h264: improve parsing of broken AVC SPS
[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
2848ce84 45//#undef NDEBUG
0da71265
MN
46#include <assert.h>
47
fcc0224e
OA
48static const uint8_t rem6[QP_MAX_NUM+1]={
490, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
acd8d10f
PI
50};
51
fcc0224e
OA
52static const uint8_t div6[QP_MAX_NUM+1]={
530, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9,10,10,10,10,
acd8d10f
PI
54};
55
0435fb16
BC
56static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
57 PIX_FMT_DXVA2_VLD,
58 PIX_FMT_VAAPI_VLD,
ac3dbb4d 59 PIX_FMT_VDA_VLD,
0435fb16
BC
60 PIX_FMT_YUVJ420P,
61 PIX_FMT_NONE
62};
63
0da71265 64/**
58c42af7
DB
65 * Check if the top & left blocks are available if needed and
66 * change the dc mode so it only uses the available blocks.
0da71265 67 */
2bedc0e8
MN
68int ff_h264_check_intra4x4_pred_mode(H264Context *h){
69 MpegEncContext * const s = &h->s;
70 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
71 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
72 int i;
73
74 if(!(h->top_samples_available&0x8000)){
75 for(i=0; i<4; i++){
76 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
77 if(status<0){
78 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
79 return -1;
80 } else if(status){
81 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
82 }
83 }
84 }
85
86 if((h->left_samples_available&0x8888)!=0x8888){
87 static const int mask[4]={0x8000,0x2000,0x80,0x20};
88 for(i=0; i<4; i++){
89 if(!(h->left_samples_available&mask[i])){
90 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
91 if(status<0){
92 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
93 return -1;
94 } else if(status){
95 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
96 }
97 }
98 }
99 }
100
101 return 0;
102} //FIXME cleanup like ff_h264_check_intra_pred_mode
103
104/**
58c42af7
DB
105 * Check if the top & left blocks are available if needed and
106 * change the dc mode so it only uses the available blocks.
2bedc0e8 107 */
45b7bd7c 108int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma){
0da71265
MN
109 MpegEncContext * const s = &h->s;
110 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
111 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
115329f1 112
43ff0714 113 if(mode > 6U) {
5175b937 114 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
7440fe83 115 return -1;
5175b937 116 }
115329f1 117
0da71265
MN
118 if(!(h->top_samples_available&0x8000)){
119 mode= top[ mode ];
120 if(mode<0){
9b879566 121 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
0da71265
MN
122 return -1;
123 }
124 }
115329f1 125
d1d10e91 126 if((h->left_samples_available&0x8080) != 0x8080){
0da71265 127 mode= left[ mode ];
45b7bd7c 128 if(is_chroma && (h->left_samples_available&0x8080)){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
d1d10e91
MN
129 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
130 }
0da71265 131 if(mode<0){
9b879566 132 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
0da71265 133 return -1;
115329f1 134 }
0da71265
MN
135 }
136
137 return mode;
138}
139
358ea75e 140const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
0da71265
MN
141 int i, si, di;
142 uint8_t *dst;
24456882 143 int bufidx;
0da71265 144
bb270c08 145// src[0]&0x80; //forbidden bit
0da71265
MN
146 h->nal_ref_idc= src[0]>>5;
147 h->nal_unit_type= src[0]&0x1F;
148
149 src++; length--;
e08715d3 150
b250f9c6
AJ
151#if HAVE_FAST_UNALIGNED
152# if HAVE_FAST_64BIT
e08715d3
MN
153# define RS 7
154 for(i=0; i+1<length; i+=9){
19769ece 155 if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
e08715d3
MN
156# else
157# define RS 3
158 for(i=0; i+1<length; i+=5){
19769ece 159 if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
e08715d3
MN
160# endif
161 continue;
162 if(i>0 && !src[i]) i--;
163 while(src[i]) i++;
164#else
165# define RS 0
0da71265
MN
166 for(i=0; i+1<length; i+=2){
167 if(src[i]) continue;
168 if(i>0 && src[i-1]==0) i--;
e08715d3 169#endif
0da71265
MN
170 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
171 if(src[i+2]!=3){
172 /* startcode, so we must be past the end */
173 length=i;
174 }
175 break;
176 }
abb27cfb 177 i-= RS;
0da71265
MN
178 }
179
180 if(i>=length-1){ //no escaped 0
181 *dst_length= length;
182 *consumed= length+1; //+1 for the header
115329f1 183 return src;
0da71265
MN
184 }
185
24456882 186 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
238ef6da 187 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
24456882 188 dst= h->rbsp_buffer[bufidx];
0da71265 189
ac658be5
FOL
190 if (dst == NULL){
191 return NULL;
192 }
193
3b66c4c5 194//printf("decoding esc\n");
593af7cd
MN
195 memcpy(dst, src, i);
196 si=di=i;
197 while(si+2<length){
0da71265 198 //remove escapes (very rare 1:2^22)
593af7cd
MN
199 if(src[si+2]>3){
200 dst[di++]= src[si++];
201 dst[di++]= src[si++];
202 }else if(src[si]==0 && src[si+1]==0){
0da71265
MN
203 if(src[si+2]==3){ //escape
204 dst[di++]= 0;
205 dst[di++]= 0;
206 si+=3;
c8470cc1 207 continue;
0da71265 208 }else //next start code
593af7cd 209 goto nsc;
0da71265
MN
210 }
211
212 dst[di++]= src[si++];
213 }
593af7cd
MN
214 while(si<length)
215 dst[di++]= src[si++];
216nsc:
0da71265 217
d4369630
AS
218 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
219
0da71265
MN
220 *dst_length= di;
221 *consumed= si + 1;//+1 for the header
