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 Libav.
7 * Libav 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 * Libav 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 Libav; 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, 3,
53 static const uint8_t div6[QP_MAX_NUM + 1] = {
54 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
55 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
56 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
59 static const uint8_t field_scan[16] = {
60 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
61 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
62 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
63 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
66 static const uint8_t field_scan8x8[64] = {
67 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
68 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
69 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
70 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
71 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
72 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
73 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
74 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
75 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
76 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
77 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
78 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
79 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
80 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
81 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
82 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
85 static const uint8_t field_scan8x8_cavlc[64] = {
86 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
87 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
88 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
89 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
90 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
91 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
92 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
93 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
94 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
95 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
96 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
97 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
98 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
99 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
100 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
101 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
104 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
105 static const uint8_t zigzag_scan8x8_cavlc[64] = {
106 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
107 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
108 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
109 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
110 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
111 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
112 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
113 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
114 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
115 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
116 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
117 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
118 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
119 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
120 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
121 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
124 static const uint8_t dequant4_coeff_init[6][3] = {
133 static const uint8_t dequant8_coeff_init_scan[16] = {
134 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
137 static const uint8_t dequant8_coeff_init[6][6] = {
138 { 20, 18, 32, 19, 25, 24 },
139 { 22, 19, 35, 21, 28, 26 },
140 { 26, 23, 42, 24, 33, 31 },
141 { 28, 25, 45, 26, 35, 33 },
142 { 32, 28, 51, 30, 40, 38 },
143 { 36, 32, 58, 34, 46, 43 },
147 static void release_unused_pictures(H264Context *h, int remove_current)
151 /* release non reference frames */
152 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
153 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
154 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
155 ff_h264_unref_picture(h, &h->DPB[i]);
160 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
162 const H264Context *h = sl->h264;
163 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
165 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
166 // edge emu needs blocksize + filter length - 1
167 // (= 21x21 for h264)
168 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
170 av_fast_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
171 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
172 av_fast_malloc(&sl->top_borders[1], &sl->top_borders_allocated[1],
173 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
175 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
176 !sl->top_borders[0] || !sl->top_borders[1]) {
177 av_freep(&sl->bipred_scratchpad);
178 av_freep(&sl->edge_emu_buffer);
179 av_freep(&sl->top_borders[0]);
180 av_freep(&sl->top_borders[1]);
182 sl->bipred_scratchpad_allocated = 0;
183 sl->edge_emu_buffer_allocated = 0;
184 sl->top_borders_allocated[0] = 0;
185 sl->top_borders_allocated[1] = 0;
186 return AVERROR(ENOMEM);
192 static int init_table_pools(H264Context *h)
194 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
195 const int mb_array_size = h->mb_stride * h->mb_height;
196 const int b4_stride = h->mb_width * 4 + 1;
197 const int b4_array_size = b4_stride * h->mb_height * 4;
199 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
201 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
202 sizeof(uint32_t), av_buffer_allocz);
203 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
204 sizeof(int16_t), av_buffer_allocz);
205 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
207 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
208 !h->ref_index_pool) {
209 av_buffer_pool_uninit(&h->qscale_table_pool);
210 av_buffer_pool_uninit(&h->mb_type_pool);
211 av_buffer_pool_uninit(&h->motion_val_pool);
212 av_buffer_pool_uninit(&h->ref_index_pool);
213 return AVERROR(ENOMEM);
219 static int alloc_picture(H264Context *h, H264Picture *pic)
223 av_assert0(!pic->f.data[0]);
226 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
227 AV_GET_BUFFER_FLAG_REF : 0);
231 h->linesize = pic->f.linesize[0];
232 h->uvlinesize = pic->f.linesize[1];
234 if (h->avctx->hwaccel) {
235 const AVHWAccel *hwaccel = h->avctx->hwaccel;
236 av_assert0(!pic->hwaccel_picture_private);
237 if (hwaccel->frame_priv_data_size) {
238 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
239 if (!pic->hwaccel_priv_buf)
240 return AVERROR(ENOMEM);
241 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
245 if (!h->qscale_table_pool) {
246 ret = init_table_pools(h);
251 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
252 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
253 if (!pic->qscale_table_buf || !pic->mb_type_buf)
256 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
257 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
259 for (i = 0; i < 2; i++) {
260 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
261 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
262 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
265 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
266 pic->ref_index[i] = pic->ref_index_buf[i]->data;
271 ff_h264_unref_picture(h, pic);
272 return (ret < 0) ? ret : AVERROR(ENOMEM);
275 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
279 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
284 static int find_unused_picture(H264Context *h)
288 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
289 if (pic_is_unused(h, &h->DPB[i]))
292 if (i == H264_MAX_PICTURE_COUNT)
293 return AVERROR_INVALIDDATA;
295 if (h->DPB[i].needs_realloc) {
296 h->DPB[i].needs_realloc = 0;
297 ff_h264_unref_picture(h, &h->DPB[i]);
304 static void init_dequant8_coeff_table(H264Context *h)
307 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
309 for (i = 0; i < 6; i++) {
310 h->dequant8_coeff[i] = h->dequant8_buffer[i];
311 for (j = 0; j < i; j++)
312 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
313 64 * sizeof(uint8_t))) {
314 h->dequant8_coeff[i] = h->dequant8_buffer[j];
320 for (q = 0; q < max_qp + 1; q++) {
323 for (x = 0; x < 64; x++)
324 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
325 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
326 h->pps.scaling_matrix8[i][x]) << shift;
331 static void init_dequant4_coeff_table(H264Context *h)
334 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
335 for (i = 0; i < 6; i++) {
336 h->dequant4_coeff[i] = h->dequant4_buffer[i];
337 for (j = 0; j < i; j++)
338 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
339 16 * sizeof(uint8_t))) {
340 h->dequant4_coeff[i] = h->dequant4_buffer[j];
346 for (q = 0; q < max_qp + 1; q++) {
347 int shift = div6[q] + 2;
349 for (x = 0; x < 16; x++)
350 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
351 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
352 h->pps.scaling_matrix4[i][x]) << shift;
357 void h264_init_dequant_tables(H264Context *h)
360 init_dequant4_coeff_table(h);
361 if (h->pps.transform_8x8_mode)
362 init_dequant8_coeff_table(h);
363 if (h->sps.transform_bypass) {
364 for (i = 0; i < 6; i++)
365 for (x = 0; x < 16; x++)
366 h->dequant4_coeff[i][0][x] = 1 << 6;
367 if (h->pps.transform_8x8_mode)
368 for (i = 0; i < 6; i++)
369 for (x = 0; x < 64; x++)
370 h->dequant8_coeff[i][0][x] = 1 << 6;
375 * Mimic alloc_tables(), but for every context thread.
