9d012c5517d8d06029d6f36245fa1b68769a7cbd
[libav.git] / libavcodec / h264.c
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 "internal.h"
31 #include "cabac.h"
32 #include "cabac_functions.h"
33 #include "dsputil.h"
34 #include "error_resilience.h"
35 #include "avcodec.h"
36 #include "mpegvideo.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 "rectangle.h"
44 #include "svq3.h"
45 #include "thread.h"
46
47 #include <assert.h>
48
49 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
50
51 static const uint8_t rem6[QP_MAX_NUM + 1] = {
52 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
53 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
54 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
55 };
56
57 static const uint8_t div6[QP_MAX_NUM + 1] = {
58 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
59 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
60 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
61 };
62
63 static const uint8_t field_scan[16] = {
64 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
65 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
66 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
67 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
68 };
69
70 static const uint8_t field_scan8x8[64] = {
71 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
72 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
73 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
74 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
75 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
76 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
77 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
78 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
79 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
80 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
81 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
82 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
83 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
84 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
85 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
86 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
87 };
88
89 static const uint8_t field_scan8x8_cavlc[64] = {
90 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
91 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
92 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
93 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
94 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
95 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
96 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
97 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
98 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
99 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
100 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
101 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
102 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
103 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
104 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
105 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
106 };
107
108 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
109 static const uint8_t zigzag_scan8x8_cavlc[64] = {
110 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
111 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
112 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
113 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
114 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
115 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
116 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
117 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
118 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
119 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
120 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
121 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
122 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
123 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
124 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
125 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
126 };
127
128 static const uint8_t dequant4_coeff_init[6][3] = {
129 { 10, 13, 16 },
130 { 11, 14, 18 },
131 { 13, 16, 20 },
132 { 14, 18, 23 },
133 { 16, 20, 25 },
134 { 18, 23, 29 },
135 };
136
137 static const uint8_t dequant8_coeff_init_scan[16] = {
138 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
139 };
140
141 static const uint8_t dequant8_coeff_init[6][6] = {
142 { 20, 18, 32, 19, 25, 24 },
143 { 22, 19, 35, 21, 28, 26 },
144 { 26, 23, 42, 24, 33, 31 },
145 { 28, 25, 45, 26, 35, 33 },
146 { 32, 28, 51, 30, 40, 38 },
147 { 36, 32, 58, 34, 46, 43 },
148 };
149
150 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
151 #if CONFIG_H264_DXVA2_HWACCEL
152 AV_PIX_FMT_DXVA2_VLD,
153 #endif
154 #if CONFIG_H264_VAAPI_HWACCEL
155 AV_PIX_FMT_VAAPI_VLD,
156 #endif
157 #if CONFIG_H264_VDA_HWACCEL
158 AV_PIX_FMT_VDA_VLD,
159 #endif
160 #if CONFIG_H264_VDPAU_HWACCEL
161 AV_PIX_FMT_VDPAU,
162 #endif
163 AV_PIX_FMT_YUV420P,
164 AV_PIX_FMT_NONE
165 };
166
167 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
168 #if CONFIG_H264_DXVA2_HWACCEL
169 AV_PIX_FMT_DXVA2_VLD,
170 #endif
171 #if CONFIG_H264_VAAPI_HWACCEL
172 AV_PIX_FMT_VAAPI_VLD,
173 #endif
174 #if CONFIG_H264_VDA_HWACCEL
175 AV_PIX_FMT_VDA_VLD,
176 #endif
177 #if CONFIG_H264_VDPAU_HWACCEL
178 AV_PIX_FMT_VDPAU,
179 #endif
180 AV_PIX_FMT_YUVJ420P,
181 AV_PIX_FMT_NONE
182 };
183
184 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
185 int (*mv)[2][4][2],
186 int mb_x, int mb_y, int mb_intra, int mb_skipped)
187 {
188 H264Context *h = opaque;
189
190 h->mb_x = mb_x;
191 h->mb_y = mb_y;
192 h->mb_xy = mb_x + mb_y * h->mb_stride;
193 memset(h->non_zero_count_cache, 0, sizeof(h->non_zero_count_cache));
194 assert(ref >= 0);
195 /* FIXME: It is possible albeit uncommon that slice references
196 * differ between slices. We take the easy approach and ignore
197 * it for now. If this turns out to have any relevance in
198 * practice then correct remapping should be added. */
199 if (ref >= h->ref_count[0])
200 ref = 0;
201 fill_rectangle(&h->cur_pic.ref_index[0][4 * h->mb_xy],
202 2, 2, 2, ref, 1);
203 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
204 fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8,
205 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
206 assert(!FRAME_MBAFF(h));
207 ff_h264_hl_decode_mb(h);
208 }
209
210 void ff_h264_draw_horiz_band(H264Context *h, int y, int height)
211 {
212 AVCodecContext *avctx = h->avctx;
213 Picture *cur = &h->cur_pic;
214 Picture *last = h->ref_list[0][0].f.data[0] ? &h->ref_list[0][0] : NULL;
215 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
216 int vshift = desc->log2_chroma_h;
217 const int field_pic = h->picture_structure != PICT_FRAME;
218 if (field_pic) {
219 height <<= 1;
220 y <<= 1;
221 }
222
223 height = FFMIN(height, avctx->height - y);
224
225 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
226 return;
227
228 if (avctx->draw_horiz_band) {
229 AVFrame *src;
230 int offset[AV_NUM_DATA_POINTERS];
231 int i;
232
233 if (cur->f.pict_type == AV_PICTURE_TYPE_B || h->low_delay ||
234 (avctx->slice_flags & SLICE_FLAG_CODED_ORDER))
235 src = &cur->f;
236 else if (last)
237 src = &last->f;
238 else
239 return;
240
241 offset[0] = y * src->linesize[0];
242 offset[1] =
243 offset[2] = (y >> vshift) * src->linesize[1];
244 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
245 offset[i] = 0;
246
247 emms_c();
248
249 avctx->draw_horiz_band(avctx, src, offset,
250 y, h->picture_structure, height);
251 }
252 }
253
254 static void unref_picture(H264Context *h, Picture *pic)
255 {
256 int off = offsetof(Picture, tf) + sizeof(pic->tf);
257 int i;
258
259 if (!pic->f.data[0])
260 return;
261
262 ff_thread_release_buffer(h->avctx, &pic->tf);
263 av_buffer_unref(&pic->hwaccel_priv_buf);
264
265 av_buffer_unref(&pic->qscale_table_buf);
266 av_buffer_unref(&pic->mb_type_buf);
267 for (i = 0; i < 2; i++) {
268 av_buffer_unref(&pic->motion_val_buf[i]);
269 av_buffer_unref(&pic->ref_index_buf[i]);
270 }
271
272 memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
273 }
274
275 static void release_unused_pictures(H264Context *h, int remove_current)
276 {
277 int i;
278
279 /* release non reference frames */
280 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
281 if (h->DPB[i].f.data[0] && !h->DPB[i].reference &&
282 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
283 unref_picture(h, &h->DPB[i]);
284 }
285 }
286 }
287
288 static int ref_picture(H264Context *h, Picture *dst, Picture *src)
289 {
290 int ret, i;
291
292 av_assert0(!dst->f.buf[0]);
293 av_assert0(src->f.buf[0]);
294
295 src->tf.f = &src->f;
296 dst->tf.f = &dst->f;
297 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
298 if (ret < 0)
299 goto fail;
300
301 dst->qscale_table_buf = av_buffer_ref(src->qscale_table_buf);
302 dst->mb_type_buf = av_buffer_ref(src->mb_type_buf);
303 if (!dst->qscale_table_buf || !dst->mb_type_buf)
304 goto fail;
305 dst->qscale_table = src->qscale_table;
306 dst->mb_type = src->mb_type;
307
308 for (i = 0; i < 2; i++) {
309 dst->motion_val_buf[i] = av_buffer_ref(src->motion_val_buf[i]);
310 dst->ref_index_buf[i] = av_buffer_ref(src->ref_index_buf[i]);
311 if (!dst->motion_val_buf[i] || !dst->ref_index_buf[i])
312 goto fail;
313 dst->motion_val[i] = src->motion_val[i];
314 dst->ref_index[i] = src->ref_index[i];
315 }
316
317 if (src->hwaccel_picture_private) {
318 dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
319 if (!dst->hwaccel_priv_buf)
320 goto fail;
321 dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
322 }
323
324 for (i = 0; i < 2; i++)
325 dst->field_poc[i] = src->field_poc[i];
326
327 memcpy(dst->ref_poc, src->ref_poc, sizeof(src->ref_poc));
328 memcpy(dst->ref_count, src->ref_count, sizeof(src->ref_count));
329
330 dst->poc = src->poc;
331 dst->frame_num = src->frame_num;
332 dst->mmco_reset = src->mmco_reset;
333 dst->pic_id = src->pic_id;
334 dst->long_ref = src->long_ref;
335 dst->mbaff = src->mbaff;
336 dst->field_picture = src->field_picture;
337 dst->needs_realloc = src->needs_realloc;
338 dst->reference = src->reference;
339
340 return 0;
341 fail:
342 unref_picture(h, dst);
343 return ret;
344 }
345
346 static int alloc_scratch_buffers(H264Context *h, int linesize)
347 {
348 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
349
350 if (h->bipred_scratchpad)
351 return 0;
352
353 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
354 // edge emu needs blocksize + filter length - 1
355 // (= 21x21 for h264)
356 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
357 h->me.scratchpad = av_mallocz(alloc_size * 2 * 16 * 2);
358
359 if (!h->bipred_scratchpad || !h->edge_emu_buffer || !h->me.scratchpad) {
360 av_freep(&h->bipred_scratchpad);
361 av_freep(&h->edge_emu_buffer);
362 av_freep(&h->me.scratchpad);
363 return AVERROR(ENOMEM);
364 }
365
366 h->me.temp = h->me.scratchpad;
367
368 return 0;
369 }
370
371 static int init_table_pools(H264Context *h)
372 {
373 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
374 const int mb_array_size = h->mb_stride * h->mb_height;
375 const int b4_stride = h->mb_width * 4 + 1;
376 const int b4_array_size = b4_stride * h->mb_height * 4;
377
378 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
379 av_buffer_allocz);
380 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
381 sizeof(uint32_t), av_buffer_allocz);
382 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
383 sizeof(int16_t), av_buffer_allocz);
384 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
385
386 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
387 !h->ref_index_pool) {
388 av_buffer_pool_uninit(&h->qscale_table_pool);
389 av_buffer_pool_uninit(&h->mb_type_pool);
390 av_buffer_pool_uninit(&h->motion_val_pool);
391 av_buffer_pool_uninit(&h->ref_index_pool);
392 return AVERROR(ENOMEM);
393 }
394
395 return 0;
396 }
397
398 static int alloc_picture(H264Context *h, Picture *pic)
399 {
400 int i, ret = 0;
401
402 av_assert0(!pic->f.data[0]);
403
404 pic->tf.f = &pic->f;
405 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
406 AV_GET_BUFFER_FLAG_REF : 0);
407 if (ret < 0)
408 goto fail;
409
410 h->linesize = pic->f.linesize[0];
411 h->uvlinesize = pic->f.linesize[1];
412
413 if (h->avctx->hwaccel) {
414 const AVHWAccel *hwaccel = h->avctx->hwaccel;
415 av_assert0(!pic->hwaccel_picture_private);
416 if (hwaccel->priv_data_size) {
417 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size);
418 if (!pic->hwaccel_priv_buf)
419 return AVERROR(ENOMEM);
420 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
421 }
422 }
423
424 if (!h->qscale_table_pool) {
425 ret = init_table_pools(h);
426 if (ret < 0)
427 goto fail;
428 }
429
430 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
431 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
432 if (!pic->qscale_table_buf || !pic->mb_type_buf)
433 goto fail;
434
435 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
436 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
437
438 for (i = 0; i < 2; i++) {
439 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
440 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
441 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
442 goto fail;
443
444 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
445 pic->ref_index[i] = pic->ref_index_buf[i]->data;
446 }
447
448 return 0;
449 fail:
450 unref_picture(h, pic);
451 return (ret < 0) ? ret : AVERROR(ENOMEM);
452 }
453
454 static inline int pic_is_unused(H264Context *h, Picture *pic)
455 {
456 if (pic->f.data[0] == NULL)
457 return 1;
458 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
459 return 1;
460 return 0;
461 }
462
463 static int find_unused_picture(H264Context *h)
464 {
465 int i;
466
467 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
468 if (pic_is_unused(h, &h->DPB[i]))
469 break;
470 }
471 if (i == MAX_PICTURE_COUNT)
472 return AVERROR_INVALIDDATA;
473
474 if (h->DPB[i].needs_realloc) {
475 h->DPB[i].needs_realloc = 0;
476 unref_picture(h, &h->DPB[i]);
477 }
478
479 return i;
480 }
481
482 /**
483 * Check if the top & left blocks are available if needed and
484 * change the dc mode so it only uses the available blocks.