90b5b51e 222//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
0da71265
MN
223 return dst;
224}
225
85297319
DEP
226/**
227 * Identify the exact end of the bitstream
228 * @return the length of the trailing, or 0 if damaged
229 */
230static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
0da71265
MN
231 int v= *src;
232 int r;
233
a9c9a240 234 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
0da71265
MN
235
236 for(r=1; r<9; r++){
237 if(v&1) return r;
238 v>>=1;
239 }
240 return 0;
241}
242
6a9c8594
AS
243static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
244 int y_offset, int list){
245 int raw_my= h->mv_cache[list][ scan8[n] ][1];
246 int filter_height= (raw_my&3) ? 2 : 0;
247 int full_my= (raw_my>>2) + y_offset;
248 int top = full_my - filter_height, bottom = full_my + height + filter_height;
249
250 return FFMAX(abs(top), bottom);
251}
252
253static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
254 int y_offset, int list0, int list1, int *nrefs){
255 MpegEncContext * const s = &h->s;
256 int my;
257
258 y_offset += 16*(s->mb_y >> MB_FIELD);
259
260 if(list0){
261 int ref_n = h->ref_cache[0][ scan8[n] ];
262 Picture *ref= &h->ref_list[0][ref_n];
263
264 // Error resilience puts the current picture in the ref list.
265 // Don't try to wait on these as it will cause a deadlock.
266 // Fields can wait on each other, though.
657ccb5a
DB
267 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
268 (ref->f.reference & 3) != s->picture_structure) {
6a9c8594
AS
269 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
270 if (refs[0][ref_n] < 0) nrefs[0] += 1;
271 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
272 }
273 }
274
275 if(list1){
276 int ref_n = h->ref_cache[1][ scan8[n] ];
277 Picture *ref= &h->ref_list[1][ref_n];
278
657ccb5a
DB
279 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
280 (ref->f.reference & 3) != s->picture_structure) {
6a9c8594
AS
281 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
282 if (refs[1][ref_n] < 0) nrefs[1] += 1;
283 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
284 }
285 }
286}
287
288/**
289 * Wait until all reference frames are available for MC operations.
290 *
291 * @param h the H264 context
292 */
293static void await_references(H264Context *h){
294 MpegEncContext * const s = &h->s;
295 const int mb_xy= h->mb_xy;
657ccb5a 296 const int mb_type = s->current_picture.f.mb_type[mb_xy];
6a9c8594
AS
297 int refs[2][48];
298 int nrefs[2] = {0};
299 int ref, list;
300
301 memset(refs, -1, sizeof(refs));
302
303 if(IS_16X16(mb_type)){
304 get_lowest_part_y(h, refs, 0, 16, 0,
305 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
306 }else if(IS_16X8(mb_type)){
307 get_lowest_part_y(h, refs, 0, 8, 0,
308 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
309 get_lowest_part_y(h, refs, 8, 8, 8,
310 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
311 }else if(IS_8X16(mb_type)){
312 get_lowest_part_y(h, refs, 0, 16, 0,
313 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
314 get_lowest_part_y(h, refs, 4, 16, 0,
315 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
316 }else{
317 int i;
318
319 assert(IS_8X8(mb_type));
320
321 for(i=0; i<4; i++){
322 const int sub_mb_type= h->sub_mb_type[i];
323 const int n= 4*i;
324 int y_offset= (i&2)<<2;
325
326 if(IS_SUB_8X8(sub_mb_type)){
327 get_lowest_part_y(h, refs, n , 8, y_offset,
328 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
329 }else if(IS_SUB_8X4(sub_mb_type)){
330 get_lowest_part_y(h, refs, n , 4, y_offset,
331 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
332 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
333 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
334 }else if(IS_SUB_4X8(sub_mb_type)){
335 get_lowest_part_y(h, refs, n , 8, y_offset,
336 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
337 get_lowest_part_y(h, refs, n+1, 8, y_offset,
338 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
339 }else{
340 int j;
341 assert(IS_SUB_4X4(sub_mb_type));
342 for(j=0; j<4; j++){
343 int sub_y_offset= y_offset + 2*(j&2);
344 get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
345 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
346 }
347 }
348 }
349 }
350
351 for(list=h->list_count-1; list>=0; list--){
352 for(ref=0; ref<48 && nrefs[list]; ref++){
353 int row = refs[list][ref];
354 if(row >= 0){
355 Picture *ref_pic = &h->ref_list[list][ref];
657ccb5a 356 int ref_field = ref_pic->f.reference - 1;
6a9c8594
AS
357 int ref_field_picture = ref_pic->field_picture;
358 int pic_height = 16*s->mb_height >> ref_field_picture;
359
360 row <<= MB_MBAFF;
361 nrefs[list]--;
362
363 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
324deaa2
DB
364 ff_thread_await_progress(&ref_pic->f, FFMIN((row >> 1) - !(row & 1), pic_height - 1), 1);
365 ff_thread_await_progress(&ref_pic->f, FFMIN((row >> 1), pic_height - 1), 0);
6a9c8594 366 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
324deaa2 367 ff_thread_await_progress(&ref_pic->f, FFMIN(row * 2 + ref_field, pic_height - 1), 0);
6a9c8594 368 }else if(FIELD_PICTURE){
324deaa2 369 ff_thread_await_progress(&ref_pic->f, FFMIN(row, pic_height - 1), ref_field);
6a9c8594 370 }else{
324deaa2 371 ff_thread_await_progress(&ref_pic->f, FFMIN(row, pic_height - 1), 0);
6a9c8594
AS
372 }
373 }
374 }
375 }
376}
377
66c6b5e2
RB
378#if 0
379/**
380 * DCT transforms the 16 dc values.
