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"
35 #include "error_resilience.h"
39 #include "h264chroma.h"
40 #include "h264_mvpred.h"
43 #include "mpegutils.h"
44 #include "rectangle.h"
48 static const uint8_t rem6[QP_MAX_NUM + 1] = {
49 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
50 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
51 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
54 static const uint8_t div6[QP_MAX_NUM + 1] = {
55 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
56 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
57 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
60 static const uint8_t field_scan[16] = {
61 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
62 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
63 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
64 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
67 static const uint8_t field_scan8x8[64] = {
68 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
69 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
70 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
71 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
72 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
73 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
74 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
75 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
76 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
77 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
78 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
79 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
80 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
81 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
82 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
83 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
86 static const uint8_t field_scan8x8_cavlc[64] = {
87 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
88 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
89 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
90 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
91 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
92 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
93 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
94 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
95 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
96 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
97 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
98 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
99 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
100 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
101 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
102 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
105 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
106 static const uint8_t zigzag_scan8x8_cavlc[64] = {
107 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
108 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
109 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
110 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
111 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
112 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
113 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
114 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
115 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
116 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
117 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
118 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
119 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
120 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
121 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
122 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
125 static const uint8_t dequant4_coeff_init[6][3] = {
134 static const uint8_t dequant8_coeff_init_scan[16] = {
135 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
138 static const uint8_t dequant8_coeff_init[6][6] = {
139 { 20, 18, 32, 19, 25, 24 },
140 { 22, 19, 35, 21, 28, 26 },
141 { 26, 23, 42, 24, 33, 31 },
142 { 28, 25, 45, 26, 35, 33 },
143 { 32, 28, 51, 30, 40, 38 },
144 { 36, 32, 58, 34, 46, 43 },
147 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
148 #if CONFIG_H264_DXVA2_HWACCEL
149 AV_PIX_FMT_DXVA2_VLD,
151 #if CONFIG_H264_VAAPI_HWACCEL
152 AV_PIX_FMT_VAAPI_VLD,
154 #if CONFIG_H264_VDA_HWACCEL
158 #if CONFIG_H264_VDPAU_HWACCEL
165 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
166 #if CONFIG_H264_DXVA2_HWACCEL
167 AV_PIX_FMT_DXVA2_VLD,
169 #if CONFIG_H264_VAAPI_HWACCEL
170 AV_PIX_FMT_VAAPI_VLD,
172 #if CONFIG_H264_VDA_HWACCEL
176 #if CONFIG_H264_VDPAU_HWACCEL
184 static void release_unused_pictures(H264Context *h, int remove_current)
188 /* release non reference frames */
189 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
190 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
191 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
192 ff_h264_unref_picture(h, &h->DPB[i]);
197 static int alloc_scratch_buffers(H264Context *h, int linesize)
199 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
201 if (h->bipred_scratchpad)
204 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
205 // edge emu needs blocksize + filter length - 1
206 // (= 21x21 for h264)
207 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
209 if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
210 av_freep(&h->bipred_scratchpad);
211 av_freep(&h->edge_emu_buffer);
212 return AVERROR(ENOMEM);
218 static int init_table_pools(H264Context *h)
220 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
221 const int mb_array_size = h->mb_stride * h->mb_height;
222 const int b4_stride = h->mb_width * 4 + 1;
223 const int b4_array_size = b4_stride * h->mb_height * 4;
225 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
227 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
228 sizeof(uint32_t), av_buffer_allocz);
229 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
230 sizeof(int16_t), av_buffer_allocz);
231 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
233 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
234 !h->ref_index_pool) {
235 av_buffer_pool_uninit(&h->qscale_table_pool);
236 av_buffer_pool_uninit(&h->mb_type_pool);
237 av_buffer_pool_uninit(&h->motion_val_pool);
238 av_buffer_pool_uninit(&h->ref_index_pool);
239 return AVERROR(ENOMEM);
245 static int alloc_picture(H264Context *h, H264Picture *pic)
249 av_assert0(!pic->f.data[0]);
252 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
253 AV_GET_BUFFER_FLAG_REF : 0);
257 h->linesize = pic->f.linesize[0];
258 h->uvlinesize = pic->f.linesize[1];
260 if (h->avctx->hwaccel) {
261 const AVHWAccel *hwaccel = h->avctx->hwaccel;
262 av_assert0(!pic->hwaccel_picture_private);
263 if (hwaccel->frame_priv_data_size) {
264 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
265 if (!pic->hwaccel_priv_buf)
266 return AVERROR(ENOMEM);
267 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
271 if (!h->qscale_table_pool) {
272 ret = init_table_pools(h);
277 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
278 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
279 if (!pic->qscale_table_buf || !pic->mb_type_buf)
282 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
283 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
285 for (i = 0; i < 2; i++) {
286 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
287 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
288 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
291 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
292 pic->ref_index[i] = pic->ref_index_buf[i]->data;
297 ff_h264_unref_picture(h, pic);
298 return (ret < 0) ? ret : AVERROR(ENOMEM);
301 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
305 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
310 static int find_unused_picture(H264Context *h)
314 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
315 if (pic_is_unused(h, &h->DPB[i]))
318 if (i == H264_MAX_PICTURE_COUNT)
319 return AVERROR_INVALIDDATA;
321 if (h->DPB[i].needs_realloc) {
322 h->DPB[i].needs_realloc = 0;
323 ff_h264_unref_picture(h, &h->DPB[i]);
330 static void init_dequant8_coeff_table(H264Context *h)
333 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
335 for (i = 0; i < 6; i++) {
336 h->dequant8_coeff[i] = h->dequant8_buffer[i];
337 for (j = 0; j < i; j++)
338 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
339 64 * sizeof(uint8_t))) {
340 h->dequant8_coeff[i] = h->dequant8_buffer[j];
346 for (q = 0; q < max_qp + 1; q++) {
349 for (x = 0; x < 64; x++)
350 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
351 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
352 h->pps.scaling_matrix8[i][x]) << shift;
357 static void init_dequant4_coeff_table(H264Context *h)
360 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
361 for (i = 0; i < 6; i++) {
362 h->dequant4_coeff[i] = h->dequant4_buffer[i];
363 for (j = 0; j < i; j++)
364 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
365 16 * sizeof(uint8_t))) {
366 h->dequant4_coeff[i] = h->dequant4_buffer[j];
372 for (q = 0; q < max_qp + 1; q++) {
373 int shift = div6[q] + 2;
375 for (x = 0; x < 16; x++)
376 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
377 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
378 h->pps.scaling_matrix4[i][x]) << shift;
383 void h264_init_dequant_tables(H264Context *h)
386 init_dequant4_coeff_table(h);
387 if (h->pps.transform_8x8_mode)
388 init_dequant8_coeff_table(h);
389 if (h->sps.transform_bypass) {
390 for (i = 0; i < 6; i++)
391 for (x = 0; x < 16; x++)
392 h->dequant4_coeff[i][0][x] = 1 << 6;
393 if (h->pps.transform_8x8_mode)
394 for (i = 0; i < 6; i++)
395 for (x = 0; x < 64; x++)
396 h->dequant8_coeff[i][0][x] = 1 << 6;
401 * Mimic alloc_tables(), but for every context thread.
