lavc: Add new VDA hwaccel
[libav.git] / libavcodec / h264_slice.c
CommitLineData
f51d0f39
LB
1/*
2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of Libav.
6 *
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.
11 *
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.
16 *
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
20 */
21
22/**
23 * @file
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28#include "libavutil/avassert.h"
29#include "libavutil/imgutils.h"
30#include "libavutil/timer.h"
31#include "internal.h"
32#include "cabac.h"
33#include "cabac_functions.h"
34#include "dsputil.h"
35#include "error_resilience.h"
36#include "avcodec.h"
37#include "h264.h"
38#include "h264data.h"
39#include "h264chroma.h"
40#include "h264_mvpred.h"
41#include "golomb.h"
42#include "mathops.h"
43#include "mpegutils.h"
44#include "rectangle.h"
45#include "thread.h"
46
47
48static 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,
52};
53
54static 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,
58};
59
60static 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,
65};
66
67static 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,
84};
85
86static 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,
103};
104
105// zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
106static 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,
123};
124
125static const uint8_t dequant4_coeff_init[6][3] = {
126 { 10, 13, 16 },
127 { 11, 14, 18 },
128 { 13, 16, 20 },
129 { 14, 18, 23 },
130 { 16, 20, 25 },
131 { 18, 23, 29 },
132};
133
134static 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
136};
137
138static 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 },
145};
146
147static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
148#if CONFIG_H264_DXVA2_HWACCEL
149 AV_PIX_FMT_DXVA2_VLD,
150#endif
151#if CONFIG_H264_VAAPI_HWACCEL
152 AV_PIX_FMT_VAAPI_VLD,
153#endif
154#if CONFIG_H264_VDA_HWACCEL
155 AV_PIX_FMT_VDA_VLD,
67afcefb 156 AV_PIX_FMT_VDA,
f51d0f39
LB
157#endif
158#if CONFIG_H264_VDPAU_HWACCEL
159 AV_PIX_FMT_VDPAU,
160#endif
161 AV_PIX_FMT_YUV420P,
162 AV_PIX_FMT_NONE
163};
164
165static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
166#if CONFIG_H264_DXVA2_HWACCEL
167 AV_PIX_FMT_DXVA2_VLD,
168#endif
169#if CONFIG_H264_VAAPI_HWACCEL
170 AV_PIX_FMT_VAAPI_VLD,
171#endif
172#if CONFIG_H264_VDA_HWACCEL
173 AV_PIX_FMT_VDA_VLD,
67afcefb 174 AV_PIX_FMT_VDA,
f51d0f39
LB
175#endif
176#if CONFIG_H264_VDPAU_HWACCEL
177 AV_PIX_FMT_VDPAU,
178#endif
179 AV_PIX_FMT_YUVJ420P,
180 AV_PIX_FMT_NONE
181};
182
183
184static void release_unused_pictures(H264Context *h, int remove_current)
185{
186 int i;
187
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]);
193 }
194 }
195}
196
197static int alloc_scratch_buffers(H264Context *h, int linesize)
198{
199 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
200
201 if (h->bipred_scratchpad)
202 return 0;
203
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);
208
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);
213 }
214
215 return 0;
216}
217
218static int init_table_pools(H264Context *h)
219{
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;
224
225 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
226 av_buffer_allocz);
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);
232
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);
240 }
241
242 return 0;
243}
244
245static int alloc_picture(H264Context *h, H264Picture *pic)
246{
247 int i, ret = 0;
248
249 av_assert0(!pic->f.data[0]);
250
251 pic->tf.f = &pic->f;
252 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
253 AV_GET_BUFFER_FLAG_REF : 0);
254 if (ret < 0)
255 goto fail;
256
257 h->linesize = pic->f.linesize[0];
258 h->uvlinesize = pic->f.linesize[1];
259
260 if (h->avctx->hwaccel) {
261 const AVHWAccel *hwaccel = h->avctx->hwaccel;
262 av_assert0(!pic->hwaccel_picture_private);
a871ef0c
AK
263 if (hwaccel->frame_priv_data_size) {
264 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
f51d0f39
LB
265 if (!pic->hwaccel_priv_buf)
266 return AVERROR(ENOMEM);
267 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
268 }
269 }
270
271 if (!h->qscale_table_pool) {
272 ret = init_table_pools(h);
273 if (ret < 0)
274 goto fail;
275 }
276
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)
280 goto fail;
281
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;
284
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])
289 goto fail;
290
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;
293 }
294
295 return 0;
296fail:
297 ff_h264_unref_picture(h, pic);
298 return (ret < 0) ? ret : AVERROR(ENOMEM);
299}
300
301static inline int pic_is_unused(H264Context *h, H264Picture *pic)
302{
303 if (!pic->f.buf[0])
304 return 1;
305 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
306 return 1;
307 return 0;
308}
309
310static int find_unused_picture(H264Context *h)
311{
312 int i;
313
314 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
315 if (pic_is_unused(h, &h->DPB[i]))
316 break;
317 }
318 if (i == H264_MAX_PICTURE_COUNT)
319 return AVERROR_INVALIDDATA;
320
321 if (h->DPB[i].needs_realloc) {
322 h->DPB[i].needs_realloc = 0;
323 ff_h264_unref_picture(h, &h->DPB[i]);
324 }
325
326 return i;
327}
328
329
330static void init_dequant8_coeff_table(H264Context *h)
331{
332 int i, j, q, x;
333 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
334
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];
341 break;
342 }
343 if (j < i)
344 continue;
345
346 for (q = 0; q < max_qp + 1; q++) {
347 int shift = div6[q];
348 int idx = rem6[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;
353 }
354 }
355}
356
357static void init_dequant4_coeff_table(H264Context *h)
358{
359 int i, j, q, x;
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];
367 break;
368 }
369 if (j < i)
370 continue;
371
372 for (q = 0; q < max_qp + 1; q++) {
373 int shift = div6[q] + 2;
374 int idx = rem6[q];
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;
379 }
380 }
381}
382
383void h264_init_dequant_tables(H264Context *h)
384{
385 int i, x;
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;
397 }
398}
399
400/**
401 * Mimic alloc_tables(), but for every context thread.