381 * @param qp quantization parameter ??? FIXME
382 */
383static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
384// const int qmul= dequant_coeff[qp][0];
385 int i;
386 int temp[16]; //FIXME check if this is a good idea
387 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
388 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
389
390 for(i=0; i<4; i++){
391 const int offset= y_offset[i];
392 const int z0= block[offset+stride*0] + block[offset+stride*4];
393 const int z1= block[offset+stride*0] - block[offset+stride*4];
394 const int z2= block[offset+stride*1] - block[offset+stride*5];
395 const int z3= block[offset+stride*1] + block[offset+stride*5];
396
397 temp[4*i+0]= z0+z3;
398 temp[4*i+1]= z1+z2;
399 temp[4*i+2]= z1-z2;
400 temp[4*i+3]= z0-z3;
401 }
402
403 for(i=0; i<4; i++){
404 const int offset= x_offset[i];
405 const int z0= temp[4*0+i] + temp[4*2+i];
406 const int z1= temp[4*0+i] - temp[4*2+i];
407 const int z2= temp[4*1+i] - temp[4*3+i];
408 const int z3= temp[4*1+i] + temp[4*3+i];
409
410 block[stride*0 +offset]= (z0 + z3)>>1;
411 block[stride*2 +offset]= (z1 + z2)>>1;
412 block[stride*8 +offset]= (z1 - z2)>>1;
413 block[stride*10+offset]= (z0 - z3)>>1;
414 }
415}
416#endif
417
418#undef xStride
419#undef stride
420
66c6b5e2
RB
421#if 0
422static void chroma_dc_dct_c(DCTELEM *block){
423 const int stride= 16*2;
424 const int xStride= 16;
425 int a,b,c,d,e;
426
427 a= block[stride*0 + xStride*0];
428 b= block[stride*0 + xStride*1];
429 c= block[stride*1 + xStride*0];
430 d= block[stride*1 + xStride*1];
431
432 e= a-b;
433 a= a+b;
434 b= c-d;
435 c= c+d;
436
437 block[stride*0 + xStride*0]= (a+c);
438 block[stride*0 + xStride*1]= (e+b);
439 block[stride*1 + xStride*0]= (a-c);
440 block[stride*1 + xStride*1]= (e-b);
441}
442#endif
443
27209bb1
RB
444static av_always_inline void
445mc_dir_part(H264Context *h, Picture *pic, int n, int square,
446 int height, int delta, int list,
447 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
448 int src_x_offset, int src_y_offset,
449 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
450 int pixel_shift, int chroma_idc)
05fb63f5 451{
0da71265
MN
452 MpegEncContext * const s = &h->s;
453 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
5d18eaad 454 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
0da71265 455 const int luma_xy= (mx&3) + ((my&3)<<2);
c90b9442 456 int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
657ccb5a 457 uint8_t * src_y = pic->f.data[0] + offset;
5d18eaad
LM
458 uint8_t * src_cb, * src_cr;
459 int extra_width= h->emu_edge_width;
460 int extra_height= h->emu_edge_height;
0da71265
MN
461 int emu=0;
462 const int full_mx= mx>>2;
463 const int full_my= my>>2;
fbd312fd 464 const int pic_width = 16*s->mb_width;
0d43dd8c 465 const int pic_height = 16*s->mb_height >> MB_FIELD;
229d263c 466 int ysh;
115329f1 467
0da71265
MN
468 if(mx&7) extra_width -= 3;
469 if(my&7) extra_height -= 3;
115329f1
DB
470
471 if( full_mx < 0-extra_width
472 || full_my < 0-extra_height
473 || full_mx + 16/*FIXME*/ > pic_width + extra_width
fbd312fd 474 || full_my + 16/*FIXME*/ > pic_height + extra_height){
229d263c
RB
475 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
476 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
6e3ef511 477 src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
0da71265
MN
478 emu=1;
479 }
115329f1 480
5d18eaad 481 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
0da71265 482 if(!square){
5d18eaad 483 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
0da71265 484 }
115329f1 485
7b442ad9
JGG
486 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
487
05fb63f5 488 if(chroma_idc == 3 /* yuv444 */){
657ccb5a 489 src_cb = pic->f.data[1] + offset;
c90b9442
JGG
490 if(emu){
491 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
492 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
493 src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
494 }
495 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
496 if(!square){
497 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
498 }
499
657ccb5a 500 src_cr = pic->f.data[2] + offset;
c90b9442
JGG
501 if(emu){
502 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
503 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
504 src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
505 }
506 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
507 if(!square){
508 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
509 }
510 return;
511 }
512
05fb63f5
RB
513 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
514 if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
5d18eaad 515 // chroma offset when predicting from a field of opposite parity
657ccb5a 516 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
5d18eaad
LM
517 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
518 }
229d263c
RB
519
520 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
521 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
5d18eaad 522
0da71265 523 if(emu){
05fb63f5
RB
524 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
525 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
526 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
0da71265
MN
527 src_cb= s->edge_emu_buffer;
528 }
05fb63f5
RB