403 static void clone_tables(H264Context *dst, H264Context *src, int i)
405 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
406 dst->non_zero_count = src->non_zero_count;
407 dst->slice_table = src->slice_table;
408 dst->cbp_table = src->cbp_table;
409 dst->mb2b_xy = src->mb2b_xy;
410 dst->mb2br_xy = src->mb2br_xy;
411 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
412 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
413 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
414 dst->direct_table = src->direct_table;
415 dst->list_counts = src->list_counts;
417 dst->cur_pic_ptr = src->cur_pic_ptr;
418 dst->cur_pic = src->cur_pic;
419 dst->bipred_scratchpad = NULL;
420 dst->edge_emu_buffer = NULL;
421 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
422 src->sps.chroma_format_idc);
425 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
426 #undef REBASE_PICTURE
427 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
428 ((pic && pic >= old_ctx->DPB && \
429 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
430 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
432 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
433 H264Context *new_base,
434 H264Context *old_base)
438 for (i = 0; i < count; i++) {
439 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
440 IN_RANGE(from[i], old_base->DPB,
441 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
443 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
447 static int copy_parameter_set(void **to, void **from, int count, int size)
451 for (i = 0; i < count; i++) {
452 if (to[i] && !from[i]) {
454 } else if (from[i] && !to[i]) {
455 to[i] = av_malloc(size);
457 return AVERROR(ENOMEM);
461 memcpy(to[i], from[i], size);
467 #define copy_fields(to, from, start_field, end_field) \
468 memcpy(&to->start_field, &from->start_field, \
469 (char *)&to->end_field - (char *)&to->start_field)
471 static int h264_slice_header_init(H264Context *h, int reinit);
473 int ff_h264_update_thread_context(AVCodecContext *dst,
474 const AVCodecContext *src)
476 H264Context *h = dst->priv_data, *h1 = src->priv_data;
477 int inited = h->context_initialized, err = 0;
478 int context_reinitialized = 0;
481 if (dst == src || !h1->context_initialized)
485 (h->width != h1->width ||
486 h->height != h1->height ||
487 h->mb_width != h1->mb_width ||
488 h->mb_height != h1->mb_height ||
489 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
490 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
491 h->sps.colorspace != h1->sps.colorspace)) {
493 /* set bits_per_raw_sample to the previous value. the check for changed
494 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
495 * the current value */
496 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
498 av_freep(&h->bipred_scratchpad);
500 h->width = h1->width;
501 h->height = h1->height;
502 h->mb_height = h1->mb_height;
503 h->mb_width = h1->mb_width;
504 h->mb_num = h1->mb_num;
505 h->mb_stride = h1->mb_stride;
506 h->b_stride = h1->b_stride;
508 if ((err = h264_slice_header_init(h, 1)) < 0) {
509 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
512 context_reinitialized = 1;
514 /* update linesize on resize. The decoder doesn't
515 * necessarily call h264_frame_start in the new thread */
516 h->linesize = h1->linesize;
517 h->uvlinesize = h1->uvlinesize;
519 /* copy block_offset since frame_start may not be called */
520 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
524 for (i = 0; i < MAX_SPS_COUNT; i++)
525 av_freep(h->sps_buffers + i);
527 for (i = 0; i < MAX_PPS_COUNT; i++)
528 av_freep(h->pps_buffers + i);
530 memcpy(h, h1, sizeof(*h1));
531 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
532 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
533 memset(&h->er, 0, sizeof(h->er));
534 memset(&h->mb, 0, sizeof(h->mb));
535 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
536 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
537 h->context_initialized = 0;
539 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
540 av_frame_unref(&h->cur_pic.f);
541 h->cur_pic.tf.f = &h->cur_pic.f;
545 h->qscale_table_pool = NULL;
546 h->mb_type_pool = NULL;
547 h->ref_index_pool = NULL;
548 h->motion_val_pool = NULL;
550 ret = ff_h264_alloc_tables(h);
552 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
555 ret = ff_h264_context_init(h);
557 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
561 for (i = 0; i < 2; i++) {
562 h->rbsp_buffer[i] = NULL;
563 h->rbsp_buffer_size[i] = 0;
565 h->bipred_scratchpad = NULL;
566 h->edge_emu_buffer = NULL;
568 h->thread_context[0] = h;
570 h->context_initialized = 1;
573 h->avctx->coded_height = h1->avctx->coded_height;
574 h->avctx->coded_width = h1->avctx->coded_width;
575 h->avctx->width = h1->avctx->width;
576 h->avctx->height = h1->avctx->height;
577 h->coded_picture_number = h1->coded_picture_number;
578 h->first_field = h1->first_field;
579 h->picture_structure = h1->picture_structure;
580 h->qscale = h1->qscale;
581 h->droppable = h1->droppable;
582 h->low_delay = h1->low_delay;
584 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
585 ff_h264_unref_picture(h, &h->DPB[i]);
586 if (h1->DPB[i].f.buf[0] &&
587 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
591 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
592 ff_h264_unref_picture(h, &h->cur_pic);
593 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
596 h->workaround_bugs = h1->workaround_bugs;
597 h->low_delay = h1->low_delay;
598 h->droppable = h1->droppable;
600 /* frame_start may not be called for the next thread (if it's decoding
601 * a bottom field) so this has to be allocated here */
602 err = alloc_scratch_buffers(h, h1->linesize);
606 // extradata/NAL handling
607 h->is_avc = h1->is_avc;
610 if ((ret = copy_parameter_set((void **)h->sps_buffers,
611 (void **)h1->sps_buffers,
612 MAX_SPS_COUNT, sizeof(SPS))) < 0)
615 if ((ret = copy_parameter_set((void **)h->pps_buffers,
616 (void **)h1->pps_buffers,
617 MAX_PPS_COUNT, sizeof(PPS))) < 0)
621 // Dequantization matrices
622 // FIXME these are big - can they be only copied when PPS changes?
623 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
625 for (i = 0; i < 6; i++)
626 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
627 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
629 for (i = 0; i < 6; i++)
630 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
631 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
633 h->dequant_coeff_pps = h1->dequant_coeff_pps;
636 copy_fields(h, h1, poc_lsb, redundant_pic_count);
639 copy_fields(h, h1, short_ref, cabac_init_idc);
641 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
642 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
643 copy_picture_range(h->delayed_pic, h1->delayed_pic,
644 MAX_DELAYED_PIC_COUNT + 2, h, h1);
646 h->last_slice_type = h1->last_slice_type;
648 if (context_reinitialized)
649 ff_h264_set_parameter_from_sps(h);
655 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
656 h->prev_poc_msb = h->poc_msb;
657 h->prev_poc_lsb = h->poc_lsb;
659 h->prev_frame_num_offset = h->frame_num_offset;
660 h->prev_frame_num = h->frame_num;
661 h->outputed_poc = h->next_outputed_poc;
663 h->recovery_frame = h1->recovery_frame;
664 h->frame_recovered = h1->frame_recovered;
669 static int h264_frame_start(H264Context *h)
673 const int pixel_shift = h->pixel_shift;
675 release_unused_pictures(h, 1);
676 h->cur_pic_ptr = NULL;
678 i = find_unused_picture(h);
680 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
685 pic->reference = h->droppable ? 0 : h->picture_structure;
686 pic->f.coded_picture_number = h->coded_picture_number++;
687 pic->field_picture = h->picture_structure != PICT_FRAME;
689 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
691 * See decode_nal_units().