377 static void clone_tables(H264Context *dst, H264SliceContext *sl,
378 H264Context *src, int i)
380 sl->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
381 sl->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
382 sl->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
384 dst->non_zero_count = src->non_zero_count;
385 dst->slice_table = src->slice_table;
386 dst->cbp_table = src->cbp_table;
387 dst->mb2b_xy = src->mb2b_xy;
388 dst->mb2br_xy = src->mb2br_xy;
389 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
390 dst->direct_table = src->direct_table;
391 dst->list_counts = src->list_counts;
393 dst->cur_pic_ptr = src->cur_pic_ptr;
394 dst->cur_pic = src->cur_pic;
395 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
396 src->sps.chroma_format_idc);
399 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
401 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
402 ((pic && pic >= old_ctx->DPB && \
403 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
404 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
406 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
407 H264Context *new_base,
408 H264Context *old_base)
412 for (i = 0; i < count; i++) {
413 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
414 IN_RANGE(from[i], old_base->DPB,
415 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
417 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
421 static int copy_parameter_set(void **to, void **from, int count, int size)
425 for (i = 0; i < count; i++) {
426 if (to[i] && !from[i]) {
428 } else if (from[i] && !to[i]) {
429 to[i] = av_malloc(size);
431 return AVERROR(ENOMEM);
435 memcpy(to[i], from[i], size);
441 #define copy_fields(to, from, start_field, end_field) \
442 memcpy(&to->start_field, &from->start_field, \
443 (char *)&to->end_field - (char *)&to->start_field)
445 static int h264_slice_header_init(H264Context *h, int reinit);
447 int ff_h264_update_thread_context(AVCodecContext *dst,
448 const AVCodecContext *src)
450 H264Context *h = dst->priv_data, *h1 = src->priv_data;
451 int inited = h->context_initialized, err = 0;
452 int context_reinitialized = 0;
455 if (dst == src || !h1->context_initialized)
459 (h->width != h1->width ||
460 h->height != h1->height ||
461 h->mb_width != h1->mb_width ||
462 h->mb_height != h1->mb_height ||
463 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
464 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
465 h->sps.colorspace != h1->sps.colorspace)) {
467 /* set bits_per_raw_sample to the previous value. the check for changed
468 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
469 * the current value */
470 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
472 h->width = h1->width;
473 h->height = h1->height;
474 h->mb_height = h1->mb_height;
475 h->mb_width = h1->mb_width;
476 h->mb_num = h1->mb_num;
477 h->mb_stride = h1->mb_stride;
478 h->b_stride = h1->b_stride;
480 if ((err = h264_slice_header_init(h, 1)) < 0) {
481 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
484 context_reinitialized = 1;
486 /* update linesize on resize. The decoder doesn't
487 * necessarily call h264_frame_start in the new thread */
488 h->linesize = h1->linesize;
489 h->uvlinesize = h1->uvlinesize;
491 /* copy block_offset since frame_start may not be called */
492 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
496 H264SliceContext *orig_slice_ctx = h->slice_ctx;
498 for (i = 0; i < MAX_SPS_COUNT; i++)
499 av_freep(h->sps_buffers + i);
501 for (i = 0; i < MAX_PPS_COUNT; i++)
502 av_freep(h->pps_buffers + i);
504 memcpy(h, h1, sizeof(*h1));
505 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
506 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
507 h->context_initialized = 0;
509 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
510 av_frame_unref(&h->cur_pic.f);
511 h->cur_pic.tf.f = &h->cur_pic.f;
513 h->slice_ctx = orig_slice_ctx;
515 memset(&h->slice_ctx[0].er, 0, sizeof(h->slice_ctx[0].er));
516 memset(&h->slice_ctx[0].mb, 0, sizeof(h->slice_ctx[0].mb));
517 memset(&h->slice_ctx[0].mb_luma_dc, 0, sizeof(h->slice_ctx[0].mb_luma_dc));
518 memset(&h->slice_ctx[0].mb_padding, 0, sizeof(h->slice_ctx[0].mb_padding));
522 h->qscale_table_pool = NULL;
523 h->mb_type_pool = NULL;
524 h->ref_index_pool = NULL;
525 h->motion_val_pool = NULL;
527 ret = ff_h264_alloc_tables(h);
529 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
532 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
534 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
538 h->rbsp_buffer = NULL;
539 h->rbsp_buffer_size = 0;
541 h->thread_context[0] = h;
543 h->context_initialized = 1;
546 h->avctx->coded_height = h1->avctx->coded_height;
547 h->avctx->coded_width = h1->avctx->coded_width;
548 h->avctx->width = h1->avctx->width;
549 h->avctx->height = h1->avctx->height;
550 h->coded_picture_number = h1->coded_picture_number;
551 h->first_field = h1->first_field;
552 h->picture_structure = h1->picture_structure;
553 h->droppable = h1->droppable;
554 h->low_delay = h1->low_delay;
556 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
557 ff_h264_unref_picture(h, &h->DPB[i]);
558 if (h1->DPB[i].f.buf[0] &&
559 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
563 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
564 ff_h264_unref_picture(h, &h->cur_pic);
565 if (h1->cur_pic.f.buf[0]) {
566 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
571 h->workaround_bugs = h1->workaround_bugs;
572 h->low_delay = h1->low_delay;
573 h->droppable = h1->droppable;
575 // extradata/NAL handling
576 h->is_avc = h1->is_avc;
579 if ((ret = copy_parameter_set((void **)h->sps_buffers,
580 (void **)h1->sps_buffers,
581 MAX_SPS_COUNT, sizeof(SPS))) < 0)
584 if ((ret = copy_parameter_set((void **)h->pps_buffers,
585 (void **)h1->pps_buffers,
586 MAX_PPS_COUNT, sizeof(PPS))) < 0)
590 // Dequantization matrices
591 // FIXME these are big - can they be only copied when PPS changes?
592 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
594 for (i = 0; i < 6; i++)
595 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
596 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
598 for (i = 0; i < 6; i++)
599 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
600 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
602 h->dequant_coeff_pps = h1->dequant_coeff_pps;
605 copy_fields(h, h1, poc_lsb, default_ref_list);
608 copy_fields(h, h1, short_ref, thread_context);
610 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
611 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
612 copy_picture_range(h->delayed_pic, h1->delayed_pic,
613 MAX_DELAYED_PIC_COUNT + 2, h, h1);
615 h->last_slice_type = h1->last_slice_type;
617 if (context_reinitialized)
618 ff_h264_set_parameter_from_sps(h);
624 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
625 h->prev_poc_msb = h->poc_msb;
626 h->prev_poc_lsb = h->poc_lsb;
628 h->prev_frame_num_offset = h->frame_num_offset;
629 h->prev_frame_num = h->frame_num;
630 h->outputed_poc = h->next_outputed_poc;
632 h->recovery_frame = h1->recovery_frame;
633 h->frame_recovered = h1->frame_recovered;
638 static int h264_frame_start(H264Context *h)
642 const int pixel_shift = h->pixel_shift;
644 release_unused_pictures(h, 1);
645 h->cur_pic_ptr = NULL;
647 i = find_unused_picture(h);
649 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
654 pic->reference = h->droppable ? 0 : h->picture_structure;
655 pic->f.coded_picture_number = h->coded_picture_number++;
656 pic->field_picture = h->picture_structure != PICT_FRAME;
658 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
660 * See decode_nal_units().