529 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
530 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
0da71265
MN
531
532 if(emu){
05fb63f5
RB
533 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
534 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
535 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
0da71265
MN
536 src_cr= s->edge_emu_buffer;
537 }
05fb63f5
RB
538 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
539 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
0da71265
MN
540}
541
27209bb1
RB
542static av_always_inline void
543mc_part_std(H264Context *h, int n, int square, int height, int delta,
544 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
545 int x_offset, int y_offset,
546 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
547 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
548 int list0, int list1, int pixel_shift, int chroma_idc)
05fb63f5 549{
0da71265
MN
550 MpegEncContext * const s = &h->s;
551 qpel_mc_func *qpix_op= qpix_put;
552 h264_chroma_mc_func chroma_op= chroma_put;
115329f1 553
c90b9442 554 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
05fb63f5 555 if (chroma_idc == 3 /* yuv444 */) {
c90b9442
JGG
556 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
557 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
05fb63f5 558 } else if (chroma_idc == 2 /* yuv422 */) {
229d263c
RB
559 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
560 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
05fb63f5 561 } else /* yuv420 */ {
c90b9442
JGG
562 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
563 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
564 }
0da71265 565 x_offset += 8*s->mb_x;
0d43dd8c 566 y_offset += 8*(s->mb_y >> MB_FIELD);
115329f1 567
0da71265 568 if(list0){
1924f3ce 569 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
c2d33742 570 mc_dir_part(h, ref, n, square, height, delta, 0,
0da71265 571 dest_y, dest_cb, dest_cr, x_offset, y_offset,
05fb63f5 572 qpix_op, chroma_op, pixel_shift, chroma_idc);
0da71265
MN
573
574 qpix_op= qpix_avg;
575 chroma_op= chroma_avg;
576 }
577
578 if(list1){
1924f3ce 579 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
c2d33742 580 mc_dir_part(h, ref, n, square, height, delta, 1,
0da71265 581 dest_y, dest_cb, dest_cr, x_offset, y_offset,
05fb63f5 582 qpix_op, chroma_op, pixel_shift, chroma_idc);
0da71265
MN
583 }
584}
585
27209bb1
RB
586static av_always_inline void
587mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
588 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
589 int x_offset, int y_offset,
590 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
591 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
592 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
593 int list0, int list1, int pixel_shift, int chroma_idc){
9f2d1b4f 594 MpegEncContext * const s = &h->s;
c2d33742 595 int chroma_height;
9f2d1b4f 596
c90b9442 597 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
05fb63f5 598 if (chroma_idc == 3 /* yuv444 */) {
c2d33742 599 chroma_height = height;
c90b9442
JGG
600 chroma_weight_avg = luma_weight_avg;
601 chroma_weight_op = luma_weight_op;
602 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
603 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
05fb63f5 604 } else if (chroma_idc == 2 /* yuv422 */) {
c2d33742 605 chroma_height = height;
229d263c
RB
606 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
607 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
05fb63f5 608 } else /* yuv420 */ {
c2d33742 609 chroma_height = height >> 1;
c90b9442
JGG
610 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
611 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
612 }
9f2d1b4f 613 x_offset += 8*s->mb_x;
0d43dd8c 614 y_offset += 8*(s->mb_y >> MB_FIELD);
115329f1 615
9f2d1b4f
LM
616 if(list0 && list1){
617 /* don't optimize for luma-only case, since B-frames usually
618 * use implicit weights => chroma too. */
619 uint8_t *tmp_cb = s->obmc_scratchpad;
c90b9442
JGG
620 uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
621 uint8_t *tmp_y = s->obmc_scratchpad + 16*h->mb_uvlinesize;
9f2d1b4f
LM
622 int refn0 = h->ref_cache[0][ scan8[n] ];
623 int refn1 = h->ref_cache[1][ scan8[n] ];
624
c2d33742 625 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
9f2d1b4f 626 dest_y, dest_cb, dest_cr,
05fb63f5
RB
627 x_offset, y_offset, qpix_put, chroma_put,
628 pixel_shift, chroma_idc);
c2d33742 629 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
9f2d1b4f 630 tmp_y, tmp_cb, tmp_cr,
05fb63f5
RB
631 x_offset, y_offset, qpix_put, chroma_put,
632 pixel_shift, chroma_idc);
9f2d1b4f
LM
633
634 if(h->use_weight == 2){
1052b76f 635 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
9f2d1b4f 636 int weight1 = 64 - weight0;
c2d33742
RB
637 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize,
638 height, 5, weight0, weight1, 0);
639 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
640 chroma_height, 5, weight0, weight1, 0);
641 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
642 chroma_height, 5, weight0, weight1, 0);
9f2d1b4f 643 }else{
c2d33742 644 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
3d9137c8
MN
645 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
646 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