402 */
403static void clone_tables(H264Context *dst, H264Context *src, int i)
404{
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;
416 dst->DPB = src->DPB;
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);
423}
424
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)
431
432static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
433 H264Context *new_base,
434 H264Context *old_base)
435{
436 int i;
437
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) ||
442 !from[i]));
443 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
444 }
445}
446
447static int copy_parameter_set(void **to, void **from, int count, int size)
448{
449 int i;
450
451 for (i = 0; i < count; i++) {
452 if (to[i] && !from[i]) {
453 av_freep(&to[i]);
454 } else if (from[i] && !to[i]) {
455 to[i] = av_malloc(size);
456 if (!to[i])
457 return AVERROR(ENOMEM);
458 }
459
460 if (from[i])
461 memcpy(to[i], from[i], size);
462 }
463
464 return 0;
465}
466
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)
470
471static int h264_slice_header_init(H264Context *h, int reinit);
472
473int ff_h264_update_thread_context(AVCodecContext *dst,
474 const AVCodecContext *src)
475{
476 H264Context *h = dst->priv_data, *h1 = src->priv_data;
477 int inited = h->context_initialized, err = 0;
478 int context_reinitialized = 0;
479 int i, ret;
480
481 if (dst == src || !h1->context_initialized)
482 return 0;
483
484 if (inited &&
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)) {
492
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;
497
498 av_freep(&h->bipred_scratchpad);
499
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;
507
508 if ((err = h264_slice_header_init(h, 1)) < 0) {
509 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
510 return err;
511 }
512 context_reinitialized = 1;
513
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;
518
519 /* copy block_offset since frame_start may not be called */
520 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
521 }
522
523 if (!inited) {
524 for (i = 0; i < MAX_SPS_COUNT; i++)
525 av_freep(h->sps_buffers + i);
526
527 for (i = 0; i < MAX_PPS_COUNT; i++)
528 av_freep(h->pps_buffers + i);
529
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;
538
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;
542
543 h->avctx = dst;
544 h->DPB = NULL;
545 h->qscale_table_pool = NULL;
546 h->mb_type_pool = NULL;
547 h->ref_index_pool = NULL;
548 h->motion_val_pool = NULL;
549
550 ret = ff_h264_alloc_tables(h);
551 if (ret < 0) {
552 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
553 return ret;
554 }
555 ret = ff_h264_context_init(h);
556 if (ret < 0) {
557 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
558 return ret;
559 }
560
561 for (i = 0; i < 2; i++) {
562 h->rbsp_buffer[i] = NULL;
563 h->rbsp_buffer_size[i] = 0;
564 }
565 h->bipred_scratchpad = NULL;
566 h->edge_emu_buffer = NULL;
567
568 h->thread_context[0] = h;
569
570 h->context_initialized = 1;
571 }
572
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;
583
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)
588 return ret;
589 }
590
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)
594 return ret;
595
596 h->workaround_bugs = h1->workaround_bugs;
597 h->low_delay = h1->low_delay;
598 h->droppable = h1->droppable;
599
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);
603 if (err < 0)
604 return err;
605
606 // extradata/NAL handling
607 h->is_avc = h1->is_avc;
608
609 // SPS/PPS
610 if ((ret = copy_parameter_set((void **)h->sps_buffers,
611 (void **)h1->sps_buffers,
612 MAX_SPS_COUNT, sizeof(SPS))) < 0)
613 return ret;
614 h->sps = h1->sps;
615 if ((ret = copy_parameter_set((void **)h->pps_buffers,
616 (void **)h1->pps_buffers,
617 MAX_PPS_COUNT, sizeof(PPS))) < 0)
618 return ret;
619 h->pps = h1->pps;
620
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);
624
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]);
628
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]);
632
633 h->dequant_coeff_pps = h1->dequant_coeff_pps;
634
635 // POC timing
636 copy_fields(h, h1, poc_lsb, redundant_pic_count);
637
638 // reference lists
639 copy_fields(h, h1, short_ref, cabac_init_idc);
640
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);
645
646 h->last_slice_type = h1->last_slice_type;
647
648 if (context_reinitialized)
649 ff_h264_set_parameter_from_sps(h);
650
651 if (!h->cur_pic_ptr)
652 return 0;
653
654 if (!h->droppable) {
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;
658 }
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;
662
663 h->recovery_frame = h1->recovery_frame;
664 h->frame_recovered = h1->frame_recovered;
665
666 return err;
667}
668
669static int h264_frame_start(H264Context *h)
670{
671 H264Picture *pic;
672 int i, ret;
673 const int pixel_shift = h->pixel_shift;
674
675 release_unused_pictures(h, 1);
676 h->cur_pic_ptr = NULL;
677
678 i = find_unused_picture(h);
679 if (i < 0) {
680 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
681 return i;
682 }
683 pic = &h->DPB[i];
684
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;
688 /*
689 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
690 * in later.
691 * See decode_nal_units().
692 */
693 pic->f.key_frame = 0;
694 pic->mmco_reset = 0;
695 pic->recovered = 0;
696
697 if ((ret = alloc_picture(h, pic)) < 0)
698 return ret;
699
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)
703 return ret;
704
705 if (CONFIG_ERROR_RESILIENCE)
706 ff_er_frame_start(&h->er);
707
708 assert(h->linesize && h->uvlinesize);
709
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);
713 }
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);
719 }
720
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);
726 if (ret < 0)
727 return ret;
728 }
729
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));
734
f51d0f39
LB
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().
738 */
739 h->cur_pic_ptr->reference = 0;
740
741 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
742
743 h->next_output_pic = NULL;
744
745 assert(h->cur_pic_ptr->long_ref == 0);
746
747 return 0;
748}
749
750static 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,
753 int simple)
754{
755 uint8_t *top_border;
756 int top_idx = 1;
757 const int pixel_shift = h->pixel_shift;
758 int chroma444 = CHROMA444(h);
759 int chroma422 = CHROMA422(h);
760
761 src_y -= linesize;
762 src_cb -= uvlinesize;
763 src_cr -= uvlinesize;
764
765 if (!simple && FRAME_MBAFF(h)) {
766 if (h->mb_y & 1) {
767 if (!MB_MBAFF(h)) {
768 top_border = h->top_borders[0][h->mb_x];
769 AV_COPY128(top_border, src_y + 15 * linesize);
770 if (pixel_shift)
771 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
772 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
773 if (chroma444) {
774 if (pixel_shift) {
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);
779 } else {
780 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
781 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
782 }
783 } else if (chroma422) {
784 if (pixel_shift) {
785 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
786 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
787 } else {
788 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
789 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
790 }
791 } else {
792 if (pixel_shift) {
793 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
794 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
795 } else {
796 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
797 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
798 }
799 }
800 }
801 }
802 } else if (MB_MBAFF(h)) {
803 top_idx = 0;
804 } else
805 return;
806 }
807
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);
812 if (pixel_shift)
813 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
814
815 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
816 if (chroma444) {
817 if (pixel_shift) {
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);
822 } else {
823 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
824 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
825 }
826 } else if (chroma422) {
827 if (pixel_shift) {
828 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
829 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
830 } else {
831 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
832 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
833 }
834 } else {
835 if (pixel_shift) {
836 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
837 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
838 } else {
839 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
840 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
841 }
842 }
843 }
844}
845
846/**
847 * Initialize implicit_weight table.
848 * @param field 0/1 initialize the weight for interlaced MBAFF
849 * -1 initializes the rest
850 */
851static void implicit_weight_table(H264Context *h, int field)
852{
853 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
854
855 for (i = 0; i < 2; i++) {
856 h->luma_weight_flag[i] = 0;
857 h->chroma_weight_flag[i] = 0;
858 }
859
860 if (field < 0) {
861 if (h->picture_structure == PICT_FRAME) {
862 cur_poc = h->cur_pic_ptr->poc;
863 } else {
864 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
865 }
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) {
868 h->use_weight = 0;
869 h->use_weight_chroma = 0;
870 return;
871 }
872 ref_start = 0;
873 ref_count0 = h->ref_count[0];
874 ref_count1 = h->ref_count[1];
875 } else {
876 cur_poc = h->cur_pic_ptr->field_poc[field];
877 ref_start = 16;
878 ref_count0 = 16 + 2 * h->ref_count[0];
879 ref_count1 = 16 + 2 * h->ref_count[1];
880 }
881
882 h->use_weight = 2;
883 h->use_weight_chroma = 2;
884 h->luma_log2_weight_denom = 5;
885 h->chroma_log2_weight_denom = 5;
886
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++) {
890 int w = 32;
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);
894 if (td) {
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;
900 }
901 }
902 if (field < 0) {
903 h->implicit_weight[ref0][ref1][0] =
904 h->implicit_weight[ref0][ref1][1] = w;
905 } else {
906 h->implicit_weight[ref0][ref1][field] = w;
907 }
908 }
909 }
910}
911
912/**
913 * initialize scan tables
914 */
915static void init_scan_tables(H264Context *h)
916{
917 int i;
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]);
922#undef TRANSPOSE
923 }
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]);
930#undef TRANSPOSE
931 }
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;
939 } else {
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;
946 }
947}
948
949/**
950 * Replicate H264 "master" context to thread contexts.
