma_format_idc = 1;
1579 ff_h264dsp_init(&h->h264dsp, 8, 1);
1580 av_assert0(h->sps.bit_depth_chroma == 0);
1581 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1582 ff_h264qpel_init(&h->h264qpel, 8);
1583 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, 8, 1);
1585 h->dequant_coeff_pps = -1;
1586 h->current_sps_id = -1;
1588 /* needed so that IDCT permutation is known early */
1589 if (CONFIG_ERROR_RESILIENCE)
1590 ff_dsputil_init(&h->dsp, h->avctx);
1591 ff_videodsp_init(&h->vdsp, 8);
1593 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1594 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1596 h->picture_structure = PICT_FRAME;
1597 h->slice_context_count = 1;
1598 h->workaround_bugs = avctx->workaround_bugs;
1599 h->flags = avctx->flags;
1602 // s->decode_mb = ff_h263_decode_mb;
1603 if (!avctx->has_b_frames)
1606 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1608 ff_h264_decode_init_vlc();
1610 ff_init_cabac_states();
1613 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1615 h->thread_context[0] = h;
1616 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1617 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1618 h->last_pocs[i] = INT_MIN;
1619 h->prev_poc_msb = 1 << 16;
1620 h->prev_frame_num = -1;
1622 h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
1623 ff_h264_reset_sei(h);
1624 if (avctx->codec_id == AV_CODEC_ID_H264) {
1625 if (avctx->ticks_per_frame == 1) {
1626 if(h->avctx->time_base.den < INT_MAX/2) {
1627 h->avctx->time_base.den *= 2;
1629 h->avctx->time_base.num /= 2;
1631 avctx->ticks_per_frame = 2;
1634 if (avctx->extradata_size > 0 && avctx->extradata) {
1635 ret = ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
1637 ff_h264_free_context(h);
1642 if (h->sps.bitstream_restriction_flag &&
1643 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
1644 h->avctx->has_b_frames = h->sps.num_reorder_frames;
1648 avctx->internal->allocate_progress = 1;
1655 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1656 #undef REBASE_PICTURE
1657 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
1658 ((pic && pic >= old_ctx->DPB && \
1659 pic < old_ctx->DPB + MAX_PICTURE_COUNT) ? \
1660 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
1662 static void copy_picture_range(Picture **to, Picture **from, int count,
1663 H264Context *new_base,
1664 H264Context *old_base)
1668 for (i = 0; i < count; i++) {
1669 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1670 IN_RANGE(from[i], old_base->DPB,
1671 sizeof(Picture) * MAX_PICTURE_COUNT) ||
1673 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1677 static void copy_parameter_set(void **to, void **from, int count, int size)
1681 for (i = 0; i < count; i++) {
1682 if (to[i] && !from[i])
1684 else if (from[i] && !to[i])
1685 to[i] = av_malloc(size);
1688 memcpy(to[i], from[i], size);
1692 static int decode_init_thread_copy(AVCodecContext *avctx)
1694 H264Context *h = avctx->priv_data;
1696 if (!avctx->internal->is_copy)
1698 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1699 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1701 h->rbsp_buffer[0] = NULL;
1702 h->rbsp_buffer[1] = NULL;
1703 h->rbsp_buffer_size[0] = 0;
1704 h->rbsp_buffer_size[1] = 0;
1705 h->context_initialized = 0;
1710 #define copy_fields(to, from, start_field, end_field) \
1711 memcpy(&to->start_field, &from->start_field, \
1712 (char *)&to->end_field - (char *)&to->start_field)
1714 static int h264_slice_header_init(H264Context *, int);
1716 static int h264_set_parameter_from_sps(H264Context *h);
1718 static int decode_update_thread_context(AVCodecContext *dst,
1719 const AVCodecContext *src)
1721 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1722 int inited = h->context_initialized, err = 0;
1723 int context_reinitialized = 0;
1730 (h->width != h1->width ||
1731 h->height != h1->height ||
1732 h->mb_width != h1->mb_width ||
1733 h->mb_height != h1->mb_height ||
1734 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
1735 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
1736 h->sps.colorspace != h1->sps.colorspace)) {
1738 /* set bits_per_raw_sample to the previous value. the check for changed
1739 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
1740 * the current value */
1741 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1743 av_freep(&h->bipred_scratchpad);
1745 h->width = h1->width;
1746 h->height = h1->height;
1747 h->mb_height = h1->mb_height;
1748 h->mb_width = h1->mb_width;
1749 h->mb_num = h1->mb_num;
1750 h->mb_stride = h1->mb_stride;
1751 h->b_stride = h1->b_stride;
1753 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1754 MAX_SPS_COUNT, sizeof(SPS));
1756 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1757 MAX_PPS_COUNT, sizeof(PPS));
1760 if ((err = h264_slice_header_init(h, 1)) < 0) {
1761 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
1764 context_reinitialized = 1;
1767 h264_set_parameter_from_sps(h);
1768 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
1769 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
1772 /* update linesize on resize for h264. The h264 decoder doesn't
1773 * necessarily call ff_MPV_frame_start in the new thread */
1774 h->linesize = h1->linesize;
1775 h->uvlinesize = h1->uvlinesize;
1777 /* copy block_offset since frame_start may not be called */
1778 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
1781 for (i = 0; i < MAX_SPS_COUNT; i++)
1782 av_freep(h->sps_buffers + i);
1784 for (i = 0; i < MAX_PPS_COUNT; i++)
1785 av_freep(h->pps_buffers + i);
1787 av_freep(&h->rbsp_buffer[0]);
1788 av_freep(&h->rbsp_buffer[1]);
1789 memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
1790 memcpy(&h->cabac, &h1->cabac,
1791 sizeof(H264Context) - offsetof(H264Context, cabac));
1792 av_assert0((void*)&h->cabac == &h->mb_padding + 1);
1794 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1795 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1797 memset(&h->er, 0, sizeof(h->er));
1798 memset(&h->me, 0, sizeof(h->me));
1799 memset(&h->mb, 0, sizeof(h->mb));
1800 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
1801 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
1802 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
1806 h->qscale_table_pool = NULL;
1807 h->mb_type_pool = NULL;
1808 h->ref_index_pool = NULL;
1809 h->motion_val_pool = NULL;
1810 for (i = 0; i < 2; i++) {
1811 h->rbsp_buffer[i] = NULL;
1812 h->rbsp_buffer_size[i] = 0;
1815 if (h1->context_initialized) {
1816 h->context_initialized = 0;
1818 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
1819 avcodec_get_frame_defaults(&h->cur_pic.f);
1820 h->cur_pic.tf.f = &h->cur_pic.f;
1822 ret = ff_h264_alloc_tables(h);
1824 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1827 ret = context_init(h);
1829 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
1834 h->bipred_scratchpad = NULL;
1835 h->edge_emu_buffer = NULL;
1837 h->thread_context[0] = h;
1838 h->context_initialized = h1->context_initialized;
1841 h->avctx->coded_height = h1->avctx->coded_height;
1842 h->avctx->coded_width = h1->avctx->coded_width;
1843 h->avctx->width = h1->avctx->width;
1844 h->avctx->height = h1->avctx->height;
1845 h->coded_picture_number = h1->coded_picture_number;
1846 h->first_field = h1->first_field;
1847 h->picture_structure = h1->picture_structure;
1848 h->qscale = h1->qscale;
1849 h->droppable = h1->droppable;
1850 h->data_partitioning = h1->data_partitioning;
1851 h->low_delay = h1->low_delay;
1853 for (i = 0; h->DPB && i < MAX_PICTURE_COUNT; i++) {
1854 unref_picture(h, &h->DPB[i]);
1855 if (h1->DPB && h1->DPB[i].f.data[0] &&
1856 (ret = ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
1860 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
1861 unref_picture(h, &h->cur_pic);
1862 if (h1->cur_pic.f.buf[0] && (ret = ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
1865 h->workaround_bugs = h1->workaround_bugs;
1866 h->low_delay = h1->low_delay;
1867 h->droppable = h1->droppable;
1869 // extradata/NAL handling
1870 h->is_avc = h1->is_avc;
1873 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1874 MAX_SPS_COUNT, sizeof(SPS));
1876 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1877 MAX_PPS_COUNT, sizeof(PPS));
1880 // Dequantization matrices
1881 // FIXME these are big - can they be only copied when PPS changes?
1882 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1884 for (i = 0; i < 6; i++)
1885 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1886 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1888 for (i = 0; i < 6; i++)
1889 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1890 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1892 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1895 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1898 copy_fields(h, h1, short_ref, cabac_init_idc);
1900 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
1901 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
1902 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1903 MAX_DELAYED_PIC_COUNT + 2, h, h1);
1907 if (context_reinitialized)
1908 h264_set_parameter_from_sps(h);
1910 if (!h->cur_pic_ptr)
1913 if (!h->droppable) {
1914 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1915 h->prev_poc_msb = h->poc_msb;
1916 h->prev_poc_lsb = h->poc_lsb;
1918 h->prev_frame_num_offset = h->frame_num_offset;
1919 h->prev_frame_num = h->frame_num;
1920 h->outputed_poc = h->next_outputed_poc;
1925 static int h264_frame_start(H264Context *h)
1929 const int pixel_shift = h->pixel_shift;
1931 1<<(h->sps.bit_depth_luma-1),
1932 1<<(h->sps.bit_depth_chroma-1),
1933 1<<(h->sps.bit_depth_chroma-1),
1937 if (!ff_thread_can_start_frame(h->avctx)) {
1938 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1942 release_unused_pictures(h, 1);
1943 h->cur_pic_ptr = NULL;
1945 i = find_unused_picture(h);
1947 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1952 pic->reference = h->droppable ? 0 : h->picture_structure;
1953 pic->f.coded_picture_number = h->coded_picture_number++;
1954 pic->field_picture = h->picture_structure != PICT_FRAME;
1957 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
1959 * See decode_nal_units().
1961 pic->f.key_frame = 0;
1963 pic->mmco_reset = 0;
1965 if ((ret = alloc_picture(h, pic)) < 0)
1967 if(!h->sync && !h->avctx->hwaccel &&
1968 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
1969 avpriv_color_frame(&pic->f, c);
1971 h->cur_pic_ptr = pic;
1972 unref_picture(h, &h->cur_pic);
1973 if (CONFIG_ERROR_RESILIENCE) {
1974 h->er.cur_pic = NULL;
1977 if ((ret = ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
1980 if (CONFIG_ERROR_RESILIENCE) {
1981 ff_er_frame_start(&h->er);
1983 h->er.next_pic = NULL;
1986 assert(h->linesize && h->uvlinesize);
1988 for (i = 0; i < 16; i++) {
1989 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1990 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1992 for (i = 0; i < 16; i++) {
1993 h->block_offset[16 + i] =
1994 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1995 h->block_offset[48 + 16 + i] =
1996 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1999 // s->decode = (h->flags & CODEC_FLAG_PSNR) || !s->encoding ||
2000 // h->cur_pic.reference /* || h->contains_intra */ || 1;
2002 /* We mark the current picture as non-reference after allocating it, so
2003 * that if we break out due to an error it can be released automatically
2004 * in the next ff_MPV_frame_start().
2006 h->cur_pic_ptr->reference = 0;
2008 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
2010 h->next_output_pic = NULL;
2012 assert(h->cur_pic_ptr->long_ref == 0);
2018 * Run setup operations that must be run after slice header decoding.
2019 * This includes finding the next displayed frame.
2021 * @param h h264 master context
2022 * @param setup_finished enough NALs have been read that we can call
2023 * ff_thread_finish_setup()
2025 static void decode_postinit(H264Context *h, int setup_finished)
2027 Picture *out = h->cur_pic_ptr;
2028 Picture *cur = h->cur_pic_ptr;
2029 int i, pics, out_of_order, out_idx;
2031 h->cur_pic_ptr->f.pict_type = h->pict_type;
2033 if (h->next_output_pic)
2036 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
2037 /* FIXME: if we have two PAFF fields in one packet, we can't start
2038 * the next thread here. If we have one field per packet, we can.
