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 if (h->avctx->hwaccel) {
232 const AVHWAccel *hwaccel = h->avctx->hwaccel;
233 av_assert0(!pic->hwaccel_picture_private);
234 if (hwaccel->frame_priv_data_size) {
235 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
236 if (!pic->hwaccel_priv_buf)
237 return AVERROR(ENOMEM);
238 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
242 if (!h->qscale_table_pool) {
243 ret = init_table_pools(h);
248 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
249 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
250 if (!pic->qscale_table_buf || !pic->mb_type_buf)
253 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
254 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
256 for (i = 0; i < 2; i++) {
257 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
258 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
259 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
262 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
263 pic->ref_index[i] = pic->ref_index_buf[i]->data;
268 ff_h264_unref_picture(h, pic);
269 return (ret < 0) ? ret : AVERROR(ENOMEM);
272 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
279 static int find_unused_picture(H264Context *h)
283 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
284 if (pic_is_unused(h, &h->DPB[i]))
287 if (i == H264_MAX_PICTURE_COUNT)
288 return AVERROR_INVALIDDATA;
294 static void init_dequant8_coeff_table(H264Context *h)
297 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
299 for (i = 0; i < 6; i++) {
300 h->dequant8_coeff[i] = h->dequant8_buffer[i];
301 for (j = 0; j < i; j++)
302 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
303 64 * sizeof(uint8_t))) {
304 h->dequant8_coeff[i] = h->dequant8_buffer[j];
310 for (q = 0; q < max_qp + 1; q++) {
313 for (x = 0; x < 64; x++)
314 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
315 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
316 h->pps.scaling_matrix8[i][x]) << shift;
321 static void init_dequant4_coeff_table(H264Context *h)
324 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
325 for (i = 0; i < 6; i++) {
326 h->dequant4_coeff[i] = h->dequant4_buffer[i];
327 for (j = 0; j < i; j++)
328 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
329 16 * sizeof(uint8_t))) {
330 h->dequant4_coeff[i] = h->dequant4_buffer[j];
336 for (q = 0; q < max_qp + 1; q++) {
337 int shift = div6[q] + 2;
339 for (x = 0; x < 16; x++)
340 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
341 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
342 h->pps.scaling_matrix4[i][x]) << shift;
347 void h264_init_dequant_tables(H264Context *h)
350 init_dequant4_coeff_table(h);
351 if (h->pps.transform_8x8_mode)
352 init_dequant8_coeff_table(h);
353 if (h->sps.transform_bypass) {
354 for (i = 0; i < 6; i++)
355 for (x = 0; x < 16; x++)
356 h->dequant4_coeff[i][0][x] = 1 << 6;
357 if (h->pps.transform_8x8_mode)
358 for (i = 0; i < 6; i++)
359 for (x = 0; x < 64; x++)
360 h->dequant8_coeff[i][0][x] = 1 << 6;
364 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
366 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
367 ((pic && pic >= old_ctx->DPB && \
368 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
369 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
371 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
372 H264Context *new_base,
373 H264Context *old_base)
377 for (i = 0; i < count; i++) {
378 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
379 IN_RANGE(from[i], old_base->DPB,
380 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
382 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
386 static int copy_parameter_set(void **to, void **from, int count, int size)
390 for (i = 0; i < count; i++) {
391 if (to[i] && !from[i]) {
393 } else if (from[i] && !to[i]) {
394 to[i] = av_malloc(size);
396 return AVERROR(ENOMEM);
400 memcpy(to[i], from[i], size);
406 #define copy_fields(to, from, start_field, end_field) \
407 memcpy(&to->start_field, &from->start_field, \
408 (char *)&to->end_field - (char *)&to->start_field)
410 static int h264_slice_header_init(H264Context *h);
412 int ff_h264_update_thread_context(AVCodecContext *dst,
413 const AVCodecContext *src)
415 H264Context *h = dst->priv_data, *h1 = src->priv_data;
416 int inited = h->context_initialized, err = 0;
420 if (dst == src || !h1->context_initialized)
424 (h->width != h1->width ||
425 h->height != h1->height ||
426 h->mb_width != h1->mb_width ||
427 h->mb_height != h1->mb_height ||
428 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
429 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
430 h->sps.colorspace != h1->sps.colorspace)) {
435 if ((ret = copy_parameter_set((void **)h->sps_buffers,
436 (void **)h1->sps_buffers,
437 MAX_SPS_COUNT, sizeof(SPS))) < 0)
440 if ((ret = copy_parameter_set((void **)h->pps_buffers,
441 (void **)h1->pps_buffers,
442 MAX_PPS_COUNT, sizeof(PPS))) < 0)
446 if (need_reinit || !inited) {
447 h->width = h1->width;
448 h->height = h1->height;
449 h->mb_height = h1->mb_height;
450 h->mb_width = h1->mb_width;
451 h->mb_num = h1->mb_num;
452 h->mb_stride = h1->mb_stride;
453 h->b_stride = h1->b_stride;
455 if ((err = h264_slice_header_init(h)) < 0) {
456 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
460 /* copy block_offset since frame_start may not be called */
461 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
464 h->avctx->coded_height = h1->avctx->coded_height;
465 h->avctx->coded_width = h1->avctx->coded_width;
466 h->avctx->width = h1->avctx->width;
467 h->avctx->height = h1->avctx->height;
468 h->coded_picture_number = h1->coded_picture_number;
469 h->first_field = h1->first_field;
470 h->picture_structure = h1->picture_structure;
471 h->droppable = h1->droppable;
472 h->low_delay = h1->low_delay;
474 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
475 ff_h264_unref_picture(h, &h->DPB[i]);
476 if (h1->DPB[i].f->buf[0] &&
477 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
481 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
482 ff_h264_unref_picture(h, &h->cur_pic);
483 if (h1->cur_pic.f->buf[0]) {
484 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
489 h->enable_er = h1->enable_er;
490 h->workaround_bugs = h1->workaround_bugs;
491 h->low_delay = h1->low_delay;
492 h->droppable = h1->droppable;
494 // extradata/NAL handling
495 h->is_avc = h1->is_avc;
496 h->nal_length_size = h1->nal_length_size;
498 // Dequantization matrices
499 // FIXME these are big - can they be only copied when PPS changes?