485 */
486 int ff_h264_check_intra4x4_pred_mode(H264Context *h)
487 {
488 static const int8_t top[12] = {
489 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
490 };
491 static const int8_t left[12] = {
492 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
493 };
494 int i;
495
496 if (!(h->top_samples_available & 0x8000)) {
497 for (i = 0; i < 4; i++) {
498 int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
499 if (status < 0) {
500 av_log(h->avctx, AV_LOG_ERROR,
501 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
502 status, h->mb_x, h->mb_y);
503 return AVERROR_INVALIDDATA;
504 } else if (status) {
505 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
506 }
507 }
508 }
509
510 if ((h->left_samples_available & 0x8888) != 0x8888) {
511 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
512 for (i = 0; i < 4; i++)
513 if (!(h->left_samples_available & mask[i])) {
514 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
515 if (status < 0) {
516 av_log(h->avctx, AV_LOG_ERROR,
517 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
518 status, h->mb_x, h->mb_y);
519 return AVERROR_INVALIDDATA;
520 } else if (status) {
521 h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
522 }
523 }
524 }
525
526 return 0;
527 } // FIXME cleanup like ff_h264_check_intra_pred_mode
528
529 /**
530 * Check if the top & left blocks are available if needed and
531 * change the dc mode so it only uses the available blocks.
532 */
533 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
534 {
535 static const int8_t top[7] = { LEFT_DC_PRED8x8, 1, -1, -1 };
536 static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
537
538 if (mode > 6U) {
539 av_log(h->avctx, AV_LOG_ERROR,
540 "out of range intra chroma pred mode at %d %d\n",
541 h->mb_x, h->mb_y);
542 return AVERROR_INVALIDDATA;
543 }
544
545 if (!(h->top_samples_available & 0x8000)) {
546 mode = top[mode];
547 if (mode < 0) {
548 av_log(h->avctx, AV_LOG_ERROR,
549 "top block unavailable for requested intra mode at %d %d\n",
550 h->mb_x, h->mb_y);
551 return AVERROR_INVALIDDATA;
552 }
553 }
554
555 if ((h->left_samples_available & 0x8080) != 0x8080) {
556 mode = left[mode];
557 if (is_chroma && (h->left_samples_available & 0x8080)) {
558 // mad cow disease mode, aka MBAFF + constrained_intra_pred
559 mode = ALZHEIMER_DC_L0T_PRED8x8 +
560 (!(h->left_samples_available & 0x8000)) +
561 2 * (mode == DC_128_PRED8x8);
562 }
563 if (mode < 0) {
564 av_log(h->avctx, AV_LOG_ERROR,
565 "left block unavailable for requested intra mode at %d %d\n",
566 h->mb_x, h->mb_y);
567 return AVERROR_INVALIDDATA;
568 }
569 }
570
571 return mode;
572 }
573
574 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
575 int *dst_length, int *consumed, int length)
576 {
577 int i, si, di;
578 uint8_t *dst;
579 int bufidx;
580
581 // src[0]&0x80; // forbidden bit
582 h->nal_ref_idc = src[0] >> 5;
583 h->nal_unit_type = src[0] & 0x1F;
584
585 src++;
586 length--;
587
588 #define STARTCODE_TEST \
589 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
590 if (src[i + 2] != 3) { \
591 /* startcode, so we must be past the end */ \
592 length = i; \
593 } \
594 break; \
595 }
596
597 #if HAVE_FAST_UNALIGNED
598 #define FIND_FIRST_ZERO \
599 if (i > 0 && !src[i]) \
600 i--; \
601 while (src[i]) \
602 i++
603
604 #if HAVE_FAST_64BIT
605 for (i = 0; i + 1 < length; i += 9) {
606 if (!((~AV_RN64A(src + i) &
607 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
608 0x8000800080008080ULL))
609 continue;
610 FIND_FIRST_ZERO;
611 STARTCODE_TEST;
612 i -= 7;
613 }
614 #else
615 for (i = 0; i + 1 < length; i += 5) {
616 if (!((~AV_RN32A(src + i) &
617 (AV_RN32A(src + i) - 0x01000101U)) &
618 0x80008080U))
619 continue;
620 FIND_FIRST_ZERO;
621 STARTCODE_TEST;
622 i -= 3;
623 }
624 #endif
625 #else
626 for (i = 0; i + 1 < length; i += 2) {
627 if (src[i])
628 continue;
629 if (i > 0 && src[i - 1] == 0)
630 i--;
631 STARTCODE_TEST;
632 }
633 #endif
634
635 if (i >= length - 1) { // no escaped 0
636 *dst_length = length;
637 *consumed = length + 1; // +1 for the header
638 return src;
639 }
640
641 // use second escape buffer for inter data
642 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
643 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx],
644 length + FF_INPUT_BUFFER_PADDING_SIZE);
645 dst = h->rbsp_buffer[bufidx];
646
647 if (dst == NULL)
648 return NULL;
649
650 memcpy(dst, src, i);
651 si = di = i;
652 while (si + 2 < length) {
653 // remove escapes (very rare 1:2^22)
654 if (src[si + 2] > 3) {
655 dst[di++] = src[si++];
656 dst[di++] = src[si++];
657 } else if (src[si] == 0 && src[si + 1] == 0) {
658 if (src[si + 2] == 3) { // escape
659 dst[di++] = 0;
660 dst[di++] = 0;
661 si += 3;
662 continue;
663 } else // next start code
664 goto nsc;
665 }
666
667 dst[di++] = src[si++];
668 }
669 while (si < length)
670 dst[di++] = src[si++];
671
672 nsc:
673 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
674
675 *dst_length = di;
676 *consumed = si + 1; // +1 for the header
677 /* FIXME store exact number of bits in the getbitcontext
678 * (it is needed for decoding) */
679 return dst;
680 }
681
682 /**
683 * Identify the exact end of the bitstream
684 * @return the length of the trailing, or 0 if damaged
685 */
686 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
687 {
688 int v = *src;
689 int r;
690
691 tprintf(h->avctx, "rbsp trailing %X\n", v);
692
693 for (r = 1; r < 9; r++) {
694 if (v & 1)
695 return r;
696 v >>= 1;
697 }
698 return 0;
699 }
700
701 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
702 int height, int y_offset, int list)
703 {
704 int raw_my = h->mv_cache[list][scan8[n]][1];
705 int filter_height_up = (raw_my & 3) ? 2 : 0;
706 int filter_height_down = (raw_my & 3) ? 3 : 0;
707 int full_my = (raw_my >> 2) + y_offset;
708 int top = full_my - filter_height_up;
709 int bottom = full_my + filter_height_down + height;
710
711 return FFMAX(abs(top), bottom);
712 }
713
714 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
715 int height, int y_offset, int list0,
716 int list1, int *nrefs)
717 {
718 int my;
719
720 y_offset += 16 * (h->mb_y >> MB_FIELD(h));
721
722 if (list0) {
723 int ref_n = h->ref_cache[0][scan8[n]];
724 Picture *ref = &h->ref_list[0][ref_n];
725
726 // Error resilience puts the current picture in the ref list.
727 // Don't try to wait on these as it will cause a deadlock.
728 // Fields can wait on each other, though.
729 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
730 (ref->reference & 3) != h->picture_structure) {
731 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
732 if (refs[0][ref_n] < 0)
733 nrefs[0] += 1;
734 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
735 }
736 }
737
738 if (list1) {
739 int ref_n = h->ref_cache[1][scan8[n]];
740 Picture *ref = &h->ref_list[1][ref_n];
741
742 if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
743 (ref->reference & 3) != h->picture_structure) {
744 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
745 if (refs[1][ref_n] < 0)
746 nrefs[1] += 1;
747 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
748 }
749 }
750 }
751
752 /**
753 * Wait until all reference frames are available for MC operations.