662 pic->f.key_frame = 0;
666 if ((ret = alloc_picture(h, pic)) < 0)
669 h->cur_pic_ptr = pic;
670 ff_h264_unref_picture(h, &h->cur_pic);
671 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
674 if (CONFIG_ERROR_RESILIENCE)
675 ff_er_frame_start(&h->slice_ctx[0].er);
677 assert(h->linesize && h->uvlinesize);
679 for (i = 0; i < 16; i++) {
680 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
681 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
683 for (i = 0; i < 16; i++) {
684 h->block_offset[16 + i] =
685 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
686 h->block_offset[48 + 16 + i] =
687 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
690 /* Some macroblocks can be accessed before they're available in case
691 * of lost slices, MBAFF or threading. */
692 memset(h->slice_table, -1,
693 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
695 /* We mark the current picture as non-reference after allocating it, so
696 * that if we break out due to an error it can be released automatically
697 * in the next ff_mpv_frame_start().
699 h->cur_pic_ptr->reference = 0;
701 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
703 h->next_output_pic = NULL;
705 assert(h->cur_pic_ptr->long_ref == 0);
710 static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *sl,
712 uint8_t *src_cb, uint8_t *src_cr,
713 int linesize, int uvlinesize,
718 const int pixel_shift = h->pixel_shift;
719 int chroma444 = CHROMA444(h);
720 int chroma422 = CHROMA422(h);
723 src_cb -= uvlinesize;
724 src_cr -= uvlinesize;
726 if (!simple && FRAME_MBAFF(h)) {
729 top_border = sl->top_borders[0][sl->mb_x];
730 AV_COPY128(top_border, src_y + 15 * linesize);
732 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
733 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
736 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
737 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
738 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
739 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
741 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
742 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
744 } else if (chroma422) {
746 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
747 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
749 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
750 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
754 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
755 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
757 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
758 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
763 } else if (MB_MBAFF(sl)) {
769 top_border = sl->top_borders[top_idx][sl->mb_x];
770 /* There are two lines saved, the line above the top macroblock
771 * of a pair, and the line above the bottom macroblock. */
772 AV_COPY128(top_border, src_y + 16 * linesize);
774 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
776 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
779 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
780 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
781 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
782 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
784 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
785 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
787 } else if (chroma422) {
789 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
790 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
792 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
793 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
797 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
798 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
800 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
801 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
808 * Initialize implicit_weight table.
809 * @param field 0/1 initialize the weight for interlaced MBAFF
810 * -1 initializes the rest
812 static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int field)
814 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
816 for (i = 0; i < 2; i++) {
817 sl->luma_weight_flag[i] = 0;
818 sl->chroma_weight_flag[i] = 0;
822 if (h->picture_structure == PICT_FRAME) {
823 cur_poc = h->cur_pic_ptr->poc;
825 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
827 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
828 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
830 sl->use_weight_chroma = 0;
834 ref_count0 = sl->ref_count[0];
835 ref_count1 = sl->ref_count[1];
837 cur_poc = h->cur_pic_ptr->field_poc[field];
839 ref_count0 = 16 + 2 * sl->ref_count[0];
840 ref_count1 = 16 + 2 * sl->ref_count[1];
844 sl->use_weight_chroma = 2;
845 sl->luma_log2_weight_denom = 5;
846 sl->chroma_log2_weight_denom = 5;
848 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
849 int poc0 = sl->ref_list[0][ref0].poc;
850 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
852 if (!sl->ref_list[0][ref0].long_ref && !sl->ref_list[1][ref1].long_ref) {
853 int poc1 = sl->ref_list[1][ref1].poc;
854 int td = av_clip_int8(poc1 - poc0);
856 int tb = av_clip_int8(cur_poc - poc0);
857 int tx = (16384 + (FFABS(td) >> 1)) / td;
858 int dist_scale_factor = (tb * tx + 32) >> 8;
859 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
860 w = 64 - dist_scale_factor;
864 sl->implicit_weight[ref0][ref1][0] =
865 sl->implicit_weight[ref0][ref1][1] = w;
867 sl->implicit_weight[ref0][ref1][field] = w;
874 * initialize scan tables
876 static void init_scan_tables(H264Context *h)
879 for (i = 0; i < 16; i++) {
880 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
881 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
882 h->field_scan[i] = TRANSPOSE(field_scan[i]);
885 for (i = 0; i < 64; i++) {
886 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
887 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
888 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
889 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
890 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
893 if (h->sps.transform_bypass) { // FIXME same ugly
894 h->zigzag_scan_q0 = zigzag_scan;
895 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
896 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
897 h->field_scan_q0 = field_scan;
898 h->field_scan8x8_q0 = field_scan8x8;
899 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
901 h->zigzag_scan_q0 = h->zigzag_scan;
902 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
903 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
904 h->field_scan_q0 = h->field_scan;
905 h->field_scan8x8_q0 = h->field_scan8x8;
906 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
911 * Replicate H264 "master" context to thread contexts.
913 static int clone_slice(H264Context *dst, H264Context *src)
915 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
916 dst->cur_pic_ptr = src->cur_pic_ptr;
917 dst->cur_pic = src->cur_pic;
918 dst->linesize = src->linesize;
919 dst->uvlinesize = src->uvlinesize;
920 dst->first_field = src->first_field;
922 dst->prev_poc_msb = src->prev_poc_msb;
923 dst->prev_poc_lsb = src->prev_poc_lsb;
924 dst->prev_frame_num_offset = src->prev_frame_num_offset;
925 dst->prev_frame_num = src->prev_frame_num;
926 dst->short_ref_count = src->short_ref_count;
928 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
929 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
930 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
932 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
933 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
938 static enum AVPixelFormat get_pixel_format(H264Context *h)
940 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
941 CONFIG_H264_VAAPI_HWACCEL + \
942 (CONFIG_H264_VDA_HWACCEL * 2) + \
943 CONFIG_H264_VDPAU_HWACCEL)
944 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
945 const enum AVPixelFormat *choices = pix_fmts;
947 switch (h->sps.bit_depth_luma) {
