asyncts: cosmetics: reindent
[libav.git] / libavcodec / h264.c
CommitLineData
0da71265 1/*
ff3d4310 2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
0da71265
MN
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
2912e87a 5 * This file is part of Libav.
b78e7197 6 *
2912e87a 7 * Libav is free software; you can redistribute it and/or
0da71265
MN
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
b78e7197 10 * version 2.1 of the License, or (at your option) any later version.
0da71265 11 *
2912e87a 12 * Libav is distributed in the hope that it will be useful,
0da71265
MN
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
2912e87a 18 * License along with Libav; if not, write to the Free Software
5509bffa 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
0da71265 20 */
115329f1 21
0da71265 22/**
ba87f080 23 * @file
0da71265
MN
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
737eb597 28#include "libavutil/imgutils.h"
40e5d31b 29#include "internal.h"
55b9ef18
DB
30#include "cabac.h"
31#include "cabac_functions.h"
0da71265
MN
32#include "dsputil.h"
33#include "avcodec.h"
34#include "mpegvideo.h"
26b4fe82 35#include "h264.h"
0da71265 36#include "h264data.h"
188d3c51 37#include "h264_mvpred.h"
0da71265 38#include "golomb.h"
199436b9 39#include "mathops.h"
626464fb 40#include "rectangle.h"
6a9c8594 41#include "thread.h"
369122dd 42#include "vdpau_internal.h"
cfa5a81e 43#include "libavutil/avassert.h"
0da71265 44
e5d40372 45// #undef NDEBUG
0da71265
MN
46#include <assert.h>
47
0becb078
DB
48const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
49
e5d40372
DB
50static const uint8_t rem6[QP_MAX_NUM + 1] = {
51 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
52 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
53 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
acd8d10f
PI
54};
55
e5d40372
DB
56static const uint8_t div6[QP_MAX_NUM + 1] = {
57 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
58 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
59 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
acd8d10f
PI
60};
61
716d413c
AK
62static const enum AVPixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
63 AV_PIX_FMT_DXVA2_VLD,
64 AV_PIX_FMT_VAAPI_VLD,
65 AV_PIX_FMT_VDA_VLD,
66 AV_PIX_FMT_YUVJ420P,
67 AV_PIX_FMT_NONE
0435fb16
BC
68};
69
0da71265 70/**
58c42af7
DB
71 * Check if the top & left blocks are available if needed and
72 * change the dc mode so it only uses the available blocks.
0da71265 73 */
e5d40372
DB
74int ff_h264_check_intra4x4_pred_mode(H264Context *h)
75{
76 MpegEncContext *const s = &h->s;
77 static const int8_t top[12] = {
78 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
79 };
80 static const int8_t left[12] = {
81 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
82 };
2bedc0e8
MN
83 int i;
84
e5d40372
DB
85 if (!(h->top_samples_available & 0x8000)) {
86 for (i = 0; i < 4; i++) {
87 int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
88 if (status < 0) {
89 av_log(h->s.avctx, AV_LOG_ERROR,
90 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
91 status, s->mb_x, s->mb_y);
2bedc0e8 92 return -1;
e5d40372
DB
93 } else if (status) {
94 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
2bedc0e8
MN
95 }
96 }
97 }
98
e5d40372
DB
99 if ((h->left_samples_available & 0x8888) != 0x8888) {
100 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
101 for (i = 0; i < 4; i++)
102 if (!(h->left_samples_available & mask[i])) {
103 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
104 if (status < 0) {
105 av_log(h->s.avctx, AV_LOG_ERROR,
106 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
107 status, s->mb_x, s->mb_y);
2bedc0e8 108 return -1;
e5d40372
DB
109 } else if (status) {
110 h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
2bedc0e8
MN
111 }
112 }
2bedc0e8
MN
113 }
114
115 return 0;
e5d40372 116} // FIXME cleanup like ff_h264_check_intra_pred_mode
2bedc0e8
MN
117
118/**
58c42af7
DB
119 * Check if the top & left blocks are available if needed and
120 * change the dc mode so it only uses the available blocks.
2bedc0e8 121 */
e5d40372
DB
122int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
123{
124 MpegEncContext *const s = &h->s;
125 static const int8_t top[7] = { LEFT_DC_PRED8x8, 1, -1, -1 };
126 static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
127
128 if (mode > 6U) {
129 av_log(h->s.avctx, AV_LOG_ERROR,
130 "out of range intra chroma pred mode at %d %d\n",
131 s->mb_x, s->mb_y);
7440fe83 132 return -1;
5175b937 133 }
115329f1 134
e5d40372
DB
135 if (!(h->top_samples_available & 0x8000)) {
136 mode = top[mode];
137 if (mode < 0) {
138 av_log(h->s.avctx, AV_LOG_ERROR,
139 "top block unavailable for requested intra mode at %d %d\n",
140 s->mb_x, s->mb_y);
0da71265
MN
141 return -1;
142 }
143 }
115329f1 144
e5d40372
DB
145 if ((h->left_samples_available & 0x8080) != 0x8080) {
146 mode = left[mode];
147 if (is_chroma && (h->left_samples_available & 0x8080)) {
148 // mad cow disease mode, aka MBAFF + constrained_intra_pred
149 mode = ALZHEIMER_DC_L0T_PRED8x8 +
150 (!(h->left_samples_available & 0x8000)) +
151 2 * (mode == DC_128_PRED8x8);
d1d10e91 152 }
e5d40372
DB
153 if (mode < 0) {
154 av_log(h->s.avctx, AV_LOG_ERROR,
155 "left block unavailable for requested intra mode at %d %d\n",
156 s->mb_x, s->mb_y);
0da71265 157 return -1;
115329f1 158 }
0da71265
MN
159 }
160
161 return mode;
162}
163
e5d40372
DB
164const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
165 int *dst_length, int *consumed, int length)
166{
0da71265
MN
167 int i, si, di;
168 uint8_t *dst;
24456882 169 int bufidx;
0da71265 170
e5d40372
DB
171 // src[0]&0x80; // forbidden bit
172 h->nal_ref_idc = src[0] >> 5;
173 h->nal_unit_type = src[0] & 0x1F;
0da71265 174
e5d40372
DB
175 src++;
176 length--;
e08715d3 177
58db34aa
RB
178#define STARTCODE_TEST \
179 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
180 if (src[i + 2] != 3) { \
181 /* startcode, so we must be past the end */ \
182 length = i; \
183 } \
184 break; \
185 }
b250f9c6 186#if HAVE_FAST_UNALIGNED
58db34aa
RB
187#define FIND_FIRST_ZERO \
188 if (i > 0 && !src[i]) \
189 i--; \
190 while (src[i]) \
191 i++
e5d40372 192#if HAVE_FAST_64BIT
e5d40372
DB
193 for (i = 0; i + 1 < length; i += 9) {
194 if (!((~AV_RN64A(src + i) &
195 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
196 0x8000800080008080ULL))
58db34aa
RB
197 continue;
198 FIND_FIRST_ZERO;
199 STARTCODE_TEST;
200 i -= 7;
201 }
e5d40372 202#else
e5d40372
DB
203 for (i = 0; i + 1 < length; i += 5) {
204 if (!((~AV_RN32A(src + i) &
205 (AV_RN32A(src + i) - 0x01000101U)) &
206 0x80008080U))
e08715d3 207 continue;
58db34aa
RB
208 FIND_FIRST_ZERO;
209 STARTCODE_TEST;
210 i -= 3;
211 }
212#endif
e08715d3 213#else
e5d40372
DB
214 for (i = 0; i + 1 < length; i += 2) {
215 if (src[i])
216 continue;
217 if (i > 0 && src[i - 1] == 0)
218 i--;
58db34aa 219 STARTCODE_TEST;
0da71265 220 }
58db34aa 221#endif
0da71265 222
e5d40372
DB
223 if (i >= length - 1) { // no escaped 0
224 *dst_length = length;
225 *consumed = length + 1; // +1 for the header
115329f1 226 return src;
0da71265
MN
227 }
228
e5d40372
DB
229 // use second escape buffer for inter data
230 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
231 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx],
232 length + FF_INPUT_BUFFER_PADDING_SIZE);
233 dst = h->rbsp_buffer[bufidx];
0da71265 234
e5d40372 235 if (dst == NULL)
ac658be5 236 return NULL;
ac658be5 237
593af7cd 238 memcpy(dst, src, i);
e5d40372
DB
239 si = di = i;
240 while (si + 2 < length) {
241 // remove escapes (very rare 1:2^22)
242 if (src[si + 2] > 3) {
243 dst[di++] = src[si++];
244 dst[di++] = src[si++];
245 } else if (src[si] == 0 && src[si + 1] == 0) {
246 if (src[si + 2] == 3) { // escape
247 dst[di++] = 0;
248 dst[di++] = 0;
249 si += 3;
c8470cc1 250 continue;
e5d40372 251 } else // next start code
593af7cd 252 goto nsc;
0da71265
MN
253 }
254
e5d40372 255 dst[di++] = src[si++];
0da71265 256 }
e5d40372
DB
257 while (si < length)
258 dst[di++] = src[si++];
593af7cd 259nsc:
0da71265 260
e5d40372 261 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
d4369630 262
e5d40372
DB
263 *dst_length = di;
264 *consumed = si + 1; // +1 for the header
265 /* FIXME store exact number of bits in the getbitcontext
266 * (it is needed for decoding) */
0da71265
MN
267 return dst;
268}
269
85297319
DEP
270/**
271 * Identify the exact end of the bitstream
272 * @return the length of the trailing, or 0 if damaged
273 */
b691fd7a 274static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
e5d40372
DB
275{
276 int v = *src;
0da71265
MN
277 int r;
278
a9c9a240 279 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
0da71265 280
e5d40372
DB
281 for (r = 1; r < 9; r++) {
282 if (v & 1)
283 return r;
284 v >>= 1;
0da71265
MN
285 }
286 return 0;
287}
288
e5d40372
DB
289static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
290 int height, int y_offset, int list)
291{
292 int raw_my = h->mv_cache[list][scan8[n]][1];
293 int filter_height = (raw_my & 3) ? 2 : 0;
294 int full_my = (raw_my >> 2) + y_offset;
295 int top = full_my - filter_height;
296 int bottom = full_my + filter_height + height;
6a9c8594
AS
297
298 return FFMAX(abs(top), bottom);
299}
300
e5d40372
DB
301static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
302 int height, int y_offset, int list0,
303 int list1, int *nrefs)
304{
305 MpegEncContext *const s = &h->s;
6a9c8594
AS
306 int my;
307
e5d40372 308 y_offset += 16 * (s->mb_y >> MB_FIELD);
6a9c8594 309
e5d40372
DB
310 if (list0) {
311 int ref_n = h->ref_cache[0][scan8[n]];
312 Picture *ref = &h->ref_list[0][ref_n];
6a9c8594
AS
313
314 // Error resilience puts the current picture in the ref list.
315 // Don't try to wait on these as it will cause a deadlock.
316 // Fields can wait on each other, though.
e5d40372
DB
317 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
318 (ref->f.reference & 3) != s->picture_structure) {
6a9c8594 319 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
e5d40372
DB
320 if (refs[0][ref_n] < 0)
321 nrefs[0] += 1;
6a9c8594
AS
322 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
323 }
324 }
325
e5d40372
DB
326 if (list1) {
327 int ref_n = h->ref_cache[1][scan8[n]];
328 Picture *ref = &h->ref_list[1][ref_n];
6a9c8594 329
e5d40372
DB
330 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
331 (ref->f.reference & 3) != s->picture_structure) {
6a9c8594 332 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
e5d40372
DB
333 if (refs[1][ref_n] < 0)
334 nrefs[1] += 1;
6a9c8594
AS
335 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
336 }
337 }
338}
339
340/**
341 * Wait until all reference frames are available for MC operations.