500 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
502 for (i = 0; i < 6; i++)
503 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
504 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
506 for (i = 0; i < 6; i++)
507 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
508 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
510 h->dequant_coeff_pps = h1->dequant_coeff_pps;
513 copy_fields(h, h1, poc_lsb, current_slice);
515 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
516 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
517 copy_picture_range(h->delayed_pic, h1->delayed_pic,
518 MAX_DELAYED_PIC_COUNT + 2, h, h1);
524 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
525 h->prev_poc_msb = h->poc_msb;
526 h->prev_poc_lsb = h->poc_lsb;
528 h->prev_frame_num_offset = h->frame_num_offset;
529 h->prev_frame_num = h->frame_num;
531 h->recovery_frame = h1->recovery_frame;
532 h->frame_recovered = h1->frame_recovered;
537 static int h264_frame_start(H264Context *h)
541 const int pixel_shift = h->pixel_shift;
543 release_unused_pictures(h, 1);
544 h->cur_pic_ptr = NULL;
546 i = find_unused_picture(h);
548 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
553 pic->reference = h->droppable ? 0 : h->picture_structure;
554 pic->f->coded_picture_number = h->coded_picture_number++;
555 pic->field_picture = h->picture_structure != PICT_FRAME;
557 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
559 * See decode_nal_units().
561 pic->f->key_frame = 0;
565 if ((ret = alloc_picture(h, pic)) < 0)
568 h->cur_pic_ptr = pic;
569 ff_h264_unref_picture(h, &h->cur_pic);
570 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
573 if (CONFIG_ERROR_RESILIENCE && h->enable_er)
574 ff_er_frame_start(&h->slice_ctx[0].er);
576 for (i = 0; i < 16; i++) {
577 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
578 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
580 for (i = 0; i < 16; i++) {
581 h->block_offset[16 + i] =
582 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
583 h->block_offset[48 + 16 + i] =
584 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
587 /* Some macroblocks can be accessed before they're available in case
588 * of lost slices, MBAFF or threading. */
589 memset(h->slice_table, -1,
590 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
592 /* We mark the current picture as non-reference after allocating it, so
593 * that if we break out due to an error it can be released automatically
594 * in the next ff_mpv_frame_start().
596 h->cur_pic_ptr->reference = 0;
598 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
600 h->next_output_pic = NULL;
602 assert(h->cur_pic_ptr->long_ref == 0);
607 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
609 uint8_t *src_cb, uint8_t *src_cr,
610 int linesize, int uvlinesize,
615 const int pixel_shift = h->pixel_shift;
616 int chroma444 = CHROMA444(h);
617 int chroma422 = CHROMA422(h);
620 src_cb -= uvlinesize;
621 src_cr -= uvlinesize;
623 if (!simple && FRAME_MBAFF(h)) {
626 top_border = sl->top_borders[0][sl->mb_x];
627 AV_COPY128(top_border, src_y + 15 * linesize);
629 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
630 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
633 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
634 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
635 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
636 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
638 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
639 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
641 } else if (chroma422) {
643 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
644 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
646 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
647 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
651 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
652 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
654 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
655 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
660 } else if (MB_MBAFF(sl)) {
666 top_border = sl->top_borders[top_idx][sl->mb_x];
667 /* There are two lines saved, the line above the top macroblock
668 * of a pair, and the line above the bottom macroblock. */
669 AV_COPY128(top_border, src_y + 16 * linesize);
671 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
673 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
676 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
677 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
678 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
679 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
681 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
682 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
684 } else if (chroma422) {
686 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
687 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
689 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
690 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
694 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
695 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
697 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
698 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
705 * Initialize implicit_weight table.
706 * @param field 0/1 initialize the weight for interlaced MBAFF
707 * -1 initializes the rest
709 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
711 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
713 for (i = 0; i < 2; i++) {
714 sl->luma_weight_flag[i] = 0;
715 sl->chroma_weight_flag[i] = 0;
719 if (h->picture_structure == PICT_FRAME) {
720 cur_poc = h->cur_pic_ptr->poc;
722 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
724 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
725 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
727 sl->use_weight_chroma = 0;
731 ref_count0 = sl->ref_count[0];
732 ref_count1 = sl->ref_count[1];
734 cur_poc = h->cur_pic_ptr->field_poc[field];
736 ref_count0 = 16 + 2 * sl->ref_count[0];
737 ref_count1 = 16 + 2 * sl->ref_count[1];
741 sl->use_weight_chroma = 2;
742 sl->luma_log2_weight_denom = 5;
743 sl->chroma_log2_weight_denom = 5;
745 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
746 int poc0 = sl->ref_list[0][ref0].poc;
747 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
749 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
750 int poc1 = sl->ref_list[1][ref1].poc;
751 int td = av_clip_int8(poc1 - poc0);
753 int tb = av_clip_int8(cur_poc - poc0);
754 int tx = (16384 + (FFABS(td) >> 1)) / td;
755 int dist_scale_factor = (tb * tx + 32) >> 8;
756 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
757 w = 64 - dist_scale_factor;
761 sl->implicit_weight[ref0][ref1][0] =
762 sl->implicit_weight[ref0][ref1][1] = w;
764 sl->implicit_weight[ref0][ref1][field] = w;
771 * initialize scan tables