950 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
951 *fmt++ = AV_PIX_FMT_GBRP9;
953 *fmt++ = AV_PIX_FMT_YUV444P9;
954 } else if (CHROMA422(h))
955 *fmt++ = AV_PIX_FMT_YUV422P9;
957 *fmt++ = AV_PIX_FMT_YUV420P9;
961 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
962 *fmt++ = AV_PIX_FMT_GBRP10;
964 *fmt++ = AV_PIX_FMT_YUV444P10;
965 } else if (CHROMA422(h))
966 *fmt++ = AV_PIX_FMT_YUV422P10;
968 *fmt++ = AV_PIX_FMT_YUV420P10;
971 #if CONFIG_H264_VDPAU_HWACCEL
972 *fmt++ = AV_PIX_FMT_VDPAU;
975 if (h->avctx->colorspace == AVCOL_SPC_RGB)
976 *fmt++ = AV_PIX_FMT_GBRP;
977 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
978 *fmt++ = AV_PIX_FMT_YUVJ444P;
980 *fmt++ = AV_PIX_FMT_YUV444P;
981 } else if (CHROMA422(h)) {
982 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
983 *fmt++ = AV_PIX_FMT_YUVJ422P;
985 *fmt++ = AV_PIX_FMT_YUV422P;
987 #if CONFIG_H264_DXVA2_HWACCEL
988 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
990 #if CONFIG_H264_VAAPI_HWACCEL
991 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
993 #if CONFIG_H264_VDA_HWACCEL
994 *fmt++ = AV_PIX_FMT_VDA_VLD;
995 *fmt++ = AV_PIX_FMT_VDA;
997 if (h->avctx->codec->pix_fmts)
998 choices = h->avctx->codec->pix_fmts;
999 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1000 *fmt++ = AV_PIX_FMT_YUVJ420P;
1002 *fmt++ = AV_PIX_FMT_YUV420P;
1006 av_log(h->avctx, AV_LOG_ERROR,
1007 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1008 return AVERROR_INVALIDDATA;
1011 *fmt = AV_PIX_FMT_NONE;
1013 return ff_get_format(h->avctx, choices);
1016 /* export coded and cropped frame dimensions to AVCodecContext */
1017 static int init_dimensions(H264Context *h)
1019 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1020 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1021 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1022 h->sps.crop_right || h->sps.crop_bottom;
1024 /* handle container cropping */
1025 if (!crop_present &&
1026 FFALIGN(h->avctx->width, 16) == h->width &&
1027 FFALIGN(h->avctx->height, 16) == h->height) {
1028 width = h->avctx->width;
1029 height = h->avctx->height;
1032 if (width <= 0 || height <= 0) {
1033 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1035 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1036 return AVERROR_INVALIDDATA;
1038 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1039 h->sps.crop_bottom =
1049 h->avctx->coded_width = h->width;
1050 h->avctx->coded_height = h->height;
1051 h->avctx->width = width;
1052 h->avctx->height = height;
1057 static int h264_slice_header_init(H264Context *h, int reinit)
1059 int nb_slices = (HAVE_THREADS &&
1060 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1061 h->avctx->thread_count : 1;
1064 ff_set_sar(h->avctx, h->sps.sar);
1065 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1066 &h->chroma_x_shift, &h->chroma_y_shift);
1068 if (h->sps.timing_info_present_flag) {
1069 int64_t den = h->sps.time_scale;
1070 if (h->x264_build < 44U)
1072 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1073 h->sps.num_units_in_tick, den, 1 << 30);
1077 ff_h264_free_tables(h, 0);
1079 h->prev_interlaced_frame = 1;
1081 init_scan_tables(h);
1082 ret = ff_h264_alloc_tables(h);
1084 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1088 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1091 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1093 max_slices = H264_MAX_THREADS;
1094 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1095 " reducing to %d\n", nb_slices, max_slices);
1096 nb_slices = max_slices;
1098 h->slice_context_count = nb_slices;
1100 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1101 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
1103 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1107 for (i = 1; i < h->slice_context_count; i++) {
1109 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1111 return AVERROR(ENOMEM);
1112 c->avctx = h->avctx;
1114 c->h264dsp = h->h264dsp;
1115 c->h264qpel = h->h264qpel;
1116 c->h264chroma = h->h264chroma;
1119 c->pixel_shift = h->pixel_shift;
1120 c->width = h->width;
1121 c->height = h->height;
1122 c->linesize = h->linesize;
1123 c->uvlinesize = h->uvlinesize;
1124 c->chroma_x_shift = h->chroma_x_shift;
1125 c->chroma_y_shift = h->chroma_y_shift;
1126 c->droppable = h->droppable;
1127 c->low_delay = h->low_delay;
1128 c->mb_width = h->mb_width;
1129 c->mb_height = h->mb_height;
1130 c->mb_stride = h->mb_stride;
1131 c->mb_num = h->mb_num;
1132 c->flags = h->flags;
1133 c->workaround_bugs = h->workaround_bugs;
1134 c->pict_type = h->pict_type;
1136 h->slice_ctx[i].h264 = c;
1138 init_scan_tables(c);
1139 clone_tables(c, &h->slice_ctx[i], h, i);
1140 c->context_initialized = 1;
1143 for (i = 0; i < h->slice_context_count; i++)
1144 if ((ret = ff_h264_slice_context_init(h, &h->slice_ctx[i])) < 0) {
1145 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1150 h->context_initialized = 1;
1156 * Decode a slice header.
1157 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1159 * @param h h264context
1160 * @param h0 h264 master context (differs from 'h' when doing sliced based
1161 * parallel decoding)
1163 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1165 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Context *h0)
1167 unsigned int first_mb_in_slice;
1168 unsigned int pps_id;
1170 unsigned int slice_type, tmp, i, j;
1171 int default_ref_list_done = 0;
1172 int last_pic_structure, last_pic_droppable;
1173 int needs_reinit = 0;
1174 int field_pic_flag, bottom_field_flag;
1176 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1177 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1179 first_mb_in_slice = get_ue_golomb(&sl->gb);
1181 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1182 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1183 ff_h264_field_end(h, sl, 1);
1186 h0->current_slice = 0;
1187 if (!h0->first_field) {
1188 if (h->cur_pic_ptr && !h->droppable) {
1189 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1190 h->picture_structure == PICT_BOTTOM_FIELD);
1192 h->cur_pic_ptr = NULL;
1196 slice_type = get_ue_golomb_31(&sl->gb);
1197 if (slice_type > 9) {
1198 av_log(h->avctx, AV_LOG_ERROR,
1199 "slice type %d too large at %d\n",
1200 slice_type, first_mb_in_slice);
1201 return AVERROR_INVALIDDATA;
1203 if (slice_type > 4) {
1205 sl->slice_type_fixed = 1;
1207 sl->slice_type_fixed = 0;
1209 slice_type = golomb_to_pict_type[slice_type];
1210 if (slice_type == AV_PICTURE_TYPE_I ||
1211 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
1212 default_ref_list_done = 1;
1214 sl->slice_type = slice_type;
1215 sl->slice_type_nos = slice_type & 3;
1217 if (h->nal_unit_type == NAL_IDR_SLICE &&
1218 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1219 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1220 return AVERROR_INVALIDDATA;
1223 // to make a few old functions happy, it's wrong though
1224 h->pict_type = sl->slice_type;
1226 pps_id = get_ue_golomb(&sl->gb);
1227 if (pps_id >= MAX_PPS_COUNT) {
1228 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1229 return AVERROR_INVALIDDATA;
1231 if (!h0->pps_buffers[pps_id]) {
1232 av_log(h->avctx, AV_LOG_ERROR,
1233 "non-existing PPS %u referenced\n",
1235 return AVERROR_INVALIDDATA;
1237 h->pps = *h0->pps_buffers[pps_id];
1239 if (!h0->sps_buffers[h->pps.sps_id]) {
1240 av_log(h->avctx, AV_LOG_ERROR,
1241 "non-existing SPS %u referenced\n",
1243 return AVERROR_INVALIDDATA;
1246 if (h->pps.sps_id != h->sps.sps_id ||
1247 h0->sps_buffers[h->pps.sps_id]->new) {
1248 h0->sps_buffers[h->pps.sps_id]->new = 0;
1250 h->sps = *h0->sps_buffers[h->pps.sps_id];
1252 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1253 h->chroma_format_idc != h->sps.chroma_format_idc) {
1254 h->bit_depth_luma = h->sps.bit_depth_luma;
1255 h->chroma_format_idc = h->sps.chroma_format_idc;
1258 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1262 h->avctx->profile = ff_h264_get_profile(&h->sps);
1263 h->avctx->level = h->sps.level_idc;
1264 h->avctx->refs = h->sps.ref_frame_count;