342 *
343 * @param h the H264 context
344 */
e5d40372
DB
345static void await_references(H264Context *h)
346{
347 MpegEncContext *const s = &h->s;
348 const int mb_xy = h->mb_xy;
657ccb5a 349 const int mb_type = s->current_picture.f.mb_type[mb_xy];
6a9c8594 350 int refs[2][48];
e5d40372 351 int nrefs[2] = { 0 };
6a9c8594
AS
352 int ref, list;
353
354 memset(refs, -1, sizeof(refs));
355
e5d40372 356 if (IS_16X16(mb_type)) {
6a9c8594 357 get_lowest_part_y(h, refs, 0, 16, 0,
e5d40372
DB
358 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
359 } else if (IS_16X8(mb_type)) {
6a9c8594 360 get_lowest_part_y(h, refs, 0, 8, 0,
e5d40372 361 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
6a9c8594 362 get_lowest_part_y(h, refs, 8, 8, 8,
e5d40372
DB
363 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
364 } else if (IS_8X16(mb_type)) {
6a9c8594 365 get_lowest_part_y(h, refs, 0, 16, 0,
e5d40372 366 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
6a9c8594 367 get_lowest_part_y(h, refs, 4, 16, 0,
e5d40372
DB
368 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
369 } else {
6a9c8594
AS
370 int i;
371
372 assert(IS_8X8(mb_type));
373
e5d40372
DB
374 for (i = 0; i < 4; i++) {
375 const int sub_mb_type = h->sub_mb_type[i];
376 const int n = 4 * i;
377 int y_offset = (i & 2) << 2;
378
379 if (IS_SUB_8X8(sub_mb_type)) {
380 get_lowest_part_y(h, refs, n, 8, y_offset,
381 IS_DIR(sub_mb_type, 0, 0),
382 IS_DIR(sub_mb_type, 0, 1),
383 nrefs);
384 } else if (IS_SUB_8X4(sub_mb_type)) {
385 get_lowest_part_y(h, refs, n, 4, y_offset,
386 IS_DIR(sub_mb_type, 0, 0),
387 IS_DIR(sub_mb_type, 0, 1),
388 nrefs);
389 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
390 IS_DIR(sub_mb_type, 0, 0),
391 IS_DIR(sub_mb_type, 0, 1),
392 nrefs);
393 } else if (IS_SUB_4X8(sub_mb_type)) {
394 get_lowest_part_y(h, refs, n, 8, y_offset,
395 IS_DIR(sub_mb_type, 0, 0),
396 IS_DIR(sub_mb_type, 0, 1),
397 nrefs);
398 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
399 IS_DIR(sub_mb_type, 0, 0),
400 IS_DIR(sub_mb_type, 0, 1),
401 nrefs);
402 } else {
6a9c8594
AS
403 int j;
404 assert(IS_SUB_4X4(sub_mb_type));
e5d40372
DB
405 for (j = 0; j < 4; j++) {
406 int sub_y_offset = y_offset + 2 * (j & 2);
407 get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
408 IS_DIR(sub_mb_type, 0, 0),
409 IS_DIR(sub_mb_type, 0, 1),
410 nrefs);
6a9c8594
AS
411 }
412 }
413 }
414 }
415
e5d40372
DB
416 for (list = h->list_count - 1; list >= 0; list--)
417 for (ref = 0; ref < 48 && nrefs[list]; ref++) {
6a9c8594 418 int row = refs[list][ref];
e5d40372
DB
419 if (row >= 0) {
420 Picture *ref_pic = &h->ref_list[list][ref];
421 int ref_field = ref_pic->f.reference - 1;
6a9c8594 422 int ref_field_picture = ref_pic->field_picture;
e5d40372 423 int pic_height = 16 * s->mb_height >> ref_field_picture;
6a9c8594
AS
424
425 row <<= MB_MBAFF;
426 nrefs[list]--;
427
e5d40372
DB
428 if (!FIELD_PICTURE && ref_field_picture) { // frame referencing two fields
429 ff_thread_await_progress(&ref_pic->f,
430 FFMIN((row >> 1) - !(row & 1),
431 pic_height - 1),
432 1);
433 ff_thread_await_progress(&ref_pic->f,
434 FFMIN((row >> 1), pic_height - 1),
435 0);
436 } else if (FIELD_PICTURE && !ref_field_picture) { // field referencing one field of a frame
437 ff_thread_await_progress(&ref_pic->f,
438 FFMIN(row * 2 + ref_field,
439 pic_height - 1),
440 0);
441 } else if (FIELD_PICTURE) {
442 ff_thread_await_progress(&ref_pic->f,
443 FFMIN(row, pic_height - 1),
444 ref_field);
445 } else {
446 ff_thread_await_progress(&ref_pic->f,
447 FFMIN(row, pic_height - 1),
448 0);
6a9c8594
AS
449 }
450 }
451 }
66c6b5e2 452}
66c6b5e2 453
e5d40372
DB
454static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
455 int n, int square, int height,
456 int delta, int list,
457 uint8_t *dest_y, uint8_t *dest_cb,
458 uint8_t *dest_cr,
459 int src_x_offset, int src_y_offset,
460 qpel_mc_func *qpix_op,
461 h264_chroma_mc_func chroma_op,
462 int pixel_shift, int chroma_idc)
05fb63f5 463{
e5d40372
DB
464 MpegEncContext *const s = &h->s;
465 const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
466 int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
467 const int luma_xy = (mx & 3) + ((my & 3) << 2);
468 int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
469 uint8_t *src_y = pic->f.data[0] + offset;
470 uint8_t *src_cb, *src_cr;
471 int extra_width = h->emu_edge_width;
472 int extra_height = h->emu_edge_height;
473 int emu = 0;
474 const int full_mx = mx >> 2;
475 const int full_my = my >> 2;
476 const int pic_width = 16 * s->mb_width;
477 const int pic_height = 16 * s->mb_height >> MB_FIELD;
229d263c 478 int ysh;
115329f1 479
e5d40372
DB
480 if (mx & 7)
481 extra_width -= 3;
482 if (my & 7)
483 extra_height -= 3;
484
485 if (full_mx < 0 - extra_width ||
486 full_my < 0 - extra_height ||
487 full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
488 full_my + 16 /*FIXME*/ > pic_height + extra_height) {
489 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
490 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
491 h->mb_linesize,
492 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
493 full_my - 2, pic_width, pic_height);
494 src_y = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
495 emu = 1;
496 }
497
498 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
499 if (!square)
5d18eaad 500 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
115329f1 501
e5d40372
DB
502 if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
503 return;
7b442ad9 504
e5d40372 505 if (chroma_idc == 3 /* yuv444 */) {
657ccb5a 506 src_cb = pic->f.data[1] + offset;
e5d40372
DB
507 if (emu) {
508 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
509 src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
510 h->mb_linesize,
511 16 + 5, 16 + 5 /*FIXME*/,
512 full_mx - 2, full_my - 2,
513 pic_width, pic_height);
514 src_cb = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
c90b9442 515 }
e5d40372
DB
516 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
517 if (!square)
c90b9442 518 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
c90b9442 519
657ccb5a 520 src_cr = pic->f.data[2] + offset;
e5d40372
DB
521 if (emu) {
522 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
523 src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
524 h->mb_linesize,
525 16 + 5, 16 + 5 /*FIXME*/,
526 full_mx - 2, full_my - 2,
527 pic_width, pic_height);
528 src_cr = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
c90b9442 529 }
e5d40372
DB
530 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
531 if (!square)
c90b9442 532 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
c90b9442
JGG
533 return;
534 }
535
05fb63f5 536 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
e5d40372 537 if (chroma_idc == 1 /* yuv420 */ && MB_FIELD) {
5d18eaad 538 // chroma offset when predicting from a field of opposite parity
e5d40372
DB
539 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
540 emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
5d18eaad 541 }
229d263c 542
e5d40372
DB
543 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
544 (my >> ysh) * h->mb_uvlinesize;
545 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
546 (my >> ysh) * h->mb_uvlinesize;
5d18eaad 547
e5d40372 548 if (emu) {
05fb63f5
RB
549 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
550 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
551 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
e5d40372 552 src_cb = s->edge_emu_buffer;
0da71265 553 }
e5d40372
DB
554 chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
555 height >> (chroma_idc == 1 /* yuv420 */),
556 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
0da71265 557
e5d40372 558 if (emu) {
05fb63f5
RB
559 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
560 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
561 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
e5d40372 562 src_cr = s->edge_emu_buffer;
0da71265 563 }
05fb63f5 564 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
e5d40372 565 mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
0da71265
MN
566}
567
e5d40372
DB
568static av_always_inline void mc_part_std(H264Context *h, int n, int square,
569 int height, int delta,
570 uint8_t *dest_y, uint8_t *dest_cb,
571 uint8_t *dest_cr,
572 int x_offset, int y_offset,
573 qpel_mc_func *qpix_put,
574 h264_chroma_mc_func chroma_put,
575 qpel_mc_func *qpix_avg,
576 h264_chroma_mc_func chroma_avg,
577 int list0, int list1,
578 int pixel_shift, int chroma_idc)
05fb63f5 579{
e5d40372
DB
580 MpegEncContext *const s = &h->s;
581 qpel_mc_func *qpix_op = qpix_put;
582 h264_chroma_mc_func chroma_op = chroma_put;
115329f1 583
e5d40372 584 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
05fb63f5 585 if (chroma_idc == 3 /* yuv444 */) {
e5d40372
DB
586 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
587 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
05fb63f5 588 } else if (chroma_idc == 2 /* yuv422 */) {
e5d40372
DB
589 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
590 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
591 } else { /* yuv420 */
592 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
593 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
c90b9442 594 }
e5d40372
DB
595 x_offset += 8 * s->mb_x;
596 y_offset += 8 * (s->mb_y >> MB_FIELD);
115329f1 597
e5d40372
DB
598 if (list0) {
599 Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
c2d33742 600 mc_dir_part(h, ref, n, square, height, delta, 0,
e5d40372
DB
601 dest_y, dest_cb, dest_cr, x_offset, y_offset,
602 qpix_op, chroma_op, pixel_shift, chroma_idc);
0da71265 603
e5d40372
DB
604 qpix_op = qpix_avg;
605 chroma_op = chroma_avg;
0da71265
MN
606 }
607
e5d40372
DB
608 if (list1) {
609 Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
c2d33742 610 mc_dir_part(h, ref, n, square, height, delta, 1,
e5d40372
DB
611 dest_y, dest_cb, dest_cr, x_offset, y_offset,
612 qpix_op, chroma_op, pixel_shift, chroma_idc);
0da71265
MN
613 }
614}
615
e5d40372
DB
616static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
617 int height, int delta,
618 uint8_t *dest_y, uint8_t *dest_cb,
619 uint8_t *dest_cr,
620 int x_offset, int y_offset,
621 qpel_mc_func *qpix_put,
622 h264_chroma_mc_func chroma_put,
623 h264_weight_func luma_weight_op,
624 h264_weight_func chroma_weight_op,
625 h264_biweight_func luma_weight_avg,
626 h264_biweight_func chroma_weight_avg,
627 int list0, int list1,
628 int pixel_shift, int chroma_idc)
629{
630 MpegEncContext *const s = &h->s;
c2d33742 631 int chroma_height;
9f2d1b4f 632
e5d40372 633 dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
05fb63f5 634 if (chroma_idc == 3 /* yuv444 */) {
e5d40372 635 chroma_height = height;
c90b9442 636 chroma_weight_avg = luma_weight_avg;
e5d40372
DB
637 chroma_weight_op = luma_weight_op;
638 dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
639 dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
05fb63f5 640 } else if (chroma_idc == 2 /* yuv422 */) {
c2d33742 641 chroma_height = height;
e5d40372
DB
642 dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
643 dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
644 } else { /* yuv420 */
c2d33742 645 chroma_height = height >> 1;
e5d40372
DB
646 dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
647 dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
c90b9442 648 }
e5d40372
DB
649 x_offset += 8 * s->mb_x;
650 y_offset += 8 * (s->mb_y >> MB_FIELD);
115329f1 651
e5d40372 652 if (list0 && list1) {
9f2d1b4f
LM
653 /* don't optimize for luma-only case, since B-frames usually
654 * use implicit weights => chroma too. */
a394959b
JG
655 uint8_t *tmp_cb = h->bipred_scratchpad;
656 uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
657 uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
e5d40372
DB
658 int refn0 = h->ref_cache[0][scan8[n]];
659 int refn1 = h->ref_cache[1][scan8[n]];
9f2d1b4f 660
c2d33742 661 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
9f2d1b4f 662 dest_y, dest_cb, dest_cr,
05fb63f5
RB
663 x_offset, y_offset, qpix_put, chroma_put,
664 pixel_shift, chroma_idc);
c2d33742 665 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
9f2d1b4f 666 tmp_y, tmp_cb, tmp_cr,
05fb63f5
RB
667 x_offset, y_offset, qpix_put, chroma_put,
668 pixel_shift, chroma_idc);
9f2d1b4f 669
e5d40372
DB
670 if (h->use_weight == 2) {
671 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y & 1];
9f2d1b4f 672 int weight1 = 64 - weight0;
e5d40372
DB
673 luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
674 height, 5, weight0, weight1, 0);
c2d33742
RB
675 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
676 chroma_height, 5, weight0, weight1, 0);
677 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
678 chroma_height, 5, weight0, weight1, 0);
e5d40372
DB
679 } else {
680 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
681 h->luma_log2_weight_denom,
682 h->luma_weight[refn0][0][0],
683 h->luma_weight[refn1][1][0],
684 h->luma_weight[refn0][0][1] +
685 h->luma_weight[refn1][1][1]);
686 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
687 h->chroma_log2_weight_denom,
688 h->chroma_weight[refn0][0][0][0],
689 h->chroma_weight[refn1][1][0][0],
690 h->chroma_weight[refn0][0][0][1] +
691 h->chroma_weight[refn1][1][0][1]);
692 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
693 h->chroma_log2_weight_denom,
694 h->chroma_weight[refn0][0][1][0],
695 h->chroma_weight[refn1][1][1][0],
696 h->chroma_weight[refn0][0][1][1] +
697 h->chroma_weight[refn1][1][1][1]);
9f2d1b4f 698 }
e5d40372
DB
699 } else {
700 int list = list1 ? 1 : 0;
701 int refn = h->ref_cache[list][scan8[n]];
702 Picture *ref = &h->ref_list[list][refn];
c2d33742 703 mc_dir_part(h, ref, n, square, height, delta, list,
9f2d1b4f 704 dest_y, dest_cb, dest_cr, x_offset, y_offset,
05fb63f5 705 qpix_put, chroma_put, pixel_shift, chroma_idc);
9f2d1b4f 706
e5d40372
DB
707 luma_weight_op(dest_y, h->mb_linesize, height,
708 h->luma_log2_weight_denom,
709 h->luma_weight[refn][list][0],
710 h->luma_weight[refn][list][1]);
711 if (h->use_weight_chroma) {
712 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
713 h->chroma_log2_weight_denom,
714 h->chroma_weight[refn][list][0][0],
715 h->chroma_weight[refn][list][0][1]);
716 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
717 h->chroma_log2_weight_denom,
718 h->chroma_weight[refn][list][1][0],
719 h->chroma_weight[refn][list][1][1]);
9f2d1b4f
LM
720 }
721 }
722}
723
e5d40372
DB
724static av_always_inline void prefetch_motion(H264Context *h, int list,
725 int pixel_shift, int chroma_idc)
27209bb1 726{
513fbd8e
LM
727 /* fetch pixels for estimated mv 4 macroblocks ahead
728 * optimized for 64byte cache lines */
e5d40372 729 MpegEncContext *const s = &h->s;
513fbd8e 730 const int refn = h->ref_cache[list][scan8[0]];
e5d40372
DB
731 if (refn >= 0) {
732 const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * s->mb_x + 8;
733 const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * s->mb_y;
657ccb5a 734 uint8_t **src = h->ref_list[list][refn].f.data;
e5d40372
DB
735 int off = (mx << pixel_shift) +
736 (my + (s->mb_x & 3) * 4) * h->mb_linesize +
737 (64 << pixel_shift);
738 s->dsp.prefetch(src[0] + off, s->linesize, 4);
05fb63f5 739 if (chroma_idc == 3 /* yuv444 */) {
e5d40372
DB
740 s->dsp.prefetch(src[1] + off, s->linesize, 4);
741 s->dsp.prefetch(src[2] + off, s->linesize, 4);
742 } else {
743 off = ((mx >> 1) << pixel_shift) +
744 ((my >> 1) + (s->mb_x & 7)) * s->uvlinesize +
745 (64 << pixel_shift);
746 s->dsp.prefetch(src[1] + off, src[2] - src[1], 2);
c90b9442 747 }
513fbd8e
LM
748 }
749}
750
e5d40372
DB
751static void free_tables(H264Context *h, int free_rbsp)
752{
7978debd 753 int i;
afebe2f7 754 H264Context *hx;
6a9c8594 755
0da71265 756 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
757 av_freep(&h->chroma_pred_mode_table);
758 av_freep(&h->cbp_table);
9e528114
LA
759 av_freep(&h->mvd_table[0]);
760 av_freep(&h->mvd_table[1]);
5ad984c9 761 av_freep(&h->direct_table);
0da71265
MN
762 av_freep(&h->non_zero_count);
763 av_freep(&h->slice_table_base);
e5d40372 764 h->slice_table = NULL;
c988f975 765 av_freep(&h->list_counts);
e5017ab8 766
0da71265 767 av_freep(&h->mb2b_xy);
d43c1922 768 av_freep(&h->mb2br_xy);
9f2d1b4f 769
e5d40372 770 for (i = 0; i < MAX_THREADS; i++) {
afebe2f7 771 hx = h->thread_context[i];
e5d40372
DB
772 if (!hx)
773 continue;
afebe2f7
774 av_freep(&hx->top_borders[1]);
775 av_freep(&hx->top_borders[0]);
a394959b 776 av_freep(&hx->bipred_scratchpad);
e5d40372 777 if (free_rbsp) {
fcb7e535
RB
778 av_freep(&hx->rbsp_buffer[1]);
779 av_freep(&hx->rbsp_buffer[0]);
780 hx->rbsp_buffer_size[0] = 0;
781 hx->rbsp_buffer_size[1] = 0;
91078926 782 }
e5d40372
DB
783 if (i)
784 av_freep(&h->thread_context[i]);
afebe2f7 785 }
0da71265
MN
786}
787
e5d40372
DB
788static void init_dequant8_coeff_table(H264Context *h)
789{
790 int i, j, q, x;
791 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
239ea04c 792
e5d40372 793 for (i = 0; i < 6; i++) {
c90b9442 794 h->dequant8_coeff[i] = h->dequant8_buffer[i];
e5d40372
DB
795 for (j = 0; j < i; j++)
796 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
797 64 * sizeof(uint8_t))) {
c90b9442
JGG
798 h->dequant8_coeff[i] = h->dequant8_buffer[j];
799 break;
800 }
e5d40372 801 if (j < i)
c90b9442 802 continue;
239ea04c 803
e5d40372 804 for (q = 0; q < max_qp + 1; q++) {
d9ec210b 805 int shift = div6[q];
e5d40372
DB
806 int idx = rem6[q];
807 for (x = 0; x < 64; x++)
808 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
809 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
810 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
811 }
812 }
813}
814
e5d40372
DB
815static void init_dequant4_coeff_table(H264Context *h)
816{
817 int i, j, q, x;
818 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
819 for (i = 0; i < 6; i++) {
239ea04c 820 h->dequant4_coeff[i] = h->dequant4_buffer[i];
e5d40372
DB
821 for (j = 0; j < i; j++)
822 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
823 16 * sizeof(uint8_t))) {
239ea04c
LM
824 h->dequant4_coeff[i] = h->dequant4_buffer[j];
825 break;
826 }
e5d40372 827 if (j < i)
239ea04c
LM
828 continue;
829
e5d40372 830 for (q = 0; q < max_qp + 1; q++) {
d9ec210b 831 int shift = div6[q] + 2;
e5d40372
DB
832 int idx = rem6[q];
833 for (x = 0; x < 16; x++)
834 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
835 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
836 h->pps.scaling_matrix4[i][x]) << shift;
239ea04c
LM
837 }
838 }
839}
840
e5d40372
DB
841static void init_dequant_tables(H264Context *h)
842{
843 int i, x;
239ea04c 844 init_dequant4_coeff_table(h);
e5d40372 845 if (h->pps.transform_8x8_mode)
239ea04c 846 init_dequant8_coeff_table(h);
e5d40372
DB
847 if (h->sps.transform_bypass) {
848 for (i = 0; i < 6; i++)
849 for (x = 0; x < 16; x++)
850 h->dequant4_coeff[i][0][x] = 1 << 6;
851 if (h->pps.transform_8x8_mode)
852 for (i = 0; i < 6; i++)
853 for (x = 0; x < 64; x++)
854 h->dequant8_coeff[i][0][x] = 1 << 6;
239ea04c
LM
855 }
856}
857
e5d40372
DB
858int ff_h264_alloc_tables(H264Context *h)
859{
860 MpegEncContext *const s = &h->s;
861 const int big_mb_num = s->mb_stride * (s->mb_height + 1);
862 const int row_mb_num = s->mb_stride * 2 * s->avctx->thread_count;
863 int x, y;
864
865 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode,
866 row_mb_num * 8 * sizeof(uint8_t), fail)
867 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count,
868 big_mb_num * 48 * sizeof(uint8_t), fail)
869 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base,
870 (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base), fail)
871 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table,
872 big_mb_num * sizeof(uint16_t), fail)
873 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table,
874 big_mb_num * sizeof(uint8_t), fail)
875 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0],
876 16 * row_mb_num * sizeof(uint8_t), fail);
877 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1],
878 16 * row_mb_num * sizeof(uint8_t), fail);
879 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table,
880 4 * big_mb_num * sizeof(uint8_t), fail);
881 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts,
882 big_mb_num * sizeof(uint8_t), fail)
883
884 memset(h->slice_table_base, -1,
885 (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base));
886 h->slice_table = h->slice_table_base + s->mb_stride * 2 + 1;
887
888 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy,
889 big_mb_num * sizeof(uint32_t), fail);
890 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy,
891 big_mb_num * sizeof(uint32_t), fail);
892 for (y = 0; y < s->mb_height; y++)
893 for (x = 0; x < s->mb_width; x++) {
894 const int mb_xy = x + y * s->mb_stride;
895 const int b_xy = 4 * x + 4 * y * h->b_stride;
896
897 h->mb2b_xy[mb_xy] = b_xy;
898 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * s->mb_stride)));
0da71265 899 }
9f2d1b4f 900
e5d40372 901 if (!h->dequant4_coeff[0])
56edbd81
LM
902 init_dequant_tables(h);
903
0da71265 904 return 0;
e5d40372 905
0da71265 906fail:
91078926 907 free_tables(h, 1);
0da71265
MN
908 return -1;
909}
910
afebe2f7
911/**
912 * Mimic alloc_tables(), but for every context thread.