2039 * The check in decode_nal_units() is not good enough to find this
2040 * yet, so we assume the worst for now. */
2041 // if (setup_finished)
2042 // ff_thread_finish_setup(h->avctx);
2046 cur->f.interlaced_frame = 0;
2047 cur->f.repeat_pict = 0;
2049 /* Signal interlacing information externally. */
2050 /* Prioritize picture timing SEI information over used
2051 * decoding process if it exists. */
2053 if (h->sps.pic_struct_present_flag) {
2054 switch (h->sei_pic_struct) {
2055 case SEI_PIC_STRUCT_FRAME:
2057 case SEI_PIC_STRUCT_TOP_FIELD:
2058 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2059 cur->f.interlaced_frame = 1;
2061 case SEI_PIC_STRUCT_TOP_BOTTOM:
2062 case SEI_PIC_STRUCT_BOTTOM_TOP:
2063 if (FIELD_OR_MBAFF_PICTURE(h))
2064 cur->f.interlaced_frame = 1;
2066 // try to flag soft telecine progressive
2067 cur->f.interlaced_frame = h->prev_interlaced_frame;
2069 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2070 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2071 /* Signal the possibility of telecined film externally
2072 * (pic_struct 5,6). From these hints, let the applications
2073 * decide if they apply deinterlacing. */
2074 cur->f.repeat_pict = 1;
2076 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2077 cur->f.repeat_pict = 2;
2079 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2080 cur->f.repeat_pict = 4;
2084 if ((h->sei_ct_type & 3) &&
2085 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2086 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
2088 /* Derive interlacing flag from used decoding process. */
2089 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
2091 h->prev_interlaced_frame = cur->f.interlaced_frame;
2093 if (cur->field_poc[0] != cur->field_poc[1]) {
2094 /* Derive top_field_first from field pocs. */
2095 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
2097 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
2098 /* Use picture timing SEI information. Even if it is a
2099 * information of a past frame, better than nothing. */
2100 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
2101 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2102 cur->f.top_field_first = 1;
2104 cur->f.top_field_first = 0;
2106 /* Most likely progressive */
2107 cur->f.top_field_first = 0;
2111 cur->mmco_reset = h->mmco_reset;
2113 // FIXME do something with unavailable reference frames
2115 /* Sort B-frames into display order */
2117 if (h->sps.bitstream_restriction_flag &&
2118 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
2119 h->avctx->has_b_frames = h->sps.num_reorder_frames;
2123 if (h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
2124 !h->sps.bitstream_restriction_flag) {
2125 h->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
2129 for (i = 0; 1; i++) {
2130 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
2132 h->last_pocs[i-1] = cur->poc;
2135 h->last_pocs[i-1]= h->last_pocs[i];
2138 out_of_order = MAX_DELAYED_PIC_COUNT - i;
2139 if( cur->f.pict_type == AV_PICTURE_TYPE_B
2140 || (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))
2141 out_of_order = FFMAX(out_of_order, 1);
2142 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
2143 av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
2144 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
2145 h->last_pocs[i] = INT_MIN;
2146 h->last_pocs[0] = cur->poc;
2147 cur->mmco_reset = 1;
2148 } else if(h->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
2149 av_log(h->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
2150 h->avctx->has_b_frames = out_of_order;
2155 while (h->delayed_pic[pics])
2158 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
2160 h->delayed_pic[pics++] = cur;
2161 if (cur->reference == 0)
2162 cur->reference = DELAYED_PIC_REF;
2164 out = h->delayed_pic[0];
2166 for (i = 1; h->delayed_pic[i] &&
2167 !h->delayed_pic[i]->f.key_frame &&
2168 !h->delayed_pic[i]->mmco_reset;
2170 if (h->delayed_pic[i]->poc < out->poc) {
2171 out = h->delayed_pic[i];
2174 if (h->avctx->has_b_frames == 0 &&
2175 (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
2176 h->next_outputed_poc = INT_MIN;
2177 out_of_order = out->poc < h->next_outputed_poc;
2179 if (out_of_order || pics > h->avctx->has_b_frames) {
2180 out->reference &= ~DELAYED_PIC_REF;
2181 // for frame threading, the owner must be the second field's thread or
2182 // else the first thread can release the picture and reuse it unsafely
2183 for (i = out_idx; h->delayed_pic[i]; i++)
2184 h->delayed_pic[i] = h->delayed_pic[i + 1];
2186 if (!out_of_order && pics > h->avctx->has_b_frames) {
2187 h->next_output_pic = out;
2188 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
2189 h->next_outputed_poc = INT_MIN;
2191 h->next_outputed_poc = out->poc;
2193 av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
2196 if (h->next_output_pic && h->next_output_pic->sync) {
2200 if (setup_finished && !h->avctx->hwaccel)
2201 ff_thread_finish_setup(h->avctx);
2204 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
2205 uint8_t *src_cb, uint8_t *src_cr,
2206 int linesize, int uvlinesize,
2209 uint8_t *top_border;
2211 const int pixel_shift = h->pixel_shift;
2212 int chroma444 = CHROMA444(h);
2213 int chroma422 = CHROMA422(h);
2216 src_cb -= uvlinesize;
2217 src_cr -= uvlinesize;
2219 if (!simple && FRAME_MBAFF(h)) {
2222 top_border = h->top_borders[0][h->mb_x];
2223 AV_COPY128(top_border, src_y + 15 * linesize);
2225 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
2226 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2229 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2230 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
2231 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
2232 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
2234 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
2235 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
2237 } else if (chroma422) {
2239 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2240 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
2242 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
2243 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
2247 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
2248 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
2250 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
2251 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
2256 } else if (MB_MBAFF(h)) {
2262 top_border = h->top_borders[top_idx][h->mb_x];
2263 /* There are two lines saved, the line above the top macroblock
2264 * of a pair, and the line above the bottom macroblock. */
2265 AV_COPY128(top_border, src_y + 16 * linesize);
2267 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
2269 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2272 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
2273 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
2274 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
2275 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
2277 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
2278 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
2280 } else if (chroma422) {
2282 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
2283 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
2285 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
2286 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
2290 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
2291 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
2293 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
2294 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
2300 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
2301 uint8_t *src_cb, uint8_t *src_cr,
2302 int linesize, int uvlinesize,
2303 int xchg, int chroma444,
2304 int simple, int pixel_shift)
2306 int deblock_topleft;
2309 uint8_t *top_border_m1;
2310 uint8_t *top_border;
2312 if (!simple && FRAME_MBAFF(h)) {
2317 top_idx = MB_MBAFF(h) ? 0 : 1;
2321 if (h->deblocking_filter == 2) {
2322 deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num;
2323 deblock_top = h->top_type;
2325 deblock_topleft = (h->mb_x > 0);
2326 deblock_top = (h->mb_y > !!MB_FIELD(h));
2329 src_y -= linesize + 1 + pixel_shift;
2330 src_cb -= uvlinesize + 1 + pixel_shift;
2331 src_cr -= uvlinesize + 1 + pixel_shift;
2333 top_border_m1 = h->top_borders[top_idx][h->mb_x - 1];
2334 top_border = h->top_borders[top_idx][h->mb_x];
2336 #define XCHG(a, b, xchg) \
2337 if (pixel_shift) { \
2339 AV_SWAP64(b + 0, a + 0); \
2340 AV_SWAP64(b + 8, a + 8); \
2350 if (deblock_topleft) {
2351 XCHG(top_border_m1 + (8 << pixel_shift),
2352 src_y - (7 << pixel_shift), 1);
2354 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
2355 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
2356 if (h->mb_x + 1 < h->mb_width) {
2357 XCHG(h->top_borders[top_idx][h->mb_x + 1],
2358 src_y + (17 << pixel_shift), 1);
2360 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2362 if (deblock_topleft) {
2363 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2364 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2366 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
2367 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
2368 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
2369 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
2370 if (h->mb_x + 1 < h->mb_width) {
2371 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
2372 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
2375 if (deblock_topleft) {
2376 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2377 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2379 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
2380 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
2386 static av_always_inline int dctcoef_get(int16_t *mb, int high_bit_depth,
2389 if (high_bit_depth) {
2390 return AV_RN32A(((int32_t *)mb) + index);
2392 return AV_RN16A(mb + index);
2395 static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth,
2396 int index, int value)
2398 if (high_bit_depth) {
2399 AV_WN32A(((int32_t *)mb) + index, value);
2401 AV_WN16A(mb + index, value);
2404 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
2405 int mb_type, int is_h264,
2407 int transform_bypass,
2411 uint8_t *dest_y, int p)
2413 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2414 void (*idct_dc_add)(uint8_t *dst, int16_t *block, int stride);
2416 int qscale = p == 0 ? h->qscale : h->chroma_qp[p - 1];
2417 block_offset += 16 * p;
2418 if (IS_INTRA4x4(mb_type)) {
2419 if (IS_8x8DCT(mb_type)) {
2420 if (transform_bypass) {
2422 idct_add = h->h264dsp.h264_add_pixels8_clear;
2424 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
2425 idct_add = h->h264dsp.h264_idct8_add;
2427 for (i = 0; i < 16; i += 4) {
2428 uint8_t *const ptr = dest_y + block_offset[i];
2429 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2430 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2431 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2433 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2434 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
2435 (h->topright_samples_available << i) & 0x4000, linesize);
2437 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2438 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2440 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2445 if (transform_bypass) {
2447 idct_add = h->h264dsp.h264_add_pixels4_clear;
2449 idct_dc_add = h->h264dsp.h264_idct_dc_add;
2450 idct_add = h->h264dsp.h264_idct_add;
2452 for (i = 0; i < 16; i++) {
2453 uint8_t *const ptr = dest_y + block_offset[i];
2454 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2456 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2457 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2462 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
2463 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
2464 av_assert2(h->mb_y || linesize <= block_offset[i]);
2465 if (!topright_avail) {
2467 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
2468 topright = (uint8_t *)&tr_high;
2470 tr = ptr[3 - linesize] * 0x01010101u;
2471 topright = (uint8_t *)&tr;
2474 topright = ptr + (4 << pixel_shift) - linesize;
2478 h->hpc.pred4x4[dir](ptr, topright, linesize);
2479 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2482 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2483 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2485 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2486 } else if (CONFIG_SVQ3_DECODER)
2487 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
2493 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
2495 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
2496 if (!transform_bypass)
2497 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
2499 h->dequant4_coeff[p][qscale][0]);
2501 static const uint8_t dc_mapping[16] = {
2502 0 * 16, 1 * 16, 4 * 16, 5 * 16,
2503 2 * 16, 3 * 16, 6 * 16, 7 * 16,
2504 8 * 16, 9 * 16, 12 * 16, 13 * 16,
2505 10 * 16, 11 * 16, 14 * 16, 15 * 16
2507 for (i = 0; i < 16; i++)
2508 dctcoef_set(h->mb + (p * 256 << pixel_shift),
2509 pixel_shift, dc_mapping[i],
