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 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
55 static const uint8_t rem6[QP_MAX_NUM + 1] = {
56 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
57 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
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,
63 static const uint8_t div6[QP_MAX_NUM + 1] = {
64 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
65 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
66 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
67 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
71 static const uint8_t field_scan[16+1] = {
72 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
73 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
74 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
75 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
78 static const uint8_t field_scan8x8[64+1] = {
79 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
80 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
81 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
82 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
83 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
84 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
85 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
86 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
87 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
88 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
89 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
90 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
91 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
92 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
93 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
94 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
97 static const uint8_t field_scan8x8_cavlc[64+1] = {
98 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
99 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
100 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
101 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
102 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
103 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
104 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
105 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
106 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
107 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
108 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
109 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
110 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
111 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
112 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
113 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
116 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
117 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
118 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
119 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
120 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
121 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
122 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
123 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
124 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
125 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
126 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
127 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
128 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
129 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
130 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
131 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
132 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
133 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
136 static const uint8_t dequant4_coeff_init[6][3] = {
145 static const uint8_t dequant8_coeff_init_scan[16] = {
146 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
149 static const uint8_t dequant8_coeff_init[6][6] = {
150 { 20, 18, 32, 19, 25, 24 },
151 { 22, 19, 35, 21, 28, 26 },
152 { 26, 23, 42, 24, 33, 31 },
153 { 28, 25, 45, 26, 35, 33 },
154 { 32, 28, 51, 30, 40, 38 },
155 { 36, 32, 58, 34, 46, 43 },
158 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
159 #if CONFIG_H264_DXVA2_HWACCEL
160 AV_PIX_FMT_DXVA2_VLD,
162 #if CONFIG_H264_VAAPI_HWACCEL
163 AV_PIX_FMT_VAAPI_VLD,
165 #if CONFIG_H264_VDA_HWACCEL
168 #if CONFIG_H264_VDPAU_HWACCEL
175 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
176 #if CONFIG_H264_DXVA2_HWACCEL
177 AV_PIX_FMT_DXVA2_VLD,
179 #if CONFIG_H264_VAAPI_HWACCEL
180 AV_PIX_FMT_VAAPI_VLD,
182 #if CONFIG_H264_VDA_HWACCEL
185 #if CONFIG_H264_VDPAU_HWACCEL
192 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
194 H264Context *h = avctx->priv_data;
195 return h ? h->sps.num_reorder_frames : 0;
198 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
200 int mb_x, int mb_y, int mb_intra, int mb_skipped)
202 H264Context *h = opaque;
206 h->mb_xy = mb_x + mb_y * h->mb_stride;
207 memset(h->non_zero_count_cache, 0, sizeof(h->non_zero_count_cache));
208 av_assert1(ref >= 0);
209 /* FIXME: It is possible albeit uncommon that slice references
210 * differ between slices. We take the easy approach and ignore
211 * it for now. If this turns out to have any relevance in
212 * practice then correct remapping should be added. */
213 if (ref >= h->ref_count[0])
215 if (!h->ref_list[0][ref].f.data[0]) {
216 av_log(h->avctx, AV_LOG_DEBUG, "Reference not available for error concealing\n");
219 if ((h->ref_list[0][ref].reference&3) != 3) {
220 av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n");
223 fill_rectangle(&h->cur_pic.ref_index[0][4 * h->mb_xy],
225 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
226 fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8,
227 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
229 h->mb_field_decoding_flag = 0;
230 ff_h264_hl_decode_mb(h);
233 void ff_h264_draw_horiz_band(H264Context *h, int y, int height)
235 AVCodecContext *avctx = h->avctx;
236 Picture *cur = &h->cur_pic;
237 Picture *last = h->ref_list[0][0].f.data[0] ? &h->ref_list[0][0] : NULL;
238 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
239 int vshift = desc->log2_chroma_h;
240 const int field_pic = h->picture_structure != PICT_FRAME;
246 height = FFMIN(height, avctx->height - y);
248 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
251 if (avctx->draw_horiz_band) {
253 int offset[AV_NUM_DATA_POINTERS];
256 if (cur->f.pict_type == AV_PICTURE_TYPE_B || h->low_delay ||
257 (avctx->slice_flags & SLICE_FLAG_CODED_ORDER))
264 offset[0] = y * src->linesize[0];
266 offset[2] = (y >> vshift) * src->linesize[1];
267 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
272 avctx->draw_horiz_band(avctx, src, offset,
273 y, h->picture_structure, height);
277 static void unref_picture(H264Context *h, Picture *pic)
279 int off = offsetof(Picture, tf) + sizeof(pic->tf);
285 ff_thread_release_buffer(h->avctx, &pic->tf);
286 av_buffer_unref(&pic->hwaccel_priv_buf);
288 av_buffer_unref(&pic->qscale_table_buf);
289 av_buffer_unref(&pic->mb_type_buf);
290 for (i = 0; i < 2; i++) {
291 av_buffer_unref(&pic->motion_val_buf[i]);
292 av_buffer_unref(&pic->ref_index_buf[i]);
295 memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
298 static void release_unused_pictures(H264Context *h, int remove_current)
302 /* release non reference frames */
303 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
304 if (h->DPB[i].f.data[0] && !h->DPB[i].reference &&
305 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
306 unref_picture(h, &h->DPB[i]);
311 static int ref_picture(H264Context *h, Picture *dst, Picture *src)
315 av_assert0(!dst->f.buf[0]);
316 av_assert0(src->f.buf[0]);
320 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
324 dst->qscale_table_buf = av_buffer_ref(src->qscale_table_buf);
325 dst->mb_type_buf = av_buffer_ref(src->mb_type_buf);
326 if (!dst->qscale_table_buf || !dst->mb_type_buf)
328 dst->qscale_table = src->qscale_table;
329 dst->mb_type = src->mb_type;
331 for (i = 0; i < 2; i++) {
332 dst->motion_val_buf[i] = av_buffer_ref(src->motion_val_buf[i]);
333 dst->ref_index_buf[i] = av_buffer_ref(src->ref_index_buf[i]);
334 if (!dst->motion_val_buf[i] || !dst->ref_index_buf[i])
336 dst->motion_val[i] = src->motion_val[i];
337 dst->ref_index[i] = src->ref_index[i];
340 if (src->hwaccel_picture_private) {
341 dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
342 if (!dst->hwaccel_priv_buf)
344 dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
347 for (i = 0; i < 2; i++)
348 dst->field_poc[i] = src->field_poc[i];
350 memcpy(dst->ref_poc, src->ref_poc, sizeof(src->ref_poc));
351 memcpy(dst->ref_count, src->ref_count, sizeof(src->ref_count));
354 dst->frame_num = src->frame_num;
355 dst->mmco_reset = src->mmco_reset;
356 dst->pic_id = src->pic_id;
357 dst->long_ref = src->long_ref;
358 dst->mbaff = src->mbaff;
359 dst->field_picture = src->field_picture;
360 dst->needs_realloc = src->needs_realloc;
361 dst->reference = src->reference;
362 dst->sync = src->sync;
363 dst->crop = src->crop;
364 dst->crop_left = src->crop_left;
365 dst->crop_top = src->crop_top;
369 unref_picture(h, dst);
373 static int alloc_scratch_buffers(H264Context *h, int linesize)
375 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
377 if (h->bipred_scratchpad)
380 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
381 // edge emu needs blocksize + filter length - 1
382 // (= 21x21 for h264)
383 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
384 h->me.scratchpad = av_mallocz(alloc_size * 2 * 16 * 2);
386 if (!h->bipred_scratchpad || !h->edge_emu_buffer || !h->me.scratchpad) {
387 av_freep(&h->bipred_scratchpad);
388 av_freep(&h->edge_emu_buffer);
389 av_freep(&h->me.scratchpad);
390 return AVERROR(ENOMEM);
393 h->me.temp = h->me.scratchpad;
398 static int init_table_pools(H264Context *h)
400 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
401 const int mb_array_size = h->mb_stride * h->mb_height;
402 const int b4_stride = h->mb_width * 4 + 1;
403 const int b4_array_size = b4_stride * h->mb_height * 4;
405 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
407 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
408 sizeof(uint32_t), av_buffer_allocz);
409 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
410 sizeof(int16_t), av_buffer_allocz);
411 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
413 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
414 !h->ref_index_pool) {
415 av_buffer_pool_uninit(&h->qscale_table_pool);
416 av_buffer_pool_uninit(&h->mb_type_pool);
417 av_buffer_pool_uninit(&h->motion_val_pool);
418 av_buffer_pool_uninit(&h->ref_index_pool);
419 return AVERROR(ENOMEM);
425 static int alloc_picture(H264Context *h, Picture *pic)
429 av_assert0(!pic->f.data[0]);
432 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
433 AV_GET_BUFFER_FLAG_REF : 0);
437 h->linesize = pic->f.linesize[0];
438 h->uvlinesize = pic->f.linesize[1];
439 pic->crop = h->sps.crop;
440 pic->crop_top = h->sps.crop_top;
441 pic->crop_left= h->sps.crop_left;
443 if (h->avctx->hwaccel) {
444 const AVHWAccel *hwaccel = h->avctx->hwaccel;
445 av_assert0(!pic->hwaccel_picture_private);
446 if (hwaccel->priv_data_size) {
447 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size);
448 if (!pic->hwaccel_priv_buf)
449 return AVERROR(ENOMEM);
450 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
454 if (!h->qscale_table_pool) {
455 ret = init_table_pools(h);
460 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
461 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
462 if (!pic->qscale_table_buf || !pic->mb_type_buf)
465 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
466 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
468 for (i = 0; i < 2; i++) {
469 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
470 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
471 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
474 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
475 pic->ref_index[i] = pic->ref_index_buf[i]->data;
480 unref_picture(h, pic);
481 return (ret < 0) ? ret : AVERROR(ENOMEM);
484 static inline int pic_is_unused(H264Context *h, Picture *pic)
486 if (pic->f.data[0] == NULL)
488 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
493 static int find_unused_picture(H264Context *h)
497 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
498 if (pic_is_unused(h, &h->DPB[i]))
501 if (i == MAX_PICTURE_COUNT)
502 return AVERROR_INVALIDDATA;
504 if (h->DPB[i].needs_realloc) {
505 h->DPB[i].needs_realloc = 0;
506 unref_picture(h, &h->DPB[i]);
513 * Check if the top & left blocks are available if needed and
514 * change the dc mode so it only uses the available blocks.
516 int ff_h264_check_intra4x4_pred_mode(H264Context *h)
518 static const int8_t top[12] = {
519 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
521 static const int8_t left[12] = {
522 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
526 if (!(h->top_samples_available & 0x8000)) {
527 for (i = 0; i < 4; i++) {
528 int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
530 av_log(h->avctx, AV_LOG_ERROR,
531 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
532 status, h->mb_x, h->mb_y);
533 return AVERROR_INVALIDDATA;
535 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
540 if ((h->left_samples_available & 0x8888) != 0x8888) {
541 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
542 for (i = 0; i < 4; i++)
543 if (!(h->left_samples_available & mask[i])) {
544 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
546 av_log(h->avctx, AV_LOG_ERROR,
547 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
548 status, h->mb_x, h->mb_y);
549 return AVERROR_INVALIDDATA;
551 h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
557 } // FIXME cleanup like ff_h264_check_intra_pred_mode
560 * Check if the top & left blocks are available if needed and
561 * change the dc mode so it only uses the available blocks.
563 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
565 static const int8_t top[4] = { LEFT_DC_PRED8x8, 1, -1, -1 };
566 static const int8_t left[5] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
569 av_log(h->avctx, AV_LOG_ERROR,
570 "out of range intra chroma pred mode at %d %d\n",
572 return AVERROR_INVALIDDATA;
575 if (!(h->top_samples_available & 0x8000)) {
578 av_log(h->avctx, AV_LOG_ERROR,
579 "top block unavailable for requested intra mode at %d %d\n",
581 return AVERROR_INVALIDDATA;
585 if ((h->left_samples_available & 0x8080) != 0x8080) {
587 if (is_chroma && (h->left_samples_available & 0x8080)) {
588 // mad cow disease mode, aka MBAFF + constrained_intra_pred
589 mode = ALZHEIMER_DC_L0T_PRED8x8 +
590 (!(h->left_samples_available & 0x8000)) +
591 2 * (mode == DC_128_PRED8x8);
594 av_log(h->avctx, AV_LOG_ERROR,
595 "left block unavailable for requested intra mode at %d %d\n",
597 return AVERROR_INVALIDDATA;
604 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
605 int *dst_length, int *consumed, int length)
611 // src[0]&0x80; // forbidden bit
612 h->nal_ref_idc = src[0] >> 5;
613 h->nal_unit_type = src[0] & 0x1F;
618 #define STARTCODE_TEST \
619 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
620 if (src[i + 2] != 3) { \
621 /* startcode, so we must be past the end */ \
627 #if HAVE_FAST_UNALIGNED
628 #define FIND_FIRST_ZERO \
629 if (i > 0 && !src[i]) \
635 for (i = 0; i + 1 < length; i += 9) {
636 if (!((~AV_RN64A(src + i) &
637 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
638 0x8000800080008080ULL))
645 for (i = 0; i + 1 < length; i += 5) {
646 if (!((~AV_RN32A(src + i) &
647 (AV_RN32A(src + i) - 0x01000101U)) &
656 for (i = 0; i + 1 < length; i += 2) {
659 if (i > 0 && src[i - 1] == 0)
665 // use second escape buffer for inter data
666 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
668 si = h->rbsp_buffer_size[bufidx];
669 av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
670 dst = h->rbsp_buffer[bufidx];
675 if(i>=length-1){ //no escaped 0
677 *consumed= length+1; //+1 for the header
678 if(h->avctx->flags2 & CODEC_FLAG2_FAST){
681 memcpy(dst, src, length);
688 while (si + 2 < length) {
689 // remove escapes (very rare 1:2^22)
690 if (src[si + 2] > 3) {
691 dst[di++] = src[si++];
692 dst[di++] = src[si++];
693 } else if (src[si] == 0 && src[si + 1] == 0) {
694 if (src[si + 2] == 3) { // escape
699 } else // next start code
703 dst[di++] = src[si++];
706 dst[di++] = src[si++];
709 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
712 *consumed = si + 1; // +1 for the header
713 /* FIXME store exact number of bits in the getbitcontext
714 * (it is needed for decoding) */
719 * Identify the exact end of the bitstream
720 * @return the length of the trailing, or 0 if damaged
722 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
727 tprintf(h->avctx, "rbsp trailing %X\n", v);
729 for (r = 1; r < 9; r++) {
737 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
738 int height, int y_offset, int list)
740 int raw_my = h->mv_cache[list][scan8[n]][1];
741 int filter_height_down = (raw_my & 3) ? 3 : 0;
742 int full_my = (raw_my >> 2) + y_offset;
743 int bottom = full_my + filter_height_down + height;
745 av_assert2(height >= 0);
747 return FFMAX(0, bottom);
750 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
751 int height, int y_offset, int list0,
752 int list1, int *nrefs)
756 y_offset += 16 * (h->mb_y >> MB_FIELD(h));
759 int ref_n = h->ref_cache[0][scan8[n]];
760 Picture *ref = &h->ref_list[0][ref_n];
762 // Error resilience puts the current picture in the ref list.
763 // Don't try to wait on these as it will cause a deadlock.
764 // Fields can wait on each other, though.
765 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
766 (ref->reference & 3) != h->picture_structure) {
767 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
768 if (refs[0][ref_n] < 0)
770 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
775 int ref_n = h->ref_cache[1][scan8[n]];
776 Picture *ref = &h->ref_list[1][ref_n];
778 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
779 (ref->reference & 3) != h->picture_structure) {
780 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
781 if (refs[1][ref_n] < 0)
783 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
789 * Wait until all reference frames are available for MC operations.