913 */
e5d40372
DB
914static void clone_tables(H264Context *dst, H264Context *src, int i)
915{
916 MpegEncContext *const s = &src->s;
917 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * s->mb_stride;
918 dst->non_zero_count = src->non_zero_count;
919 dst->slice_table = src->slice_table;
920 dst->cbp_table = src->cbp_table;
921 dst->mb2b_xy = src->mb2b_xy;
922 dst->mb2br_xy = src->mb2br_xy;
923 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
924 dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * s->mb_stride;
925 dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * s->mb_stride;
926 dst->direct_table = src->direct_table;
927 dst->list_counts = src->list_counts;
a394959b 928 dst->bipred_scratchpad = NULL;
e5d40372
DB
929 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma,
930 src->sps.chroma_format_idc);
afebe2f7
931}
932
933/**
934 * Init context
935 * Allocate buffers which are not shared amongst multiple threads.
936 */
e5d40372
DB
937static int context_init(H264Context *h)
938{
939 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0],
940 h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
941 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1],
942 h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
943
944 h->ref_cache[0][scan8[5] + 1] =
945 h->ref_cache[0][scan8[7] + 1] =
946 h->ref_cache[0][scan8[13] + 1] =
947 h->ref_cache[1][scan8[5] + 1] =
948 h->ref_cache[1][scan8[7] + 1] =
949 h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
145061a1 950
afebe2f7 951 return 0;
e5d40372 952
afebe2f7
953fail:
954 return -1; // free_tables will clean up for us
955}
956
9855b2e3
MN
957static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
958
e5d40372
DB
959static av_cold void common_init(H264Context *h)
960{
961 MpegEncContext *const s = &h->s;
0da71265 962
e5d40372
DB
963 s->width = s->avctx->width;
964 s->height = s->avctx->height;
965 s->codec_id = s->avctx->codec->id;
115329f1 966
76741b0e
BC
967 ff_h264dsp_init(&h->h264dsp, 8, 1);
968 ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
0da71265 969
e5d40372
DB
970 h->dequant_coeff_pps = -1;
971 s->unrestricted_mv = 1;
56edbd81 972
e5d40372
DB
973 /* needed so that IDCT permutation is known early */
974 ff_dsputil_init(&s->dsp, s->avctx);
a5805aa9 975
e5d40372
DB
976 memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
977 memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
0da71265
MN
978}
979
05e95319
HC
980int ff_h264_decode_extradata(H264Context *h)
981{
982 AVCodecContext *avctx = h->s.avctx;
9855b2e3 983
e5d40372 984 if (avctx->extradata[0] == 1) {
9855b2e3
MN
985 int i, cnt, nalsize;
986 unsigned char *p = avctx->extradata;
987
988 h->is_avc = 1;
989
e5d40372 990 if (avctx->extradata_size < 7) {
9855b2e3
MN
991 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
992 return -1;
993 }
994 /* sps and pps in the avcC always have length coded with 2 bytes,
e5d40372 995 * so put a fake nal_length_size = 2 while parsing them */
9855b2e3
MN
996 h->nal_length_size = 2;
997 // Decode sps from avcC
e5d40372
DB
998 cnt = *(p + 5) & 0x1f; // Number of sps
999 p += 6;
9855b2e3
MN
1000 for (i = 0; i < cnt; i++) {
1001 nalsize = AV_RB16(p) + 2;
d1186ff7
LA
1002 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1003 return -1;
e5d40372
DB
1004 if (decode_nal_units(h, p, nalsize) < 0) {
1005 av_log(avctx, AV_LOG_ERROR,
1006 "Decoding sps %d from avcC failed\n", i);
9855b2e3
MN
1007 return -1;
1008 }
1009 p += nalsize;
1010 }
1011 // Decode pps from avcC
1012 cnt = *(p++); // Number of pps
1013 for (i = 0; i < cnt; i++) {
1014 nalsize = AV_RB16(p) + 2;
d1186ff7
LA
1015 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1016 return -1;
108f318d 1017 if (decode_nal_units(h, p, nalsize) < 0) {
e5d40372
DB
1018 av_log(avctx, AV_LOG_ERROR,
1019 "Decoding pps %d from avcC failed\n", i);
9855b2e3
MN
1020 return -1;
1021 }
1022 p += nalsize;
1023 }
e5d40372 1024 // Now store right nal length size, that will be used to parse all other nals
0c17beba 1025 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
9855b2e3
MN
1026 } else {
1027 h->is_avc = 0;
e5d40372 1028 if (decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
9855b2e3
MN
1029 return -1;
1030 }
05e95319
HC
1031 return 0;
1032}
1033
e5d40372
DB
1034av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1035{
1036 H264Context *h = avctx->priv_data;
1037 MpegEncContext *const s = &h->s;
ea2bb12e 1038 int i;
05e95319 1039
efd29844 1040 ff_MPV_decode_defaults(s);
05e95319
HC
1041
1042 s->avctx = avctx;
1043 common_init(h);
1044
e5d40372
DB
1045 s->out_format = FMT_H264;
1046 s->workaround_bugs = avctx->workaround_bugs;
05e95319 1047
e5d40372
DB
1048 /* set defaults */
1049 // s->decode_mb = ff_h263_decode_mb;
05e95319 1050 s->quarter_sample = 1;
e5d40372
DB
1051 if (!avctx->has_b_frames)
1052 s->low_delay = 1;
05e95319
HC
1053
1054 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1055
1056 ff_h264_decode_init_vlc();
1057
6e3ef511 1058 h->pixel_shift = 0;
19a0729b 1059 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
6e3ef511 1060
05e95319 1061 h->thread_context[0] = h;
e5d40372 1062 h->outputed_poc = h->next_outputed_poc = INT_MIN;
ea2bb12e
RB
1063 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1064 h->last_pocs[i] = INT_MIN;
e5d40372
DB
1065 h->prev_poc_msb = 1 << 16;
1066 h->x264_build = -1;
05e95319 1067 ff_h264_reset_sei(h);
36ef5369 1068 if (avctx->codec_id == AV_CODEC_ID_H264) {
e5d40372
DB
1069 if (avctx->ticks_per_frame == 1)
1070 s->avctx->time_base.den *= 2;
05e95319
HC
1071 avctx->ticks_per_frame = 2;
1072 }
1073
e5d40372 1074 if (avctx->extradata_size > 0 && avctx->extradata &&
05e95319
HC
1075 ff_h264_decode_extradata(h))
1076 return -1;
1077
e5d40372
DB
1078 if (h->sps.bitstream_restriction_flag &&
1079 s->avctx->has_b_frames < h->sps.num_reorder_frames) {
db8cb47d 1080 s->avctx->has_b_frames = h->sps.num_reorder_frames;
e5d40372 1081 s->low_delay = 0;
db8cb47d 1082 }
9855b2e3 1083
0da71265
MN
1084 return 0;
1085}
1086
e5d40372
DB
1087#define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1088
1089static void copy_picture_range(Picture **to, Picture **from, int count,
1090 MpegEncContext *new_base,
1091 MpegEncContext *old_base)
6a9c8594
AS
1092{
1093 int i;
1094
e5d40372 1095 for (i = 0; i < count; i++) {
6a9c8594 1096 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
e5d40372
DB
1097 IN_RANGE(from[i], old_base->picture,
1098 sizeof(Picture) * old_base->picture_count) ||
6a9c8594
AS
1099 !from[i]));
1100 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1101 }
1102}
1103
1104static void copy_parameter_set(void **to, void **from, int count, int size)
1105{
1106 int i;
1107
e5d40372
DB
1108 for (i = 0; i < count; i++) {
1109 if (to[i] && !from[i])
1110 av_freep(&to[i]);
1111 else if (from[i] && !to[i])
1112 to[i] = av_malloc(size);
6a9c8594 1113
e5d40372
DB
1114 if (from[i])
1115 memcpy(to[i], from[i], size);
6a9c8594
AS
1116 }
1117}
1118
e5d40372
DB
1119static int decode_init_thread_copy(AVCodecContext *avctx)
1120{
1121 H264Context *h = avctx->priv_data;
6a9c8594 1122
f3a29b75
JR
1123 if (!avctx->internal->is_copy)
1124 return 0;
6a9c8594
AS
1125 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1126 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1127
1128 return 0;
1129}
1130
e5d40372
DB
1131#define copy_fields(to, from, start_field, end_field) \
1132 memcpy(&to->start_field, &from->start_field, \
1133 (char *)&to->end_field - (char *)&to->start_field)
1134
1135static int decode_update_thread_context(AVCodecContext *dst,
1136 const AVCodecContext *src)
1137{
1138 H264Context *h = dst->priv_data, *h1 = src->priv_data;
1139 MpegEncContext *const s = &h->s, *const s1 = &h1->s;
6a9c8594
AS
1140 int inited = s->context_initialized, err;
1141 int i;
1142
e5d40372
DB
1143 if (dst == src || !s1->context_initialized)
1144 return 0;
6a9c8594
AS
1145
1146 err = ff_mpeg_update_thread_context(dst, src);
e5d40372
DB
1147 if (err)
1148 return err;
6a9c8594 1149
e5d40372
DB
1150 // FIXME handle width/height changing
1151 if (!inited) {
1152 for (i = 0; i < MAX_SPS_COUNT; i++)
6a9c8594
AS
1153 av_freep(h->sps_buffers + i);
1154
e5d40372 1155 for (i = 0; i < MAX_PPS_COUNT; i++)
6a9c8594
AS
1156 av_freep(h->pps_buffers + i);
1157
e5d40372
DB
1158 // copy all fields after MpegEnc
1159 memcpy(&h->s + 1, &h1->s + 1,
1160 sizeof(H264Context) - sizeof(MpegEncContext));
6a9c8594
AS
1161 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1162 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
bac3ab13
DB
1163 if (ff_h264_alloc_tables(h) < 0) {
1164 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1165 return AVERROR(ENOMEM);
1166 }
6a9c8594
AS
1167 context_init(h);
1168
e5d40372
DB
1169 for (i = 0; i < 2; i++) {
1170 h->rbsp_buffer[i] = NULL;
6a9c8594
AS
1171 h->rbsp_buffer_size[i] = 0;
1172 }
a394959b 1173 h->bipred_scratchpad = NULL;
6a9c8594
AS
1174
1175 h->thread_context[0] = h;
1176
6a9c8594 1177 s->dsp.clear_blocks(h->mb);
e5d40372 1178 s->dsp.clear_blocks(h->mb + (24 * 16 << h->pixel_shift));
6a9c8594
AS
1179 }
1180
a394959b
JG
1181 /* frame_start may not be called for the next thread (if it's decoding
1182 * a bottom field) so this has to be allocated here */
1183 if (!h->bipred_scratchpad)
1184 h->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1185
e5d40372
DB
1186 // extradata/NAL handling
1187 h->is_avc = h1->is_avc;
6a9c8594 1188
e5d40372
DB
1189 // SPS/PPS
1190 copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1191 MAX_SPS_COUNT, sizeof(SPS));
1192 h->sps = h1->sps;
1193 copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1194 MAX_PPS_COUNT, sizeof(PPS));
1195 h->pps = h1->pps;
6a9c8594 1196
e5d40372
DB
1197 // Dequantization matrices
1198 // FIXME these are big - can they be only copied when PPS changes?
6a9c8594
AS
1199 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1200
e5d40372
DB
1201 for (i = 0; i < 6; i++)
1202 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1203 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
6a9c8594 1204
e5d40372
DB
1205 for (i = 0; i < 6; i++)
1206 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1207 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
6a9c8594
AS
1208
1209 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1210
e5d40372 1211 // POC timing
6a9c8594
AS
1212 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1213
e5d40372 1214 // reference lists
6a9c8594 1215 copy_fields(h, h1, ref_count, list_count);
e5d40372 1216 copy_fields(h, h1, ref_list, intra_gb);
6a9c8594
AS
1217 copy_fields(h, h1, short_ref, cabac_init_idc);
1218
e5d40372
DB
1219 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1220 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1221 copy_picture_range(h->delayed_pic, h1->delayed_pic,
1222 MAX_DELAYED_PIC_COUNT + 2, s, s1);
6a9c8594
AS
1223
1224 h->last_slice_type = h1->last_slice_type;
1225
e5d40372
DB
1226 if (!s->current_picture_ptr)
1227 return 0;
6a9c8594 1228
ba0c8981 1229 if (!s->droppable) {
12fe7594 1230 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
e5d40372
DB
1231 h->prev_poc_msb = h->poc_msb;
1232 h->prev_poc_lsb = h->poc_lsb;
6a9c8594 1233 }
e5d40372
DB
1234 h->prev_frame_num_offset = h->frame_num_offset;
1235 h->prev_frame_num = h->frame_num;
1236 h->outputed_poc = h->next_outputed_poc;
6a9c8594 1237
12fe7594 1238 return err;
6a9c8594
AS
1239}
1240
e5d40372
DB
1241int ff_h264_frame_start(H264Context *h)
1242{
1243 MpegEncContext *const s = &h->s;
0da71265 1244 int i;
6e3ef511 1245 const int pixel_shift = h->pixel_shift;
0da71265 1246
e5d40372 1247 if (ff_MPV_frame_start(s, s->avctx) < 0)
af8aa846 1248 return -1;
0da71265 1249 ff_er_frame_start(s);
3a22d7fa 1250 /*
efd29844 1251 * ff_MPV_frame_start uses pict_type to derive key_frame.
3a22d7fa 1252 * This is incorrect for H.264; IDR markings must be used.
1412060e 1253 * Zero here; IDR markings per slice in frame or fields are ORed in later.
3a22d7fa
JD
1254 * See decode_nal_units().
1255 */
657ccb5a 1256 s->current_picture_ptr->f.key_frame = 0;
e5d40372 1257 s->current_picture_ptr->mmco_reset = 0;
0da71265
MN
1258
1259 assert(s->linesize && s->uvlinesize);
1260
e5d40372
DB
1261 for (i = 0; i < 16; i++) {
1262 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1263 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
0da71265 1264 }
e5d40372
DB
1265 for (i = 0; i < 16; i++) {
1266 h->block_offset[16 + i] =
1267 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1268 h->block_offset[48 + 16 + i] =
1269 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
0da71265
MN
1270 }
1271
934b0821
LM
1272 /* can't be in alloc_tables because linesize isn't known there.