2510 dctcoef_get(h->mb_luma_dc[p],
2514 } else if (CONFIG_SVQ3_DECODER)
2515 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
2516 h->mb_luma_dc[p], qscale);
2520 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
2521 int is_h264, int simple,
2522 int transform_bypass,
2526 uint8_t *dest_y, int p)
2528 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2530 block_offset += 16 * p;
2531 if (!IS_INTRA4x4(mb_type)) {
2533 if (IS_INTRA16x16(mb_type)) {
2534 if (transform_bypass) {
2535 if (h->sps.profile_idc == 244 &&
2536 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
2537 h->intra16x16_pred_mode == HOR_PRED8x8)) {
2538 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
2539 h->mb + (p * 256 << pixel_shift),
2542 for (i = 0; i < 16; i++)
2543 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
2544 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2545 h->h264dsp.h264_add_pixels4_clear(dest_y + block_offset[i],
2546 h->mb + (i * 16 + p * 256 << pixel_shift),
2550 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
2551 h->mb + (p * 256 << pixel_shift),
2553 h->non_zero_count_cache + p * 5 * 8);
2555 } else if (h->cbp & 15) {
2556 if (transform_bypass) {
2557 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2558 idct_add = IS_8x8DCT(mb_type) ? h->h264dsp.h264_add_pixels8_clear
2559 : h->h264dsp.h264_add_pixels4_clear;
2560 for (i = 0; i < 16; i += di)
2561 if (h->non_zero_count_cache[scan8[i + p * 16]])
2562 idct_add(dest_y + block_offset[i],
2563 h->mb + (i * 16 + p * 256 << pixel_shift),
2566 if (IS_8x8DCT(mb_type))
2567 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
2568 h->mb + (p * 256 << pixel_shift),
2570 h->non_zero_count_cache + p * 5 * 8);
2572 h->h264dsp.h264_idct_add16(dest_y, block_offset,
2573 h->mb + (p * 256 << pixel_shift),
2575 h->non_zero_count_cache + p * 5 * 8);
2578 } else if (CONFIG_SVQ3_DECODER) {
2579 for (i = 0; i < 16; i++)
2580 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
2581 // FIXME benchmark weird rule, & below
2582 uint8_t *const ptr = dest_y + block_offset[i];
2583 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
2584 h->qscale, IS_INTRA(mb_type) ? 1 : 0);
2592 #include "h264_mb_template.c"
2596 #include "h264_mb_template.c"
2600 #include "h264_mb_template.c"
2602 void ff_h264_hl_decode_mb(H264Context *h)
2604 const int mb_xy = h->mb_xy;
2605 const int mb_type = h->cur_pic.mb_type[mb_xy];
2606 int is_complex = CONFIG_SMALL || h->is_complex ||
2607 IS_INTRA_PCM(mb_type) || h->qscale == 0;
2610 if (is_complex || h->pixel_shift)
2611 hl_decode_mb_444_complex(h);
2613 hl_decode_mb_444_simple_8(h);
2614 } else if (is_complex) {
2615 hl_decode_mb_complex(h);
2616 } else if (h->pixel_shift) {
2617 hl_decode_mb_simple_16(h);
2619 hl_decode_mb_simple_8(h);
2622 int ff_pred_weight_table(H264Context *h)
2625 int luma_def, chroma_def;
2628 h->use_weight_chroma = 0;
2629 h->luma_log2_weight_denom = get_ue_golomb(&h->gb);
2630 if (h->sps.chroma_format_idc)
2631 h->chroma_log2_weight_denom = get_ue_golomb(&h->gb);
2632 luma_def = 1 << h->luma_log2_weight_denom;
2633 chroma_def = 1 << h->chroma_log2_weight_denom;
2635 for (list = 0; list < 2; list++) {
2636 h->luma_weight_flag[list] = 0;
2637 h->chroma_weight_flag[list] = 0;
2638 for (i = 0; i < h->ref_count[list]; i++) {
2639 int luma_weight_flag, chroma_weight_flag;
2641 luma_weight_flag = get_bits1(&h->gb);
2642 if (luma_weight_flag) {
2643 h->luma_weight[i][list][0] = get_se_golomb(&h->gb);
2644 h->luma_weight[i][list][1] = get_se_golomb(&h->gb);
2645 if (h->luma_weight[i][list][0] != luma_def ||
2646 h->luma_weight[i][list][1] != 0) {
2648 h->luma_weight_flag[list] = 1;
2651 h->luma_weight[i][list][0] = luma_def;
2652 h->luma_weight[i][list][1] = 0;
2655 if (h->sps.chroma_format_idc) {
2656 chroma_weight_flag = get_bits1(&h->gb);
2657 if (chroma_weight_flag) {
2659 for (j = 0; j < 2; j++) {
2660 h->chroma_weight[i][list][j][0] = get_se_golomb(&h->gb);
2661 h->chroma_weight[i][list][j][1] = get_se_golomb(&h->gb);
2662 if (h->chroma_weight[i][list][j][0] != chroma_def ||
2663 h->chroma_weight[i][list][j][1] != 0) {
2664 h->use_weight_chroma = 1;
2665 h->chroma_weight_flag[list] = 1;
2670 for (j = 0; j < 2; j++) {
2671 h->chroma_weight[i][list][j][0] = chroma_def;
2672 h->chroma_weight[i][list][j][1] = 0;
2677 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2680 h->use_weight = h->use_weight || h->use_weight_chroma;
2685 * Initialize implicit_weight table.
2686 * @param field 0/1 initialize the weight for interlaced MBAFF
2687 * -1 initializes the rest
2689 static void implicit_weight_table(H264Context *h, int field)
2691 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2693 for (i = 0; i < 2; i++) {
2694 h->luma_weight_flag[i] = 0;
2695 h->chroma_weight_flag[i] = 0;
2699 if (h->picture_structure == PICT_FRAME) {
2700 cur_poc = h->cur_pic_ptr->poc;
2702 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
2704 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
2705 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2707 h->use_weight_chroma = 0;
2711 ref_count0 = h->ref_count[0];
2712 ref_count1 = h->ref_count[1];
2714 cur_poc = h->cur_pic_ptr->field_poc[field];
2716 ref_count0 = 16 + 2 * h->ref_count[0];
2717 ref_count1 = 16 + 2 * h->ref_count[1];
2721 h->use_weight_chroma = 2;
2722 h->luma_log2_weight_denom = 5;
2723 h->chroma_log2_weight_denom = 5;
2725 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2726 int poc0 = h->ref_list[0][ref0].poc;
2727 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2729 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2730 int poc1 = h->ref_list[1][ref1].poc;
2731 int td = av_clip(poc1 - poc0, -128, 127);
2733 int tb = av_clip(cur_poc - poc0, -128, 127);
2734 int tx = (16384 + (FFABS(td) >> 1)) / td;
2735 int dist_scale_factor = (tb * tx + 32) >> 8;
2736 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2737 w = 64 - dist_scale_factor;
2741 h->implicit_weight[ref0][ref1][0] =
2742 h->implicit_weight[ref0][ref1][1] = w;
2744 h->implicit_weight[ref0][ref1][field] = w;
2751 * instantaneous decoder refresh.
2753 static void idr(H264Context *h)
2756 ff_h264_remove_all_refs(h);
2757 h->prev_frame_num = 0;
2758 h->prev_frame_num_offset = 0;
2759 h->prev_poc_msb = 1<<16;
2760 h->prev_poc_lsb = 0;
2761 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2762 h->last_pocs[i] = INT_MIN;
2765 /* forget old pics after a seek */
2766 static void flush_change(H264Context *h)
2770 h->outputed_poc = h->next_outputed_poc = INT_MIN;
2771 h->prev_interlaced_frame = 1;
2774 h->prev_frame_num = -1;
2775 if (h->cur_pic_ptr) {
2776 h->cur_pic_ptr->reference = 0;
2777 for (j=i=0; h->delayed_pic[i]; i++)
2778 if (h->delayed_pic[i] != h->cur_pic_ptr)
2779 h->delayed_pic[j++] = h->delayed_pic[i];
2780 h->delayed_pic[j] = NULL;
2783 memset(h->ref_list[0], 0, sizeof(h->ref_list[0]));
2784 memset(h->ref_list[1], 0, sizeof(h->ref_list[1]));
2785 memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0]));
2786 memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1]));
2787 ff_h264_reset_sei(h);
2788 h->recovery_frame= -1;
2791 h->current_slice = 0;
2795 /* forget old pics after a seek */
2796 static void flush_dpb(AVCodecContext *avctx)
2798 H264Context *h = avctx->priv_data;
2801 for (i = 0; i <= MAX_DELAYED_PIC_COUNT; i++) {
2802 if (h->delayed_pic[i])
2803 h->delayed_pic[i]->reference = 0;
2804 h->delayed_pic[i] = NULL;
2810 for (i = 0; i < MAX_PICTURE_COUNT; i++)
2811 unref_picture(h, &h->DPB[i]);
2812 h->cur_pic_ptr = NULL;
2813 unref_picture(h, &h->cur_pic);
2815 h->mb_x = h->mb_y = 0;
2817 h->parse_context.state = -1;
2818 h->parse_context.frame_start_found = 0;
2819 h->parse_context.overread = 0;
2820 h->parse_context.overread_index = 0;
2821 h->parse_context.index = 0;
2822 h->parse_context.last_index = 0;
2825 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
2827 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2830 h->frame_num_offset = h->prev_frame_num_offset;
2831 if (h->frame_num < h->prev_frame_num)
2832 h->frame_num_offset += max_frame_num;
2834 if (h->sps.poc_type == 0) {
2835 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2837 if (h->poc_lsb < h->prev_poc_lsb &&
2838 h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2839 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2840 else if (h->poc_lsb > h->prev_poc_lsb &&
2841 h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2842 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2844 h->poc_msb = h->prev_poc_msb;
2846 field_poc[1] = h->poc_msb + h->poc_lsb;
2847 if (h->picture_structure == PICT_FRAME)
2848 field_poc[1] += h->delta_poc_bottom;
2849 } else if (h->sps.poc_type == 1) {
2850 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2853 if (h->sps.poc_cycle_length != 0)
2854 abs_frame_num = h->frame_num_offset + h->frame_num;
2858 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2861 expected_delta_per_poc_cycle = 0;
2862 for (i = 0; i < h->sps.poc_cycle_length; i++)
2863 // FIXME integrate during sps parse
2864 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2866 if (abs_frame_num > 0) {
2867 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2868 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2870 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2871 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2872 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2876 if (h->nal_ref_idc == 0)
2877 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2879 field_poc[0] = expectedpoc + h->delta_poc[0];
2880 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2882 if (h->picture_structure == PICT_FRAME)
2883 field_poc[1] += h->delta_poc[1];
2885 int poc = 2 * (h->frame_num_offset + h->frame_num);
2887 if (!h->nal_ref_idc)
2894 if (h->picture_structure != PICT_BOTTOM_FIELD)
2895 pic_field_poc[0] = field_poc[0];
2896 if (h->picture_structure != PICT_TOP_FIELD)
2897 pic_field_poc[1] = field_poc[1];
2898 *pic_poc = FFMIN(pic_field_poc[0], pic_field_poc[1]);
2904 * initialize scan tables
2906 static void init_scan_tables(H264Context *h)
2909 for (i = 0; i < 16; i++) {
2910 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2911 h->zigzag_scan[i] = T(zigzag_scan[i]);
2912 h->field_scan[i] = T(field_scan[i]);
2915 for (i = 0; i < 64; i++) {
2916 #define T(x) (x >> 3) | ((x & 7) << 3)
2917 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2918 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2919 h->field_scan8x8[i] = T(field_scan8x8[i]);
2920 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2923 if (h->sps.transform_bypass) { // FIXME same ugly
2924 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2925 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
2926 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2927 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
2928 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2929 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2931 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2932 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
2933 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2934 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
2935 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2936 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2940 static int field_end(H264Context *h, int in_setup)
2942 AVCodecContext *const avctx = h->avctx;
2946 if (CONFIG_H264_VDPAU_DECODER &&
2947 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2948 ff_vdpau_h264_set_reference_frames(h);
2950 if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2951 if (!h->droppable) {
2952 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2953 h->prev_poc_msb = h->poc_msb;
2954 h->prev_poc_lsb = h->poc_lsb;
2956 h->prev_frame_num_offset = h->frame_num_offset;
2957 h->prev_frame_num = h->frame_num;
2958 h->outputed_poc = h->next_outputed_poc;
2961 if (avctx->hwaccel) {
2962 if (avctx->hwaccel->end_frame(avctx) < 0)
2963 av_log(avctx, AV_LOG_ERROR,
2964 "hardware accelerator failed to decode picture\n");
2967 if (CONFIG_H264_VDPAU_DECODER &&
2968 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2969 ff_vdpau_h264_picture_complete(h);
2972 * FIXME: Error handling code does not seem to support interlaced
2973 * when slices span multiple rows
2974 * The ff_er_add_slice calls don't work right for bottom
2975 * fields; they cause massive erroneous error concealing
2976 * Error marking covers both fields (top and bottom).
2977 * This causes a mismatched s->error_count
2978 * and a bad error table. Further, the error count goes to
2979 * INT_MAX when called for bottom field, because mb_y is
2980 * past end by one (callers fault) and resync_mb_y != 0
2981 * causes problems for the first MB line, too.
2983 if (CONFIG_ERROR_RESILIENCE &&
2984 !FIELD_PICTURE(h) && h->current_slice && !h->sps.new) {
2985 h->er.cur_pic = h->cur_pic_ptr;
2986 ff_er_frame_end(&h->er);
2988 if (!in_setup && !h->droppable)
2989 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
2990 h->picture_structure == PICT_BOTTOM_FIELD);
2993 h->current_slice = 0;
2999 * Replicate H264 "master" context to thread contexts.
