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 #include "libavutil/avassert.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/timer.h"
33 #include "cabac_functions.h"
34 #include "error_resilience.h"
38 #include "h264chroma.h"
39 #include "h264_mvpred.h"
42 #include "mpegutils.h"
43 #include "rectangle.h"
47 static const uint8_t rem6[QP_MAX_NUM + 1] = {
48 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
49 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
50 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
51 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
55 static const uint8_t div6[QP_MAX_NUM + 1] = {
56 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
57 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
58 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
59 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
63 static const uint8_t field_scan[16+1] = {
64 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
65 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
66 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
67 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
70 static const uint8_t field_scan8x8[64+1] = {
71 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
72 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
73 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
74 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
75 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
76 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
77 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
78 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
79 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
80 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
81 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
82 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
83 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
84 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
85 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
86 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
89 static const uint8_t field_scan8x8_cavlc[64+1] = {
90 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
91 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
92 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
93 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
94 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
95 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
96 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
97 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
98 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
99 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
100 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
101 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
102 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
103 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
104 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
105 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
108 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
109 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
110 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
111 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
112 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
113 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
114 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
115 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
116 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
117 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
118 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
119 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
120 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
121 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
122 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
123 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
124 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
125 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
128 static const uint8_t dequant4_coeff_init[6][3] = {
137 static const uint8_t dequant8_coeff_init_scan[16] = {
138 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
141 static const uint8_t dequant8_coeff_init[6][6] = {
142 { 20, 18, 32, 19, 25, 24 },
143 { 22, 19, 35, 21, 28, 26 },
144 { 26, 23, 42, 24, 33, 31 },
145 { 28, 25, 45, 26, 35, 33 },
146 { 32, 28, 51, 30, 40, 38 },
147 { 36, 32, 58, 34, 46, 43 },
151 static void release_unused_pictures(H264Context *h, int remove_current)
155 /* release non reference frames */
156 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
157 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
158 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
159 ff_h264_unref_picture(h, &h->DPB[i]);
164 static int alloc_scratch_buffers(H264Context *h, int linesize)
166 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
168 if (h->bipred_scratchpad)
171 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
172 // edge emu needs blocksize + filter length - 1
173 // (= 21x21 for h264)
174 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
176 if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
177 av_freep(&h->bipred_scratchpad);
178 av_freep(&h->edge_emu_buffer);
179 return AVERROR(ENOMEM);
185 static int init_table_pools(H264Context *h)
187 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
188 const int mb_array_size = h->mb_stride * h->mb_height;
189 const int b4_stride = h->mb_width * 4 + 1;
190 const int b4_array_size = b4_stride * h->mb_height * 4;
192 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
194 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
195 sizeof(uint32_t), av_buffer_allocz);
196 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
197 sizeof(int16_t), av_buffer_allocz);
198 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
200 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
201 !h->ref_index_pool) {
202 av_buffer_pool_uninit(&h->qscale_table_pool);
203 av_buffer_pool_uninit(&h->mb_type_pool);
204 av_buffer_pool_uninit(&h->motion_val_pool);
205 av_buffer_pool_uninit(&h->ref_index_pool);
206 return AVERROR(ENOMEM);
212 static int alloc_picture(H264Context *h, H264Picture *pic)
216 av_assert0(!pic->f.data[0]);
219 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
220 AV_GET_BUFFER_FLAG_REF : 0);
224 h->linesize = pic->f.linesize[0];
225 h->uvlinesize = pic->f.linesize[1];
226 pic->crop = h->sps.crop;
227 pic->crop_top = h->sps.crop_top;
228 pic->crop_left= h->sps.crop_left;
230 if (h->avctx->hwaccel) {
231 const AVHWAccel *hwaccel = h->avctx->hwaccel;
232 av_assert0(!pic->hwaccel_picture_private);
233 if (hwaccel->frame_priv_data_size) {
234 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
235 if (!pic->hwaccel_priv_buf)
236 return AVERROR(ENOMEM);
237 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
240 if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
241 int h_chroma_shift, v_chroma_shift;
242 av_pix_fmt_get_chroma_sub_sample(pic->f.format,
243 &h_chroma_shift, &v_chroma_shift);
245 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
246 memset(pic->f.data[1] + pic->f.linesize[1]*i,
247 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
248 memset(pic->f.data[2] + pic->f.linesize[2]*i,
249 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
253 if (!h->qscale_table_pool) {
254 ret = init_table_pools(h);
259 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
260 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
261 if (!pic->qscale_table_buf || !pic->mb_type_buf)
264 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
265 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
267 for (i = 0; i < 2; i++) {
268 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
269 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
270 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
273 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
274 pic->ref_index[i] = pic->ref_index_buf[i]->data;
279 ff_h264_unref_picture(h, pic);
280 return (ret < 0) ? ret : AVERROR(ENOMEM);
283 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
287 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
292 static int find_unused_picture(H264Context *h)
296 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
297 if (pic_is_unused(h, &h->DPB[i]))
300 if (i == H264_MAX_PICTURE_COUNT)
301 return AVERROR_INVALIDDATA;
303 if (h->DPB[i].needs_realloc) {
304 h->DPB[i].needs_realloc = 0;
305 ff_h264_unref_picture(h, &h->DPB[i]);
312 static void init_dequant8_coeff_table(H264Context *h)
315 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
317 for (i = 0; i < 6; i++) {
318 h->dequant8_coeff[i] = h->dequant8_buffer[i];
319 for (j = 0; j < i; j++)
320 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
321 64 * sizeof(uint8_t))) {
322 h->dequant8_coeff[i] = h->dequant8_buffer[j];
328 for (q = 0; q < max_qp + 1; q++) {
331 for (x = 0; x < 64; x++)
332 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
333 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
334 h->pps.scaling_matrix8[i][x]) << shift;
339 static void init_dequant4_coeff_table(H264Context *h)
342 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
343 for (i = 0; i < 6; i++) {
344 h->dequant4_coeff[i] = h->dequant4_buffer[i];
345 for (j = 0; j < i; j++)
346 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
347 16 * sizeof(uint8_t))) {
348 h->dequant4_coeff[i] = h->dequant4_buffer[j];
354 for (q = 0; q < max_qp + 1; q++) {
355 int shift = div6[q] + 2;
357 for (x = 0; x < 16; x++)
358 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
359 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
360 h->pps.scaling_matrix4[i][x]) << shift;
365 void h264_init_dequant_tables(H264Context *h)
368 init_dequant4_coeff_table(h);
369 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
371 if (h->pps.transform_8x8_mode)
372 init_dequant8_coeff_table(h);
373 if (h->sps.transform_bypass) {
374 for (i = 0; i < 6; i++)
375 for (x = 0; x < 16; x++)
376 h->dequant4_coeff[i][0][x] = 1 << 6;
377 if (h->pps.transform_8x8_mode)
378 for (i = 0; i < 6; i++)
379 for (x = 0; x < 64; x++)
380 h->dequant8_coeff[i][0][x] = 1 << 6;
385 * Mimic alloc_tables(), but for every context thread.