951 */
952static int clone_slice(H264Context *dst, H264Context *src)
953{
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;
960
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;
966
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));
970
971 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
972 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
973
974 return 0;
975}
976
977static enum AVPixelFormat get_pixel_format(H264Context *h)
978{
979 switch (h->sps.bit_depth_luma) {
980 case 9:
981 if (CHROMA444(h)) {
982 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
983 return AV_PIX_FMT_GBRP9;
984 } else
985 return AV_PIX_FMT_YUV444P9;
986 } else if (CHROMA422(h))
987 return AV_PIX_FMT_YUV422P9;
988 else
989 return AV_PIX_FMT_YUV420P9;
990 break;
991 case 10:
992 if (CHROMA444(h)) {
993 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
994 return AV_PIX_FMT_GBRP10;
995 } else
996 return AV_PIX_FMT_YUV444P10;
997 } else if (CHROMA422(h))
998 return AV_PIX_FMT_YUV422P10;
999 else
1000 return AV_PIX_FMT_YUV420P10;
1001 break;
1002 case 8:
1003 if (CHROMA444(h)) {
1004 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1005 return AV_PIX_FMT_GBRP;
1006 } else
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;
1012 } else {
632ad224
AK
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);
f51d0f39
LB
1018 }
1019 break;
1020 default:
1021 av_log(h->avctx, AV_LOG_ERROR,
1022 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1023 return AVERROR_INVALIDDATA;
1024 }
1025}
1026
1027/* export coded and cropped frame dimensions to AVCodecContext */
1028static int init_dimensions(H264Context *h)
1029{
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);
1032
1033 /* handle container cropping */
1034 if (!h->sps.crop &&
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;
1039 }
1040
1041 if (width <= 0 || height <= 0) {
1042 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1043 width, height);
1044 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1045 return AVERROR_INVALIDDATA;
1046
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;
1049 h->sps.crop = 0;
1050
1051 width = h->width;
1052 height = h->height;
1053 }
1054
1055 h->avctx->coded_width = h->width;
1056 h->avctx->coded_height = h->height;
1057 h->avctx->width = width;
1058 h->avctx->height = height;
1059
1060 return 0;
1061}
1062
1063static int h264_slice_header_init(H264Context *h, int reinit)
1064{
1065 int nb_slices = (HAVE_THREADS &&
1066 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1067 h->avctx->thread_count : 1;
1068 int i, ret;
1069
1070 h->avctx->sample_aspect_ratio = h->sps.sar;
1071 av_assert0(h->avctx->sample_aspect_ratio.den);
1072 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1073 &h->chroma_x_shift, &h->chroma_y_shift);
1074
1075 if (h->sps.timing_info_present_flag) {
1076 int64_t den = h->sps.time_scale;
1077 if (h->x264_build < 44U)
1078 den *= 2;
1079 av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
1080 h->sps.num_units_in_tick, den, 1 << 30);
1081 }
1082
f51d0f39
LB
1083 if (reinit)
1084 ff_h264_free_tables(h, 0);
1085 h->first_field = 0;
1086 h->prev_interlaced_frame = 1;
1087
1088 init_scan_tables(h);
1089 ret = ff_h264_alloc_tables(h);
1090 if (ret < 0) {
1091 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1092 return ret;
1093 }
1094
1095 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1096 int max_slices;
1097 if (h->mb_height)
1098 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1099 else
1100 max_slices = H264_MAX_THREADS;
1101 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1102 " reducing to %d\n", nb_slices, max_slices);
1103 nb_slices = max_slices;
1104 }
1105 h->slice_context_count = nb_slices;
1106
1107 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1108 ret = ff_h264_context_init(h);
1109 if (ret < 0) {
1110 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1111 return ret;
1112 }
1113 } else {
1114 for (i = 1; i < h->slice_context_count; i++) {
1115 H264Context *c;
1116 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1117 if (!c)
1118 return AVERROR(ENOMEM);
1119 c->avctx = h->avctx;
1120 c->dsp = h->dsp;
1121 c->vdsp = h->vdsp;
1122 c->h264dsp = h->h264dsp;
1123 c->h264qpel = h->h264qpel;
1124 c->h264chroma = h->h264chroma;
1125 c->sps = h->sps;
1126 c->pps = h->pps;
1127 c->pixel_shift = h->pixel_shift;
1128 c->width = h->width;
1129 c->height = h->height;
1130 c->linesize = h->linesize;
1131 c->uvlinesize = h->uvlinesize;
1132 c->chroma_x_shift = h->chroma_x_shift;
1133 c->chroma_y_shift = h->chroma_y_shift;
1134 c->qscale = h->qscale;
1135 c->droppable = h->droppable;
1136 c->data_partitioning = h->data_partitioning;
1137 c->low_delay = h->low_delay;
1138 c->mb_width = h->mb_width;
1139 c->mb_height = h->mb_height;
1140 c->mb_stride = h->mb_stride;
1141 c->mb_num = h->mb_num;
1142 c->flags = h->flags;
1143 c->workaround_bugs = h->workaround_bugs;
1144 c->pict_type = h->pict_type;
1145
1146 init_scan_tables(c);
1147 clone_tables(c, h, i);
1148 c->context_initialized = 1;
1149 }
1150
1151 for (i = 0; i < h->slice_context_count; i++)
1152 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1153 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1154 return ret;
1155 }
1156 }
1157
1158 h->context_initialized = 1;
1159
1160 return 0;
1161}
1162
1163/**
1164 * Decode a slice header.