754 *
755 * @param h the H264 context
756 */
757 static void await_references(H264Context *h)
758 {
759 const int mb_xy = h->mb_xy;
760 const int mb_type = h->cur_pic.mb_type[mb_xy];
761 int refs[2][48];
762 int nrefs[2] = { 0 };
763 int ref, list;
764
765 memset(refs, -1, sizeof(refs));
766
767 if (IS_16X16(mb_type)) {
768 get_lowest_part_y(h, refs, 0, 16, 0,
769 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
770 } else if (IS_16X8(mb_type)) {
771 get_lowest_part_y(h, refs, 0, 8, 0,
772 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
773 get_lowest_part_y(h, refs, 8, 8, 8,
774 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
775 } else if (IS_8X16(mb_type)) {
776 get_lowest_part_y(h, refs, 0, 16, 0,
777 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
778 get_lowest_part_y(h, refs, 4, 16, 0,
779 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
780 } else {
781 int i;
782
783 assert(IS_8X8(mb_type));
784
785 for (i = 0; i < 4; i++) {
786 const int sub_mb_type = h->sub_mb_type[i];
787 const int n = 4 * i;
788 int y_offset = (i & 2) << 2;
789
790 if (IS_SUB_8X8(sub_mb_type)) {
791 get_lowest_part_y(h, refs, n, 8, y_offset,
792 IS_DIR(sub_mb_type, 0, 0),
793 IS_DIR(sub_mb_type, 0, 1),
794 nrefs);
795 } else if (IS_SUB_8X4(sub_mb_type)) {
796 get_lowest_part_y(h, refs, n, 4, y_offset,
797 IS_DIR(sub_mb_type, 0, 0),
798 IS_DIR(sub_mb_type, 0, 1),
799 nrefs);
800 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
801 IS_DIR(sub_mb_type, 0, 0),
802 IS_DIR(sub_mb_type, 0, 1),
803 nrefs);
804 } else if (IS_SUB_4X8(sub_mb_type)) {
805 get_lowest_part_y(h, refs, n, 8, y_offset,
806 IS_DIR(sub_mb_type, 0, 0),
807 IS_DIR(sub_mb_type, 0, 1),
808 nrefs);
809 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
810 IS_DIR(sub_mb_type, 0, 0),
811 IS_DIR(sub_mb_type, 0, 1),
812 nrefs);
813 } else {
814 int j;
815 assert(IS_SUB_4X4(sub_mb_type));
816 for (j = 0; j < 4; j++) {
817 int sub_y_offset = y_offset + 2 * (j & 2);
818 get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
819 IS_DIR(sub_mb_type, 0, 0),
820 IS_DIR(sub_mb_type, 0, 1),
821 nrefs);
822 }
823 }
824 }
825 }
826
827 for (list = h->list_count - 1; list >= 0; list--)
828 for (ref = 0; ref < 48 && nrefs[list]; ref++) {
829 int row = refs[list][ref];
830 if (row >= 0) {
831 Picture *ref_pic = &h->ref_list[list][ref];
832 int ref_field = ref_pic->reference - 1;
833 int ref_field_picture = ref_pic->field_picture;
834 int pic_height = 16 * h->mb_height >> ref_field_picture;
835
836 row <<= MB_MBAFF(h);
837 nrefs[list]--;
838
839 if (!FIELD_PICTURE(h) && ref_field_picture) { // frame referencing two fields
840 ff_thread_await_progress(&ref_pic->tf,
841 FFMIN((row >> 1) - !(row & 1),
842 pic_height - 1),
843 1);
844 ff_thread_await_progress(&ref_pic->tf,
845 FFMIN((row >> 1), pic_height - 1),
846 0);
847 } else if (FIELD_PICTURE(h) && !ref_field_picture) { // field referencing one field of a frame
848 ff_thread_await_progress(&ref_pic->tf,
849 FFMIN(row * 2 + ref_field,
850 pic_height - 1),
851 0);
852 } else if (FIELD_PICTURE(h)) {
853 ff_thread_await_progress(&ref_pic->tf,
854 FFMIN(row, pic_height - 1),
855 ref_field);
856 } else {
857 ff_thread_await_progress(&ref_pic->tf,
858 FFMIN(row, pic_height - 1),
859 0);
860 }
861 }
862 }
863 }
864
865 static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
866 int n, int square, int height,
867 int delta, int list,
868 uint8_t *dest_y, uint8_t *dest_cb,
869 uint8_t *dest_cr,
870 int src_x_offset, int src_y_offset,
871 qpel_mc_func *qpix_op,
872 h264_chroma_mc_func chroma_op,
873 int pixel_shift, int chroma_idc)
874 {
875 const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
876 int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
877 const int luma_xy = (mx & 3) + ((my & 3) << 2);
878 int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
879 uint8_t *src_y = pic->f.data[0] + offset;
880 uint8_t *src_cb, *src_cr;
881 int extra_width = 0;
882 int extra_height = 0;
883 int emu = 0;
884 const int full_mx = mx >> 2;
885 const int full_my = my >> 2;
886 const int pic_width = 16 * h->mb_width;
887 const int pic_height = 16 * h->mb_height >> MB_FIELD(h);
888 int ysh;
889
890 if (mx & 7)
891 extra_width -= 3;
892 if (my & 7)
893 extra_height -= 3;
894
895 if (full_mx < 0 - extra_width ||
896 full_my < 0 - extra_height ||
897 full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
898 full_my + 16 /*FIXME*/ > pic_height + extra_height) {
899 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
900 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
901 h->mb_linesize,
902 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
903 full_my - 2, pic_width, pic_height);
904 src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
905 emu = 1;
906 }
907
908 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
909 if (!square)
910 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
911
912 if (CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY)
913 return;
914
915 if (chroma_idc == 3 /* yuv444 */) {
916 src_cb = pic->f.data[1] + offset;
917 if (emu) {
918 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
919 src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
920 h->mb_linesize,
921 16 + 5, 16 + 5 /*FIXME*/,
922 full_mx - 2, full_my - 2,
923 pic_width, pic_height);
924 src_cb = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
925 }
926 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
927 if (!square)
928 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
929
930 src_cr = pic->f.data[2] + offset;
931 if (emu) {
932 h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
933 src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
934 h->mb_linesize,
935 16 + 5, 16 + 5 /*FIXME*/,
936 full_mx - 2, full_my - 2,
937 pic_width, pic_height);
938 src_cr = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
939 }
940 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
941 if (!square)
942 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
943 return;
944 }
945
946 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
947 if (chroma_idc == 1 /* yuv420 */ && MB_FIELD(h)) {
948 // chroma offset when predicting from a field of opposite parity
949 my += 2 * ((h->mb_y & 1) - (pic->reference - 1));
950 emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
951 }
952
953 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
954 (my >> ysh) * h->mb_uvlinesize;
955 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
956 (my >> ysh) * h->mb_uvlinesize;
957
958 if (emu) {
959 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->mb_uvlinesize,
960 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
961 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
962 src_cb = h->edge_emu_buffer;
963 }
964 chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
965 height >> (chroma_idc == 1 /* yuv420 */),
966 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
967
968 if (emu) {
969 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->mb_uvlinesize,
970 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
971 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
972 src_cr = h->edge_emu_buffer;
973 }
974 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
975 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
976 }
977
978 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
979 int height, int delta,
980 uint8_t *dest_y, uint8_t *dest_cb,
981 uint8_t *dest_cr,
982 int x_offset, int y_offset,
983 qpel_mc_func *qpix_put,
984 h264_chroma_mc_func chroma_put,
985 qpel_mc_func *qpix_avg,
986 h264_chroma_mc_func chroma_avg,
987 int list0, int list1,
988 int pixel_shift, int chroma_idc)
989 {
990 qpel_mc_func *qpix_op = qpix_put;
991 h264_chroma_mc_func chroma_op = chroma_put;
992
993 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
994 if (chroma_idc == 3 /* yuv444 */) {
995 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
996 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
997 } else if (chroma_idc == 2 /* yuv422 */) {
998 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
999 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1000 } else { /* yuv420 */
1001 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1002 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1003 }
1004 x_offset += 8 * h->mb_x;
1005 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
1006
1007 if (list0) {
1008 Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
1009 mc_dir_part(h, ref, n, square, height, delta, 0,
1010 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1011 qpix_op, chroma_op, pixel_shift, chroma_idc);
1012
1013 qpix_op = qpix_avg;
1014 chroma_op = chroma_avg;
1015 }
1016
1017 if (list1) {
1018 Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
1019 mc_dir_part(h, ref, n, square, height, delta, 1,
1020 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1021 qpix_op, chroma_op, pixel_shift, chroma_idc);
1022 }
1023 }
1024
1025 static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
1026 int height, int delta,
1027 uint8_t *dest_y, uint8_t *dest_cb,
1028 uint8_t *dest_cr,
1029 int x_offset, int y_offset,
1030 qpel_mc_func *qpix_put,
1031 h264_chroma_mc_func chroma_put,
1032 h264_weight_func luma_weight_op,
1033 h264_weight_func chroma_weight_op,
1034 h264_biweight_func luma_weight_avg,
1035 h264_biweight_func chroma_weight_avg,
1036 int list0, int list1,
1037 int pixel_shift, int chroma_idc)
1038 {
1039 int chroma_height;
1040
1041 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1042 if (chroma_idc == 3 /* yuv444 */) {
1043 chroma_height = height;
1044 chroma_weight_avg = luma_weight_avg;
1045 chroma_weight_op = luma_weight_op;
1046 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1047 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
1048 } else if (chroma_idc == 2 /* yuv422 */) {
1049 chroma_height = height;
1050 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1051 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
1052 } else { /* yuv420 */
1053 chroma_height = height >> 1;
1054 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1055 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
1056 }
1057 x_offset += 8 * h->mb_x;
1058 y_offset += 8 * (h->mb_y >> MB_FIELD(h));
1059
1060 if (list0 && list1) {
1061 /* don't optimize for luma-only case, since B-frames usually
1062 * use implicit weights => chroma too. */
1063 uint8_t *tmp_cb = h->bipred_scratchpad;
1064 uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
1065 uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
1066 int refn0 = h->ref_cache[0][scan8[n]];