3001 static int clone_slice(H264Context *dst, H264Context *src)
3003 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
3004 dst->cur_pic_ptr = src->cur_pic_ptr;
3005 dst->cur_pic = src->cur_pic;
3006 dst->linesize = src->linesize;
3007 dst->uvlinesize = src->uvlinesize;
3008 dst->first_field = src->first_field;
3010 dst->prev_poc_msb = src->prev_poc_msb;
3011 dst->prev_poc_lsb = src->prev_poc_lsb;
3012 dst->prev_frame_num_offset = src->prev_frame_num_offset;
3013 dst->prev_frame_num = src->prev_frame_num;
3014 dst->short_ref_count = src->short_ref_count;
3016 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
3017 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
3018 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
3020 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
3021 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
3027 * Compute profile from profile_idc and constraint_set?_flags.
3031 * @return profile as defined by FF_PROFILE_H264_*
3033 int ff_h264_get_profile(SPS *sps)
3035 int profile = sps->profile_idc;
3037 switch (sps->profile_idc) {
3038 case FF_PROFILE_H264_BASELINE:
3039 // constraint_set1_flag set to 1
3040 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
3042 case FF_PROFILE_H264_HIGH_10:
3043 case FF_PROFILE_H264_HIGH_422:
3044 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
3045 // constraint_set3_flag set to 1
3046 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
3053 static int h264_set_parameter_from_sps(H264Context *h)
3055 if (h->flags & CODEC_FLAG_LOW_DELAY ||
3056 (h->sps.bitstream_restriction_flag &&
3057 !h->sps.num_reorder_frames)) {
3058 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
3059 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
3060 "Reenabling low delay requires a codec flush.\n");
3065 if (h->avctx->has_b_frames < 2)
3066 h->avctx->has_b_frames = !h->low_delay;
3068 if (h->sps.bit_depth_luma != h->sps.bit_depth_chroma) {
3069 avpriv_request_sample(h->avctx,
3070 "Different chroma and luma bit depth");
3071 return AVERROR_PATCHWELCOME;
3074 if (h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3075 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
3076 if (h->avctx->codec &&
3077 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU &&
3078 (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
3079 av_log(h->avctx, AV_LOG_ERROR,
3080 "VDPAU decoding does not support video colorspace.\n");
3081 return AVERROR_INVALIDDATA;
3083 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 &&
3084 h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13) {
3085 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3086 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
3087 h->pixel_shift = h->sps.bit_depth_luma > 8;
3089 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
3090 h->sps.chroma_format_idc);
3091 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
3092 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
3093 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
3094 h->sps.chroma_format_idc);
3096 if (CONFIG_ERROR_RESILIENCE)
3097 ff_dsputil_init(&h->dsp, h->avctx);
3098 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
3100 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
3101 h->sps.bit_depth_luma);
3102 return AVERROR_INVALIDDATA;
3108 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
3110 switch (h->sps.bit_depth_luma) {
3113 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3114 return AV_PIX_FMT_GBRP9;
3116 return AV_PIX_FMT_YUV444P9;
3117 } else if (CHROMA422(h))
3118 return AV_PIX_FMT_YUV422P9;
3120 return AV_PIX_FMT_YUV420P9;
3124 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3125 return AV_PIX_FMT_GBRP10;
3127 return AV_PIX_FMT_YUV444P10;
3128 } else if (CHROMA422(h))
3129 return AV_PIX_FMT_YUV422P10;
3131 return AV_PIX_FMT_YUV420P10;
3135 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3136 return AV_PIX_FMT_GBRP12;
3138 return AV_PIX_FMT_YUV444P12;
3139 } else if (CHROMA422(h))
3140 return AV_PIX_FMT_YUV422P12;
3142 return AV_PIX_FMT_YUV420P12;
3146 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3147 return AV_PIX_FMT_GBRP14;
3149 return AV_PIX_FMT_YUV444P14;
3150 } else if (CHROMA422(h))
3151 return AV_PIX_FMT_YUV422P14;
3153 return AV_PIX_FMT_YUV420P14;
3157 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3158 av_log(h->avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
3159 return AV_PIX_FMT_GBR24P;
3160 } else if (h->avctx->colorspace == AVCOL_SPC_YCGCO) {
3161 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
3163 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
3164 : AV_PIX_FMT_YUV444P;
3165 } else if (CHROMA422(h)) {
3166 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
3167 : AV_PIX_FMT_YUV422P;
3170 const enum AVPixelFormat * fmt = h->avctx->codec->pix_fmts ?
3171 h->avctx->codec->pix_fmts :
3172 h->avctx->color_range == AVCOL_RANGE_JPEG ?
3173 h264_hwaccel_pixfmt_list_jpeg_420 :
3174 h264_hwaccel_pixfmt_list_420;
3176 for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++)
3177 if (fmt[i] == h->avctx->pix_fmt && !force_callback)
3179 return ff_thread_get_format(h->avctx, fmt);
3183 av_log(h->avctx, AV_LOG_ERROR,
3184 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3185 return AVERROR_INVALIDDATA;
3189 /* export coded and cropped frame dimensions to AVCodecContext */
3190 static int init_dimensions(H264Context *h)
3192 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
3193 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
3194 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
3195 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
3197 /* handle container cropping */
3199 FFALIGN(h->avctx->width, 16) == h->width &&
3200 FFALIGN(h->avctx->height, 16) == h->height) {
3201 width = h->avctx->width;
3202 height = h->avctx->height;
3205 if (width <= 0 || height <= 0) {
3206 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
3208 if (h->avctx->err_recognition & AV_EF_EXPLODE)
3209 return AVERROR_INVALIDDATA;
3211 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
3212 h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
3219 h->avctx->coded_width = h->width;
3220 h->avctx->coded_height = h->height;
3221 h->avctx->width = width;
3222 h->avctx->height = height;
3227 static int h264_slice_header_init(H264Context *h, int reinit)
3229 int nb_slices = (HAVE_THREADS &&
3230 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
3231 h->avctx->thread_count : 1;
3234 h->avctx->sample_aspect_ratio = h->sps.sar;
3235 av_assert0(h->avctx->sample_aspect_ratio.den);
3236 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
3237 &h->chroma_x_shift, &h->chroma_y_shift);
3239 if (h->sps.timing_info_present_flag) {
3240 int64_t den = h->sps.time_scale;
3241 if (h->x264_build < 44U)
3243 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
3244 h->sps.num_units_in_tick, den, 1 << 30);
3247 h->avctx->hwaccel = ff_find_hwaccel(h->avctx->codec->id, h->avctx->pix_fmt);
3252 h->prev_interlaced_frame = 1;
3254 init_scan_tables(h);
3255 ret = ff_h264_alloc_tables(h);
3257 av_log(h->avctx, AV_LOG_ERROR,
3258 "Could not allocate memory for h264\n");
3262 if (nb_slices > MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
3265 max_slices = FFMIN(MAX_THREADS, h->mb_height);
3267 max_slices = MAX_THREADS;
3268 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
3269 " reducing to %d\n", nb_slices, max_slices);
3270 nb_slices = max_slices;
3272 h->slice_context_count = nb_slices;
3274 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
3275 ret = context_init(h);
3277 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
3281 for (i = 1; i < h->slice_context_count; i++) {
3283 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
3284 c->avctx = h->avctx;
3285 if (CONFIG_ERROR_RESILIENCE) {
3289 c->h264dsp = h->h264dsp;
3290 c->h264qpel = h->h264qpel;
3291 c->h264chroma = h->h264chroma;
3294 c->pixel_shift = h->pixel_shift;
3295 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
3296 c->width = h->width;
3297 c->height = h->height;
3298 c->linesize = h->linesize;
3299 c->uvlinesize = h->uvlinesize;
3300 c->chroma_x_shift = h->chroma_x_shift;
3301 c->chroma_y_shift = h->chroma_y_shift;
3302 c->qscale = h->qscale;
3303 c->droppable = h->droppable;
3304 c->data_partitioning = h->data_partitioning;
3305 c->low_delay = h->low_delay;
3306 c->mb_width = h->mb_width;
3307 c->mb_height = h->mb_height;
3308 c->mb_stride = h->mb_stride;
3309 c->mb_num = h->mb_num;
3310 c->flags = h->flags;
3311 c->workaround_bugs = h->workaround_bugs;
3312 c->pict_type = h->pict_type;
3314 init_scan_tables(c);
3315 clone_tables(c, h, i);
3316 c->context_initialized = 1;
3319 for (i = 0; i < h->slice_context_count; i++)
3320 if ((ret = context_init(h->thread_context[i])) < 0) {
3321 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
3326 h->context_initialized = 1;
3331 int ff_set_ref_count(H264Context *h)
3333 int num_ref_idx_active_override_flag;
3335 // set defaults, might be overridden a few lines later
3336 h->ref_count[0] = h->pps.ref_count[0];
3337 h->ref_count[1] = h->pps.ref_count[1];
3339 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3341 max[0] = max[1] = h->picture_structure == PICT_FRAME ? 15 : 31;
3343 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3344 h->direct_spatial_mv_pred = get_bits1(&h->gb);
3345 num_ref_idx_active_override_flag = get_bits1(&h->gb);
3347 if (num_ref_idx_active_override_flag) {
3348 h->ref_count[0] = get_ue_golomb(&h->gb) + 1;
3349 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3350 h->ref_count[1] = get_ue_golomb(&h->gb) + 1;
3352 // full range is spec-ok in this case, even for frames
3353 h->ref_count[1] = 1;
3356 if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
3357 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]);
3358 h->ref_count[0] = h->ref_count[1] = 0;
3359 return AVERROR_INVALIDDATA;
3362 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3368 h->ref_count[0] = h->ref_count[1] = 0;
3374 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
3377 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
3378 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
3379 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
3386 * Decode a slice header.
3387 * This will also call ff_MPV_common_init() and frame_start() as needed.