773 static void init_scan_tables(H264Context *h)
776 for (i = 0; i < 16; i++) {
777 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
778 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
779 h->field_scan[i] = TRANSPOSE(field_scan[i]);
782 for (i = 0; i < 64; i++) {
783 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
784 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
785 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
786 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
787 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
790 if (h->sps.transform_bypass) { // FIXME same ugly
791 h->zigzag_scan_q0 = zigzag_scan;
792 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
793 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
794 h->field_scan_q0 = field_scan;
795 h->field_scan8x8_q0 = field_scan8x8;
796 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
798 h->zigzag_scan_q0 = h->zigzag_scan;
799 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
800 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
801 h->field_scan_q0 = h->field_scan;
802 h->field_scan8x8_q0 = h->field_scan8x8;
803 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
807 static enum AVPixelFormat get_pixel_format(H264Context *h)
809 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
810 CONFIG_H264_D3D11VA_HWACCEL + \
811 CONFIG_H264_VAAPI_HWACCEL + \
812 (CONFIG_H264_VDA_HWACCEL * 2) + \
813 CONFIG_H264_VDPAU_HWACCEL)
814 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
815 const enum AVPixelFormat *choices = pix_fmts;
817 switch (h->sps.bit_depth_luma) {
820 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
821 *fmt++ = AV_PIX_FMT_GBRP9;
823 *fmt++ = AV_PIX_FMT_YUV444P9;
824 } else if (CHROMA422(h))
825 *fmt++ = AV_PIX_FMT_YUV422P9;
827 *fmt++ = AV_PIX_FMT_YUV420P9;
831 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
832 *fmt++ = AV_PIX_FMT_GBRP10;
834 *fmt++ = AV_PIX_FMT_YUV444P10;
835 } else if (CHROMA422(h))
836 *fmt++ = AV_PIX_FMT_YUV422P10;
838 *fmt++ = AV_PIX_FMT_YUV420P10;
841 #if CONFIG_H264_VDPAU_HWACCEL
842 *fmt++ = AV_PIX_FMT_VDPAU;
845 if (h->avctx->colorspace == AVCOL_SPC_RGB)
846 *fmt++ = AV_PIX_FMT_GBRP;
847 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
848 *fmt++ = AV_PIX_FMT_YUVJ444P;
850 *fmt++ = AV_PIX_FMT_YUV444P;
851 } else if (CHROMA422(h)) {
852 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
853 *fmt++ = AV_PIX_FMT_YUVJ422P;
855 *fmt++ = AV_PIX_FMT_YUV422P;
857 #if CONFIG_H264_DXVA2_HWACCEL
858 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
860 #if CONFIG_H264_D3D11VA_HWACCEL
861 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
863 #if CONFIG_H264_VAAPI_HWACCEL
864 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
866 #if CONFIG_H264_VDA_HWACCEL
867 *fmt++ = AV_PIX_FMT_VDA_VLD;
868 *fmt++ = AV_PIX_FMT_VDA;
870 if (h->avctx->codec->pix_fmts)
871 choices = h->avctx->codec->pix_fmts;
872 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
873 *fmt++ = AV_PIX_FMT_YUVJ420P;
875 *fmt++ = AV_PIX_FMT_YUV420P;
879 av_log(h->avctx, AV_LOG_ERROR,
880 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
881 return AVERROR_INVALIDDATA;
884 *fmt = AV_PIX_FMT_NONE;
886 return ff_get_format(h->avctx, choices);
889 /* export coded and cropped frame dimensions to AVCodecContext */
890 static int init_dimensions(H264Context *h)
892 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
893 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
895 /* handle container cropping */
896 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
897 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16)) {
898 width = h->avctx->width;
899 height = h->avctx->height;
902 if (width <= 0 || height <= 0) {
903 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
905 if (h->avctx->err_recognition & AV_EF_EXPLODE)
906 return AVERROR_INVALIDDATA;
908 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
919 h->avctx->coded_width = h->width;
920 h->avctx->coded_height = h->height;
921 h->avctx->width = width;
922 h->avctx->height = height;
927 static int h264_slice_header_init(H264Context *h)
929 int nb_slices = (HAVE_THREADS &&
930 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
931 h->avctx->thread_count : 1;
934 ff_set_sar(h->avctx, h->sps.sar);
935 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
936 &h->chroma_x_shift, &h->chroma_y_shift);
938 if (h->sps.timing_info_present_flag) {
939 int64_t den = h->sps.time_scale;
940 if (h->x264_build < 44U)
942 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
943 h->sps.num_units_in_tick, den, 1 << 30);
946 ff_h264_free_tables(h);
949 h->prev_interlaced_frame = 1;
952 ret = ff_h264_alloc_tables(h);
954 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
958 if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 10) {
959 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
960 h->sps.bit_depth_luma);
961 return AVERROR_INVALIDDATA;
964 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
965 h->pixel_shift = h->sps.bit_depth_luma > 8;
966 h->chroma_format_idc = h->sps.chroma_format_idc;
967 h->bit_depth_luma = h->sps.bit_depth_luma;
969 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
970 h->sps.chroma_format_idc);
971 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
972 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
973 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
974 h->sps.chroma_format_idc);
975 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
977 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
980 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
982 max_slices = H264_MAX_THREADS;
983 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
984 " reducing to %d\n", nb_slices, max_slices);
985 nb_slices = max_slices;
987 h->slice_context_count = nb_slices;
989 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
990 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
992 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
996 for (i = 0; i < h->slice_context_count; i++) {
997 H264SliceContext *sl = &h->slice_ctx[i];
1000 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1001 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1002 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1004 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1005 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1011 h->context_initialized = 1;
1017 * Decode a slice header.
1018 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1020 * @param h h264context
1022 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1024 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1026 unsigned int first_mb_in_slice;
1027 unsigned int pps_id;
1029 unsigned int slice_type, tmp, i, j;
1030 int last_pic_structure, last_pic_droppable;
1031 int needs_reinit = 0;
1032 int field_pic_flag, bottom_field_flag;
1033 int frame_num, droppable, picture_structure;