387 static void clone_tables(H264Context *dst, H264Context *src, int i)
389 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
390 dst->non_zero_count = src->non_zero_count;
391 dst->slice_table = src->slice_table;
392 dst->cbp_table = src->cbp_table;
393 dst->mb2b_xy = src->mb2b_xy;
394 dst->mb2br_xy = src->mb2br_xy;
395 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
396 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
397 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
398 dst->direct_table = src->direct_table;
399 dst->list_counts = src->list_counts;
401 dst->cur_pic_ptr = src->cur_pic_ptr;
402 dst->cur_pic = src->cur_pic;
403 dst->bipred_scratchpad = NULL;
404 dst->edge_emu_buffer = NULL;
405 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
406 src->sps.chroma_format_idc);
409 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
411 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
412 (((pic) && (pic) >= (old_ctx)->DPB && \
413 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
414 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
416 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
417 H264Context *new_base,
418 H264Context *old_base)
422 for (i = 0; i < count; i++) {
423 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
424 IN_RANGE(from[i], old_base->DPB,
425 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
427 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
431 static int copy_parameter_set(void **to, void **from, int count, int size)
435 for (i = 0; i < count; i++) {
436 if (to[i] && !from[i]) {
438 } else if (from[i] && !to[i]) {
439 to[i] = av_malloc(size);
441 return AVERROR(ENOMEM);
445 memcpy(to[i], from[i], size);
451 #define copy_fields(to, from, start_field, end_field) \
452 memcpy(&(to)->start_field, &(from)->start_field, \
453 (char *)&(to)->end_field - (char *)&(to)->start_field)
455 static int h264_slice_header_init(H264Context *h, int reinit);
457 int ff_h264_update_thread_context(AVCodecContext *dst,
458 const AVCodecContext *src)
460 H264Context *h = dst->priv_data, *h1 = src->priv_data;
461 int inited = h->context_initialized, err = 0;
462 int context_reinitialized = 0;
469 (h->width != h1->width ||
470 h->height != h1->height ||
471 h->mb_width != h1->mb_width ||
472 h->mb_height != h1->mb_height ||
473 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
474 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
475 h->sps.colorspace != h1->sps.colorspace)) {
477 /* set bits_per_raw_sample to the previous value. the check for changed
478 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
479 * the current value */
480 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
482 av_freep(&h->bipred_scratchpad);
484 h->width = h1->width;
485 h->height = h1->height;
486 h->mb_height = h1->mb_height;
487 h->mb_width = h1->mb_width;
488 h->mb_num = h1->mb_num;
489 h->mb_stride = h1->mb_stride;
490 h->b_stride = h1->b_stride;
492 if ((ret = copy_parameter_set((void **)h->sps_buffers,
493 (void **)h1->sps_buffers,
494 MAX_SPS_COUNT, sizeof(SPS))) < 0)
497 if ((ret = copy_parameter_set((void **)h->pps_buffers,
498 (void **)h1->pps_buffers,
499 MAX_PPS_COUNT, sizeof(PPS))) < 0)
503 if ((err = h264_slice_header_init(h, 1)) < 0) {
504 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
507 context_reinitialized = 1;
510 h264_set_parameter_from_sps(h);
511 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
512 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
515 /* update linesize on resize for h264. The h264 decoder doesn't
516 * necessarily call ff_mpv_frame_start in the new thread */
517 h->linesize = h1->linesize;
518 h->uvlinesize = h1->uvlinesize;
520 /* copy block_offset since frame_start may not be called */
521 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
524 for (i = 0; i < MAX_SPS_COUNT; i++)
525 av_freep(h->sps_buffers + i);
527 for (i = 0; i < MAX_PPS_COUNT; i++)
528 av_freep(h->pps_buffers + i);
530 av_freep(&h->rbsp_buffer[0]);
531 av_freep(&h->rbsp_buffer[1]);
532 ff_h264_unref_picture(h, &h->last_pic_for_ec);
533 memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
534 memcpy(&h->cabac, &h1->cabac,
535 sizeof(H264Context) - offsetof(H264Context, cabac));
536 av_assert0((void*)&h->cabac == &h->mb_padding + 1);
538 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
539 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
541 memset(&h->er, 0, sizeof(h->er));
542 memset(&h->mb, 0, sizeof(h->mb));
543 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
544 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
545 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
546 memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
550 h->qscale_table_pool = NULL;
551 h->mb_type_pool = NULL;
552 h->ref_index_pool = NULL;
553 h->motion_val_pool = NULL;
554 h->intra4x4_pred_mode= NULL;
555 h->non_zero_count = NULL;
556 h->slice_table_base = NULL;
557 h->slice_table = NULL;
559 h->chroma_pred_mode_table = NULL;
560 memset(h->mvd_table, 0, sizeof(h->mvd_table));
561 h->direct_table = NULL;
562 h->list_counts = NULL;
565 for (i = 0; i < 2; i++) {
566 h->rbsp_buffer[i] = NULL;
567 h->rbsp_buffer_size[i] = 0;
570 if (h1->context_initialized) {
571 h->context_initialized = 0;
573 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
574 av_frame_unref(&h->cur_pic.f);
575 h->cur_pic.tf.f = &h->cur_pic.f;
577 ret = ff_h264_alloc_tables(h);
579 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
582 ret = ff_h264_context_init(h);
584 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
589 h->bipred_scratchpad = NULL;
590 h->edge_emu_buffer = NULL;
592 h->thread_context[0] = h;
593 h->context_initialized = h1->context_initialized;
596 h->avctx->coded_height = h1->avctx->coded_height;
597 h->avctx->coded_width = h1->avctx->coded_width;
598 h->avctx->width = h1->avctx->width;
599 h->avctx->height = h1->avctx->height;
600 h->coded_picture_number = h1->coded_picture_number;
601 h->first_field = h1->first_field;
602 h->picture_structure = h1->picture_structure;
603 h->qscale = h1->qscale;
604 h->droppable = h1->droppable;
605 h->low_delay = h1->low_delay;
607 for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
608 ff_h264_unref_picture(h, &h->DPB[i]);
609 if (h1->DPB && h1->DPB[i].f.buf[0] &&
610 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
614 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
615 ff_h264_unref_picture(h, &h->cur_pic);
616 if (h1->cur_pic.f.buf[0] && (ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
619 h->workaround_bugs = h1->workaround_bugs;
620 h->low_delay = h1->low_delay;
621 h->droppable = h1->droppable;
623 // extradata/NAL handling
624 h->is_avc = h1->is_avc;
627 if ((ret = copy_parameter_set((void **)h->sps_buffers,
628 (void **)h1->sps_buffers,
629 MAX_SPS_COUNT, sizeof(SPS))) < 0)
632 if ((ret = copy_parameter_set((void **)h->pps_buffers,
633 (void **)h1->pps_buffers,
634 MAX_PPS_COUNT, sizeof(PPS))) < 0)
638 // Dequantization matrices
639 // FIXME these are big - can they be only copied when PPS changes?
640 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
642 for (i = 0; i < 6; i++)
643 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
644 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
646 for (i = 0; i < 6; i++)
647 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
648 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
650 h->dequant_coeff_pps = h1->dequant_coeff_pps;
653 copy_fields(h, h1, poc_lsb, redundant_pic_count);
656 copy_fields(h, h1, short_ref, cabac_init_idc);
658 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
659 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
660 copy_picture_range(h->delayed_pic, h1->delayed_pic,
661 MAX_DELAYED_PIC_COUNT + 2, h, h1);
663 h->frame_recovered = h1->frame_recovered;
665 if (context_reinitialized)
666 ff_h264_set_parameter_from_sps(h);
672 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
673 h->prev_poc_msb = h->poc_msb;
674 h->prev_poc_lsb = h->poc_lsb;
676 h->prev_frame_num_offset = h->frame_num_offset;
677 h->prev_frame_num = h->frame_num;
678 h->outputed_poc = h->next_outputed_poc;
680 h->recovery_frame = h1->recovery_frame;
685 static int h264_frame_start(H264Context *h)
689 const int pixel_shift = h->pixel_shift;
691 1<<(h->sps.bit_depth_luma-1),
692 1<<(h->sps.bit_depth_chroma-1),
693 1<<(h->sps.bit_depth_chroma-1),
697 if (!ff_thread_can_start_frame(h->avctx)) {
698 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
702 release_unused_pictures(h, 1);
703 h->cur_pic_ptr = NULL;
705 i = find_unused_picture(h);
707 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
712 pic->reference = h->droppable ? 0 : h->picture_structure;
713 pic->f.coded_picture_number = h->coded_picture_number++;
714 pic->field_picture = h->picture_structure != PICT_FRAME;
717 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
719 * See decode_nal_units().