3389 * @param h h264context
3390 * @param h0 h264 master context (differs from 'h' when doing sliced based
3391 * parallel decoding)
3393 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3395 static int decode_slice_header(H264Context *h, H264Context *h0)
3397 unsigned int first_mb_in_slice;
3398 unsigned int pps_id;
3400 unsigned int slice_type, tmp, i, j;
3401 int last_pic_structure, last_pic_droppable;
3403 int needs_reinit = 0;
3404 int field_pic_flag, bottom_field_flag;
3406 h->me.qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
3407 h->me.qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
3409 first_mb_in_slice = get_ue_golomb_long(&h->gb);
3411 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
3412 if (h0->current_slice && FIELD_PICTURE(h)) {
3416 h0->current_slice = 0;
3417 if (!h0->first_field) {
3418 if (h->cur_pic_ptr && !h->droppable) {
3419 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
3420 h->picture_structure == PICT_BOTTOM_FIELD);
3422 h->cur_pic_ptr = NULL;
3426 slice_type = get_ue_golomb_31(&h->gb);
3427 if (slice_type > 9) {
3428 av_log(h->avctx, AV_LOG_ERROR,
3429 "slice type too large (%d) at %d %d\n",
3430 slice_type, h->mb_x, h->mb_y);
3431 return AVERROR_INVALIDDATA;
3433 if (slice_type > 4) {
3435 h->slice_type_fixed = 1;
3437 h->slice_type_fixed = 0;
3439 slice_type = golomb_to_pict_type[slice_type];
3440 h->slice_type = slice_type;
3441 h->slice_type_nos = slice_type & 3;
3443 // to make a few old functions happy, it's wrong though
3444 h->pict_type = h->slice_type;
3446 pps_id = get_ue_golomb(&h->gb);
3447 if (pps_id >= MAX_PPS_COUNT) {
3448 av_log(h->avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
3449 return AVERROR_INVALIDDATA;
3451 if (!h0->pps_buffers[pps_id]) {
3452 av_log(h->avctx, AV_LOG_ERROR,
3453 "non-existing PPS %u referenced\n",
3455 return AVERROR_INVALIDDATA;
3457 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
3458 av_log(h->avctx, AV_LOG_ERROR,
3459 "PPS change from %d to %d forbidden\n",
3460 h0->au_pps_id, pps_id);
3461 return AVERROR_INVALIDDATA;
3463 h->pps = *h0->pps_buffers[pps_id];
3465 if (!h0->sps_buffers[h->pps.sps_id]) {
3466 av_log(h->avctx, AV_LOG_ERROR,
3467 "non-existing SPS %u referenced\n",
3469 return AVERROR_INVALIDDATA;
3472 if (h->pps.sps_id != h->current_sps_id ||
3473 h0->sps_buffers[h->pps.sps_id]->new) {
3475 h->sps = *h0->sps_buffers[h->pps.sps_id];
3477 if (h->mb_width != h->sps.mb_width ||
3478 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
3479 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3480 h->cur_chroma_format_idc != h->sps.chroma_format_idc
3484 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
3485 h->chroma_format_idc != h->sps.chroma_format_idc) {
3486 h->bit_depth_luma = h->sps.bit_depth_luma;
3487 h->chroma_format_idc = h->sps.chroma_format_idc;
3490 if ((ret = h264_set_parameter_from_sps(h)) < 0)
3494 h->avctx->profile = ff_h264_get_profile(&h->sps);
3495 h->avctx->level = h->sps.level_idc;
3496 h->avctx->refs = h->sps.ref_frame_count;
3498 must_reinit = (h->context_initialized &&
3499 ( 16*h->sps.mb_width != h->avctx->coded_width
3500 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
3501 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
3502 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
3503 || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)
3504 || h->mb_width != h->sps.mb_width
3505 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
3507 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
3510 h->mb_width = h->sps.mb_width;
3511 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3512 h->mb_num = h->mb_width * h->mb_height;
3513 h->mb_stride = h->mb_width + 1;
3515 h->b_stride = h->mb_width * 4;
3517 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
3519 h->width = 16 * h->mb_width;
3520 h->height = 16 * h->mb_height;
3522 ret = init_dimensions(h);
3526 if (h->sps.video_signal_type_present_flag) {
3527 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
3529 if (h->sps.colour_description_present_flag) {
3530 if (h->avctx->colorspace != h->sps.colorspace)
3532 h->avctx->color_primaries = h->sps.color_primaries;
3533 h->avctx->color_trc = h->sps.color_trc;
3534 h->avctx->colorspace = h->sps.colorspace;
3538 if (h->context_initialized &&
3539 (h->width != h->avctx->coded_width ||
3540 h->height != h->avctx->coded_height ||
3544 av_log(h->avctx, AV_LOG_ERROR, "changing width/height on "
3545 "slice %d\n", h0->current_slice + 1);
3546 return AVERROR_INVALIDDATA;
3551 if ((ret = get_pixel_format(h, 1)) < 0)
3553 h->avctx->pix_fmt = ret;
3555 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
3556 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
3558 if ((ret = h264_slice_header_init(h, 1)) < 0) {
3559 av_log(h->avctx, AV_LOG_ERROR,
3560 "h264_slice_header_init() failed\n");
3564 if (!h->context_initialized) {
3566 av_log(h->avctx, AV_LOG_ERROR,
3567 "Cannot (re-)initialize context during parallel decoding.\n");
3568 return AVERROR_PATCHWELCOME;
3571 if ((ret = get_pixel_format(h, 1)) < 0)
3573 h->avctx->pix_fmt = ret;
3575 if ((ret = h264_slice_header_init(h, 0)) < 0) {
3576 av_log(h->avctx, AV_LOG_ERROR,
3577 "h264_slice_header_init() failed\n");
3582 if (h == h0 && h->dequant_coeff_pps != pps_id) {
3583 h->dequant_coeff_pps = pps_id;
3584 init_dequant_tables(h);
3587 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
3590 h->mb_aff_frame = 0;
3591 last_pic_structure = h0->picture_structure;
3592 last_pic_droppable = h0->droppable;
3593 h->droppable = h->nal_ref_idc == 0;
3594 if (h->sps.frame_mbs_only_flag) {
3595 h->picture_structure = PICT_FRAME;
3597 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
3598 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
3601 field_pic_flag = get_bits1(&h->gb);
3602 if (field_pic_flag) {
3603 bottom_field_flag = get_bits1(&h->gb);
3604 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
3606 h->picture_structure = PICT_FRAME;
3607 h->mb_aff_frame = h->sps.mb_aff;
3610 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
3612 if (h0->current_slice != 0) {
3613 if (last_pic_structure != h->picture_structure ||
3614 last_pic_droppable != h->droppable) {
3615 av_log(h->avctx, AV_LOG_ERROR,
3616 "Changing field mode (%d -> %d) between slices is not allowed\n",
3617 last_pic_structure, h->picture_structure);
3618 h->picture_structure = last_pic_structure;
3619 h->droppable = last_pic_droppable;
3620 return AVERROR_INVALIDDATA;
3621 } else if (!h0->cur_pic_ptr) {
3622 av_log(h->avctx, AV_LOG_ERROR,
3623 "unset cur_pic_ptr on %d. slice\n",
3624 h0->current_slice + 1);
3625 return AVERROR_INVALIDDATA;
3628 /* Shorten frame num gaps so we don't have to allocate reference
3629 * frames just to throw them away */
3630 if (h->frame_num != h->prev_frame_num) {
3631 int unwrap_prev_frame_num = h->prev_frame_num;
3632 int max_frame_num = 1 << h->sps.log2_max_frame_num;
3634 if (unwrap_prev_frame_num > h->frame_num)
3635 unwrap_prev_frame_num -= max_frame_num;
3637 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
3638 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
3639 if (unwrap_prev_frame_num < 0)
3640 unwrap_prev_frame_num += max_frame_num;
3642 h->prev_frame_num = unwrap_prev_frame_num;
3646 /* See if we have a decoded first field looking for a pair...
3647 * Here, we're using that to see if we should mark previously
3648 * decode frames as "finished".
3649 * We have to do that before the "dummy" in-between frame allocation,
3650 * since that can modify h->cur_pic_ptr. */
3651 if (h0->first_field) {
3652 assert(h0->cur_pic_ptr);
3653 assert(h0->cur_pic_ptr->f.data[0]);
3654 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
3656 /* Mark old field/frame as completed */
3657 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
3658 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3659 last_pic_structure == PICT_BOTTOM_FIELD);
3662 /* figure out if we have a complementary field pair */
3663 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
3664 /* Previous field is unmatched. Don't display it, but let it
3665 * remain for reference if marked as such. */
3666 if (last_pic_structure != PICT_FRAME) {
3667 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3668 last_pic_structure == PICT_TOP_FIELD);
3671 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
3672 /* This and previous field were reference, but had
3673 * different frame_nums. Consider this field first in
3674 * pair. Throw away previous field except for reference
3676 if (last_pic_structure != PICT_FRAME) {
3677 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3678 last_pic_structure == PICT_TOP_FIELD);
3681 /* Second field in complementary pair */
3682 if (!((last_pic_structure == PICT_TOP_FIELD &&
3683 h->picture_structure == PICT_BOTTOM_FIELD) ||
3684 (last_pic_structure == PICT_BOTTOM_FIELD &&
3685 h->picture_structure == PICT_TOP_FIELD))) {
3686 av_log(h->avctx, AV_LOG_ERROR,
3687 "Invalid field mode combination %d/%d\n",
3688 last_pic_structure, h->picture_structure);
3689 h->picture_structure = last_pic_structure;
3690 h->droppable = last_pic_droppable;
3691 return AVERROR_INVALIDDATA;
3692 } else if (last_pic_droppable != h->droppable) {
3693 avpriv_request_sample(h->avctx,
3694 "Found reference and non-reference fields in the same frame, which");
3695 h->picture_structure = last_pic_structure;
3696 h->droppable = last_pic_droppable;
3697 return AVERROR_PATCHWELCOME;
3703 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
3704 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
3705 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
3706 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
3707 h->frame_num, h->prev_frame_num);
3708 if (!h->sps.gaps_in_frame_num_allowed_flag)
3709 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
3710 h->last_pocs[i] = INT_MIN;
3711 ret = h264_frame_start(h);
3714 h->prev_frame_num++;
3715 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
3716 h->cur_pic_ptr->frame_num = h->prev_frame_num;
3717 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
3718 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
3719 ret = ff_generate_sliding_window_mmcos(h, 1);
3720 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
3722 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
3723 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
3725 /* Error concealment: If a ref is missing, copy the previous ref
3727 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
3728 * many assumptions about there being no actual duplicates.
3729 * FIXME: This does not copy padding for out-of-frame motion
3730 * vectors. Given we are concealing a lost frame, this probably
3731 * is not noticeable by comparison, but it should be fixed. */
3732 if (h->short_ref_count) {
3734 av_image_copy(h->short_ref[0]->f.data,
3735 h->short_ref[0]->f.linesize,
3736 (const uint8_t **)prev->f.data,
3741 h->short_ref[0]->poc = prev->poc + 2;
3743 h->short_ref[0]->frame_num = h->prev_frame_num;
3747 /* See if we have a decoded first field looking for a pair...
3748 * We're using that to see whether to continue decoding in that
3749 * frame, or to allocate a new one. */
3750 if (h0->first_field) {
3751 assert(h0->cur_pic_ptr);
3752 assert(h0->cur_pic_ptr->f.data[0]);
3753 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
3755 /* figure out if we have a complementary field pair */
3756 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
3757 /* Previous field is unmatched. Don't display it, but let it
3758 * remain for reference if marked as such. */
3759 h0->cur_pic_ptr = NULL;
3760 h0->first_field = FIELD_PICTURE(h);
3762 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
3763 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3764 h0->picture_structure==PICT_BOTTOM_FIELD);
3765 /* This and the previous field had different frame_nums.
3766 * Consider this field first in pair. Throw away previous
3767 * one except for reference purposes. */
3768 h0->first_field = 1;
3769 h0->cur_pic_ptr = NULL;
3771 /* Second field in complementary pair */
3772 h0->first_field = 0;
3776 /* Frame or first field in a potentially complementary pair */
3777 h0->first_field = FIELD_PICTURE(h);
3780 if (!FIELD_PICTURE(h) || h0->first_field) {
3781 if (h264_frame_start(h) < 0) {
3782 h0->first_field = 0;
3783 return AVERROR_INVALIDDATA;
3786 release_unused_pictures(h, 0);
3788 /* Some macroblocks can be accessed before they're available in case
3789 * of lost slices, MBAFF or threading. */
3790 if (FIELD_PICTURE(h)) {
3791 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
3792 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
3794 memset(h->slice_table, -1,
3795 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
3797 h0->last_slice_type = -1;
3799 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
3802 /* can't be in alloc_tables because linesize isn't known there.
3803 * FIXME: redo bipred weight to not require extra buffer? */
3804 for (i = 0; i < h->slice_context_count; i++)
3805 if (h->thread_context[i]) {
3806 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
3811 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
3813 av_assert1(h->mb_num == h->mb_width * h->mb_height);
3814 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
3815 first_mb_in_slice >= h->mb_num) {
3816 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3817 return AVERROR_INVALIDDATA;
3819 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
3820 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
3821 FIELD_OR_MBAFF_PICTURE(h);
3822 if (h->picture_structure == PICT_BOTTOM_FIELD)
3823 h->resync_mb_y = h->mb_y = h->mb_y + 1;
3824 av_assert1(h->mb_y < h->mb_height);
3826 if (h->picture_structure == PICT_FRAME) {
3827 h->curr_pic_num = h->frame_num;
3828 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
3830 h->curr_pic_num = 2 * h->frame_num + 1;
3831 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
3834 if (h->nal_unit_type == NAL_IDR_SLICE)
3835 get_ue_golomb(&h->gb); /* idr_pic_id */
3837 if (h->sps.poc_type == 0) {
3838 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
3840 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
3841 h->delta_poc_bottom = get_se_golomb(&h->gb);
3844 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
3845 h->delta_poc[0] = get_se_golomb(&h->gb);
3847 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
3848 h->delta_poc[1] = get_se_golomb(&h->gb);
3851 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
3853 if (h->pps.redundant_pic_cnt_present)
3854 h->redundant_pic_count = get_ue_golomb(&h->gb);
3856 ret = ff_set_ref_count(h);
3860 if (slice_type != AV_PICTURE_TYPE_I &&
3861 (h0->current_slice == 0 ||
3862 slice_type != h0->last_slice_type ||
3863 memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
3865 ff_h264_fill_default_ref_list(h);
3868 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3869 ret = ff_h264_decode_ref_pic_list_reordering(h);
3871 h->ref_count[1] = h->ref_count[0] = 0;
3876 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
3877 (h->pps.weighted_bipred_idc == 1 &&
3878 h->slice_type_nos == AV_PICTURE_TYPE_B))
3879 ff_pred_weight_table(h);
3880 else if (h->pps.weighted_bipred_idc == 2 &&
3881 h->slice_type_nos == AV_PICTURE_TYPE_B) {
3882 implicit_weight_table(h, -1);
3885 for (i = 0; i < 2; i++) {
3886 h->luma_weight_flag[i] = 0;
3887 h->chroma_weight_flag[i] = 0;
3891 // If frame-mt is enabled, only update mmco tables for the first slice
3892 // in a field. Subsequent slices can temporarily clobber h->mmco_index
3893 // or h->mmco, which will cause ref list mix-ups and decoding errors
3894 // further down the line. This may break decoding if the first slice is
3895 // corrupt, thus we only do this if frame-mt is enabled.