693 pic->f.key_frame = 0;
697 if ((ret = alloc_picture(h, pic)) < 0)
700 h->cur_pic_ptr = pic;
701 ff_h264_unref_picture(h, &h->cur_pic);
702 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
705 if (CONFIG_ERROR_RESILIENCE)
706 ff_er_frame_start(&h->er);
708 assert(h->linesize && h->uvlinesize);
710 for (i = 0; i < 16; i++) {
711 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
712 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
714 for (i = 0; i < 16; i++) {
715 h->block_offset[16 + i] =
716 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
717 h->block_offset[48 + 16 + i] =
718 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
721 /* can't be in alloc_tables because linesize isn't known there.
722 * FIXME: redo bipred weight to not require extra buffer? */
723 for (i = 0; i < h->slice_context_count; i++)
724 if (h->thread_context[i]) {
725 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
730 /* Some macroblocks can be accessed before they're available in case
731 * of lost slices, MBAFF or threading. */
732 memset(h->slice_table, -1,
733 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
735 /* We mark the current picture as non-reference after allocating it, so
736 * that if we break out due to an error it can be released automatically
737 * in the next ff_MPV_frame_start().
739 h->cur_pic_ptr->reference = 0;
741 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
743 h->next_output_pic = NULL;
745 assert(h->cur_pic_ptr->long_ref == 0);
750 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
751 uint8_t *src_cb, uint8_t *src_cr,
752 int linesize, int uvlinesize,
757 const int pixel_shift = h->pixel_shift;
758 int chroma444 = CHROMA444(h);
759 int chroma422 = CHROMA422(h);
762 src_cb -= uvlinesize;
763 src_cr -= uvlinesize;
765 if (!simple && FRAME_MBAFF(h)) {
768 top_border = h->top_borders[0][h->mb_x];
769 AV_COPY128(top_border, src_y + 15 * linesize);
771 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
772 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
775 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
776 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
777 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
778 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
780 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
781 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
783 } else if (chroma422) {
785 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
786 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
788 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
789 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
793 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
794 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
796 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
797 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
802 } else if (MB_MBAFF(h)) {
808 top_border = h->top_borders[top_idx][h->mb_x];
809 /* There are two lines saved, the line above the top macroblock
810 * of a pair, and the line above the bottom macroblock. */
811 AV_COPY128(top_border, src_y + 16 * linesize);
813 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
815 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
818 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
819 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
820 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
821 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
823 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
824 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
826 } else if (chroma422) {
828 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
829 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
831 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
832 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
836 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
837 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
839 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
840 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
847 * Initialize implicit_weight table.
848 * @param field 0/1 initialize the weight for interlaced MBAFF
849 * -1 initializes the rest
851 static void implicit_weight_table(H264Context *h, int field)
853 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
855 for (i = 0; i < 2; i++) {
856 h->luma_weight_flag[i] = 0;
857 h->chroma_weight_flag[i] = 0;
861 if (h->picture_structure == PICT_FRAME) {
862 cur_poc = h->cur_pic_ptr->poc;
864 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
866 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
867 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
869 h->use_weight_chroma = 0;
873 ref_count0 = h->ref_count[0];
874 ref_count1 = h->ref_count[1];
876 cur_poc = h->cur_pic_ptr->field_poc[field];
878 ref_count0 = 16 + 2 * h->ref_count[0];
879 ref_count1 = 16 + 2 * h->ref_count[1];
883 h->use_weight_chroma = 2;
884 h->luma_log2_weight_denom = 5;
885 h->chroma_log2_weight_denom = 5;
887 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
888 int poc0 = h->ref_list[0][ref0].poc;
889 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
891 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
892 int poc1 = h->ref_list[1][ref1].poc;
893 int td = av_clip(poc1 - poc0, -128, 127);
895 int tb = av_clip(cur_poc - poc0, -128, 127);
896 int tx = (16384 + (FFABS(td) >> 1)) / td;
897 int dist_scale_factor = (tb * tx + 32) >> 8;
898 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
899 w = 64 - dist_scale_factor;
903 h->implicit_weight[ref0][ref1][0] =
904 h->implicit_weight[ref0][ref1][1] = w;
906 h->implicit_weight[ref0][ref1][field] = w;
913 * initialize scan tables
915 static void init_scan_tables(H264Context *h)
918 for (i = 0; i < 16; i++) {
919 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
920 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
921 h->field_scan[i] = TRANSPOSE(field_scan[i]);
924 for (i = 0; i < 64; i++) {
925 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
926 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
927 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
928 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
929 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
932 if (h->sps.transform_bypass) { // FIXME same ugly
933 h->zigzag_scan_q0 = zigzag_scan;
934 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
935 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
936 h->field_scan_q0 = field_scan;
937 h->field_scan8x8_q0 = field_scan8x8;
938 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
940 h->zigzag_scan_q0 = h->zigzag_scan;
941 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
942 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
943 h->field_scan_q0 = h->field_scan;
944 h->field_scan8x8_q0 = h->field_scan8x8;
945 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
950 * Replicate H264 "master" context to thread contexts.
952 static int clone_slice(H264Context *dst, H264Context *src)
954 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
955 dst->cur_pic_ptr = src->cur_pic_ptr;
956 dst->cur_pic = src->cur_pic;
957 dst->linesize = src->linesize;
958 dst->uvlinesize = src->uvlinesize;
959 dst->first_field = src->first_field;
961 dst->prev_poc_msb = src->prev_poc_msb;
962 dst->prev_poc_lsb = src->prev_poc_lsb;
963 dst->prev_frame_num_offset = src->prev_frame_num_offset;
964 dst->prev_frame_num = src->prev_frame_num;
965 dst->short_ref_count = src->short_ref_count;
967 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
968 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
969 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
971 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
972 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
977 static enum AVPixelFormat get_pixel_format(H264Context *h)
979 switch (h->sps.bit_depth_luma) {
982 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
983 return AV_PIX_FMT_GBRP9;
985 return AV_PIX_FMT_YUV444P9;
986 } else if (CHROMA422(h))
987 return AV_PIX_FMT_YUV422P9;
989 return AV_PIX_FMT_YUV420P9;
993 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
994 return AV_PIX_FMT_GBRP10;
996 return AV_PIX_FMT_YUV444P10;
997 } else if (CHROMA422(h))
998 return AV_PIX_FMT_YUV422P10;
1000 return AV_PIX_FMT_YUV420P10;
1004 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1005 return AV_PIX_FMT_GBRP;
1007 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
1008 : AV_PIX_FMT_YUV444P;
1009 } else if (CHROMA422(h)) {
1010 return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
1011 : AV_PIX_FMT_YUV422P;
1013 return ff_get_format(h->avctx, h->avctx->codec->pix_fmts ?
1014 h->avctx->codec->pix_fmts :
1015 h->avctx->color_range == AVCOL_RANGE_JPEG ?