1165 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1166 *
1167 * @param h h264context
1168 * @param h0 h264 master context (differs from 'h' when doing sliced based
1169 * parallel decoding)
1170 *
1171 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1172 */
1173int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1174{
1175 unsigned int first_mb_in_slice;
1176 unsigned int pps_id;
1177 int ret;
1178 unsigned int slice_type, tmp, i, j;
1179 int default_ref_list_done = 0;
1180 int last_pic_structure, last_pic_droppable;
1181 int needs_reinit = 0;
1182 int field_pic_flag, bottom_field_flag;
1183
1184 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1185 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1186
1187 first_mb_in_slice = get_ue_golomb(&h->gb);
1188
1189 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1190 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1191 ff_h264_field_end(h, 1);
1192 }
1193
1194 h0->current_slice = 0;
1195 if (!h0->first_field) {
1196 if (h->cur_pic_ptr && !h->droppable) {
1197 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1198 h->picture_structure == PICT_BOTTOM_FIELD);
1199 }
1200 h->cur_pic_ptr = NULL;
1201 }
1202 }
1203
1204 slice_type = get_ue_golomb_31(&h->gb);
1205 if (slice_type > 9) {
1206 av_log(h->avctx, AV_LOG_ERROR,
1207 "slice type %d too large at %d %d\n",
1208 slice_type, h->mb_x, h->mb_y);
1209 return AVERROR_INVALIDDATA;
1210 }
1211 if (slice_type > 4) {
1212 slice_type -= 5;
1213 h->slice_type_fixed = 1;
1214 } else
1215 h->slice_type_fixed = 0;
1216
1217 slice_type = golomb_to_pict_type[slice_type];
1218 if (slice_type == AV_PICTURE_TYPE_I ||
1219 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
1220 default_ref_list_done = 1;
1221 }
1222 h->slice_type = slice_type;
1223 h->slice_type_nos = slice_type & 3;
1224
1225 if (h->nal_unit_type == NAL_IDR_SLICE &&
1226 h->slice_type_nos != AV_PICTURE_TYPE_I) {
1227 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1228 return AVERROR_INVALIDDATA;
1229 }
1230
1231 // to make a few old functions happy, it's wrong though
1232 h->pict_type = h->slice_type;
1233
1234 pps_id = get_ue_golomb(&h->gb);
1235 if (pps_id >= MAX_PPS_COUNT) {
1236 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1237 return AVERROR_INVALIDDATA;
1238 }
1239 if (!h0->pps_buffers[pps_id]) {
1240 av_log(h->avctx, AV_LOG_ERROR,
1241 "non-existing PPS %u referenced\n",
1242 pps_id);
1243 return AVERROR_INVALIDDATA;
1244 }
1245 h->pps = *h0->pps_buffers[pps_id];
1246
1247 if (!h0->sps_buffers[h->pps.sps_id]) {
1248 av_log(h->avctx, AV_LOG_ERROR,
1249 "non-existing SPS %u referenced\n",
1250 h->pps.sps_id);
1251 return AVERROR_INVALIDDATA;
1252 }
1253
1254 if (h->pps.sps_id != h->sps.sps_id ||
1255 h0->sps_buffers[h->pps.sps_id]->new) {
1256 h0->sps_buffers[h->pps.sps_id]->new = 0;
1257
1258 h->sps = *h0->sps_buffers[h->pps.sps_id];
1259
1260 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1261 h->chroma_format_idc != h->sps.chroma_format_idc) {
1262 h->bit_depth_luma = h->sps.bit_depth_luma;
1263 h->chroma_format_idc = h->sps.chroma_format_idc;
1264 needs_reinit = 1;
1265 }
1266 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1267 return ret;
1268 }
1269
1270 h->avctx->profile = ff_h264_get_profile(&h->sps);
1271 h->avctx->level = h->sps.level_idc;
1272 h->avctx->refs = h->sps.ref_frame_count;
1273
1274 if (h->mb_width != h->sps.mb_width ||
1275 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1276 needs_reinit = 1;
1277
1278 h->mb_width = h->sps.mb_width;
1279 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1280 h->mb_num = h->mb_width * h->mb_height;
1281 h->mb_stride = h->mb_width + 1;
1282
1283 h->b_stride = h->mb_width * 4;
1284
1285 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1286
1287 h->width = 16 * h->mb_width;
1288 h->height = 16 * h->mb_height;
1289
1290 ret = init_dimensions(h);
1291 if (ret < 0)
1292 return ret;
1293
1294 if (h->sps.video_signal_type_present_flag) {
1295 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1296 : AVCOL_RANGE_MPEG;
1297 if (h->sps.colour_description_present_flag) {
1298 if (h->avctx->colorspace != h->sps.colorspace)
1299 needs_reinit = 1;
1300 h->avctx->color_primaries = h->sps.color_primaries;
1301 h->avctx->color_trc = h->sps.color_trc;
1302 h->avctx->colorspace = h->sps.colorspace;
1303 }
1304 }
1305
34c5a666 1306 if (h->context_initialized && needs_reinit) {
f51d0f39
LB
1307 if (h != h0) {
1308 av_log(h->avctx, AV_LOG_ERROR,
1309 "changing width %d -> %d / height %d -> %d on "
1310 "slice %d\n",
1311 h->width, h->avctx->coded_width,
1312 h->height, h->avctx->coded_height,
1313 h0->current_slice + 1);
1314 return AVERROR_INVALIDDATA;
1315 }
1316
1317 ff_h264_flush_change(h);
1318
1319 if ((ret = get_pixel_format(h)) < 0)
1320 return ret;
1321 h->avctx->pix_fmt = ret;
1322
1323 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1324 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1325
1326 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1327 av_log(h->avctx, AV_LOG_ERROR,
1328 "h264_slice_header_init() failed\n");
1329 return ret;
1330 }
1331 }
1332 if (!h->context_initialized) {
1333 if (h != h0) {
1334 av_log(h->avctx, AV_LOG_ERROR,
1335 "Cannot (re-)initialize context during parallel decoding.\n");
1336 return AVERROR_PATCHWELCOME;
1337 }
1338
1339 if ((ret = get_pixel_format(h)) < 0)
1340 return ret;
1341 h->avctx->pix_fmt = ret;
1342
1343 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1344 av_log(h->avctx, AV_LOG_ERROR,
1345 "h264_slice_header_init() failed\n");
1346 return ret;
1347 }
1348 }
1349
1350 if (h == h0 && h->dequant_coeff_pps != pps_id) {
1351 h->dequant_coeff_pps = pps_id;
1352 h264_init_dequant_tables(h);
1353 }
1354
1355 h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1356
1357 h->mb_mbaff = 0;
1358 h->mb_aff_frame = 0;
1359 last_pic_structure = h0->picture_structure;
1360 last_pic_droppable = h0->droppable;
1361 h->droppable = h->nal_ref_idc == 0;
1362 if (h->sps.frame_mbs_only_flag) {
1363 h->picture_structure = PICT_FRAME;
1364 } else {
1365 field_pic_flag = get_bits1(&h->gb);
1366 if (field_pic_flag) {
1367 bottom_field_flag = get_bits1(&h->gb);
1368 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1369 } else {
1370 h->picture_structure = PICT_FRAME;
1371 h->mb_aff_frame = h->sps.mb_aff;
1372 }
1373 }
1374 h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1375
1376 if (h0->current_slice != 0) {
1377 if (last_pic_structure != h->picture_structure ||
1378 last_pic_droppable != h->droppable) {
1379 av_log(h->avctx, AV_LOG_ERROR,
1380 "Changing field mode (%d -> %d) between slices is not allowed\n",
1381 last_pic_structure, h->picture_structure);
1382 h->picture_structure = last_pic_structure;
1383 h->droppable = last_pic_droppable;
1384 return AVERROR_INVALIDDATA;
1385 } else if (!h0->cur_pic_ptr) {
1386 av_log(h->avctx, AV_LOG_ERROR,
1387 "unset cur_pic_ptr on slice %d\n",
1388 h0->current_slice + 1);
1389 return AVERROR_INVALIDDATA;
1390 }
1391 } else {
1392 /* Shorten frame num gaps so we don't have to allocate reference
1393 * frames just to throw them away */
1394 if (h->frame_num != h->prev_frame_num) {
1395 int unwrap_prev_frame_num = h->prev_frame_num;
1396 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1397
1398 if (unwrap_prev_frame_num > h->frame_num)
1399 unwrap_prev_frame_num -= max_frame_num;
1400
1401 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1402 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1403 if (unwrap_prev_frame_num < 0)
1404 unwrap_prev_frame_num += max_frame_num;
1405
1406 h->prev_frame_num = unwrap_prev_frame_num;
1407 }
1408 }
1409
1410 /* See if we have a decoded first field looking for a pair...
1411 * Here, we're using that to see if we should mark previously
1412 * decode frames as "finished".