3896 if (h->nal_ref_idc) {
3897 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
3898 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
3899 h0->current_slice == 0);
3900 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
3901 return AVERROR_INVALIDDATA;
3904 if (FRAME_MBAFF(h)) {
3905 ff_h264_fill_mbaff_ref_list(h);
3907 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
3908 implicit_weight_table(h, 0);
3909 implicit_weight_table(h, 1);
3913 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3914 ff_h264_direct_dist_scale_factor(h);
3915 ff_h264_direct_ref_list_init(h);
3917 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3918 tmp = get_ue_golomb_31(&h->gb);
3920 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3921 return AVERROR_INVALIDDATA;
3923 h->cabac_init_idc = tmp;
3926 h->last_qscale_diff = 0;
3927 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
3928 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3929 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3930 return AVERROR_INVALIDDATA;
3933 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
3934 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
3935 // FIXME qscale / qp ... stuff
3936 if (h->slice_type == AV_PICTURE_TYPE_SP)
3937 get_bits1(&h->gb); /* sp_for_switch_flag */
3938 if (h->slice_type == AV_PICTURE_TYPE_SP ||
3939 h->slice_type == AV_PICTURE_TYPE_SI)
3940 get_se_golomb(&h->gb); /* slice_qs_delta */
3942 h->deblocking_filter = 1;
3943 h->slice_alpha_c0_offset = 52;
3944 h->slice_beta_offset = 52;
3945 if (h->pps.deblocking_filter_parameters_present) {
3946 tmp = get_ue_golomb_31(&h->gb);
3948 av_log(h->avctx, AV_LOG_ERROR,
3949 "deblocking_filter_idc %u out of range\n", tmp);
3950 return AVERROR_INVALIDDATA;
3952 h->deblocking_filter = tmp;
3953 if (h->deblocking_filter < 2)
3954 h->deblocking_filter ^= 1; // 1<->0
3956 if (h->deblocking_filter) {
3957 h->slice_alpha_c0_offset += get_se_golomb(&h->gb) << 1;
3958 h->slice_beta_offset += get_se_golomb(&h->gb) << 1;
3959 if (h->slice_alpha_c0_offset < 52 - 12 || h->slice_alpha_c0_offset > 52 + 12 ||
3960 h->slice_beta_offset < 52 - 12 || h->slice_beta_offset > 52 + 12) {
3961 av_log(h->avctx, AV_LOG_ERROR,
3962 "deblocking filter parameters %d %d out of range\n",
3963 h->slice_alpha_c0_offset, h->slice_beta_offset);
3964 return AVERROR_INVALIDDATA;
3969 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3970 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3971 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3972 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
3973 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3974 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3975 h->nal_ref_idc == 0))
3976 h->deblocking_filter = 0;
3978 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3979 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
3980 /* Cheat slightly for speed:
3981 * Do not bother to deblock across slices. */
3982 h->deblocking_filter = 2;
3984 h0->max_contexts = 1;
3985 if (!h0->single_decode_warning) {
3986 av_log(h->avctx, AV_LOG_INFO,
3987 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3988 h0->single_decode_warning = 1;
3991 av_log(h->avctx, AV_LOG_ERROR,
3992 "Deblocking switched inside frame.\n");
3997 h->qp_thresh = 15 + 52 -
3998 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
4000 h->pps.chroma_qp_index_offset[0],
4001 h->pps.chroma_qp_index_offset[1]) +
4002 6 * (h->sps.bit_depth_luma - 8);
4004 h0->last_slice_type = slice_type;
4005 memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
4006 h->slice_num = ++h0->current_slice;
4009 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
4010 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
4011 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
4012 && h->slice_num >= MAX_SLICES) {
4013 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
4014 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);
4017 for (j = 0; j < 2; j++) {
4019 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
4020 for (i = 0; i < 16; i++) {
4022 if (j < h->list_count && i < h->ref_count[j] &&
4023 h->ref_list[j][i].f.buf[0]) {
4025 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
4026 for (k = 0; k < h->short_ref_count; k++)
4027 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
4031 for (k = 0; k < h->long_ref_count; k++)
4032 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
4033 id_list[i] = h->short_ref_count + k;
4041 for (i = 0; i < 16; i++)
4042 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
4044 ref2frm[18 + 1] = -1;
4045 for (i = 16; i < 48; i++)
4046 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
4047 (h->ref_list[j][i].reference & 3);
4050 if (h->ref_count[0]) h->er.last_pic = &h->ref_list[0][0];
4051 if (h->ref_count[1]) h->er.next_pic = &h->ref_list[1][0];
4052 h->er.ref_count = h->ref_count[0];
4053 h0->au_pps_id = pps_id;
4055 h0->sps_buffers[h->pps.sps_id]->new = 0;
4056 h->current_sps_id = h->pps.sps_id;
4058 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
4059 av_log(h->avctx, AV_LOG_DEBUG,
4060 "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",
4062 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
4064 av_get_picture_type_char(h->slice_type),
4065 h->slice_type_fixed ? " fix" : "",
4066 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
4067 pps_id, h->frame_num,
4068 h->cur_pic_ptr->field_poc[0],
4069 h->cur_pic_ptr->field_poc[1],
4070 h->ref_count[0], h->ref_count[1],
4072 h->deblocking_filter,
4073 h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
4075 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
4076 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
4082 int ff_h264_get_slice_type(const H264Context *h)
4084 switch (h->slice_type) {
4085 case AV_PICTURE_TYPE_P:
4087 case AV_PICTURE_TYPE_B:
4089 case AV_PICTURE_TYPE_I:
4091 case AV_PICTURE_TYPE_SP:
4093 case AV_PICTURE_TYPE_SI:
4096 return AVERROR_INVALIDDATA;
4100 static av_always_inline void fill_filter_caches_inter(H264Context *h,
4101 int mb_type, int top_xy,
4102 int left_xy[LEFT_MBS],
4104 int left_type[LEFT_MBS],
4105 int mb_xy, int list)
4107 int b_stride = h->b_stride;
4108 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
4109 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
4110 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
4111 if (USES_LIST(top_type, list)) {
4112 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
4113 const int b8_xy = 4 * top_xy + 2;
4114 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
4115 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
4116 ref_cache[0 - 1 * 8] =
4117 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
4118 ref_cache[2 - 1 * 8] =
4119 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
4121 AV_ZERO128(mv_dst - 1 * 8);
4122 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4125 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
4126 if (USES_LIST(left_type[LTOP], list)) {
4127 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
4128 const int b8_xy = 4 * left_xy[LTOP] + 1;
4129 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
4130 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
4131 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
4132 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
4133 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
4135 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
4136 ref_cache[-1 + 16] =
4137 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
4139 AV_ZERO32(mv_dst - 1 + 0);
4140 AV_ZERO32(mv_dst - 1 + 8);
4141 AV_ZERO32(mv_dst - 1 + 16);
4142 AV_ZERO32(mv_dst - 1 + 24);
4145 ref_cache[-1 + 16] =
4146 ref_cache[-1 + 24] = LIST_NOT_USED;
4151 if (!USES_LIST(mb_type, list)) {
4152 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
4153 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4154 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4155 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4156 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4161 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
4162 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
4163 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
4164 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
4165 AV_WN32A(&ref_cache[0 * 8], ref01);
4166 AV_WN32A(&ref_cache[1 * 8], ref01);
4167 AV_WN32A(&ref_cache[2 * 8], ref23);
4168 AV_WN32A(&ref_cache[3 * 8], ref23);
4172 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
4173 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
4174 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
4175 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
4176 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
4182 * @return non zero if the loop filter can be skipped
4184 static int fill_filter_caches(H264Context *h, int mb_type)
4186 const int mb_xy = h->mb_xy;
4187 int top_xy, left_xy[LEFT_MBS];
4188 int top_type, left_type[LEFT_MBS];
4192 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
4194 /* Wow, what a mess, why didn't they simplify the interlacing & intra
4195 * stuff, I can't imagine that these complex rules are worth it. */
4197 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
4198 if (FRAME_MBAFF(h)) {
4199 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
4200 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
4202 if (left_mb_field_flag != curr_mb_field_flag)
4203 left_xy[LTOP] -= h->mb_stride;
4205 if (curr_mb_field_flag)
4206 top_xy += h->mb_stride &
4207 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
4208 if (left_mb_field_flag != curr_mb_field_flag)
4209 left_xy[LBOT] += h->mb_stride;
4213 h->top_mb_xy = top_xy;
4214 h->left_mb_xy[LTOP] = left_xy[LTOP];
4215 h->left_mb_xy[LBOT] = left_xy[LBOT];
4217 /* For sufficiently low qp, filtering wouldn't do anything.
4218 * This is a conservative estimate: could also check beta_offset
4219 * and more accurate chroma_qp. */
4220 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
4221 int qp = h->cur_pic.qscale_table[mb_xy];
4222 if (qp <= qp_thresh &&
4223 (left_xy[LTOP] < 0 ||
4224 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
4226 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
4227 if (!FRAME_MBAFF(h))
4229 if ((left_xy[LTOP] < 0 ||
4230 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
4231 (top_xy < h->mb_stride ||
4232 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
4237 top_type = h->cur_pic.mb_type[top_xy];
4238 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
4239 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
4240 if (h->deblocking_filter == 2) {
4241 if (h->slice_table[top_xy] != h->slice_num)
4243 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
4244 left_type[LTOP] = left_type[LBOT] = 0;
4246 if (h->slice_table[top_xy] == 0xFFFF)
4248 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
4249 left_type[LTOP] = left_type[LBOT] = 0;
4251 h->top_type = top_type;
4252 h->left_type[LTOP] = left_type[LTOP];
4253 h->left_type[LBOT] = left_type[LBOT];
4255 if (IS_INTRA(mb_type))
4258 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
4259 top_type, left_type, mb_xy, 0);
4260 if (h->list_count == 2)
4261 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
4262 top_type, left_type, mb_xy, 1);
4264 nnz = h->non_zero_count[mb_xy];
4265 nnz_cache = h->non_zero_count_cache;
4266 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
4267 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
4268 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
4269 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
4270 h->cbp = h->cbp_table[mb_xy];
4273 nnz = h->non_zero_count[top_xy];
4274 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
4277 if (left_type[LTOP]) {
4278 nnz = h->non_zero_count[left_xy[LTOP]];
4279 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
4280 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
4281 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
4282 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
4285 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
4286 * from what the loop filter needs */
4287 if (!CABAC(h) && h->pps.transform_8x8_mode) {
4288 if (IS_8x8DCT(top_type)) {
4289 nnz_cache[4 + 8 * 0] =
4290 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
4291 nnz_cache[6 + 8 * 0] =
4292 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
4294 if (IS_8x8DCT(left_type[LTOP])) {
4295 nnz_cache[3 + 8 * 1] =
4296 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
4298 if (IS_8x8DCT(left_type[LBOT])) {
4299 nnz_cache[3 + 8 * 3] =
4300 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
4303 if (IS_8x8DCT(mb_type)) {
4304 nnz_cache[scan8[0]] =
4305 nnz_cache[scan8[1]] =
4306 nnz_cache[scan8[2]] =
4307 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
4309 nnz_cache[scan8[0 + 4]] =
4310 nnz_cache[scan8[1 + 4]] =
4311 nnz_cache[scan8[2 + 4]] =
4312 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
4314 nnz_cache[scan8[0 + 8]] =
4315 nnz_cache[scan8[1 + 8]] =
4316 nnz_cache[scan8[2 + 8]] =
4317 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
4319 nnz_cache[scan8[0 + 12]] =
4320 nnz_cache[scan8[1 + 12]] =
4321 nnz_cache[scan8[2 + 12]] =
4322 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
4329 static void loop_filter(H264Context *h, int start_x, int end_x)
4331 uint8_t *dest_y, *dest_cb, *dest_cr;
4332 int linesize, uvlinesize, mb_x, mb_y;
4333 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
4334 const int old_slice_type = h->slice_type;
4335 const int pixel_shift = h->pixel_shift;
4336 const int block_h = 16 >> h->chroma_y_shift;
4338 if (h->deblocking_filter) {
4339 for (mb_x = start_x; mb_x < end_x; mb_x++)
4340 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
4342 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
4343 h->slice_num = h->slice_table[mb_xy];
4344 mb_type = h->cur_pic.mb_type[mb_xy];
4345 h->list_count = h->list_counts[mb_xy];
4349 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
4353 dest_y = h->cur_pic.f.data[0] +
4354 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
4355 dest_cb = h->cur_pic.f.data[1] +
4356 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
4357 mb_y * h->uvlinesize * block_h;
4358 dest_cr = h->cur_pic.f.data[2] +
4359 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
4360 mb_y * h->uvlinesize * block_h;
4361 // FIXME simplify above
4364 linesize = h->mb_linesize = h->linesize * 2;
4365 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
4366 if (mb_y & 1) { // FIXME move out of this function?