721 pic->f.key_frame = 0;
724 pic->invalid_gap = 0;
725 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
727 if ((ret = alloc_picture(h, pic)) < 0)
729 if(!h->frame_recovered && !h->avctx->hwaccel &&
730 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
731 avpriv_color_frame(&pic->f, c);
733 h->cur_pic_ptr = pic;
734 ff_h264_unref_picture(h, &h->cur_pic);
735 if (CONFIG_ERROR_RESILIENCE) {
736 ff_h264_set_erpic(&h->er.cur_pic, NULL);
739 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
742 if (CONFIG_ERROR_RESILIENCE) {
743 ff_er_frame_start(&h->er);
744 ff_h264_set_erpic(&h->er.last_pic, NULL);
745 ff_h264_set_erpic(&h->er.next_pic, NULL);
748 assert(h->linesize && h->uvlinesize);
750 for (i = 0; i < 16; i++) {
751 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
752 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
754 for (i = 0; i < 16; i++) {
755 h->block_offset[16 + i] =
756 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
757 h->block_offset[48 + 16 + i] =
758 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
761 /* We mark the current picture as non-reference after allocating it, so
762 * that if we break out due to an error it can be released automatically
763 * in the next ff_mpv_frame_start().
765 h->cur_pic_ptr->reference = 0;
767 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
769 h->next_output_pic = NULL;
771 assert(h->cur_pic_ptr->long_ref == 0);
776 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
777 uint8_t *src_cb, uint8_t *src_cr,
778 int linesize, int uvlinesize,
783 const int pixel_shift = h->pixel_shift;
784 int chroma444 = CHROMA444(h);
785 int chroma422 = CHROMA422(h);
788 src_cb -= uvlinesize;
789 src_cr -= uvlinesize;
791 if (!simple && FRAME_MBAFF(h)) {
794 top_border = h->top_borders[0][h->mb_x];
795 AV_COPY128(top_border, src_y + 15 * linesize);
797 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
798 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
801 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
802 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
803 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
804 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
806 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
807 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
809 } else if (chroma422) {
811 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
812 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
814 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
815 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
819 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
820 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
822 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
823 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
828 } else if (MB_MBAFF(h)) {
834 top_border = h->top_borders[top_idx][h->mb_x];
835 /* There are two lines saved, the line above the top macroblock
836 * of a pair, and the line above the bottom macroblock. */
837 AV_COPY128(top_border, src_y + 16 * linesize);
839 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
841 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
844 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
845 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
846 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
847 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
849 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
850 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
852 } else if (chroma422) {
854 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
855 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
857 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
858 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
862 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
863 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
865 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
866 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
873 * Initialize implicit_weight table.
874 * @param field 0/1 initialize the weight for interlaced MBAFF
875 * -1 initializes the rest
877 static void implicit_weight_table(H264Context *h, int field)
879 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
881 for (i = 0; i < 2; i++) {
882 h->luma_weight_flag[i] = 0;
883 h->chroma_weight_flag[i] = 0;
887 if (h->picture_structure == PICT_FRAME) {
888 cur_poc = h->cur_pic_ptr->poc;
890 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
892 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
893 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
895 h->use_weight_chroma = 0;
899 ref_count0 = h->ref_count[0];
900 ref_count1 = h->ref_count[1];
902 cur_poc = h->cur_pic_ptr->field_poc[field];
904 ref_count0 = 16 + 2 * h->ref_count[0];
905 ref_count1 = 16 + 2 * h->ref_count[1];
909 h->use_weight_chroma = 2;
910 h->luma_log2_weight_denom = 5;
911 h->chroma_log2_weight_denom = 5;
913 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
914 int poc0 = h->ref_list[0][ref0].poc;
915 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
917 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
918 int poc1 = h->ref_list[1][ref1].poc;
919 int td = av_clip(poc1 - poc0, -128, 127);
921 int tb = av_clip(cur_poc - poc0, -128, 127);
922 int tx = (16384 + (FFABS(td) >> 1)) / td;
923 int dist_scale_factor = (tb * tx + 32) >> 8;
924 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
925 w = 64 - dist_scale_factor;
929 h->implicit_weight[ref0][ref1][0] =
930 h->implicit_weight[ref0][ref1][1] = w;
932 h->implicit_weight[ref0][ref1][field] = w;
939 * initialize scan tables
941 static void init_scan_tables(H264Context *h)
944 for (i = 0; i < 16; i++) {
945 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
946 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
947 h->field_scan[i] = TRANSPOSE(field_scan[i]);
950 for (i = 0; i < 64; i++) {
951 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
952 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
953 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
954 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
955 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
958 if (h->sps.transform_bypass) { // FIXME same ugly
959 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
960 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
961 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
962 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
963 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
964 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
966 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
967 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
968 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
969 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
970 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
971 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
976 * Replicate H264 "master" context to thread contexts.
978 static int clone_slice(H264Context *dst, H264Context *src)
980 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
981 dst->cur_pic_ptr = src->cur_pic_ptr;
982 dst->cur_pic = src->cur_pic;
983 dst->linesize = src->linesize;
984 dst->uvlinesize = src->uvlinesize;
985 dst->first_field = src->first_field;
987 dst->prev_poc_msb = src->prev_poc_msb;
988 dst->prev_poc_lsb = src->prev_poc_lsb;
989 dst->prev_frame_num_offset = src->prev_frame_num_offset;
990 dst->prev_frame_num = src->prev_frame_num;
991 dst->short_ref_count = src->short_ref_count;
993 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
994 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
995 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
997 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
998 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1003 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1005 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
1006 CONFIG_H264_VAAPI_HWACCEL + \
1007 (CONFIG_H264_VDA_HWACCEL * 2) + \
1008 CONFIG_H264_VDPAU_HWACCEL)
1009 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
1010 const enum AVPixelFormat *choices = pix_fmts;
1013 switch (h->sps.bit_depth_luma) {
1016 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1017 *fmt++ = AV_PIX_FMT_GBRP9;
1019 *fmt++ = AV_PIX_FMT_YUV444P9;
1020 } else if (CHROMA422(h))
1021 *fmt++ = AV_PIX_FMT_YUV422P9;
1023 *fmt++ = AV_PIX_FMT_YUV420P9;
1027 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1028 *fmt++ = AV_PIX_FMT_GBRP10;
1030 *fmt++ = AV_PIX_FMT_YUV444P10;
1031 } else if (CHROMA422(h))
1032 *fmt++ = AV_PIX_FMT_YUV422P10;
1034 *fmt++ = AV_PIX_FMT_YUV420P10;
1038 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1039 *fmt++ = AV_PIX_FMT_GBRP12;
1041 *fmt++ = AV_PIX_FMT_YUV444P12;
1042 } else if (CHROMA422(h))
1043 *fmt++ = AV_PIX_FMT_YUV422P12;