1016 h264_hwaccel_pixfmt_list_jpeg_420 :
1017 h264_hwaccel_pixfmt_list_420);
1021 av_log(h->avctx, AV_LOG_ERROR,
1022 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1023 return AVERROR_INVALIDDATA;
1027 /* export coded and cropped frame dimensions to AVCodecContext */
1028 static int init_dimensions(H264Context *h)
1030 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1031 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1033 /* handle container cropping */
1035 FFALIGN(h->avctx->width, 16) == h->width &&
1036 FFALIGN(h->avctx->height, 16) == h->height) {
1037 width = h->avctx->width;
1038 height = h->avctx->height;
1041 if (width <= 0 || height <= 0) {
1042 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1044 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1045 return AVERROR_INVALIDDATA;
1047 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1048 h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
1055 h->avctx->coded_width = h->width;
1056 h->avctx->coded_height = h->height;
1057 h->avctx->width = width;
1058 h->avctx->height = height;
1063 static int h264_slice_header_init(H264Context *h, int reinit)
1065 int nb_slices = (HAVE_THREADS &&
1066 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1067 h->avctx->thread_count : 1;
1070 ff_set_sar(h->avctx, h->sps.sar);
1071 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1072 &h->chroma_x_shift, &h->chroma_y_shift);
1074 if (h->sps.timing_info_present_flag) {
1075 int64_t den = h->sps.time_scale;
1076 if (h->x264_build < 44U)
1078 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
1079 h->sps.num_units_in_tick, den, 1 << 30);
1083 ff_h264_free_tables(h, 0);
1085 h->prev_interlaced_frame = 1;
1087 init_scan_tables(h);
1088 ret = ff_h264_alloc_tables(h);
1090 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1094 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1097 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1099 max_slices = H264_MAX_THREADS;
1100 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1101 " reducing to %d\n", nb_slices, max_slices);
1102 nb_slices = max_slices;
1104 h->slice_context_count = nb_slices;
1106 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1107 ret = ff_h264_context_init(h);
1109 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1113 for (i = 1; i < h->slice_context_count; i++) {
1115 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1117 return AVERROR(ENOMEM);
1118 c->avctx = h->avctx;
1121 c->h264dsp = h->h264dsp;
1122 c->h264qpel = h->h264qpel;
1123 c->h264chroma = h->h264chroma;
1126 c->pixel_shift = h->pixel_shift;
1127 c->width = h->width;
1128 c->height = h->height;
1129 c->linesize = h->linesize;
1130 c->uvlinesize = h->uvlinesize;
1131 c->chroma_x_shift = h->chroma_x_shift;
1132 c->chroma_y_shift = h->chroma_y_shift;
1133 c->qscale = h->qscale;
1134 c->droppable = h->droppable;
1135 c->data_partitioning = h->data_partitioning;
1136 c->low_delay = h->low_delay;
1137 c->mb_width = h->mb_width;
1138 c->mb_height = h->mb_height;
1139 c->mb_stride = h->mb_stride;
1140 c->mb_num = h->mb_num;
1141 c->flags = h->flags;
1142 c->workaround_bugs = h->workaround_bugs;
1143 c->pict_type = h->pict_type;
1145 init_scan_tables(c);
1146 clone_tables(c, h, i);
1147 c->context_initialized = 1;
1150 for (i = 0; i < h->slice_context_count; i++)
1151 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1152 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1157 h->context_initialized = 1;
1163 * Decode a slice header.
1164 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1166 * @param h h264context
1167 * @param h0 h264 master context (differs from 'h' when doing sliced based
1168 * parallel decoding)
1170 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1172 int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1174 unsigned int first_mb_in_slice;
1175 unsigned int pps_id;
1177 unsigned int slice_type, tmp, i, j;
1178 int default_ref_list_done = 0;
1179 int last_pic_structure, last_pic_droppable;
1180 int needs_reinit = 0;
1181 int field_pic_flag, bottom_field_flag;
1183 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1184 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1186 first_mb_in_slice = get_ue_golomb(&h->gb);
1188 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1189 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1190 ff_h264_field_end(h, 1);
1193 h0->current_slice = 0;
1194 if (!h0->first_field) {
1195 if (h->cur_pic_ptr && !h->droppable) {
1196 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1197 h->picture_structure == PICT_BOTTOM_FIELD);
1199 h->cur_pic_ptr = NULL;
1203 slice_type = get_ue_golomb_31(&h->gb);
1204 if (slice_type > 9) {
1205 av_log(h->avctx, AV_LOG_ERROR,
1206 "slice type %d too large at %d %d\n",
1207 slice_type, h->mb_x, h->mb_y);
1208 return AVERROR_INVALIDDATA;
1210 if (slice_type > 4) {
1212 h->slice_type_fixed = 1;
1214 h->slice_type_fixed = 0;
1216 slice_type = golomb_to_pict_type[slice_type];
1217 if (slice_type == AV_PICTURE_TYPE_I ||
1218 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
1219 default_ref_list_done = 1;
1221 h->slice_type = slice_type;
1222 h->slice_type_nos = slice_type & 3;
1224 if (h->nal_unit_type == NAL_IDR_SLICE &&
1225 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1226 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1227 return AVERROR_INVALIDDATA;
1230 // to make a few old functions happy, it's wrong though
1231 h->pict_type = h->slice_type;
1233 pps_id = get_ue_golomb(&h->gb);
1234 if (pps_id >= MAX_PPS_COUNT) {
1235 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1236 return AVERROR_INVALIDDATA;
1238 if (!h0->pps_buffers[pps_id]) {
1239 av_log(h->avctx, AV_LOG_ERROR,
1240 "non-existing PPS %u referenced\n",
1242 return AVERROR_INVALIDDATA;
1244 h->pps = *h0->pps_buffers[pps_id];
1246 if (!h0->sps_buffers[h->pps.sps_id]) {
1247 av_log(h->avctx, AV_LOG_ERROR,
1248 "non-existing SPS %u referenced\n",
1250 return AVERROR_INVALIDDATA;
1253 if (h->pps.sps_id != h->sps.sps_id ||
1254 h0->sps_buffers[h->pps.sps_id]->new) {
1255 h0->sps_buffers[h->pps.sps_id]->new = 0;
1257 h->sps = *h0->sps_buffers[h->pps.sps_id];
1259 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1260 h->chroma_format_idc != h->sps.chroma_format_idc) {
1261 h->bit_depth_luma = h->sps.bit_depth_luma;
1262 h->chroma_format_idc = h->sps.chroma_format_idc;
1265 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1269 h->avctx->profile = ff_h264_get_profile(&h->sps);
1270 h->avctx->level = h->sps.level_idc;
1271 h->avctx->refs = h->sps.ref_frame_count;
1273 if (h->mb_width != h->sps.mb_width ||
1274 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1277 h->mb_width = h->sps.mb_width;
1278 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1279 h->mb_num = h->mb_width * h->mb_height;
1280 h->mb_stride = h->mb_width + 1;
1282 h->b_stride = h->mb_width * 4;
1284 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1286 h->width = 16 * h->mb_width;
1287 h->height = 16 * h->mb_height;
1289 ret = init_dimensions(h);
1293 if (h->sps.video_signal_type_present_flag) {
1294 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1296 if (h->sps.colour_description_present_flag) {
1297 if (h->avctx->colorspace != h->sps.colorspace)