1034 int mb_aff_frame = 0;
1036 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1037 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1039 first_mb_in_slice = get_ue_golomb(&sl->gb);
1041 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1042 if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1043 ff_h264_field_end(h, sl, 1);
1046 h->current_slice = 0;
1047 if (!h->first_field) {
1048 if (h->cur_pic_ptr && !h->droppable) {
1049 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1050 h->picture_structure == PICT_BOTTOM_FIELD);
1052 h->cur_pic_ptr = NULL;
1056 slice_type = get_ue_golomb_31(&sl->gb);
1057 if (slice_type > 9) {
1058 av_log(h->avctx, AV_LOG_ERROR,
1059 "slice type %d too large at %d\n",
1060 slice_type, first_mb_in_slice);
1061 return AVERROR_INVALIDDATA;
1063 if (slice_type > 4) {
1065 sl->slice_type_fixed = 1;
1067 sl->slice_type_fixed = 0;
1069 slice_type = golomb_to_pict_type[slice_type];
1070 sl->slice_type = slice_type;
1071 sl->slice_type_nos = slice_type & 3;
1073 if (h->nal_unit_type == NAL_IDR_SLICE &&
1074 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1075 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1076 return AVERROR_INVALIDDATA;
1079 // to make a few old functions happy, it's wrong though
1080 if (!h->setup_finished)
1081 h->pict_type = sl->slice_type;
1083 pps_id = get_ue_golomb(&sl->gb);
1084 if (pps_id >= MAX_PPS_COUNT) {
1085 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1086 return AVERROR_INVALIDDATA;
1088 if (!h->pps_buffers[pps_id]) {
1089 av_log(h->avctx, AV_LOG_ERROR,
1090 "non-existing PPS %u referenced\n",
1092 return AVERROR_INVALIDDATA;
1094 if (!h->setup_finished) {
1095 h->pps = *h->pps_buffers[pps_id];
1096 } else if (h->dequant_coeff_pps != pps_id) {
1097 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1098 return AVERROR_INVALIDDATA;
1101 if (!h->sps_buffers[h->pps.sps_id]) {
1102 av_log(h->avctx, AV_LOG_ERROR,
1103 "non-existing SPS %u referenced\n",
1105 return AVERROR_INVALIDDATA;
1108 if (h->pps.sps_id != h->sps.sps_id ||
1109 h->sps_buffers[h->pps.sps_id]->new) {
1110 h->sps_buffers[h->pps.sps_id]->new = 0;
1112 h->sps = *h->sps_buffers[h->pps.sps_id];
1114 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1115 h->chroma_format_idc != h->sps.chroma_format_idc)
1118 if (h->flags & AV_CODEC_FLAG_LOW_DELAY ||
1119 (h->sps.bitstream_restriction_flag &&
1120 !h->sps.num_reorder_frames)) {
1121 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1122 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1123 "Reenabling low delay requires a codec flush.\n");
1128 if (h->avctx->has_b_frames < 2)
1129 h->avctx->has_b_frames = !h->low_delay;
1133 if (!h->setup_finished) {
1134 h->avctx->profile = ff_h264_get_profile(&h->sps);
1135 h->avctx->level = h->sps.level_idc;
1136 h->avctx->refs = h->sps.ref_frame_count;
1138 if (h->mb_width != h->sps.mb_width ||
1139 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1142 h->mb_width = h->sps.mb_width;
1143 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1144 h->mb_num = h->mb_width * h->mb_height;
1145 h->mb_stride = h->mb_width + 1;
1147 h->b_stride = h->mb_width * 4;
1149 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1151 h->width = 16 * h->mb_width;
1152 h->height = 16 * h->mb_height;
1154 ret = init_dimensions(h);
1158 if (h->sps.video_signal_type_present_flag) {
1159 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1161 if (h->sps.colour_description_present_flag) {
1162 if (h->avctx->colorspace != h->sps.colorspace)
1164 h->avctx->color_primaries = h->sps.color_primaries;
1165 h->avctx->color_trc = h->sps.color_trc;
1166 h->avctx->colorspace = h->sps.colorspace;
1171 if (h->context_initialized && needs_reinit) {
1172 h->context_initialized = 0;
1173 if (sl != h->slice_ctx) {
1174 av_log(h->avctx, AV_LOG_ERROR,
1175 "changing width %d -> %d / height %d -> %d on "
1177 h->width, h->avctx->coded_width,
1178 h->height, h->avctx->coded_height,
1179 h->current_slice + 1);
1180 return AVERROR_INVALIDDATA;
1183 ff_h264_flush_change(h);
1185 if ((ret = get_pixel_format(h)) < 0)
1187 h->avctx->pix_fmt = ret;
1189 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1190 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1192 if ((ret = h264_slice_header_init(h)) < 0) {
1193 av_log(h->avctx, AV_LOG_ERROR,
1194 "h264_slice_header_init() failed\n");
1198 if (!h->context_initialized) {
1199 if (sl != h->slice_ctx) {
1200 av_log(h->avctx, AV_LOG_ERROR,
1201 "Cannot (re-)initialize context during parallel decoding.\n");
1202 return AVERROR_PATCHWELCOME;
1205 if ((ret = get_pixel_format(h)) < 0)
1207 h->avctx->pix_fmt = ret;
1209 if ((ret = h264_slice_header_init(h)) < 0) {
1210 av_log(h->avctx, AV_LOG_ERROR,
1211 "h264_slice_header_init() failed\n");
1216 if (sl == h->slice_ctx && h->dequant_coeff_pps != pps_id) {
1217 h->dequant_coeff_pps = pps_id;
1218 h264_init_dequant_tables(h);
1221 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1222 if (!h->setup_finished)
1223 h->frame_num = frame_num;
1227 last_pic_structure = h->picture_structure;
1228 last_pic_droppable = h->droppable;
1230 droppable = h->nal_ref_idc == 0;
1231 if (h->sps.frame_mbs_only_flag) {
1232 picture_structure = PICT_FRAME;
1234 field_pic_flag = get_bits1(&sl->gb);
1235 if (field_pic_flag) {
1236 bottom_field_flag = get_bits1(&sl->gb);
1237 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1239 picture_structure = PICT_FRAME;
1240 mb_aff_frame = h->sps.mb_aff;
1243 if (!h->setup_finished) {
1244 h->droppable = droppable;
1245 h->picture_structure = picture_structure;
1246 h->mb_aff_frame = mb_aff_frame;
1248 sl->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1250 if (h->current_slice != 0) {
1251 if (last_pic_structure != picture_structure ||
1252 last_pic_droppable != droppable) {
1253 av_log(h->avctx, AV_LOG_ERROR,
1254 "Changing field mode (%d -> %d) between slices is not allowed\n",
1255 last_pic_structure, h->picture_structure);
1256 return AVERROR_INVALIDDATA;
1257 } else if (!h->cur_pic_ptr) {
1258 av_log(h->avctx, AV_LOG_ERROR,
1259 "unset cur_pic_ptr on slice %d\n",
1260 h->current_slice + 1);
1261 return AVERROR_INVALIDDATA;
1264 /* Shorten frame num gaps so we don't have to allocate reference
1265 * frames just to throw them away */
1266 if (h->frame_num != h->prev_frame_num) {
1267 int unwrap_prev_frame_num = h->prev_frame_num;
1268 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1270 if (unwrap_prev_frame_num > h->frame_num)
1271 unwrap_prev_frame_num -= max_frame_num;
1273 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1274 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1275 if (unwrap_prev_frame_num < 0)
1276 unwrap_prev_frame_num += max_frame_num;
1278 h->prev_frame_num = unwrap_prev_frame_num;
1282 /* See if we have a decoded first field looking for a pair...
1283 * Here, we're using that to see if we should mark previously
1284 * decode frames as "finished".