1067 int refn1 = h->ref_cache[1][scan8[n]];
1068
1069 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
1070 dest_y, dest_cb, dest_cr,
1071 x_offset, y_offset, qpix_put, chroma_put,
1072 pixel_shift, chroma_idc);
1073 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
1074 tmp_y, tmp_cb, tmp_cr,
1075 x_offset, y_offset, qpix_put, chroma_put,
1076 pixel_shift, chroma_idc);
1077
1078 if (h->use_weight == 2) {
1079 int weight0 = h->implicit_weight[refn0][refn1][h->mb_y & 1];
1080 int weight1 = 64 - weight0;
1081 luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
1082 height, 5, weight0, weight1, 0);
1083 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
1084 chroma_height, 5, weight0, weight1, 0);
1085 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
1086 chroma_height, 5, weight0, weight1, 0);
1087 } else {
1088 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
1089 h->luma_log2_weight_denom,
1090 h->luma_weight[refn0][0][0],
1091 h->luma_weight[refn1][1][0],
1092 h->luma_weight[refn0][0][1] +
1093 h->luma_weight[refn1][1][1]);
1094 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
1095 h->chroma_log2_weight_denom,
1096 h->chroma_weight[refn0][0][0][0],
1097 h->chroma_weight[refn1][1][0][0],
1098 h->chroma_weight[refn0][0][0][1] +
1099 h->chroma_weight[refn1][1][0][1]);
1100 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
1101 h->chroma_log2_weight_denom,
1102 h->chroma_weight[refn0][0][1][0],
1103 h->chroma_weight[refn1][1][1][0],
1104 h->chroma_weight[refn0][0][1][1] +
1105 h->chroma_weight[refn1][1][1][1]);
1106 }
1107 } else {
1108 int list = list1 ? 1 : 0;
1109 int refn = h->ref_cache[list][scan8[n]];
1110 Picture *ref = &h->ref_list[list][refn];
1111 mc_dir_part(h, ref, n, square, height, delta, list,
1112 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1113 qpix_put, chroma_put, pixel_shift, chroma_idc);
1114
1115 luma_weight_op(dest_y, h->mb_linesize, height,
1116 h->luma_log2_weight_denom,
1117 h->luma_weight[refn][list][0],
1118 h->luma_weight[refn][list][1]);
1119 if (h->use_weight_chroma) {
1120 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
1121 h->chroma_log2_weight_denom,
1122 h->chroma_weight[refn][list][0][0],
1123 h->chroma_weight[refn][list][0][1]);
1124 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
1125 h->chroma_log2_weight_denom,
1126 h->chroma_weight[refn][list][1][0],
1127 h->chroma_weight[refn][list][1][1]);
1128 }
1129 }
1130 }
1131
1132 static av_always_inline void prefetch_motion(H264Context *h, int list,
1133 int pixel_shift, int chroma_idc)
1134 {
1135 /* fetch pixels for estimated mv 4 macroblocks ahead
1136 * optimized for 64byte cache lines */
1137 const int refn = h->ref_cache[list][scan8[0]];
1138 if (refn >= 0) {
1139 const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * h->mb_x + 8;
1140 const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * h->mb_y;
1141 uint8_t **src = h->ref_list[list][refn].f.data;
1142 int off = (mx << pixel_shift) +
1143 (my + (h->mb_x & 3) * 4) * h->mb_linesize +
1144 (64 << pixel_shift);
1145 h->vdsp.prefetch(src[0] + off, h->linesize, 4);
1146 if (chroma_idc == 3 /* yuv444 */) {
1147 h->vdsp.prefetch(src[1] + off, h->linesize, 4);
1148 h->vdsp.prefetch(src[2] + off, h->linesize, 4);
1149 } else {
1150 off = ((mx >> 1) << pixel_shift) +
1151 ((my >> 1) + (h->mb_x & 7)) * h->uvlinesize +
1152 (64 << pixel_shift);
1153 h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
1154 }
1155 }
1156 }
1157
1158 static void free_tables(H264Context *h, int free_rbsp)
1159 {
1160 int i;
1161 H264Context *hx;
1162
1163 av_freep(&h->intra4x4_pred_mode);
1164 av_freep(&h->chroma_pred_mode_table);
1165 av_freep(&h->cbp_table);
1166 av_freep(&h->mvd_table[0]);
1167 av_freep(&h->mvd_table[1]);
1168 av_freep(&h->direct_table);
1169 av_freep(&h->non_zero_count);
1170 av_freep(&h->slice_table_base);
1171 h->slice_table = NULL;
1172 av_freep(&h->list_counts);
1173
1174 av_freep(&h->mb2b_xy);
1175 av_freep(&h->mb2br_xy);
1176
1177 av_buffer_pool_uninit(&h->qscale_table_pool);
1178 av_buffer_pool_uninit(&h->mb_type_pool);
1179 av_buffer_pool_uninit(&h->motion_val_pool);
1180 av_buffer_pool_uninit(&h->ref_index_pool);
1181
1182 if (free_rbsp && h->DPB) {
1183 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1184 unref_picture(h, &h->DPB[i]);
1185 av_freep(&h->DPB);
1186 } else if (h->DPB) {
1187 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1188 h->DPB[i].needs_realloc = 1;
1189 }
1190
1191 h->cur_pic_ptr = NULL;
1192
1193 for (i = 0; i < MAX_THREADS; i++) {
1194 hx = h->thread_context[i];
1195 if (!hx)
1196 continue;
1197 av_freep(&hx->top_borders[1]);
1198 av_freep(&hx->top_borders[0]);
1199 av_freep(&hx->bipred_scratchpad);
1200 av_freep(&hx->edge_emu_buffer);
1201 av_freep(&hx->dc_val_base);
1202 av_freep(&hx->me.scratchpad);
1203 av_freep(&hx->er.mb_index2xy);
1204 av_freep(&hx->er.error_status_table);
1205 av_freep(&hx->er.er_temp_buffer);
1206 av_freep(&hx->er.mbintra_table);
1207 av_freep(&hx->er.mbskip_table);
1208
1209 if (free_rbsp) {
1210 av_freep(&hx->rbsp_buffer[1]);
1211 av_freep(&hx->rbsp_buffer[0]);
1212 hx->rbsp_buffer_size[0] = 0;
1213 hx->rbsp_buffer_size[1] = 0;
1214 }
1215 if (i)
1216 av_freep(&h->thread_context[i]);
1217 }
1218 }
1219
1220 static void init_dequant8_coeff_table(H264Context *h)
1221 {
1222 int i, j, q, x;
1223 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1224
1225 for (i = 0; i < 6; i++) {
1226 h->dequant8_coeff[i] = h->dequant8_buffer[i];
1227 for (j = 0; j < i; j++)
1228 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
1229 64 * sizeof(uint8_t))) {
1230 h->dequant8_coeff[i] = h->dequant8_buffer[j];
1231 break;
1232 }
1233 if (j < i)
1234 continue;
1235
1236 for (q = 0; q < max_qp + 1; q++) {
1237 int shift = div6[q];
1238 int idx = rem6[q];
1239 for (x = 0; x < 64; x++)
1240 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
1241 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
1242 h->pps.scaling_matrix8[i][x]) << shift;
1243 }
1244 }
1245 }
1246
1247 static void init_dequant4_coeff_table(H264Context *h)
1248 {
1249 int i, j, q, x;
1250 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
1251 for (i = 0; i < 6; i++) {
1252 h->dequant4_coeff[i] = h->dequant4_buffer[i];
1253 for (j = 0; j < i; j++)
1254 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
1255 16 * sizeof(uint8_t))) {
1256 h->dequant4_coeff[i] = h->dequant4_buffer[j];
1257 break;
1258 }
1259 if (j < i)
1260 continue;
1261
1262 for (q = 0; q < max_qp + 1; q++) {
1263 int shift = div6[q] + 2;
1264 int idx = rem6[q];
1265 for (x = 0; x < 16; x++)
1266 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
1267 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
1268 h->pps.scaling_matrix4[i][x]) << shift;
1269 }
1270 }
1271 }
1272
1273 static void init_dequant_tables(H264Context *h)
1274 {
1275 int i, x;
1276 init_dequant4_coeff_table(h);
1277 if (h->pps.transform_8x8_mode)
1278 init_dequant8_coeff_table(h);
1279 if (h->sps.transform_bypass) {
1280 for (i = 0; i < 6; i++)
1281 for (x = 0; x < 16; x++)
1282 h->dequant4_coeff[i][0][x] = 1 << 6;
1283 if (h->pps.transform_8x8_mode)
1284 for (i = 0; i < 6; i++)
1285 for (x = 0; x < 64; x++)
1286 h->dequant8_coeff[i][0][x] = 1 << 6;
1287 }
1288 }
1289
1290 int ff_h264_alloc_tables(H264Context *h)
1291 {
1292 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
1293 const int row_mb_num = h->mb_stride * 2 * h->avctx->thread_count;
1294 int x, y, i;
1295
1296 FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
1297 row_mb_num * 8 * sizeof(uint8_t), fail)
1298 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
1299 big_mb_num * 48 * sizeof(uint8_t), fail)
1300 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
1301 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
1302 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
1303 big_mb_num * sizeof(uint16_t), fail)
1304 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
1305 big_mb_num * sizeof(uint8_t), fail)
1306 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0],
1307 16 * row_mb_num * sizeof(uint8_t), fail);
1308 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1],
1309 16 * row_mb_num * sizeof(uint8_t), fail);
1310 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
1311 4 * big_mb_num * sizeof(uint8_t), fail);
1312 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
1313 big_mb_num * sizeof(uint8_t), fail)
1314
1315 memset(h->slice_table_base, -1,
1316 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
1317 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
1318
1319 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
1320 big_mb_num * sizeof(uint32_t), fail);
1321 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
1322 big_mb_num * sizeof(uint32_t), fail);
1323 for (y = 0; y < h->mb_height; y++)
1324 for (x = 0; x < h->mb_width; x++) {
1325 const int mb_xy = x + y * h->mb_stride;
1326 const int b_xy = 4 * x + 4 * y * h->b_stride;
1327
1328 h->mb2b_xy[mb_xy] = b_xy;
1329 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
1330 }
1331
1332 if (!h->dequant4_coeff[0])
1333 init_dequant_tables(h);
1334
1335 if (!h->DPB) {
1336 h->DPB = av_mallocz_array(MAX_PICTURE_COUNT, sizeof(*h->DPB));
1337 if (!h->DPB)
1338 return AVERROR(ENOMEM);
1339 for (i = 0; i < MAX_PICTURE_COUNT; i++)
1340 avcodec_get_frame_defaults(&h->DPB[i].f);
1341 avcodec_get_frame_defaults(&h->cur_pic.f);
1342 }
1343
1344 return 0;
1345
1346 fail:
1347 free_tables(h, 1);
1348 return AVERROR(ENOMEM);
1349 }
1350
1351 /**
1352 * Mimic alloc_tables(), but for every context thread.
1353 */
1354 static void clone_tables(H264Context *dst, H264Context *src, int i)
1355 {
1356 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
1357 dst->non_zero_count = src->non_zero_count;
1358 dst->slice_table = src->slice_table;
1359 dst->cbp_table = src->cbp_table;
1360 dst->mb2b_xy = src->mb2b_xy;
1361 dst->mb2br_xy = src->mb2br_xy;
1362 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
1363 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
1364 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
1365 dst->direct_table = src->direct_table;
1366 dst->list_counts = src->list_counts;
1367 dst->DPB = src->DPB;
1368 dst->cur_pic_ptr = src->cur_pic_ptr;
1369 dst->cur_pic = src->cur_pic;
1370 dst->bipred_scratchpad = NULL;
1371 dst->edge_emu_buffer = NULL;
1372 dst->me.scratchpad = NULL;
1373 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
1374 src->sps.chroma_format_idc);
1375 }
1376
1377 /**
1378 * Init context
1379 * Allocate buffers which are not shared amongst multiple threads.