4367 dest_y -= h->linesize * 15;
4368 dest_cb -= h->uvlinesize * (block_h - 1);
4369 dest_cr -= h->uvlinesize * (block_h - 1);
4372 linesize = h->mb_linesize = h->linesize;
4373 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
4375 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
4377 if (fill_filter_caches(h, mb_type))
4379 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
4380 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
4382 if (FRAME_MBAFF(h)) {
4383 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
4384 linesize, uvlinesize);
4386 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
4387 dest_cr, linesize, uvlinesize);
4391 h->slice_type = old_slice_type;
4393 h->mb_y = end_mb_y - FRAME_MBAFF(h);
4394 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
4395 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
4398 static void predict_field_decoding_flag(H264Context *h)
4400 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
4401 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
4402 h->cur_pic.mb_type[mb_xy - 1] :
4403 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
4404 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
4405 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4409 * Draw edges and report progress for the last MB row.
4411 static void decode_finish_row(H264Context *h)
4413 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
4414 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
4415 int height = 16 << FRAME_MBAFF(h);
4416 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
4418 if (h->deblocking_filter) {
4419 if ((top + height) >= pic_height)
4420 height += deblock_border;
4421 top -= deblock_border;
4424 if (top >= pic_height || (top + height) < 0)
4427 height = FFMIN(height, pic_height - top);
4429 height = top + height;
4433 ff_h264_draw_horiz_band(h, top, height);
4435 if (h->droppable || h->er.error_occurred)
4438 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
4439 h->picture_structure == PICT_BOTTOM_FIELD);
4442 static void er_add_slice(H264Context *h, int startx, int starty,
4443 int endx, int endy, int status)
4445 if (CONFIG_ERROR_RESILIENCE) {
4446 ERContext *er = &h->er;
4448 ff_er_add_slice(er, startx, starty, endx, endy, status);
4452 static int decode_slice(struct AVCodecContext *avctx, void *arg)
4454 H264Context *h = *(void **)arg;
4455 int lf_x_start = h->mb_x;
4457 h->mb_skip_run = -1;
4459 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
4461 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
4462 avctx->codec_id != AV_CODEC_ID_H264 ||
4463 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
4465 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
4466 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
4468 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
4469 prev_status &= ~ VP_START;
4470 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
4471 h->er.error_occurred = 1;
4477 align_get_bits(&h->gb);
4480 ff_init_cabac_decoder(&h->cabac,
4481 h->gb.buffer + get_bits_count(&h->gb) / 8,
4482 (get_bits_left(&h->gb) + 7) / 8);
4484 ff_h264_init_cabac_states(h);
4488 int ret = ff_h264_decode_mb_cabac(h);
4490 // STOP_TIMER("decode_mb_cabac")
4493 ff_h264_hl_decode_mb(h);
4495 // FIXME optimal? or let mb_decode decode 16x32 ?
4496 if (ret >= 0 && FRAME_MBAFF(h)) {
4499 ret = ff_h264_decode_mb_cabac(h);
4502 ff_h264_hl_decode_mb(h);
4505 eos = get_cabac_terminate(&h->cabac);
4507 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
4508 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
4509 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
4510 h->mb_y, ER_MB_END);
4511 if (h->mb_x >= lf_x_start)
4512 loop_filter(h, lf_x_start, h->mb_x + 1);
4515 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
4516 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
4517 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
4518 av_log(h->avctx, AV_LOG_ERROR,
4519 "error while decoding MB %d %d, bytestream (%td)\n",
4521 h->cabac.bytestream_end - h->cabac.bytestream);
4522 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4523 h->mb_y, ER_MB_ERROR);
4524 return AVERROR_INVALIDDATA;
4527 if (++h->mb_x >= h->mb_width) {
4528 loop_filter(h, lf_x_start, h->mb_x);
4529 h->mb_x = lf_x_start = 0;
4530 decode_finish_row(h);
4532 if (FIELD_OR_MBAFF_PICTURE(h)) {
4534 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
4535 predict_field_decoding_flag(h);
4539 if (eos || h->mb_y >= h->mb_height) {
4540 tprintf(h->avctx, "slice end %d %d\n",
4541 get_bits_count(&h->gb), h->gb.size_in_bits);
4542 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
4543 h->mb_y, ER_MB_END);
4544 if (h->mb_x > lf_x_start)
4545 loop_filter(h, lf_x_start, h->mb_x);
4551 int ret = ff_h264_decode_mb_cavlc(h);
4554 ff_h264_hl_decode_mb(h);
4556 // FIXME optimal? or let mb_decode decode 16x32 ?
4557 if (ret >= 0 && FRAME_MBAFF(h)) {
4559 ret = ff_h264_decode_mb_cavlc(h);
4562 ff_h264_hl_decode_mb(h);
4567 av_log(h->avctx, AV_LOG_ERROR,
4568 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
4569 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4570 h->mb_y, ER_MB_ERROR);
4574 if (++h->mb_x >= h->mb_width) {
4575 loop_filter(h, lf_x_start, h->mb_x);
4576 h->mb_x = lf_x_start = 0;
4577 decode_finish_row(h);
4579 if (FIELD_OR_MBAFF_PICTURE(h)) {
4581 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
4582 predict_field_decoding_flag(h);
4584 if (h->mb_y >= h->mb_height) {
4585 tprintf(h->avctx, "slice end %d %d\n",
4586 get_bits_count(&h->gb), h->gb.size_in_bits);
4588 if ( get_bits_left(&h->gb) == 0
4589 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
4590 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4591 h->mb_x - 1, h->mb_y,
4596 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4600 return AVERROR_INVALIDDATA;
4605 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
4606 tprintf(h->avctx, "slice end %d %d\n",
4607 get_bits_count(&h->gb), h->gb.size_in_bits);
4609 if (get_bits_left(&h->gb) == 0) {
4610 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4611 h->mb_x - 1, h->mb_y,
4613 if (h->mb_x > lf_x_start)
4614 loop_filter(h, lf_x_start, h->mb_x);
4618 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4619 h->mb_y, ER_MB_ERROR);
4621 return AVERROR_INVALIDDATA;
4629 * Call decode_slice() for each context.
4631 * @param h h264 master context
4632 * @param context_count number of contexts to execute
4634 static int execute_decode_slices(H264Context *h, int context_count)
4636 AVCodecContext *const avctx = h->avctx;
4640 av_assert0(h->mb_y < h->mb_height);
4642 if (h->avctx->hwaccel ||
4643 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4645 if (context_count == 1) {
4646 return decode_slice(avctx, &h);
4648 av_assert0(context_count > 0);
4649 for (i = 1; i < context_count; i++) {
4650 hx = h->thread_context[i];
4651 if (CONFIG_ERROR_RESILIENCE) {
4652 hx->er.error_count = 0;
4654 hx->x264_build = h->x264_build;
4657 avctx->execute(avctx, decode_slice, h->thread_context,
4658 NULL, context_count, sizeof(void *));
4660 /* pull back stuff from slices to master context */
4661 hx = h->thread_context[context_count - 1];
4664 h->droppable = hx->droppable;
4665 h->picture_structure = hx->picture_structure;
4666 if (CONFIG_ERROR_RESILIENCE) {
4667 for (i = 1; i < context_count; i++)
4668 h->er.error_count += h->thread_context[i]->er.error_count;
4675 static const uint8_t start_code[] = { 0x00, 0x00, 0x01 };
4677 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
4678 int parse_extradata)
4680 AVCodecContext *const avctx = h->avctx;
4681 H264Context *hx; ///< thread context
4685 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
4686 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
4689 int first_slice = 0;
4692 h->nal_unit_type= 0;
4694 if(!h->slice_context_count)
4695 h->slice_context_count= 1;
4696 h->max_contexts = h->slice_context_count;
4697 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS)) {
4698 h->current_slice = 0;
4699 if (!h->first_field)
4700 h->cur_pic_ptr = NULL;
4701 ff_h264_reset_sei(h);
4704 if (h->nal_length_size == 4) {
4705 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
4707 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
4711 for (; pass <= 1; pass++) {
4714 next_avc = h->is_avc ? 0 : buf_size;
4724 if (buf_index >= next_avc) {
4725 if (buf_index >= buf_size - h->nal_length_size)
4728 for (i = 0; i < h->nal_length_size; i++)
4729 nalsize = (nalsize << 8) | buf[buf_index++];
4730 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
4731 av_log(h->avctx, AV_LOG_ERROR,
4732 "AVC: nal size %d\n", nalsize);
4735 next_avc = buf_index + nalsize;
4737 // start code prefix search
4738 for (; buf_index + 3 < next_avc; buf_index++)
4739 // This should always succeed in the first iteration.