c2d33742 647 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
3d9137c8
MN
648 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
649 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
c2d33742 650 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
3d9137c8
MN
651 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
652 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
9f2d1b4f
LM
653 }
654 }else{
655 int list = list1 ? 1 : 0;
656 int refn = h->ref_cache[list][ scan8[n] ];
657 Picture *ref= &h->ref_list[list][refn];
c2d33742 658 mc_dir_part(h, ref, n, square, height, delta, list,
9f2d1b4f 659 dest_y, dest_cb, dest_cr, x_offset, y_offset,
05fb63f5 660 qpix_put, chroma_put, pixel_shift, chroma_idc);
9f2d1b4f 661
c2d33742 662 luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
3d9137c8 663 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
9f2d1b4f 664 if(h->use_weight_chroma){
c2d33742 665 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
3d9137c8 666 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
c2d33742 667 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
3d9137c8 668 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
9f2d1b4f
LM
669 }
670 }
671}
672
27209bb1
RB
673static av_always_inline void
674mc_part(H264Context *h, int n, int square, int height, int delta,
675 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
676 int x_offset, int y_offset,
677 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
678 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
679 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
680 int list0, int list1, int pixel_shift, int chroma_idc)
05fb63f5 681{
9f2d1b4f 682 if((h->use_weight==2 && list0 && list1
1052b76f 683 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
9f2d1b4f 684 || h->use_weight==1)
c2d33742 685 mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
9f2d1b4f 686 x_offset, y_offset, qpix_put, chroma_put,
c2d33742 687 weight_op[0], weight_op[1], weight_avg[0],
05fb63f5 688 weight_avg[1], list0, list1, pixel_shift, chroma_idc);
9f2d1b4f 689 else
c2d33742 690 mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
6e3ef511 691 x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
05fb63f5 692 chroma_avg, list0, list1, pixel_shift, chroma_idc);
9f2d1b4f
LM
693}
694
27209bb1
RB
695static av_always_inline void
696prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
697{
513fbd8e
LM
698 /* fetch pixels for estimated mv 4 macroblocks ahead
699 * optimized for 64byte cache lines */
700 MpegEncContext * const s = &h->s;
701 const int refn = h->ref_cache[list][scan8[0]];
702 if(refn >= 0){
703 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
704 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
657ccb5a 705 uint8_t **src = h->ref_list[list][refn].f.data;
6e3ef511 706 int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
513fbd8e 707 s->dsp.prefetch(src[0]+off, s->linesize, 4);
05fb63f5 708 if (chroma_idc == 3 /* yuv444 */) {
c90b9442
JGG
709 s->dsp.prefetch(src[1]+off, s->linesize, 4);
710 s->dsp.prefetch(src[2]+off, s->linesize, 4);
711 }else{
712 off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
713 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
714 }
513fbd8e
LM
715 }
716}
717
6e3ef511 718static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
0da71265 719 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f 720 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
6e3ef511 721 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
05fb63f5
RB
722 int pixel_shift, int chroma_idc)
723{
0da71265 724 MpegEncContext * const s = &h->s;
64514ee8 725 const int mb_xy= h->mb_xy;
657ccb5a 726 const int mb_type = s->current_picture.f.mb_type[mb_xy];
115329f1 727
0da71265 728 assert(IS_INTER(mb_type));
115329f1 729
27237d52 730 if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
6a9c8594 731 await_references(h);
05fb63f5 732 prefetch_motion(h, 0, pixel_shift, chroma_idc);
513fbd8e 733
0da71265 734 if(IS_16X16(mb_type)){
c2d33742 735 mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 736 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
7231ccf4 737 weight_op, weight_avg,
6e3ef511 738 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
05fb63f5 739 pixel_shift, chroma_idc);
0da71265 740 }else if(IS_16X8(mb_type)){
c2d33742 741 mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 742 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
c2d33742 743 weight_op, weight_avg,
6e3ef511 744 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
05fb63f5 745 pixel_shift, chroma_idc);
c2d33742 746 mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
0da71265 747 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
c2d33742 748 weight_op, weight_avg,
6e3ef511 749 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
05fb63f5 750 pixel_shift, chroma_idc);
0da71265 751 }else if(IS_8X16(mb_type)){
c2d33742 752 mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 753 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
c2d33742 754 &weight_op[1], &weight_avg[1],
6e3ef511 755 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
05fb63f5 756 pixel_shift, chroma_idc);
c2d33742 757 mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
0da71265 758 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