1380 */
1381 static int context_init(H264Context *h)
1382 {
1383 ERContext *er = &h->er;
1384 int mb_array_size = h->mb_height * h->mb_stride;
1385 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
1386 int c_size = h->mb_stride * (h->mb_height + 1);
1387 int yc_size = y_size + 2 * c_size;
1388 int x, y, i;
1389
1390 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[0],
1391 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1392 FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[1],
1393 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
1394
1395 h->ref_cache[0][scan8[5] + 1] =
1396 h->ref_cache[0][scan8[7] + 1] =
1397 h->ref_cache[0][scan8[13] + 1] =
1398 h->ref_cache[1][scan8[5] + 1] =
1399 h->ref_cache[1][scan8[7] + 1] =
1400 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
1401
1402 if (CONFIG_ERROR_RESILIENCE) {
1403 /* init ER */
1404 er->avctx = h->avctx;
1405 er->dsp = &h->dsp;
1406 er->decode_mb = h264_er_decode_mb;
1407 er->opaque = h;
1408 er->quarter_sample = 1;
1409
1410 er->mb_num = h->mb_num;
1411 er->mb_width = h->mb_width;
1412 er->mb_height = h->mb_height;
1413 er->mb_stride = h->mb_stride;
1414 er->b8_stride = h->mb_width * 2 + 1;
1415
1416 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy, (h->mb_num + 1) * sizeof(int),
1417 fail); // error ressilience code looks cleaner with this
1418 for (y = 0; y < h->mb_height; y++)
1419 for (x = 0; x < h->mb_width; x++)
1420 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
1421
1422 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
1423 h->mb_stride + h->mb_width;
1424
1425 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
1426 mb_array_size * sizeof(uint8_t), fail);
1427
1428 FF_ALLOC_OR_GOTO(h->avctx, er->mbintra_table, mb_array_size, fail);
1429 memset(er->mbintra_table, 1, mb_array_size);
1430
1431 FF_ALLOCZ_OR_GOTO(h->avctx, er->mbskip_table, mb_array_size + 2, fail);
1432
1433 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer, h->mb_height * h->mb_stride,
1434 fail);
1435
1436 FF_ALLOCZ_OR_GOTO(h->avctx, h->dc_val_base, yc_size * sizeof(int16_t), fail);
1437 er->dc_val[0] = h->dc_val_base + h->mb_width * 2 + 2;
1438 er->dc_val[1] = h->dc_val_base + y_size + h->mb_stride + 1;
1439 er->dc_val[2] = er->dc_val[1] + c_size;
1440 for (i = 0; i < yc_size; i++)
1441 h->dc_val_base[i] = 1024;
1442 }
1443
1444 return 0;
1445
1446 fail:
1447 return AVERROR(ENOMEM); // free_tables will clean up for us
1448 }
1449
1450 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1451 int parse_extradata);
1452
1453 int ff_h264_decode_extradata(H264Context *h)
1454 {
1455 AVCodecContext *avctx = h->avctx;
1456 int ret;
1457
1458 if (avctx->extradata[0] == 1) {
1459 int i, cnt, nalsize;
1460 unsigned char *p = avctx->extradata;
1461
1462 h->is_avc = 1;
1463
1464 if (avctx->extradata_size < 7) {
1465 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1466 return AVERROR_INVALIDDATA;
1467 }
1468 /* sps and pps in the avcC always have length coded with 2 bytes,
1469 * so put a fake nal_length_size = 2 while parsing them */
1470 h->nal_length_size = 2;
1471 // Decode sps from avcC
1472 cnt = *(p + 5) & 0x1f; // Number of sps
1473 p += 6;
1474 for (i = 0; i < cnt; i++) {
1475 nalsize = AV_RB16(p) + 2;
1476 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1477 return AVERROR_INVALIDDATA;
1478 ret = decode_nal_units(h, p, nalsize, 1);
1479 if (ret < 0) {
1480 av_log(avctx, AV_LOG_ERROR,
1481 "Decoding sps %d from avcC failed\n", i);
1482 return ret;
1483 }
1484 p += nalsize;
1485 }
1486 // Decode pps from avcC
1487 cnt = *(p++); // Number of pps
1488 for (i = 0; i < cnt; i++) {
1489 nalsize = AV_RB16(p) + 2;
1490 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1491 return AVERROR_INVALIDDATA;
1492 ret = decode_nal_units(h, p, nalsize, 1);
1493 if (ret < 0) {
1494 av_log(avctx, AV_LOG_ERROR,
1495 "Decoding pps %d from avcC failed\n", i);
1496 return ret;
1497 }
1498 p += nalsize;
1499 }
1500 // Now store right nal length size, that will be used to parse all other nals
1501 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1502 } else {
1503 h->is_avc = 0;
1504 ret = decode_nal_units(h, avctx->extradata, avctx->extradata_size, 1);
1505 if (ret < 0)
1506 return ret;
1507 }
1508 return 0;
1509 }
1510
1511 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1512 {
1513 H264Context *h = avctx->priv_data;
1514 int i;
1515 int ret;
1516
1517 h->avctx = avctx;
1518
1519 h->bit_depth_luma = 8;
1520 h->chroma_format_idc = 1;
1521
1522 ff_h264dsp_init(&h->h264dsp, 8, 1);
1523 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1524 ff_h264qpel_init(&h->h264qpel, 8);
1525 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, 8, 1);
1526
1527 h->dequant_coeff_pps = -1;
1528
1529 /* needed so that IDCT permutation is known early */
1530 if (CONFIG_ERROR_RESILIENCE)
1531 ff_dsputil_init(&h->dsp, h->avctx);
1532 ff_videodsp_init(&h->vdsp, 8);
1533
1534 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1535 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1536
1537 h->picture_structure = PICT_FRAME;
1538 h->slice_context_count = 1;
1539 h->workaround_bugs = avctx->workaround_bugs;
1540 h->flags = avctx->flags;
1541
1542 /* set defaults */
1543 // s->decode_mb = ff_h263_decode_mb;
1544 if (!avctx->has_b_frames)
1545 h->low_delay = 1;
1546
1547 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1548
1549 ff_h264_decode_init_vlc();
1550
1551 h->pixel_shift = 0;
1552 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1553
1554 h->thread_context[0] = h;
1555 h->outputed_poc = h->next_outputed_poc = INT_MIN;
1556 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1557 h->last_pocs[i] = INT_MIN;
1558 h->prev_poc_msb = 1 << 16;
1559 h->x264_build = -1;
1560 ff_h264_reset_sei(h);
1561 if (avctx->codec_id == AV_CODEC_ID_H264) {
1562 if (avctx->ticks_per_frame == 1)
1563 h->avctx->time_base.den *= 2;
1564 avctx->ticks_per_frame = 2;
1565 }
1566
1567 if (avctx->extradata_size > 0 && avctx->extradata) {
1568 ret = ff_h264_decode_extradata(h);
1569 if (ret < 0)
1570 return ret;
1571 }
1572
1573 if (h->sps.bitstream_restriction_flag &&
1574 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
1575 h->avctx->has_b_frames = h->sps.num_reorder_frames;
1576 h->low_delay = 0;
1577 }
1578
1579 avctx->internal->allocate_progress = 1;
1580
1581 return 0;
1582 }
1583
1584 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1585 #undef REBASE_PICTURE
1586 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
1587 ((pic && pic >= old_ctx->DPB && \
1588 pic < old_ctx->DPB + MAX_PICTURE_COUNT) ? \
1589 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
1590
1591 static void copy_picture_range(Picture **to, Picture **from, int count,
1592 H264Context *new_base,
1593 H264Context *old_base)
1594 {
1595 int i;
1596
1597 for (i = 0; i < count; i++) {
1598 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1599 IN_RANGE(from[i], old_base->DPB,
1600 sizeof(Picture) * MAX_PICTURE_COUNT) ||
1601 !from[i]));
1602 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1603 }
1604 }
1605
1606 static void copy_parameter_set(void **to, void **from, int count, int size)
1607 {
1608 int i;
1609
1610 for (i = 0; i < count; i++) {
1611 if (to[i] && !from[i])
1612 av_freep(&to[i]);
1613 else if (from[i] && !to[i])
1614 to[i] = av_malloc(size);
1615
1616 if (from[i])
1617 memcpy(to[i], from[i], size);
1618 }
1619 }
1620
1621 static int decode_init_thread_copy(AVCodecContext *avctx)
1622 {
1623 H264Context *h = avctx->priv_data;
1624
1625 if (!avctx->internal->is_copy)
1626 return 0;
1627 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1628 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1629
1630 h->context_initialized = 0;
1631
1632 return 0;
1633 }
1634
1635 #define copy_fields(to, from, start_field, end_field) \
1636 memcpy(&to->start_field, &from->start_field, \
1637 (char *)&to->end_field - (char *)&to->start_field)
1638
1639 static int h264_slice_header_init(H264Context *, int);
1640
1641 static int h264_set_parameter_from_sps(H264Context *h);
1642
1643 static int decode_update_thread_context(AVCodecContext *dst,
1644 const AVCodecContext *src)
1645 {
1646 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1647 int inited = h->context_initialized, err = 0;
1648 int context_reinitialized = 0;
1649 int i, ret;
1650
1651 if (dst == src || !h1->context_initialized)
1652 return 0;
1653
1654 if (inited &&
1655 (h->width != h1->width ||
1656 h->height != h1->height ||
1657 h->mb_width != h1->mb_width ||
1658 h->mb_height != h1->mb_height ||
1659 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
1660 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
1661 h->sps.colorspace != h1->sps.colorspace)) {
1662
1663 /* set bits_per_raw_sample to the previous value. the check for changed
1664 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
1665 * the current value */
1666 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1667
1668 av_freep(&h->bipred_scratchpad);
1669
1670 h->width = h1->width;
1671 h->height = h1->height;
1672 h->mb_height = h1->mb_height;
1673 h->mb_width = h1->mb_width;
1674 h->mb_num = h1->mb_num;
1675 h->mb_stride = h1->mb_stride;
1676 h->b_stride = h1->b_stride;
1677
1678 if ((err = h264_slice_header_init(h, 1)) < 0) {
1679 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
1680 return err;
1681 }
1682 context_reinitialized = 1;
1683
1684 /* update linesize on resize. The decoder doesn't
1685 * necessarily call h264_frame_start in the new thread */
1686 h->linesize = h1->linesize;
1687 h->uvlinesize = h1->uvlinesize;
1688
1689 /* copy block_offset since frame_start may not be called */
1690 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
1691 }
1692
1693 if (!inited) {
1694 for (i = 0; i < MAX_SPS_COUNT; i++)
1695 av_freep(h->sps_buffers + i);
1696
1697 for (i = 0; i < MAX_PPS_COUNT; i++)
1698 av_freep(h->pps_buffers + i);
1699
1700 memcpy(h, h1, sizeof(*h1));
1701 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1702 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1703 memset(&h->er, 0, sizeof(h->er));
1704 memset(&h->me, 0, sizeof(h->me));
1705 memset(&h->mb, 0, sizeof(h->mb));
1706 memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
1707 memset(&h->mb_padding, 0, sizeof(h->mb_padding));
1708 h->context_initialized = 0;
1709
1710 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
1711 avcodec_get_frame_defaults(&h->cur_pic.f);
1712 h->cur_pic.tf.f = &h->cur_pic.f;
1713
1714 h->avctx = dst;
1715 h->DPB = NULL;
1716 h->qscale_table_pool = NULL;
1717 h->mb_type_pool = NULL;
1718 h->ref_index_pool = NULL;
1719 h->motion_val_pool = NULL;
1720
1721 ret = ff_h264_alloc_tables(h);
1722 if (ret < 0) {
1723 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1724 return ret;
1725 }
1726 ret = context_init(h);
1727 if (ret < 0) {
1728 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
1729 return ret;
1730 }
1731
1732 for (i = 0; i < 2; i++) {
1733 h->rbsp_buffer[i] = NULL;
1734 h->rbsp_buffer_size[i] = 0;
1735 }
1736 h->bipred_scratchpad = NULL;
1737 h->edge_emu_buffer = NULL;
1738
1739 h->thread_context[0] = h;
1740
1741 h->context_initialized = 1;
1742 }
1743
1744 h->avctx->coded_height = h1->avctx->coded_height;
1745 h->avctx->coded_width = h1->avctx->coded_width;
1746 h->avctx->width = h1->avctx->width;
1747 h->avctx->height = h1->avctx->height;
1748 h->coded_picture_number = h1->coded_picture_number;
1749 h->first_field = h1->first_field;
1750 h->picture_structure = h1->picture_structure;
1751 h->qscale = h1->qscale;
1752 h->droppable = h1->droppable;
1753 h->data_partitioning = h1->data_partitioning;
1754 h->low_delay = h1->low_delay;
1755
1756 for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1757 unref_picture(h, &h->DPB[i]);
1758 if (h1->DPB[i].f.data[0] &&
1759 (ret = ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
1760 return ret;
1761 }
1762
1763 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
1764 unref_picture(h, &h->cur_pic);
1765 if ((ret = ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
1766 return ret;
1767
1768 h->workaround_bugs = h1->workaround_bugs;
1769 h->low_delay = h1->low_delay;
1770 h->droppable = h1->droppable;
1771
1772 /* frame_start may not be called for the next thread (if it's decoding
1773 * a bottom field) so this has to be allocated here */
1774 err = alloc_scratch_buffers(h, h1->linesize);
1775 if (err < 0)
1776 return err;
1777
1778 // extradata/NAL handling
1779 h->is_avc = h1->is_avc;
1780
1781 // SPS/PPS
1782 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1783 MAX_SPS_COUNT, sizeof(SPS));
1784 h->sps = h1->sps;
1785 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1786 MAX_PPS_COUNT, sizeof(PPS));
1787 h->pps = h1->pps;
1788
1789 // Dequantization matrices
1790 // FIXME these are big - can they be only copied when PPS changes?