1266 if (h->mb_width != h->sps.mb_width ||
1267 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1270 h->mb_width = h->sps.mb_width;
1271 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1272 h->mb_num = h->mb_width * h->mb_height;
1273 h->mb_stride = h->mb_width + 1;
1275 h->b_stride = h->mb_width * 4;
1277 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1279 h->width = 16 * h->mb_width;
1280 h->height = 16 * h->mb_height;
1282 ret = init_dimensions(h);
1286 if (h->sps.video_signal_type_present_flag) {
1287 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1289 if (h->sps.colour_description_present_flag) {
1290 if (h->avctx->colorspace != h->sps.colorspace)
1292 h->avctx->color_primaries = h->sps.color_primaries;
1293 h->avctx->color_trc = h->sps.color_trc;
1294 h->avctx->colorspace = h->sps.colorspace;
1298 if (h->context_initialized && needs_reinit) {
1300 av_log(h->avctx, AV_LOG_ERROR,
1301 "changing width %d -> %d / height %d -> %d on "
1303 h->width, h->avctx->coded_width,
1304 h->height, h->avctx->coded_height,
1305 h0->current_slice + 1);
1306 return AVERROR_INVALIDDATA;
1309 ff_h264_flush_change(h);
1311 if ((ret = get_pixel_format(h)) < 0)
1313 h->avctx->pix_fmt = ret;
1315 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1316 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1318 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1319 av_log(h->avctx, AV_LOG_ERROR,
1320 "h264_slice_header_init() failed\n");
1324 if (!h->context_initialized) {
1326 av_log(h->avctx, AV_LOG_ERROR,
1327 "Cannot (re-)initialize context during parallel decoding.\n");
1328 return AVERROR_PATCHWELCOME;
1331 if ((ret = get_pixel_format(h)) < 0)
1333 h->avctx->pix_fmt = ret;
1335 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1336 av_log(h->avctx, AV_LOG_ERROR,
1337 "h264_slice_header_init() failed\n");
1342 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1343 h->dequant_coeff_pps = pps_id;
1344 h264_init_dequant_tables(h);
1347 h->frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1350 h->mb_aff_frame = 0;
1351 last_pic_structure = h0->picture_structure;
1352 last_pic_droppable = h0->droppable;
1353 h->droppable = h->nal_ref_idc == 0;
1354 if (h->sps.frame_mbs_only_flag) {
1355 h->picture_structure = PICT_FRAME;
1357 field_pic_flag = get_bits1(&sl->gb);
1358 if (field_pic_flag) {
1359 bottom_field_flag = get_bits1(&sl->gb);
1360 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1362 h->picture_structure = PICT_FRAME;
1363 h->mb_aff_frame = h->sps.mb_aff;
1366 sl->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1368 if (h0->current_slice != 0) {
1369 if (last_pic_structure != h->picture_structure ||
1370 last_pic_droppable != h->droppable) {
1371 av_log(h->avctx, AV_LOG_ERROR,
1372 "Changing field mode (%d -> %d) between slices is not allowed\n",
1373 last_pic_structure, h->picture_structure);
1374 h->picture_structure = last_pic_structure;
1375 h->droppable = last_pic_droppable;
1376 return AVERROR_INVALIDDATA;
1377 } else if (!h0->cur_pic_ptr) {
1378 av_log(h->avctx, AV_LOG_ERROR,
1379 "unset cur_pic_ptr on slice %d\n",
1380 h0->current_slice + 1);
1381 return AVERROR_INVALIDDATA;
1384 /* Shorten frame num gaps so we don't have to allocate reference
1385 * frames just to throw them away */
1386 if (h->frame_num != h->prev_frame_num) {
1387 int unwrap_prev_frame_num = h->prev_frame_num;
1388 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1390 if (unwrap_prev_frame_num > h->frame_num)
1391 unwrap_prev_frame_num -= max_frame_num;
1393 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1394 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1395 if (unwrap_prev_frame_num < 0)
1396 unwrap_prev_frame_num += max_frame_num;
1398 h->prev_frame_num = unwrap_prev_frame_num;
1402 /* See if we have a decoded first field looking for a pair...
1403 * Here, we're using that to see if we should mark previously
1404 * decode frames as "finished".
1405 * We have to do that before the "dummy" in-between frame allocation,
1406 * since that can modify s->current_picture_ptr. */
1407 if (h0->first_field) {
1408 assert(h0->cur_pic_ptr);
1409 assert(h0->cur_pic_ptr->f.buf[0]);
1410 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1412 /* figure out if we have a complementary field pair */
1413 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1414 /* Previous field is unmatched. Don't display it, but let it
1415 * remain for reference if marked as such. */
1416 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1417 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1418 last_pic_structure == PICT_TOP_FIELD);
1421 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1422 /* This and previous field were reference, but had
1423 * different frame_nums. Consider this field first in
1424 * pair. Throw away previous field except for reference
1426 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1427 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1428 last_pic_structure == PICT_TOP_FIELD);
1431 /* Second field in complementary pair */
1432 if (!((last_pic_structure == PICT_TOP_FIELD &&
1433 h->picture_structure == PICT_BOTTOM_FIELD) ||
1434 (last_pic_structure == PICT_BOTTOM_FIELD &&
1435 h->picture_structure == PICT_TOP_FIELD))) {
1436 av_log(h->avctx, AV_LOG_ERROR,
1437 "Invalid field mode combination %d/%d\n",
1438 last_pic_structure, h->picture_structure);
1439 h->picture_structure = last_pic_structure;
1440 h->droppable = last_pic_droppable;
1441 return AVERROR_INVALIDDATA;
1442 } else if (last_pic_droppable != h->droppable) {
1443 avpriv_request_sample(h->avctx,
1444 "Found reference and non-reference fields in the same frame, which");
1445 h->picture_structure = last_pic_structure;
1446 h->droppable = last_pic_droppable;
1447 return AVERROR_PATCHWELCOME;
1453 while (h->frame_num != h->prev_frame_num &&
1454 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1455 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1456 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1457 h->frame_num, h->prev_frame_num);
1458 ret = h264_frame_start(h);
1460 h0->first_field = 0;
1464 h->prev_frame_num++;
1465 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1466 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1467 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1468 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1469 ret = ff_generate_sliding_window_mmcos(h, 1);
1470 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1472 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1473 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1475 /* Error concealment: If a ref is missing, copy the previous ref
1477 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1478 * many assumptions about there being no actual duplicates.
1479 * FIXME: This does not copy padding for out-of-frame motion
1480 * vectors. Given we are concealing a lost frame, this probably
1481 * is not noticeable by comparison, but it should be fixed. */
1482 if (h->short_ref_count) {
1484 av_image_copy(h->short_ref[0]->f.data,
1485 h->short_ref[0]->f.linesize,
1486 (const uint8_t **)prev->f.data,
1491 h->short_ref[0]->poc = prev->poc + 2;
1493 h->short_ref[0]->frame_num = h->prev_frame_num;
1497 /* See if we have a decoded first field looking for a pair...
1498 * We're using that to see whether to continue decoding in that
1499 * frame, or to allocate a new one. */
1500 if (h0->first_field) {
1501 assert(h0->cur_pic_ptr);
1502 assert(h0->cur_pic_ptr->f.buf[0]);
1503 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1505 /* figure out if we have a complementary field pair */
1506 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1507 /* Previous field is unmatched. Don't display it, but let it
1508 * remain for reference if marked as such. */
1509 h0->cur_pic_ptr = NULL;
1510 h0->first_field = FIELD_PICTURE(h);
1512 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1513 /* This and the previous field had different frame_nums.