1413 * We have to do that before the "dummy" in-between frame allocation,
1414 * since that can modify s->current_picture_ptr. */
1415 if (h0->first_field) {
1416 assert(h0->cur_pic_ptr);
1417 assert(h0->cur_pic_ptr->f.buf[0]);
1418 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1419
1420 /* figure out if we have a complementary field pair */
1421 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1422 /* Previous field is unmatched. Don't display it, but let it
1423 * remain for reference if marked as such. */
1424 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1425 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1426 last_pic_structure == PICT_TOP_FIELD);
1427 }
1428 } else {
1429 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1430 /* This and previous field were reference, but had
1431 * different frame_nums. Consider this field first in
1432 * pair. Throw away previous field except for reference
1433 * purposes. */
1434 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1435 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1436 last_pic_structure == PICT_TOP_FIELD);
1437 }
1438 } else {
1439 /* Second field in complementary pair */
1440 if (!((last_pic_structure == PICT_TOP_FIELD &&
1441 h->picture_structure == PICT_BOTTOM_FIELD) ||
1442 (last_pic_structure == PICT_BOTTOM_FIELD &&
1443 h->picture_structure == PICT_TOP_FIELD))) {
1444 av_log(h->avctx, AV_LOG_ERROR,
1445 "Invalid field mode combination %d/%d\n",
1446 last_pic_structure, h->picture_structure);
1447 h->picture_structure = last_pic_structure;
1448 h->droppable = last_pic_droppable;
1449 return AVERROR_INVALIDDATA;
1450 } else if (last_pic_droppable != h->droppable) {
1451 avpriv_request_sample(h->avctx,
1452 "Found reference and non-reference fields in the same frame, which");
1453 h->picture_structure = last_pic_structure;
1454 h->droppable = last_pic_droppable;
1455 return AVERROR_PATCHWELCOME;
1456 }
1457 }
1458 }
1459 }
1460
1461 while (h->frame_num != h->prev_frame_num &&
1462 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1463 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1464 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1465 h->frame_num, h->prev_frame_num);
1466 ret = h264_frame_start(h);
1467 if (ret < 0) {
1468 h0->first_field = 0;
1469 return ret;
1470 }
1471
1472 h->prev_frame_num++;
1473 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1474 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1475 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1476 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1477 ret = ff_generate_sliding_window_mmcos(h, 1);
1478 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1479 return ret;
1480 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1481 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1482 return ret;
1483 /* Error concealment: If a ref is missing, copy the previous ref
1484 * in its place.
1485 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1486 * many assumptions about there being no actual duplicates.
1487 * FIXME: This does not copy padding for out-of-frame motion
1488 * vectors. Given we are concealing a lost frame, this probably
1489 * is not noticeable by comparison, but it should be fixed. */
1490 if (h->short_ref_count) {
1491 if (prev) {
1492 av_image_copy(h->short_ref[0]->f.data,
1493 h->short_ref[0]->f.linesize,
1494 (const uint8_t **)prev->f.data,
1495 prev->f.linesize,
1496 h->avctx->pix_fmt,
1497 h->mb_width * 16,
1498 h->mb_height * 16);
1499 h->short_ref[0]->poc = prev->poc + 2;
1500 }
1501 h->short_ref[0]->frame_num = h->prev_frame_num;
1502 }
1503 }
1504
1505 /* See if we have a decoded first field looking for a pair...
1506 * We're using that to see whether to continue decoding in that
1507 * frame, or to allocate a new one. */
1508 if (h0->first_field) {
1509 assert(h0->cur_pic_ptr);
1510 assert(h0->cur_pic_ptr->f.buf[0]);
1511 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1512
1513 /* figure out if we have a complementary field pair */
1514 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1515 /* Previous field is unmatched. Don't display it, but let it
1516 * remain for reference if marked as such. */
1517 h0->cur_pic_ptr = NULL;
1518 h0->first_field = FIELD_PICTURE(h);
1519 } else {
1520 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1521 /* This and the previous field had different frame_nums.
1522 * Consider this field first in pair. Throw away previous
1523 * one except for reference purposes. */
1524 h0->first_field = 1;
1525 h0->cur_pic_ptr = NULL;
1526 } else {
1527 /* Second field in complementary pair */
1528 h0->first_field = 0;
1529 }
1530 }
1531 } else {
1532 /* Frame or first field in a potentially complementary pair */
1533 h0->first_field = FIELD_PICTURE(h);
1534 }
1535
1536 if (!FIELD_PICTURE(h) || h0->first_field) {
1537 if (h264_frame_start(h) < 0) {
1538 h0->first_field = 0;
1539 return AVERROR_INVALIDDATA;
1540 }
1541 } else {
1542 release_unused_pictures(h, 0);
1543 }
1544 }
1545 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1546 return ret;
1547
1548 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1549
1550 assert(h->mb_num == h->mb_width * h->mb_height);
1551 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1552 first_mb_in_slice >= h->mb_num) {
1553 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1554 return AVERROR_INVALIDDATA;
1555 }
1556 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1557 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1558 FIELD_OR_MBAFF_PICTURE(h);
1559 if (h->picture_structure == PICT_BOTTOM_FIELD)
1560 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1561 assert(h->mb_y < h->mb_height);
1562
1563 if (h->picture_structure == PICT_FRAME) {
1564 h->curr_pic_num = h->frame_num;
1565 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1566 } else {
1567 h->curr_pic_num = 2 * h->frame_num + 1;
1568 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1569 }
1570
1571 if (h->nal_unit_type == NAL_IDR_SLICE)
1572 get_ue_golomb(&h->gb); /* idr_pic_id */
1573
1574 if (h->sps.poc_type == 0) {
1575 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1576
1577 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1578 h->delta_poc_bottom = get_se_golomb(&h->gb);
1579 }
1580
1581 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1582 h->delta_poc[0] = get_se_golomb(&h->gb);
1583
1584 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1585 h->delta_poc[1] = get_se_golomb(&h->gb);
1586 }
1587
1588 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1589
1590 if (h->pps.redundant_pic_cnt_present)
1591 h->redundant_pic_count = get_ue_golomb(&h->gb);
1592
1593 ret = ff_set_ref_count(h);
1594 if (ret < 0)
1595 return ret;
1596 else if (ret == 1)
1597 default_ref_list_done = 0;
1598
1599 if (!default_ref_list_done)
1600 ff_h264_fill_default_ref_list(h);
1601
1602 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1603 ret = ff_h264_decode_ref_pic_list_reordering(h);
1604 if (ret < 0) {
1605 h->ref_count[1] = h->ref_count[0] = 0;
1606 return ret;
1607 }
1608 }
1609
1610 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1611 (h->pps.weighted_bipred_idc == 1 &&
1612 h->slice_type_nos == AV_PICTURE_TYPE_B))
1613 ff_pred_weight_table(h);
1614 else if (h->pps.weighted_bipred_idc == 2 &&
1615 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1616 implicit_weight_table(h, -1);
1617 } else {
1618 h->use_weight = 0;
1619 for (i = 0; i < 2; i++) {
1620 h->luma_weight_flag[i] = 0;
1621 h->chroma_weight_flag[i] = 0;
1622 }
1623 }
1624
1625 // If frame-mt is enabled, only update mmco tables for the first slice
1626 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1627 // or h->mmco, which will cause ref list mix-ups and decoding errors
1628 // further down the line. This may break decoding if the first slice is
1629 // corrupt, thus we only do this if frame-mt is enabled.