1045 *fmt++ = AV_PIX_FMT_YUV420P12;
1049 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1050 *fmt++ = AV_PIX_FMT_GBRP14;
1052 *fmt++ = AV_PIX_FMT_YUV444P14;
1053 } else if (CHROMA422(h))
1054 *fmt++ = AV_PIX_FMT_YUV422P14;
1056 *fmt++ = AV_PIX_FMT_YUV420P14;
1059 #if CONFIG_H264_VDPAU_HWACCEL
1060 *fmt++ = AV_PIX_FMT_VDPAU;
1063 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1064 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1065 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1066 *fmt++ = AV_PIX_FMT_GBRP;
1067 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1068 *fmt++ = AV_PIX_FMT_YUVJ444P;
1070 *fmt++ = AV_PIX_FMT_YUV444P;
1071 } else if (CHROMA422(h)) {
1072 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1073 *fmt++ = AV_PIX_FMT_YUVJ422P;
1075 *fmt++ = AV_PIX_FMT_YUV422P;
1077 #if CONFIG_H264_DXVA2_HWACCEL
1078 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1080 #if CONFIG_H264_VAAPI_HWACCEL
1081 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1083 #if CONFIG_H264_VDA_HWACCEL
1084 *fmt++ = AV_PIX_FMT_VDA_VLD;
1085 *fmt++ = AV_PIX_FMT_VDA;
1087 if (h->avctx->codec->pix_fmts)
1088 choices = h->avctx->codec->pix_fmts;
1089 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1090 *fmt++ = AV_PIX_FMT_YUVJ420P;
1092 *fmt++ = AV_PIX_FMT_YUV420P;
1096 av_log(h->avctx, AV_LOG_ERROR,
1097 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1098 return AVERROR_INVALIDDATA;
1101 *fmt = AV_PIX_FMT_NONE;
1103 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1104 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1106 return ff_thread_get_format(h->avctx, choices);
1109 /* export coded and cropped frame dimensions to AVCodecContext */
1110 static int init_dimensions(H264Context *h)
1112 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1113 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1114 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1115 h->sps.crop_right || h->sps.crop_bottom;
1116 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1117 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1119 /* handle container cropping */
1120 if (!crop_present &&
1121 FFALIGN(h->avctx->width, 16) == h->width &&
1122 FFALIGN(h->avctx->height, 16) == h->height) {
1123 width = h->avctx->width;
1124 height = h->avctx->height;
1127 if (width <= 0 || height <= 0) {
1128 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1130 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1131 return AVERROR_INVALIDDATA;
1133 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1134 h->sps.crop_bottom =
1144 h->avctx->coded_width = h->width;
1145 h->avctx->coded_height = h->height;
1146 h->avctx->width = width;
1147 h->avctx->height = height;
1152 static int h264_slice_header_init(H264Context *h, int reinit)
1154 int nb_slices = (HAVE_THREADS &&
1155 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1156 h->avctx->thread_count : 1;
1159 ff_set_sar(h->avctx, h->sps.sar);
1160 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1161 &h->chroma_x_shift, &h->chroma_y_shift);
1163 if (h->sps.timing_info_present_flag) {
1164 int64_t den = h->sps.time_scale;
1165 if (h->x264_build < 44U)
1167 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1168 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1172 ff_h264_free_tables(h, 0);
1174 h->prev_interlaced_frame = 1;
1176 init_scan_tables(h);
1177 ret = ff_h264_alloc_tables(h);
1179 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1183 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1186 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1188 max_slices = H264_MAX_THREADS;
1189 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1190 " reducing to %d\n", nb_slices, max_slices);
1191 nb_slices = max_slices;
1193 h->slice_context_count = nb_slices;
1195 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1196 ret = ff_h264_context_init(h);
1198 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1202 for (i = 1; i < h->slice_context_count; i++) {
1204 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1206 ret = AVERROR(ENOMEM);
1209 c->avctx = h->avctx;
1211 c->h264dsp = h->h264dsp;
1212 c->h264qpel = h->h264qpel;
1213 c->h264chroma = h->h264chroma;
1216 c->pixel_shift = h->pixel_shift;
1217 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
1218 c->width = h->width;
1219 c->height = h->height;
1220 c->linesize = h->linesize;
1221 c->uvlinesize = h->uvlinesize;
1222 c->chroma_x_shift = h->chroma_x_shift;
1223 c->chroma_y_shift = h->chroma_y_shift;
1224 c->qscale = h->qscale;
1225 c->droppable = h->droppable;
1226 c->low_delay = h->low_delay;
1227 c->mb_width = h->mb_width;
1228 c->mb_height = h->mb_height;
1229 c->mb_stride = h->mb_stride;
1230 c->mb_num = h->mb_num;
1231 c->flags = h->flags;
1232 c->workaround_bugs = h->workaround_bugs;
1233 c->pict_type = h->pict_type;
1235 init_scan_tables(c);
1236 clone_tables(c, h, i);
1237 c->context_initialized = 1;
1240 for (i = 0; i < h->slice_context_count; i++)
1241 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1242 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1247 h->context_initialized = 1;
1251 ff_h264_free_tables(h, 0);
1252 h->context_initialized = 0;
1256 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1259 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1260 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1261 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1268 * Decode a slice header.
1269 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1271 * @param h h264context
1272 * @param h0 h264 master context (differs from 'h' when doing sliced based
1273 * parallel decoding)
1275 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1277 int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1279 unsigned int first_mb_in_slice;
1280 unsigned int pps_id;
1282 unsigned int slice_type, tmp, i, j;
1283 int last_pic_structure, last_pic_droppable;
1285 int needs_reinit = 0;
1286 int field_pic_flag, bottom_field_flag;
1287 int first_slice = h == h0 && !h0->current_slice;
1288 int frame_num, picture_structure, droppable;
1291 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1292 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1294 first_mb_in_slice = get_ue_golomb_long(&h->gb);
1296 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1297 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1298 ff_h264_field_end(h, 1);
1301 h0->current_slice = 0;
1302 if (!h0->first_field) {
1303 if (h->cur_pic_ptr && !h->droppable) {
1304 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1305 h->picture_structure == PICT_BOTTOM_FIELD);
1307 h->cur_pic_ptr = NULL;
1311 slice_type = get_ue_golomb_31(&h->gb);
1312 if (slice_type > 9) {
1313 av_log(h->avctx, AV_LOG_ERROR,
1314 "slice type %d too large at %d %d\n",
1315 slice_type, h->mb_x, h->mb_y);
1316 return AVERROR_INVALIDDATA;
1318 if (slice_type > 4) {
1320 h->slice_type_fixed = 1;
1322 h->slice_type_fixed = 0;
1324 slice_type = golomb_to_pict_type[slice_type];
1325 h->slice_type = slice_type;
1326 h->slice_type_nos = slice_type & 3;
1328 if (h->nal_unit_type == NAL_IDR_SLICE &&
1329 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1330 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1331 return AVERROR_INVALIDDATA;
1335 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1336 (h->avctx->skip_frame >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1337 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1338 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1339 h->avctx->skip_frame >= AVDISCARD_ALL) {
1340 return SLICE_SKIPED;
1343 // to make a few old functions happy, it's wrong though
1344 h->pict_type = h->slice_type;
1346 pps_id = get_ue_golomb(&h->gb);
1347 if (pps_id >= MAX_PPS_COUNT) {
1348 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1349 return AVERROR_INVALIDDATA;
1351 if (!h0->pps_buffers[pps_id]) {
1352 av_log(h->avctx, AV_LOG_ERROR,
1353 "non-existing PPS %u referenced\n",
1355 return AVERROR_INVALIDDATA;
1357 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1358 av_log(h->avctx, AV_LOG_ERROR,
1359 "PPS change from %d to %d forbidden\n",
1360 h0->au_pps_id, pps_id);
1361 return AVERROR_INVALIDDATA;
1364 pps = h0->pps_buffers[pps_id];
1366 if (!h0->sps_buffers[pps->sps_id]) {
1367 av_log(h->avctx, AV_LOG_ERROR,
1368 "non-existing SPS %u referenced\n",
1370 return AVERROR_INVALIDDATA;
1373 h->pps = *h0->pps_buffers[pps_id];
1375 if (pps->sps_id != h->sps.sps_id ||
1376 pps->sps_id != h->current_sps_id ||
1377 h0->sps_buffers[pps->sps_id]->new) {
1380 av_log(h->avctx, AV_LOG_ERROR,
1381 "SPS changed in the middle of the frame\n");
1382 return AVERROR_INVALIDDATA;
1385 h->sps = *h0->sps_buffers[h->pps.sps_id];
1387 if (h->mb_width != h->sps.mb_width ||