791 * @param h the H264 context
793 static void await_references(H264Context *h)
795 const int mb_xy = h->mb_xy;
796 const int mb_type = h->cur_pic.mb_type[mb_xy];
798 int nrefs[2] = { 0 };
801 memset(refs, -1, sizeof(refs));
803 if (IS_16X16(mb_type)) {
804 get_lowest_part_y(h, refs, 0, 16, 0,
805 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
806 } else if (IS_16X8(mb_type)) {
807 get_lowest_part_y(h, refs, 0, 8, 0,
808 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
809 get_lowest_part_y(h, refs, 8, 8, 8,
810 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
811 } else if (IS_8X16(mb_type)) {
812 get_lowest_part_y(h, refs, 0, 16, 0,
813 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
814 get_lowest_part_y(h, refs, 4, 16, 0,
815 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
819 av_assert2(IS_8X8(mb_type));
821 for (i = 0; i < 4; i++) {
822 const int sub_mb_type = h->sub_mb_type[i];
824 int y_offset = (i & 2) << 2;
826 if (IS_SUB_8X8(sub_mb_type)) {
827 get_lowest_part_y(h, refs, n, 8, y_offset,
828 IS_DIR(sub_mb_type, 0, 0),
829 IS_DIR(sub_mb_type, 0, 1),
831 } else if (IS_SUB_8X4(sub_mb_type)) {
832 get_lowest_part_y(h, refs, n, 4, y_offset,
833 IS_DIR(sub_mb_type, 0, 0),
834 IS_DIR(sub_mb_type, 0, 1),
836 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
837 IS_DIR(sub_mb_type, 0, 0),
838 IS_DIR(sub_mb_type, 0, 1),
840 } else if (IS_SUB_4X8(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 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
846 IS_DIR(sub_mb_type, 0, 0),
847 IS_DIR(sub_mb_type, 0, 1),
851 av_assert2(IS_SUB_4X4(sub_mb_type));
852 for (j = 0; j < 4; j++) {
853 int sub_y_offset = y_offset + 2 * (j & 2);
854 get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
855 IS_DIR(sub_mb_type, 0, 0),
856 IS_DIR(sub_mb_type, 0, 1),
863 for (list = h->list_count - 1; list >= 0; list--)
864 for (ref = 0; ref < 48 && nrefs[list]; ref++) {
865 int row = refs[list][ref];
867 Picture *ref_pic = &h->ref_list[list][ref];
868 int ref_field = ref_pic->reference - 1;
869 int ref_field_picture = ref_pic->field_picture;
870 int pic_height = 16 * h->mb_height >> ref_field_picture;
875 if (!FIELD_PICTURE(h) && ref_field_picture) { // frame referencing two fields
876 ff_thread_await_progress(&ref_pic->tf,
877 FFMIN((row >> 1) - !(row & 1),
880 ff_thread_await_progress(&ref_pic->tf,
881 FFMIN((row >> 1), pic_height - 1),
883 } else if (FIELD_PICTURE(h) && !ref_field_picture) { // field referencing one field of a frame
884 ff_thread_await_progress(&ref_pic->tf,
885 FFMIN(row * 2 + ref_field,
888 } else if (FIELD_PICTURE(h)) {
889 ff_thread_await_progress(&ref_pic->tf,
890 FFMIN(row, pic_height - 1),
893 ff_thread_await_progress(&ref_pic->tf,
894 FFMIN(row, pic_height - 1),
901 static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
902 int n, int square, int height,
904 uint8_t *dest_y, uint8_t *dest_cb,
906 int src_x_offset, int src_y_offset,
907 qpel_mc_func *qpix_op,
908 h264_chroma_mc_func chroma_op,
909 int pixel_shift, int chroma_idc)
911 const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
912 int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
913 const int luma_xy = (mx & 3) + ((my & 3) << 2);
914 int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
915 uint8_t *src_y = pic->f.data[0] + offset;
916 uint8_t *src_cb, *src_cr;
918 int extra_height = 0;
920 const int full_mx = mx >> 2;
921 const int full_my = my >> 2;
922 const int pic_width = 16 * h->mb_width;
923 const int pic_height = 16 * h->mb_height >> MB_FIELD(h);
931 if (full_mx < 0 - extra_width ||
932 full_my < 0 - extra_height ||
933 full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
934 full_my + 16 /*FIXME*/ > pic_height + extra_height) {
935 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
936 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
938 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
939 full_my - 2, pic_width, pic_height);
940 src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
944 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
946 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
948 if (CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY)
951 if (chroma_idc == 3 /* yuv444 */) {
952 src_cb = pic->f.data[1] + offset;
954 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
955 src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
957 16 + 5, 16 + 5 /*FIXME*/,
958 full_mx - 2, full_my - 2,
959 pic_width, pic_height);
960 src_cb = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
962 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
964 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
966 src_cr = pic->f.data[2] + offset;
968 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
969 src_cr - (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_cr = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
976 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
978 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
982 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
983 if (chroma_idc == 1 /* yuv420 */ && MB_FIELD(h)) {
984 // chroma offset when predicting from a field of opposite parity
985 my += 2 * ((h->mb_y & 1) - (pic->reference - 1));
986 emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
989 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
990 (my >> ysh) * h->mb_uvlinesize;
991 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
992 (my >> ysh) * h->mb_uvlinesize;
995 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->mb_uvlinesize,
996 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
997 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
998 src_cb = h->edge_emu_buffer;
1000 chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
1001 height >> (chroma_idc == 1 /* yuv420 */),
1002 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
1005 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->mb_uvlinesize,
1006 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
1007 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
1008 src_cr = h->edge_emu_buffer;
1010 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
1011 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
1014 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
1015 int height, int delta,
1016 uint8_t *dest_y, uint8_t *dest_cb,
1018 int x_offset, int y_offset,
1019 qpel_mc_func *qpix_put,
1020 h264_chroma_mc_func chroma_put,
1021 qpel_mc_func *qpix_avg,
1022 h264_chroma_mc_func chroma_avg,
1023 int list0, int list1,
1024 int pixel_shift, int chroma_idc)
1026 qpel_mc_func *qpix_op = qpix_put;
1027 h264_chroma_mc_func chroma_op = chroma_put;
1029 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1030 if (chroma_idc == 3 /* yuv444 */) {
1031 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1032 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1033 } else if (chroma_idc == 2 /* yuv422 */) {
1034 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1035 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1036 } else { /* yuv420 */
1037 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1038 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1040 x_offset += 8 * h->mb_x;
1041 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
1044 Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
1045 mc_dir_part(h, ref, n, square, height, delta, 0,
1046 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1047 qpix_op, chroma_op, pixel_shift, chroma_idc);
1050 chroma_op = chroma_avg;
1054 Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
1055 mc_dir_part(h, ref, n, square, height, delta, 1,
1056 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1057 qpix_op, chroma_op, pixel_shift, chroma_idc);
1061 static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
1062 int height, int delta,
1063 uint8_t *dest_y, uint8_t *dest_cb,
1065 int x_offset, int y_offset,
1066 qpel_mc_func *qpix_put,
1067 h264_chroma_mc_func chroma_put,
1068 h264_weight_func luma_weight_op,
1069 h264_weight_func chroma_weight_op,
1070 h264_biweight_func luma_weight_avg,
1071 h264_biweight_func chroma_weight_avg,
1072 int list0, int list1,
1073 int pixel_shift, int chroma_idc)
1077 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1078 if (chroma_idc == 3 /* yuv444 */) {
1079 chroma_height = height;
1080 chroma_weight_avg = luma_weight_avg;
1081 chroma_weight_op = luma_weight_op;
1082 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1083 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1084 } else if (chroma_idc == 2 /* yuv422 */) {
1085 chroma_height = height;
1086 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1087 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1088 } else { /* yuv420 */
1089 chroma_height = height >> 1;
1090 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1091 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1093 x_offset += 8 * h->mb_x;
1094 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
1096 if (list0 && list1) {
1097 /* don't optimize for luma-only case, since B-frames usually
1098 * use implicit weights => chroma too. */
1099 uint8_t *tmp_cb = h->bipred_scratchpad;
1100 uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
1101 uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
1102 int refn0 = h->ref_cache[0][scan8[n]];
1103 int refn1 = h->ref_cache[1][scan8[n]];
1105 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
1106 dest_y, dest_cb, dest_cr,
1107 x_offset, y_offset, qpix_put, chroma_put,
1108 pixel_shift, chroma_idc);
1109 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
1110 tmp_y, tmp_cb, tmp_cr,
1111 x_offset, y_offset, qpix_put, chroma_put,
1112 pixel_shift, chroma_idc);
1114 if (h->use_weight == 2) {
1115 int weight0 = h->implicit_weight[refn0][refn1][h->mb_y & 1];
1116 int weight1 = 64 - weight0;
1117 luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
1118 height, 5, weight0, weight1, 0);
1119 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
1120 chroma_height, 5, weight0, weight1, 0);
1121 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
1122 chroma_height, 5, weight0, weight1, 0);
1124 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
1125 h->luma_log2_weight_denom,
1126 h->luma_weight[refn0][0][0],
1127 h->luma_weight[refn1][1][0],
1128 h->luma_weight[refn0][0][1] +
1129 h->luma_weight[refn1][1][1]);
1130 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
1131 h->chroma_log2_weight_denom,
1132 h->chroma_weight[refn0][0][0][0],
1133 h->chroma_weight[refn1][1][0][0],
1134 h->chroma_weight[refn0][0][0][1] +
1135 h->chroma_weight[refn1][1][0][1]);
1136 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
1137 h->chroma_log2_weight_denom,
1138 h->chroma_weight[refn0][0][1][0],
1139 h->chroma_weight[refn1][1][1][0],
1140 h->chroma_weight[refn0][0][1][1] +
1141 h->chroma_weight[refn1][1][1][1]);
1144 int list = list1 ? 1 : 0;
1145 int refn = h->ref_cache[list][scan8[n]];
1146 Picture *ref = &h->ref_list[list][refn];
1147 mc_dir_part(h, ref, n, square, height, delta, list,
1148 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1149 qpix_put, chroma_put, pixel_shift, chroma_idc);
1151 luma_weight_op(dest_y, h->mb_linesize, height,
1152 h->luma_log2_weight_denom,
1153 h->luma_weight[refn][list][0],
1154 h->luma_weight[refn][list][1]);
1155 if (h->use_weight_chroma) {
1156 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
1157 h->chroma_log2_weight_denom,
1158 h->chroma_weight[refn][list][0][0],
1159 h->chroma_weight[refn][list][0][1]);
1160 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
1161 h->chroma_log2_weight_denom,
1162 h->chroma_weight[refn][list][1][0],
1163 h->chroma_weight[refn][list][1][1]);
1168 static av_always_inline void prefetch_motion(H264Context *h, int list,
1169 int pixel_shift, int chroma_idc)
1171 /* fetch pixels for estimated mv 4 macroblocks ahead
1172 * optimized for 64byte cache lines */
1173 const int refn = h->ref_cache[list][scan8[0]];
1175 const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * h->mb_x + 8;
1176 const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * h->mb_y;
1177 uint8_t **src = h->ref_list[list][refn].f.data;
1178 int off = (mx << pixel_shift) +
1179 (my + (h->mb_x & 3) * 4) * h->mb_linesize +
1180 (64 << pixel_shift);
1181 h->vdsp.prefetch(src[0] + off, h->linesize, 4);
1182 if (chroma_idc == 3 /* yuv444 */) {
1183 h->vdsp.prefetch(src[1] + off, h->linesize, 4);
1184 h->vdsp.prefetch(src[2] + off, h->linesize, 4);
1186 off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (h->mb_x&7))*h->uvlinesize;
1187 h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
1192 static void free_tables(H264Context *h, int free_rbsp)
1197 av_freep(&h->intra4x4_pred_mode);
1198 av_freep(&h->chroma_pred_mode_table);
1199 av_freep(&h->cbp_table);
1200 av_freep(&h->mvd_table[0]);
1201 av_freep(&h->mvd_table[1]);
1202 av_freep(&h->direct_table);
1203 av_freep(&h->non_zero_count);
1204 av_freep(&h->slice_table_base);
1205 h->slice_table = NULL;
1206 av_freep(&h->list_counts);
1208 av_freep(&h->mb2b_xy);
1209 av_freep(&h->mb2br_xy);
1211 for (i = 0; i < 3; i++)
1212 av_freep(&h->visualization_buffer[i]);
1214 av_buffer_pool_uninit(&h->qscale_table_pool);
1215 av_buffer_pool_uninit(&h->mb_type_pool);
1216 av_buffer_pool_uninit(&h->motion_val_pool);
1217 av_buffer_pool_uninit(&h->ref_index_pool);
1219 if (free_rbsp && h->DPB) {
1220 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1221 unref_picture(h, &h->DPB[i]);
1223 } else if (h->DPB) {
1224 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1225 h->DPB[i].needs_realloc = 1;
1228 h->cur_pic_ptr = NULL;
1230 for (i = 0; i < MAX_THREADS; i++) {
1231 hx = h->thread_context[i];
1234 av_freep(&hx->top_borders[1]);
1235 av_freep(&hx->top_borders[0]);
1236 av_freep(&hx->bipred_scratchpad);
1237 av_freep(&hx->edge_emu_buffer);
1238 av_freep(&hx->dc_val_base);
1239 av_freep(&hx->me.scratchpad);
1240 av_freep(&hx->er.mb_index2xy);
1241 av_freep(&hx->er.error_status_table);
1242 av_freep(&hx->er.er_temp_buffer);
1243 av_freep(&hx->er.mbintra_table);
1244 av_freep(&hx->er.mbskip_table);
1247 av_freep(&hx->rbsp_buffer[1]);
1248 av_freep(&hx->rbsp_buffer[0]);
1249 hx->rbsp_buffer_size[0] = 0;
1250 hx->rbsp_buffer_size[1] = 0;
1253 av_freep(&h->thread_context[i]);
1257 static void init_dequant8_coeff_table(H264Context *h)
1260 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1262 for (i = 0; i < 6; i++) {
1263 h->dequant8_coeff[i] = h->dequant8_buffer[i];
1264 for (j = 0; j < i; j++)
1265 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
1266 64 * sizeof(uint8_t))) {
1267 h->dequant8_coeff[i] = h->dequant8_buffer[j];
1273 for (q = 0; q < max_qp + 1; q++) {
1274 int shift = div6[q];
1276 for (x = 0; x < 64; x++)
1277 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
1278 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
1279 h->pps.scaling_matrix8[i][x]) << shift;
1284 static void init_dequant4_coeff_table(H264Context *h)
1287 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1288 for (i = 0; i < 6; i++) {
1289 h->dequant4_coeff[i] = h->dequant4_buffer[i];
1290 for (j = 0; j < i; j++)
1291 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
1292 16 * sizeof(uint8_t))) {
1293 h->dequant4_coeff[i] = h->dequant4_buffer[j];
1299 for (q = 0; q < max_qp + 1; q++) {
1300 int shift = div6[q] + 2;
1302 for (x = 0; x < 16; x++)
1303 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
1304 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
1305 h->pps.scaling_matrix4[i][x]) << shift;
1310 static void init_dequant_tables(H264Context *h)
1313 init_dequant4_coeff_table(h);
1314 if (h->pps.transform_8x8_mode)
1315 init_dequant8_coeff_table(h);
1316 if (h->sps.transform_bypass) {
1317 for (i = 0; i < 6; i++)
1318 for (x = 0; x < 16; x++)
1319 h->dequant4_coeff[i][0][x] = 1 << 6;
1320 if (h->pps.transform_8x8_mode)
1321 for (i = 0; i < 6; i++)
1322 for (x = 0; x < 64; x++)
1323 h->dequant8_coeff[i][0][x] = 1 << 6;
1327 int ff_h264_alloc_tables(H264Context *h)
1329 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
1330 const int row_mb_num = 2*h->mb_stride*FFMAX(h->avctx->thread_count, 1);
1333 FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
1334 row_mb_num * 8 * sizeof(uint8_t), fail)
1335 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
1336 big_mb_num * 48 * sizeof(uint8_t), fail)
1337 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
1338 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
1339 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
1340 big_mb_num * sizeof(uint16_t), fail)
1341 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
1342 big_mb_num * sizeof(uint8_t), fail)
1343 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0],
1344 16 * row_mb_num * sizeof(uint8_t), fail);
1345 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1],
1346 16 * row_mb_num * sizeof(uint8_t), fail);
1347 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
1348 4 * big_mb_num * sizeof(uint8_t), fail);
1349 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
1350 big_mb_num * sizeof(uint8_t), fail)
1352 memset(h->slice_table_base, -1,
1353 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
1354 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
1356 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
1357 big_mb_num * sizeof(uint32_t), fail);
1358 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
1359 big_mb_num * sizeof(uint32_t), fail);
1360 for (y = 0; y < h->mb_height; y++)
1361 for (x = 0; x < h->mb_width; x++) {
1362 const int mb_xy = x + y * h->mb_stride;
1363 const int b_xy = 4 * x + 4 * y * h->b_stride;
1365 h->mb2b_xy[mb_xy] = b_xy;
1366 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
1369 if (!h->dequant4_coeff[0])
1370 init_dequant_tables(h);
1373 h->DPB = av_mallocz_array(MAX_PICTURE_COUNT, sizeof(*h->DPB));
1375 return AVERROR(ENOMEM);
1376 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1377 avcodec_get_frame_defaults(&h->DPB[i].f);
1378 avcodec_get_frame_defaults(&h->cur_pic.f);
1385 return AVERROR(ENOMEM);
1389 * Mimic alloc_tables(), but for every context thread.