1285 * We have to do that before the "dummy" in-between frame allocation,
1286 * since that can modify s->current_picture_ptr. */
1287 if (h->first_field) {
1288 assert(h->cur_pic_ptr);
1289 assert(h->cur_pic_ptr->f->buf[0]);
1290 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1292 /* figure out if we have a complementary field pair */
1293 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1294 /* Previous field is unmatched. Don't display it, but let it
1295 * remain for reference if marked as such. */
1296 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1297 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1298 last_pic_structure == PICT_TOP_FIELD);
1301 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1302 /* This and previous field were reference, but had
1303 * different frame_nums. Consider this field first in
1304 * pair. Throw away previous field except for reference
1306 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1307 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1308 last_pic_structure == PICT_TOP_FIELD);
1311 /* Second field in complementary pair */
1312 if (!((last_pic_structure == PICT_TOP_FIELD &&
1313 h->picture_structure == PICT_BOTTOM_FIELD) ||
1314 (last_pic_structure == PICT_BOTTOM_FIELD &&
1315 h->picture_structure == PICT_TOP_FIELD))) {
1316 av_log(h->avctx, AV_LOG_ERROR,
1317 "Invalid field mode combination %d/%d\n",
1318 last_pic_structure, h->picture_structure);
1319 h->picture_structure = last_pic_structure;
1320 h->droppable = last_pic_droppable;
1321 return AVERROR_INVALIDDATA;
1322 } else if (last_pic_droppable != h->droppable) {
1323 avpriv_request_sample(h->avctx,
1324 "Found reference and non-reference fields in the same frame, which");
1325 h->picture_structure = last_pic_structure;
1326 h->droppable = last_pic_droppable;
1327 return AVERROR_PATCHWELCOME;
1333 while (h->frame_num != h->prev_frame_num &&
1334 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1335 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1336 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1337 h->frame_num, h->prev_frame_num);
1338 ret = h264_frame_start(h);
1344 h->prev_frame_num++;
1345 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1346 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1347 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1348 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1349 ret = ff_generate_sliding_window_mmcos(h, 1);
1350 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1352 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1353 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1355 /* Error concealment: If a ref is missing, copy the previous ref
1357 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1358 * many assumptions about there being no actual duplicates.
1359 * FIXME: This does not copy padding for out-of-frame motion
1360 * vectors. Given we are concealing a lost frame, this probably
1361 * is not noticeable by comparison, but it should be fixed. */
1362 if (h->short_ref_count) {
1364 h->short_ref[0]->f->width == prev->f->width &&
1365 h->short_ref[0]->f->height == prev->f->height &&
1366 h->short_ref[0]->f->format == prev->f->format) {
1367 av_image_copy(h->short_ref[0]->f->data,
1368 h->short_ref[0]->f->linesize,
1369 (const uint8_t **)prev->f->data,
1374 h->short_ref[0]->poc = prev->poc + 2;
1376 h->short_ref[0]->frame_num = h->prev_frame_num;
1380 /* See if we have a decoded first field looking for a pair...
1381 * We're using that to see whether to continue decoding in that
1382 * frame, or to allocate a new one. */
1383 if (h->first_field) {
1384 assert(h->cur_pic_ptr);
1385 assert(h->cur_pic_ptr->f->buf[0]);
1386 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1388 /* figure out if we have a complementary field pair */
1389 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1390 /* Previous field is unmatched. Don't display it, but let it
1391 * remain for reference if marked as such. */
1392 h->cur_pic_ptr = NULL;
1393 h->first_field = FIELD_PICTURE(h);
1395 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1396 /* This and the previous field had different frame_nums.
1397 * Consider this field first in pair. Throw away previous
1398 * one except for reference purposes. */
1400 h->cur_pic_ptr = NULL;
1402 /* Second field in complementary pair */
1407 /* Frame or first field in a potentially complementary pair */
1408 h->first_field = FIELD_PICTURE(h);
1411 if (!FIELD_PICTURE(h) || h->first_field) {
1412 if (h264_frame_start(h) < 0) {
1414 return AVERROR_INVALIDDATA;
1417 release_unused_pictures(h, 0);
1421 if (!h->setup_finished)
1422 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1424 assert(h->mb_num == h->mb_width * h->mb_height);
1425 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1426 first_mb_in_slice >= h->mb_num) {
1427 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1428 return AVERROR_INVALIDDATA;
1430 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1431 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1432 FIELD_OR_MBAFF_PICTURE(h);
1433 if (h->picture_structure == PICT_BOTTOM_FIELD)
1434 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1435 assert(sl->mb_y < h->mb_height);
1437 if (h->picture_structure == PICT_FRAME) {
1438 h->curr_pic_num = h->frame_num;
1439 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1441 h->curr_pic_num = 2 * h->frame_num + 1;
1442 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1445 if (h->nal_unit_type == NAL_IDR_SLICE)
1446 get_ue_golomb(&sl->gb); /* idr_pic_id */
1448 if (h->sps.poc_type == 0) {
1449 int poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1451 if (!h->setup_finished)
1452 h->poc_lsb = poc_lsb;
1454 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1455 int delta_poc_bottom = get_se_golomb(&sl->gb);
1456 if (!h->setup_finished)
1457 h->delta_poc_bottom = delta_poc_bottom;
1461 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1462 int delta_poc = get_se_golomb(&sl->gb);
1464 if (!h->setup_finished)
1465 h->delta_poc[0] = delta_poc;
1467 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1468 delta_poc = get_se_golomb(&sl->gb);
1470 if (!h->setup_finished)
1471 h->delta_poc[1] = delta_poc;
1475 if (!h->setup_finished)
1476 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1478 if (h->pps.redundant_pic_cnt_present)
1479 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1481 ret = ff_set_ref_count(h, sl);
1485 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1486 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1488 sl->ref_count[1] = sl->ref_count[0] = 0;
1493 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1494 (h->pps.weighted_bipred_idc == 1 &&
1495 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1496 ff_pred_weight_table(h, sl);
1497 else if (h->pps.weighted_bipred_idc == 2 &&
1498 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1499 implicit_weight_table(h, sl, -1);
1502 for (i = 0; i < 2; i++) {
1503 sl->luma_weight_flag[i] = 0;
1504 sl->chroma_weight_flag[i] = 0;
1508 // If frame-mt is enabled, only update mmco tables for the first slice
1509 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1510 // or h->mmco, which will cause ref list mix-ups and decoding errors
1511 // further down the line. This may break decoding if the first slice is
1512 // corrupt, thus we only do this if frame-mt is enabled.