1514 * Consider this field first in pair. Throw away previous
1515 * one except for reference purposes. */
1516 h0->first_field = 1;
1517 h0->cur_pic_ptr = NULL;
1519 /* Second field in complementary pair */
1520 h0->first_field = 0;
1524 /* Frame or first field in a potentially complementary pair */
1525 h0->first_field = FIELD_PICTURE(h);
1528 if (!FIELD_PICTURE(h) || h0->first_field) {
1529 if (h264_frame_start(h) < 0) {
1530 h0->first_field = 0;
1531 return AVERROR_INVALIDDATA;
1534 release_unused_pictures(h, 0);
1537 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1540 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1542 assert(h->mb_num == h->mb_width * h->mb_height);
1543 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1544 first_mb_in_slice >= h->mb_num) {
1545 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1546 return AVERROR_INVALIDDATA;
1548 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1549 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1550 FIELD_OR_MBAFF_PICTURE(h);
1551 if (h->picture_structure == PICT_BOTTOM_FIELD)
1552 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1553 assert(sl->mb_y < h->mb_height);
1555 if (h->picture_structure == PICT_FRAME) {
1556 h->curr_pic_num = h->frame_num;
1557 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1559 h->curr_pic_num = 2 * h->frame_num + 1;
1560 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1563 if (h->nal_unit_type == NAL_IDR_SLICE)
1564 get_ue_golomb(&sl->gb); /* idr_pic_id */
1566 if (h->sps.poc_type == 0) {
1567 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1569 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1570 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1573 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1574 h->delta_poc[0] = get_se_golomb(&sl->gb);
1576 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1577 h->delta_poc[1] = get_se_golomb(&sl->gb);
1580 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1582 if (h->pps.redundant_pic_cnt_present)
1583 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1585 ret = ff_set_ref_count(h, sl);
1589 default_ref_list_done = 0;
1591 if (!default_ref_list_done)
1592 ff_h264_fill_default_ref_list(h, sl);
1594 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1595 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1597 sl->ref_count[1] = sl->ref_count[0] = 0;
1602 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1603 (h->pps.weighted_bipred_idc == 1 &&
1604 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1605 ff_pred_weight_table(h, sl);
1606 else if (h->pps.weighted_bipred_idc == 2 &&
1607 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1608 implicit_weight_table(h, sl, -1);
1611 for (i = 0; i < 2; i++) {
1612 sl->luma_weight_flag[i] = 0;
1613 sl->chroma_weight_flag[i] = 0;
1617 // If frame-mt is enabled, only update mmco tables for the first slice
1618 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1619 // or h->mmco, which will cause ref list mix-ups and decoding errors
1620 // further down the line. This may break decoding if the first slice is
1621 // corrupt, thus we only do this if frame-mt is enabled.
1622 if (h->nal_ref_idc) {
1623 ret = ff_h264_decode_ref_pic_marking(h0, &sl->gb,
1624 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1625 h0->current_slice == 0);
1626 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1627 return AVERROR_INVALIDDATA;
1630 if (FRAME_MBAFF(h)) {
1631 ff_h264_fill_mbaff_ref_list(h, sl);
1633 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1634 implicit_weight_table(h, sl, 0);
1635 implicit_weight_table(h, sl, 1);
1639 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1640 ff_h264_direct_dist_scale_factor(h, sl);
1641 ff_h264_direct_ref_list_init(h, sl);
1643 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1644 tmp = get_ue_golomb_31(&sl->gb);
1646 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1647 return AVERROR_INVALIDDATA;
1649 sl->cabac_init_idc = tmp;
1652 sl->last_qscale_diff = 0;
1653 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1654 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1655 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1656 return AVERROR_INVALIDDATA;
1659 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1660 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1661 // FIXME qscale / qp ... stuff
1662 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1663 get_bits1(&sl->gb); /* sp_for_switch_flag */
1664 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1665 sl->slice_type == AV_PICTURE_TYPE_SI)
1666 get_se_golomb(&sl->gb); /* slice_qs_delta */
1668 sl->deblocking_filter = 1;
1669 sl->slice_alpha_c0_offset = 0;
1670 sl->slice_beta_offset = 0;
1671 if (h->pps.deblocking_filter_parameters_present) {
1672 tmp = get_ue_golomb_31(&sl->gb);
1674 av_log(h->avctx, AV_LOG_ERROR,
1675 "deblocking_filter_idc %u out of range\n", tmp);
1676 return AVERROR_INVALIDDATA;
1678 sl->deblocking_filter = tmp;
1679 if (sl->deblocking_filter < 2)
1680 sl->deblocking_filter ^= 1; // 1<->0
1682 if (sl->deblocking_filter) {
1683 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1684 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1685 if (sl->slice_alpha_c0_offset > 12 ||
1686 sl->slice_alpha_c0_offset < -12 ||
1687 sl->slice_beta_offset > 12 ||
1688 sl->slice_beta_offset < -12) {
1689 av_log(h->avctx, AV_LOG_ERROR,
1690 "deblocking filter parameters %d %d out of range\n",
1691 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1692 return AVERROR_INVALIDDATA;
1697 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1698 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1699 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1700 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1701 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1702 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1703 h->nal_ref_idc == 0))
1704 sl->deblocking_filter = 0;
1706 if (sl->deblocking_filter == 1 && h0->max_contexts > 1) {
1707 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1708 /* Cheat slightly for speed:
1709 * Do not bother to deblock across slices. */
1710 sl->deblocking_filter = 2;
1712 h0->max_contexts = 1;
1713 if (!h0->single_decode_warning) {
1714 av_log(h->avctx, AV_LOG_INFO,
1715 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1716 h0->single_decode_warning = 1;
1719 av_log(h->avctx, AV_LOG_ERROR,
1720 "Deblocking switched inside frame.\n");
1725 sl->qp_thresh = 15 -
1726 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1728 h->pps.chroma_qp_index_offset[0],
1729 h->pps.chroma_qp_index_offset[1]) +
1730 6 * (h->sps.bit_depth_luma - 8);
1732 h0->last_slice_type = slice_type;
1733 sl->slice_num = ++h0->current_slice;
1734 if (sl->slice_num >= MAX_SLICES) {
1735 av_log(h->avctx, AV_LOG_ERROR,
1736 "Too many slices, increase MAX_SLICES and recompile\n");
1739 for (j = 0; j < 2; j++) {
1741 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1742 for (i = 0; i < 16; i++) {
1744 if (j < sl->list_count && i < sl->ref_count[j] &&
1745 sl->ref_list[j][i].f.buf[0]) {
1747 AVBuffer *buf = sl->ref_list[j][i].f.buf[0]->buffer;
1748 for (k = 0; k < h->short_ref_count; k++)
1749 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1753 for (k = 0; k < h->long_ref_count; k++)
1754 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1755 id_list[i] = h->short_ref_count + k;
1763 for (i = 0; i < 16; i++)
1764 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1766 ref2frm[18 + 1] = -1;
1767 for (i = 16; i < 48; i++)
1768 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1769 (sl->ref_list[j][i].reference & 3);
1772 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1773 av_log(h->avctx, AV_LOG_DEBUG,
1774 "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",
1776 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1778 av_get_picture_type_char(sl->slice_type),
1779 sl->slice_type_fixed ? " fix" : "",
1780 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1781 pps_id, h->frame_num,
1782 h->cur_pic_ptr->field_poc[0],
1783 h->cur_pic_ptr->field_poc[1],
1784 sl->ref_count[0], sl->ref_count[1],
1786 sl->deblocking_filter,
1787 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1789 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
1790 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1796 int ff_h264_get_slice_type(const H264SliceContext *sl)
1798 switch (sl->slice_type) {
1799 case AV_PICTURE_TYPE_P:
1801 case AV_PICTURE_TYPE_B:
1803 case AV_PICTURE_TYPE_I:
1805 case AV_PICTURE_TYPE_SP:
1807 case AV_PICTURE_TYPE_SI:
1810 return AVERROR_INVALIDDATA;