1391 static void clone_tables(H264Context *dst, H264Context *src, int i)
1393 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
1394 dst->non_zero_count = src->non_zero_count;
1395 dst->slice_table = src->slice_table;
1396 dst->cbp_table = src->cbp_table;
1397 dst->mb2b_xy = src->mb2b_xy;
1398 dst->mb2br_xy = src->mb2br_xy;
1399 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
1400 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
1401 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
1402 dst->direct_table = src->direct_table;
1403 dst->list_counts = src->list_counts;
1404 dst->DPB = src->DPB;
1405 dst->cur_pic_ptr = src->cur_pic_ptr;
1406 dst->cur_pic = src->cur_pic;
1407 dst->bipred_scratchpad = NULL;
1408 dst->edge_emu_buffer = NULL;
1409 dst->me.scratchpad = NULL;
1410 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
1411 src->sps.chroma_format_idc);
1416 * Allocate buffers which are not shared amongst multiple threads.
1418 static int context_init(H264Context *h)
1420 ERContext *er = &h->er;
1421 int mb_array_size = h->mb_height * h->mb_stride;
1422 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
1423 int c_size = h->mb_stride * (h->mb_height + 1);
1424 int yc_size = y_size + 2 * c_size;
1427 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[0],
1428 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1429 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[1],
1430 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1432 h->ref_cache[0][scan8[5] + 1] =
1433 h->ref_cache[0][scan8[7] + 1] =
1434 h->ref_cache[0][scan8[13] + 1] =
1435 h->ref_cache[1][scan8[5] + 1] =
1436 h->ref_cache[1][scan8[7] + 1] =
1437 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
1439 if (CONFIG_ERROR_RESILIENCE) {
1441 er->avctx = h->avctx;
1443 er->decode_mb = h264_er_decode_mb;
1445 er->quarter_sample = 1;
1447 er->mb_num = h->mb_num;
1448 er->mb_width = h->mb_width;
1449 er->mb_height = h->mb_height;
1450 er->mb_stride = h->mb_stride;
1451 er->b8_stride = h->mb_width * 2 + 1;
1453 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy, (h->mb_num + 1) * sizeof(int),
1454 fail); // error ressilience code looks cleaner with this
1455 for (y = 0; y < h->mb_height; y++)
1456 for (x = 0; x < h->mb_width; x++)
1457 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
1459 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
1460 h->mb_stride + h->mb_width;
1462 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
1463 mb_array_size * sizeof(uint8_t), fail);
1465 FF_ALLOC_OR_GOTO(h->avctx, er->mbintra_table, mb_array_size, fail);
1466 memset(er->mbintra_table, 1, mb_array_size);
1468 FF_ALLOCZ_OR_GOTO(h->avctx, er->mbskip_table, mb_array_size + 2, fail);
1470 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer, h->mb_height * h->mb_stride,
1473 FF_ALLOCZ_OR_GOTO(h->avctx, h->dc_val_base, yc_size * sizeof(int16_t), fail);
1474 er->dc_val[0] = h->dc_val_base + h->mb_width * 2 + 2;
1475 er->dc_val[1] = h->dc_val_base + y_size + h->mb_stride + 1;
1476 er->dc_val[2] = er->dc_val[1] + c_size;
1477 for (i = 0; i < yc_size; i++)
1478 h->dc_val_base[i] = 1024;
1484 return AVERROR(ENOMEM); // free_tables will clean up for us
1487 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1488 int parse_extradata);
1490 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1492 AVCodecContext *avctx = h->avctx;
1495 if (!buf || size <= 0)
1499 int i, cnt, nalsize;
1500 const unsigned char *p = buf;
1505 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1506 return AVERROR_INVALIDDATA;
1508 /* sps and pps in the avcC always have length coded with 2 bytes,
1509 * so put a fake nal_length_size = 2 while parsing them */
1510 h->nal_length_size = 2;
1511 // Decode sps from avcC
1512 cnt = *(p + 5) & 0x1f; // Number of sps
1514 for (i = 0; i < cnt; i++) {
1515 nalsize = AV_RB16(p) + 2;
1516 if(nalsize > size - (p-buf))
1517 return AVERROR_INVALIDDATA;
1518 ret = decode_nal_units(h, p, nalsize, 1);
1520 av_log(avctx, AV_LOG_ERROR,
1521 "Decoding sps %d from avcC failed\n", i);
1526 // Decode pps from avcC
1527 cnt = *(p++); // Number of pps
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 pps %d from avcC failed\n", i);
1540 // Now store right nal length size, that will be used to parse all other nals
1541 h->nal_length_size = (buf[4] & 0x03) + 1;
1544 ret = decode_nal_units(h, buf, size, 1);
1551 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1553 H264Context *h = avctx->priv_data;
1559 h->bit_depth_luma = 8;
1560 h->chroma_format_idc = 1;
1562 h->avctx->bits_per_raw_sample = 8;
1563 h->cur_chroma_format_idc = 1;
1565 ff_h264dsp_init(&h->h264dsp, 8, 1);
1566 av_assert0(h->sps.bit_depth_chroma == 0);
1567 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1568 ff_h264qpel_init(&h->h264qpel, 8);
1569 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, 8, 1);
1571 h->dequant_coeff_pps = -1;
1572 h->current_sps_id = -1;
1574 /* needed so that IDCT permutation is known early */
1575 if (CONFIG_ERROR_RESILIENCE)
1576 ff_dsputil_init(&h->dsp, h->avctx);
1577 ff_videodsp_init(&h->vdsp, 8);
1579 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1580 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1582 h->picture_structure = PICT_FRAME;
1583 h->slice_context_count = 1;
1584 h->workaround_bugs = avctx->workaround_bugs;
1585 h->flags = avctx->flags;
1588 // s->decode_mb = ff_h263_decode_mb;
1589 if (!avctx->has_b_frames)
1592 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1594 ff_h264_decode_init_vlc();
1597 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1599 h->thread_context[0] = h;
1600 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1601 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1602 h->last_pocs[i] = INT_MIN;
1603 h->prev_poc_msb = 1 << 16;
1604 h->prev_frame_num = -1;
1606 h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
1607 ff_h264_reset_sei(h);
1608 if (avctx->codec_id == AV_CODEC_ID_H264) {
1609 if (avctx->ticks_per_frame == 1) {
1610 if(h->avctx->time_base.den < INT_MAX/2) {
1611 h->avctx->time_base.den *= 2;
1613 h->avctx->time_base.num /= 2;
1615 avctx->ticks_per_frame = 2;
1618 if (avctx->extradata_size > 0 && avctx->extradata) {
1619 ret = ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
1621 ff_h264_free_context(h);
1626 if (h->sps.bitstream_restriction_flag &&
1627 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
1628 h->avctx->has_b_frames = h->sps.num_reorder_frames;
1632 ff_init_cabac_states();
1633 avctx->internal->allocate_progress = 1;
1638 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1639 #undef REBASE_PICTURE
1640 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
1641 ((pic && pic >= old_ctx->DPB && \
1642 pic < old_ctx->DPB + MAX_PICTURE_COUNT) ? \
1643 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
1645 static void copy_picture_range(Picture **to, Picture **from, int count,
1646 H264Context *new_base,
1647 H264Context *old_base)
1651 for (i = 0; i < count; i++) {
1652 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1653 IN_RANGE(from[i], old_base->DPB,
1654 sizeof(Picture) * MAX_PICTURE_COUNT) ||
1656 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1660 static void copy_parameter_set(void **to, void **from, int count, int size)
1664 for (i = 0; i < count; i++) {
1665 if (to[i] && !from[i])
1667 else if (from[i] && !to[i])
1668 to[i] = av_malloc(size);
1671 memcpy(to[i], from[i], size);
1675 static int decode_init_thread_copy(AVCodecContext *avctx)
1677 H264Context *h = avctx->priv_data;
1679 if (!avctx->internal->is_copy)
1681 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1682 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1684 h->rbsp_buffer[0] = NULL;
1685 h->rbsp_buffer[1] = NULL;
1686 h->rbsp_buffer_size[0] = 0;
1687 h->rbsp_buffer_size[1] = 0;
1688 h->context_initialized = 0;
1693 #define copy_fields(to, from, start_field, end_field) \
1694 memcpy(&to->start_field, &from->start_field, \
1695 (char *)&to->end_field - (char *)&to->start_field)
1697 static int h264_slice_header_init(H264Context *, int);
1699 static int h264_set_parameter_from_sps(H264Context *h);
1701 static int decode_update_thread_context(AVCodecContext *dst,
1702 const AVCodecContext *src)
1704 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1705 int inited = h->context_initialized, err = 0;
1706 int context_reinitialized = 0;
1713 (h->width != h1->width ||
1714 h->height != h1->height ||
1715 h->mb_width != h1->mb_width ||
1716 h->mb_height != h1->mb_height ||
1717 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
1718 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
1719 h->sps.colorspace != h1->sps.colorspace)) {
1721 /* set bits_per_raw_sample to the previous value. the check for changed
1722 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
1723 * the current value */
1724 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1726 av_freep(&h->bipred_scratchpad);
1728 h->width = h1->width;
1729 h->height = h1->height;
1730 h->mb_height = h1->mb_height;
1731 h->mb_width = h1->mb_width;
1732 h->mb_num = h1->mb_num;
1733 h->mb_stride = h1->mb_stride;
1734 h->b_stride = h1->b_stride;
1736 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1737 MAX_SPS_COUNT, sizeof(SPS));
1739 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1740 MAX_PPS_COUNT, sizeof(PPS));
1743 if ((err = h264_slice_header_init(h, 1)) < 0) {
1744 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
1747 context_reinitialized = 1;
1750 h264_set_parameter_from_sps(h);
1751 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
1752 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
1755 /* update linesize on resize for h264. The h264 decoder doesn't
1756 * necessarily call ff_MPV_frame_start in the new thread */
1757 h->linesize = h1->linesize;
1758 h->uvlinesize = h1->uvlinesize;
1760 /* copy block_offset since frame_start may not be called */
1761 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
1764 for (i = 0; i < MAX_SPS_COUNT; i++)
1765 av_freep(h->sps_buffers + i);
1767 for (i = 0; i < MAX_PPS_COUNT; i++)
1768 av_freep(h->pps_buffers + i);
1770 av_freep(&h->rbsp_buffer[0]);
1771 av_freep(&h->rbsp_buffer[1]);
1772 memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
1773 memcpy(&h->cabac, &h1->cabac,
1774 sizeof(H264Context) - offsetof(H264Context, cabac));
1775 av_assert0((void*)&h->cabac == &h->mb_padding + 1);
1777 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1778 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1780 memset(&h->er, 0, sizeof(h->er));
1781 memset(&h->me, 0, sizeof(h->me));
1782 memset(&h->mb, 0, sizeof(h->mb));
1783 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
1784 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
1788 h->qscale_table_pool = NULL;
1789 h->mb_type_pool = NULL;
1790 h->ref_index_pool = NULL;
1791 h->motion_val_pool = NULL;
1792 for (i = 0; i < 2; i++) {
1793 h->rbsp_buffer[i] = NULL;
1794 h->rbsp_buffer_size[i] = 0;
1797 if (h1->context_initialized) {
1798 h->context_initialized = 0;
1800 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
1801 avcodec_get_frame_defaults(&h->cur_pic.f);
1802 h->cur_pic.tf.f = &h->cur_pic.f;
1804 ret = ff_h264_alloc_tables(h);
1806 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1809 ret = context_init(h);
1811 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
1816 h->bipred_scratchpad = NULL;
1817 h->edge_emu_buffer = NULL;
1819 h->thread_context[0] = h;
1820 h->context_initialized = h1->context_initialized;
1823 h->avctx->coded_height = h1->avctx->coded_height;
1824 h->avctx->coded_width = h1->avctx->coded_width;
1825 h->avctx->width = h1->avctx->width;
1826 h->avctx->height = h1->avctx->height;
1827 h->coded_picture_number = h1->coded_picture_number;
1828 h->first_field = h1->first_field;
1829 h->picture_structure = h1->picture_structure;
1830 h->qscale = h1->qscale;
1831 h->droppable = h1->droppable;
1832 h->data_partitioning = h1->data_partitioning;
1833 h->low_delay = h1->low_delay;
1835 for (i = 0; h->DPB && i < MAX_PICTURE_COUNT; i++) {
1836 unref_picture(h, &h->DPB[i]);
1837 if (h1->DPB[i].f.data[0] &&
1838 (ret = ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
1842 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
1843 unref_picture(h, &h->cur_pic);
1844 if (h1->cur_pic.f.buf[0] && (ret = ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
1847 h->workaround_bugs = h1->workaround_bugs;
1848 h->low_delay = h1->low_delay;
1849 h->droppable = h1->droppable;
1851 // extradata/NAL handling
1852 h->is_avc = h1->is_avc;
1855 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1856 MAX_SPS_COUNT, sizeof(SPS));
1858 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1859 MAX_PPS_COUNT, sizeof(PPS));
1862 // Dequantization matrices
1863 // FIXME these are big - can they be only copied when PPS changes?
1864 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1866 for (i = 0; i < 6; i++)
1867 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1868 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1870 for (i = 0; i < 6; i++)
1871 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1872 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1874 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1877 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1880 copy_fields(h, h1, short_ref, cabac_init_idc);
1882 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
1883 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
1884 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1885 MAX_DELAYED_PIC_COUNT + 2, h, h1);
1889 if (context_reinitialized)
1890 h264_set_parameter_from_sps(h);
1892 if (!h->cur_pic_ptr)
1895 if (!h->droppable) {
1896 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1897 h->prev_poc_msb = h->poc_msb;
1898 h->prev_poc_lsb = h->poc_lsb;
1900 h->prev_frame_num_offset = h->frame_num_offset;
1901 h->prev_frame_num = h->frame_num;
1902 h->outputed_poc = h->next_outputed_poc;
1907 static int h264_frame_start(H264Context *h)
1911 const int pixel_shift = h->pixel_shift;
1913 1<<(h->sps.bit_depth_luma-1),
1914 1<<(h->sps.bit_depth_chroma-1),
1915 1<<(h->sps.bit_depth_chroma-1),
1919 if (!ff_thread_can_start_frame(h->avctx)) {
1920 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1924 release_unused_pictures(h, 1);
1925 h->cur_pic_ptr = NULL;
1927 i = find_unused_picture(h);
1929 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1934 pic->reference = h->droppable ? 0 : h->picture_structure;
1935 pic->f.coded_picture_number = h->coded_picture_number++;
1936 pic->field_picture = h->picture_structure != PICT_FRAME;
1939 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
1941 * See decode_nal_units().
1943 pic->f.key_frame = 0;
1945 pic->mmco_reset = 0;
1947 if ((ret = alloc_picture(h, pic)) < 0)
1949 if(!h->sync && !h->avctx->hwaccel &&
1950 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
1951 avpriv_color_frame(&pic->f, c);
1953 h->cur_pic_ptr = pic;
1954 unref_picture(h, &h->cur_pic);
1955 if ((ret = ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
1958 if (CONFIG_ERROR_RESILIENCE) {
1959 ff_er_frame_start(&h->er);
1961 h->er.next_pic = NULL;
1964 assert(h->linesize && h->uvlinesize);
1966 for (i = 0; i < 16; i++) {
1967 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1968 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1970 for (i = 0; i < 16; i++) {
1971 h->block_offset[16 + i] =
1972 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1973 h->block_offset[48 + 16 + i] =
1974 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1977 // s->decode = (h->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1978 // h->cur_pic.reference /* || h->contains_intra */ || 1;
1980 /* We mark the current picture as non-reference after allocating it, so
1981 * that if we break out due to an error it can be released automatically
1982 * in the next ff_MPV_frame_start().
1984 h->cur_pic_ptr->reference = 0;
1986 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
1988 h->next_output_pic = NULL;
1990 assert(h->cur_pic_ptr->long_ref == 0);
1996 * Run setup operations that must be run after slice header decoding.
1997 * This includes finding the next displayed frame.
1999 * @param h h264 master context
2000 * @param setup_finished enough NALs have been read that we can call
2001 * ff_thread_finish_setup()
2003 static void decode_postinit(H264Context *h, int setup_finished)
2005 Picture *out = h->cur_pic_ptr;
2006 Picture *cur = h->cur_pic_ptr;
2007 int i, pics, out_of_order, out_idx;
2009 h->cur_pic_ptr->f.pict_type = h->pict_type;
2011 if (h->next_output_pic)
2014 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
2015 /* FIXME: if we have two PAFF fields in one packet, we can't start
2016 * the next thread here. If we have one field per packet, we can.