1791 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1792
1793 for (i = 0; i < 6; i++)
1794 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1795 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1796
1797 for (i = 0; i < 6; i++)
1798 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1799 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1800
1801 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1802
1803 // POC timing
1804 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1805
1806 // reference lists
1807 copy_fields(h, h1, short_ref, cabac_init_idc);
1808
1809 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
1810 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
1811 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1812 MAX_DELAYED_PIC_COUNT + 2, h, h1);
1813
1814 h->last_slice_type = h1->last_slice_type;
1815
1816 if (context_reinitialized)
1817 h264_set_parameter_from_sps(h);
1818
1819 if (!h->cur_pic_ptr)
1820 return 0;
1821
1822 if (!h->droppable) {
1823 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1824 h->prev_poc_msb = h->poc_msb;
1825 h->prev_poc_lsb = h->poc_lsb;
1826 }
1827 h->prev_frame_num_offset = h->frame_num_offset;
1828 h->prev_frame_num = h->frame_num;
1829 h->outputed_poc = h->next_outputed_poc;
1830
1831 return err;
1832 }
1833
1834 static int h264_frame_start(H264Context *h)
1835 {
1836 Picture *pic;
1837 int i, ret;
1838 const int pixel_shift = h->pixel_shift;
1839
1840 release_unused_pictures(h, 1);
1841 h->cur_pic_ptr = NULL;
1842
1843 i = find_unused_picture(h);
1844 if (i < 0) {
1845 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1846 return i;
1847 }
1848 pic = &h->DPB[i];
1849
1850 pic->reference = h->droppable ? 0 : h->picture_structure;
1851 pic->f.coded_picture_number = h->coded_picture_number++;
1852 pic->field_picture = h->picture_structure != PICT_FRAME;
1853 /*
1854 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
1855 * in later.
1856 * See decode_nal_units().
1857 */
1858 pic->f.key_frame = 0;
1859 pic->mmco_reset = 0;
1860
1861 if ((ret = alloc_picture(h, pic)) < 0)
1862 return ret;
1863
1864 h->cur_pic_ptr = pic;
1865 unref_picture(h, &h->cur_pic);
1866 if ((ret = ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
1867 return ret;
1868
1869 if (CONFIG_ERROR_RESILIENCE)
1870 ff_er_frame_start(&h->er);
1871
1872 assert(h->linesize && h->uvlinesize);
1873
1874 for (i = 0; i < 16; i++) {
1875 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1876 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
1877 }
1878 for (i = 0; i < 16; i++) {
1879 h->block_offset[16 + i] =
1880 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1881 h->block_offset[48 + 16 + i] =
1882 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1883 }
1884
1885 /* can't be in alloc_tables because linesize isn't known there.
1886 * FIXME: redo bipred weight to not require extra buffer? */
1887 for (i = 0; i < h->slice_context_count; i++)
1888 if (h->thread_context[i]) {
1889 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
1890 if (ret < 0)
1891 return ret;
1892 }
1893
1894 /* Some macroblocks can be accessed before they're available in case
1895 * of lost slices, MBAFF or threading. */
1896 memset(h->slice_table, -1,
1897 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1898
1899 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1900 // s->current_picture.f.reference /* || h->contains_intra */ || 1;
1901
1902 /* We mark the current picture as non-reference after allocating it, so
1903 * that if we break out due to an error it can be released automatically
1904 * in the next ff_MPV_frame_start().
1905 */
1906 h->cur_pic_ptr->reference = 0;
1907
1908 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
1909
1910 h->next_output_pic = NULL;
1911
1912 assert(h->cur_pic_ptr->long_ref == 0);
1913
1914 return 0;
1915 }
1916
1917 /**
1918 * Run setup operations that must be run after slice header decoding.
1919 * This includes finding the next displayed frame.
1920 *
1921 * @param h h264 master context
1922 * @param setup_finished enough NALs have been read that we can call
1923 * ff_thread_finish_setup()
1924 */
1925 static void decode_postinit(H264Context *h, int setup_finished)
1926 {
1927 Picture *out = h->cur_pic_ptr;
1928 Picture *cur = h->cur_pic_ptr;
1929 int i, pics, out_of_order, out_idx;
1930 int invalid = 0, cnt = 0;
1931
1932 h->cur_pic_ptr->f.pict_type = h->pict_type;
1933
1934 if (h->next_output_pic)
1935 return;
1936
1937 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1938 /* FIXME: if we have two PAFF fields in one packet, we can't start
1939 * the next thread here. If we have one field per packet, we can.
1940 * The check in decode_nal_units() is not good enough to find this
1941 * yet, so we assume the worst for now. */
1942 // if (setup_finished)
1943 // ff_thread_finish_setup(h->avctx);
1944 return;
1945 }
1946
1947 cur->f.interlaced_frame = 0;
1948 cur->f.repeat_pict = 0;
1949
1950 /* Signal interlacing information externally. */
1951 /* Prioritize picture timing SEI information over used
1952 * decoding process if it exists. */
1953
1954 if (h->sps.pic_struct_present_flag) {
1955 switch (h->sei_pic_struct) {
1956 case SEI_PIC_STRUCT_FRAME:
1957 break;
1958 case SEI_PIC_STRUCT_TOP_FIELD:
1959 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1960 cur->f.interlaced_frame = 1;
1961 break;
1962 case SEI_PIC_STRUCT_TOP_BOTTOM:
1963 case SEI_PIC_STRUCT_BOTTOM_TOP:
1964 if (FIELD_OR_MBAFF_PICTURE(h))
1965 cur->f.interlaced_frame = 1;
1966 else
1967 // try to flag soft telecine progressive
1968 cur->f.interlaced_frame = h->prev_interlaced_frame;
1969 break;
1970 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1971 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1972 /* Signal the possibility of telecined film externally
1973 * (pic_struct 5,6). From these hints, let the applications
1974 * decide if they apply deinterlacing. */
1975 cur->f.repeat_pict = 1;
1976 break;
1977 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1978 cur->f.repeat_pict = 2;
1979 break;
1980 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1981 cur->f.repeat_pict = 4;
1982 break;
1983 }
1984
1985 if ((h->sei_ct_type & 3) &&
1986 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1987 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1988 } else {
1989 /* Derive interlacing flag from used decoding process. */
1990 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
1991 }
1992 h->prev_interlaced_frame = cur->f.interlaced_frame;
1993
1994 if (cur->field_poc[0] != cur->field_poc[1]) {
1995 /* Derive top_field_first from field pocs. */
1996 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1997 } else {
1998 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1999 /* Use picture timing SEI information. Even if it is a
2000 * information of a past frame, better than nothing. */
2001 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
2002 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
2003 cur->f.top_field_first = 1;
2004 else
2005 cur->f.top_field_first = 0;
2006 } else {
2007 /* Most likely progressive */
2008 cur->f.top_field_first = 0;
2009 }
2010 }
2011
2012 // FIXME do something with unavailable reference frames
2013
2014 /* Sort B-frames into display order */
2015
2016 if (h->sps.bitstream_restriction_flag &&
2017 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
2018 h->avctx->has_b_frames = h->sps.num_reorder_frames;
2019 h->low_delay = 0;
2020 }
2021
2022 if (h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
2023 !h->sps.bitstream_restriction_flag) {
2024 h->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
2025 h->low_delay = 0;
2026 }
2027
2028 pics = 0;
2029 while (h->delayed_pic[pics])
2030 pics++;
2031
2032 assert(pics <= MAX_DELAYED_PIC_COUNT);
2033
2034 h->delayed_pic[pics++] = cur;
2035 if (cur->reference == 0)
2036 cur->reference = DELAYED_PIC_REF;
2037
2038 /* Frame reordering. This code takes pictures from coding order and sorts
2039 * them by their incremental POC value into display order. It supports POC
2040 * gaps, MMCO reset codes and random resets.
2041 * A "display group" can start either with a IDR frame (f.key_frame = 1),
2042 * and/or can be closed down with a MMCO reset code. In sequences where
2043 * there is no delay, we can't detect that (since the frame was already
2044 * output to the user), so we also set h->mmco_reset to detect the MMCO
2045 * reset code.