1814 static av_always_inline void fill_filter_caches_inter(H264Context *h,
1815 H264SliceContext *sl,
1816 int mb_type, int top_xy,
1817 int left_xy[LEFT_MBS],
1819 int left_type[LEFT_MBS],
1820 int mb_xy, int list)
1822 int b_stride = h->b_stride;
1823 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1824 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1825 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1826 if (USES_LIST(top_type, list)) {
1827 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1828 const int b8_xy = 4 * top_xy + 2;
1829 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1830 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1831 ref_cache[0 - 1 * 8] =
1832 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1833 ref_cache[2 - 1 * 8] =
1834 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1836 AV_ZERO128(mv_dst - 1 * 8);
1837 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1840 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1841 if (USES_LIST(left_type[LTOP], list)) {
1842 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1843 const int b8_xy = 4 * left_xy[LTOP] + 1;
1844 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1845 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1846 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1847 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1848 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1850 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1851 ref_cache[-1 + 16] =
1852 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1854 AV_ZERO32(mv_dst - 1 + 0);
1855 AV_ZERO32(mv_dst - 1 + 8);
1856 AV_ZERO32(mv_dst - 1 + 16);
1857 AV_ZERO32(mv_dst - 1 + 24);
1860 ref_cache[-1 + 16] =
1861 ref_cache[-1 + 24] = LIST_NOT_USED;
1866 if (!USES_LIST(mb_type, list)) {
1867 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1868 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1869 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1870 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1871 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1876 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1877 int (*ref2frm)[64] = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1878 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1879 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1880 AV_WN32A(&ref_cache[0 * 8], ref01);
1881 AV_WN32A(&ref_cache[1 * 8], ref01);
1882 AV_WN32A(&ref_cache[2 * 8], ref23);
1883 AV_WN32A(&ref_cache[3 * 8], ref23);
1887 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1888 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1889 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1890 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1891 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1897 * @return non zero if the loop filter can be skipped
1899 static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
1901 const int mb_xy = sl->mb_xy;
1902 int top_xy, left_xy[LEFT_MBS];
1903 int top_type, left_type[LEFT_MBS];
1907 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1909 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1910 * stuff, I can't imagine that these complex rules are worth it. */
1912 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1913 if (FRAME_MBAFF(h)) {
1914 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1915 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1917 if (left_mb_field_flag != curr_mb_field_flag)
1918 left_xy[LTOP] -= h->mb_stride;
1920 if (curr_mb_field_flag)
1921 top_xy += h->mb_stride &
1922 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1923 if (left_mb_field_flag != curr_mb_field_flag)
1924 left_xy[LBOT] += h->mb_stride;
1928 sl->top_mb_xy = top_xy;
1929 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1930 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1932 /* For sufficiently low qp, filtering wouldn't do anything.
1933 * This is a conservative estimate: could also check beta_offset
1934 * and more accurate chroma_qp. */
1935 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1936 int qp = h->cur_pic.qscale_table[mb_xy];
1937 if (qp <= qp_thresh &&
1938 (left_xy[LTOP] < 0 ||
1939 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1941 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1942 if (!FRAME_MBAFF(h))
1944 if ((left_xy[LTOP] < 0 ||
1945 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1946 (top_xy < h->mb_stride ||
1947 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1952 top_type = h->cur_pic.mb_type[top_xy];
1953 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1954 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1955 if (sl->deblocking_filter == 2) {
1956 if (h->slice_table[top_xy] != sl->slice_num)
1958 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
1959 left_type[LTOP] = left_type[LBOT] = 0;
1961 if (h->slice_table[top_xy] == 0xFFFF)
1963 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1964 left_type[LTOP] = left_type[LBOT] = 0;
1966 sl->top_type = top_type;
1967 sl->left_type[LTOP] = left_type[LTOP];
1968 sl->left_type[LBOT] = left_type[LBOT];
1970 if (IS_INTRA(mb_type))
1973 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1974 top_type, left_type, mb_xy, 0);
1975 if (sl->list_count == 2)
1976 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1977 top_type, left_type, mb_xy, 1);
1979 nnz = h->non_zero_count[mb_xy];
1980 nnz_cache = sl->non_zero_count_cache;
1981 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1982 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1983 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1984 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1985 sl->cbp = h->cbp_table[mb_xy];
1988 nnz = h->non_zero_count[top_xy];
1989 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1992 if (left_type[LTOP]) {
1993 nnz = h->non_zero_count[left_xy[LTOP]];
1994 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
1995 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
1996 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
1997 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2000 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2001 * from what the loop filter needs */
2002 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2003 if (IS_8x8DCT(top_type)) {
2004 nnz_cache[4 + 8 * 0] =
2005 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2006 nnz_cache[6 + 8 * 0] =
2007 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2009 if (IS_8x8DCT(left_type[LTOP])) {
2010 nnz_cache[3 + 8 * 1] =
2011 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2013 if (IS_8x8DCT(left_type[LBOT])) {
2014 nnz_cache[3 + 8 * 3] =
2015 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2018 if (IS_8x8DCT(mb_type)) {
2019 nnz_cache[scan8[0]] =
2020 nnz_cache[scan8[1]] =
2021 nnz_cache[scan8[2]] =
2022 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2024 nnz_cache[scan8[0 + 4]] =
2025 nnz_cache[scan8[1 + 4]] =
2026 nnz_cache[scan8[2 + 4]] =
2027 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2029 nnz_cache[scan8[0 + 8]] =
2030 nnz_cache[scan8[1 + 8]] =
2031 nnz_cache[scan8[2 + 8]] =
2032 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2034 nnz_cache[scan8[0 + 12]] =
2035 nnz_cache[scan8[1 + 12]] =
2036 nnz_cache[scan8[2 + 12]] =
2037 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2044 static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2046 uint8_t *dest_y, *dest_cb, *dest_cr;
2047 int linesize, uvlinesize, mb_x, mb_y;
2048 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2049 const int old_slice_type = sl->slice_type;
2050 const int pixel_shift = h->pixel_shift;
2051 const int block_h = 16 >> h->chroma_y_shift;
2053 if (sl->deblocking_filter) {
2054 for (mb_x = start_x; mb_x < end_x; mb_x++)
2055 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2057 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2058 sl->slice_num = h->slice_table[mb_xy];
2059 mb_type = h->cur_pic.mb_type[mb_xy];
2060 sl->list_count = h->list_counts[mb_xy];
2064 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2068 dest_y = h->cur_pic.f.data[0] +
2069 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2070 dest_cb = h->cur_pic.f.data[1] +
2071 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2072 mb_y * h->uvlinesize * block_h;
2073 dest_cr = h->cur_pic.f.data[2] +
2074 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2075 mb_y * h->uvlinesize * block_h;
2076 // FIXME simplify above
2079 linesize = sl->mb_linesize = h->linesize * 2;
2080 uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2;
2081 if (mb_y & 1) { // FIXME move out of this function?