2017 * The check in decode_nal_units() is not good enough to find this
2018 * yet, so we assume the worst for now. */
2019 // if (setup_finished)
2020 // ff_thread_finish_setup(h->avctx);
2024 cur->f.interlaced_frame = 0;
2025 cur->f.repeat_pict = 0;
2027 /* Signal interlacing information externally. */
2028 /* Prioritize picture timing SEI information over used
2029 * decoding process if it exists. */
2031 if (h->sps.pic_struct_present_flag) {
2032 switch (h->sei_pic_struct) {
2033 case SEI_PIC_STRUCT_FRAME:
2035 case SEI_PIC_STRUCT_TOP_FIELD:
2036 case SEI_PIC_STRUCT_BOTTOM_FIELD:
2037 cur->f.interlaced_frame = 1;
2039 case SEI_PIC_STRUCT_TOP_BOTTOM:
2040 case SEI_PIC_STRUCT_BOTTOM_TOP:
2041 if (FIELD_OR_MBAFF_PICTURE(h))
2042 cur->f.interlaced_frame = 1;
2044 // try to flag soft telecine progressive
2045 cur->f.interlaced_frame = h->prev_interlaced_frame;
2047 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
2048 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
2049 /* Signal the possibility of telecined film externally
2050 * (pic_struct 5,6). From these hints, let the applications
2051 * decide if they apply deinterlacing. */
2052 cur->f.repeat_pict = 1;
2054 case SEI_PIC_STRUCT_FRAME_DOUBLING:
2055 cur->f.repeat_pict = 2;
2057 case SEI_PIC_STRUCT_FRAME_TRIPLING:
2058 cur->f.repeat_pict = 4;
2062 if ((h->sei_ct_type & 3) &&
2063 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
2064 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
2066 /* Derive interlacing flag from used decoding process. */
2067 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
2069 h->prev_interlaced_frame = cur->f.interlaced_frame;
2071 if (cur->field_poc[0] != cur->field_poc[1]) {
2072 /* Derive top_field_first from field pocs. */
2073 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
2075 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
2076 /* Use picture timing SEI information. Even if it is a
2077 * information of a past frame, better than nothing. */
2078 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
2079 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2080 cur->f.top_field_first = 1;
2082 cur->f.top_field_first = 0;
2084 /* Most likely progressive */
2085 cur->f.top_field_first = 0;
2089 cur->mmco_reset = h->mmco_reset;
2091 // FIXME do something with unavailable reference frames
2093 /* Sort B-frames into display order */
2095 if (h->sps.bitstream_restriction_flag &&
2096 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
2097 h->avctx->has_b_frames = h->sps.num_reorder_frames;
2101 if (h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
2102 !h->sps.bitstream_restriction_flag) {
2103 h->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
2107 for (i = 0; 1; i++) {
2108 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
2110 h->last_pocs[i-1] = cur->poc;
2113 h->last_pocs[i-1]= h->last_pocs[i];
2116 out_of_order = MAX_DELAYED_PIC_COUNT - i;
2117 if( cur->f.pict_type == AV_PICTURE_TYPE_B
2118 || (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))
2119 out_of_order = FFMAX(out_of_order, 1);
2120 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
2121 av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
2122 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
2123 h->last_pocs[i] = INT_MIN;
2124 h->last_pocs[0] = cur->poc;
2125 cur->mmco_reset = 1;
2126 } else if(h->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
2127 av_log(h->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
2128 h->avctx->has_b_frames = out_of_order;
2133 while (h->delayed_pic[pics])
2136 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
2138 h->delayed_pic[pics++] = cur;
2139 if (cur->reference == 0)
2140 cur->reference = DELAYED_PIC_REF;
2142 out = h->delayed_pic[0];
2144 for (i = 1; h->delayed_pic[i] &&
2145 !h->delayed_pic[i]->f.key_frame &&
2146 !h->delayed_pic[i]->mmco_reset;
2148 if (h->delayed_pic[i]->poc < out->poc) {
2149 out = h->delayed_pic[i];
2152 if (h->avctx->has_b_frames == 0 &&
2153 (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
2154 h->next_outputed_poc = INT_MIN;
2155 out_of_order = out->poc < h->next_outputed_poc;
2157 if (out_of_order || pics > h->avctx->has_b_frames) {
2158 out->reference &= ~DELAYED_PIC_REF;
2159 // for frame threading, the owner must be the second field's thread or
2160 // else the first thread can release the picture and reuse it unsafely
2161 for (i = out_idx; h->delayed_pic[i]; i++)
2162 h->delayed_pic[i] = h->delayed_pic[i + 1];
2164 if (!out_of_order && pics > h->avctx->has_b_frames) {
2165 h->next_output_pic = out;
2166 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
2167 h->next_outputed_poc = INT_MIN;
2169 h->next_outputed_poc = out->poc;
2171 av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
2174 if (h->next_output_pic && h->next_output_pic->sync) {
2178 if (setup_finished && !h->avctx->hwaccel)
2179 ff_thread_finish_setup(h->avctx);
2182 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
2183 uint8_t *src_cb, uint8_t *src_cr,
2184 int linesize, int uvlinesize,
2187 uint8_t *top_border;
2189 const int pixel_shift = h->pixel_shift;
2190 int chroma444 = CHROMA444(h);
2191 int chroma422 = CHROMA422(h);
2194 src_cb -= uvlinesize;
2195 src_cr -= uvlinesize;
2197 if (!simple && FRAME_MBAFF(h)) {
2200 top_border = h->top_borders[0][h->mb_x];
2201 AV_COPY128(top_border, src_y + 15 * linesize);
2203 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
2204 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2207 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2208 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
2209 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
2210 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
2212 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
2213 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
2215 } else if (chroma422) {
2217 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2218 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
2220 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
2221 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
2225 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
2226 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
2228 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
2229 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
2234 } else if (MB_MBAFF(h)) {
2240 top_border = h->top_borders[top_idx][h->mb_x];
2241 /* There are two lines saved, the line above the top macroblock
2242 * of a pair, and the line above the bottom macroblock. */
2243 AV_COPY128(top_border, src_y + 16 * linesize);
2245 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
2247 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2250 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
2251 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
2252 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
2253 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
2255 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
2256 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
2258 } else if (chroma422) {
2260 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
2261 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
2263 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
2264 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
2268 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
2269 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
2271 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
2272 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
2278 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
2279 uint8_t *src_cb, uint8_t *src_cr,
2280 int linesize, int uvlinesize,
2281 int xchg, int chroma444,
2282 int simple, int pixel_shift)
2284 int deblock_topleft;
2287 uint8_t *top_border_m1;
2288 uint8_t *top_border;
2290 if (!simple && FRAME_MBAFF(h)) {
2295 top_idx = MB_MBAFF(h) ? 0 : 1;
2299 if (h->deblocking_filter == 2) {
2300 deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num;
2301 deblock_top = h->top_type;
2303 deblock_topleft = (h->mb_x > 0);
2304 deblock_top = (h->mb_y > !!MB_FIELD(h));
2307 src_y -= linesize + 1 + pixel_shift;
2308 src_cb -= uvlinesize + 1 + pixel_shift;
2309 src_cr -= uvlinesize + 1 + pixel_shift;
2311 top_border_m1 = h->top_borders[top_idx][h->mb_x - 1];
2312 top_border = h->top_borders[top_idx][h->mb_x];
2314 #define XCHG(a, b, xchg) \
2315 if (pixel_shift) { \
2317 AV_SWAP64(b + 0, a + 0); \
2318 AV_SWAP64(b + 8, a + 8); \
2328 if (deblock_topleft) {
2329 XCHG(top_border_m1 + (8 << pixel_shift),
2330 src_y - (7 << pixel_shift), 1);
2332 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
2333 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
2334 if (h->mb_x + 1 < h->mb_width) {
2335 XCHG(h->top_borders[top_idx][h->mb_x + 1],
2336 src_y + (17 << pixel_shift), 1);
2338 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2340 if (deblock_topleft) {
2341 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2342 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2344 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
2345 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
2346 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
2347 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
2348 if (h->mb_x + 1 < h->mb_width) {
2349 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
2350 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
2353 if (deblock_topleft) {
2354 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2355 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2357 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
2358 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
2364 static av_always_inline int dctcoef_get(int16_t *mb, int high_bit_depth,
2367 if (high_bit_depth) {
2368 return AV_RN32A(((int32_t *)mb) + index);
2370 return AV_RN16A(mb + index);
2373 static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth,
2374 int index, int value)
2376 if (high_bit_depth) {
2377 AV_WN32A(((int32_t *)mb) + index, value);
2379 AV_WN16A(mb + index, value);
2382 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
2383 int mb_type, int is_h264,
2385 int transform_bypass,
2389 uint8_t *dest_y, int p)
2391 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2392 void (*idct_dc_add)(uint8_t *dst, int16_t *block, int stride);
2394 int qscale = p == 0 ? h->qscale : h->chroma_qp[p - 1];
2395 block_offset += 16 * p;
2396 if (IS_INTRA4x4(mb_type)) {
2397 if (IS_8x8DCT(mb_type)) {
2398 if (transform_bypass) {
2400 idct_add = h->h264dsp.h264_add_pixels8_clear;
2402 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
2403 idct_add = h->h264dsp.h264_idct8_add;
2405 for (i = 0; i < 16; i += 4) {
2406 uint8_t *const ptr = dest_y + block_offset[i];
2407 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2408 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2409 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2411 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2412 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
2413 (h->topright_samples_available << i) & 0x4000, linesize);
2415 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2416 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2418 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2423 if (transform_bypass) {
2425 idct_add = h->h264dsp.h264_add_pixels4_clear;
2427 idct_dc_add = h->h264dsp.h264_idct_dc_add;
2428 idct_add = h->h264dsp.h264_idct_add;
2430 for (i = 0; i < 16; i++) {
2431 uint8_t *const ptr = dest_y + block_offset[i];
2432 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2434 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2435 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2440 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
2441 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
2442 av_assert2(h->mb_y || linesize <= block_offset[i]);
2443 if (!topright_avail) {
2445 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
2446 topright = (uint8_t *)&tr_high;
2448 tr = ptr[3 - linesize] * 0x01010101u;
2449 topright = (uint8_t *)&tr;
2452 topright = ptr + (4 << pixel_shift) - linesize;
2456 h->hpc.pred4x4[dir](ptr, topright, linesize);
2457 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2460 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2461 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2463 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2464 } else if (CONFIG_SVQ3_DECODER)
2465 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
2471 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
2473 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
2474 if (!transform_bypass)
2475 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
2477 h->dequant4_coeff[p][qscale][0]);
2479 static const uint8_t dc_mapping[16] = {
2480 0 * 16, 1 * 16, 4 * 16, 5 * 16,
2481 2 * 16, 3 * 16, 6 * 16, 7 * 16,
2482 8 * 16, 9 * 16, 12 * 16, 13 * 16,
2483 10 * 16, 11 * 16, 14 * 16, 15 * 16
2485 for (i = 0; i < 16; i++)
2486 dctcoef_set(h->mb + (p * 256 << pixel_shift),
2487 pixel_shift, dc_mapping[i],
2488 dctcoef_get(h->mb_luma_dc[p],
2492 } else if (CONFIG_SVQ3_DECODER)
2493 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
2494 h->mb_luma_dc[p], qscale);
2498 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
2499 int is_h264, int simple,
2500 int transform_bypass,
2504 uint8_t *dest_y, int p)
2506 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2508 block_offset += 16 * p;
2509 if (!IS_INTRA4x4(mb_type)) {
2511 if (IS_INTRA16x16(mb_type)) {
2512 if (transform_bypass) {
2513 if (h->sps.profile_idc == 244 &&
2514 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
2515 h->intra16x16_pred_mode == HOR_PRED8x8)) {
2516 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
2517 h->mb + (p * 256 << pixel_shift),
2520 for (i = 0; i < 16; i++)
2521 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
2522 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2523 h->h264dsp.h264_add_pixels4_clear(dest_y + block_offset[i],
2524 h->mb + (i * 16 + p * 256 << pixel_shift),
2528 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
2529 h->mb + (p * 256 << pixel_shift),
2531 h->non_zero_count_cache + p * 5 * 8);
2533 } else if (h->cbp & 15) {
2534 if (transform_bypass) {
2535 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2536 idct_add = IS_8x8DCT(mb_type) ? h->h264dsp.h264_add_pixels8_clear
2537 : h->h264dsp.h264_add_pixels4_clear;
2538 for (i = 0; i < 16; i += di)
2539 if (h->non_zero_count_cache[scan8[i + p * 16]])
2540 idct_add(dest_y + block_offset[i],
2541 h->mb + (i * 16 + p * 256 << pixel_shift),
2544 if (IS_8x8DCT(mb_type))
2545 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
2546 h->mb + (p * 256 << pixel_shift),
2548 h->non_zero_count_cache + p * 5 * 8);
2550 h->h264dsp.h264_idct_add16(dest_y, block_offset,
2551 h->mb + (p * 256 << pixel_shift),
2553 h->non_zero_count_cache + p * 5 * 8);
2556 } else if (CONFIG_SVQ3_DECODER) {
2557 for (i = 0; i < 16; i++)
2558 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
2559 // FIXME benchmark weird rule, & below
2560 uint8_t *const ptr = dest_y + block_offset[i];
2561 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
2562 h->qscale, IS_INTRA(mb_type) ? 1 : 0);
2570 #include "h264_mb_template.c"
2574 #include "h264_mb_template.c"
2578 #include "h264_mb_template.c"
2580 void ff_h264_hl_decode_mb(H264Context *h)
2582 const int mb_xy = h->mb_xy;
2583 const int mb_type = h->cur_pic.mb_type[mb_xy];
2584 int is_complex = CONFIG_SMALL || h->is_complex ||
2585 IS_INTRA_PCM(mb_type) || h->qscale == 0;
2588 if (is_complex || h->pixel_shift)
2589 hl_decode_mb_444_complex(h);
2591 hl_decode_mb_444_simple_8(h);
2592 } else if (is_complex) {
2593 hl_decode_mb_complex(h);
2594 } else if (h->pixel_shift) {
2595 hl_decode_mb_simple_16(h);
2597 hl_decode_mb_simple_8(h);
2600 static int pred_weight_table(H264Context *h)
2603 int luma_def, chroma_def;
2606 h->use_weight_chroma = 0;
2607 h->luma_log2_weight_denom = get_ue_golomb(&h->gb);
2608 if (h->sps.chroma_format_idc)
2609 h->chroma_log2_weight_denom = get_ue_golomb(&h->gb);
2610 luma_def = 1 << h->luma_log2_weight_denom;
2611 chroma_def = 1 << h->chroma_log2_weight_denom;
2613 for (list = 0; list < 2; list++) {
2614 h->luma_weight_flag[list] = 0;
2615 h->chroma_weight_flag[list] = 0;
2616 for (i = 0; i < h->ref_count[list]; i++) {
2617 int luma_weight_flag, chroma_weight_flag;
2619 luma_weight_flag = get_bits1(&h->gb);
2620 if (luma_weight_flag) {
2621 h->luma_weight[i][list][0] = get_se_golomb(&h->gb);
2622 h->luma_weight[i][list][1] = get_se_golomb(&h->gb);
2623 if (h->luma_weight[i][list][0] != luma_def ||
2624 h->luma_weight[i][list][1] != 0) {
2626 h->luma_weight_flag[list] = 1;
2629 h->luma_weight[i][list][0] = luma_def;
2630 h->luma_weight[i][list][1] = 0;
2633 if (h->sps.chroma_format_idc) {
2634 chroma_weight_flag = get_bits1(&h->gb);
2635 if (chroma_weight_flag) {
2637 for (j = 0; j < 2; j++) {
2638 h->chroma_weight[i][list][j][0] = get_se_golomb(&h->gb);
2639 h->chroma_weight[i][list][j][1] = get_se_golomb(&h->gb);
2640 if (h->chroma_weight[i][list][j][0] != chroma_def ||
2641 h->chroma_weight[i][list][j][1] != 0) {
2642 h->use_weight_chroma = 1;
2643 h->chroma_weight_flag[list] = 1;
2648 for (j = 0; j < 2; j++) {
2649 h->chroma_weight[i][list][j][0] = chroma_def;
2650 h->chroma_weight[i][list][j][1] = 0;
2655 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2658 h->use_weight = h->use_weight || h->use_weight_chroma;
2663 * Initialize implicit_weight table.
2664 * @param field 0/1 initialize the weight for interlaced MBAFF
2665 * -1 initializes the rest
2667 static void implicit_weight_table(H264Context *h, int field)
2669 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2671 for (i = 0; i < 2; i++) {
2672 h->luma_weight_flag[i] = 0;
2673 h->chroma_weight_flag[i] = 0;
2677 if (h->picture_structure == PICT_FRAME) {
2678 cur_poc = h->cur_pic_ptr->poc;
2680 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
2682 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
2683 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2685 h->use_weight_chroma = 0;
2689 ref_count0 = h->ref_count[0];
2690 ref_count1 = h->ref_count[1];
2692 cur_poc = h->cur_pic_ptr->field_poc[field];
2694 ref_count0 = 16 + 2 * h->ref_count[0];
2695 ref_count1 = 16 + 2 * h->ref_count[1];
2699 h->use_weight_chroma = 2;
2700 h->luma_log2_weight_denom = 5;
2701 h->chroma_log2_weight_denom = 5;
2703 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2704 int poc0 = h->ref_list[0][ref0].poc;
2705 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2707 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2708 int poc1 = h->ref_list[1][ref1].poc;
2709 int td = av_clip(poc1 - poc0, -128, 127);
2711 int tb = av_clip(cur_poc - poc0, -128, 127);
2712 int tx = (16384 + (FFABS(td) >> 1)) / td;
2713 int dist_scale_factor = (tb * tx + 32) >> 8;
2714 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2715 w = 64 - dist_scale_factor;
2719 h->implicit_weight[ref0][ref1][0] =
2720 h->implicit_weight[ref0][ref1][1] = w;
2722 h->implicit_weight[ref0][ref1][field] = w;
2729 * instantaneous decoder refresh.