1273 * FIXME: redo bipred weight to not require extra buffer? */
e5d40372 1274 for (i = 0; i < s->slice_context_count; i++)
a394959b
JG
1275 if (h->thread_context[i] && !h->thread_context[i]->bipred_scratchpad)
1276 h->thread_context[i]->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
e5d40372
DB
1277
1278 /* Some macroblocks can be accessed before they're available in case
1279 * of lost slices, MBAFF or threading. */
1280 memset(h->slice_table, -1,
1281 (s->mb_height * s->mb_stride - 1) * sizeof(*h->slice_table));
1282
1283 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1284 // s->current_picture.f.reference /* || h->contains_intra */ || 1;
1285
1286 /* We mark the current picture as non-reference after allocating it, so
1287 * that if we break out due to an error it can be released automatically
1288 * in the next ff_MPV_frame_start().
1289 * SVQ3 as well as most other codecs have only last/next/current and thus
1290 * get released even with set reference, besides SVQ3 and others do not
1291 * mark frames as reference later "naturally". */
36ef5369 1292 if (s->codec_id != AV_CODEC_ID_SVQ3)
657ccb5a 1293 s->current_picture_ptr->f.reference = 0;
357282c6 1294
e5d40372
DB
1295 s->current_picture_ptr->field_poc[0] =
1296 s->current_picture_ptr->field_poc[1] = INT_MAX;
6a9c8594
AS
1297
1298 h->next_output_pic = NULL;
1299
e5d40372 1300 assert(s->current_picture_ptr->long_ref == 0);
357282c6 1301
af8aa846 1302 return 0;
0da71265
MN
1303}
1304
6a9c8594 1305/**
e5d40372
DB
1306 * Run setup operations that must be run after slice header decoding.
1307 * This includes finding the next displayed frame.
1308 *
1309 * @param h h264 master context
1310 * @param setup_finished enough NALs have been read that we can call
1311 * ff_thread_finish_setup()
1312 */
1313static void decode_postinit(H264Context *h, int setup_finished)
1314{
1315 MpegEncContext *const s = &h->s;
6a9c8594
AS
1316 Picture *out = s->current_picture_ptr;
1317 Picture *cur = s->current_picture_ptr;
1318 int i, pics, out_of_order, out_idx;
adedd840 1319 int invalid = 0, cnt = 0;
6a9c8594 1320
657ccb5a
DB
1321 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1322 s->current_picture_ptr->f.pict_type = s->pict_type;
6a9c8594 1323
e5d40372
DB
1324 if (h->next_output_pic)
1325 return;
6a9c8594 1326
e5d40372
DB
1327 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1328 /* FIXME: if we have two PAFF fields in one packet, we can't start
1329 * the next thread here. If we have one field per packet, we can.
1330 * The check in decode_nal_units() is not good enough to find this
1331 * yet, so we assume the worst for now. */
1332 // if (setup_finished)
6a9c8594
AS
1333 // ff_thread_finish_setup(s->avctx);
1334 return;
1335 }
1336
657ccb5a
DB
1337 cur->f.interlaced_frame = 0;
1338 cur->f.repeat_pict = 0;
6a9c8594
AS
1339
1340 /* Signal interlacing information externally. */
e5d40372
DB
1341 /* Prioritize picture timing SEI information over used
1342 * decoding process if it exists. */
6a9c8594 1343
e5d40372
DB
1344 if (h->sps.pic_struct_present_flag) {
1345 switch (h->sei_pic_struct) {
6a9c8594
AS
1346 case SEI_PIC_STRUCT_FRAME:
1347 break;
1348 case SEI_PIC_STRUCT_TOP_FIELD:
1349 case SEI_PIC_STRUCT_BOTTOM_FIELD:
657ccb5a 1350 cur->f.interlaced_frame = 1;
6a9c8594
AS
1351 break;
1352 case SEI_PIC_STRUCT_TOP_BOTTOM:
1353 case SEI_PIC_STRUCT_BOTTOM_TOP:
1354 if (FIELD_OR_MBAFF_PICTURE)
657ccb5a 1355 cur->f.interlaced_frame = 1;
6a9c8594
AS
1356 else
1357 // try to flag soft telecine progressive
657ccb5a 1358 cur->f.interlaced_frame = h->prev_interlaced_frame;
6a9c8594
AS
1359 break;
1360 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1361 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
e5d40372
DB
1362 /* Signal the possibility of telecined film externally
1363 * (pic_struct 5,6). From these hints, let the applications
1364 * decide if they apply deinterlacing. */
657ccb5a 1365 cur->f.repeat_pict = 1;
6a9c8594
AS
1366 break;
1367 case SEI_PIC_STRUCT_FRAME_DOUBLING:
e5d40372 1368 // Force progressive here, doubling interlaced frame is a bad idea.
657ccb5a 1369 cur->f.repeat_pict = 2;
6a9c8594
AS
1370 break;
1371 case SEI_PIC_STRUCT_FRAME_TRIPLING:
657ccb5a 1372 cur->f.repeat_pict = 4;
6a9c8594
AS
1373 break;
1374 }
1375
e5d40372
DB
1376 if ((h->sei_ct_type & 3) &&
1377 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
657ccb5a 1378 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
e5d40372 1379 } else {
6a9c8594 1380 /* Derive interlacing flag from used decoding process. */
657ccb5a 1381 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
6a9c8594 1382 }
657ccb5a 1383 h->prev_interlaced_frame = cur->f.interlaced_frame;
6a9c8594 1384
e5d40372 1385 if (cur->field_poc[0] != cur->field_poc[1]) {
6a9c8594 1386 /* Derive top_field_first from field pocs. */
657ccb5a 1387 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
e5d40372 1388 } else {
657ccb5a 1389 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
e5d40372
DB
1390 /* Use picture timing SEI information. Even if it is a
1391 * information of a past frame, better than nothing. */
1392 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1393 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
657ccb5a 1394 cur->f.top_field_first = 1;
6a9c8594 1395 else
657ccb5a 1396 cur->f.top_field_first = 0;
e5d40372 1397 } else {
6a9c8594 1398 /* Most likely progressive */
657ccb5a 1399 cur->f.top_field_first = 0;
6a9c8594
AS
1400 }
1401 }
1402
e5d40372 1403 // FIXME do something with unavailable reference frames
6a9c8594
AS
1404
1405 /* Sort B-frames into display order */
1406
e5d40372
DB
1407 if (h->sps.bitstream_restriction_flag &&
1408 s->avctx->has_b_frames < h->sps.num_reorder_frames) {
6a9c8594 1409 s->avctx->has_b_frames = h->sps.num_reorder_frames;
e5d40372 1410 s->low_delay = 0;
6a9c8594
AS
1411 }
1412
e5d40372
DB
1413 if (s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
1414 !h->sps.bitstream_restriction_flag) {
2574f08d 1415 s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
e5d40372 1416 s->low_delay = 0;
6a9c8594
AS
1417 }
1418
1419 pics = 0;
e5d40372
DB
1420 while (h->delayed_pic[pics])
1421 pics++;
6a9c8594
AS
1422
1423 assert(pics <= MAX_DELAYED_PIC_COUNT);
1424
1425 h->delayed_pic[pics++] = cur;
657ccb5a
DB
1426 if (cur->f.reference == 0)
1427 cur->f.reference = DELAYED_PIC_REF;
6a9c8594 1428
adedd840
RB
1429 /* Frame reordering. This code takes pictures from coding order and sorts
1430 * them by their incremental POC value into display order. It supports POC
1431 * gaps, MMCO reset codes and random resets.
1432 * A "display group" can start either with a IDR frame (f.key_frame = 1),
1433 * and/or can be closed down with a MMCO reset code. In sequences where
1434 * there is no delay, we can't detect that (since the frame was already
1435 * output to the user), so we also set h->mmco_reset to detect the MMCO
1436 * reset code.
1437 * FIXME: if we detect insufficient delays (as per s->avctx->has_b_frames),
1438 * we increase the delay between input and output. All frames affected by
1439 * the lag (e.g. those that should have been output before another frame
1440 * that we already returned to the user) will be dropped. This is a bug
1441 * that we will fix later. */
1442 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
1443 cnt += out->poc < h->last_pocs[i];
1444 invalid += out->poc == INT_MIN;
1445 }
e5d40372
DB
1446 if (!h->mmco_reset && !cur->f.key_frame &&
1447 cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
adedd840
RB
1448 h->mmco_reset = 2;
1449 if (pics > 1)
1450 h->delayed_pic[pics - 2]->mmco_reset = 2;
1451 }
1452 if (h->mmco_reset || cur->f.key_frame) {
1453 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1454 h->last_pocs[i] = INT_MIN;
1455 cnt = 0;
1456 invalid = MAX_DELAYED_PIC_COUNT;
1457 }
e5d40372 1458 out = h->delayed_pic[0];
6a9c8594 1459 out_idx = 0;
e5d40372
DB
1460 for (i = 1; i < MAX_DELAYED_PIC_COUNT &&
1461 h->delayed_pic[i] &&
1462 !h->delayed_pic[i - 1]->mmco_reset &&
1463 !h->delayed_pic[i]->f.key_frame;
1464 i++)
1465 if (h->delayed_pic[i]->poc < out->poc) {
1466 out = h->delayed_pic[i];
6a9c8594
AS
1467 out_idx = i;
1468 }
e5d40372
DB
1469 if (s->avctx->has_b_frames == 0 &&
1470 (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
adedd840 1471 h->next_outputed_poc = INT_MIN;
e5d40372
DB
1472 out_of_order = !out->f.key_frame && !h->mmco_reset &&
1473 (out->poc < h->next_outputed_poc);
6a9c8594 1474
e5d40372
DB
1475 if (h->sps.bitstream_restriction_flag &&
1476 s->avctx->has_b_frames >= h->sps.num_reorder_frames) {
1477 } else if (out_of_order && pics - 1 == s->avctx->has_b_frames &&
1478 s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
ea2bb12e
RB
1479 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
1480 s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
ea2bb12e
RB
1481 }
1482 s->low_delay = 0;
1483 } else if (s->low_delay &&
e5d40372
DB
1484 ((h->next_outputed_poc != INT_MIN &&
1485 out->poc > h->next_outputed_poc + 2) ||
ea2bb12e 1486 cur->f.pict_type == AV_PICTURE_TYPE_B)) {
6a9c8594
AS
1487 s->low_delay = 0;
1488 s->avctx->has_b_frames++;
1489 }
1490
e5d40372 1491 if (pics > s->avctx->has_b_frames) {
657ccb5a 1492 out->f.reference &= ~DELAYED_PIC_REF;
e5d40372
DB
1493 // for frame threading, the owner must be the second field's thread or
1494 // else the first thread can release the picture and reuse it unsafely
1495 out->owner2 = s;
1496 for (i = out_idx; h->delayed_pic[i]; i++)
1497 h->delayed_pic[i] = h->delayed_pic[i + 1];
1498 }
1499 memmove(h->last_pocs, &h->last_pocs[1],
1500 sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
adedd840 1501 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
e5d40372 1502 if (!out_of_order && pics > s->avctx->has_b_frames) {
6a9c8594 1503 h->next_output_pic = out;
adedd840
RB
1504 if (out->mmco_reset) {
1505 if (out_idx > 0) {
e5d40372 1506 h->next_outputed_poc = out->poc;
adedd840
RB
1507 h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
1508 } else {
1509 h->next_outputed_poc = INT_MIN;
1510 }
1511 } else {
0b4c3232
RB
1512 if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f.key_frame) {
1513 h->next_outputed_poc = INT_MIN;
1514 } else {
1515 h->next_outputed_poc = out->poc;
1516 }
adedd840
RB
1517 }
1518 h->mmco_reset = 0;
e5d40372 1519 } else {
6a9c8594
AS
1520 av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1521 }
1522
1523 if (setup_finished)
1524 ff_thread_finish_setup(s->avctx);
1525}
1526
76741b0e
BC
1527static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1528 uint8_t *src_cb, uint8_t *src_cr,
e5d40372
DB
1529 int linesize, int uvlinesize,
1530 int simple)
76741b0e 1531{
e5d40372 1532 MpegEncContext *const s = &h->s;
0b69d625 1533 uint8_t *top_border;
5f7f9719 1534 int top_idx = 1;
6e3ef511 1535 const int pixel_shift = h->pixel_shift;
76741b0e
BC
1536 int chroma444 = CHROMA444;
1537 int chroma422 = CHROMA422;
115329f1 1538
e5d40372 1539 src_y -= linesize;
53c05b1e
MN
1540 src_cb -= uvlinesize;
1541 src_cr -= uvlinesize;
1542
e5d40372
DB
1543 if (!simple && FRAME_MBAFF) {
1544 if (s->mb_y & 1) {
1545 if (!MB_MBAFF) {
0b69d625 1546 top_border = h->top_borders[0][s->mb_x];
e5d40372 1547 AV_COPY128(top_border, src_y + 15 * linesize);
6e3ef511 1548 if (pixel_shift)
e5d40372
DB
1549 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
1550 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1551 if (chroma444) {
1552 if (pixel_shift) {
1553 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1554 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
1555 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
1556 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
c90b9442 1557 } else {
e5d40372
DB
1558 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
1559 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
c90b9442 1560 }
e5d40372 1561 } else if (chroma422) {
76741b0e 1562 if (pixel_shift) {
e5d40372
DB
1563 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1564 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
76741b0e 1565 } else {
e5d40372
DB
1566 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
1567 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
76741b0e 1568 }
6e3ef511 1569 } else {
c90b9442 1570 if (pixel_shift) {
e5d40372
DB
1571 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
1572 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
c90b9442 1573 } else {
e5d40372
DB
1574 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1575 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
c90b9442 1576 }
6e3ef511 1577 }
5f7f9719
MN
1578 }
1579 }
e5d40372 1580 } else if (MB_MBAFF) {
c988f975 1581 top_idx = 0;
e5d40372 1582 } else
c988f975 1583 return;
5f7f9719
MN
1584 }
1585
0b69d625 1586 top_border = h->top_borders[top_idx][s->mb_x];
09f21198 1587 /* There are two lines saved, the line above the top macroblock
e5d40372
DB
1588 * of a pair, and the line above the bottom macroblock. */
1589 AV_COPY128(top_border, src_y + 16 * linesize);
6e3ef511 1590 if (pixel_shift)
e5d40372
DB
1591 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
1592
1593 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1594 if (chroma444) {
1595 if (pixel_shift) {
1596 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
1597 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
1598 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
1599 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
c90b9442 1600 } else {
e5d40372
DB
1601 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
1602 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
c90b9442 1603 }
e5d40372 1604 } else if (chroma422) {
76741b0e 1605 if (pixel_shift) {
e5d40372
DB
1606 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
1607 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
76741b0e 1608 } else {
e5d40372
DB
1609 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
1610 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
76741b0e 1611 }
6e3ef511 1612 } else {