c2d33742 759 &weight_op[1], &weight_avg[1],
6e3ef511 760 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
05fb63f5 761 pixel_shift, chroma_idc);
0da71265
MN
762 }else{
763 int i;
115329f1 764
0da71265
MN
765 assert(IS_8X8(mb_type));
766
767 for(i=0; i<4; i++){
768 const int sub_mb_type= h->sub_mb_type[i];
769 const int n= 4*i;
770 int x_offset= (i&1)<<2;
771 int y_offset= (i&2)<<1;
772
773 if(IS_SUB_8X8(sub_mb_type)){
c2d33742 774 mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 775 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
c2d33742 776 &weight_op[1], &weight_avg[1],
6e3ef511 777 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 778 pixel_shift, chroma_idc);
0da71265 779 }else if(IS_SUB_8X4(sub_mb_type)){
c2d33742 780 mc_part(h, n , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 781 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
c2d33742 782 &weight_op[1], &weight_avg[1],
6e3ef511 783 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 784 pixel_shift, chroma_idc);
c2d33742 785 mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
0da71265 786 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
c2d33742 787 &weight_op[1], &weight_avg[1],
6e3ef511 788 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 789 pixel_shift, chroma_idc);
0da71265 790 }else if(IS_SUB_4X8(sub_mb_type)){
c2d33742 791 mc_part(h, n , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 792 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
c2d33742 793 &weight_op[2], &weight_avg[2],
6e3ef511 794 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 795 pixel_shift, chroma_idc);
c2d33742 796 mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
0da71265 797 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
c2d33742 798 &weight_op[2], &weight_avg[2],
6e3ef511 799 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 800 pixel_shift, chroma_idc);
0da71265
MN
801 }else{
802 int j;
803 assert(IS_SUB_4X4(sub_mb_type));
804 for(j=0; j<4; j++){
805 int sub_x_offset= x_offset + 2*(j&1);
806 int sub_y_offset= y_offset + (j&2);
c2d33742 807 mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
0da71265 808 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
c2d33742 809 &weight_op[2], &weight_avg[2],
6e3ef511 810 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 811 pixel_shift, chroma_idc);
0da71265
MN
812 }
813 }
814 }
815 }
513fbd8e 816
05fb63f5
RB
817 prefetch_motion(h, 1, pixel_shift, chroma_idc);
818}
819
820static av_always_inline void
821hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
822 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
823 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
824 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
825 int pixel_shift)
826{
827 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
828 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
829}
830
831static av_always_inline void
832hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
833 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
834 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
835 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
836 int pixel_shift)
837{
838 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
839 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
0da71265
MN
840}
841
91078926 842static void free_tables(H264Context *h, int free_rbsp){
7978debd 843 int i;
afebe2f7 844 H264Context *hx;
6a9c8594 845
0da71265 846 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
847 av_freep(&h->chroma_pred_mode_table);
848 av_freep(&h->cbp_table);
9e528114
LA
849 av_freep(&h->mvd_table[0]);
850 av_freep(&h->mvd_table[1]);
5ad984c9 851 av_freep(&h->direct_table);
0da71265
MN
852 av_freep(&h->non_zero_count);
853 av_freep(&h->slice_table_base);
854 h->slice_table= NULL;
c988f975 855 av_freep(&h->list_counts);
e5017ab8 856
0da71265 857 av_freep(&h->mb2b_xy);
d43c1922 858 av_freep(&h->mb2br_xy);
9f2d1b4f 859
6752dd5a 860 for(i = 0; i < MAX_THREADS; i++) {
afebe2f7
861 hx = h->thread_context[i];
862 if(!hx) continue;
863 av_freep(&hx->top_borders[1]);
864 av_freep(&hx->top_borders[0]);
865 av_freep(&hx->s.obmc_scratchpad);
91078926 866 if (free_rbsp){
fcb7e535
RB
867 av_freep(&hx->rbsp_buffer[1]);
868 av_freep(&hx->rbsp_buffer[0]);
869 hx->rbsp_buffer_size[0] = 0;
870 hx->rbsp_buffer_size[1] = 0;
91078926 871 }
d2d5e067 872 if (i) av_freep(&h->thread_context[i]);
afebe2f7 873 }
0da71265
MN
874}
875
239ea04c 876static void init_dequant8_coeff_table(H264Context *h){
c90b9442 877 int i,j,q,x;
fcc0224e 878 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
239ea04c 879
c90b9442
JGG
880 for(i=0; i<6; i++ ){
881 h->dequant8_coeff[i] = h->dequant8_buffer[i];
882 for(j=0; j<i; j++){
883 if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
884 h->dequant8_coeff[i] = h->dequant8_buffer[j];
885 break;
886 }
239ea04c 887 }
c90b9442
JGG
888 if(j<i)
889 continue;
239ea04c 890
fcc0224e 891 for(q=0; q<max_qp+1; q++){
d9ec210b
DP
892 int shift = div6[q];
893 int idx = rem6[q];
239ea04c 894 for(x=0; x<64; x++)
ca32f7f2 895 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
548a1c8a
LM
896 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
897 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
898 }
899 }
900}
901