2731 static void idr(H264Context *h)
2734 ff_h264_remove_all_refs(h);
2735 h->prev_frame_num = 0;
2736 h->prev_frame_num_offset = 0;
2737 h->prev_poc_msb = 1<<16;
2738 h->prev_poc_lsb = 0;
2739 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2740 h->last_pocs[i] = INT_MIN;
2743 /* forget old pics after a seek */
2744 static void flush_change(H264Context *h)
2748 h->outputed_poc = h->next_outputed_poc = INT_MIN;
2749 h->prev_interlaced_frame = 1;
2752 h->prev_frame_num = -1;
2753 if (h->cur_pic_ptr) {
2754 h->cur_pic_ptr->reference = 0;
2755 for (j=i=0; h->delayed_pic[i]; i++)
2756 if (h->delayed_pic[i] != h->cur_pic_ptr)
2757 h->delayed_pic[j++] = h->delayed_pic[i];
2758 h->delayed_pic[j] = NULL;
2761 memset(h->ref_list[0], 0, sizeof(h->ref_list[0]));
2762 memset(h->ref_list[1], 0, sizeof(h->ref_list[1]));
2763 memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0]));
2764 memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1]));
2765 ff_h264_reset_sei(h);
2766 h->recovery_frame= -1;
2769 h->current_slice = 0;
2772 /* forget old pics after a seek */
2773 static void flush_dpb(AVCodecContext *avctx)
2775 H264Context *h = avctx->priv_data;
2778 for (i = 0; i <= MAX_DELAYED_PIC_COUNT; i++) {
2779 if (h->delayed_pic[i])
2780 h->delayed_pic[i]->reference = 0;
2781 h->delayed_pic[i] = NULL;
2787 for (i = 0; i < MAX_PICTURE_COUNT; i++)
2788 unref_picture(h, &h->DPB[i]);
2789 h->cur_pic_ptr = NULL;
2790 unref_picture(h, &h->cur_pic);
2792 h->mb_x = h->mb_y = 0;
2794 h->parse_context.state = -1;
2795 h->parse_context.frame_start_found = 0;
2796 h->parse_context.overread = 0;
2797 h->parse_context.overread_index = 0;
2798 h->parse_context.index = 0;
2799 h->parse_context.last_index = 0;
2802 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
2804 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2807 h->frame_num_offset = h->prev_frame_num_offset;
2808 if (h->frame_num < h->prev_frame_num)
2809 h->frame_num_offset += max_frame_num;
2811 if (h->sps.poc_type == 0) {
2812 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2814 if (h->poc_lsb < h->prev_poc_lsb &&
2815 h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2816 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2817 else if (h->poc_lsb > h->prev_poc_lsb &&
2818 h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2819 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2821 h->poc_msb = h->prev_poc_msb;
2823 field_poc[1] = h->poc_msb + h->poc_lsb;
2824 if (h->picture_structure == PICT_FRAME)
2825 field_poc[1] += h->delta_poc_bottom;
2826 } else if (h->sps.poc_type == 1) {
2827 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2830 if (h->sps.poc_cycle_length != 0)
2831 abs_frame_num = h->frame_num_offset + h->frame_num;
2835 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2838 expected_delta_per_poc_cycle = 0;
2839 for (i = 0; i < h->sps.poc_cycle_length; i++)
2840 // FIXME integrate during sps parse
2841 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2843 if (abs_frame_num > 0) {
2844 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2845 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2847 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2848 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2849 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2853 if (h->nal_ref_idc == 0)
2854 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2856 field_poc[0] = expectedpoc + h->delta_poc[0];
2857 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2859 if (h->picture_structure == PICT_FRAME)
2860 field_poc[1] += h->delta_poc[1];
2862 int poc = 2 * (h->frame_num_offset + h->frame_num);
2864 if (!h->nal_ref_idc)
2871 if (h->picture_structure != PICT_BOTTOM_FIELD)
2872 pic_field_poc[0] = field_poc[0];
2873 if (h->picture_structure != PICT_TOP_FIELD)
2874 pic_field_poc[1] = field_poc[1];
2875 *pic_poc = FFMIN(pic_field_poc[0], pic_field_poc[1]);
2881 * initialize scan tables
2883 static void init_scan_tables(H264Context *h)
2886 for (i = 0; i < 16; i++) {
2887 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2888 h->zigzag_scan[i] = T(zigzag_scan[i]);
2889 h->field_scan[i] = T(field_scan[i]);
2892 for (i = 0; i < 64; i++) {
2893 #define T(x) (x >> 3) | ((x & 7) << 3)
2894 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2895 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2896 h->field_scan8x8[i] = T(field_scan8x8[i]);
2897 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
2900 if (h->sps.transform_bypass) { // FIXME same ugly
2901 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2902 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
2903 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2904 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
2905 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2906 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2908 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2909 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
2910 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2911 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
2912 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2913 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
2917 static int field_end(H264Context *h, int in_setup)
2919 AVCodecContext *const avctx = h->avctx;
2923 if (CONFIG_H264_VDPAU_DECODER &&
2924 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2925 ff_vdpau_h264_set_reference_frames(h);
2927 if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2928 if (!h->droppable) {
2929 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2930 h->prev_poc_msb = h->poc_msb;
2931 h->prev_poc_lsb = h->poc_lsb;
2933 h->prev_frame_num_offset = h->frame_num_offset;
2934 h->prev_frame_num = h->frame_num;
2935 h->outputed_poc = h->next_outputed_poc;
2938 if (avctx->hwaccel) {
2939 if (avctx->hwaccel->end_frame(avctx) < 0)
2940 av_log(avctx, AV_LOG_ERROR,
2941 "hardware accelerator failed to decode picture\n");
2944 if (CONFIG_H264_VDPAU_DECODER &&
2945 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2946 ff_vdpau_h264_picture_complete(h);
2949 * FIXME: Error handling code does not seem to support interlaced
2950 * when slices span multiple rows
2951 * The ff_er_add_slice calls don't work right for bottom
2952 * fields; they cause massive erroneous error concealing
2953 * Error marking covers both fields (top and bottom).
2954 * This causes a mismatched s->error_count
2955 * and a bad error table. Further, the error count goes to
2956 * INT_MAX when called for bottom field, because mb_y is
2957 * past end by one (callers fault) and resync_mb_y != 0
2958 * causes problems for the first MB line, too.
2960 if (CONFIG_ERROR_RESILIENCE &&
2961 !FIELD_PICTURE(h) && h->current_slice && !h->sps.new) {
2962 h->er.cur_pic = h->cur_pic_ptr;
2963 ff_er_frame_end(&h->er);
2965 if (!in_setup && !h->droppable)
2966 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
2967 h->picture_structure == PICT_BOTTOM_FIELD);
2970 h->current_slice = 0;
2976 * Replicate H264 "master" context to thread contexts.
2978 static int clone_slice(H264Context *dst, H264Context *src)
2980 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2981 dst->cur_pic_ptr = src->cur_pic_ptr;
2982 dst->cur_pic = src->cur_pic;
2983 dst->linesize = src->linesize;
2984 dst->uvlinesize = src->uvlinesize;
2985 dst->first_field = src->first_field;
2987 dst->prev_poc_msb = src->prev_poc_msb;
2988 dst->prev_poc_lsb = src->prev_poc_lsb;
2989 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2990 dst->prev_frame_num = src->prev_frame_num;
2991 dst->short_ref_count = src->short_ref_count;
2993 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2994 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2995 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2997 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2998 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
3004 * Compute profile from profile_idc and constraint_set?_flags.
3008 * @return profile as defined by FF_PROFILE_H264_*
3010 int ff_h264_get_profile(SPS *sps)
3012 int profile = sps->profile_idc;
3014 switch (sps->profile_idc) {
3015 case FF_PROFILE_H264_BASELINE:
3016 // constraint_set1_flag set to 1
3017 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
3019 case FF_PROFILE_H264_HIGH_10:
3020 case FF_PROFILE_H264_HIGH_422:
3021 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
3022 // constraint_set3_flag set to 1
3023 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
3030 static int h264_set_parameter_from_sps(H264Context *h)
3032 if (h->flags & CODEC_FLAG_LOW_DELAY ||
3033 (h->sps.bitstream_restriction_flag &&
3034 !h->sps.num_reorder_frames)) {
3035 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
3036 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
3037 "Reenabling low delay requires a codec flush.\n");
3042 if (h->avctx->has_b_frames < 2)
3043 h->avctx->has_b_frames = !h->low_delay;
3045 if (h->sps.bit_depth_luma != h->sps.bit_depth_chroma) {
3046 avpriv_request_sample(h->avctx,
3047 "Different chroma and luma bit depth");
3048 return AVERROR_PATCHWELCOME;
3051 if (h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3052 h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
3053 if (h->avctx->codec &&
3054 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU &&
3055 (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
3056 av_log(h->avctx, AV_LOG_ERROR,
3057 "VDPAU decoding does not support video colorspace.\n");
3058 return AVERROR_INVALIDDATA;
3060 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 &&
3061 h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13) {
3062 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
3063 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
3064 h->pixel_shift = h->sps.bit_depth_luma > 8;
3066 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
3067 h->sps.chroma_format_idc);
3068 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
3069 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
3070 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
3071 h->sps.chroma_format_idc);
3073 if (CONFIG_ERROR_RESILIENCE)
3074 ff_dsputil_init(&h->dsp, h->avctx);
3075 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
3077 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
3078 h->sps.bit_depth_luma);
3079 return AVERROR_INVALIDDATA;
3085 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
3087 switch (h->sps.bit_depth_luma) {
3090 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3091 return AV_PIX_FMT_GBRP9;
3093 return AV_PIX_FMT_YUV444P9;
3094 } else if (CHROMA422(h))
3095 return AV_PIX_FMT_YUV422P9;
3097 return AV_PIX_FMT_YUV420P9;
3101 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3102 return AV_PIX_FMT_GBRP10;
3104 return AV_PIX_FMT_YUV444P10;
3105 } else if (CHROMA422(h))
3106 return AV_PIX_FMT_YUV422P10;
3108 return AV_PIX_FMT_YUV420P10;
3112 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3113 return AV_PIX_FMT_GBRP12;
3115 return AV_PIX_FMT_YUV444P12;
3116 } else if (CHROMA422(h))
3117 return AV_PIX_FMT_YUV422P12;
3119 return AV_PIX_FMT_YUV420P12;
3123 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3124 return AV_PIX_FMT_GBRP14;
3126 return AV_PIX_FMT_YUV444P14;
3127 } else if (CHROMA422(h))
3128 return AV_PIX_FMT_YUV422P14;
3130 return AV_PIX_FMT_YUV420P14;
3134 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
3135 av_log(h->avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
3136 return AV_PIX_FMT_GBR24P;
3137 } else if (h->avctx->colorspace == AVCOL_SPC_YCGCO) {
3138 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
3140 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
3141 : AV_PIX_FMT_YUV444P;
3142 } else if (CHROMA422(h)) {
3143 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
3144 : AV_PIX_FMT_YUV422P;
3147 const enum AVPixelFormat * fmt = h->avctx->codec->pix_fmts ?
3148 h->avctx->codec->pix_fmts :
3149 h->avctx->color_range == AVCOL_RANGE_JPEG ?
3150 h264_hwaccel_pixfmt_list_jpeg_420 :
3151 h264_hwaccel_pixfmt_list_420;
3153 for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++)
3154 if (fmt[i] == h->avctx->pix_fmt && !force_callback)
3156 return ff_thread_get_format(h->avctx, fmt);
3160 av_log(h->avctx, AV_LOG_ERROR,
3161 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
3162 return AVERROR_INVALIDDATA;
3166 /* export coded and cropped frame dimensions to AVCodecContext */
3167 static int init_dimensions(H264Context *h)
3169 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
3170 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
3171 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
3172 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
3174 /* handle container cropping */
3176 FFALIGN(h->avctx->width, 16) == h->width &&
3177 FFALIGN(h->avctx->height, 16) == h->height) {
3178 width = h->avctx->width;
3179 height = h->avctx->height;
3182 if (width <= 0 || height <= 0) {
3183 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
3185 if (h->avctx->err_recognition & AV_EF_EXPLODE)
3186 return AVERROR_INVALIDDATA;
3188 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
3189 h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
3196 h->avctx->coded_width = h->width;
3197 h->avctx->coded_height = h->height;
3198 h->avctx->width = width;
3199 h->avctx->height = height;
3204 static int h264_slice_header_init(H264Context *h, int reinit)
3206 int nb_slices = (HAVE_THREADS &&
3207 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
3208 h->avctx->thread_count : 1;
3211 h->avctx->sample_aspect_ratio = h->sps.sar;
3212 av_assert0(h->avctx->sample_aspect_ratio.den);
3213 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
3214 &h->chroma_x_shift, &h->chroma_y_shift);
3216 if (h->sps.timing_info_present_flag) {
3217 int64_t den = h->sps.time_scale;
3218 if (h->x264_build < 44U)
3220 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
3221 h->sps.num_units_in_tick, den, 1 << 30);
3224 h->avctx->hwaccel = ff_find_hwaccel(h->avctx->codec->id, h->avctx->pix_fmt);
3229 h->prev_interlaced_frame = 1;
3231 init_scan_tables(h);
3232 ret = ff_h264_alloc_tables(h);
3234 av_log(h->avctx, AV_LOG_ERROR,
3235 "Could not allocate memory for h264\n");
3239 if (nb_slices > MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
3242 max_slices = FFMIN(MAX_THREADS, h->mb_height);
3244 max_slices = MAX_THREADS;
3245 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
3246 " reducing to %d\n", nb_slices, max_slices);
3247 nb_slices = max_slices;
3249 h->slice_context_count = nb_slices;
3251 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
3252 ret = context_init(h);
3254 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
3258 for (i = 1; i < h->slice_context_count; i++) {
3260 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
3261 c->avctx = h->avctx;
3262 if (CONFIG_ERROR_RESILIENCE) {
3266 c->h264dsp = h->h264dsp;
3267 c->h264qpel = h->h264qpel;
3268 c->h264chroma = h->h264chroma;
3271 c->pixel_shift = h->pixel_shift;
3272 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
3273 c->width = h->width;
3274 c->height = h->height;
3275 c->linesize = h->linesize;
3276 c->uvlinesize = h->uvlinesize;
3277 c->chroma_x_shift = h->chroma_x_shift;
3278 c->chroma_y_shift = h->chroma_y_shift;
3279 c->qscale = h->qscale;
3280 c->droppable = h->droppable;
3281 c->data_partitioning = h->data_partitioning;
3282 c->low_delay = h->low_delay;
3283 c->mb_width = h->mb_width;
3284 c->mb_height = h->mb_height;
3285 c->mb_stride = h->mb_stride;
3286 c->mb_num = h->mb_num;
3287 c->flags = h->flags;
3288 c->workaround_bugs = h->workaround_bugs;
3289 c->pict_type = h->pict_type;
3291 init_scan_tables(c);
3292 clone_tables(c, h, i);
3293 c->context_initialized = 1;
3296 for (i = 0; i < h->slice_context_count; i++)
3297 if ((ret = context_init(h->thread_context[i])) < 0) {
3298 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
3303 h->context_initialized = 1;
3309 * Decode a slice header.
3310 * This will also call ff_MPV_common_init() and frame_start() as needed.