4740 if (buf[buf_index] == 0 &&
4741 buf[buf_index + 1] == 0 &&
4742 buf[buf_index + 2] == 1)
4745 if (buf_index + 3 >= buf_size) {
4746 buf_index = buf_size;
4751 if (buf_index >= next_avc)
4755 hx = h->thread_context[context_count];
4757 ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
4758 &consumed, next_avc - buf_index);
4759 if (ptr == NULL || dst_length < 0) {
4763 i = buf_index + consumed;
4764 if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
4765 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
4766 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
4767 h->workaround_bugs |= FF_BUG_TRUNCATED;
4769 if (!(h->workaround_bugs & FF_BUG_TRUNCATED))
4770 while(dst_length > 0 && ptr[dst_length - 1] == 0)
4772 bit_length = !dst_length ? 0
4774 decode_rbsp_trailing(h, ptr + dst_length - 1));
4776 if (h->avctx->debug & FF_DEBUG_STARTCODE)
4777 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);
4779 if (h->is_avc && (nalsize != consumed) && nalsize)
4780 av_log(h->avctx, AV_LOG_DEBUG,
4781 "AVC: Consumed only %d bytes instead of %d\n",
4784 buf_index += consumed;
4788 /* packets can sometimes contain multiple PPS/SPS,
4789 * e.g. two PAFF field pictures in one packet, or a demuxer
4790 * which splits NALs strangely if so, when frame threading we
4791 * can't start the next thread until we've read all of them */
4792 switch (hx->nal_unit_type) {
4795 nals_needed = nal_index;
4800 init_get_bits(&hx->gb, ptr, bit_length);
4801 if (!get_ue_golomb(&hx->gb) || !first_slice)
4802 nals_needed = nal_index;
4804 first_slice = hx->nal_unit_type;
4810 switch (hx->nal_unit_type) {
4814 first_slice = hx->nal_unit_type;
4817 if (avctx->skip_frame >= AVDISCARD_NONREF &&
4818 h->nal_ref_idc == 0 &&
4819 h->nal_unit_type != NAL_SEI)
4823 if ( !(avctx->active_thread_type & FF_THREAD_FRAME)
4824 || nals_needed >= nal_index)
4826 /* Ignore per frame NAL unit type during extradata
4827 * parsing. Decoding slices is not possible in codec init
4829 if (parse_extradata) {
4830 switch (hx->nal_unit_type) {
4836 av_log(h->avctx, AV_LOG_WARNING,
4837 "Ignoring NAL %d in global header/extradata\n",
4839 // fall through to next case
4840 case NAL_AUXILIARY_SLICE:
4841 hx->nal_unit_type = NAL_FF_IGNORE;
4847 switch (hx->nal_unit_type) {
4849 if (first_slice != NAL_IDR_SLICE) {
4850 av_log(h->avctx, AV_LOG_ERROR,
4851 "Invalid mix of idr and non-idr slices\n");
4856 idr(h); // FIXME ensure we don't lose some frames if there is reordering
4858 h->has_recovery_point = 1;
4860 init_get_bits(&hx->gb, ptr, bit_length);
4862 hx->inter_gb_ptr = &hx->gb;
4863 hx->data_partitioning = 0;
4865 if ((err = decode_slice_header(hx, h)))
4868 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
4869 h->valid_recovery_point = 1;
4871 if ( h->sei_recovery_frame_cnt >= 0
4872 && ( h->recovery_frame<0
4873 || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
4874 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
4875 (1 << h->sps.log2_max_frame_num);
4877 if (!h->valid_recovery_point)
4878 h->recovery_frame = h->frame_num;
4881 h->cur_pic_ptr->f.key_frame |=
4882 (hx->nal_unit_type == NAL_IDR_SLICE);
4884 if (h->recovery_frame == h->frame_num) {
4885 h->cur_pic_ptr->sync |= 1;
4886 h->recovery_frame = -1;
4889 h->sync |= !!h->cur_pic_ptr->f.key_frame;
4890 h->sync |= 3*!!(avctx->flags2 & CODEC_FLAG2_SHOW_ALL);
4891 h->cur_pic_ptr->sync |= h->sync;
4893 if (h->current_slice == 1) {
4894 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS))
4895 decode_postinit(h, nal_index >= nals_needed);
4897 if (h->avctx->hwaccel &&
4898 (ret = h->avctx->hwaccel->start_frame(h->avctx, NULL, 0)) < 0)
4900 if (CONFIG_H264_VDPAU_DECODER &&
4901 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4902 ff_vdpau_h264_picture_start(h);
4905 if (hx->redundant_pic_count == 0 &&
4906 (avctx->skip_frame < AVDISCARD_NONREF ||
4908 (avctx->skip_frame < AVDISCARD_BIDIR ||
4909 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4910 (avctx->skip_frame < AVDISCARD_NONKEY ||
4911 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4912 avctx->skip_frame < AVDISCARD_ALL) {
4913 if (avctx->hwaccel) {
4914 ret = avctx->hwaccel->decode_slice(avctx,
4915 &buf[buf_index - consumed],
4919 } else if (CONFIG_H264_VDPAU_DECODER &&
4920 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
4921 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0],
4923 sizeof(start_code));
4924 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0],
4925 &buf[buf_index - consumed],
4932 init_get_bits(&hx->gb, ptr, bit_length);
4934 hx->inter_gb_ptr = NULL;
4936 if ((err = decode_slice_header(hx, h)) < 0)
4939 hx->data_partitioning = 1;
4942 init_get_bits(&hx->intra_gb, ptr, bit_length);
4943 hx->intra_gb_ptr = &hx->intra_gb;
4946 init_get_bits(&hx->inter_gb, ptr, bit_length);
4947 hx->inter_gb_ptr = &hx->inter_gb;
4949 av_log(h->avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
4952 if (hx->redundant_pic_count == 0 &&
4954 hx->data_partitioning &&
4955 h->cur_pic_ptr && h->context_initialized &&
4956 (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4957 (avctx->skip_frame < AVDISCARD_BIDIR ||
4958 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4959 (avctx->skip_frame < AVDISCARD_NONKEY ||
4960 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4961 avctx->skip_frame < AVDISCARD_ALL)
4965 init_get_bits(&h->gb, ptr, bit_length);
4966 ff_h264_decode_sei(h);
4969 init_get_bits(&h->gb, ptr, bit_length);
4970 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? nalsize : 1)) {
4971 av_log(h->avctx, AV_LOG_DEBUG,
4972 "SPS decoding failure, trying again with the complete NAL\n");
4974 av_assert0(next_avc - buf_index + consumed == nalsize);
4975 if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4977 init_get_bits(&h->gb, &buf[buf_index + 1 - consumed],
4978 8*(next_avc - buf_index + consumed - 1));
4979 ff_h264_decode_seq_parameter_set(h);
4984 init_get_bits(&h->gb, ptr, bit_length);
4985 ff_h264_decode_picture_parameter_set(h, bit_length);
4988 case NAL_END_SEQUENCE:
4989 case NAL_END_STREAM:
4990 case NAL_FILLER_DATA:
4992 case NAL_AUXILIARY_SLICE:
4997 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4998 hx->nal_unit_type, bit_length);
5001 if (context_count == h->max_contexts) {
5002 execute_decode_slices(h, context_count);
5007 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
5008 else if (err == 1) {
5009 /* Slice could not be decoded in parallel mode, copy down
5010 * NAL unit stuff to context 0 and restart. Note that
5011 * rbsp_buffer is not transferred, but since we no longer
5012 * run in parallel mode this should not be an issue. */
5013 h->nal_unit_type = hx->nal_unit_type;
5014 h->nal_ref_idc = hx->nal_ref_idc;
5021 execute_decode_slices(h, context_count);
5025 if (h->cur_pic_ptr && !h->droppable) {
5026 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
5027 h->picture_structure == PICT_BOTTOM_FIELD);
5030 return (ret < 0) ? ret : buf_index;
5034 * Return the number of bytes consumed for building the current frame.
5036 static int get_consumed_bytes(int pos, int buf_size)
5039 pos = 1; // avoid infinite loops (i doubt that is needed but ...)
5040 if (pos + 10 > buf_size)
5041 pos = buf_size; // oops ;)
5046 static int output_frame(H264Context *h, AVFrame *dst, Picture *srcp)
5048 AVFrame *src = &srcp->f;
5049 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(src->format);
5051 int ret = av_frame_ref(dst, src);
5055 av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(h), 0);
5060 for (i = 0; i < desc->nb_components; i++) {
5061 int hshift = (i > 0) ? desc->log2_chroma_w : 0;
5062 int vshift = (i > 0) ? desc->log2_chroma_h : 0;
5063 int off = ((srcp->crop_left >> hshift) << h->pixel_shift) +
5064 (srcp->crop_top >> vshift) * dst->linesize[i];
5065 dst->data[i] += off;
5070 static int decode_frame(AVCodecContext *avctx, void *data,
5071 int *got_frame, AVPacket *avpkt)
5073 const uint8_t *buf = avpkt->data;
5074 int buf_size = avpkt->size;
5075 H264Context *h = avctx->priv_data;
5076 AVFrame *pict = data;
5082 h->flags = avctx->flags;
5084 /* end of stream, output what is still in the buffers */
5085 if (buf_size == 0) {
5088 h->cur_pic_ptr = NULL;
5091 // FIXME factorize this with the output code below
5092 out = h->delayed_pic[0];
5095 h->delayed_pic[i] &&
5096 !h->delayed_pic[i]->f.key_frame &&
5097 !h->delayed_pic[i]->mmco_reset;
5099 if (h->delayed_pic[i]->poc < out->poc) {
5100 out = h->delayed_pic[i];
5104 for (i = out_idx; h->delayed_pic[i]; i++)
5105 h->delayed_pic[i] = h->delayed_pic[i + 1];
5108 out->reference &= ~DELAYED_PIC_REF;
5109 ret = output_frame(h, pict, out);
5117 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
5118 int cnt= buf[5]&0x1f;
5119 const uint8_t *p= buf+6;
5121 int nalsize= AV_RB16(p) + 2;
5122 if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
5130 int nalsize= AV_RB16(p) + 2;
5131 if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
5136 return ff_h264_decode_extradata(h, buf, buf_size);
5140 buf_index = decode_nal_units(h, buf, buf_size, 0);
5142 return AVERROR_INVALIDDATA;
5144 if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
5145 av_assert0(buf_index <= buf_size);
5149 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
5150 if (avctx->skip_frame >= AVDISCARD_NONREF ||
5151 buf_size >= 4 && !memcmp("Q264", buf, 4))
5153 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
5154 return AVERROR_INVALIDDATA;
5157 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) ||
5158 (h->mb_y >= h->mb_height && h->mb_height)) {
5159 if (avctx->flags2 & CODEC_FLAG2_CHUNKS)
5160 decode_postinit(h, 1);
5164 /* Wait for second field. */
5166 if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
5167 ret = output_frame(h, pict, h->next_output_pic);
5171 if (CONFIG_MPEGVIDEO) {
5172 ff_print_debug_info2(h->avctx, h->next_output_pic, pict, h->er.mbskip_table,
5174 h->mb_width, h->mb_height, h->mb_stride, 1);
5179 assert(pict->data[0] || !*got_frame);
5181 return get_consumed_bytes(buf_index, buf_size);
5184 av_cold void ff_h264_free_context(H264Context *h)
5188 free_tables(h, 1); // FIXME cleanup init stuff perhaps
5190 for (i = 0; i < MAX_SPS_COUNT; i++)
5191 av_freep(h->sps_buffers + i);
5193 for (i = 0; i < MAX_PPS_COUNT; i++)
5194 av_freep(h->pps_buffers + i);
5197 static av_cold int h264_decode_end(AVCodecContext *avctx)
5199 H264Context *h = avctx->priv_data;
5201 ff_h264_remove_all_refs(h);
5202 ff_h264_free_context(h);
5204 unref_picture(h, &h->cur_pic);
5209 static const AVProfile profiles[] = {
5210 { FF_PROFILE_H264_BASELINE, "Baseline" },
5211 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
5212 { FF_PROFILE_H264_MAIN, "Main" },
5213 { FF_PROFILE_H264_EXTENDED, "Extended" },
5214 { FF_PROFILE_H264_HIGH, "High" },
5215 { FF_PROFILE_H264_HIGH_10, "High 10" },
5216 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
5217 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
5218 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
5219 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
5220 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
5221 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
5222 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
5223 { FF_PROFILE_UNKNOWN },
5226 static const AVOption h264_options[] = {
5227 {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
5228 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
5232 static const AVClass h264_class = {
5233 .class_name = "H264 Decoder",
5234 .item_name = av_default_item_name,
5235 .option = h264_options,
5236 .version = LIBAVUTIL_VERSION_INT,
5239 static const AVClass h264_vdpau_class = {
5240 .class_name = "H264 VDPAU Decoder",
5241 .item_name = av_default_item_name,
5242 .option = h264_options,
5243 .version = LIBAVUTIL_VERSION_INT,
5246 AVCodec ff_h264_decoder = {
5248 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
5249 .type = AVMEDIA_TYPE_VIDEO,
5250 .id = AV_CODEC_ID_H264,
5251 .priv_data_size = sizeof(H264Context),
5252 .init = ff_h264_decode_init,
5253 .close = h264_decode_end,
5254 .decode = decode_frame,
5255 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
5256 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
5257 CODEC_CAP_FRAME_THREADS,
5259 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
5260 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
5261 .profiles = NULL_IF_CONFIG_SMALL(profiles),
5262 .priv_class = &h264_class,
5265 #if CONFIG_H264_VDPAU_DECODER
5266 AVCodec ff_h264_vdpau_decoder = {
5267 .name = "h264_vdpau",
5268 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
5269 .type = AVMEDIA_TYPE_VIDEO,
5270 .id = AV_CODEC_ID_H264,
5271 .priv_data_size = sizeof(H264Context),
5272 .init = ff_h264_decode_init,
5273 .close = h264_decode_end,
5274 .decode = decode_frame,
5275 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
5277 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
5279 .profiles = NULL_IF_CONFIG_SMALL(profiles),
5280 .priv_class = &h264_vdpau_class,