1299 h->avctx->color_primaries = h->sps.color_primaries;
1300 h->avctx->color_trc = h->sps.color_trc;
1301 h->avctx->colorspace = h->sps.colorspace;
1305 if (h->context_initialized && needs_reinit) {
1307 av_log(h->avctx, AV_LOG_ERROR,
1308 "changing width %d -> %d / height %d -> %d on "
1310 h->width, h->avctx->coded_width,
1311 h->height, h->avctx->coded_height,
1312 h0->current_slice + 1);
1313 return AVERROR_INVALIDDATA;
1316 ff_h264_flush_change(h);
1318 if ((ret = get_pixel_format(h)) < 0)
1320 h->avctx->pix_fmt = ret;
1322 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1323 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1325 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1326 av_log(h->avctx, AV_LOG_ERROR,
1327 "h264_slice_header_init() failed\n");
1331 if (!h->context_initialized) {
1333 av_log(h->avctx, AV_LOG_ERROR,
1334 "Cannot (re-)initialize context during parallel decoding.\n");
1335 return AVERROR_PATCHWELCOME;
1338 if ((ret = get_pixel_format(h)) < 0)
1340 h->avctx->pix_fmt = ret;
1342 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1343 av_log(h->avctx, AV_LOG_ERROR,
1344 "h264_slice_header_init() failed\n");
1349 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1350 h->dequant_coeff_pps = pps_id;
1351 h264_init_dequant_tables(h);
1354 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1357 h->mb_aff_frame = 0;
1358 last_pic_structure = h0->picture_structure;
1359 last_pic_droppable = h0->droppable;
1360 h->droppable = h->nal_ref_idc == 0;
1361 if (h->sps.frame_mbs_only_flag) {
1362 h->picture_structure = PICT_FRAME;
1364 field_pic_flag = get_bits1(&h->gb);
1365 if (field_pic_flag) {
1366 bottom_field_flag = get_bits1(&h->gb);
1367 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1369 h->picture_structure = PICT_FRAME;
1370 h->mb_aff_frame = h->sps.mb_aff;
1373 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1375 if (h0->current_slice != 0) {
1376 if (last_pic_structure != h->picture_structure ||
1377 last_pic_droppable != h->droppable) {
1378 av_log(h->avctx, AV_LOG_ERROR,
1379 "Changing field mode (%d -> %d) between slices is not allowed\n",
1380 last_pic_structure, h->picture_structure);
1381 h->picture_structure = last_pic_structure;
1382 h->droppable = last_pic_droppable;
1383 return AVERROR_INVALIDDATA;
1384 } else if (!h0->cur_pic_ptr) {
1385 av_log(h->avctx, AV_LOG_ERROR,
1386 "unset cur_pic_ptr on slice %d\n",
1387 h0->current_slice + 1);
1388 return AVERROR_INVALIDDATA;
1391 /* Shorten frame num gaps so we don't have to allocate reference
1392 * frames just to throw them away */
1393 if (h->frame_num != h->prev_frame_num) {
1394 int unwrap_prev_frame_num = h->prev_frame_num;
1395 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1397 if (unwrap_prev_frame_num > h->frame_num)
1398 unwrap_prev_frame_num -= max_frame_num;
1400 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1401 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1402 if (unwrap_prev_frame_num < 0)
1403 unwrap_prev_frame_num += max_frame_num;
1405 h->prev_frame_num = unwrap_prev_frame_num;
1409 /* See if we have a decoded first field looking for a pair...
1410 * Here, we're using that to see if we should mark previously
1411 * decode frames as "finished".
1412 * We have to do that before the "dummy" in-between frame allocation,
1413 * since that can modify s->current_picture_ptr. */
1414 if (h0->first_field) {
1415 assert(h0->cur_pic_ptr);
1416 assert(h0->cur_pic_ptr->f.buf[0]);
1417 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1419 /* figure out if we have a complementary field pair */
1420 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1421 /* Previous field is unmatched. Don't display it, but let it
1422 * remain for reference if marked as such. */
1423 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1424 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1425 last_pic_structure == PICT_TOP_FIELD);
1428 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1429 /* This and previous field were reference, but had
1430 * different frame_nums. Consider this field first in
1431 * pair. Throw away previous field except for reference
1433 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1434 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1435 last_pic_structure == PICT_TOP_FIELD);
1438 /* Second field in complementary pair */
1439 if (!((last_pic_structure == PICT_TOP_FIELD &&
1440 h->picture_structure == PICT_BOTTOM_FIELD) ||
1441 (last_pic_structure == PICT_BOTTOM_FIELD &&
1442 h->picture_structure == PICT_TOP_FIELD))) {
1443 av_log(h->avctx, AV_LOG_ERROR,
1444 "Invalid field mode combination %d/%d\n",
1445 last_pic_structure, h->picture_structure);
1446 h->picture_structure = last_pic_structure;
1447 h->droppable = last_pic_droppable;
1448 return AVERROR_INVALIDDATA;
1449 } else if (last_pic_droppable != h->droppable) {
1450 avpriv_request_sample(h->avctx,
1451 "Found reference and non-reference fields in the same frame, which");
1452 h->picture_structure = last_pic_structure;
1453 h->droppable = last_pic_droppable;
1454 return AVERROR_PATCHWELCOME;
1460 while (h->frame_num != h->prev_frame_num &&
1461 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1462 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1463 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1464 h->frame_num, h->prev_frame_num);
1465 ret = h264_frame_start(h);
1467 h0->first_field = 0;
1471 h->prev_frame_num++;
1472 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1473 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1474 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1475 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1476 ret = ff_generate_sliding_window_mmcos(h, 1);
1477 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1479 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1480 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1482 /* Error concealment: If a ref is missing, copy the previous ref
1484 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1485 * many assumptions about there being no actual duplicates.
1486 * FIXME: This does not copy padding for out-of-frame motion
1487 * vectors. Given we are concealing a lost frame, this probably
1488 * is not noticeable by comparison, but it should be fixed. */
1489 if (h->short_ref_count) {
1491 av_image_copy(h->short_ref[0]->f.data,
1492 h->short_ref[0]->f.linesize,
1493 (const uint8_t **)prev->f.data,
1498 h->short_ref[0]->poc = prev->poc + 2;
1500 h->short_ref[0]->frame_num = h->prev_frame_num;
1504 /* See if we have a decoded first field looking for a pair...
1505 * We're using that to see whether to continue decoding in that
1506 * frame, or to allocate a new one. */
1507 if (h0->first_field) {
1508 assert(h0->cur_pic_ptr);
1509 assert(h0->cur_pic_ptr->f.buf[0]);
1510 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1512 /* figure out if we have a complementary field pair */
1513 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1514 /* Previous field is unmatched. Don't display it, but let it
1515 * remain for reference if marked as such. */
1516 h0->cur_pic_ptr = NULL;
1517 h0->first_field = FIELD_PICTURE(h);
1519 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1520 /* This and the previous field had different frame_nums.