1513 if (h->nal_ref_idc) {
1514 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1515 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1516 h->current_slice == 0);
1517 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1518 return AVERROR_INVALIDDATA;
1521 if (FRAME_MBAFF(h)) {
1522 ff_h264_fill_mbaff_ref_list(h, sl);
1524 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1525 implicit_weight_table(h, sl, 0);
1526 implicit_weight_table(h, sl, 1);
1530 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1531 ff_h264_direct_dist_scale_factor(h, sl);
1532 ff_h264_direct_ref_list_init(h, sl);
1534 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1535 tmp = get_ue_golomb_31(&sl->gb);
1537 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1538 return AVERROR_INVALIDDATA;
1540 sl->cabac_init_idc = tmp;
1543 sl->last_qscale_diff = 0;
1544 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1545 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1546 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1547 return AVERROR_INVALIDDATA;
1550 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1551 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1552 // FIXME qscale / qp ... stuff
1553 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1554 get_bits1(&sl->gb); /* sp_for_switch_flag */
1555 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1556 sl->slice_type == AV_PICTURE_TYPE_SI)
1557 get_se_golomb(&sl->gb); /* slice_qs_delta */
1559 sl->deblocking_filter = 1;
1560 sl->slice_alpha_c0_offset = 0;
1561 sl->slice_beta_offset = 0;
1562 if (h->pps.deblocking_filter_parameters_present) {
1563 tmp = get_ue_golomb_31(&sl->gb);
1565 av_log(h->avctx, AV_LOG_ERROR,
1566 "deblocking_filter_idc %u out of range\n", tmp);
1567 return AVERROR_INVALIDDATA;
1569 sl->deblocking_filter = tmp;
1570 if (sl->deblocking_filter < 2)
1571 sl->deblocking_filter ^= 1; // 1<->0
1573 if (sl->deblocking_filter) {
1574 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1575 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1576 if (sl->slice_alpha_c0_offset > 12 ||
1577 sl->slice_alpha_c0_offset < -12 ||
1578 sl->slice_beta_offset > 12 ||
1579 sl->slice_beta_offset < -12) {
1580 av_log(h->avctx, AV_LOG_ERROR,
1581 "deblocking filter parameters %d %d out of range\n",
1582 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1583 return AVERROR_INVALIDDATA;
1588 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1589 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1590 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1591 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1592 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1593 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1594 h->nal_ref_idc == 0))
1595 sl->deblocking_filter = 0;
1597 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1598 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1599 /* Cheat slightly for speed:
1600 * Do not bother to deblock across slices. */
1601 sl->deblocking_filter = 2;
1603 h->max_contexts = 1;
1604 if (!h->single_decode_warning) {
1605 av_log(h->avctx, AV_LOG_INFO,
1606 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1607 h->single_decode_warning = 1;
1609 if (sl != h->slice_ctx) {
1610 av_log(h->avctx, AV_LOG_ERROR,
1611 "Deblocking switched inside frame.\n");
1616 sl->qp_thresh = 15 -
1617 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1619 h->pps.chroma_qp_index_offset[0],
1620 h->pps.chroma_qp_index_offset[1]) +
1621 6 * (h->sps.bit_depth_luma - 8);
1623 sl->slice_num = ++h->current_slice;
1624 if (sl->slice_num >= MAX_SLICES) {
1625 av_log(h->avctx, AV_LOG_ERROR,
1626 "Too many slices, increase MAX_SLICES and recompile\n");
1629 for (j = 0; j < 2; j++) {
1631 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1632 for (i = 0; i < 16; i++) {
1634 if (j < sl->list_count && i < sl->ref_count[j] &&
1635 sl->ref_list[j][i].parent->f->buf[0]) {
1637 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1638 for (k = 0; k < h->short_ref_count; k++)
1639 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1643 for (k = 0; k < h->long_ref_count; k++)
1644 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1645 id_list[i] = h->short_ref_count + k;
1653 for (i = 0; i < 16; i++)
1654 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1656 ref2frm[18 + 1] = -1;
1657 for (i = 16; i < 48; i++)
1658 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1659 (sl->ref_list[j][i].reference & 3);
1662 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1663 av_log(h->avctx, AV_LOG_DEBUG,
1664 "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",
1666 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1668 av_get_picture_type_char(sl->slice_type),
1669 sl->slice_type_fixed ? " fix" : "",
1670 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1671 pps_id, h->frame_num,
1672 h->cur_pic_ptr->field_poc[0],
1673 h->cur_pic_ptr->field_poc[1],
1674 sl->ref_count[0], sl->ref_count[1],
1676 sl->deblocking_filter,
1677 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1679 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
1680 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1686 int ff_h264_get_slice_type(const H264SliceContext *sl)
1688 switch (sl->slice_type) {
1689 case AV_PICTURE_TYPE_P:
1691 case AV_PICTURE_TYPE_B:
1693 case AV_PICTURE_TYPE_I:
1695 case AV_PICTURE_TYPE_SP:
1697 case AV_PICTURE_TYPE_SI:
1700 return AVERROR_INVALIDDATA;
1704 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1705 H264SliceContext *sl,
1706 int mb_type, int top_xy,
1707 int left_xy[LEFT_MBS],
1709 int left_type[LEFT_MBS],
1710 int mb_xy, int list)
1712 int b_stride = h->b_stride;
1713 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1714 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1715 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1716 if (USES_LIST(top_type, list)) {
1717 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1718 const int b8_xy = 4 * top_xy + 2;
1719 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1720 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1721 ref_cache[0 - 1 * 8] =
1722 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1723 ref_cache[2 - 1 * 8] =
1724 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1726 AV_ZERO128(mv_dst - 1 * 8);
1727 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1730 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1731 if (USES_LIST(left_type[LTOP], list)) {
1732 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1733 const int b8_xy = 4 * left_xy[LTOP] + 1;
1734 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1735 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1736 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1737 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1738 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1740 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1741 ref_cache[-1 + 16] =
1742 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1744 AV_ZERO32(mv_dst - 1 + 0);
1745 AV_ZERO32(mv_dst - 1 + 8);
1746 AV_ZERO32(mv_dst - 1 + 16);
1747 AV_ZERO32(mv_dst - 1 + 24);
1750 ref_cache[-1 + 16] =
1751 ref_cache[-1 + 24] = LIST_NOT_USED;
1756 if (!USES_LIST(mb_type, list)) {
1757 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1758 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1759 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1760 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1761 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1766 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1767 int (*ref2frm)[64] = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1768 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1769 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1770 AV_WN32A(&ref_cache[0 * 8], ref01);
1771 AV_WN32A(&ref_cache[1 * 8], ref01);
1772 AV_WN32A(&ref_cache[2 * 8], ref23);
1773 AV_WN32A(&ref_cache[3 * 8], ref23);
1777 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1778 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1779 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1780 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1781 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1787 * @return non zero if the loop filter can be skipped
1789 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
1791 const int mb_xy = sl->mb_xy;
1792 int top_xy, left_xy[LEFT_MBS];
1793 int top_type, left_type[LEFT_MBS];
1797 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1799 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1800 * stuff, I can't imagine that these complex rules are worth it. */
1802 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1803 if (FRAME_MBAFF(h)) {
1804 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1805 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1807 if (left_mb_field_flag != curr_mb_field_flag)
1808 left_xy[LTOP] -= h->mb_stride;
1810 if (curr_mb_field_flag)
1811 top_xy += h->mb_stride &
1812 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1813 if (left_mb_field_flag != curr_mb_field_flag)
1814 left_xy[LBOT] += h->mb_stride;
1818 sl->top_mb_xy = top_xy;
1819 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1820 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1822 /* For sufficiently low qp, filtering wouldn't do anything.