c90b9442 1613 if (pixel_shift) {
e5d40372
DB
1614 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
1615 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
c90b9442 1616 } else {
e5d40372
DB
1617 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
1618 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
c90b9442 1619 }
6e3ef511 1620 }
53c05b1e
MN
1621 }
1622}
1623
bbdd52ed 1624static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
e5d40372
DB
1625 uint8_t *src_cb, uint8_t *src_cr,
1626 int linesize, int uvlinesize,
1627 int xchg, int chroma444,
1628 int simple, int pixel_shift)
1629{
1630 MpegEncContext *const s = &h->s;
4e987f82 1631 int deblock_topleft;
b69378e2 1632 int deblock_top;
5f7f9719 1633 int top_idx = 1;
1e4f1c56
AS
1634 uint8_t *top_border_m1;
1635 uint8_t *top_border;
5f7f9719 1636
e5d40372
DB
1637 if (!simple && FRAME_MBAFF) {
1638 if (s->mb_y & 1) {
1639 if (!MB_MBAFF)
c988f975 1640 return;
e5d40372 1641 } else {
5f7f9719
MN
1642 top_idx = MB_MBAFF ? 0 : 1;
1643 }
5f7f9719 1644 }
b69378e2 1645
e5d40372 1646 if (h->deblocking_filter == 2) {
4e987f82
RB
1647 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1648 deblock_top = h->top_type;
b69378e2 1649 } else {
4e987f82
RB
1650 deblock_topleft = (s->mb_x > 0);
1651 deblock_top = (s->mb_y > !!MB_FIELD);
b69378e2 1652 }
53c05b1e 1653
e5d40372 1654 src_y -= linesize + 1 + pixel_shift;
6e3ef511
OA
1655 src_cb -= uvlinesize + 1 + pixel_shift;
1656 src_cr -= uvlinesize + 1 + pixel_shift;
53c05b1e 1657
e5d40372 1658 top_border_m1 = h->top_borders[top_idx][s->mb_x - 1];
1e4f1c56
AS
1659 top_border = h->top_borders[top_idx][s->mb_x];
1660
e5d40372
DB
1661#define XCHG(a, b, xchg) \
1662 if (pixel_shift) { \
1663 if (xchg) { \
1664 AV_SWAP64(b + 0, a + 0); \
1665 AV_SWAP64(b + 8, a + 8); \
1666 } else { \
1667 AV_COPY128(b, a); \
1668 } \
1669 } else if (xchg) \
1670 AV_SWAP64(b, a); \
1671 else \
1672 AV_COPY64(b, a);
1673
1674 if (deblock_top) {
1675 if (deblock_topleft) {
1676 XCHG(top_border_m1 + (8 << pixel_shift),
1677 src_y - (7 << pixel_shift), 1);
c988f975 1678 }
6e3ef511
OA
1679 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1680 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
e5d40372
DB
1681 if (s->mb_x + 1 < s->mb_width) {
1682 XCHG(h->top_borders[top_idx][s->mb_x + 1],
1683 src_y + (17 << pixel_shift), 1);
43efd19a 1684 }
53c05b1e 1685 }
e5d40372
DB
1686 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1687 if (chroma444) {
1688 if (deblock_topleft) {
c90b9442
JGG
1689 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1690 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1691 }
1692 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1693 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1694 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1695 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
e5d40372
DB
1696 if (s->mb_x + 1 < s->mb_width) {
1697 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1698 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
c90b9442
JGG
1699 }
1700 } else {
e5d40372
DB
1701 if (deblock_top) {
1702 if (deblock_topleft) {
c90b9442
JGG
1703 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1704 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1705 }
e5d40372
DB
1706 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
1707 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
c988f975 1708 }
53c05b1e 1709 }
53c05b1e
MN
1710 }
1711}
1712
e5d40372
DB
1713static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth,
1714 int index)
1715{
6e3ef511 1716 if (high_bit_depth) {
e5d40372 1717 return AV_RN32A(((int32_t *)mb) + index);
6e3ef511
OA
1718 } else
1719 return AV_RN16A(mb + index);
1720}
1721
e5d40372
DB
1722static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth,
1723 int index, int value)
1724{
6e3ef511 1725 if (high_bit_depth) {
e5d40372 1726 AV_WN32A(((int32_t *)mb) + index, value);
6e3ef511
OA
1727 } else
1728 AV_WN16A(mb + index, value);
1729}
1730
e5d40372
DB
1731static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
1732 int mb_type, int is_h264,
1733 int simple,
1734 int transform_bypass,
1735 int pixel_shift,
1736 int *block_offset,
1737 int linesize,
1738 uint8_t *dest_y, int p)
c90b9442 1739{
e5d40372 1740 MpegEncContext *const s = &h->s;
c90b9442
JGG
1741 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1742 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1743 int i;
e5d40372
DB
1744 int qscale = p == 0 ? s->qscale : h->chroma_qp[p - 1];
1745 block_offset += 16 * p;
1746 if (IS_INTRA4x4(mb_type)) {
1747 if (simple || !s->encoding) {
1748 if (IS_8x8DCT(mb_type)) {
1749 if (transform_bypass) {
1750 idct_dc_add =
1751 idct_add = s->dsp.add_pixels8;
1752 } else {
c90b9442
JGG
1753 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1754 idct_add = h->h264dsp.h264_idct8_add;
1755 }
e5d40372
DB
1756 for (i = 0; i < 16; i += 4) {
1757 uint8_t *const ptr = dest_y + block_offset[i];
1758 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1759 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1760 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1761 } else {
1762 const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1763 h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
1764 (h->topright_samples_available << i) & 0x4000, linesize);
1765 if (nnz) {
1766 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1767 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
c90b9442 1768 else
e5d40372 1769 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
c90b9442
JGG
1770 }
1771 }
1772 }
e5d40372
DB
1773 } else {
1774 if (transform_bypass) {
1775 idct_dc_add =
1776 idct_add = s->dsp.add_pixels4;
1777 } else {
c90b9442
JGG
1778 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1779 idct_add = h->h264dsp.h264_idct_add;
1780 }
e5d40372
DB
1781 for (i = 0; i < 16; i++) {
1782 uint8_t *const ptr = dest_y + block_offset[i];
1783 const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
c90b9442 1784
e5d40372
DB
1785 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1786 h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1787 } else {
c90b9442
JGG
1788 uint8_t *topright;
1789 int nnz, tr;
1790 uint64_t tr_high;
e5d40372
DB
1791 if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
1792 const int topright_avail = (h->topright_samples_available << i) & 0x8000;
6371ce4b 1793 assert(s->mb_y || linesize <= block_offset[i]);
e5d40372 1794 if (!topright_avail) {
c90b9442 1795 if (pixel_shift) {
e5d40372
DB
1796 tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
1797 topright = (uint8_t *)&tr_high;
c90b9442 1798 } else {
e5d40372
DB
1799 tr = ptr[3 - linesize] * 0x01010101u;
1800 topright = (uint8_t *)&tr;
c90b9442 1801 }
e5d40372
DB
1802 } else
1803 topright = ptr + (4 << pixel_shift) - linesize;
1804 } else
1805 topright = NULL;
1806
1807 h->hpc.pred4x4[dir](ptr, topright, linesize);
1808 nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1809 if (nnz) {
1810 if (is_h264) {
1811 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1812 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
c90b9442 1813 else
e5d40372 1814 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
301fb921 1815 } else if (CONFIG_SVQ3_DECODER)
e5d40372 1816 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
c90b9442
JGG
1817 }
1818 }
1819 }
1820 }
1821 }
e5d40372
DB
1822 } else {
1823 h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
1824 if (is_h264) {
1825 if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
1826 if (!transform_bypass)
1827 h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
1828 h->mb_luma_dc[p],
1829 h->dequant4_coeff[p][qscale][0]);
1830 else {
1831 static const uint8_t dc_mapping[16] = {
1832 0 * 16, 1 * 16, 4 * 16, 5 * 16,
1833 2 * 16, 3 * 16, 6 * 16, 7 * 16,
1834 8 * 16, 9 * 16, 12 * 16, 13 * 16,
1835 10 * 16, 11 * 16, 14 * 16, 15 * 16 };
1836 for (i = 0; i < 16; i++)
1837 dctcoef_set(h->mb + (p * 256 << pixel_shift),
1838 pixel_shift, dc_mapping[i],
1839 dctcoef_get(h->mb_luma_dc[p],
1840 pixel_shift, i));
c90b9442
JGG
1841 }
1842 }
301fb921 1843 } else if (CONFIG_SVQ3_DECODER)
e5d40372
DB
1844 ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
1845 h->mb_luma_dc[p], qscale);
c90b9442
JGG
1846 }
1847}
1848
e5d40372
DB
1849static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
1850 int is_h264, int simple,
1851 int transform_bypass,
1852 int pixel_shift,
1853 int *block_offset,
1854 int linesize,
1855 uint8_t *dest_y, int p)
c90b9442 1856{
e5d40372 1857 MpegEncContext *const s = &h->s;
c90b9442
JGG
1858 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1859 int i;
e5d40372
DB
1860 block_offset += 16 * p;
1861 if (!IS_INTRA4x4(mb_type)) {
1862 if (is_h264) {
1863 if (IS_INTRA16x16(mb_type)) {
1864 if (transform_bypass) {
1865 if (h->sps.profile_idc == 244 &&
1866 (h->intra16x16_pred_mode == VERT_PRED8x8 ||
1867 h->intra16x16_pred_mode == HOR_PRED8x8)) {
1868 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
1869 h->mb + (p * 256 << pixel_shift),
1870 linesize);
1871 } else {
1872 for (i = 0; i < 16; i++)
1873 if (h->non_zero_count_cache[scan8[i + p * 16]] ||
1874 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1875 s->dsp.add_pixels4(dest_y + block_offset[i],
1876 h->mb + (i * 16 + p * 256 << pixel_shift),
1877 linesize);
c90b9442 1878 }
e5d40372
DB
1879 } else {
1880 h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
1881 h->mb + (p * 256 << pixel_shift),
1882 linesize,
1883 h->non_zero_count_cache + p * 5 * 8);
c90b9442 1884 }
e5d40372
DB
1885 } else if (h->cbp & 15) {
1886 if (transform_bypass) {
c90b9442 1887 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
e5d40372
DB
1888 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8
1889 : s->dsp.add_pixels4;
1890 for (i = 0; i < 16; i += di)
1891 if (h->non_zero_count_cache[scan8[i + p * 16]])
1892 idct_add(dest_y + block_offset[i],
1893 h->mb + (i * 16 + p * 256 << pixel_shift),
1894 linesize);
1895 } else {
1896 if (IS_8x8DCT(mb_type))
1897 h->h264dsp.h264_idct8_add4(dest_y, block_offset,
1898 h->mb + (p * 256 << pixel_shift),
1899 linesize,
1900 h->non_zero_count_cache + p * 5 * 8);
1901 else
1902 h->h264dsp.h264_idct_add16(dest_y, block_offset,
1903 h->mb + (p * 256 << pixel_shift),
1904 linesize,
1905 h->non_zero_count_cache + p * 5 * 8);
c90b9442
JGG
1906 }
1907 }
301fb921 1908 } else if (CONFIG_SVQ3_DECODER) {
e5d40372
DB
1909 for (i = 0; i < 16; i++)
1910 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
1911 // FIXME benchmark weird rule, & below
1912 uint8_t *const ptr = dest_y + block_offset[i];
1913 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
1914 s->qscale, IS_INTRA(mb_type) ? 1 : 0);
c90b9442 1915 }
c90b9442
JGG
1916 }
1917 }
1918}
1919
28fff0d9
MR
1920#define BITS 8
1921#define SIMPLE 1
1922#include "h264_mb_template.c"
e5d40372 1923
28fff0d9
MR
1924#undef BITS
1925#define BITS 16
1926#include "h264_mb_template.c"
bd91fee3 1927
28fff0d9
MR
1928#undef SIMPLE
1929#define SIMPLE 0
1930#include "h264_mb_template.c"
c90b9442 1931
e5d40372
DB
1932void ff_h264_hl_decode_mb(H264Context *h)
1933{
1934 MpegEncContext *const s = &h->s;
1935 const int mb_xy = h->mb_xy;
657ccb5a 1936 const int mb_type = s->current_picture.f.mb_type[mb_xy];
e5d40372 1937 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
bd91fee3 1938
c90b9442 1939 if (CHROMA444) {
e5d40372 1940 if (is_complex || h->pixel_shift)
c90b9442
JGG
1941 hl_decode_mb_444_complex(h);
1942 else
28fff0d9 1943 hl_decode_mb_444_simple_8(h);
c90b9442 1944 } else if (is_complex) {
bd91fee3 1945 hl_decode_mb_complex(h);
6e3ef511
OA
1946 } else if (h->pixel_shift) {
1947 hl_decode_mb_simple_16(h);
1948 } else
1949 hl_decode_mb_simple_8(h);
bd91fee3
AS
1950}
1951
e5d40372
DB
1952static int pred_weight_table(H264Context *h)
1953{
1954 MpegEncContext *const s = &h->s;
0da71265 1955 int list, i;
9f2d1b4f 1956 int luma_def, chroma_def;
115329f1 1957
e5d40372
DB
1958 h->use_weight = 0;
1959 h->use_weight_chroma = 0;
1960 h->luma_log2_weight_denom = get_ue_golomb(&s->gb);
1961 if (h->sps.chroma_format_idc)
1962 h->chroma_log2_weight_denom = get_ue_golomb(&s->gb);
1963 luma_def = 1 << h->luma_log2_weight_denom;
1964 chroma_def = 1 << h->chroma_log2_weight_denom;
0da71265 1965
e5d40372 1966 for (list = 0; list < 2; list++) {
cb99c652
GB
1967 h->luma_weight_flag[list] = 0;
1968 h->chroma_weight_flag[list] = 0;
e5d40372 1969 for (i = 0; i < h->ref_count[list]; i++) {
0da71265 1970 int luma_weight_flag, chroma_weight_flag;
115329f1 1971
e5d40372
DB
1972 luma_weight_flag = get_bits1(&s->gb);
1973 if (luma_weight_flag) {
1974 h->luma_weight[i][list][0] = get_se_golomb(&s->gb);
1975 h->luma_weight[i][list][1] = get_se_golomb(&s->gb);
1976 if (h->luma_weight[i][list][0] != luma_def ||
1977 h->luma_weight[i][list][1] != 0) {
1978 h->use_weight = 1;
1979 h->luma_weight_flag[list] = 1;
cb99c652 1980 }
e5d40372
DB
1981 } else {
1982 h->luma_weight[i][list][0] = luma_def;
1983 h->luma_weight[i][list][1] = 0;
0da71265
MN
1984 }
1985
e5d40372
DB
1986 if (h->sps.chroma_format_idc) {
1987 chroma_weight_flag = get_bits1(&s->gb);
1988 if (chroma_weight_flag) {
fef744d4 1989 int j;
e5d40372
DB
1990 for (j = 0; j < 2; j++) {
1991 h->chroma_weight[i][list][j][0] = get_se_golomb(&s->gb);
1992 h->chroma_weight[i][list][j][1] = get_se_golomb(&s->gb);
1993 if (h->chroma_weight[i][list][j][0] != chroma_def ||
1994 h->chroma_weight[i][list][j][1] != 0) {
1995 h->use_weight_chroma = 1;
1996 h->chroma_weight_flag[list] = 1;
cb99c652 1997 }
fef744d4 1998 }
e5d40372 1999 } else {
fef744d4 2000 int j;
e5d40372
DB
2001 for (j = 0; j < 2; j++) {
2002 h->chroma_weight[i][list][j][0] = chroma_def;
2003 h->chroma_weight[i][list][j][1] = 0;
fef744d4 2004 }
0da71265
MN
2005 }
2006 }
2007 }
e5d40372
DB
2008 if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2009 break;
0da71265 2010 }
e5d40372 2011 h->use_weight = h->use_weight || h->use_weight_chroma;
0da71265
MN
2012 return 0;
2013}
2014
1052b76f
MN
2015/**
2016 * Initialize implicit_weight table.