1630 if (h->nal_ref_idc) {
1631 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1632 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1633 h0->current_slice == 0);
1634 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1635 return AVERROR_INVALIDDATA;
1636 }
1637
1638 if (FRAME_MBAFF(h)) {
1639 ff_h264_fill_mbaff_ref_list(h);
1640
1641 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1642 implicit_weight_table(h, 0);
1643 implicit_weight_table(h, 1);
1644 }
1645 }
1646
1647 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1648 ff_h264_direct_dist_scale_factor(h);
1649 ff_h264_direct_ref_list_init(h);
1650
1651 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1652 tmp = get_ue_golomb_31(&h->gb);
1653 if (tmp > 2) {
1654 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1655 return AVERROR_INVALIDDATA;
1656 }
1657 h->cabac_init_idc = tmp;
1658 }
1659
1660 h->last_qscale_diff = 0;
1661 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1662 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1663 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1664 return AVERROR_INVALIDDATA;
1665 }
1666 h->qscale = tmp;
1667 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1668 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1669 // FIXME qscale / qp ... stuff
1670 if (h->slice_type == AV_PICTURE_TYPE_SP)
1671 get_bits1(&h->gb); /* sp_for_switch_flag */
1672 if (h->slice_type == AV_PICTURE_TYPE_SP ||
1673 h->slice_type == AV_PICTURE_TYPE_SI)
1674 get_se_golomb(&h->gb); /* slice_qs_delta */
1675
1676 h->deblocking_filter = 1;
1677 h->slice_alpha_c0_offset = 0;
1678 h->slice_beta_offset = 0;
1679 if (h->pps.deblocking_filter_parameters_present) {
1680 tmp = get_ue_golomb_31(&h->gb);
1681 if (tmp > 2) {
1682 av_log(h->avctx, AV_LOG_ERROR,
1683 "deblocking_filter_idc %u out of range\n", tmp);
1684 return AVERROR_INVALIDDATA;
1685 }
1686 h->deblocking_filter = tmp;
1687 if (h->deblocking_filter < 2)
1688 h->deblocking_filter ^= 1; // 1<->0
1689
1690 if (h->deblocking_filter) {
1691 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1692 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1693 if (h->slice_alpha_c0_offset > 12 ||
1694 h->slice_alpha_c0_offset < -12 ||
1695 h->slice_beta_offset > 12 ||
1696 h->slice_beta_offset < -12) {
1697 av_log(h->avctx, AV_LOG_ERROR,
1698 "deblocking filter parameters %d %d out of range\n",
1699 h->slice_alpha_c0_offset, h->slice_beta_offset);
1700 return AVERROR_INVALIDDATA;
1701 }
1702 }
1703 }
1704
1705 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1706 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1707 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1708 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1709 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1710 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1711 h->nal_ref_idc == 0))
1712 h->deblocking_filter = 0;
1713
1714 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1715 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1716 /* Cheat slightly for speed:
1717 * Do not bother to deblock across slices. */
1718 h->deblocking_filter = 2;
1719 } else {
1720 h0->max_contexts = 1;
1721 if (!h0->single_decode_warning) {
1722 av_log(h->avctx, AV_LOG_INFO,
1723 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1724 h0->single_decode_warning = 1;
1725 }
1726 if (h != h0) {
1727 av_log(h->avctx, AV_LOG_ERROR,
1728 "Deblocking switched inside frame.\n");
1729 return 1;
1730 }
1731 }
1732 }
1733 h->qp_thresh = 15 -
1734 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1735 FFMAX3(0,
1736 h->pps.chroma_qp_index_offset[0],
1737 h->pps.chroma_qp_index_offset[1]) +
1738 6 * (h->sps.bit_depth_luma - 8);
1739
1740 h0->last_slice_type = slice_type;
1741 h->slice_num = ++h0->current_slice;
1742 if (h->slice_num >= MAX_SLICES) {
1743 av_log(h->avctx, AV_LOG_ERROR,
1744 "Too many slices, increase MAX_SLICES and recompile\n");
1745 }
1746
1747 for (j = 0; j < 2; j++) {
1748 int id_list[16];
1749 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1750 for (i = 0; i < 16; i++) {
1751 id_list[i] = 60;
1752 if (j < h->list_count && i < h->ref_count[j] &&
1753 h->ref_list[j][i].f.buf[0]) {
1754 int k;
1755 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1756 for (k = 0; k < h->short_ref_count; k++)
1757 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1758 id_list[i] = k;
1759 break;
1760 }
1761 for (k = 0; k < h->long_ref_count; k++)
1762 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1763 id_list[i] = h->short_ref_count + k;
1764 break;
1765 }
1766 }
1767 }
1768
1769 ref2frm[0] =
1770 ref2frm[1] = -1;
1771 for (i = 0; i < 16; i++)
1772 ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1773 ref2frm[18 + 0] =
1774 ref2frm[18 + 1] = -1;
1775 for (i = 16; i < 48; i++)
1776 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1777 (h->ref_list[j][i].reference & 3);
1778 }
1779
1780 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1781 av_log(h->avctx, AV_LOG_DEBUG,
1782 "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->slice_num,
1784 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1785 first_mb_in_slice,
1786 av_get_picture_type_char(h->slice_type),
1787 h->slice_type_fixed ? " fix" : "",
1788 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1789 pps_id, h->frame_num,
1790 h->cur_pic_ptr->field_poc[0],
1791 h->cur_pic_ptr->field_poc[1],
1792 h->ref_count[0], h->ref_count[1],
1793 h->qscale,
1794 h->deblocking_filter,
1795 h->slice_alpha_c0_offset, h->slice_beta_offset,
1796 h->use_weight,
1797 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
1798 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1799 }
1800
1801 return 0;
1802}
1803
1804int ff_h264_get_slice_type(const H264Context *h)
1805{
1806 switch (h->slice_type) {
1807 case AV_PICTURE_TYPE_P:
1808 return 0;
1809 case AV_PICTURE_TYPE_B:
1810 return 1;
1811 case AV_PICTURE_TYPE_I:
1812 return 2;
1813 case AV_PICTURE_TYPE_SP:
1814 return 3;
1815 case AV_PICTURE_TYPE_SI:
1816 return 4;
1817 default:
1818 return AVERROR_INVALIDDATA;
1819 }
1820}
1821
1822static av_always_inline void fill_filter_caches_inter(H264Context *h,
1823 int mb_type, int top_xy,
1824 int left_xy[LEFT_MBS],
1825 int top_type,
1826 int left_type[LEFT_MBS],
1827 int mb_xy, int list)
1828{
1829 int b_stride = h->b_stride;
1830 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
1831 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
1832 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1833 if (USES_LIST(top_type, list)) {
1834 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1835 const int b8_xy = 4 * top_xy + 2;
1836 int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1837 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1838 ref_cache[0 - 1 * 8] =
1839 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1840 ref_cache[2 - 1 * 8] =
1841 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1842 } else {
1843 AV_ZERO128(mv_dst - 1 * 8);
1844 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1845 }
1846
1847 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1848 if (USES_LIST(left_type[LTOP], list)) {
1849 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1850 const int b8_xy = 4 * left_xy[LTOP] + 1;
1851 int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1852 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1853 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1854 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1855 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1856 ref_cache[-1 + 0] =
1857 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1858 ref_cache[-1 + 16] =
1859 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1860 } else {
1861 AV_ZERO32(mv_dst - 1 + 0);
1862 AV_ZERO32(mv_dst - 1 + 8);
1863 AV_ZERO32(mv_dst - 1 + 16);
1864 AV_ZERO32(mv_dst - 1 + 24);
1865 ref_cache[-1 + 0] =
1866 ref_cache[-1 + 8] =
1867 ref_cache[-1 + 16] =
1868 ref_cache[-1 + 24] = LIST_NOT_USED;
1869 }
1870 }
1871 }
1872
1873 if (!USES_LIST(mb_type, list)) {
1874 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1875 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1876 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1877 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1878 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1879 return;
1880 }
1881
1882 {
1883 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1884 int (*ref2frm)[64] = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
1885 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1886 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1887 AV_WN32A(&ref_cache[0 * 8], ref01);
1888 AV_WN32A(&ref_cache[1 * 8], ref01);
1889 AV_WN32A(&ref_cache[2 * 8], ref23);
1890 AV_WN32A(&ref_cache[3 * 8], ref23);
1891 }
1892
1893 {
1894 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
1895 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1896 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1897 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1898 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1899 }
1900}
1901
1902/**
1903 *
1904 * @return non zero if the loop filter can be skipped
1905 */
1906static int fill_filter_caches(H264Context *h, int mb_type)
1907{
1908 const int mb_xy = h->mb_xy;
1909 int top_xy, left_xy[LEFT_MBS];
1910 int top_type, left_type[LEFT_MBS];
1911 uint8_t *nnz;
1912 uint8_t *nnz_cache;
1913
1914 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
1915
1916 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1917 * stuff, I can't imagine that these complex rules are worth it. */
1918
1919 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1920 if (FRAME_MBAFF(h)) {
1921 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1922 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1923 if (h->mb_y & 1) {
1924 if (left_mb_field_flag != curr_mb_field_flag)
1925 left_xy[LTOP] -= h->mb_stride;
1926 } else {
1927 if (curr_mb_field_flag)
1928 top_xy += h->mb_stride &
1929 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1930 if (left_mb_field_flag != curr_mb_field_flag)
1931 left_xy[LBOT] += h->mb_stride;
1932 }
1933 }
1934
1935 h->top_mb_xy = top_xy;
1936 h->left_mb_xy[LTOP] = left_xy[LTOP];
1937 h->left_mb_xy[LBOT] = left_xy[LBOT];
1938 {
1939 /* For sufficiently low qp, filtering wouldn't do anything.