3312 * @param h h264context
3313 * @param h0 h264 master context (differs from 'h' when doing sliced based
3314 * parallel decoding)
3316 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
3318 static int decode_slice_header(H264Context *h, H264Context *h0)
3320 unsigned int first_mb_in_slice;
3321 unsigned int pps_id;
3322 int num_ref_idx_active_override_flag, ret;
3323 unsigned int slice_type, tmp, i, j;
3324 int last_pic_structure, last_pic_droppable;
3326 int needs_reinit = 0;
3327 int field_pic_flag, bottom_field_flag;
3329 h->me.qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
3330 h->me.qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
3332 first_mb_in_slice = get_ue_golomb_long(&h->gb);
3334 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
3335 if (h0->current_slice && FIELD_PICTURE(h)) {
3339 h0->current_slice = 0;
3340 if (!h0->first_field) {
3341 if (h->cur_pic_ptr && !h->droppable) {
3342 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
3343 h->picture_structure == PICT_BOTTOM_FIELD);
3345 h->cur_pic_ptr = NULL;
3349 slice_type = get_ue_golomb_31(&h->gb);
3350 if (slice_type > 9) {
3351 av_log(h->avctx, AV_LOG_ERROR,
3352 "slice type too large (%d) at %d %d\n",
3353 slice_type, h->mb_x, h->mb_y);
3354 return AVERROR_INVALIDDATA;
3356 if (slice_type > 4) {
3358 h->slice_type_fixed = 1;
3360 h->slice_type_fixed = 0;
3362 slice_type = golomb_to_pict_type[slice_type];
3363 h->slice_type = slice_type;
3364 h->slice_type_nos = slice_type & 3;
3366 // to make a few old functions happy, it's wrong though
3367 h->pict_type = h->slice_type;
3369 pps_id = get_ue_golomb(&h->gb);
3370 if (pps_id >= MAX_PPS_COUNT) {
3371 av_log(h->avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
3372 return AVERROR_INVALIDDATA;
3374 if (!h0->pps_buffers[pps_id]) {
3375 av_log(h->avctx, AV_LOG_ERROR,
3376 "non-existing PPS %u referenced\n",
3378 return AVERROR_INVALIDDATA;
3380 h->pps = *h0->pps_buffers[pps_id];
3382 if (!h0->sps_buffers[h->pps.sps_id]) {
3383 av_log(h->avctx, AV_LOG_ERROR,
3384 "non-existing SPS %u referenced\n",
3386 return AVERROR_INVALIDDATA;
3389 if (h->pps.sps_id != h->current_sps_id ||
3390 h0->sps_buffers[h->pps.sps_id]->new) {
3391 h0->sps_buffers[h->pps.sps_id]->new = 0;
3393 h->current_sps_id = h->pps.sps_id;
3394 h->sps = *h0->sps_buffers[h->pps.sps_id];
3396 if (h->mb_width != h->sps.mb_width ||
3397 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
3398 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
3399 h->cur_chroma_format_idc != h->sps.chroma_format_idc
3403 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
3404 h->chroma_format_idc != h->sps.chroma_format_idc) {
3405 h->bit_depth_luma = h->sps.bit_depth_luma;
3406 h->chroma_format_idc = h->sps.chroma_format_idc;
3409 if ((ret = h264_set_parameter_from_sps(h)) < 0)
3413 h->avctx->profile = ff_h264_get_profile(&h->sps);
3414 h->avctx->level = h->sps.level_idc;
3415 h->avctx->refs = h->sps.ref_frame_count;
3417 must_reinit = (h->context_initialized &&
3418 ( 16*h->sps.mb_width != h->avctx->coded_width
3419 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
3420 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
3421 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
3422 || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)
3423 || h->mb_width != h->sps.mb_width
3424 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
3426 if (h0->avctx->pix_fmt != get_pixel_format(h0, 0))
3429 h->mb_width = h->sps.mb_width;
3430 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3431 h->mb_num = h->mb_width * h->mb_height;
3432 h->mb_stride = h->mb_width + 1;
3434 h->b_stride = h->mb_width * 4;
3436 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
3438 h->width = 16 * h->mb_width;
3439 h->height = 16 * h->mb_height;
3441 ret = init_dimensions(h);
3445 if (h->sps.video_signal_type_present_flag) {
3446 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
3448 if (h->sps.colour_description_present_flag) {
3449 if (h->avctx->colorspace != h->sps.colorspace)
3451 h->avctx->color_primaries = h->sps.color_primaries;
3452 h->avctx->color_trc = h->sps.color_trc;
3453 h->avctx->colorspace = h->sps.colorspace;
3457 if (h->context_initialized &&
3458 (h->width != h->avctx->coded_width ||
3459 h->height != h->avctx->coded_height ||
3463 av_log(h->avctx, AV_LOG_ERROR, "changing width/height on "
3464 "slice %d\n", h0->current_slice + 1);
3465 return AVERROR_INVALIDDATA;
3470 if ((ret = get_pixel_format(h, 1)) < 0)
3472 h->avctx->pix_fmt = ret;
3474 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
3475 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
3477 if ((ret = h264_slice_header_init(h, 1)) < 0) {
3478 av_log(h->avctx, AV_LOG_ERROR,
3479 "h264_slice_header_init() failed\n");
3483 if (!h->context_initialized) {
3485 av_log(h->avctx, AV_LOG_ERROR,
3486 "Cannot (re-)initialize context during parallel decoding.\n");
3487 return AVERROR_PATCHWELCOME;
3490 if ((ret = get_pixel_format(h, 1)) < 0)
3492 h->avctx->pix_fmt = ret;
3494 if ((ret = h264_slice_header_init(h, 0)) < 0) {
3495 av_log(h->avctx, AV_LOG_ERROR,
3496 "h264_slice_header_init() failed\n");
3501 if (h == h0 && h->dequant_coeff_pps != pps_id) {
3502 h->dequant_coeff_pps = pps_id;
3503 init_dequant_tables(h);
3506 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
3509 h->mb_aff_frame = 0;
3510 last_pic_structure = h0->picture_structure;
3511 last_pic_droppable = h0->droppable;
3512 h->droppable = h->nal_ref_idc == 0;
3513 if (h->sps.frame_mbs_only_flag) {
3514 h->picture_structure = PICT_FRAME;
3516 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
3517 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
3520 field_pic_flag = get_bits1(&h->gb);
3521 if (field_pic_flag) {
3522 bottom_field_flag = get_bits1(&h->gb);
3523 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
3525 h->picture_structure = PICT_FRAME;
3526 h->mb_aff_frame = h->sps.mb_aff;
3529 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
3531 if (h0->current_slice != 0) {
3532 if (last_pic_structure != h->picture_structure ||
3533 last_pic_droppable != h->droppable) {
3534 av_log(h->avctx, AV_LOG_ERROR,
3535 "Changing field mode (%d -> %d) between slices is not allowed\n",
3536 last_pic_structure, h->picture_structure);
3537 h->picture_structure = last_pic_structure;
3538 h->droppable = last_pic_droppable;
3539 return AVERROR_INVALIDDATA;
3540 } else if (!h0->cur_pic_ptr) {
3541 av_log(h->avctx, AV_LOG_ERROR,
3542 "unset cur_pic_ptr on %d. slice\n",
3543 h0->current_slice + 1);
3544 return AVERROR_INVALIDDATA;
3547 /* Shorten frame num gaps so we don't have to allocate reference
3548 * frames just to throw them away */
3549 if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
3550 int unwrap_prev_frame_num = h->prev_frame_num;
3551 int max_frame_num = 1 << h->sps.log2_max_frame_num;
3553 if (unwrap_prev_frame_num > h->frame_num)
3554 unwrap_prev_frame_num -= max_frame_num;
3556 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
3557 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
3558 if (unwrap_prev_frame_num < 0)
3559 unwrap_prev_frame_num += max_frame_num;
3561 h->prev_frame_num = unwrap_prev_frame_num;
3565 /* See if we have a decoded first field looking for a pair...
3566 * Here, we're using that to see if we should mark previously
3567 * decode frames as "finished".
3568 * We have to do that before the "dummy" in-between frame allocation,
3569 * since that can modify h->cur_pic_ptr. */
3570 if (h0->first_field) {
3571 assert(h0->cur_pic_ptr);
3572 assert(h0->cur_pic_ptr->f.data[0]);
3573 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
3575 /* Mark old field/frame as completed */
3576 if (!last_pic_droppable && h0->cur_pic_ptr->tf.owner == h0->avctx) {
3577 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3578 last_pic_structure == PICT_BOTTOM_FIELD);
3581 /* figure out if we have a complementary field pair */
3582 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
3583 /* Previous field is unmatched. Don't display it, but let it
3584 * remain for reference if marked as such. */
3585 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
3586 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3587 last_pic_structure == PICT_TOP_FIELD);
3590 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
3591 /* This and previous field were reference, but had
3592 * different frame_nums. Consider this field first in
3593 * pair. Throw away previous field except for reference
3595 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
3596 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3597 last_pic_structure == PICT_TOP_FIELD);
3600 /* Second field in complementary pair */
3601 if (!((last_pic_structure == PICT_TOP_FIELD &&
3602 h->picture_structure == PICT_BOTTOM_FIELD) ||
3603 (last_pic_structure == PICT_BOTTOM_FIELD &&
3604 h->picture_structure == PICT_TOP_FIELD))) {
3605 av_log(h->avctx, AV_LOG_ERROR,
3606 "Invalid field mode combination %d/%d\n",
3607 last_pic_structure, h->picture_structure);
3608 h->picture_structure = last_pic_structure;
3609 h->droppable = last_pic_droppable;
3610 return AVERROR_INVALIDDATA;
3611 } else if (last_pic_droppable != h->droppable) {
3612 avpriv_request_sample(h->avctx,
3613 "Found reference and non-reference fields in the same frame, which");
3614 h->picture_structure = last_pic_structure;
3615 h->droppable = last_pic_droppable;
3616 return AVERROR_PATCHWELCOME;
3622 while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 && !h0->first_field &&
3623 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
3624 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
3625 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
3626 h->frame_num, h->prev_frame_num);
3627 if (!h->sps.gaps_in_frame_num_allowed_flag)
3628 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
3629 h->last_pocs[i] = INT_MIN;
3630 ret = h264_frame_start(h);
3633 h->prev_frame_num++;
3634 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
3635 h->cur_pic_ptr->frame_num = h->prev_frame_num;
3636 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
3637 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
3638 ret = ff_generate_sliding_window_mmcos(h, 1);
3639 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
3641 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
3642 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
3644 /* Error concealment: If a ref is missing, copy the previous ref
3646 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
3647 * many assumptions about there being no actual duplicates.
3648 * FIXME: This does not copy padding for out-of-frame motion
3649 * vectors. Given we are concealing a lost frame, this probably
3650 * is not noticeable by comparison, but it should be fixed. */
3651 if (h->short_ref_count) {
3653 av_image_copy(h->short_ref[0]->f.data,
3654 h->short_ref[0]->f.linesize,
3655 (const uint8_t **)prev->f.data,
3660 h->short_ref[0]->poc = prev->poc + 2;
3662 h->short_ref[0]->frame_num = h->prev_frame_num;
3666 /* See if we have a decoded first field looking for a pair...
3667 * We're using that to see whether to continue decoding in that
3668 * frame, or to allocate a new one. */
3669 if (h0->first_field) {
3670 assert(h0->cur_pic_ptr);
3671 assert(h0->cur_pic_ptr->f.data[0]);
3672 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
3674 /* figure out if we have a complementary field pair */
3675 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
3676 /* Previous field is unmatched. Don't display it, but let it
3677 * remain for reference if marked as such. */
3678 h0->cur_pic_ptr = NULL;
3679 h0->first_field = FIELD_PICTURE(h);
3681 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
3682 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
3683 h0->picture_structure==PICT_BOTTOM_FIELD);
3684 /* This and the previous field had different frame_nums.
3685 * Consider this field first in pair. Throw away previous
3686 * one except for reference purposes. */
3687 h0->first_field = 1;
3688 h0->cur_pic_ptr = NULL;
3690 /* Second field in complementary pair */
3691 h0->first_field = 0;
3695 /* Frame or first field in a potentially complementary pair */
3696 h0->first_field = FIELD_PICTURE(h);
3699 if (!FIELD_PICTURE(h) || h0->first_field) {
3700 if (h264_frame_start(h) < 0) {
3701 h0->first_field = 0;
3702 return AVERROR_INVALIDDATA;
3705 release_unused_pictures(h, 0);
3707 /* Some macroblocks can be accessed before they're available in case
3708 * of lost slices, MBAFF or threading. */
3709 if (FIELD_PICTURE(h)) {
3710 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
3711 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
3713 memset(h->slice_table, -1,
3714 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
3716 h0->last_slice_type = -1;
3718 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
3721 /* can't be in alloc_tables because linesize isn't known there.
3722 * FIXME: redo bipred weight to not require extra buffer? */
3723 for (i = 0; i < h->slice_context_count; i++)
3724 if (h->thread_context[i]) {
3725 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
3730 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
3732 av_assert1(h->mb_num == h->mb_width * h->mb_height);
3733 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
3734 first_mb_in_slice >= h->mb_num) {
3735 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3736 return AVERROR_INVALIDDATA;
3738 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
3739 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
3740 FIELD_OR_MBAFF_PICTURE(h);
3741 if (h->picture_structure == PICT_BOTTOM_FIELD)
3742 h->resync_mb_y = h->mb_y = h->mb_y + 1;
3743 av_assert1(h->mb_y < h->mb_height);
3745 if (h->picture_structure == PICT_FRAME) {
3746 h->curr_pic_num = h->frame_num;
3747 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
3749 h->curr_pic_num = 2 * h->frame_num + 1;
3750 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
3753 if (h->nal_unit_type == NAL_IDR_SLICE)
3754 get_ue_golomb(&h->gb); /* idr_pic_id */
3756 if (h->sps.poc_type == 0) {
3757 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
3759 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
3760 h->delta_poc_bottom = get_se_golomb(&h->gb);
3763 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
3764 h->delta_poc[0] = get_se_golomb(&h->gb);
3766 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
3767 h->delta_poc[1] = get_se_golomb(&h->gb);
3770 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
3772 if (h->pps.redundant_pic_cnt_present)
3773 h->redundant_pic_count = get_ue_golomb(&h->gb);
3775 // set defaults, might be overridden a few lines later
3776 h->ref_count[0] = h->pps.ref_count[0];
3777 h->ref_count[1] = h->pps.ref_count[1];
3779 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3781 max[0] = max[1] = h->picture_structure == PICT_FRAME ? 15 : 31;
3783 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3784 h->direct_spatial_mv_pred = get_bits1(&h->gb);
3785 num_ref_idx_active_override_flag = get_bits1(&h->gb);
3787 if (num_ref_idx_active_override_flag) {
3788 h->ref_count[0] = get_ue_golomb(&h->gb) + 1;
3789 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3790 h->ref_count[1] = get_ue_golomb(&h->gb) + 1;
3792 // full range is spec-ok in this case, even for frames
3793 h->ref_count[1] = 1;
3796 if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
3797 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]);
3798 h->ref_count[0] = h->ref_count[1] = 0;
3799 return AVERROR_INVALIDDATA;
3802 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3808 h->ref_count[0] = h->ref_count[1] = 0;
3810 if (slice_type != AV_PICTURE_TYPE_I &&
3811 (h0->current_slice == 0 ||
3812 slice_type != h0->last_slice_type ||
3813 memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
3814 ff_h264_fill_default_ref_list(h);
3817 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3818 ret = ff_h264_decode_ref_pic_list_reordering(h);
3820 h->ref_count[1] = h->ref_count[0] = 0;
3825 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
3826 (h->pps.weighted_bipred_idc == 1 &&
3827 h->slice_type_nos == AV_PICTURE_TYPE_B))
3828 pred_weight_table(h);
3829 else if (h->pps.weighted_bipred_idc == 2 &&
3830 h->slice_type_nos == AV_PICTURE_TYPE_B) {
3831 implicit_weight_table(h, -1);
3834 for (i = 0; i < 2; i++) {
3835 h->luma_weight_flag[i] = 0;
3836 h->chroma_weight_flag[i] = 0;
3840 // If frame-mt is enabled, only update mmco tables for the first slice
3841 // in a field. Subsequent slices can temporarily clobber h->mmco_index
3842 // or h->mmco, which will cause ref list mix-ups and decoding errors
3843 // further down the line. This may break decoding if the first slice is
3844 // corrupt, thus we only do this if frame-mt is enabled.