2046 * FIXME: if we detect insufficient delays (as per h->avctx->has_b_frames),
2047 * we increase the delay between input and output. All frames affected by
2048 * the lag (e.g. those that should have been output before another frame
2049 * that we already returned to the user) will be dropped. This is a bug
2050 * that we will fix later. */
2051 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
2052 cnt += out->poc < h->last_pocs[i];
2053 invalid += out->poc == INT_MIN;
2054 }
2055 if (!h->mmco_reset && !cur->f.key_frame &&
2056 cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
2057 h->mmco_reset = 2;
2058 if (pics > 1)
2059 h->delayed_pic[pics - 2]->mmco_reset = 2;
2060 }
2061 if (h->mmco_reset || cur->f.key_frame) {
2062 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2063 h->last_pocs[i] = INT_MIN;
2064 cnt = 0;
2065 invalid = MAX_DELAYED_PIC_COUNT;
2066 }
2067 out = h->delayed_pic[0];
2068 out_idx = 0;
2069 for (i = 1; i < MAX_DELAYED_PIC_COUNT &&
2070 h->delayed_pic[i] &&
2071 !h->delayed_pic[i - 1]->mmco_reset &&
2072 !h->delayed_pic[i]->f.key_frame;
2073 i++)
2074 if (h->delayed_pic[i]->poc < out->poc) {
2075 out = h->delayed_pic[i];
2076 out_idx = i;
2077 }
2078 if (h->avctx->has_b_frames == 0 &&
2079 (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
2080 h->next_outputed_poc = INT_MIN;
2081 out_of_order = !out->f.key_frame && !h->mmco_reset &&
2082 (out->poc < h->next_outputed_poc);
2083
2084 if (h->sps.bitstream_restriction_flag &&
2085 h->avctx->has_b_frames >= h->sps.num_reorder_frames) {
2086 } else if (out_of_order && pics - 1 == h->avctx->has_b_frames &&
2087 h->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
2088 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
2089 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, cnt);
2090 }
2091 h->low_delay = 0;
2092 } else if (h->low_delay &&
2093 ((h->next_outputed_poc != INT_MIN &&
2094 out->poc > h->next_outputed_poc + 2) ||
2095 cur->f.pict_type == AV_PICTURE_TYPE_B)) {
2096 h->low_delay = 0;
2097 h->avctx->has_b_frames++;
2098 }
2099
2100 if (pics > h->avctx->has_b_frames) {
2101 out->reference &= ~DELAYED_PIC_REF;
2102 // for frame threading, the owner must be the second field's thread or
2103 // else the first thread can release the picture and reuse it unsafely
2104 for (i = out_idx; h->delayed_pic[i]; i++)
2105 h->delayed_pic[i] = h->delayed_pic[i + 1];
2106 }
2107 memmove(h->last_pocs, &h->last_pocs[1],
2108 sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
2109 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
2110 if (!out_of_order && pics > h->avctx->has_b_frames) {
2111 h->next_output_pic = out;
2112 if (out->mmco_reset) {
2113 if (out_idx > 0) {
2114 h->next_outputed_poc = out->poc;
2115 h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
2116 } else {
2117 h->next_outputed_poc = INT_MIN;
2118 }
2119 } else {
2120 if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f.key_frame) {
2121 h->next_outputed_poc = INT_MIN;
2122 } else {
2123 h->next_outputed_poc = out->poc;
2124 }
2125 }
2126 h->mmco_reset = 0;
2127 } else {
2128 av_log(h->avctx, AV_LOG_DEBUG, "no picture\n");
2129 }
2130
2131 if (setup_finished && !h->avctx->hwaccel)
2132 ff_thread_finish_setup(h->avctx);
2133 }
2134
2135 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
2136 uint8_t *src_cb, uint8_t *src_cr,
2137 int linesize, int uvlinesize,
2138 int simple)
2139 {
2140 uint8_t *top_border;
2141 int top_idx = 1;
2142 const int pixel_shift = h->pixel_shift;
2143 int chroma444 = CHROMA444(h);
2144 int chroma422 = CHROMA422(h);
2145
2146 src_y -= linesize;
2147 src_cb -= uvlinesize;
2148 src_cr -= uvlinesize;
2149
2150 if (!simple && FRAME_MBAFF(h)) {
2151 if (h->mb_y & 1) {
2152 if (!MB_MBAFF(h)) {
2153 top_border = h->top_borders[0][h->mb_x];
2154 AV_COPY128(top_border, src_y + 15 * linesize);
2155 if (pixel_shift)
2156 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
2157 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2158 if (chroma444) {
2159 if (pixel_shift) {
2160 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2161 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
2162 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
2163 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
2164 } else {
2165 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
2166 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
2167 }
2168 } else if (chroma422) {
2169 if (pixel_shift) {
2170 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
2171 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
2172 } else {
2173 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
2174 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
2175 }
2176 } else {
2177 if (pixel_shift) {
2178 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
2179 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
2180 } else {
2181 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
2182 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
2183 }
2184 }
2185 }
2186 }
2187 } else if (MB_MBAFF(h)) {
2188 top_idx = 0;
2189 } else
2190 return;
2191 }
2192
2193 top_border = h->top_borders[top_idx][h->mb_x];
2194 /* There are two lines saved, the line above the top macroblock
2195 * of a pair, and the line above the bottom macroblock. */
2196 AV_COPY128(top_border, src_y + 16 * linesize);
2197 if (pixel_shift)
2198 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
2199
2200 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2201 if (chroma444) {
2202 if (pixel_shift) {
2203 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
2204 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
2205 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
2206 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
2207 } else {
2208 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
2209 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
2210 }
2211 } else if (chroma422) {
2212 if (pixel_shift) {
2213 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
2214 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
2215 } else {
2216 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
2217 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
2218 }
2219 } else {
2220 if (pixel_shift) {
2221 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
2222 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
2223 } else {
2224 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
2225 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
2226 }
2227 }
2228 }
2229 }
2230
2231 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
2232 uint8_t *src_cb, uint8_t *src_cr,
2233 int linesize, int uvlinesize,
2234 int xchg, int chroma444,
2235 int simple, int pixel_shift)
2236 {
2237 int deblock_topleft;
2238 int deblock_top;
2239 int top_idx = 1;
2240 uint8_t *top_border_m1;
2241 uint8_t *top_border;
2242
2243 if (!simple && FRAME_MBAFF(h)) {
2244 if (h->mb_y & 1) {
2245 if (!MB_MBAFF(h))
2246 return;
2247 } else {
2248 top_idx = MB_MBAFF(h) ? 0 : 1;
2249 }
2250 }
2251
2252 if (h->deblocking_filter == 2) {
2253 deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num;
2254 deblock_top = h->top_type;
2255 } else {
2256 deblock_topleft = (h->mb_x > 0);
2257 deblock_top = (h->mb_y > !!MB_FIELD(h));
2258 }
2259
2260 src_y -= linesize + 1 + pixel_shift;
2261 src_cb -= uvlinesize + 1 + pixel_shift;
2262 src_cr -= uvlinesize + 1 + pixel_shift;
2263
2264 top_border_m1 = h->top_borders[top_idx][h->mb_x - 1];
2265 top_border = h->top_borders[top_idx][h->mb_x];
2266
2267 #define XCHG(a, b, xchg) \
2268 if (pixel_shift) { \
2269 if (xchg) { \
2270 AV_SWAP64(b + 0, a + 0); \
2271 AV_SWAP64(b + 8, a + 8); \
2272 } else { \
2273 AV_COPY128(b, a); \
2274 } \
2275 } else if (xchg) \
2276 AV_SWAP64(b, a); \
2277 else \
2278 AV_COPY64(b, a);
2279
2280 if (deblock_top) {
2281 if (deblock_topleft) {
2282 XCHG(top_border_m1 + (8 << pixel_shift),
2283 src_y - (7 << pixel_shift), 1);
2284 }
2285 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
2286 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
2287 if (h->mb_x + 1 < h->mb_width) {
2288 XCHG(h->top_borders[top_idx][h->mb_x + 1],
2289 src_y + (17 << pixel_shift), 1);
2290 }
2291 }
2292 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
2293 if (chroma444) {
2294 if (deblock_topleft) {
2295 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2296 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2297 }
2298 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
2299 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
2300 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
2301 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
2302 if (h->mb_x + 1 < h->mb_width) {
2303 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
2304 XCHG(h->top_borders[top_idx][h->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
2305 }
2306 } else {
2307 if (deblock_top) {
2308 if (deblock_topleft) {
2309 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
2310 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
2311 }
2312 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
2313 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
2314 }
2315 }
2316 }
2317 }
2318
2319 static av_always_inline int dctcoef_get(int16_t *mb, int high_bit_depth,
2320 int index)
2321 {
2322 if (high_bit_depth) {
2323 return AV_RN32A(((int32_t *)mb) + index);
2324 } else
2325 return AV_RN16A(mb + index);
2326 }
2327
2328 static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth,
2329 int index, int value)
2330 {
2331 if (high_bit_depth) {
2332 AV_WN32A(((int32_t *)mb) + index, value);
2333 } else
2334 AV_WN16A(mb + index, value);
2335 }
2336
2337 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
2338 int mb_type, int is_h264,
2339 int simple,
2340 int transform_bypass,
2341 int pixel_shift,
2342 int *block_offset,
2343 int linesize,
2344 uint8_t *dest_y, int p)
2345 {
2346 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2347 void (*idct_dc_add)(uint8_t *dst, int16_t *block, int stride);
2348 int i;
2349 int qscale = p == 0 ? h->qscale : h->chroma_qp[p - 1];
2350 block_offset += 16 * p;
2351 if (IS_INTRA4x4(mb_type)) {
2352 if (IS_8x8DCT(mb_type)) {
2353 if (transform_bypass) {
2354 idct_dc_add =
2355 idct_add = h->h264dsp.h264_add_pixels8_clear;
2356 } else {
2357 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
2358 idct_add = h->h264dsp.h264_idct8_add;
2359 }
2360 for (i = 0; i < 16; i += 4) {
2361 uint8_t *const ptr = dest_y + block_offset[i];
2362 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2363 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2364 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2365 } else {
2366 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2367 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
2368 (h->topright_samples_available << i) & 0x4000, linesize);
2369 if (nnz) {
2370 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2371 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2372 else
2373 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2374 }
2375 }
2376 }
2377 } else {
2378 if (transform_bypass) {
2379 idct_dc_add =
2380 idct_add = h->h264dsp.h264_add_pixels4_clear;
2381 } else {
2382 idct_dc_add = h->h264dsp.h264_idct_dc_add;
2383 idct_add = h->h264dsp.h264_idct_add;
2384 }
2385 for (i = 0; i < 16; i++) {
2386 uint8_t *const ptr = dest_y + block_offset[i];
2387 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
2388
2389 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
2390 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2391 } else {
2392 uint8_t *topright;
2393 int nnz, tr;
2394 uint64_t tr_high;
2395 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
2396 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
2397 assert(h->mb_y || linesize <= block_offset[i]);
2398 if (!topright_avail) {
2399 if (pixel_shift) {
2400 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
2401 topright = (uint8_t *)&tr_high;
2402 } else {
2403 tr = ptr[3 - linesize] * 0x01010101u;
2404 topright = (uint8_t *)&tr;
2405 }
2406 } else
2407 topright = ptr + (4 << pixel_shift) - linesize;
2408 } else
2409 topright = NULL;
2410
2411 h->hpc.pred4x4[dir](ptr, topright, linesize);
2412 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