1940 * This is a conservative estimate: could also check beta_offset
1941 * and more accurate chroma_qp. */
1942 int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1943 int qp = h->cur_pic.qscale_table[mb_xy];
1944 if (qp <= qp_thresh &&
1945 (left_xy[LTOP] < 0 ||
1946 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1947 (top_xy < 0 ||
1948 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1949 if (!FRAME_MBAFF(h))
1950 return 1;
1951 if ((left_xy[LTOP] < 0 ||
1952 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1953 (top_xy < h->mb_stride ||
1954 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1955 return 1;
1956 }
1957 }
1958
1959 top_type = h->cur_pic.mb_type[top_xy];
1960 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1961 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1962 if (h->deblocking_filter == 2) {
1963 if (h->slice_table[top_xy] != h->slice_num)
1964 top_type = 0;
1965 if (h->slice_table[left_xy[LBOT]] != h->slice_num)
1966 left_type[LTOP] = left_type[LBOT] = 0;
1967 } else {
1968 if (h->slice_table[top_xy] == 0xFFFF)
1969 top_type = 0;
1970 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1971 left_type[LTOP] = left_type[LBOT] = 0;
1972 }
1973 h->top_type = top_type;
1974 h->left_type[LTOP] = left_type[LTOP];
1975 h->left_type[LBOT] = left_type[LBOT];
1976
1977 if (IS_INTRA(mb_type))
1978 return 0;
1979
1980 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1981 top_type, left_type, mb_xy, 0);
1982 if (h->list_count == 2)
1983 fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
1984 top_type, left_type, mb_xy, 1);
1985
1986 nnz = h->non_zero_count[mb_xy];
1987 nnz_cache = h->non_zero_count_cache;
1988 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1989 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1990 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1991 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1992 h->cbp = h->cbp_table[mb_xy];
1993
1994 if (top_type) {
1995 nnz = h->non_zero_count[top_xy];
1996 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1997 }
1998
1999 if (left_type[LTOP]) {
2000 nnz = h->non_zero_count[left_xy[LTOP]];
2001 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2002 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2003 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2004 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2005 }
2006
2007 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2008 * from what the loop filter needs */
2009 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2010 if (IS_8x8DCT(top_type)) {
2011 nnz_cache[4 + 8 * 0] =
2012 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2013 nnz_cache[6 + 8 * 0] =
2014 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2015 }
2016 if (IS_8x8DCT(left_type[LTOP])) {
2017 nnz_cache[3 + 8 * 1] =
2018 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2019 }
2020 if (IS_8x8DCT(left_type[LBOT])) {
2021 nnz_cache[3 + 8 * 3] =
2022 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2023 }
2024
2025 if (IS_8x8DCT(mb_type)) {
2026 nnz_cache[scan8[0]] =
2027 nnz_cache[scan8[1]] =
2028 nnz_cache[scan8[2]] =
2029 nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2030
2031 nnz_cache[scan8[0 + 4]] =
2032 nnz_cache[scan8[1 + 4]] =
2033 nnz_cache[scan8[2 + 4]] =
2034 nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2035
2036 nnz_cache[scan8[0 + 8]] =
2037 nnz_cache[scan8[1 + 8]] =
2038 nnz_cache[scan8[2 + 8]] =
2039 nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2040
2041 nnz_cache[scan8[0 + 12]] =
2042 nnz_cache[scan8[1 + 12]] =
2043 nnz_cache[scan8[2 + 12]] =
2044 nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2045 }
2046 }
2047
2048 return 0;
2049}
2050
2051static void loop_filter(H264Context *h, int start_x, int end_x)
2052{
2053 uint8_t *dest_y, *dest_cb, *dest_cr;
2054 int linesize, uvlinesize, mb_x, mb_y;
2055 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2056 const int old_slice_type = h->slice_type;
2057 const int pixel_shift = h->pixel_shift;
2058 const int block_h = 16 >> h->chroma_y_shift;
2059
2060 if (h->deblocking_filter) {
2061 for (mb_x = start_x; mb_x < end_x; mb_x++)
2062 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2063 int mb_xy, mb_type;
2064 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2065 h->slice_num = h->slice_table[mb_xy];
2066 mb_type = h->cur_pic.mb_type[mb_xy];
2067 h->list_count = h->list_counts[mb_xy];
2068
2069 if (FRAME_MBAFF(h))
2070 h->mb_mbaff =
2071 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2072
2073 h->mb_x = mb_x;
2074 h->mb_y = mb_y;
2075 dest_y = h->cur_pic.f.data[0] +
2076 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2077 dest_cb = h->cur_pic.f.data[1] +
2078 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2079 mb_y * h->uvlinesize * block_h;
2080 dest_cr = h->cur_pic.f.data[2] +
2081 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2082 mb_y * h->uvlinesize * block_h;
2083 // FIXME simplify above
2084
2085 if (MB_FIELD(h)) {
2086 linesize = h->mb_linesize = h->linesize * 2;
2087 uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2088 if (mb_y & 1) { // FIXME move out of this function?
2089 dest_y -= h->linesize * 15;
2090 dest_cb -= h->uvlinesize * (block_h - 1);
2091 dest_cr -= h->uvlinesize * (block_h - 1);
2092 }
2093 } else {
2094 linesize = h->mb_linesize = h->linesize;
2095 uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2096 }
2097 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2098 uvlinesize, 0);
2099 if (fill_filter_caches(h, mb_type))
2100 continue;
2101 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2102 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2103
2104 if (FRAME_MBAFF(h)) {
2105 ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2106 linesize, uvlinesize);
2107 } else {
2108 ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2109 dest_cr, linesize, uvlinesize);
2110 }
2111 }
2112 }
2113 h->slice_type = old_slice_type;
2114 h->mb_x = end_x;
2115 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2116 h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2117 h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2118}
2119
2120static void predict_field_decoding_flag(H264Context *h)
2121{
2122 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2123 int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2124 h->cur_pic.mb_type[mb_xy - 1] :
2125 (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2126 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2127 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2128}
2129
2130/**
2131 * Draw edges and report progress for the last MB row.