6da88bd3 2017 * @param field 0/1 initialize the weight for interlaced MBAFF
1052b76f
MN
2018 * -1 initializes the rest
2019 */
e5d40372
DB
2020static void implicit_weight_table(H264Context *h, int field)
2021{
2022 MpegEncContext *const s = &h->s;
1052b76f 2023 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
9f2d1b4f 2024
ce09f927
GB
2025 for (i = 0; i < 2; i++) {
2026 h->luma_weight_flag[i] = 0;
2027 h->chroma_weight_flag[i] = 0;
2028 }
2029
e5d40372 2030 if (field < 0) {
4418aa9c
RB
2031 if (s->picture_structure == PICT_FRAME) {
2032 cur_poc = s->current_picture_ptr->poc;
2033 } else {
2034 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2035 }
e5d40372
DB
2036 if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF &&
2037 h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2038 h->use_weight = 0;
2039 h->use_weight_chroma = 0;
2040 return;
2041 }
2042 ref_start = 0;
2043 ref_count0 = h->ref_count[0];
2044 ref_count1 = h->ref_count[1];
2045 } else {
2046 cur_poc = s->current_picture_ptr->field_poc[field];
2047 ref_start = 16;
2048 ref_count0 = 16 + 2 * h->ref_count[0];
2049 ref_count1 = 16 + 2 * h->ref_count[1];
1052b76f 2050 }
9f2d1b4f 2051
e5d40372
DB
2052 h->use_weight = 2;
2053 h->use_weight_chroma = 2;
2054 h->luma_log2_weight_denom = 5;
2055 h->chroma_log2_weight_denom = 5;
9f2d1b4f 2056
e5d40372 2057 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
9f2d1b4f 2058 int poc0 = h->ref_list[0][ref0].poc;
e5d40372 2059 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
87cf70eb
JD
2060 int w = 32;
2061 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2062 int poc1 = h->ref_list[1][ref1].poc;
e5d40372
DB
2063 int td = av_clip(poc1 - poc0, -128, 127);
2064 if (td) {
87cf70eb
JD
2065 int tb = av_clip(cur_poc - poc0, -128, 127);
2066 int tx = (16384 + (FFABS(td) >> 1)) / td;
e5d40372
DB
2067 int dist_scale_factor = (tb * tx + 32) >> 8;
2068 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
87cf70eb
JD
2069 w = 64 - dist_scale_factor;
2070 }
1052b76f 2071 }
e5d40372
DB
2072 if (field < 0) {
2073 h->implicit_weight[ref0][ref1][0] =
2074 h->implicit_weight[ref0][ref1][1] = w;
2075 } else {
2076 h->implicit_weight[ref0][ref1][field] = w;
72f86ec0 2077 }
9f2d1b4f
LM
2078 }
2079 }
2080}
2081
8fd57a66 2082/**
5175b937 2083 * instantaneous decoder refresh.
0da71265 2084 */
e5d40372
DB
2085static void idr(H264Context *h)
2086{
ea6f00c4 2087 ff_h264_remove_all_refs(h);
e5d40372
DB
2088 h->prev_frame_num = 0;
2089 h->prev_frame_num_offset = 0;
2090 h->prev_poc_msb =
2091 h->prev_poc_lsb = 0;
0da71265
MN
2092}
2093
7c33ad19 2094/* forget old pics after a seek */
e5d40372
DB
2095static void flush_dpb(AVCodecContext *avctx)
2096{
2097 H264Context *h = avctx->priv_data;
7c33ad19 2098 int i;
e5d40372
DB
2099 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
2100 if (h->delayed_pic[i])
657ccb5a 2101 h->delayed_pic[i]->f.reference = 0;
e5d40372 2102 h->delayed_pic[i] = NULL;
285b570f 2103 }
adedd840
RB
2104 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2105 h->last_pocs[i] = INT_MIN;
e5d40372 2106 h->outputed_poc = h->next_outputed_poc = INT_MIN;
b19d493f 2107 h->prev_interlaced_frame = 1;
7c33ad19 2108 idr(h);
e5d40372 2109 if (h->s.current_picture_ptr)
657ccb5a 2110 h->s.current_picture_ptr->f.reference = 0;
e5d40372 2111 h->s.first_field = 0;
9c095463 2112 ff_h264_reset_sei(h);
e240f898 2113 ff_mpeg_flush(avctx);
7c33ad19
LM
2114}
2115
e5d40372
DB
2116static int init_poc(H264Context *h)
2117{
2118 MpegEncContext *const s = &h->s;
2119 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
0da71265 2120 int field_poc[2];
357282c6 2121 Picture *cur = s->current_picture_ptr;
0da71265 2122
e5d40372
DB
2123 h->frame_num_offset = h->prev_frame_num_offset;
2124 if (h->frame_num < h->prev_frame_num)
b78a6baa 2125 h->frame_num_offset += max_frame_num;
0da71265 2126
e5d40372
DB
2127 if (h->sps.poc_type == 0) {
2128 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
0da71265 2129
e5d40372 2130 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
0da71265 2131 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
e5d40372 2132 else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
0da71265
MN
2133 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2134 else
2135 h->poc_msb = h->prev_poc_msb;
115329f1 2136 field_poc[0] =
0da71265 2137 field_poc[1] = h->poc_msb + h->poc_lsb;
e5d40372 2138 if (s->picture_structure == PICT_FRAME)
0da71265 2139 field_poc[1] += h->delta_poc_bottom;
e5d40372 2140 } else if (h->sps.poc_type == 1) {
0da71265
MN
2141 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2142 int i;
2143
e5d40372 2144 if (h->sps.poc_cycle_length != 0)
0da71265
MN
2145 abs_frame_num = h->frame_num_offset + h->frame_num;
2146 else
2147 abs_frame_num = 0;
2148
e5d40372 2149 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
0da71265 2150 abs_frame_num--;
115329f1 2151
0da71265 2152 expected_delta_per_poc_cycle = 0;
e5d40372
DB
2153 for (i = 0; i < h->sps.poc_cycle_length; i++)
2154 // FIXME integrate during sps parse
2155 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
0da71265 2156
e5d40372 2157 if (abs_frame_num > 0) {
0da71265
MN
2158 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2159 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2160
2161 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
e5d40372
DB
2162 for (i = 0; i <= frame_num_in_poc_cycle; i++)
2163 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
0da71265
MN
2164 } else
2165 expectedpoc = 0;
2166
e5d40372 2167 if (h->nal_ref_idc == 0)
0da71265 2168 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
115329f1 2169
0da71265
MN
2170 field_poc[0] = expectedpoc + h->delta_poc[0];
2171 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2172
e5d40372 2173 if (s->picture_structure == PICT_FRAME)
0da71265 2174 field_poc[1] += h->delta_poc[1];
e5d40372
DB
2175 } else {
2176 int poc = 2 * (h->frame_num_offset + h->frame_num);
5710b371 2177
e5d40372 2178 if (!h->nal_ref_idc)
b78a6baa 2179 poc--;
5710b371 2180
e5d40372
DB
2181 field_poc[0] = poc;
2182 field_poc[1] = poc;
0da71265 2183 }
115329f1 2184
e5d40372
DB
2185 if (s->picture_structure != PICT_BOTTOM_FIELD)
2186 s->current_picture_ptr->field_poc[0] = field_poc[0];
2187 if (s->picture_structure != PICT_TOP_FIELD)
2188 s->current_picture_ptr->field_poc[1] = field_poc[1];
2189 cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]);
0da71265
MN
2190
2191 return 0;
2192}
2193
b41c1db3
2194/**
2195 * initialize scan tables
2196 */
e5d40372
DB
2197static void init_scan_tables(H264Context *h)
2198{
b41c1db3 2199 int i;
e5d40372
DB
2200 for (i = 0; i < 16; i++) {
2201#define T(x) (x >> 2) | ((x << 2) & 0xF)
ca32f7f2 2202 h->zigzag_scan[i] = T(zigzag_scan[i]);
e5d40372 2203 h->field_scan[i] = T(field_scan[i]);
b41c1db3 2204#undef T
b41c1db3 2205 }
e5d40372
DB
2206 for (i = 0; i < 64; i++) {
2207#define T(x) (x >> 3) | ((x & 7) << 3)
ca32f7f2
JGG
2208 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2209 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2210 h->field_scan8x8[i] = T(field_scan8x8[i]);
2211 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
b41c1db3 2212#undef T
b41c1db3 2213 }
e5d40372 2214 if (h->sps.transform_bypass) { // FIXME same ugly
b41c1db3 2215 h->zigzag_scan_q0 = zigzag_scan;
45beb850 2216 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
b41c1db3
2217 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2218 h->field_scan_q0 = field_scan;
2219 h->field_scan8x8_q0 = field_scan8x8;
2220 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
e5d40372 2221 } else {
b41c1db3
2222 h->zigzag_scan_q0 = h->zigzag_scan;
2223 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
2224 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2225 h->field_scan_q0 = h->field_scan;
2226 h->field_scan8x8_q0 = h->field_scan8x8;
2227 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
2228 }
2229}
afebe2f7 2230
e5d40372
DB
2231static int field_end(H264Context *h, int in_setup)
2232{
2233 MpegEncContext *const s = &h->s;
2234 AVCodecContext *const avctx = s->avctx;
12fe7594 2235 int err = 0;
e5d40372 2236 s->mb_y = 0;
256299d3 2237
ba0c8981 2238 if (!in_setup && !s->droppable)
1e26a48f 2239 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
47c0ac96 2240 s->picture_structure == PICT_BOTTOM_FIELD);
256299d3 2241
e5d40372
DB
2242 if (CONFIG_H264_VDPAU_DECODER &&
2243 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
256299d3
MN
2244 ff_vdpau_h264_set_reference_frames(s);
2245
e5d40372 2246 if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
ba0c8981 2247 if (!s->droppable) {
12fe7594 2248 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
e5d40372
DB
2249 h->prev_poc_msb = h->poc_msb;
2250 h->prev_poc_lsb = h->poc_lsb;
6a9c8594 2251 }
e5d40372
DB
2252 h->prev_frame_num_offset = h->frame_num_offset;
2253 h->prev_frame_num = h->frame_num;
2254 h->outputed_poc = h->next_outputed_poc;
256299d3 2255 }
256299d3
MN
2256
2257 if (avctx->hwaccel) {
2258 if (avctx->hwaccel->end_frame(avctx) < 0)
e5d40372
DB
2259 av_log(avctx, AV_LOG_ERROR,
2260 "hardware accelerator failed to decode picture\n");
256299d3
MN
2261 }
2262
e5d40372
DB
2263 if (CONFIG_H264_VDPAU_DECODER &&
2264 s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
256299d3
MN
2265 ff_vdpau_h264_picture_complete(s);
2266
2267 /*
2268 * FIXME: Error handling code does not seem to support interlaced
2269 * when slices span multiple rows
2270 * The ff_er_add_slice calls don't work right for bottom
2271 * fields; they cause massive erroneous error concealing
2272 * Error marking covers both fields (top and bottom).
2273 * This causes a mismatched s->error_count
2274 * and a bad error table. Further, the error count goes to
2275 * INT_MAX when called for bottom field, because mb_y is
2276 * past end by one (callers fault) and resync_mb_y != 0
2277 * causes problems for the first MB line, too.
2278 */
2279 if (!FIELD_PICTURE)
2280 ff_er_frame_end(s);
2281
efd29844 2282 ff_MPV_frame_end(s);
d225a1e2 2283
e5d40372 2284 h->current_slice = 0;
12fe7594
DB
2285
2286 return err;
256299d3
MN
2287}
2288
afebe2f7 2289/**
49bd8e4b 2290 * Replicate H264 "master" context to thread contexts.
afebe2f7
2291 */
2292static void clone_slice(H264Context *dst, H264Context *src)
2293{
e5d40372
DB
2294 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2295 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2296 dst->s.current_picture = src->s.current_picture;
2297 dst->s.linesize = src->s.linesize;
2298 dst->s.uvlinesize = src->s.uvlinesize;
2299 dst->s.first_field = src->s.first_field;
2300
2301 dst->prev_poc_msb = src->prev_poc_msb;
2302 dst->prev_poc_lsb = src->prev_poc_lsb;
2303 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2304 dst->prev_frame_num = src->prev_frame_num;
2305 dst->short_ref_count = src->short_ref_count;
afebe2f7
2306
2307 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2308 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2309 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2310 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
50c21814
2311
2312 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2313 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
afebe2f7
2314}
2315
0da71265 2316/**
58c42af7 2317 * Compute profile from profile_idc and constraint_set?_flags.
fe9a3fbe
JG
2318 *
2319 * @param sps SPS
2320 *
2321 * @return profile as defined by FF_PROFILE_H264_*
2322 */
2323int ff_h264_get_profile(SPS *sps)
2324{
2325 int profile = sps->profile_idc;
2326
e5d40372 2327 switch (sps->profile_idc) {
fe9a3fbe
JG
2328 case FF_PROFILE_H264_BASELINE:
2329 // constraint_set1_flag set to 1
e5d40372 2330 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
fe9a3fbe
JG
2331 break;
2332 case FF_PROFILE_H264_HIGH_10:
2333 case FF_PROFILE_H264_HIGH_422:
2334 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2335 // constraint_set3_flag set to 1
e5d40372 2336 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
fe9a3fbe
JG
2337 break;
2338 }
2339
2340 return profile;
2341}
2342
2343/**
58c42af7 2344 * Decode a slice header.
efd29844 2345 * This will also call ff_MPV_common_init() and frame_start() as needed.