2413 if (nnz) {
2414 if (is_h264) {
2415 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2416 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2417 else
2418 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
2419 } else if (CONFIG_SVQ3_DECODER)
2420 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
2421 }
2422 }
2423 }
2424 }
2425 } else {
2426 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
2427 if (is_h264) {
2428 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
2429 if (!transform_bypass)
2430 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
2431 h->mb_luma_dc[p],
2432 h->dequant4_coeff[p][qscale][0]);
2433 else {
2434 static const uint8_t dc_mapping[16] = {
2435 0 * 16, 1 * 16, 4 * 16, 5 * 16,
2436 2 * 16, 3 * 16, 6 * 16, 7 * 16,
2437 8 * 16, 9 * 16, 12 * 16, 13 * 16,
2438 10 * 16, 11 * 16, 14 * 16, 15 * 16
2439 };
2440 for (i = 0; i < 16; i++)
2441 dctcoef_set(h->mb + (p * 256 << pixel_shift),
2442 pixel_shift, dc_mapping[i],
2443 dctcoef_get(h->mb_luma_dc[p],
2444 pixel_shift, i));
2445 }
2446 }
2447 } else if (CONFIG_SVQ3_DECODER)
2448 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
2449 h->mb_luma_dc[p], qscale);
2450 }
2451 }
2452
2453 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
2454 int is_h264, int simple,
2455 int transform_bypass,
2456 int pixel_shift,
2457 int *block_offset,
2458 int linesize,
2459 uint8_t *dest_y, int p)
2460 {
2461 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
2462 int i;
2463 block_offset += 16 * p;
2464 if (!IS_INTRA4x4(mb_type)) {
2465 if (is_h264) {
2466 if (IS_INTRA16x16(mb_type)) {
2467 if (transform_bypass) {
2468 if (h->sps.profile_idc == 244 &&
2469 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
2470 h->intra16x16_pred_mode == HOR_PRED8x8)) {
2471 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
2472 h->mb + (p * 256 << pixel_shift),
2473 linesize);
2474 } else {
2475 for (i = 0; i < 16; i++)
2476 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
2477 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
2478 h->h264dsp.h264_add_pixels4_clear(dest_y + block_offset[i],
2479 h->mb + (i * 16 + p * 256 << pixel_shift),
2480 linesize);
2481 }
2482 } else {
2483 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
2484 h->mb + (p * 256 << pixel_shift),
2485 linesize,
2486 h->non_zero_count_cache + p * 5 * 8);
2487 }
2488 } else if (h->cbp & 15) {
2489 if (transform_bypass) {
2490 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2491 idct_add = IS_8x8DCT(mb_type) ? h->h264dsp.h264_add_pixels8_clear
2492 : h->h264dsp.h264_add_pixels4_clear;
2493 for (i = 0; i < 16; i += di)
2494 if (h->non_zero_count_cache[scan8[i + p * 16]])
2495 idct_add(dest_y + block_offset[i],
2496 h->mb + (i * 16 + p * 256 << pixel_shift),
2497 linesize);
2498 } else {
2499 if (IS_8x8DCT(mb_type))
2500 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
2501 h->mb + (p * 256 << pixel_shift),
2502 linesize,
2503 h->non_zero_count_cache + p * 5 * 8);
2504 else
2505 h->h264dsp.h264_idct_add16(dest_y, block_offset,
2506 h->mb + (p * 256 << pixel_shift),
2507 linesize,
2508 h->non_zero_count_cache + p * 5 * 8);
2509 }
2510 }
2511 } else if (CONFIG_SVQ3_DECODER) {
2512 for (i = 0; i < 16; i++)
2513 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
2514 // FIXME benchmark weird rule, & below
2515 uint8_t *const ptr = dest_y + block_offset[i];
2516 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
2517 h->qscale, IS_INTRA(mb_type) ? 1 : 0);
2518 }
2519 }
2520 }
2521 }
2522
2523 #define BITS 8
2524 #define SIMPLE 1
2525 #include "h264_mb_template.c"
2526
2527 #undef BITS
2528 #define BITS 16
2529 #include "h264_mb_template.c"
2530
2531 #undef SIMPLE
2532 #define SIMPLE 0
2533 #include "h264_mb_template.c"
2534
2535 void ff_h264_hl_decode_mb(H264Context *h)
2536 {
2537 const int mb_xy = h->mb_xy;
2538 const int mb_type = h->cur_pic.mb_type[mb_xy];
2539 int is_complex = CONFIG_SMALL || h->is_complex ||
2540 IS_INTRA_PCM(mb_type) || h->qscale == 0;
2541
2542 if (CHROMA444(h)) {
2543 if (is_complex || h->pixel_shift)
2544 hl_decode_mb_444_complex(h);
2545 else
2546 hl_decode_mb_444_simple_8(h);
2547 } else if (is_complex) {
2548 hl_decode_mb_complex(h);
2549 } else if (h->pixel_shift) {
2550 hl_decode_mb_simple_16(h);
2551 } else
2552 hl_decode_mb_simple_8(h);
2553 }
2554
2555 static int pred_weight_table(H264Context *h)
2556 {
2557 int list, i;
2558 int luma_def, chroma_def;
2559
2560 h->use_weight = 0;
2561 h->use_weight_chroma = 0;
2562 h->luma_log2_weight_denom = get_ue_golomb(&h->gb);
2563 if (h->sps.chroma_format_idc)
2564 h->chroma_log2_weight_denom = get_ue_golomb(&h->gb);
2565 luma_def = 1 << h->luma_log2_weight_denom;
2566 chroma_def = 1 << h->chroma_log2_weight_denom;
2567
2568 for (list = 0; list < 2; list++) {
2569 h->luma_weight_flag[list] = 0;
2570 h->chroma_weight_flag[list] = 0;
2571 for (i = 0; i < h->ref_count[list]; i++) {
2572 int luma_weight_flag, chroma_weight_flag;
2573
2574 luma_weight_flag = get_bits1(&h->gb);
2575 if (luma_weight_flag) {
2576 h->luma_weight[i][list][0] = get_se_golomb(&h->gb);
2577 h->luma_weight[i][list][1] = get_se_golomb(&h->gb);
2578 if (h->luma_weight[i][list][0] != luma_def ||
2579 h->luma_weight[i][list][1] != 0) {
2580 h->use_weight = 1;
2581 h->luma_weight_flag[list] = 1;
2582 }
2583 } else {
2584 h->luma_weight[i][list][0] = luma_def;
2585 h->luma_weight[i][list][1] = 0;
2586 }
2587
2588 if (h->sps.chroma_format_idc) {
2589 chroma_weight_flag = get_bits1(&h->gb);
2590 if (chroma_weight_flag) {
2591 int j;
2592 for (j = 0; j < 2; j++) {
2593 h->chroma_weight[i][list][j][0] = get_se_golomb(&h->gb);
2594 h->chroma_weight[i][list][j][1] = get_se_golomb(&h->gb);
2595 if (h->chroma_weight[i][list][j][0] != chroma_def ||
2596 h->chroma_weight[i][list][j][1] != 0) {
2597 h->use_weight_chroma = 1;
2598 h->chroma_weight_flag[list] = 1;
2599 }
2600 }
2601 } else {
2602 int j;
2603 for (j = 0; j < 2; j++) {
2604 h->chroma_weight[i][list][j][0] = chroma_def;
2605 h->chroma_weight[i][list][j][1] = 0;
2606 }
2607 }
2608 }
2609 }
2610 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2611 break;
2612 }
2613 h->use_weight = h->use_weight || h->use_weight_chroma;
2614 return 0;
2615 }
2616
2617 /**
2618 * Initialize implicit_weight table.
2619 * @param field 0/1 initialize the weight for interlaced MBAFF
2620 * -1 initializes the rest
2621 */
2622 static void implicit_weight_table(H264Context *h, int field)
2623 {
2624 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2625
2626 for (i = 0; i < 2; i++) {
2627 h->luma_weight_flag[i] = 0;
2628 h->chroma_weight_flag[i] = 0;
2629 }
2630
2631 if (field < 0) {
2632 if (h->picture_structure == PICT_FRAME) {
2633 cur_poc = h->cur_pic_ptr->poc;
2634 } else {
2635 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
2636 }
2637 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
2638 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2639 h->use_weight = 0;
2640 h->use_weight_chroma = 0;
2641 return;
2642 }
2643 ref_start = 0;
2644 ref_count0 = h->ref_count[0];
2645 ref_count1 = h->ref_count[1];
2646 } else {
2647 cur_poc = h->cur_pic_ptr->field_poc[field];
2648 ref_start = 16;
2649 ref_count0 = 16 + 2 * h->ref_count[0];
2650 ref_count1 = 16 + 2 * h->ref_count[1];
2651 }
2652
2653 h->use_weight = 2;
2654 h->use_weight_chroma = 2;
2655 h->luma_log2_weight_denom = 5;
2656 h->chroma_log2_weight_denom = 5;
2657
2658 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2659 int poc0 = h->ref_list[0][ref0].poc;
2660 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2661 int w = 32;
2662 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2663 int poc1 = h->ref_list[1][ref1].poc;
2664 int td = av_clip(poc1 - poc0, -128, 127);
2665 if (td) {
2666 int tb = av_clip(cur_poc - poc0, -128, 127);
2667 int tx = (16384 + (FFABS(td) >> 1)) / td;
2668 int dist_scale_factor = (tb * tx + 32) >> 8;
2669 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2670 w = 64 - dist_scale_factor;
2671 }
2672 }
2673 if (field < 0) {
2674 h->implicit_weight[ref0][ref1][0] =
2675 h->implicit_weight[ref0][ref1][1] = w;
2676 } else {
2677 h->implicit_weight[ref0][ref1][field] = w;
2678 }
2679 }
2680 }
2681 }
2682
2683 /**
2684 * instantaneous decoder refresh.
2685 */
2686 static void idr(H264Context *h)
2687 {
2688 ff_h264_remove_all_refs(h);
2689 h->prev_frame_num = 0;
2690 h->prev_frame_num_offset = 0;
2691 h->prev_poc_msb =
2692 h->prev_poc_lsb = 0;
2693 }
2694
2695 /* forget old pics after a seek */
2696 static void flush_change(H264Context *h)
2697 {
2698 int i;
2699 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2700 h->last_pocs[i] = INT_MIN;
2701 h->outputed_poc = h->next_outputed_poc = INT_MIN;
2702 h->prev_interlaced_frame = 1;
2703 idr(h);
2704 if (h->cur_pic_ptr)
2705 h->cur_pic_ptr->reference = 0;
2706 h->first_field = 0;
2707 memset(h->ref_list[0], 0, sizeof(h->ref_list[0]));
2708 memset(h->ref_list[1], 0, sizeof(h->ref_list[1]));
2709 memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0]));
2710 memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1]));
2711 ff_h264_reset_sei(h);
2712 }
2713
2714 /* forget old pics after a seek */
2715 static void flush_dpb(AVCodecContext *avctx)
2716 {
2717 H264Context *h = avctx->priv_data;
2718 int i;
2719
2720 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
2721 if (h->delayed_pic[i])
2722 h->delayed_pic[i]->reference = 0;
2723 h->delayed_pic[i] = NULL;
2724 }
2725
2726 flush_change(h);
2727
2728 if (h->DPB)
2729 for (i = 0; i < MAX_PICTURE_COUNT; i++)
2730 unref_picture(h, &h->DPB[i]);
2731 h->cur_pic_ptr = NULL;
2732 unref_picture(h, &h->cur_pic);
2733
2734 h->mb_x = h->mb_y = 0;
2735
2736 h->parse_context.state = -1;
2737 h->parse_context.frame_start_found = 0;
2738 h->parse_context.overread = 0;
2739 h->parse_context.overread_index = 0;
2740 h->parse_context.index = 0;
2741 h->parse_context.last_index = 0;
2742 }
2743
2744 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
2745 {
2746 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2747 int field_poc[2];
2748
2749 h->frame_num_offset = h->prev_frame_num_offset;
2750 if (h->frame_num < h->prev_frame_num)
2751 h->frame_num_offset += max_frame_num;
2752
2753 if (h->sps.poc_type == 0) {
2754 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2755
2756 if (h->poc_lsb < h->prev_poc_lsb &&
2757 h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2758 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2759 else if (h->poc_lsb > h->prev_poc_lsb &&
2760 h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2761 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2762 else
2763 h->poc_msb = h->prev_poc_msb;
2764 field_poc[0] =
2765 field_poc[1] = h->poc_msb + h->poc_lsb;
2766 if (h->picture_structure == PICT_FRAME)
2767 field_poc[1] += h->delta_poc_bottom;
2768 } else if (h->sps.poc_type == 1) {
2769 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2770 int i;
2771
2772 if (h->sps.poc_cycle_length != 0)
2773 abs_frame_num = h->frame_num_offset + h->frame_num;
2774 else
2775 abs_frame_num = 0;
2776
2777 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2778 abs_frame_num--;
2779
2780 expected_delta_per_poc_cycle = 0;
2781 for (i = 0; i < h->sps.poc_cycle_length; i++)
2782 // FIXME integrate during sps parse
2783 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2784
2785 if (abs_frame_num > 0) {
2786 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2787 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2788
2789 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2790 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2791 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2792 } else
2793 expectedpoc = 0;
2794
2795 if (h->nal_ref_idc == 0)
2796 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2797
2798 field_poc[0] = expectedpoc + h->delta_poc[0];
2799 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2800
2801 if (h->picture_structure == PICT_FRAME)
2802 field_poc[1] += h->delta_poc[1];
2803 } else {
2804 int poc = 2 * (h->frame_num_offset + h->frame_num);
2805
2806 if (!h->nal_ref_idc)
2807 poc--;
2808
2809 field_poc[0] = poc;
2810 field_poc[1] = poc;
2811 }
2812
2813 if (h->picture_structure != PICT_BOTTOM_FIELD)
2814 pic_field_poc[0] = field_poc[0];
2815 if (h->picture_structure