2132 */
2133static void decode_finish_row(H264Context *h)
2134{
2135 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2136 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2137 int height = 16 << FRAME_MBAFF(h);
2138 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2139
2140 if (h->deblocking_filter) {
2141 if ((top + height) >= pic_height)
2142 height += deblock_border;
2143 top -= deblock_border;
2144 }
2145
2146 if (top >= pic_height || (top + height) < 0)
2147 return;
2148
2149 height = FFMIN(height, pic_height - top);
2150 if (top < 0) {
2151 height = top + height;
2152 top = 0;
2153 }
2154
2155 ff_h264_draw_horiz_band(h, top, height);
2156
2157 if (h->droppable)
2158 return;
2159
2160 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2161 h->picture_structure == PICT_BOTTOM_FIELD);
2162}
2163
2164static void er_add_slice(H264Context *h, int startx, int starty,
2165 int endx, int endy, int status)
2166{
2167#if CONFIG_ERROR_RESILIENCE
2168 ERContext *er = &h->er;
2169
2170 er->ref_count = h->ref_count[0];
2171 ff_er_add_slice(er, startx, starty, endx, endy, status);
2172#endif
2173}
2174
2175static int decode_slice(struct AVCodecContext *avctx, void *arg)
2176{
2177 H264Context *h = *(void **)arg;
2178 int lf_x_start = h->mb_x;
2179
2180 h->mb_skip_run = -1;
2181
2182 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2183 avctx->codec_id != AV_CODEC_ID_H264 ||
2184 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2185
2186 if (h->pps.cabac) {
2187 /* realign */
2188 align_get_bits(&h->gb);
2189
2190 /* init cabac */
2191 ff_init_cabac_decoder(&h->cabac,
2192 h->gb.buffer + get_bits_count(&h->gb) / 8,
2193 (get_bits_left(&h->gb) + 7) / 8);
2194
2195 ff_h264_init_cabac_states(h);
2196
2197 for (;;) {
2198 // START_TIMER
2199 int ret = ff_h264_decode_mb_cabac(h);
2200 int eos;
2201 // STOP_TIMER("decode_mb_cabac")
2202
2203 if (ret >= 0)
2204 ff_h264_hl_decode_mb(h);
2205
2206 // FIXME optimal? or let mb_decode decode 16x32 ?
2207 if (ret >= 0 && FRAME_MBAFF(h)) {
2208 h->mb_y++;
2209
2210 ret = ff_h264_decode_mb_cabac(h);
2211
2212 if (ret >= 0)
2213 ff_h264_hl_decode_mb(h);
2214 h->mb_y--;
2215 }
2216 eos = get_cabac_terminate(&h->cabac);
2217
2218 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2219 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2220 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2221 h->mb_y, ER_MB_END);
2222 if (h->mb_x >= lf_x_start)
2223 loop_filter(h, lf_x_start, h->mb_x + 1);
2224 return 0;
2225 }
2226 if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2227 av_log(h->avctx, AV_LOG_ERROR,
2228 "error while decoding MB %d %d, bytestream %td\n",
2229 h->mb_x, h->mb_y,
2230 h->cabac.bytestream_end - h->cabac.bytestream);
2231 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2232 h->mb_y, ER_MB_ERROR);
2233 return AVERROR_INVALIDDATA;
2234 }
2235
2236 if (++h->mb_x >= h->mb_width) {
2237 loop_filter(h, lf_x_start, h->mb_x);
2238 h->mb_x = lf_x_start = 0;
2239 decode_finish_row(h);
2240 ++h->mb_y;
2241 if (FIELD_OR_MBAFF_PICTURE(h)) {
2242 ++h->mb_y;
2243 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2244 predict_field_decoding_flag(h);
2245 }
2246 }
2247
2248 if (eos || h->mb_y >= h->mb_height) {
2249 tprintf(h->avctx, "slice end %d %d\n",
2250 get_bits_count(&h->gb), h->gb.size_in_bits);
2251 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2252 h->mb_y, ER_MB_END);
2253 if (h->mb_x > lf_x_start)
2254 loop_filter(h, lf_x_start, h->mb_x);
2255 return 0;
2256 }
2257 }
2258 } else {
2259 for (;;) {
2260 int ret = ff_h264_decode_mb_cavlc(h);
2261
2262 if (ret >= 0)
2263 ff_h264_hl_decode_mb(h);
2264
2265 // FIXME optimal? or let mb_decode decode 16x32 ?
2266 if (ret >= 0 && FRAME_MBAFF(h)) {
2267 h->mb_y++;
2268 ret = ff_h264_decode_mb_cavlc(h);
2269
2270 if (ret >= 0)
2271 ff_h264_hl_decode_mb(h);
2272 h->mb_y--;
2273 }
2274
2275 if (ret < 0) {
2276 av_log(h->avctx, AV_LOG_ERROR,
2277 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2278 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2279 h->mb_y, ER_MB_ERROR);
2280 return ret;
2281 }
2282
2283 if (++h->mb_x >= h->mb_width) {
2284 loop_filter(h, lf_x_start, h->mb_x);
2285 h->mb_x = lf_x_start = 0;
2286 decode_finish_row(h);
2287 ++h->mb_y;
2288 if (FIELD_OR_MBAFF_PICTURE(h)) {
2289 ++h->mb_y;
2290 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2291 predict_field_decoding_flag(h);
2292 }
2293 if (h->mb_y >= h->mb_height) {
2294 tprintf(h->avctx, "slice end %d %d\n",
2295 get_bits_count(&h->gb), h->gb.size_in_bits);
2296
2297 if (get_bits_left(&h->gb) == 0) {
2298 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2299 h->mb_x - 1, h->mb_y,
2300 ER_MB_END);
2301
2302 return 0;
2303 } else {
2304 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2305 h->mb_x - 1, h->mb_y,
2306 ER_MB_END);
2307
2308 return AVERROR_INVALIDDATA;
2309 }
2310 }
2311 }
2312
2313 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2314 tprintf(h->avctx, "slice end %d %d\n",
2315 get_bits_count(&h->gb), h->gb.size_in_bits);
2316
2317 if (get_bits_left(&h->gb) == 0) {
2318 er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2319 h->mb_x - 1, h->mb_y,
2320 ER_MB_END);
2321 if (h->mb_x > lf_x_start)
2322 loop_filter(h, lf_x_start, h->mb_x);
2323
2324 return 0;
2325 } else {
2326 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2327 h->mb_y, ER_MB_ERROR);
2328
2329 return AVERROR_INVALIDDATA;
2330 }
2331 }
2332 }
2333 }
2334}
2335
2336/**
2337 * Call decode_slice() for each context.
2338 *
2339 * @param h h264 master context
2340 * @param context_count number of contexts to execute
2341 */
2342int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2343{
2344 AVCodecContext *const avctx = h->avctx;
2345 H264Context *hx;
2346 int i;
2347
2348 if (h->mb_y >= h->mb_height) {
2349 av_log(h->avctx, AV_LOG_ERROR,
2350 "Input contains more MB rows than the frame height.\n");
2351 return AVERROR_INVALIDDATA;
2352 }
2353
2354 if (h->avctx->hwaccel)
2355 return 0;
2356 if (context_count == 1) {
2357 return decode_slice(avctx, &h);
2358 } else {
2359 for (i = 1; i < context_count; i++) {
2360 hx = h->thread_context[i];
2361 hx->er.error_count = 0;
2362 }
2363
2364 avctx->execute(avctx, decode_slice, h->thread_context,
2365 NULL, context_count, sizeof(void *));
2366
2367 /* pull back stuff from slices to master context */
2368 hx = h->thread_context[context_count - 1];
2369 h->mb_x = hx->mb_x;
2370 h->mb_y = hx->mb_y;
2371 h->droppable = hx->droppable;
2372 h->picture_structure = hx->picture_structure;
2373 for (i = 1; i < context_count; i++)
2374 h->er.error_count += h->thread_context[i]->er.error_count;
2375 }
2376
2377 return 0;
2378}