3845 if (h->nal_ref_idc) {
3846 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
3847 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
3848 h0->current_slice == 0);
3849 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
3850 return AVERROR_INVALIDDATA;
3853 if (FRAME_MBAFF(h)) {
3854 ff_h264_fill_mbaff_ref_list(h);
3856 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
3857 implicit_weight_table(h, 0);
3858 implicit_weight_table(h, 1);
3862 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3863 ff_h264_direct_dist_scale_factor(h);
3864 ff_h264_direct_ref_list_init(h);
3866 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3867 tmp = get_ue_golomb_31(&h->gb);
3869 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3870 return AVERROR_INVALIDDATA;
3872 h->cabac_init_idc = tmp;
3875 h->last_qscale_diff = 0;
3876 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
3877 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3878 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3879 return AVERROR_INVALIDDATA;
3882 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
3883 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
3884 // FIXME qscale / qp ... stuff
3885 if (h->slice_type == AV_PICTURE_TYPE_SP)
3886 get_bits1(&h->gb); /* sp_for_switch_flag */
3887 if (h->slice_type == AV_PICTURE_TYPE_SP ||
3888 h->slice_type == AV_PICTURE_TYPE_SI)
3889 get_se_golomb(&h->gb); /* slice_qs_delta */
3891 h->deblocking_filter = 1;
3892 h->slice_alpha_c0_offset = 52;
3893 h->slice_beta_offset = 52;
3894 if (h->pps.deblocking_filter_parameters_present) {
3895 tmp = get_ue_golomb_31(&h->gb);
3897 av_log(h->avctx, AV_LOG_ERROR,
3898 "deblocking_filter_idc %u out of range\n", tmp);
3899 return AVERROR_INVALIDDATA;
3901 h->deblocking_filter = tmp;
3902 if (h->deblocking_filter < 2)
3903 h->deblocking_filter ^= 1; // 1<->0
3905 if (h->deblocking_filter) {
3906 h->slice_alpha_c0_offset += get_se_golomb(&h->gb) << 1;
3907 h->slice_beta_offset += get_se_golomb(&h->gb) << 1;
3908 if (h->slice_alpha_c0_offset > 104U ||
3909 h->slice_beta_offset > 104U) {
3910 av_log(h->avctx, AV_LOG_ERROR,
3911 "deblocking filter parameters %d %d out of range\n",
3912 h->slice_alpha_c0_offset, h->slice_beta_offset);
3913 return AVERROR_INVALIDDATA;
3918 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3919 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3920 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3921 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
3922 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3923 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3924 h->nal_ref_idc == 0))
3925 h->deblocking_filter = 0;
3927 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3928 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
3929 /* Cheat slightly for speed:
3930 * Do not bother to deblock across slices. */
3931 h->deblocking_filter = 2;
3933 h0->max_contexts = 1;
3934 if (!h0->single_decode_warning) {
3935 av_log(h->avctx, AV_LOG_INFO,
3936 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3937 h0->single_decode_warning = 1;
3940 av_log(h->avctx, AV_LOG_ERROR,
3941 "Deblocking switched inside frame.\n");
3946 h->qp_thresh = 15 + 52 -
3947 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3949 h->pps.chroma_qp_index_offset[0],
3950 h->pps.chroma_qp_index_offset[1]) +
3951 6 * (h->sps.bit_depth_luma - 8);
3953 h0->last_slice_type = slice_type;
3954 memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
3955 h->slice_num = ++h0->current_slice;
3958 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
3959 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
3960 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
3961 && h->slice_num >= MAX_SLICES) {
3962 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3963 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);
3966 for (j = 0; j < 2; j++) {
3968 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3969 for (i = 0; i < 16; i++) {
3971 if (j < h->list_count && i < h->ref_count[j] &&
3972 h->ref_list[j][i].f.buf[0]) {
3974 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
3975 for (k = 0; k < h->short_ref_count; k++)
3976 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
3980 for (k = 0; k < h->long_ref_count; k++)
3981 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
3982 id_list[i] = h->short_ref_count + k;
3990 for (i = 0; i < 16; i++)
3991 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
3993 ref2frm[18 + 1] = -1;
3994 for (i = 16; i < 48; i++)
3995 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3996 (h->ref_list[j][i].reference & 3);
3999 if (h->ref_count[0]) h->er.last_pic = &h->ref_list[0][0];
4000 if (h->ref_count[1]) h->er.next_pic = &h->ref_list[1][0];
4001 h->er.ref_count = h->ref_count[0];
4003 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
4004 av_log(h->avctx, AV_LOG_DEBUG,
4005 "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",
4007 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
4009 av_get_picture_type_char(h->slice_type),
4010 h->slice_type_fixed ? " fix" : "",
4011 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
4012 pps_id, h->frame_num,
4013 h->cur_pic_ptr->field_poc[0],
4014 h->cur_pic_ptr->field_poc[1],
4015 h->ref_count[0], h->ref_count[1],
4017 h->deblocking_filter,
4018 h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
4020 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
4021 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
4027 int ff_h264_get_slice_type(const H264Context *h)
4029 switch (h->slice_type) {
4030 case AV_PICTURE_TYPE_P:
4032 case AV_PICTURE_TYPE_B:
4034 case AV_PICTURE_TYPE_I:
4036 case AV_PICTURE_TYPE_SP:
4038 case AV_PICTURE_TYPE_SI:
4041 return AVERROR_INVALIDDATA;
4045 static av_always_inline void fill_filter_caches_inter(H264Context *h,
4046 int mb_type, int top_xy,
4047 int left_xy[LEFT_MBS],
4049 int left_type[LEFT_MBS],
4050 int mb_xy, int list)
4052 int b_stride = h->b_stride;
4053 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
4054 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
4055 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
4056 if (USES_LIST(top_type, list)) {
4057 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
4058 const int b8_xy = 4 * top_xy + 2;
4059 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
4060 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
4061 ref_cache[0 - 1 * 8] =
4062 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
4063 ref_cache[2 - 1 * 8] =
4064 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
4066 AV_ZERO128(mv_dst - 1 * 8);
4067 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4070 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
4071 if (USES_LIST(left_type[LTOP], list)) {
4072 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
4073 const int b8_xy = 4 * left_xy[LTOP] + 1;
4074 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
4075 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
4076 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
4077 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
4078 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
4080 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
4081 ref_cache[-1 + 16] =
4082 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
4084 AV_ZERO32(mv_dst - 1 + 0);
4085 AV_ZERO32(mv_dst - 1 + 8);
4086 AV_ZERO32(mv_dst - 1 + 16);
4087 AV_ZERO32(mv_dst - 1 + 24);
4090 ref_cache[-1 + 16] =
4091 ref_cache[-1 + 24] = LIST_NOT_USED;
4096 if (!USES_LIST(mb_type, list)) {
4097 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
4098 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4099 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4100 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4101 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
4106 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
4107 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
4108 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
4109 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
4110 AV_WN32A(&ref_cache[0 * 8], ref01);
4111 AV_WN32A(&ref_cache[1 * 8], ref01);
4112 AV_WN32A(&ref_cache[2 * 8], ref23);
4113 AV_WN32A(&ref_cache[3 * 8], ref23);
4117 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
4118 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
4119 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
4120 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
4121 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
4127 * @return non zero if the loop filter can be skipped
4129 static int fill_filter_caches(H264Context *h, int mb_type)
4131 const int mb_xy = h->mb_xy;
4132 int top_xy, left_xy[LEFT_MBS];
4133 int top_type, left_type[LEFT_MBS];
4137 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
4139 /* Wow, what a mess, why didn't they simplify the interlacing & intra
4140 * stuff, I can't imagine that these complex rules are worth it. */
4142 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
4143 if (FRAME_MBAFF(h)) {
4144 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
4145 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
4147 if (left_mb_field_flag != curr_mb_field_flag)
4148 left_xy[LTOP] -= h->mb_stride;
4150 if (curr_mb_field_flag)
4151 top_xy += h->mb_stride &
4152 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
4153 if (left_mb_field_flag != curr_mb_field_flag)
4154 left_xy[LBOT] += h->mb_stride;
4158 h->top_mb_xy = top_xy;
4159 h->left_mb_xy[LTOP] = left_xy[LTOP];
4160 h->left_mb_xy[LBOT] = left_xy[LBOT];
4162 /* For sufficiently low qp, filtering wouldn't do anything.
4163 * This is a conservative estimate: could also check beta_offset
4164 * and more accurate chroma_qp. */
4165 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
4166 int qp = h->cur_pic.qscale_table[mb_xy];
4167 if (qp <= qp_thresh &&
4168 (left_xy[LTOP] < 0 ||
4169 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
4171 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
4172 if (!FRAME_MBAFF(h))
4174 if ((left_xy[LTOP] < 0 ||
4175 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
4176 (top_xy < h->mb_stride ||
4177 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
4182 top_type = h->cur_pic.mb_type[top_xy];
4183 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
4184 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
4185 if (h->deblocking_filter == 2) {
4186 if (h->slice_table[top_xy] != h->slice_num)
4188 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
4189 left_type[LTOP] = left_type[LBOT] = 0;
4191 if (h->slice_table[top_xy] == 0xFFFF)
4193 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
4194 left_type[LTOP] = left_type[LBOT] = 0;
4196 h->top_type = top_type;
4197 h->left_type[LTOP] = left_type[LTOP];
4198 h->left_type[LBOT] = left_type[LBOT];
4200 if (IS_INTRA(mb_type))
4203 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
4204 top_type, left_type, mb_xy, 0);
4205 if (h->list_count == 2)
4206 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
4207 top_type, left_type, mb_xy, 1);
4209 nnz = h->non_zero_count[mb_xy];
4210 nnz_cache = h->non_zero_count_cache;
4211 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
4212 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
4213 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
4214 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
4215 h->cbp = h->cbp_table[mb_xy];
4218 nnz = h->non_zero_count[top_xy];
4219 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
4222 if (left_type[LTOP]) {
4223 nnz = h->non_zero_count[left_xy[LTOP]];
4224 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
4225 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
4226 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
4227 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
4230 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
4231 * from what the loop filter needs */
4232 if (!CABAC(h) && h->pps.transform_8x8_mode) {
4233 if (IS_8x8DCT(top_type)) {
4234 nnz_cache[4 + 8 * 0] =
4235 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
4236 nnz_cache[6 + 8 * 0] =
4237 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
4239 if (IS_8x8DCT(left_type[LTOP])) {
4240 nnz_cache[3 + 8 * 1] =
4241 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
4243 if (IS_8x8DCT(left_type[LBOT])) {
4244 nnz_cache[3 + 8 * 3] =
4245 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
4248 if (IS_8x8DCT(mb_type)) {
4249 nnz_cache[scan8[0]] =
4250 nnz_cache[scan8[1]] =
4251 nnz_cache[scan8[2]] =
4252 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
4254 nnz_cache[scan8[0 + 4]] =
4255 nnz_cache[scan8[1 + 4]] =
4256 nnz_cache[scan8[2 + 4]] =
4257 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
4259 nnz_cache[scan8[0 + 8]] =
4260 nnz_cache[scan8[1 + 8]] =
4261 nnz_cache[scan8[2 + 8]] =
4262 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
4264 nnz_cache[scan8[0 + 12]] =
4265 nnz_cache[scan8[1 + 12]] =
4266 nnz_cache[scan8[2 + 12]] =
4267 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
4274 static void loop_filter(H264Context *h, int start_x, int end_x)
4276 uint8_t *dest_y, *dest_cb, *dest_cr;
4277 int linesize, uvlinesize, mb_x, mb_y;
4278 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
4279 const int old_slice_type = h->slice_type;
4280 const int pixel_shift = h->pixel_shift;
4281 const int block_h = 16 >> h->chroma_y_shift;
4283 if (h->deblocking_filter) {
4284 for (mb_x = start_x; mb_x < end_x; mb_x++)
4285 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
4287 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
4288 h->slice_num = h->slice_table[mb_xy];
4289 mb_type = h->cur_pic.mb_type[mb_xy];
4290 h->list_count = h->list_counts[mb_xy];
4294 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
4298 dest_y = h->cur_pic.f.data[0] +
4299 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
4300 dest_cb = h->cur_pic.f.data[1] +
4301 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
4302 mb_y * h->uvlinesize * block_h;
4303 dest_cr = h->cur_pic.f.data[2] +
4304 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
4305 mb_y * h->uvlinesize * block_h;
4306 // FIXME simplify above
4309 linesize = h->mb_linesize = h->linesize * 2;
4310 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
4311 if (mb_y & 1) { // FIXME move out of this function?
4312 dest_y -= h->linesize * 15;
4313 dest_cb -= h->uvlinesize * (block_h - 1);
4314 dest_cr -= h->uvlinesize * (block_h - 1);
4317 linesize = h->mb_linesize = h->linesize;
4318 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
4320 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
4322 if (fill_filter_caches(h, mb_type))
4324 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
4325 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
4327 if (FRAME_MBAFF(h)) {
4328 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
4329 linesize, uvlinesize);
4331 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
4332 dest_cr, linesize, uvlinesize);
4336 h->slice_type = old_slice_type;
4338 h->mb_y = end_mb_y - FRAME_MBAFF(h);
4339 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
4340 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
4343 static void predict_field_decoding_flag(H264Context *h)
4345 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
4346 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
4347 h->cur_pic.mb_type[mb_xy - 1] :
4348 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
4349 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
4350 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
4354 * Draw edges and report progress for the last MB row.
4356 static void decode_finish_row(H264Context *h)
4358 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
4359 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
4360 int height = 16 << FRAME_MBAFF(h);
4361 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
4363 if (h->deblocking_filter) {
4364 if ((top + height) >= pic_height)
4365 height += deblock_border;
4366 top -= deblock_border;
4369 if (top >= pic_height || (top + height) < 0)
4372 height = FFMIN(height, pic_height - top);
4374 height = top + height;
4378 ff_h264_draw_horiz_band(h, top, height);
4380 if (h->droppable || h->er.error_occurred)
4383 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
4384 h->picture_structure == PICT_BOTTOM_FIELD);
4387 static void er_add_slice(H264Context *h, int startx, int starty,
4388 int endx, int endy, int status)
4390 if (CONFIG_ERROR_RESILIENCE) {
4391 ERContext *er = &h->er;
4393 ff_er_add_slice(er, startx, starty, endx, endy, status);
4397 static int decode_slice(struct AVCodecContext *avctx, void *arg)
4399 H264Context *h = *(void **)arg;
4400 int lf_x_start = h->mb_x;
4402 h->mb_skip_run = -1;
4404 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
4406 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
4407 avctx->codec_id != AV_CODEC_ID_H264 ||
4408 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
4410 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
4411 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
4413 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
4414 prev_status &= ~ VP_START;
4415 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
4416 h->er.error_occurred = 1;
4422 align_get_bits(&h->gb);
4425 ff_init_cabac_decoder(&h->cabac,
4426 h->gb.buffer + get_bits_count(&h->gb) / 8,
4427 (get_bits_left(&h->gb) + 7) / 8);
4429 ff_h264_init_cabac_states(h);
4433 int ret = ff_h264_decode_mb_cabac(h);
4435 // STOP_TIMER("decode_mb_cabac")
4438 ff_h264_hl_decode_mb(h);
4440 // FIXME optimal? or let mb_decode decode 16x32 ?
4441 if (ret >= 0 && FRAME_MBAFF(h)) {
4444 ret = ff_h264_decode_mb_cabac(h);
4447 ff_h264_hl_decode_mb(h);
4450 eos = get_cabac_terminate(&h->cabac);
4452 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
4453 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
4454 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
4455 h->mb_y, ER_MB_END);
4456 if (h->mb_x >= lf_x_start)
4457 loop_filter(h, lf_x_start, h->mb_x + 1);
4460 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
4461 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
4462 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
4463 av_log(h->avctx, AV_LOG_ERROR,
4464 "error while decoding MB %d %d, bytestream (%td)\n",
4466 h->cabac.bytestream_end - h->cabac.bytestream);
4467 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4468 h->mb_y, ER_MB_ERROR);
4469 return AVERROR_INVALIDDATA;
4472 if (++h->mb_x >= h->mb_width) {
4473 loop_filter(h, lf_x_start, h->mb_x);
4474 h->mb_x = lf_x_start = 0;
4475 decode_finish_row(h);
4477 if (FIELD_OR_MBAFF_PICTURE(h)) {
4479 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
4480 predict_field_decoding_flag(h);
4484 if (eos || h->mb_y >= h->mb_height) {
4485 tprintf(h->avctx, "slice end %d %d\n",
4486 get_bits_count(&h->gb), h->gb.size_in_bits);
4487 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
4488 h->mb_y, ER_MB_END);
4489 if (h->mb_x > lf_x_start)
4490 loop_filter(h, lf_x_start, h->mb_x);
4496 int ret = ff_h264_decode_mb_cavlc(h);
4499 ff_h264_hl_decode_mb(h);
4501 // FIXME optimal? or let mb_decode decode 16x32 ?
4502 if (ret >= 0 && FRAME_MBAFF(h)) {
4504 ret = ff_h264_decode_mb_cavlc(h);
4507 ff_h264_hl_decode_mb(h);
4512 av_log(h->avctx, AV_LOG_ERROR,
4513 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
4514 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4515 h->mb_y, ER_MB_ERROR);
4519 if (++h->mb_x >= h->mb_width) {
4520 loop_filter(h, lf_x_start, h->mb_x);
4521 h->mb_x = lf_x_start = 0;
4522 decode_finish_row(h);
4524 if (FIELD_OR_MBAFF_PICTURE(h)) {
4526 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
4527 predict_field_decoding_flag(h);
4529 if (h->mb_y >= h->mb_height) {
4530 tprintf(h->avctx, "slice end %d %d\n",
4531 get_bits_count(&h->gb), h->gb.size_in_bits);
4533 if ( get_bits_left(&h->gb) == 0
4534 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
4535 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4536 h->mb_x - 1, h->mb_y,
4541 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4545 return AVERROR_INVALIDDATA;
4550 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
4551 tprintf(h->avctx, "slice end %d %d\n",
4552 get_bits_count(&h->gb), h->gb.size_in_bits);
4554 if (get_bits_left(&h->gb) == 0) {
4555 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
4556 h->mb_x - 1, h->mb_y,
4558 if (h->mb_x > lf_x_start)
4559 loop_filter(h, lf_x_start, h->mb_x);
4563 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
4564 h->mb_y, ER_MB_ERROR);
4566 return AVERROR_INVALIDDATA;
4574 * Call decode_slice() for each context.