2082 dest_y -= h->linesize * 15;
2083 dest_cb -= h->uvlinesize * (block_h - 1);
2084 dest_cr -= h->uvlinesize * (block_h - 1);
2087 linesize = sl->mb_linesize = h->linesize;
2088 uvlinesize = sl->mb_uvlinesize = h->uvlinesize;
2090 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2092 if (fill_filter_caches(h, sl, mb_type))
2094 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2095 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2097 if (FRAME_MBAFF(h)) {
2098 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2099 linesize, uvlinesize);
2101 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2102 dest_cr, linesize, uvlinesize);
2106 sl->slice_type = old_slice_type;
2108 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2109 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2110 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2113 static void predict_field_decoding_flag(H264Context *h, H264SliceContext *sl)
2115 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2116 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2117 h->cur_pic.mb_type[mb_xy - 1] :
2118 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2119 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2120 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2124 * Draw edges and report progress for the last MB row.
2126 static void decode_finish_row(H264Context *h, H264SliceContext *sl)
2128 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2129 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2130 int height = 16 << FRAME_MBAFF(h);
2131 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2133 if (sl->deblocking_filter) {
2134 if ((top + height) >= pic_height)
2135 height += deblock_border;
2136 top -= deblock_border;
2139 if (top >= pic_height || (top + height) < 0)
2142 height = FFMIN(height, pic_height - top);
2144 height = top + height;
2148 ff_h264_draw_horiz_band(h, sl, top, height);
2153 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2154 h->picture_structure == PICT_BOTTOM_FIELD);
2157 static void er_add_slice(H264SliceContext *sl,
2158 int startx, int starty,
2159 int endx, int endy, int status)
2161 #if CONFIG_ERROR_RESILIENCE
2162 ERContext *er = &sl->er;
2164 er->ref_count = sl->ref_count[0];
2165 ff_er_add_slice(er, startx, starty, endx, endy, status);
2169 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2171 H264SliceContext *sl = arg;
2172 H264Context *h = sl->h264;
2173 int lf_x_start = sl->mb_x;
2176 ret = alloc_scratch_buffers(sl, h->linesize);
2180 sl->mb_skip_run = -1;
2182 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2183 avctx->codec_id != AV_CODEC_ID_H264 ||
2184 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2188 align_get_bits(&sl->gb);
2191 ff_init_cabac_decoder(&sl->cabac,
2192 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2193 (get_bits_left(&sl->gb) + 7) / 8);
2195 ff_h264_init_cabac_states(h, sl);
2199 int ret = ff_h264_decode_mb_cabac(h, sl);
2201 // STOP_TIMER("decode_mb_cabac")
2204 ff_h264_hl_decode_mb(h, sl);
2206 // FIXME optimal? or let mb_decode decode 16x32 ?
2207 if (ret >= 0 && FRAME_MBAFF(h)) {
2210 ret = ff_h264_decode_mb_cabac(h, sl);
2213 ff_h264_hl_decode_mb(h, sl);
2216 eos = get_cabac_terminate(&sl->cabac);
2218 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2219 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2220 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2221 sl->mb_y, ER_MB_END);
2222 if (sl->mb_x >= lf_x_start)
2223 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2226 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2227 av_log(h->avctx, AV_LOG_ERROR,
2228 "error while decoding MB %d %d, bytestream %td\n",
2230 sl->cabac.bytestream_end - sl->cabac.bytestream);
2231 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2232 sl->mb_y, ER_MB_ERROR);
2233 return AVERROR_INVALIDDATA;
2236 if (++sl->mb_x >= h->mb_width) {
2237 loop_filter(h, sl, lf_x_start, sl->mb_x);
2238 sl->mb_x = lf_x_start = 0;
2239 decode_finish_row(h, sl);
2241 if (FIELD_OR_MBAFF_PICTURE(h)) {
2243 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2244 predict_field_decoding_flag(h, sl);
2248 if (eos || sl->mb_y >= h->mb_height) {
2249 tprintf(h->avctx, "slice end %d %d\n",
2250 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2251 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2252 sl->mb_y, ER_MB_END);
2253 if (sl->mb_x > lf_x_start)
2254 loop_filter(h, sl, lf_x_start, sl->mb_x);
2260 int ret = ff_h264_decode_mb_cavlc(h, sl);
2263 ff_h264_hl_decode_mb(h, sl);
2265 // FIXME optimal? or let mb_decode decode 16x32 ?
2266 if (ret >= 0 && FRAME_MBAFF(h)) {
2268 ret = ff_h264_decode_mb_cavlc(h, sl);
2271 ff_h264_hl_decode_mb(h, sl);
2276 av_log(h->avctx, AV_LOG_ERROR,
2277 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2278 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2279 sl->mb_y, ER_MB_ERROR);
2283 if (++sl->mb_x >= h->mb_width) {
2284 loop_filter(h, sl, lf_x_start, sl->mb_x);
2285 sl->mb_x = lf_x_start = 0;
2286 decode_finish_row(h, sl);
2288 if (FIELD_OR_MBAFF_PICTURE(h)) {
2290 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2291 predict_field_decoding_flag(h, sl);
2293 if (sl->mb_y >= h->mb_height) {
2294 tprintf(h->avctx, "slice end %d %d\n",
2295 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2297 if (get_bits_left(&sl->gb) == 0) {
2298 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2299 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2303 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2304 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2306 return AVERROR_INVALIDDATA;
2311 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2312 tprintf(h->avctx, "slice end %d %d\n",
2313 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2315 if (get_bits_left(&sl->gb) == 0) {
2316 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2317 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2318 if (sl->mb_x > lf_x_start)
2319 loop_filter(h, sl, lf_x_start, sl->mb_x);
2323 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2324 sl->mb_y, ER_MB_ERROR);
2326 return AVERROR_INVALIDDATA;
2334 * Call decode_slice() for each context.
2336 * @param h h264 master context
2337 * @param context_count number of contexts to execute
2339 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2341 AVCodecContext *const avctx = h->avctx;
2343 H264SliceContext *sl;
2346 if (h->avctx->hwaccel)
2348 if (context_count == 1) {
2349 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2350 h->mb_y = h->slice_ctx[0].mb_y;
2353 for (i = 1; i < context_count; i++) {
2354 sl = &h->slice_ctx[i];
2355 sl->er.error_count = 0;
2358 avctx->execute(avctx, decode_slice, h->slice_ctx,
2359 NULL, context_count, sizeof(h->slice_ctx[0]));
2361 /* pull back stuff from slices to master context */
2362 hx = h->thread_context[context_count - 1];
2363 sl = &h->slice_ctx[context_count - 1];
2365 h->droppable = hx->droppable;
2366 h->picture_structure = hx->picture_structure;
2367 for (i = 1; i < context_count; i++)
2368 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;