1521 * Consider this field first in pair. Throw away previous
1522 * one except for reference purposes. */
1523 h0->first_field = 1;
1524 h0->cur_pic_ptr = NULL;
1526 /* Second field in complementary pair */
1527 h0->first_field = 0;
1531 /* Frame or first field in a potentially complementary pair */
1532 h0->first_field = FIELD_PICTURE(h);
1535 if (!FIELD_PICTURE(h) || h0->first_field) {
1536 if (h264_frame_start(h) < 0) {
1537 h0->first_field = 0;
1538 return AVERROR_INVALIDDATA;
1541 release_unused_pictures(h, 0);
1544 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1547 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1549 assert(h->mb_num == h->mb_width * h->mb_height);
1550 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1551 first_mb_in_slice >= h->mb_num) {
1552 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1553 return AVERROR_INVALIDDATA;
1555 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1556 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1557 FIELD_OR_MBAFF_PICTURE(h);
1558 if (h->picture_structure == PICT_BOTTOM_FIELD)
1559 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1560 assert(h->mb_y < h->mb_height);
1562 if (h->picture_structure == PICT_FRAME) {
1563 h->curr_pic_num = h->frame_num;
1564 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1566 h->curr_pic_num = 2 * h->frame_num + 1;
1567 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1570 if (h->nal_unit_type == NAL_IDR_SLICE)
1571 get_ue_golomb(&h->gb); /* idr_pic_id */
1573 if (h->sps.poc_type == 0) {
1574 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1576 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1577 h->delta_poc_bottom = get_se_golomb(&h->gb);
1580 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1581 h->delta_poc[0] = get_se_golomb(&h->gb);
1583 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1584 h->delta_poc[1] = get_se_golomb(&h->gb);
1587 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1589 if (h->pps.redundant_pic_cnt_present)
1590 h->redundant_pic_count = get_ue_golomb(&h->gb);
1592 ret = ff_set_ref_count(h);
1596 default_ref_list_done = 0;
1598 if (!default_ref_list_done)
1599 ff_h264_fill_default_ref_list(h);
1601 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1602 ret = ff_h264_decode_ref_pic_list_reordering(h);
1604 h->ref_count[1] = h->ref_count[0] = 0;
1609 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1610 (h->pps.weighted_bipred_idc == 1 &&
1611 h->slice_type_nos == AV_PICTURE_TYPE_B))
1612 ff_pred_weight_table(h);
1613 else if (h->pps.weighted_bipred_idc == 2 &&
1614 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1615 implicit_weight_table(h, -1);
1618 for (i = 0; i < 2; i++) {
1619 h->luma_weight_flag[i] = 0;
1620 h->chroma_weight_flag[i] = 0;
1624 // If frame-mt is enabled, only update mmco tables for the first slice
1625 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1626 // or h->mmco, which will cause ref list mix-ups and decoding errors
1627 // further down the line. This may break decoding if the first slice is
1628 // corrupt, thus we only do this if frame-mt is enabled.
1629 if (h->nal_ref_idc) {
1630 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1631 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1632 h0->current_slice == 0);
1633 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1634 return AVERROR_INVALIDDATA;
1637 if (FRAME_MBAFF(h)) {
1638 ff_h264_fill_mbaff_ref_list(h);
1640 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1641 implicit_weight_table(h, 0);
1642 implicit_weight_table(h, 1);
1646 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1647 ff_h264_direct_dist_scale_factor(h);
1648 ff_h264_direct_ref_list_init(h);
1650 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1651 tmp = get_ue_golomb_31(&h->gb);
1653 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1654 return AVERROR_INVALIDDATA;
1656 h->cabac_init_idc = tmp;
1659 h->last_qscale_diff = 0;
1660 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1661 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1662 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1663 return AVERROR_INVALIDDATA;
1666 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1667 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1668 // FIXME qscale / qp ... stuff
1669 if (h->slice_type == AV_PICTURE_TYPE_SP)
1670 get_bits1(&h->gb); /* sp_for_switch_flag */
1671 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1672 h->slice_type == AV_PICTURE_TYPE_SI)
1673 get_se_golomb(&h->gb); /* slice_qs_delta */
1675 h->deblocking_filter = 1;
1676 h->slice_alpha_c0_offset = 0;
1677 h->slice_beta_offset = 0;
1678 if (h->pps.deblocking_filter_parameters_present) {
1679 tmp = get_ue_golomb_31(&h->gb);
1681 av_log(h->avctx, AV_LOG_ERROR,
1682 "deblocking_filter_idc %u out of range\n", tmp);
1683 return AVERROR_INVALIDDATA;
1685 h->deblocking_filter = tmp;
1686 if (h->deblocking_filter < 2)
1687 h->deblocking_filter ^= 1; // 1<->0
1689 if (h->deblocking_filter) {
1690 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1691 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1692 if (h->slice_alpha_c0_offset > 12 ||
1693 h->slice_alpha_c0_offset < -12 ||
1694 h->slice_beta_offset > 12 ||
1695 h->slice_beta_offset < -12) {
1696 av_log(h->avctx, AV_LOG_ERROR,
1697 "deblocking filter parameters %d %d out of range\n",
1698 h->slice_alpha_c0_offset, h->slice_beta_offset);
1699 return AVERROR_INVALIDDATA;
1704 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1705 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1706 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1707 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1708 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1709 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1710 h->nal_ref_idc == 0))
1711 h->deblocking_filter = 0;
1713 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1714 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1715 /* Cheat slightly for speed:
1716 * Do not bother to deblock across slices. */
1717 h->deblocking_filter = 2;
1719 h0->max_contexts = 1;
1720 if (!h0->single_decode_warning) {
1721 av_log(h->avctx, AV_LOG_INFO,
1722 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1723 h0->single_decode_warning = 1;
1726 av_log(h->avctx, AV_LOG_ERROR,
1727 "Deblocking switched inside frame.\n");
1733 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1735 h->pps.chroma_qp_index_offset[0],
1736 h->pps.chroma_qp_index_offset[1]) +
1737 6 * (h->sps.bit_depth_luma - 8);
1739 h0->last_slice_type = slice_type;
1740 h->slice_num = ++h0->current_slice;
1741 if (h->slice_num >= MAX_SLICES) {
1742 av_log(h->avctx, AV_LOG_ERROR,
1743 "Too many slices, increase MAX_SLICES and recompile\n");
1746 for (j = 0; j < 2; j++) {
1748 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1749 for (i = 0; i < 16; i++) {
1751 if (j < h->list_count && i < h->ref_count[j] &&
1752 h->ref_list[j][i].f.buf[0]) {
1754 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1755 for (k = 0; k < h->short_ref_count; k++)
1756 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1760 for (k = 0; k < h->long_ref_count; k++)
1761 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1762 id_list[i] = h->short_ref_count + k;
1770 for (i = 0; i < 16; i++)
1771 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1773 ref2frm[18 + 1] = -1;
1774 for (i = 16; i < 48; i++)
1775 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1776 (h->ref_list[j][i].reference & 3);
1779 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1780 av_log(h->avctx, AV_LOG_DEBUG,
1781 "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",
1783 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1785 av_get_picture_type_char(h->slice_type),
1786 h->slice_type_fixed ? " fix" : "",
1787 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1788 pps_id, h->frame_num,
1789 h->cur_pic_ptr->field_poc[0],
1790 h->cur_pic_ptr->field_poc[1],
1791 h->ref_count[0], h->ref_count[1],
1793 h->deblocking_filter,
1794 h->slice_alpha_c0_offset, h->slice_beta_offset,
1796 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
1797 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1803 int ff_h264_get_slice_type(const H264Context *h)
1805 switch (h->slice_type) {
1806 case AV_PICTURE_TYPE_P:
1808 case AV_PICTURE_TYPE_B:
1810 case AV_PICTURE_TYPE_I:
1812 case AV_PICTURE_TYPE_SP:
1814 case AV_PICTURE_TYPE_SI:
1817 return AVERROR_INVALIDDATA;
1821 static av_always_inline void fill_filter_caches_inter(H264Context *h,