1388 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1389 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1390 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1394 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1395 h->chroma_format_idc != h->sps.chroma_format_idc) {
1396 h->bit_depth_luma = h->sps.bit_depth_luma;
1397 h->chroma_format_idc = h->sps.chroma_format_idc;
1400 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1404 h->avctx->profile = ff_h264_get_profile(&h->sps);
1405 h->avctx->level = h->sps.level_idc;
1406 h->avctx->refs = h->sps.ref_frame_count;
1408 must_reinit = (h->context_initialized &&
1409 ( 16*h->sps.mb_width != h->avctx->coded_width
1410 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1411 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1412 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1413 || h->mb_width != h->sps.mb_width
1414 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1416 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1419 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1422 h->mb_width = h->sps.mb_width;
1423 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1424 h->mb_num = h->mb_width * h->mb_height;
1425 h->mb_stride = h->mb_width + 1;
1427 h->b_stride = h->mb_width * 4;
1429 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1431 h->width = 16 * h->mb_width;
1432 h->height = 16 * h->mb_height;
1434 ret = init_dimensions(h);
1438 if (h->sps.video_signal_type_present_flag) {
1439 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1441 if (h->sps.colour_description_present_flag) {
1442 if (h->avctx->colorspace != h->sps.colorspace)
1444 h->avctx->color_primaries = h->sps.color_primaries;
1445 h->avctx->color_trc = h->sps.color_trc;
1446 h->avctx->colorspace = h->sps.colorspace;
1450 if (h->context_initialized &&
1451 (must_reinit || needs_reinit)) {
1453 av_log(h->avctx, AV_LOG_ERROR,
1454 "changing width %d -> %d / height %d -> %d on "
1456 h->width, h->avctx->coded_width,
1457 h->height, h->avctx->coded_height,
1458 h0->current_slice + 1);
1459 return AVERROR_INVALIDDATA;
1462 av_assert1(first_slice);
1464 ff_h264_flush_change(h);
1466 if ((ret = get_pixel_format(h, 1)) < 0)
1468 h->avctx->pix_fmt = ret;
1470 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1471 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1473 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1474 av_log(h->avctx, AV_LOG_ERROR,
1475 "h264_slice_header_init() failed\n");
1479 if (!h->context_initialized) {
1481 av_log(h->avctx, AV_LOG_ERROR,
1482 "Cannot (re-)initialize context during parallel decoding.\n");
1483 return AVERROR_PATCHWELCOME;
1486 if ((ret = get_pixel_format(h, 1)) < 0)
1488 h->avctx->pix_fmt = ret;
1490 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1491 av_log(h->avctx, AV_LOG_ERROR,
1492 "h264_slice_header_init() failed\n");
1497 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1498 h->dequant_coeff_pps = pps_id;
1499 h264_init_dequant_tables(h);
1502 frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1504 if (h0->frame_num != frame_num) {
1505 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1506 h0->frame_num, frame_num);
1507 return AVERROR_INVALIDDATA;
1512 h->mb_aff_frame = 0;
1513 last_pic_structure = h0->picture_structure;
1514 last_pic_droppable = h0->droppable;
1515 droppable = h->nal_ref_idc == 0;
1516 if (h->sps.frame_mbs_only_flag) {
1517 picture_structure = PICT_FRAME;
1519 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1520 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1523 field_pic_flag = get_bits1(&h->gb);
1525 if (field_pic_flag) {
1526 bottom_field_flag = get_bits1(&h->gb);
1527 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1529 picture_structure = PICT_FRAME;
1530 h->mb_aff_frame = h->sps.mb_aff;
1533 if (h0->current_slice) {
1534 if (last_pic_structure != picture_structure ||
1535 last_pic_droppable != droppable) {
1536 av_log(h->avctx, AV_LOG_ERROR,
1537 "Changing field mode (%d -> %d) between slices is not allowed\n",
1538 last_pic_structure, h->picture_structure);
1539 return AVERROR_INVALIDDATA;
1540 } else if (!h0->cur_pic_ptr) {
1541 av_log(h->avctx, AV_LOG_ERROR,
1542 "unset cur_pic_ptr on slice %d\n",
1543 h0->current_slice + 1);
1544 return AVERROR_INVALIDDATA;
1548 h->picture_structure = picture_structure;
1549 h->droppable = droppable;
1550 h->frame_num = frame_num;
1551 h->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1553 if (h0->current_slice == 0) {
1554 /* Shorten frame num gaps so we don't have to allocate reference
1555 * frames just to throw them away */
1556 if (h->frame_num != h->prev_frame_num) {
1557 int unwrap_prev_frame_num = h->prev_frame_num;
1558 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1560 if (unwrap_prev_frame_num > h->frame_num)
1561 unwrap_prev_frame_num -= max_frame_num;
1563 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1564 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1565 if (unwrap_prev_frame_num < 0)
1566 unwrap_prev_frame_num += max_frame_num;
1568 h->prev_frame_num = unwrap_prev_frame_num;
1572 /* See if we have a decoded first field looking for a pair...
1573 * Here, we're using that to see if we should mark previously
1574 * decode frames as "finished".
1575 * We have to do that before the "dummy" in-between frame allocation,
1576 * since that can modify h->cur_pic_ptr. */
1577 if (h0->first_field) {
1578 assert(h0->cur_pic_ptr);
1579 assert(h0->cur_pic_ptr->f.buf[0]);
1580 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1582 /* Mark old field/frame as completed */
1583 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1584 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1585 last_pic_structure == PICT_BOTTOM_FIELD);
1588 /* figure out if we have a complementary field pair */
1589 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1590 /* Previous field is unmatched. Don't display it, but let it
1591 * remain for reference if marked as such. */
1592 if (last_pic_structure != PICT_FRAME) {
1593 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1594 last_pic_structure == PICT_TOP_FIELD);
1597 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1598 /* This and previous field were reference, but had
1599 * different frame_nums. Consider this field first in
1600 * pair. Throw away previous field except for reference
1602 if (last_pic_structure != PICT_FRAME) {
1603 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1604 last_pic_structure == PICT_TOP_FIELD);
1607 /* Second field in complementary pair */
1608 if (!((last_pic_structure == PICT_TOP_FIELD &&
1609 h->picture_structure == PICT_BOTTOM_FIELD) ||
1610 (last_pic_structure == PICT_BOTTOM_FIELD &&
1611 h->picture_structure == PICT_TOP_FIELD))) {
1612 av_log(h->avctx, AV_LOG_ERROR,
1613 "Invalid field mode combination %d/%d\n",
1614 last_pic_structure, h->picture_structure);
1615 h->picture_structure = last_pic_structure;
1616 h->droppable = last_pic_droppable;
1617 return AVERROR_INVALIDDATA;
1618 } else if (last_pic_droppable != h->droppable) {
1619 avpriv_request_sample(h->avctx,
1620 "Found reference and non-reference fields in the same frame, which");
1621 h->picture_structure = last_pic_structure;
1622 h->droppable = last_pic_droppable;
1623 return AVERROR_PATCHWELCOME;
1629 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1630 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1631 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1632 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1633 h->frame_num, h->prev_frame_num);
1634 if (!h->sps.gaps_in_frame_num_allowed_flag)
1635 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1636 h->last_pocs[i] = INT_MIN;
1637 ret = h264_frame_start(h);
1639 h0->first_field = 0;
1643 h->prev_frame_num++;
1644 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1645 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1646 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1647 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1648 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1649 ret = ff_generate_sliding_window_mmcos(h, 1);
1650 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1652 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1653 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1655 /* Error concealment: If a ref is missing, copy the previous ref
1657 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1658 * many assumptions about there being no actual duplicates.
1659 * FIXME: This does not copy padding for out-of-frame motion
1660 * vectors. Given we are concealing a lost frame, this probably
1661 * is not noticeable by comparison, but it should be fixed. */
1662 if (h->short_ref_count) {
1664 av_image_copy(h->short_ref[0]->f.data,
1665 h->short_ref[0]->f.linesize,
1666 (const uint8_t **)prev->f.data,
1671 h->short_ref[0]->poc = prev->poc + 2;
1673 h->short_ref[0]->frame_num = h->prev_frame_num;
1677 /* See if we have a decoded first field looking for a pair...