afebe2f7
2346 *
2347 * @param h h264context
e5d40372
DB
2348 * @param h0 h264 master context (differs from 'h' when doing sliced based
2349 * parallel decoding)
afebe2f7 2350 *
d9526386 2351 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
0da71265 2352 */
e5d40372
DB
2353static int decode_slice_header(H264Context *h, H264Context *h0)
2354{
2355 MpegEncContext *const s = &h->s;
2356 MpegEncContext *const s0 = &h0->s;
88e7a4d1 2357 unsigned int first_mb_in_slice;
ac658be5 2358 unsigned int pps_id;
60b6b8c0 2359 int num_ref_idx_active_override_flag, max_refs;
41f5c62f 2360 unsigned int slice_type, tmp, i, j;
0bf79634 2361 int default_ref_list_done = 0;
ba0c8981 2362 int last_pic_structure, last_pic_droppable;
0da71265 2363
9a0dda8b 2364 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
e5d40372
DB
2365 if ((s->avctx->flags2 & CODEC_FLAG2_FAST) &&
2366 !h->nal_ref_idc && !h->pixel_shift) {
2367 s->me.qpel_put = s->dsp.put_2tap_qpel_pixels_tab;
2368 s->me.qpel_avg = s->dsp.avg_2tap_qpel_pixels_tab;
2369 } else {
2370 s->me.qpel_put = s->dsp.put_h264_qpel_pixels_tab;
2371 s->me.qpel_avg = s->dsp.avg_h264_qpel_pixels_tab;
cf653d08
JD
2372 }
2373
e5d40372 2374 first_mb_in_slice = get_ue_golomb(&s->gb);
0da71265 2375
e5d40372
DB
2376 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
2377 if (h0->current_slice && FIELD_PICTURE) {
6a9c8594 2378 field_end(h, 1);
d225a1e2
MN
2379 }
2380
afebe2f7 2381 h0->current_slice = 0;
1e26a48f 2382 if (!s0->first_field) {
ba0c8981 2383 if (s->current_picture_ptr && !s->droppable &&
1e26a48f
RB
2384 s->current_picture_ptr->owner2 == s) {
2385 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2386 s->picture_structure == PICT_BOTTOM_FIELD);
2387 }
e5d40372 2388 s->current_picture_ptr = NULL;
1e26a48f 2389 }
66a4b2c1
MN
2390 }
2391
e5d40372
DB
2392 slice_type = get_ue_golomb_31(&s->gb);
2393 if (slice_type > 9) {
2394 av_log(h->s.avctx, AV_LOG_ERROR,
2395 "slice type too large (%d) at %d %d\n",
2396 h->slice_type, s->mb_x, s->mb_y);
5175b937 2397 return -1;
0da71265 2398 }
e5d40372 2399 if (slice_type > 4) {
0bf79634 2400 slice_type -= 5;
e5d40372
DB
2401 h->slice_type_fixed = 1;
2402 } else
2403 h->slice_type_fixed = 0;
115329f1 2404
e5d40372
DB
2405 slice_type = golomb_to_pict_type[slice_type];
2406 if (slice_type == AV_PICTURE_TYPE_I ||
2407 (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
0bf79634
LLL
2408 default_ref_list_done = 1;
2409 }
e5d40372
DB
2410 h->slice_type = slice_type;
2411 h->slice_type_nos = slice_type & 3;
0bf79634 2412
e5d40372
DB
2413 // to make a few old functions happy, it's wrong though
2414 s->pict_type = h->slice_type;
115329f1 2415
e5d40372
DB
2416 pps_id = get_ue_golomb(&s->gb);
2417 if (pps_id >= MAX_PPS_COUNT) {
9b879566 2418 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
0da71265
MN
2419 return -1;
2420 }
e5d40372
DB
2421 if (!h0->pps_buffers[pps_id]) {
2422 av_log(h->s.avctx, AV_LOG_ERROR,
2423 "non-existing PPS %u referenced\n",
2424 pps_id);
8b92b792
MN
2425 return -1;
2426 }
e5d40372 2427 h->pps = *h0->pps_buffers[pps_id];
8b92b792 2428
e5d40372
DB
2429 if (!h0->sps_buffers[h->pps.sps_id]) {
2430 av_log(h->s.avctx, AV_LOG_ERROR,
2431 "non-existing SPS %u referenced\n",
2432 h->pps.sps_id);
8b92b792
MN
2433 return -1;
2434 }
afebe2f7 2435 h->sps = *h0->sps_buffers[h->pps.sps_id];
239ea04c 2436
fe9a3fbe 2437 s->avctx->profile = ff_h264_get_profile(&h->sps);
fa37cf0d 2438 s->avctx->level = h->sps.level_idc;
6752a3cc 2439 s->avctx->refs = h->sps.ref_frame_count;
b08e38e8 2440
e5d40372
DB
2441 s->mb_width = h->sps.mb_width;
2442 s->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
115329f1 2443
e5d40372 2444 h->b_stride = s->mb_width * 4;
0da71265 2445
76741b0e
BC
2446 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2447
e5d40372
DB
2448 s->width = 16 * s->mb_width - (2 >> CHROMA444) * FFMIN(h->sps.crop_right, (8 << CHROMA444) - 1);
2449 if (h->sps.frame_mbs_only_flag)
2450 s->height = 16 * s->mb_height - (1 << s->chroma_y_shift) * FFMIN(h->sps.crop_bottom, (16 >> s->chroma_y_shift) - 1);
0da71265 2451 else
e5d40372 2452 s->height = 16 * s->mb_height - (2 << s->chroma_y_shift) * FFMIN(h->sps.crop_bottom, (16 >> s->chroma_y_shift) - 1);
115329f1 2453
30f51509
MR
2454 if (FFALIGN(s->avctx->width, 16) == s->width &&
2455 FFALIGN(s->avctx->height, 16) == s->height) {
2456 s->width = s->avctx->width;
2457 s->height = s->avctx->height;
2458 }
2459
e5d40372
DB
2460 if (s->context_initialized &&
2461 (s->width != s->avctx->width || s->height != s->avctx->height ||
2462 av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
732f9fcf 2463 if (h != h0 || (HAVE_THREADS && h->s.avctx->active_thread_type & FF_THREAD_FRAME)) {
e5d40372 2464 av_log_missing_feature(s->avctx,
f75f4194 2465 "Width/height changing with threads", 0);
732f9fcf 2466 return AVERROR_PATCHWELCOME; // width / height changed during parallelized decoding
6a9c8594 2467 }
91078926 2468 free_tables(h, 0);
ff7f75e1 2469 flush_dpb(s->avctx);
efd29844 2470 ff_MPV_common_end(s);
0da71265
MN
2471 }
2472 if (!s->context_initialized) {
33aec3f4 2473 if (h != h0) {
e5d40372
DB
2474 av_log(h->s.avctx, AV_LOG_ERROR,
2475 "Cannot (re-)initialize context during parallel decoding.\n");
33aec3f4
MN
2476 return -1;
2477 }
f3bdc3da
RD
2478
2479 avcodec_set_dimensions(s->avctx, s->width, s->height);
e5d40372 2480 s->avctx->sample_aspect_ratio = h->sps.sar;
cfa5a81e 2481 av_assert0(s->avctx->sample_aspect_ratio.den);
f3bdc3da 2482
e5d40372
DB
2483 if (h->sps.video_signal_type_present_flag) {
2484 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
2485 : AVCOL_RANGE_MPEG;
2486 if (h->sps.colour_description_present_flag) {
c4dffe7e
DC
2487 s->avctx->color_primaries = h->sps.color_primaries;
2488 s->avctx->color_trc = h->sps.color_trc;
2489 s->avctx->colorspace = h->sps.colorspace;
2490 }
2491 }
2492
e5d40372
DB
2493 if (h->sps.timing_info_present_flag) {
2494 int64_t den = h->sps.time_scale;
2495 if (h->x264_build < 44U)
3102d180 2496 den *= 2;
f3bdc3da 2497 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
e5d40372 2498 h->sps.num_units_in_tick, den, 1 << 30);
f3bdc3da 2499 }
d545cf80
OA
2500
2501 switch (h->sps.bit_depth_luma) {
e5d40372
DB
2502 case 9:
2503 if (CHROMA444) {
2504 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
716d413c 2505 s->avctx->pix_fmt = AV_PIX_FMT_GBRP9;
e5d40372 2506 } else
716d413c 2507 s->avctx->pix_fmt = AV_PIX_FMT_YUV444P9;
e5d40372 2508 } else if (CHROMA422)
716d413c 2509 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P9;
e5d40372 2510 else
716d413c 2511 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P9;
e5d40372
DB
2512 break;
2513 case 10:
2514 if (CHROMA444) {
2515 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
716d413c 2516 s->avctx->pix_fmt = AV_PIX_FMT_GBRP10;
e5d40372 2517 } else
716d413c 2518 s->avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
e5d40372 2519 } else if (CHROMA422)
716d413c 2520 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
e5d40372 2521 else
716d413c 2522 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P10;
e5d40372
DB
2523 break;
2524 case 8:
2525 if (CHROMA444) {
2526 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
716d413c 2527 s->avctx->pix_fmt = AV_PIX_FMT_GBRP;
e5d40372 2528 } else
716d413c
AK
2529 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
2530 : AV_PIX_FMT_YUV444P;
e5d40372 2531 } else if (CHROMA422) {
716d413c
AK
2532 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
2533 : AV_PIX_FMT_YUV422P;
e5d40372
DB
2534 } else {
2535 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2536 s->avctx->codec->pix_fmts ?
2537 s->avctx->codec->pix_fmts :
2538 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2539 hwaccel_pixfmt_list_h264_jpeg_420 :
2540 ff_hwaccel_pixfmt_list_420);
2541 }
2542 break;
2543 default:
2544 av_log(s->avctx, AV_LOG_ERROR,
2545 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2546 return AVERROR_INVALIDDATA;
d545cf80
OA
2547 }
2548
e5d40372
DB
2549 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id,
2550 s->avctx->pix_fmt);
f3bdc3da 2551
efd29844
MS
2552 if (ff_MPV_common_init(s) < 0) {
2553 av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
0da71265 2554 return -1;
33aec3f4 2555 }
12d96de3 2556 s->first_field = 0;
b19d493f 2557 h->prev_interlaced_frame = 1;
115329f1 2558
b41c1db3 2559 init_scan_tables(h);
bac3ab13 2560 if (ff_h264_alloc_tables(h) < 0) {
e5d40372
DB
2561 av_log(h->s.avctx, AV_LOG_ERROR,
2562 "Could not allocate memory for h264\n");
bac3ab13
DB
2563 return AVERROR(ENOMEM);
2564 }
0da71265 2565
e5d40372 2566 if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_SLICE)) {
33aec3f4
MN
2567 if (context_init(h) < 0) {
2568 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
afebe2f7 2569 return -1;
33aec3f4 2570 }
6a9c8594 2571 } else {
e5d40372 2572 for (i = 1; i < s->slice_context_count; i++) {
6a9c8594
AS
2573 H264Context *c;
2574 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2575 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2576 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
e5d40372
DB
2577 c->h264dsp = h->h264dsp;
2578 c->sps = h->sps;
2579 c->pps = h->pps;
6a9c8594
AS
2580 c->pixel_shift = h->pixel_shift;
2581 init_scan_tables(c);
2582 clone_tables(c, h, i);
2583 }
2584
e5d40372 2585 for (i = 0; i < s->slice_context_count; i++)
33aec3f4 2586 if (context_init(h->thread_context[i]) < 0) {
e5d40372
DB
2587 av_log(h->s.avctx, AV_LOG_ERROR,
2588 "context_init() failed.\n");
6a9c8594 2589 return -1;
33aec3f4 2590 }
6a9c8594 2591 }
0da71265
MN
2592 }
2593
e5d40372 2594 if (h == h0 && h->dequant_coeff_pps != pps_id) {
0ce4fe48
RB
2595 h->dequant_coeff_pps = pps_id;
2596 init_dequant_tables(h);
2597 }
2598
e5d40372 2599 h->frame_num = get_bits(&s->gb, h->sps.log2_max_frame_num);
0da71265 2600
e5d40372
DB
2601 h->mb_mbaff = 0;
2602 h->mb_aff_frame = 0;
12d96de3 2603 last_pic_structure = s0->picture_structure;
ba0c8981
DB
2604 last_pic_droppable = s0->droppable;
2605 s->droppable = h->nal_ref_idc == 0;
e5d40372
DB
2606 if (h->sps.frame_mbs_only_flag) {
2607 s->picture_structure = PICT_FRAME;
2608 } else {
2609 if (get_bits1(&s->gb)) { // field_pic_flag
2610 s->picture_structure = PICT_TOP_FIELD + get_bits1(&s->gb); // bottom_field_flag
6ba71fc4 2611 } else {
e5d40372
DB
2612 s->picture_structure = PICT_FRAME;
2613 h->mb_aff_frame = h->sps.mb_aff;
6867a90b 2614 }
0da71265 2615 }
e5d40372 2616 h->mb_field_decoding_flag = s->picture_structure != PICT_FRAME;
2ddcf84b 2617
1e26a48f
RB
2618 if (h0->current_slice != 0) {
2619 if (last_pic_structure != s->picture_structure ||
ba0c8981 2620 last_pic_droppable != s->droppable) {
1e26a48f
RB
2621 av_log(h->s.avctx, AV_LOG_ERROR,
2622 "Changing field mode (%d -> %d) between slices is not allowed\n",
2623 last_pic_structure, s->picture_structure);
2624 s->picture_structure = last_pic_structure;
ba0c8981 2625 s->droppable = last_pic_droppable;
1e26a48f 2626 return AVERROR_INVALIDDATA;
5945c7b3 2627 } else if (!s0->current_picture_ptr) {
0b300daa
JG
2628 av_log(s->avctx, AV_LOG_ERROR,
2629 "unset current_picture_ptr on %d. slice\n",
2630 h0->current_slice + 1);
2631 return AVERROR_INVALIDDATA;
1e26a48f
RB
2632 }
2633 } else {
e5d40372
DB
2634 /* Shorten frame num gaps so we don't have to allocate reference
2635 * frames just to throw them away */
2636 if (h->frame_num != h->prev_frame_num) {
2637 int unwrap_prev_frame_num = h->prev_frame_num;
2638 int max_frame_num = 1 << h->sps.log2_max_frame_num;
3803af22 2639
e5d40372
DB
2640 if (unwrap_prev_frame_num > h->frame_num)
2641 unwrap_prev_frame_num -= max_frame_num;
3803af22
AS
2642
2643 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2644 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2645 if (unwrap_prev_frame_num < 0)
2646 unwrap_prev_frame_num += max_frame_num;
2647
2648 h->prev_frame_num = unwrap_prev_frame_num;
2649 }
2650 }
6a9c8594 2651
1e26a48f
RB
2652 /* See if we have a decoded first field looking for a pair...
2653 * Here, we're using that to see if we should mark previously
2654 * decode frames as "finished".
2655 * We have to do that before the "dummy" in-between frame allocation,
2656 * since that can modify s->current_picture_ptr. */
2657 if (s0->first_field) {
2658 assert(s0->current_picture_ptr);
2659 assert(s0->current_picture_ptr->f.data[0]);
2660 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2661
2662 /* Mark old field/frame as completed */
ba0c8981 2663 if (!last_pic_droppable && s0->current_picture_ptr->owner2 == s0) {
1e26a48f
RB
2664 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2665 last_pic_structure == PICT_BOTTOM_FIELD);
2666 }
2667
2668 /* figure out if we have a complementary field pair */
2669 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2670 /* Previous field is unmatched. Don't display it, but let it
2671 * remain for reference if marked as such. */
ba0c8981 2672 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1e26a48f
RB
2673 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2674 last_pic_structure == PICT_TOP_FIELD);
2675 }
2676 } else {
2677 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2678 /* This and previous field were reference, but had
2679 * different frame_nums. Consider this field first in
2680 * pair. Throw away previous field except for reference
2681 * purposes. */
ba0c8981 2682 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1e26a48f
RB
2683 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2684 last_pic_structure == PICT_TOP_FIELD);
2685 }
2686 } else {
2687 /* Second field in complementary pair */
2688 if (!((last_pic_structure == PICT_TOP_FIELD &&
2689 s->picture_structure == PICT_BOTTOM_FIELD) ||
2690 (last_pic_structure == PICT_BOTTOM_FIELD &&
2691 s->picture_structure == PICT_TOP_FIELD))) {
2692 av_log(s->avctx, AV_LOG_ERROR,
2693 "Invalid field mode combination %d/%d\n",
2694 last_pic_structure, s->picture_structure);
2695 s->picture_structure = last_pic_structure;
ba0c8981 2696 s->droppable = last_pic_droppable;
1e26a48f 2697 return AVERROR_INVALIDDATA;
ba0c8981 2698 } else if (last_pic_droppable != s->droppable) {
1e26a48f
RB
2699 av_log(s->avctx, AV_LOG_ERROR,
2700 "Cannot combine reference and non-reference fields in the same frame\n");
2701 av_log_ask_for_sample(s->avctx, NULL);
2702 s->picture_structure = last_pic_structure;
ba0c8981 2703 s->droppable = last_pic_droppable;
1e26a48f
RB
2704 return AVERROR_INVALIDDATA;
2705 }
2706
2707 /* Take ownership of this buffer. Note that if another thread owned
2708 * the first field of this buffer, we're not operating on that pointer,
2709 * so the original thread is still responsible for reporting progress
2710 * on that first field (or if that was us, we just did that above).
2711 * By taking ownership, we assign responsibility to ourselves to
2712 * report progress on the second field. */
2713 s0->current_picture_ptr->owner2 = s0;
2714 }
2715 }
2716 }
2717
e5d40372
DB
2718 while (h->frame_num != h->prev_frame_num &&
2719 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
4dece8c7 2720 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
e5d40372
DB
2721 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
2722 h->frame_num, h->prev_frame_num);
903d58f6 2723 if (ff_h264_frame_start(h) < 0)
66e6038c 2724 return -1;
26b86e47 2725 h->prev_frame_num++;
e5d40372
DB
2726 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
2727 s->current_picture_ptr->frame_num = h->prev_frame_num;
47c0ac96
DB
2728 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
2729 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 1);
3d542120 2730 ff_generate_sliding_window_mmcos(h);
12fe7594 2731 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
9abc9873 2732 (s->avctx->err_recognition & AV_EF_EXPLODE))
12fe7594 2733 return AVERROR_INVALIDDATA;
e2983d6e
JGG
2734 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2735 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2736 * about there being no actual duplicates.
2737 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
da9cea77 2738 * concealing a lost frame, this probably isn't noticeable by comparison, but it should
e2983d6e 2739 * be fixed. */
4dece8c7
JGG
2740 if (h->short_ref_count) {
2741 if (prev) {
657ccb5a 2742 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
e5d40372
DB
2743 (const uint8_t **)prev->f.data, prev->f.linesize,
2744 s->avctx->pix_fmt, s->mb_width * 16, s->mb_height * 16);
2745 h->short_ref[0]->poc = prev->poc + 2;
4dece8c7
JGG
2746 }
2747 h->short_ref[0]->frame_num = h->prev_frame_num;
2748 }
26b86e47
MN
2749 }
2750
1e26a48f
RB
2751 /* See if we have a decoded first field looking for a pair...
2752 * We're using that to see whether to continue decoding in that
2753 * frame, or to allocate a new one. */
12d96de3
JD
2754 if (s0->first_field) {
2755 assert(s0->current_picture_ptr);
657ccb5a 2756 assert(s0->current_picture_ptr->f.data[0]);
95a06eb4 2757 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
12d96de3
JD
2758
2759 /* figure out if we have a complementary field pair */
2760 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
e5d40372
DB
2761 /* Previous field is unmatched. Don't display it, but let it
2762 * remain for reference if marked as such. */
12d96de3 2763 s0->current_picture_ptr = NULL;
e5d40372 2764 s0->first_field = FIELD_PICTURE;
12d96de3 2765 } else {
1e26a48f
RB
2766 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2767 /* This and the previous field had different frame_nums.