1823 * This is a conservative estimate: could also check beta_offset
1824 * and more accurate chroma_qp. */
1825 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1826 int qp = h->cur_pic.qscale_table[mb_xy];
1827 if (qp <= qp_thresh &&
1828 (left_xy[LTOP] < 0 ||
1829 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1831 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1832 if (!FRAME_MBAFF(h))
1834 if ((left_xy[LTOP] < 0 ||
1835 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1836 (top_xy < h->mb_stride ||
1837 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1842 top_type = h->cur_pic.mb_type[top_xy];
1843 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1844 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1845 if (sl->deblocking_filter == 2) {
1846 if (h->slice_table[top_xy] != sl->slice_num)
1848 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
1849 left_type[LTOP] = left_type[LBOT] = 0;
1851 if (h->slice_table[top_xy] == 0xFFFF)
1853 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1854 left_type[LTOP] = left_type[LBOT] = 0;
1856 sl->top_type = top_type;
1857 sl->left_type[LTOP] = left_type[LTOP];
1858 sl->left_type[LBOT] = left_type[LBOT];
1860 if (IS_INTRA(mb_type))
1863 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1864 top_type, left_type, mb_xy, 0);
1865 if (sl->list_count == 2)
1866 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1867 top_type, left_type, mb_xy, 1);
1869 nnz = h->non_zero_count[mb_xy];
1870 nnz_cache = sl->non_zero_count_cache;
1871 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1872 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1873 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1874 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1875 sl->cbp = h->cbp_table[mb_xy];
1878 nnz = h->non_zero_count[top_xy];
1879 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1882 if (left_type[LTOP]) {
1883 nnz = h->non_zero_count[left_xy[LTOP]];
1884 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
1885 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
1886 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
1887 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
1890 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
1891 * from what the loop filter needs */
1892 if (!CABAC(h) && h->pps.transform_8x8_mode) {
1893 if (IS_8x8DCT(top_type)) {
1894 nnz_cache[4 + 8 * 0] =
1895 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
1896 nnz_cache[6 + 8 * 0] =
1897 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
1899 if (IS_8x8DCT(left_type[LTOP])) {
1900 nnz_cache[3 + 8 * 1] =
1901 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
1903 if (IS_8x8DCT(left_type[LBOT])) {
1904 nnz_cache[3 + 8 * 3] =
1905 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
1908 if (IS_8x8DCT(mb_type)) {
1909 nnz_cache[scan8[0]] =
1910 nnz_cache[scan8[1]] =
1911 nnz_cache[scan8[2]] =
1912 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
1914 nnz_cache[scan8[0 + 4]] =
1915 nnz_cache[scan8[1 + 4]] =
1916 nnz_cache[scan8[2 + 4]] =
1917 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
1919 nnz_cache[scan8[0 + 8]] =
1920 nnz_cache[scan8[1 + 8]] =
1921 nnz_cache[scan8[2 + 8]] =
1922 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
1924 nnz_cache[scan8[0 + 12]] =
1925 nnz_cache[scan8[1 + 12]] =
1926 nnz_cache[scan8[2 + 12]] =
1927 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
1934 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
1936 uint8_t *dest_y, *dest_cb, *dest_cr;
1937 int linesize, uvlinesize, mb_x, mb_y;
1938 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
1939 const int old_slice_type = sl->slice_type;
1940 const int pixel_shift = h->pixel_shift;
1941 const int block_h = 16 >> h->chroma_y_shift;
1943 if (sl->deblocking_filter) {
1944 for (mb_x = start_x; mb_x < end_x; mb_x++)
1945 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
1947 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
1948 sl->slice_num = h->slice_table[mb_xy];
1949 mb_type = h->cur_pic.mb_type[mb_xy];
1950 sl->list_count = h->list_counts[mb_xy];
1954 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
1958 dest_y = h->cur_pic.f->data[0] +
1959 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
1960 dest_cb = h->cur_pic.f->data[1] +
1961 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1962 mb_y * sl->uvlinesize * block_h;
1963 dest_cr = h->cur_pic.f->data[2] +
1964 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1965 mb_y * sl->uvlinesize * block_h;
1966 // FIXME simplify above
1969 linesize = sl->mb_linesize = sl->linesize * 2;
1970 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
1971 if (mb_y & 1) { // FIXME move out of this function?
1972 dest_y -= sl->linesize * 15;
1973 dest_cb -= sl->uvlinesize * (block_h - 1);
1974 dest_cr -= sl->uvlinesize * (block_h - 1);
1977 linesize = sl->mb_linesize = sl->linesize;
1978 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
1980 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
1982 if (fill_filter_caches(h, sl, mb_type))
1984 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
1985 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
1987 if (FRAME_MBAFF(h)) {
1988 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
1989 linesize, uvlinesize);
1991 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
1992 dest_cr, linesize, uvlinesize);
1996 sl->slice_type = old_slice_type;
1998 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
1999 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2000 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2003 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2005 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2006 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2007 h->cur_pic.mb_type[mb_xy - 1] :
2008 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2009 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2010 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2014 * Draw edges and report progress for the last MB row.