1678 * We're using that to see whether to continue decoding in that
1679 * frame, or to allocate a new one. */
1680 if (h0->first_field) {
1681 assert(h0->cur_pic_ptr);
1682 assert(h0->cur_pic_ptr->f.buf[0]);
1683 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1685 /* figure out if we have a complementary field pair */
1686 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1687 /* Previous field is unmatched. Don't display it, but let it
1688 * remain for reference if marked as such. */
1689 h0->missing_fields ++;
1690 h0->cur_pic_ptr = NULL;
1691 h0->first_field = FIELD_PICTURE(h);
1693 h0->missing_fields = 0;
1694 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1695 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1696 h0->picture_structure==PICT_BOTTOM_FIELD);
1697 /* This and the previous field had different frame_nums.
1698 * Consider this field first in pair. Throw away previous
1699 * one except for reference purposes. */
1700 h0->first_field = 1;
1701 h0->cur_pic_ptr = NULL;
1703 /* Second field in complementary pair */
1704 h0->first_field = 0;
1708 /* Frame or first field in a potentially complementary pair */
1709 h0->first_field = FIELD_PICTURE(h);
1712 if (!FIELD_PICTURE(h) || h0->first_field) {
1713 if (h264_frame_start(h) < 0) {
1714 h0->first_field = 0;
1715 return AVERROR_INVALIDDATA;
1718 release_unused_pictures(h, 0);
1720 /* Some macroblocks can be accessed before they're available in case
1721 * of lost slices, MBAFF or threading. */
1722 if (FIELD_PICTURE(h)) {
1723 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1724 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1726 memset(h->slice_table, -1,
1727 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1729 h0->last_slice_type = -1;
1731 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1734 /* can't be in alloc_tables because linesize isn't known there.
1735 * FIXME: redo bipred weight to not require extra buffer? */
1736 for (i = 0; i < h->slice_context_count; i++)
1737 if (h->thread_context[i]) {
1738 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
1743 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1745 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1746 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1747 first_mb_in_slice >= h->mb_num) {
1748 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1749 return AVERROR_INVALIDDATA;
1751 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1752 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1753 FIELD_OR_MBAFF_PICTURE(h);
1754 if (h->picture_structure == PICT_BOTTOM_FIELD)
1755 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1756 av_assert1(h->mb_y < h->mb_height);
1758 if (h->picture_structure == PICT_FRAME) {
1759 h->curr_pic_num = h->frame_num;
1760 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1762 h->curr_pic_num = 2 * h->frame_num + 1;
1763 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1766 if (h->nal_unit_type == NAL_IDR_SLICE)
1767 get_ue_golomb(&h->gb); /* idr_pic_id */
1769 if (h->sps.poc_type == 0) {
1770 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1772 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1773 h->delta_poc_bottom = get_se_golomb(&h->gb);
1776 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1777 h->delta_poc[0] = get_se_golomb(&h->gb);
1779 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1780 h->delta_poc[1] = get_se_golomb(&h->gb);
1783 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1785 if (h->pps.redundant_pic_cnt_present)
1786 h->redundant_pic_count = get_ue_golomb(&h->gb);
1788 ret = ff_set_ref_count(h);
1792 if (slice_type != AV_PICTURE_TYPE_I &&
1793 (h0->current_slice == 0 ||
1794 slice_type != h0->last_slice_type ||
1795 memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
1797 ff_h264_fill_default_ref_list(h);
1800 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1801 ret = ff_h264_decode_ref_pic_list_reordering(h);
1803 h->ref_count[1] = h->ref_count[0] = 0;
1808 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1809 (h->pps.weighted_bipred_idc == 1 &&
1810 h->slice_type_nos == AV_PICTURE_TYPE_B))
1811 ff_pred_weight_table(h);
1812 else if (h->pps.weighted_bipred_idc == 2 &&
1813 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1814 implicit_weight_table(h, -1);
1817 for (i = 0; i < 2; i++) {
1818 h->luma_weight_flag[i] = 0;
1819 h->chroma_weight_flag[i] = 0;
1823 // If frame-mt is enabled, only update mmco tables for the first slice
1824 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1825 // or h->mmco, which will cause ref list mix-ups and decoding errors
1826 // further down the line. This may break decoding if the first slice is
1827 // corrupt, thus we only do this if frame-mt is enabled.
1828 if (h->nal_ref_idc) {
1829 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1830 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1831 h0->current_slice == 0);
1832 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1833 return AVERROR_INVALIDDATA;
1836 if (FRAME_MBAFF(h)) {
1837 ff_h264_fill_mbaff_ref_list(h);
1839 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1840 implicit_weight_table(h, 0);
1841 implicit_weight_table(h, 1);
1845 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1846 ff_h264_direct_dist_scale_factor(h);
1847 ff_h264_direct_ref_list_init(h);
1849 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1850 tmp = get_ue_golomb_31(&h->gb);
1852 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1853 return AVERROR_INVALIDDATA;
1855 h->cabac_init_idc = tmp;
1858 h->last_qscale_diff = 0;
1859 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1860 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1861 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1862 return AVERROR_INVALIDDATA;
1865 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1866 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1867 // FIXME qscale / qp ... stuff
1868 if (h->slice_type == AV_PICTURE_TYPE_SP)
1869 get_bits1(&h->gb); /* sp_for_switch_flag */
1870 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1871 h->slice_type == AV_PICTURE_TYPE_SI)
1872 get_se_golomb(&h->gb); /* slice_qs_delta */
1874 h->deblocking_filter = 1;
1875 h->slice_alpha_c0_offset = 0;
1876 h->slice_beta_offset = 0;
1877 if (h->pps.deblocking_filter_parameters_present) {
1878 tmp = get_ue_golomb_31(&h->gb);
1880 av_log(h->avctx, AV_LOG_ERROR,
1881 "deblocking_filter_idc %u out of range\n", tmp);
1882 return AVERROR_INVALIDDATA;
1884 h->deblocking_filter = tmp;
1885 if (h->deblocking_filter < 2)
1886 h->deblocking_filter ^= 1; // 1<->0
1888 if (h->deblocking_filter) {
1889 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1890 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1891 if (h->slice_alpha_c0_offset > 12 ||
1892 h->slice_alpha_c0_offset < -12 ||
1893 h->slice_beta_offset > 12 ||
1894 h->slice_beta_offset < -12) {
1895 av_log(h->avctx, AV_LOG_ERROR,
1896 "deblocking filter parameters %d %d out of range\n",
1897 h->slice_alpha_c0_offset, h->slice_beta_offset);
1898 return AVERROR_INVALIDDATA;
1903 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1904 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1905 h->nal_unit_type != NAL_IDR_SLICE) ||
1906 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1907 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1908 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1909 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1910 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1911 h->nal_ref_idc == 0))
1912 h->deblocking_filter = 0;
1914 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1915 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1916 /* Cheat slightly for speed:
1917 * Do not bother to deblock across slices. */
1918 h->deblocking_filter = 2;
1920 h0->max_contexts = 1;
1921 if (!h0->single_decode_warning) {
1922 av_log(h->avctx, AV_LOG_INFO,
1923 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1924 "To parallelize slice decoding you need video encoded with disable_deblocking_filter_idc set to 2 (deblock only edges that do not cross slices).\n"
1925 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1926 "but will generate non-standard-compliant output.\n");
1927 h0->single_decode_warning = 1;
1930 av_log(h->avctx, AV_LOG_ERROR,
1931 "Deblocking switched inside frame.\n");
1932 return SLICE_SINGLETHREAD;
1937 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1939 h->pps.chroma_qp_index_offset[0],
1940 h->pps.chroma_qp_index_offset[1]) +
1941 6 * (h->sps.bit_depth_luma - 8);
1943 h0->last_slice_type = slice_type;
1944 memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
1945 h->slice_num = ++h0->current_slice;
1948 h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
1949 if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
1950 && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
1951 && h->slice_num >= MAX_SLICES) {
1952 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1953 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);
1956 for (j = 0; j < 2; j++) {
1958 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1959 for (i = 0; i < 16; i++) {
1961 if (j < h->list_count && i < h->ref_count[j] &&
1962 h->ref_list[j][i].f.buf[0]) {
1964 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1965 for (k = 0; k < h->short_ref_count; k++)
1966 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1970 for (k = 0; k < h->long_ref_count; k++)
1971 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1972 id_list[i] = h->short_ref_count + k;
1980 for (i = 0; i < 16; i++)
1981 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1983 ref2frm[18 + 1] = -1;
1984 for (i = 16; i < 48; i++)