902static void init_dequant4_coeff_table(H264Context *h){
903 int i,j,q,x;
fcc0224e 904 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
239ea04c
LM
905 for(i=0; i<6; i++ ){
906 h->dequant4_coeff[i] = h->dequant4_buffer[i];
907 for(j=0; j<i; j++){
908 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
909 h->dequant4_coeff[i] = h->dequant4_buffer[j];
910 break;
911 }
912 }
913 if(j<i)
914 continue;
915
fcc0224e 916 for(q=0; q<max_qp+1; q++){
d9ec210b
DP
917 int shift = div6[q] + 2;
918 int idx = rem6[q];
239ea04c 919 for(x=0; x<16; x++)
ca32f7f2 920 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
ab2e3e2c 921 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
239ea04c
LM
922 h->pps.scaling_matrix4[i][x]) << shift;
923 }
924 }
925}
926
927static void init_dequant_tables(H264Context *h){
928 int i,x;
929 init_dequant4_coeff_table(h);
930 if(h->pps.transform_8x8_mode)
931 init_dequant8_coeff_table(h);
932 if(h->sps.transform_bypass){
933 for(i=0; i<6; i++)
934 for(x=0; x<16; x++)
935 h->dequant4_coeff[i][0][x] = 1<<6;
936 if(h->pps.transform_8x8_mode)
c90b9442 937 for(i=0; i<6; i++)
239ea04c
LM
938 for(x=0; x<64; x++)
939 h->dequant8_coeff[i][0][x] = 1<<6;
940 }
941}
942
943
903d58f6 944int ff_h264_alloc_tables(H264Context *h){
0da71265 945 MpegEncContext * const s = &h->s;
7bc9090a 946 const int big_mb_num= s->mb_stride * (s->mb_height+1);
145061a1 947 const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
239ea04c 948 int x,y;
0da71265 949
145061a1 950 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
e5017ab8 951
c90b9442 952 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 48 * sizeof(uint8_t), fail)
d31dbec3
RP
953 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
954 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
0da71265 955
d31dbec3 956 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
145061a1
MN
957 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
958 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
36b54927 959 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
c988f975 960 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
e5017ab8 961
b735aeea 962 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
5d18eaad 963 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
0da71265 964
d31dbec3 965 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
d43c1922 966 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
0da71265
MN
967 for(y=0; y<s->mb_height; y++){
968 for(x=0; x<s->mb_width; x++){
7bc9090a 969 const int mb_xy= x + y*s->mb_stride;
0da71265 970 const int b_xy = 4*x + 4*y*h->b_stride;
115329f1 971
0da71265 972 h->mb2b_xy [mb_xy]= b_xy;
e1c88a21 973 h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
0da71265
MN
974 }
975 }
9f2d1b4f 976
9c6221ae
GV
977 s->obmc_scratchpad = NULL;
978
56edbd81
LM
979 if(!h->dequant4_coeff[0])
980 init_dequant_tables(h);
981
0da71265
MN
982 return 0;
983fail:
91078926 984 free_tables(h, 1);
0da71265
MN
985 return -1;
986}
987
afebe2f7
988/**
989 * Mimic alloc_tables(), but for every context thread.
990 */
145061a1
MN
991static void clone_tables(H264Context *dst, H264Context *src, int i){
992 MpegEncContext * const s = &src->s;
993 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
afebe2f7
994 dst->non_zero_count = src->non_zero_count;
995 dst->slice_table = src->slice_table;
996 dst->cbp_table = src->cbp_table;
997 dst->mb2b_xy = src->mb2b_xy;
d43c1922 998 dst->mb2br_xy = src->mb2br_xy;
afebe2f7 999 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
145061a1
MN
1000 dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
1001 dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
afebe2f7 1002 dst->direct_table = src->direct_table;
fb823b77 1003 dst->list_counts = src->list_counts;
afebe2f7 1004
afebe2f7 1005 dst->s.obmc_scratchpad = NULL;
76741b0e 1006 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
afebe2f7
1007}
1008
1009/**
1010 * Init context
1011 * Allocate buffers which are not shared amongst multiple threads.
1012 */
1013static int context_init(H264Context *h){
c90b9442
JGG
1014 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1015 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
afebe2f7 1016
145061a1
MN
1017 h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1018 h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1019
afebe2f7
1020 return 0;
1021fail:
1022 return -1; // free_tables will clean up for us
1023}
1024
9855b2e3
MN
1025static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1026
98a6fff9 1027static av_cold void common_init(H264Context *h){
0da71265 1028 MpegEncContext * const s = &h->s;
0da71265
MN
1029
1030 s->width = s->avctx->width;
1031 s->height = s->avctx->height;
1032 s->codec_id= s->avctx->codec->id;
115329f1 1033
76741b0e
BC
1034 ff_h264dsp_init(&h->h264dsp, 8, 1);
1035 ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
0da71265 1036
239ea04c 1037 h->dequant_coeff_pps= -1;
9a41c2c7 1038 s->unrestricted_mv=1;
56edbd81 1039
9cf0841e 1040 ff_dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
a5805aa9 1041
56edbd81
LM
1042 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1043 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
0da71265
MN
1044}
1045
05e95319
HC
1046int ff_h264_decode_extradata(H264Context *h)
1047{