2768 * Consider this field first in pair. Throw away previous
2769 * one except for reference purposes. */
e5d40372 2770 s0->first_field = 1;
12d96de3 2771 s0->current_picture_ptr = NULL;
12d96de3
JD
2772 } else {
2773 /* Second field in complementary pair */
2774 s0->first_field = 0;
2775 }
2776 }
12d96de3
JD
2777 } else {
2778 /* Frame or first field in a potentially complementary pair */
12d96de3
JD
2779 s0->first_field = FIELD_PICTURE;
2780 }
2781
e5d40372 2782 if (!FIELD_PICTURE || s0->first_field) {
6a9c8594
AS
2783 if (ff_h264_frame_start(h) < 0) {
2784 s0->first_field = 0;
2785 return -1;
2786 }
2787 } else {
2788 ff_release_unused_pictures(s, 0);
12d96de3 2789 }
2ddcf84b 2790 }
e5d40372 2791 if (h != h0)
2ddcf84b
JD
2792 clone_slice(h, h0);
2793
e5d40372 2794 s->current_picture_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
2ddcf84b 2795
88e7a4d1 2796 assert(s->mb_num == s->mb_width * s->mb_height);
e5d40372
DB
2797 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2798 first_mb_in_slice >= s->mb_num) {
88e7a4d1 2799 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
6b53b87e
MN
2800 return -1;
2801 }
e5d40372 2802 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
f3e53d9f
JD
2803 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2804 if (s->picture_structure == PICT_BOTTOM_FIELD)
2805 s->resync_mb_y = s->mb_y = s->mb_y + 1;
88e7a4d1 2806 assert(s->mb_y < s->mb_height);
115329f1 2807
e5d40372
DB
2808 if (s->picture_structure == PICT_FRAME) {
2809 h->curr_pic_num = h->frame_num;
2810 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
2811 } else {
2812 h->curr_pic_num = 2 * h->frame_num + 1;
2813 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
0da71265 2814 }
115329f1 2815
e5d40372 2816 if (h->nal_unit_type == NAL_IDR_SLICE)
1df1df0b 2817 get_ue_golomb(&s->gb); /* idr_pic_id */
115329f1 2818
e5d40372
DB
2819 if (h->sps.poc_type == 0) {
2820 h->poc_lsb = get_bits(&s->gb, h->sps.log2_max_poc_lsb);
115329f1 2821
e5d40372
DB
2822 if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
2823 h->delta_poc_bottom = get_se_golomb(&s->gb);
0da71265 2824 }
115329f1 2825
e5d40372
DB
2826 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
2827 h->delta_poc[0] = get_se_golomb(&s->gb);
115329f1 2828
e5d40372
DB
2829 if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
2830 h->delta_poc[1] = get_se_golomb(&s->gb);
0da71265 2831 }
115329f1 2832
0da71265 2833 init_poc(h);
115329f1 2834
e5d40372
DB
2835 if (h->pps.redundant_pic_cnt_present)
2836 h->redundant_pic_count = get_ue_golomb(&s->gb);
0da71265 2837
e5d40372
DB
2838 // set defaults, might be overridden a few lines later
2839 h->ref_count[0] = h->pps.ref_count[0];
2840 h->ref_count[1] = h->pps.ref_count[1];
0da71265 2841
e5d40372 2842 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
e5d40372
DB
2843 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2844 h->direct_spatial_mv_pred = get_bits1(&s->gb);
2845 num_ref_idx_active_override_flag = get_bits1(&s->gb);
115329f1 2846
e5d40372
DB
2847 if (num_ref_idx_active_override_flag) {
2848 h->ref_count[0] = get_ue_golomb(&s->gb) + 1;
6e5cdf26
JG
2849 if (h->ref_count[0] < 1)
2850 return AVERROR_INVALIDDATA;
2851 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
e5d40372 2852 h->ref_count[1] = get_ue_golomb(&s->gb) + 1;
6e5cdf26
JG
2853 if (h->ref_count[1] < 1)
2854 return AVERROR_INVALIDDATA;
2855 }
e0febda2 2856 }
0da71265 2857
e5d40372
DB
2858 if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2859 h->list_count = 2;
187696fa 2860 else
e5d40372
DB
2861 h->list_count = 1;
2862 } else
2863 h->list_count = 0;
0da71265 2864
60b6b8c0
JG
2865 max_refs = s->picture_structure == PICT_FRAME ? 16 : 32;
2866
2867 if (h->ref_count[0] > max_refs || h->ref_count[1] > max_refs) {
2868 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2869 h->ref_count[0] = h->ref_count[1] = 1;
2870 return AVERROR_INVALIDDATA;
2871 }
2872
e5d40372 2873 if (!default_ref_list_done)
ea6f00c4 2874 ff_h264_fill_default_ref_list(h);
0da71265 2875
e5d40372
DB
2876 if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
2877 ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2878 h->ref_count[1] = h->ref_count[0] = 0;
806bb93f 2879 return -1;
4c7a232f 2880 }
0da71265 2881
e5d40372
DB
2882 if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
2883 s->last_picture_ptr = &h->ref_list[0][0];
d5e83122 2884 s->last_picture_ptr->owner2 = s;
8d2fc163 2885 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
07dff5c7 2886 }
e5d40372
DB
2887 if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
2888 s->next_picture_ptr = &h->ref_list[1][0];
d5e83122 2889 s->next_picture_ptr->owner2 = s;
8d2fc163 2890 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
07dff5c7
MN
2891 }
2892
e5d40372
DB
2893 if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
2894 (h->pps.weighted_bipred_idc == 1 &&
2895 h->slice_type_nos == AV_PICTURE_TYPE_B))
0da71265 2896 pred_weight_table(h);
e5d40372
DB
2897 else if (h->pps.weighted_bipred_idc == 2 &&
2898 h->slice_type_nos == AV_PICTURE_TYPE_B) {
1052b76f 2899 implicit_weight_table(h, -1);
e5d40372 2900 } else {
9f2d1b4f 2901 h->use_weight = 0;
cb99c652
GB
2902 for (i = 0; i < 2; i++) {
2903 h->luma_weight_flag[i] = 0;
2904 h->chroma_weight_flag[i] = 0;
2905 }
2906 }
115329f1 2907
e5d40372
DB
2908 if (h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
2909 (s->avctx->err_recognition & AV_EF_EXPLODE))
12fe7594 2910 return AVERROR_INVALIDDATA;
0da71265 2911
e5d40372 2912 if (FRAME_MBAFF) {
ea6f00c4 2913 ff_h264_fill_mbaff_ref_list(h);
5d18eaad 2914
e5d40372 2915 if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1052b76f
MN
2916 implicit_weight_table(h, 0);
2917 implicit_weight_table(h, 1);
2918 }
2919 }
2920
e5d40372 2921 if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
943f69a6
MN
2922 ff_h264_direct_dist_scale_factor(h);
2923 ff_h264_direct_ref_list_init(h);
8f56e219 2924
e5d40372 2925 if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
9963b332 2926 tmp = get_ue_golomb_31(&s->gb);
e5d40372 2927 if (tmp > 2) {
88e7a4d1
MN
2928 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
2929 return -1;
2930 }
e5d40372 2931 h->cabac_init_idc = tmp;
88e7a4d1 2932 }
e5017ab8
LA
2933
2934 h->last_qscale_diff = 0;
88e7a4d1 2935 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
e5d40372 2936 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
88e7a4d1 2937 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3ebc7e04
MN
2938 return -1;
2939 }
e5d40372 2940 s->qscale = tmp;
4691a77d
2941 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
2942 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
e5d40372
DB
2943 // FIXME qscale / qp ... stuff
2944 if (h->slice_type == AV_PICTURE_TYPE_SP)
1df1df0b 2945 get_bits1(&s->gb); /* sp_for_switch_flag */
e5d40372
DB
2946 if (h->slice_type == AV_PICTURE_TYPE_SP ||
2947 h->slice_type == AV_PICTURE_TYPE_SI)
1df1df0b 2948 get_se_golomb(&s->gb); /* slice_qs_delta */
0da71265 2949
e5d40372 2950 h->deblocking_filter = 1;
0c32e19d 2951 h->slice_alpha_c0_offset = 52;
e5d40372
DB
2952 h->slice_beta_offset = 52;
2953 if (h->pps.deblocking_filter_parameters_present) {
2954 tmp = get_ue_golomb_31(&s->gb);
2955 if (tmp > 2) {
2956 av_log(s->avctx, AV_LOG_ERROR,
2957 "deblocking_filter_idc %u out of range\n", tmp);
88e7a4d1
MN
2958 return -1;
2959 }
e5d40372
DB
2960 h->deblocking_filter = tmp;
2961 if (h->deblocking_filter < 2)
2962 h->deblocking_filter ^= 1; // 1<->0
53c05b1e 2963
e5d40372 2964 if (h->deblocking_filter) {
0c32e19d
MN
2965 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
2966 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
e5d40372
DB
2967 if (h->slice_alpha_c0_offset > 104U ||
2968 h->slice_beta_offset > 104U) {
2969 av_log(s->avctx, AV_LOG_ERROR,
2970 "deblocking filter parameters %d %d out of range\n",
2971 h->slice_alpha_c0_offset, h->slice_beta_offset);
0c32e19d
MN
2972 return -1;
2973 }
0da71265 2974 }
980a82b7 2975 }
afebe2f7 2976
e5d40372
DB
2977 if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
2978 (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
2979 h->slice_type_nos != AV_PICTURE_TYPE_I) ||
2980 (s->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
2981 h->slice_type_nos == AV_PICTURE_TYPE_B) ||
2982 (s->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
2983 h->nal_ref_idc == 0))
2984 h->deblocking_filter = 0;
61858a76 2985
e5d40372
DB
2986 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
2987 if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
ec970c21 2988 /* Cheat slightly for speed:
e5d40372 2989 * Do not bother to deblock across slices. */
ec970c21
2990 h->deblocking_filter = 2;
2991 } else {
7ae94d52 2992 h0->max_contexts = 1;
e5d40372
DB
2993 if (!h0->single_decode_warning) {
2994 av_log(s->avctx, AV_LOG_INFO,
2995 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
7ae94d52
2996 h0->single_decode_warning = 1;
2997 }
33aec3f4 2998 if (h != h0) {
e5d40372
DB
2999 av_log(h->s.avctx, AV_LOG_ERROR,
3000 "Deblocking switched inside frame.\n");
33aec3f4
MN
3001 return 1;
3002 }
ec970c21 3003 }
afebe2f7 3004 }
e5d40372
DB
3005 h->qp_thresh = 15 + 52 -
3006 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3007 FFMAX3(0,
3008 h->pps.chroma_qp_index_offset[0],
3009 h->pps.chroma_qp_index_offset[1]) +
3010 6 * (h->sps.bit_depth_luma - 8);
0da71265 3011
afebe2f7
3012 h0->last_slice_type = slice_type;
3013 h->slice_num = ++h0->current_slice;
e5d40372
DB
3014 if (h->slice_num >= MAX_SLICES) {
3015 av_log(s->avctx, AV_LOG_ERROR,
3016 "Too many slices, increase MAX_SLICES and recompile\n");
b735aeea 3017 }
5175b937 3018
e5d40372 3019 for (j = 0; j < 2; j++) {
6d7e6b26 3020 int id_list[16];
e5d40372
DB
3021 int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3022 for (i = 0; i < 16; i++) {
3023 id_list[i] = 60;
657ccb5a 3024 if (h->ref_list[j][i].f.data[0]) {
6d7e6b26 3025 int k;
657ccb5a 3026 uint8_t *base = h->ref_list[j][i].f.base[0];
e5d40372 3027 for (k = 0; k < h->short_ref_count; k++)
657ccb5a 3028 if (h->short_ref[k]->f.base[0] == base) {
e5d40372 3029 id_list[i] = k;
6d7e6b26
MN
3030 break;
3031 }
e5d40372 3032 for (k = 0; k < h->long_ref_count; k++)
657ccb5a 3033 if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
e5d40372 3034 id_list[i] = h->short_ref_count + k;
6d7e6b26
MN
3035 break;
3036 }
3037 }
3038 }
3039
e5d40372
DB
3040 ref2frm[0] =
3041 ref2frm[1] = -1;
3042 for (i = 0; i < 16; i++)
3043 ref2frm[i + 2] = 4 * id_list[i] +
3044 (h->ref_list[j][i].f.reference & 3);
3045 ref2frm[18 + 0] =
3046 ref2frm[18 + 1] = -1;
3047 for (i = 16; i < 48; i++)
3048 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3049 (h->ref_list[j][i].f.reference & 3);
3050 }
3051
3052 // FIXME: fix draw_edges + PAFF + frame threads
3053 h->emu_edge_width = (s->flags & CODEC_FLAG_EMU_EDGE ||
3054 (!h->sps.frame_mbs_only_flag &&
3055 s->avctx->active_thread_type))
3056 ? 0 : 16;
3057 h->emu_edge_height = (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3058
3059 if (s->avctx->debug & FF_DEBUG_PICT_INFO) {
3060 av_log(h->s.avctx, AV_LOG_DEBUG,
3061 "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",
6867a90b 3062 h->slice_num,
e5d40372 3063 (s->picture_structure == PICT_FRAME ? "F" : s->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
115329f1 3064 first_mb_in_slice,
e5d40372
DB
3065 av_get_picture_type_char(h->slice_type),
3066 h->slice_type_fixed ? " fix" : "",
3067 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
0da71265 3068 pps_id, h->frame_num,
e5d40372
DB
3069 s->current_picture_ptr->field_poc[0],
3070 s->current_picture_ptr->field_poc[1],
0da71265
MN
3071 h->ref_count[0], h->ref_count[1],
3072 s->qscale,
e5d40372
DB
3073 h->deblocking_filter,
3074 h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
9f2d1b4f 3075 h->use_weight,
e5d40372
DB
3076 h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3077 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
0da71265
MN
3078 }
3079
3080 return 0;
3081}
3082
0dc343d4 3083int ff_h264_get_slice_type(const H264Context *h)
75dd6938
LA
3084{
3085 switch (h->slice_type) {
e5d40372
DB
3086 case AV_PICTURE_TYPE_P:
3087 return 0;
3088 case AV_PICTURE_TYPE_B:
3089 return 1;
3090 case AV_PICTURE_TYPE_I:
3091 return 2;
3092 case AV_PICTURE_TYPE_SP:
3093 return 3;
3094 case AV_PICTURE_TYPE_SI:
3095 return 4;
3096 default:
3097 return -1;
75dd6938
LA
3098 }
3099}
3100
e5d40372
DB
3101static av_always_inline void fill_filter_caches_inter(H264Context *h,
3102 MpegEncContext *const s,
3103 int mb_type, int top_xy,
3104 int left_xy[LEFT_MBS],
3105 int top_type,
3106 int left_type[LEFT_MBS],
3107 int mb_xy, int list)
cb546946
JGG
3108{
3109 int b_stride = h->b_stride;
e5d40372 3110 int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
cb546946 3111 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
e5d40372
DB
3112 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3113 if (USES_LIST(top_type, list)) {
3114 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
3115 const int b8_xy = 4 * top_xy + 2;
3116 int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2);
3117 AV_COPY128(mv_dst - 1 * 8, s->current_picture.f.motion_val[list][b_xy + 0]);
3118 ref_cache[0 - 1 * 8] =
3119 ref_cache[1 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3120 ref_cache[2 - 1 * 8] =
3121 ref_cache[3 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3122 } else {
3123 AV_ZERO128(mv_dst - 1 * 8);
3124 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
cb546946
JGG
3125 }
3126
e5d40372
DB
3127 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3128 if (USES_LIST(left_type[LTOP], list)) {
3129 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
3130 const int b8_xy = 4 * left_xy[LTOP] + 1;
3131 int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2);
3132 AV_COPY32(mv_dst - 1 + 0, s->current_picture.f.motion_val[list][b_xy + b_stride * 0]);
3133 AV_COPY32(mv_dst - 1 + 8, s->current_picture.f.motion_val[list][b_xy + b_stride * 1]);
3134 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride * 2]);
3135 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride * 3]);
3136 ref_cache[-1 + 0] =
3137 ref_cache[-1 + 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 0]];
3138 ref_cache[-1 + 16] =
3139 ref_cache[-1 + 24] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 1]];
3140 } else {
3141 AV_ZERO32(mv_dst - 1 + 0);
3142 AV_ZERO32(mv_dst - 1 + 8);
3143 AV_ZERO32(mv_dst - 1 + 16);
3144 AV_ZERO32(mv_dst - 1 + 24);