1985 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1986 (h->ref_list[j][i].reference & 3);
1989 h0->au_pps_id = pps_id;
1991 h0->sps_buffers[h->pps.sps_id]->new = 0;
1992 h->current_sps_id = h->pps.sps_id;
1994 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1995 av_log(h->avctx, AV_LOG_DEBUG,
1996 "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",
1998 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
2000 av_get_picture_type_char(h->slice_type),
2001 h->slice_type_fixed ? " fix" : "",
2002 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2003 pps_id, h->frame_num,
2004 h->cur_pic_ptr->field_poc[0],
2005 h->cur_pic_ptr->field_poc[1],
2006 h->ref_count[0], h->ref_count[1],
2008 h->deblocking_filter,
2009 h->slice_alpha_c0_offset, h->slice_beta_offset,
2011 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
2012 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2018 int ff_h264_get_slice_type(const H264Context *h)
2020 switch (h->slice_type) {
2021 case AV_PICTURE_TYPE_P:
2023 case AV_PICTURE_TYPE_B:
2025 case AV_PICTURE_TYPE_I:
2027 case AV_PICTURE_TYPE_SP:
2029 case AV_PICTURE_TYPE_SI:
2032 return AVERROR_INVALIDDATA;
2036 static av_always_inline void fill_filter_caches_inter(H264Context *h,
2037 int mb_type, int top_xy,
2038 int left_xy[LEFT_MBS],
2040 int left_type[LEFT_MBS],
2041 int mb_xy, int list)
2043 int b_stride = h->b_stride;
2044 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
2045 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
2046 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2047 if (USES_LIST(top_type, list)) {
2048 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2049 const int b8_xy = 4 * top_xy + 2;
2050 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2051 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2052 ref_cache[0 - 1 * 8] =
2053 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2054 ref_cache[2 - 1 * 8] =
2055 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2057 AV_ZERO128(mv_dst - 1 * 8);
2058 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2061 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2062 if (USES_LIST(left_type[LTOP], list)) {
2063 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2064 const int b8_xy = 4 * left_xy[LTOP] + 1;
2065 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2066 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2067 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2068 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2069 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2071 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2072 ref_cache[-1 + 16] =
2073 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2075 AV_ZERO32(mv_dst - 1 + 0);
2076 AV_ZERO32(mv_dst - 1 + 8);
2077 AV_ZERO32(mv_dst - 1 + 16);
2078 AV_ZERO32(mv_dst - 1 + 24);
2081 ref_cache[-1 + 16] =
2082 ref_cache[-1 + 24] = LIST_NOT_USED;
2087 if (!USES_LIST(mb_type, list)) {
2088 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2089 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2090 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2091 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2092 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2097 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2098 int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2099 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2100 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2101 AV_WN32A(&ref_cache[0 * 8], ref01);
2102 AV_WN32A(&ref_cache[1 * 8], ref01);
2103 AV_WN32A(&ref_cache[2 * 8], ref23);
2104 AV_WN32A(&ref_cache[3 * 8], ref23);
2108 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
2109 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2110 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2111 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2112 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2118 * @return non zero if the loop filter can be skipped
2120 static int fill_filter_caches(H264Context *h, int mb_type)
2122 const int mb_xy = h->mb_xy;
2123 int top_xy, left_xy[LEFT_MBS];
2124 int top_type, left_type[LEFT_MBS];
2128 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
2130 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2131 * stuff, I can't imagine that these complex rules are worth it. */
2133 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2134 if (FRAME_MBAFF(h)) {
2135 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2136 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2138 if (left_mb_field_flag != curr_mb_field_flag)
2139 left_xy[LTOP] -= h->mb_stride;
2141 if (curr_mb_field_flag)
2142 top_xy += h->mb_stride &
2143 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2144 if (left_mb_field_flag != curr_mb_field_flag)
2145 left_xy[LBOT] += h->mb_stride;
2149 h->top_mb_xy = top_xy;
2150 h->left_mb_xy[LTOP] = left_xy[LTOP];
2151 h->left_mb_xy[LBOT] = left_xy[LBOT];
2153 /* For sufficiently low qp, filtering wouldn't do anything.
2154 * This is a conservative estimate: could also check beta_offset
2155 * and more accurate chroma_qp. */
2156 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2157 int qp = h->cur_pic.qscale_table[mb_xy];
2158 if (qp <= qp_thresh &&
2159 (left_xy[LTOP] < 0 ||
2160 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2162 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2163 if (!FRAME_MBAFF(h))
2165 if ((left_xy[LTOP] < 0 ||
2166 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2167 (top_xy < h->mb_stride ||
2168 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2173 top_type = h->cur_pic.mb_type[top_xy];
2174 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2175 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2176 if (h->deblocking_filter == 2) {
2177 if (h->slice_table[top_xy] != h->slice_num)
2179 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
2180 left_type[LTOP] = left_type[LBOT] = 0;
2182 if (h->slice_table[top_xy] == 0xFFFF)
2184 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2185 left_type[LTOP] = left_type[LBOT] = 0;
2187 h->top_type = top_type;
2188 h->left_type[LTOP] = left_type[LTOP];
2189 h->left_type[LBOT] = left_type[LBOT];
2191 if (IS_INTRA(mb_type))
2194 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2195 top_type, left_type, mb_xy, 0);
2196 if (h->list_count == 2)
2197 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2198 top_type, left_type, mb_xy, 1);
2200 nnz = h->non_zero_count[mb_xy];
2201 nnz_cache = h->non_zero_count_cache;
2202 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2203 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2204 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2205 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2206 h->cbp = h->cbp_table[mb_xy];
2209 nnz = h->non_zero_count[top_xy];
2210 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2213 if (left_type[LTOP]) {
2214 nnz = h->non_zero_count[left_xy[LTOP]];
2215 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2216 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2217 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2218 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2221 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2222 * from what the loop filter needs */
2223 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2224 if (IS_8x8DCT(top_type)) {
2225 nnz_cache[4 + 8 * 0] =
2226 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2227 nnz_cache[6 + 8 * 0] =
2228 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2230 if (IS_8x8DCT(left_type[LTOP])) {
2231 nnz_cache[3 + 8 * 1] =
2232 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2234 if (IS_8x8DCT(left_type[LBOT])) {
2235 nnz_cache[3 + 8 * 3] =
2236 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2239 if (IS_8x8DCT(mb_type)) {
2240 nnz_cache[scan8[0]] =
2241 nnz_cache[scan8[1]] =
2242 nnz_cache[scan8[2]] =
2243 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2245 nnz_cache[scan8[0 + 4]] =
2246 nnz_cache[scan8[1 + 4]] =
2247 nnz_cache[scan8[2 + 4]] =
2248 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2250 nnz_cache[scan8[0 + 8]] =
2251 nnz_cache[scan8[1 + 8]] =
2252 nnz_cache[scan8[2 + 8]] =
2253 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2255 nnz_cache[scan8[0 + 12]] =
2256 nnz_cache[scan8[1 + 12]] =
2257 nnz_cache[scan8[2 + 12]] =
2258 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2265 static void loop_filter(H264Context *h, int start_x, int end_x)
2267 uint8_t *dest_y, *dest_cb, *dest_cr;
2268 int linesize, uvlinesize, mb_x, mb_y;
2269 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2270 const int old_slice_type = h->slice_type;
2271 const int pixel_shift = h->pixel_shift;
2272 const int block_h = 16 >> h->chroma_y_shift;
2274 if (h->deblocking_filter) {
2275 for (mb_x = start_x; mb_x < end_x; mb_x++)
2276 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2278 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2279 h->slice_num = h->slice_table[mb_xy];
2280 mb_type = h->cur_pic.mb_type[mb_xy];
2281 h->list_count = h->list_counts[mb_xy];
2285 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2289 dest_y = h->cur_pic.f.data[0] +
2290 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2291 dest_cb = h->cur_pic.f.data[1] +
2292 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2293 mb_y * h->uvlinesize * block_h;
2294 dest_cr = h->cur_pic.f.data[2] +
2295 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2296 mb_y * h->uvlinesize * block_h;
2297 // FIXME simplify above
2300 linesize = h->mb_linesize = h->linesize * 2;
2301 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2302 if (mb_y & 1) { // FIXME move out of this function?