4576 * @param h h264 master context
4577 * @param context_count number of contexts to execute
4579 static int execute_decode_slices(H264Context *h, int context_count)
4581 AVCodecContext *const avctx = h->avctx;
4585 if (h->avctx->hwaccel ||
4586 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4588 if (context_count == 1) {
4589 return decode_slice(avctx, &h);
4591 av_assert0(context_count > 0);
4592 for (i = 1; i < context_count; i++) {
4593 hx = h->thread_context[i];
4594 if (CONFIG_ERROR_RESILIENCE) {
4595 hx->er.error_count = 0;
4597 hx->x264_build = h->x264_build;
4600 avctx->execute(avctx, decode_slice, h->thread_context,
4601 NULL, context_count, sizeof(void *));
4603 /* pull back stuff from slices to master context */
4604 hx = h->thread_context[context_count - 1];
4607 h->droppable = hx->droppable;
4608 h->picture_structure = hx->picture_structure;
4609 if (CONFIG_ERROR_RESILIENCE) {
4610 for (i = 1; i < context_count; i++)
4611 h->er.error_count += h->thread_context[i]->er.error_count;
4618 static const uint8_t start_code[] = { 0x00, 0x00, 0x01 };
4620 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
4621 int parse_extradata)
4623 AVCodecContext *const avctx = h->avctx;
4624 H264Context *hx; ///< thread context
4628 int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
4629 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
4632 int first_slice = 0;
4635 h->nal_unit_type= 0;
4637 if(!h->slice_context_count)
4638 h->slice_context_count= 1;
4639 h->max_contexts = h->slice_context_count;
4640 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS)) {
4641 h->current_slice = 0;
4642 if (!h->first_field)
4643 h->cur_pic_ptr = NULL;
4644 ff_h264_reset_sei(h);
4647 if (h->nal_length_size == 4) {
4648 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
4650 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
4654 for (; pass <= 1; pass++) {
4657 next_avc = h->is_avc ? 0 : buf_size;
4667 if (buf_index >= next_avc) {
4668 if (buf_index >= buf_size - h->nal_length_size)
4671 for (i = 0; i < h->nal_length_size; i++)
4672 nalsize = (nalsize << 8) | buf[buf_index++];
4673 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
4674 av_log(h->avctx, AV_LOG_ERROR,
4675 "AVC: nal size %d\n", nalsize);
4678 next_avc = buf_index + nalsize;
4680 // start code prefix search
4681 for (; buf_index + 3 < next_avc; buf_index++)
4682 // This should always succeed in the first iteration.
4683 if (buf[buf_index] == 0 &&
4684 buf[buf_index + 1] == 0 &&
4685 buf[buf_index + 2] == 1)
4688 if (buf_index + 3 >= buf_size) {
4689 buf_index = buf_size;
4694 if (buf_index >= next_avc)
4698 hx = h->thread_context[context_count];
4700 ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
4701 &consumed, next_avc - buf_index);
4702 if (ptr == NULL || dst_length < 0) {
4706 i = buf_index + consumed;
4707 if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
4708 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
4709 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
4710 h->workaround_bugs |= FF_BUG_TRUNCATED;
4712 if (!(h->workaround_bugs & FF_BUG_TRUNCATED))
4713 while(dst_length > 0 && ptr[dst_length - 1] == 0)
4715 bit_length = !dst_length ? 0
4717 decode_rbsp_trailing(h, ptr + dst_length - 1));
4719 if (h->avctx->debug & FF_DEBUG_STARTCODE)
4720 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);
4722 if (h->is_avc && (nalsize != consumed) && nalsize)
4723 av_log(h->avctx, AV_LOG_DEBUG,
4724 "AVC: Consumed only %d bytes instead of %d\n",
4727 buf_index += consumed;
4731 /* packets can sometimes contain multiple PPS/SPS,
4732 * e.g. two PAFF field pictures in one packet, or a demuxer
4733 * which splits NALs strangely if so, when frame threading we
4734 * can't start the next thread until we've read all of them */
4735 switch (hx->nal_unit_type) {
4738 nals_needed = nal_index;
4743 init_get_bits(&hx->gb, ptr, bit_length);
4744 if (!get_ue_golomb(&hx->gb) || !first_slice)
4745 nals_needed = nal_index;
4747 first_slice = hx->nal_unit_type;
4753 switch (hx->nal_unit_type) {
4757 first_slice = hx->nal_unit_type;
4760 // FIXME do not discard SEI id
4761 if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
4765 /* Ignore per frame NAL unit type during extradata
4766 * parsing. Decoding slices is not possible in codec init
4768 if (parse_extradata) {
4769 switch (hx->nal_unit_type) {
4775 case NAL_AUXILIARY_SLICE:
4776 av_log(h->avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header/extradata\n", hx->nal_unit_type);
4777 hx->nal_unit_type = NAL_FF_IGNORE;
4783 switch (hx->nal_unit_type) {
4785 if (first_slice != NAL_IDR_SLICE) {
4786 av_log(h->avctx, AV_LOG_ERROR,
4787 "Invalid mix of idr and non-idr slices\n");
4792 idr(h); // FIXME ensure we don't lose some frames if there is reordering
4795 init_get_bits(&hx->gb, ptr, bit_length);
4797 hx->inter_gb_ptr = &hx->gb;
4798 hx->data_partitioning = 0;
4800 if ((err = decode_slice_header(hx, h)))
4803 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
4804 h->valid_recovery_point = 1;
4806 if ( h->sei_recovery_frame_cnt >= 0
4807 && ( h->recovery_frame<0
4808 || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
4809 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
4810 (1 << h->sps.log2_max_frame_num);
4812 if (!h->valid_recovery_point)
4813 h->recovery_frame = h->frame_num;
4816 h->cur_pic_ptr->f.key_frame |=
4817 (hx->nal_unit_type == NAL_IDR_SLICE);
4819 if (h->recovery_frame == h->frame_num) {
4820 h->cur_pic_ptr->sync |= 1;
4821 h->recovery_frame = -1;
4824 h->sync |= !!h->cur_pic_ptr->f.key_frame;
4825 h->sync |= 3*!!(avctx->flags2 & CODEC_FLAG2_SHOW_ALL);
4826 h->cur_pic_ptr->sync |= h->sync;
4828 if (h->current_slice == 1) {
4829 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS))
4830 decode_postinit(h, nal_index >= nals_needed);
4832 if (h->avctx->hwaccel &&
4833 (ret = h->avctx->hwaccel->start_frame(h->avctx, NULL, 0)) < 0)
4835 if (CONFIG_H264_VDPAU_DECODER &&
4836 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4837 ff_vdpau_h264_picture_start(h);
4840 if (hx->redundant_pic_count == 0 &&
4841 (avctx->skip_frame < AVDISCARD_NONREF ||
4843 (avctx->skip_frame < AVDISCARD_BIDIR ||
4844 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4845 (avctx->skip_frame < AVDISCARD_NONKEY ||
4846 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4847 avctx->skip_frame < AVDISCARD_ALL) {
4848 if (avctx->hwaccel) {
4849 ret = avctx->hwaccel->decode_slice(avctx,
4850 &buf[buf_index - consumed],
4854 } else if (CONFIG_H264_VDPAU_DECODER &&
4855 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
4856 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0],
4858 sizeof(start_code));
4859 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0],
4860 &buf[buf_index - consumed],
4867 init_get_bits(&hx->gb, ptr, bit_length);
4869 hx->inter_gb_ptr = NULL;
4871 if ((err = decode_slice_header(hx, h)) < 0)
4874 hx->data_partitioning = 1;
4877 init_get_bits(&hx->intra_gb, ptr, bit_length);
4878 hx->intra_gb_ptr = &hx->intra_gb;
4881 init_get_bits(&hx->inter_gb, ptr, bit_length);
4882 hx->inter_gb_ptr = &hx->inter_gb;
4884 av_log(h->avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
4887 if (hx->redundant_pic_count == 0 &&
4889 hx->data_partitioning &&
4890 h->cur_pic_ptr && h->context_initialized &&
4891 (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4892 (avctx->skip_frame < AVDISCARD_BIDIR ||
4893 hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4894 (avctx->skip_frame < AVDISCARD_NONKEY ||
4895 hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4896 avctx->skip_frame < AVDISCARD_ALL)
4900 init_get_bits(&h->gb, ptr, bit_length);
4901 ff_h264_decode_sei(h);
4904 init_get_bits(&h->gb, ptr, bit_length);
4905 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? nalsize : 1)) {
4906 av_log(h->avctx, AV_LOG_DEBUG,
4907 "SPS decoding failure, trying again with the complete NAL\n");
4909 av_assert0(next_avc - buf_index + consumed == nalsize);
4910 if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4912 init_get_bits(&h->gb, &buf[buf_index + 1 - consumed],
4913 8*(next_avc - buf_index + consumed - 1));
4914 ff_h264_decode_seq_parameter_set(h);
4919 init_get_bits(&h->gb, ptr, bit_length);
4920 ff_h264_decode_picture_parameter_set(h, bit_length);
4923 case NAL_END_SEQUENCE:
4924 case NAL_END_STREAM:
4925 case NAL_FILLER_DATA:
4927 case NAL_AUXILIARY_SLICE:
4932 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4933 hx->nal_unit_type, bit_length);
4936 if (context_count == h->max_contexts) {
4937 execute_decode_slices(h, context_count);
4942 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4943 else if (err == 1) {
4944 /* Slice could not be decoded in parallel mode, copy down
4945 * NAL unit stuff to context 0 and restart. Note that
4946 * rbsp_buffer is not transferred, but since we no longer
4947 * run in parallel mode this should not be an issue. */
4948 h->nal_unit_type = hx->nal_unit_type;
4949 h->nal_ref_idc = hx->nal_ref_idc;
4956 execute_decode_slices(h, context_count);
4960 if (h->cur_pic_ptr && !h->droppable) {
4961 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
4962 h->picture_structure == PICT_BOTTOM_FIELD);
4965 return (ret < 0) ? ret : buf_index;
4969 * Return the number of bytes consumed for building the current frame.
4971 static int get_consumed_bytes(int pos, int buf_size)
4974 pos = 1; // avoid infinite loops (i doubt that is needed but ...)
4975 if (pos + 10 > buf_size)
4976 pos = buf_size; // oops ;)
4981 static int output_frame(H264Context *h, AVFrame *dst, Picture *srcp)
4983 AVFrame *src = &srcp->f;
4985 int ret = av_frame_ref(dst, src);
4989 av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(h), 0);
4994 for (i = 0; i < 3; i++) {
4995 int hshift = (i > 0) ? h->chroma_x_shift : 0;
4996 int vshift = (i > 0) ? h->chroma_y_shift : 0;
4997 int off = ((srcp->crop_left >> hshift) << h->pixel_shift) +
4998 (srcp->crop_top >> vshift) * dst->linesize[i];
4999 dst->data[i] += off;
5004 static int decode_frame(AVCodecContext *avctx, void *data,
5005 int *got_frame, AVPacket *avpkt)
5007 const uint8_t *buf = avpkt->data;
5008 int buf_size = avpkt->size;
5009 H264Context *h = avctx->priv_data;
5010 AVFrame *pict = data;
5016 h->flags = avctx->flags;
5018 /* end of stream, output what is still in the buffers */
5019 if (buf_size == 0) {
5022 h->cur_pic_ptr = NULL;
5025 // FIXME factorize this with the output code below
5026 out = h->delayed_pic[0];
5029 h->delayed_pic[i] &&
5030 !h->delayed_pic[i]->f.key_frame &&
5031 !h->delayed_pic[i]->mmco_reset;
5033 if (h->delayed_pic[i]->poc < out->poc) {
5034 out = h->delayed_pic[i];
5038 for (i = out_idx; h->delayed_pic[i]; i++)
5039 h->delayed_pic[i] = h->delayed_pic[i + 1];
5042 out->reference &= ~DELAYED_PIC_REF;
5043 ret = output_frame(h, pict, out);
5051 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
5052 int cnt= buf[5]&0x1f;
5053 const uint8_t *p= buf+6;
5055 int nalsize= AV_RB16(p) + 2;
5056 if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
5064 int nalsize= AV_RB16(p) + 2;
5065 if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
5070 return ff_h264_decode_extradata(h, buf, buf_size);
5074 buf_index = decode_nal_units(h, buf, buf_size, 0);
5076 return AVERROR_INVALIDDATA;
5078 if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
5079 av_assert0(buf_index <= buf_size);
5083 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
5084 if (avctx->skip_frame >= AVDISCARD_NONREF ||
5085 buf_size >= 4 && !memcmp("Q264", buf, 4))
5087 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
5088 return AVERROR_INVALIDDATA;
5091 if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) ||
5092 (h->mb_y >= h->mb_height && h->mb_height)) {
5093 if (avctx->flags2 & CODEC_FLAG2_CHUNKS)
5094 decode_postinit(h, 1);
5098 /* Wait for second field. */
5100 if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
5101 ret = output_frame(h, pict, h->next_output_pic);
5105 if (CONFIG_MPEGVIDEO) {
5106 ff_print_debug_info2(h->avctx, h->next_output_pic, pict, h->er.mbskip_table,
5108 h->mb_width, h->mb_height, h->mb_stride, 1);
5113 assert(pict->data[0] || !*got_frame);
5115 return get_consumed_bytes(buf_index, buf_size);
5118 av_cold void ff_h264_free_context(H264Context *h)
5122 free_tables(h, 1); // FIXME cleanup init stuff perhaps
5124 for (i = 0; i < MAX_SPS_COUNT; i++)
5125 av_freep(h->sps_buffers + i);
5127 for (i = 0; i < MAX_PPS_COUNT; i++)
5128 av_freep(h->pps_buffers + i);
5131 static av_cold int h264_decode_end(AVCodecContext *avctx)
5133 H264Context *h = avctx->priv_data;
5135 ff_h264_remove_all_refs(h);
5136 ff_h264_free_context(h);
5138 unref_picture(h, &h->cur_pic);
5143 static const AVProfile profiles[] = {
5144 { FF_PROFILE_H264_BASELINE, "Baseline" },
5145 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
5146 { FF_PROFILE_H264_MAIN, "Main" },
5147 { FF_PROFILE_H264_EXTENDED, "Extended" },
5148 { FF_PROFILE_H264_HIGH, "High" },
5149 { FF_PROFILE_H264_HIGH_10, "High 10" },
5150 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
5151 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
5152 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
5153 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
5154 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
5155 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
5156 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
5157 { FF_PROFILE_UNKNOWN },
5160 static const AVOption h264_options[] = {
5161 {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
5162 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
5166 static const AVClass h264_class = {
5167 .class_name = "H264 Decoder",
5168 .item_name = av_default_item_name,
5169 .option = h264_options,
5170 .version = LIBAVUTIL_VERSION_INT,
5173 static const AVClass h264_vdpau_class = {
5174 .class_name = "H264 VDPAU Decoder",
5175 .item_name = av_default_item_name,
5176 .option = h264_options,
5177 .version = LIBAVUTIL_VERSION_INT,
5180 AVCodec ff_h264_decoder = {
5182 .type = AVMEDIA_TYPE_VIDEO,
5183 .id = AV_CODEC_ID_H264,
5184 .priv_data_size = sizeof(H264Context),
5185 .init = ff_h264_decode_init,
5186 .close = h264_decode_end,
5187 .decode = decode_frame,
5188 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
5189 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
5190 CODEC_CAP_FRAME_THREADS,
5192 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
5193 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
5194 .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
5195 .profiles = NULL_IF_CONFIG_SMALL(profiles),
5196 .priv_class = &h264_class,
5199 #if CONFIG_H264_VDPAU_DECODER
5200 AVCodec ff_h264_vdpau_decoder = {
5201 .name = "h264_vdpau",
5202 .type = AVMEDIA_TYPE_VIDEO,
5203 .id = AV_CODEC_ID_H264,
5204 .priv_data_size = sizeof(H264Context),
5205 .init = ff_h264_decode_init,
5206 .close = h264_decode_end,
5207 .decode = decode_frame,
5208 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
5210 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
5211 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
5213 .profiles = NULL_IF_CONFIG_SMALL(profiles),
5214 .priv_class = &h264_vdpau_class,