1822 int mb_type, int top_xy,
1823 int left_xy[LEFT_MBS],
1825 int left_type[LEFT_MBS],
1826 int mb_xy, int list)
1828 int b_stride = h->b_stride;
1829 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
1830 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
1831 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1832 if (USES_LIST(top_type, list)) {
1833 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1834 const int b8_xy = 4 * top_xy + 2;
1835 int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1836 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1837 ref_cache[0 - 1 * 8] =
1838 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1839 ref_cache[2 - 1 * 8] =
1840 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1842 AV_ZERO128(mv_dst - 1 * 8);
1843 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1846 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1847 if (USES_LIST(left_type[LTOP], list)) {
1848 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1849 const int b8_xy = 4 * left_xy[LTOP] + 1;
1850 int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1851 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1852 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1853 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1854 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1856 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1857 ref_cache[-1 + 16] =
1858 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1860 AV_ZERO32(mv_dst - 1 + 0);
1861 AV_ZERO32(mv_dst - 1 + 8);
1862 AV_ZERO32(mv_dst - 1 + 16);
1863 AV_ZERO32(mv_dst - 1 + 24);
1866 ref_cache[-1 + 16] =
1867 ref_cache[-1 + 24] = LIST_NOT_USED;
1872 if (!USES_LIST(mb_type, list)) {
1873 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1874 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1875 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1876 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1877 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1882 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1883 int (*ref2frm)[64] = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1884 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1885 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1886 AV_WN32A(&ref_cache[0 * 8], ref01);
1887 AV_WN32A(&ref_cache[1 * 8], ref01);
1888 AV_WN32A(&ref_cache[2 * 8], ref23);
1889 AV_WN32A(&ref_cache[3 * 8], ref23);
1893 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
1894 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1895 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1896 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1897 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1903 * @return non zero if the loop filter can be skipped
1905 static int fill_filter_caches(H264Context *h, int mb_type)
1907 const int mb_xy = h->mb_xy;
1908 int top_xy, left_xy[LEFT_MBS];
1909 int top_type, left_type[LEFT_MBS];
1913 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
1915 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1916 * stuff, I can't imagine that these complex rules are worth it. */
1918 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1919 if (FRAME_MBAFF(h)) {
1920 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1921 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1923 if (left_mb_field_flag != curr_mb_field_flag)
1924 left_xy[LTOP] -= h->mb_stride;
1926 if (curr_mb_field_flag)
1927 top_xy += h->mb_stride &
1928 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1929 if (left_mb_field_flag != curr_mb_field_flag)
1930 left_xy[LBOT] += h->mb_stride;
1934 h->top_mb_xy = top_xy;
1935 h->left_mb_xy[LTOP] = left_xy[LTOP];
1936 h->left_mb_xy[LBOT] = left_xy[LBOT];
1938 /* For sufficiently low qp, filtering wouldn't do anything.
1939 * This is a conservative estimate: could also check beta_offset
1940 * and more accurate chroma_qp. */
1941 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1942 int qp = h->cur_pic.qscale_table[mb_xy];
1943 if (qp <= qp_thresh &&
1944 (left_xy[LTOP] < 0 ||
1945 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1947 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1948 if (!FRAME_MBAFF(h))
1950 if ((left_xy[LTOP] < 0 ||
1951 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1952 (top_xy < h->mb_stride ||
1953 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1958 top_type = h->cur_pic.mb_type[top_xy];
1959 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1960 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1961 if (h->deblocking_filter == 2) {
1962 if (h->slice_table[top_xy] != h->slice_num)
1964 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
1965 left_type[LTOP] = left_type[LBOT] = 0;
1967 if (h->slice_table[top_xy] == 0xFFFF)
1969 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1970 left_type[LTOP] = left_type[LBOT] = 0;
1972 h->top_type = top_type;
1973 h->left_type[LTOP] = left_type[LTOP];
1974 h->left_type[LBOT] = left_type[LBOT];
1976 if (IS_INTRA(mb_type))
1979 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1980 top_type, left_type, mb_xy, 0);
1981 if (h->list_count == 2)
1982 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1983 top_type, left_type, mb_xy, 1);
1985 nnz = h->non_zero_count[mb_xy];
1986 nnz_cache = h->non_zero_count_cache;
1987 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1988 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1989 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1990 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1991 h->cbp = h->cbp_table[mb_xy];
1994 nnz = h->non_zero_count[top_xy];
1995 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1998 if (left_type[LTOP]) {
1999 nnz = h->non_zero_count[left_xy[LTOP]];
2000 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2001 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2002 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2003 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2006 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2007 * from what the loop filter needs */
2008 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2009 if (IS_8x8DCT(top_type)) {
2010 nnz_cache[4 + 8 * 0] =
2011 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2012 nnz_cache[6 + 8 * 0] =
2013 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2015 if (IS_8x8DCT(left_type[LTOP])) {
2016 nnz_cache[3 + 8 * 1] =
2017 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2019 if (IS_8x8DCT(left_type[LBOT])) {
2020 nnz_cache[3 + 8 * 3] =
2021 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2024 if (IS_8x8DCT(mb_type)) {
2025 nnz_cache[scan8[0]] =
2026 nnz_cache[scan8[1]] =
2027 nnz_cache[scan8[2]] =
2028 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2030 nnz_cache[scan8[0 + 4]] =
2031 nnz_cache[scan8[1 + 4]] =
2032 nnz_cache[scan8[2 + 4]] =
2033 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2035 nnz_cache[scan8[0 + 8]] =
2036 nnz_cache[scan8[1 + 8]] =
2037 nnz_cache[scan8[2 + 8]] =
2038 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2040 nnz_cache[scan8[0 + 12]] =
2041 nnz_cache[scan8[1 + 12]] =
2042 nnz_cache[scan8[2 + 12]] =
2043 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2050 static void loop_filter(H264Context *h, int start_x, int end_x)
2052 uint8_t *dest_y, *dest_cb, *dest_cr;
2053 int linesize, uvlinesize, mb_x, mb_y;
2054 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2055 const int old_slice_type = h->slice_type;
2056 const int pixel_shift = h->pixel_shift;
2057 const int block_h = 16 >> h->chroma_y_shift;
2059 if (h->deblocking_filter) {
2060 for (mb_x = start_x; mb_x < end_x; mb_x++)
2061 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2063 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2064 h->slice_num = h->slice_table[mb_xy];
2065 mb_type = h->cur_pic.mb_type[mb_xy];
2066 h->list_count = h->list_counts[mb_xy];
2070 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2074 dest_y = h->cur_pic.f.data[0] +
2075 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2076 dest_cb = h->cur_pic.f.data[1] +
2077 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2078 mb_y * h->uvlinesize * block_h;
2079 dest_cr = h->cur_pic.f.data[2] +
2080 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2081 mb_y * h->uvlinesize * block_h;
2082 // FIXME simplify above
2085 linesize = h->mb_linesize = h->linesize * 2;
2086 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2087 if (mb_y & 1) { // FIXME move out of this function?