2303 dest_y -= h->linesize * 15;
2304 dest_cb -= h->uvlinesize * (block_h - 1);
2305 dest_cr -= h->uvlinesize * (block_h - 1);
2308 linesize = h->mb_linesize = h->linesize;
2309 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2311 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2313 if (fill_filter_caches(h, mb_type))
2315 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2316 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2318 if (FRAME_MBAFF(h)) {
2319 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2320 linesize, uvlinesize);
2322 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2323 dest_cr, linesize, uvlinesize);
2327 h->slice_type = old_slice_type;
2329 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2330 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2331 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2334 static void predict_field_decoding_flag(H264Context *h)
2336 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2337 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2338 h->cur_pic.mb_type[mb_xy - 1] :
2339 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2340 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2341 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2345 * Draw edges and report progress for the last MB row.
2347 static void decode_finish_row(H264Context *h)
2349 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2350 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2351 int height = 16 << FRAME_MBAFF(h);
2352 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2354 if (h->deblocking_filter) {
2355 if ((top + height) >= pic_height)
2356 height += deblock_border;
2357 top -= deblock_border;
2360 if (top >= pic_height || (top + height) < 0)
2363 height = FFMIN(height, pic_height - top);
2365 height = top + height;
2369 ff_h264_draw_horiz_band(h, top, height);
2371 if (h->droppable || h->er.error_occurred)
2374 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2375 h->picture_structure == PICT_BOTTOM_FIELD);
2378 static void er_add_slice(H264Context *h, int startx, int starty,
2379 int endx, int endy, int status)
2381 if (CONFIG_ERROR_RESILIENCE) {
2382 ERContext *er = &h->er;
2384 ff_er_add_slice(er, startx, starty, endx, endy, status);
2388 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2390 H264Context *h = *(void **)arg;
2391 int lf_x_start = h->mb_x;
2393 h->mb_skip_run = -1;
2395 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2397 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2398 avctx->codec_id != AV_CODEC_ID_H264 ||
2399 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2401 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
2402 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2404 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
2405 prev_status &= ~ VP_START;
2406 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2407 h->er.error_occurred = 1;
2413 align_get_bits(&h->gb);
2416 ff_init_cabac_decoder(&h->cabac,
2417 h->gb.buffer + get_bits_count(&h->gb) / 8,
2418 (get_bits_left(&h->gb) + 7) / 8);
2420 ff_h264_init_cabac_states(h);
2424 int ret = ff_h264_decode_mb_cabac(h);
2426 // STOP_TIMER("decode_mb_cabac")
2429 ff_h264_hl_decode_mb(h);
2431 // FIXME optimal? or let mb_decode decode 16x32 ?
2432 if (ret >= 0 && FRAME_MBAFF(h)) {
2435 ret = ff_h264_decode_mb_cabac(h);
2438 ff_h264_hl_decode_mb(h);
2441 eos = get_cabac_terminate(&h->cabac);
2443 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2444 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2445 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2446 h->mb_y, ER_MB_END);
2447 if (h->mb_x >= lf_x_start)
2448 loop_filter(h, lf_x_start, h->mb_x + 1);
2451 if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
2452 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", h->cabac.bytestream_end - h->cabac.bytestream);
2453 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
2454 av_log(h->avctx, AV_LOG_ERROR,
2455 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2457 h->cabac.bytestream_end - h->cabac.bytestream);
2458 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2459 h->mb_y, ER_MB_ERROR);
2460 return AVERROR_INVALIDDATA;
2463 if (++h->mb_x >= h->mb_width) {
2464 loop_filter(h, lf_x_start, h->mb_x);
2465 h->mb_x = lf_x_start = 0;
2466 decode_finish_row(h);
2468 if (FIELD_OR_MBAFF_PICTURE(h)) {
2470 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2471 predict_field_decoding_flag(h);
2475 if (eos || h->mb_y >= h->mb_height) {
2476 tprintf(h->avctx, "slice end %d %d\n",
2477 get_bits_count(&h->gb), h->gb.size_in_bits);
2478 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2479 h->mb_y, ER_MB_END);
2480 if (h->mb_x > lf_x_start)
2481 loop_filter(h, lf_x_start, h->mb_x);
2487 int ret = ff_h264_decode_mb_cavlc(h);
2490 ff_h264_hl_decode_mb(h);
2492 // FIXME optimal? or let mb_decode decode 16x32 ?
2493 if (ret >= 0 && FRAME_MBAFF(h)) {
2495 ret = ff_h264_decode_mb_cavlc(h);
2498 ff_h264_hl_decode_mb(h);
2503 av_log(h->avctx, AV_LOG_ERROR,
2504 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2505 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2506 h->mb_y, ER_MB_ERROR);
2510 if (++h->mb_x >= h->mb_width) {
2511 loop_filter(h, lf_x_start, h->mb_x);
2512 h->mb_x = lf_x_start = 0;
2513 decode_finish_row(h);
2515 if (FIELD_OR_MBAFF_PICTURE(h)) {
2517 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2518 predict_field_decoding_flag(h);
2520 if (h->mb_y >= h->mb_height) {
2521 tprintf(h->avctx, "slice end %d %d\n",
2522 get_bits_count(&h->gb), h->gb.size_in_bits);
2524 if ( get_bits_left(&h->gb) == 0
2525 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2526 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2527 h->mb_x - 1, h->mb_y, ER_MB_END);
2531 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2532 h->mb_x, h->mb_y, ER_MB_END);
2534 return AVERROR_INVALIDDATA;
2539 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2540 tprintf(h->avctx, "slice end %d %d\n",
2541 get_bits_count(&h->gb), h->gb.size_in_bits);
2543 if (get_bits_left(&h->gb) == 0) {
2544 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2545 h->mb_x - 1, h->mb_y, ER_MB_END);
2546 if (h->mb_x > lf_x_start)
2547 loop_filter(h, lf_x_start, h->mb_x);
2551 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2552 h->mb_y, ER_MB_ERROR);
2554 return AVERROR_INVALIDDATA;
2562 * Call decode_slice() for each context.
2564 * @param h h264 master context
2565 * @param context_count number of contexts to execute
2567 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2569 AVCodecContext *const avctx = h->avctx;
2573 av_assert0(h->mb_y < h->mb_height);
2575 if (h->avctx->hwaccel ||
2576 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2578 if (context_count == 1) {
2579 return decode_slice(avctx, &h);
2581 av_assert0(context_count > 0);
2582 for (i = 1; i < context_count; i++) {
2583 hx = h->thread_context[i];
2584 if (CONFIG_ERROR_RESILIENCE) {
2585 hx->er.error_count = 0;
2587 hx->x264_build = h->x264_build;
2590 avctx->execute(avctx, decode_slice, h->thread_context,
2591 NULL, context_count, sizeof(void *));
2593 /* pull back stuff from slices to master context */
2594 hx = h->thread_context[context_count - 1];
2597 h->droppable = hx->droppable;
2598 h->picture_structure = hx->picture_structure;
2599 if (CONFIG_ERROR_RESILIENCE) {
2600 for (i = 1; i < context_count; i++)
2601 h->er.error_count += h->thread_context[i]->er.error_count;