2016 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2018 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2019 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2020 int height = 16 << FRAME_MBAFF(h);
2021 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2023 if (sl->deblocking_filter) {
2024 if ((top + height) >= pic_height)
2025 height += deblock_border;
2026 top -= deblock_border;
2029 if (top >= pic_height || (top + height) < 0)
2032 height = FFMIN(height, pic_height - top);
2034 height = top + height;
2038 ff_h264_draw_horiz_band(h, sl, top, height);
2043 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2044 h->picture_structure == PICT_BOTTOM_FIELD);
2047 static void er_add_slice(H264SliceContext *sl,
2048 int startx, int starty,
2049 int endx, int endy, int status)
2051 #if CONFIG_ERROR_RESILIENCE
2052 ERContext *er = &sl->er;
2054 if (!sl->h264->enable_er)
2057 er->ref_count = sl->ref_count[0];
2058 ff_er_add_slice(er, startx, starty, endx, endy, status);
2062 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2064 H264SliceContext *sl = arg;
2065 const H264Context *h = sl->h264;
2066 int lf_x_start = sl->mb_x;
2069 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2070 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2072 ret = alloc_scratch_buffers(sl, sl->linesize);
2076 sl->mb_skip_run = -1;
2078 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2079 avctx->codec_id != AV_CODEC_ID_H264 ||
2080 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2084 align_get_bits(&sl->gb);
2087 ff_init_cabac_decoder(&sl->cabac,
2088 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2089 (get_bits_left(&sl->gb) + 7) / 8);
2091 ff_h264_init_cabac_states(h, sl);
2097 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2098 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2099 sl->next_slice_idx);
2100 return AVERROR_INVALIDDATA;
2103 ret = ff_h264_decode_mb_cabac(h, sl);
2104 // STOP_TIMER("decode_mb_cabac")
2107 ff_h264_hl_decode_mb(h, sl);
2109 // FIXME optimal? or let mb_decode decode 16x32 ?
2110 if (ret >= 0 && FRAME_MBAFF(h)) {
2113 ret = ff_h264_decode_mb_cabac(h, sl);
2116 ff_h264_hl_decode_mb(h, sl);
2119 eos = get_cabac_terminate(&sl->cabac);
2121 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2122 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2123 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2124 sl->mb_y, ER_MB_END);
2125 if (sl->mb_x >= lf_x_start)
2126 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2129 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2130 av_log(h->avctx, AV_LOG_ERROR,
2131 "error while decoding MB %d %d, bytestream %td\n",
2133 sl->cabac.bytestream_end - sl->cabac.bytestream);
2134 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2135 sl->mb_y, ER_MB_ERROR);
2136 return AVERROR_INVALIDDATA;
2139 if (++sl->mb_x >= h->mb_width) {
2140 loop_filter(h, sl, lf_x_start, sl->mb_x);
2141 sl->mb_x = lf_x_start = 0;
2142 decode_finish_row(h, sl);
2144 if (FIELD_OR_MBAFF_PICTURE(h)) {
2146 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2147 predict_field_decoding_flag(h, sl);
2151 if (eos || sl->mb_y >= h->mb_height) {
2152 ff_tlog(h->avctx, "slice end %d %d\n",
2153 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2154 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2155 sl->mb_y, ER_MB_END);
2156 if (sl->mb_x > lf_x_start)
2157 loop_filter(h, sl, lf_x_start, sl->mb_x);
2165 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2166 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2167 sl->next_slice_idx);
2168 return AVERROR_INVALIDDATA;
2171 ret = ff_h264_decode_mb_cavlc(h, sl);
2174 ff_h264_hl_decode_mb(h, sl);
2176 // FIXME optimal? or let mb_decode decode 16x32 ?
2177 if (ret >= 0 && FRAME_MBAFF(h)) {
2179 ret = ff_h264_decode_mb_cavlc(h, sl);
2182 ff_h264_hl_decode_mb(h, sl);
2187 av_log(h->avctx, AV_LOG_ERROR,
2188 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2189 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2190 sl->mb_y, ER_MB_ERROR);
2194 if (++sl->mb_x >= h->mb_width) {
2195 loop_filter(h, sl, lf_x_start, sl->mb_x);
2196 sl->mb_x = lf_x_start = 0;
2197 decode_finish_row(h, sl);
2199 if (FIELD_OR_MBAFF_PICTURE(h)) {
2201 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2202 predict_field_decoding_flag(h, sl);
2204 if (sl->mb_y >= h->mb_height) {
2205 ff_tlog(h->avctx, "slice end %d %d\n",
2206 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2208 if (get_bits_left(&sl->gb) == 0) {
2209 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2210 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2214 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2215 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2217 return AVERROR_INVALIDDATA;
2222 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2223 ff_tlog(h->avctx, "slice end %d %d\n",
2224 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2226 if (get_bits_left(&sl->gb) == 0) {
2227 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2228 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2229 if (sl->mb_x > lf_x_start)
2230 loop_filter(h, sl, lf_x_start, sl->mb_x);
2234 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2235 sl->mb_y, ER_MB_ERROR);
2237 return AVERROR_INVALIDDATA;
2245 * Call decode_slice() for each context.
2247 * @param h h264 master context
2248 * @param context_count number of contexts to execute
2250 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2252 AVCodecContext *const avctx = h->avctx;
2253 H264SliceContext *sl;
2256 if (h->avctx->hwaccel)
2258 if (context_count == 1) {
2261 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2263 ret = decode_slice(avctx, &h->slice_ctx[0]);
2264 h->mb_y = h->slice_ctx[0].mb_y;
2267 for (i = 0; i < context_count; i++) {
2268 int next_slice_idx = h->mb_width * h->mb_height;
2271 sl = &h->slice_ctx[i];
2272 sl->er.error_count = 0;
2274 /* make sure none of those slices overlap */
2275 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2276 for (j = 0; j < context_count; j++) {
2277 H264SliceContext *sl2 = &h->slice_ctx[j];
2278 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2280 if (i == j || slice_idx2 < slice_idx)
2282 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2284 sl->next_slice_idx = next_slice_idx;
2287 avctx->execute(avctx, decode_slice, h->slice_ctx,
2288 NULL, context_count, sizeof(h->slice_ctx[0]));
2290 /* pull back stuff from slices to master context */
2291 sl = &h->slice_ctx[context_count - 1];
2293 for (i = 1; i < context_count; i++)
2294 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;