2088 dest_y -= h->linesize * 15;
2089 dest_cb -= h->uvlinesize * (block_h - 1);
2090 dest_cr -= h->uvlinesize * (block_h - 1);
2093 linesize = h->mb_linesize = h->linesize;
2094 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2096 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2098 if (fill_filter_caches(h, mb_type))
2100 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2101 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2103 if (FRAME_MBAFF(h)) {
2104 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2105 linesize, uvlinesize);
2107 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2108 dest_cr, linesize, uvlinesize);
2112 h->slice_type = old_slice_type;
2114 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2115 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2116 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2119 static void predict_field_decoding_flag(H264Context *h)
2121 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2122 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2123 h->cur_pic.mb_type[mb_xy - 1] :
2124 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2125 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2126 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2130 * Draw edges and report progress for the last MB row.
2132 static void decode_finish_row(H264Context *h)
2134 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2135 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2136 int height = 16 << FRAME_MBAFF(h);
2137 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2139 if (h->deblocking_filter) {
2140 if ((top + height) >= pic_height)
2141 height += deblock_border;
2142 top -= deblock_border;
2145 if (top >= pic_height || (top + height) < 0)
2148 height = FFMIN(height, pic_height - top);
2150 height = top + height;
2154 ff_h264_draw_horiz_band(h, top, height);
2159 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2160 h->picture_structure == PICT_BOTTOM_FIELD);
2163 static void er_add_slice(H264Context *h, int startx, int starty,
2164 int endx, int endy, int status)
2166 #if CONFIG_ERROR_RESILIENCE
2167 ERContext *er = &h->er;
2169 er->ref_count = h->ref_count[0];
2170 ff_er_add_slice(er, startx, starty, endx, endy, status);
2174 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2176 H264Context *h = *(void **)arg;
2177 int lf_x_start = h->mb_x;
2179 h->mb_skip_run = -1;
2181 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2182 avctx->codec_id != AV_CODEC_ID_H264 ||
2183 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2187 align_get_bits(&h->gb);
2190 ff_init_cabac_decoder(&h->cabac,
2191 h->gb.buffer + get_bits_count(&h->gb) / 8,
2192 (get_bits_left(&h->gb) + 7) / 8);
2194 ff_h264_init_cabac_states(h);
2198 int ret = ff_h264_decode_mb_cabac(h);
2200 // STOP_TIMER("decode_mb_cabac")
2203 ff_h264_hl_decode_mb(h);
2205 // FIXME optimal? or let mb_decode decode 16x32 ?
2206 if (ret >= 0 && FRAME_MBAFF(h)) {
2209 ret = ff_h264_decode_mb_cabac(h);
2212 ff_h264_hl_decode_mb(h);
2215 eos = get_cabac_terminate(&h->cabac);
2217 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2218 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2219 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2220 h->mb_y, ER_MB_END);
2221 if (h->mb_x >= lf_x_start)
2222 loop_filter(h, lf_x_start, h->mb_x + 1);
2225 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2226 av_log(h->avctx, AV_LOG_ERROR,
2227 "error while decoding MB %d %d, bytestream %td\n",
2229 h->cabac.bytestream_end - h->cabac.bytestream);
2230 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2231 h->mb_y, ER_MB_ERROR);
2232 return AVERROR_INVALIDDATA;
2235 if (++h->mb_x >= h->mb_width) {
2236 loop_filter(h, lf_x_start, h->mb_x);
2237 h->mb_x = lf_x_start = 0;
2238 decode_finish_row(h);
2240 if (FIELD_OR_MBAFF_PICTURE(h)) {
2242 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2243 predict_field_decoding_flag(h);
2247 if (eos || h->mb_y >= h->mb_height) {
2248 tprintf(h->avctx, "slice end %d %d\n",
2249 get_bits_count(&h->gb), h->gb.size_in_bits);
2250 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2251 h->mb_y, ER_MB_END);
2252 if (h->mb_x > lf_x_start)
2253 loop_filter(h, lf_x_start, h->mb_x);
2259 int ret = ff_h264_decode_mb_cavlc(h);
2262 ff_h264_hl_decode_mb(h);
2264 // FIXME optimal? or let mb_decode decode 16x32 ?
2265 if (ret >= 0 && FRAME_MBAFF(h)) {
2267 ret = ff_h264_decode_mb_cavlc(h);
2270 ff_h264_hl_decode_mb(h);
2275 av_log(h->avctx, AV_LOG_ERROR,
2276 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2277 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2278 h->mb_y, ER_MB_ERROR);
2282 if (++h->mb_x >= h->mb_width) {
2283 loop_filter(h, lf_x_start, h->mb_x);
2284 h->mb_x = lf_x_start = 0;
2285 decode_finish_row(h);
2287 if (FIELD_OR_MBAFF_PICTURE(h)) {
2289 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2290 predict_field_decoding_flag(h);
2292 if (h->mb_y >= h->mb_height) {
2293 tprintf(h->avctx, "slice end %d %d\n",
2294 get_bits_count(&h->gb), h->gb.size_in_bits);
2296 if (get_bits_left(&h->gb) == 0) {
2297 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2298 h->mb_x - 1, h->mb_y,
2303 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2304 h->mb_x - 1, h->mb_y,
2307 return AVERROR_INVALIDDATA;
2312 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2313 tprintf(h->avctx, "slice end %d %d\n",
2314 get_bits_count(&h->gb), h->gb.size_in_bits);
2316 if (get_bits_left(&h->gb) == 0) {
2317 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2318 h->mb_x - 1, h->mb_y,
2320 if (h->mb_x > lf_x_start)
2321 loop_filter(h, lf_x_start, h->mb_x);
2325 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2326 h->mb_y, ER_MB_ERROR);
2328 return AVERROR_INVALIDDATA;
2336 * Call decode_slice() for each context.
2338 * @param h h264 master context
2339 * @param context_count number of contexts to execute
2341 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2343 AVCodecContext *const avctx = h->avctx;
2347 if (h->mb_y >= h->mb_height) {
2348 av_log(h->avctx, AV_LOG_ERROR,
2349 "Input contains more MB rows than the frame height.\n");
2350 return AVERROR_INVALIDDATA;
2353 if (h->avctx->hwaccel)
2355 if (context_count == 1) {
2356 return decode_slice(avctx, &h);
2358 for (i = 1; i < context_count; i++) {
2359 hx = h->thread_context[i];
2360 hx->er.error_count = 0;
2363 avctx->execute(avctx, decode_slice, h->thread_context,
2364 NULL, context_count, sizeof(void *));
2366 /* pull back stuff from slices to master context */
2367 hx = h->thread_context[context_count - 1];
2370 h->droppable = hx->droppable;
2371 h->picture_structure = hx->picture_structure;
2372 for (i = 1; i < context_count; i++)
2373 h->er.error_count += h->thread_context[i]->er.error_count;