1048 AVCodecContext *avctx = h->s.avctx;
9855b2e3 1049
0c17beba 1050 if(avctx->extradata[0] == 1){
9855b2e3
MN
1051 int i, cnt, nalsize;
1052 unsigned char *p = avctx->extradata;
1053
1054 h->is_avc = 1;
1055
1056 if(avctx->extradata_size < 7) {
1057 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1058 return -1;
1059 }
1060 /* sps and pps in the avcC always have length coded with 2 bytes,
1061 so put a fake nal_length_size = 2 while parsing them */
1062 h->nal_length_size = 2;
1063 // Decode sps from avcC
1064 cnt = *(p+5) & 0x1f; // Number of sps
1065 p += 6;
1066 for (i = 0; i < cnt; i++) {
1067 nalsize = AV_RB16(p) + 2;
d1186ff7
LA
1068 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1069 return -1;
9855b2e3
MN
1070 if(decode_nal_units(h, p, nalsize) < 0) {
1071 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1072 return -1;
1073 }
1074 p += nalsize;
1075 }
1076 // Decode pps from avcC
1077 cnt = *(p++); // Number of pps
1078 for (i = 0; i < cnt; i++) {
1079 nalsize = AV_RB16(p) + 2;
d1186ff7
LA
1080 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1081 return -1;
108f318d 1082 if (decode_nal_units(h, p, nalsize) < 0) {
9855b2e3
MN
1083 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1084 return -1;
1085 }
1086 p += nalsize;
1087 }
1088 // Now store right nal length size, that will be use to parse all other nals
0c17beba 1089 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
9855b2e3
MN
1090 } else {
1091 h->is_avc = 0;
05e95319 1092 if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
9855b2e3
MN
1093 return -1;
1094 }
05e95319
HC
1095 return 0;
1096}
1097
1098av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1099 H264Context *h= avctx->priv_data;
1100 MpegEncContext * const s = &h->s;
ea2bb12e 1101 int i;
05e95319 1102
efd29844 1103 ff_MPV_decode_defaults(s);
05e95319
HC
1104
1105 s->avctx = avctx;
1106 common_init(h);
1107
1108 s->out_format = FMT_H264;
1109 s->workaround_bugs= avctx->workaround_bugs;
1110
1111 // set defaults
1112// s->decode_mb= ff_h263_decode_mb;
1113 s->quarter_sample = 1;
1114 if(!avctx->has_b_frames)
1115 s->low_delay= 1;
1116
1117 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1118
1119 ff_h264_decode_init_vlc();
1120
6e3ef511 1121 h->pixel_shift = 0;
19a0729b 1122 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
6e3ef511 1123
05e95319 1124 h->thread_context[0] = h;
6a9c8594 1125 h->outputed_poc = h->next_outputed_poc = INT_MIN;
ea2bb12e
RB
1126 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1127 h->last_pocs[i] = INT_MIN;
05e95319
HC
1128 h->prev_poc_msb= 1<<16;
1129 h->x264_build = -1;
1130 ff_h264_reset_sei(h);
1131 if(avctx->codec_id == CODEC_ID_H264){
1132 if(avctx->ticks_per_frame == 1){
1133 s->avctx->time_base.den *=2;
1134 }
1135 avctx->ticks_per_frame = 2;
1136 }
1137
1138 if(avctx->extradata_size > 0 && avctx->extradata &&
1139 ff_h264_decode_extradata(h))
1140 return -1;
1141
db8cb47d
MN
1142 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1143 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1144 s->low_delay = 0;
1145 }
9855b2e3 1146
0da71265
MN
1147 return 0;
1148}
1149
6a9c8594
AS
1150#define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1151static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1152{
1153 int i;
1154
1155 for (i=0; i<count; i++){
1156 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1157 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1158 !from[i]));
1159 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1160 }
1161}
1162
1163static void copy_parameter_set(void **to, void **from, int count, int size)
1164{
1165 int i;
1166
1167 for (i=0; i<count; i++){
1168 if (to[i] && !from[i]) av_freep(&to[i]);
1169 else if (from[i] && !to[i]) to[i] = av_malloc(size);
1170
1171 if (from[i]) memcpy(to[i], from[i], size);
1172 }
1173}
1174
1175static int decode_init_thread_copy(AVCodecContext *avctx){
1176 H264Context *h= avctx->priv_data;
1177
f3a29b75
JR
1178 if (!avctx->internal->is_copy)
1179 return 0;
6a9c8594
AS
1180 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1181 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1182
1183 return 0;
1184}
1185
1186#define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1187static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1188 H264Context *h= dst->priv_data, *h1= src->priv_data;
1189 MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1190 int inited = s->context_initialized, err;
1191 int i;
1192
1193 if(dst == src || !s1->context_initialized) return 0;
1194
1195 err = ff_mpeg_update_thread_context(dst, src);
1196 if(err) return err;
1197
1198 //FIXME handle width/height changing
1199 if(!inited){
1200 for(i = 0; i < MAX_SPS_COUNT; i++)
1201 av_freep(h->sps_buffers + i);
1202
1203 for(i = 0; i < MAX_PPS_COUNT; i++)
1204 av_freep(h->pps_buffers + i);
1205
1206 memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1207 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1208 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
bac3ab13
DB
1209 if (ff_h264_alloc_tables(h) < 0) {
1210 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1211 return AVERROR(ENOMEM);
1212 }
6a9c8594
AS
1213 context_init(h);
1214
1215 for(i=0; i<2; i++){
1216 h->rbsp_buffer[i] = NULL;
1217 h->rbsp_buffer_size[i] = 0;
1218 }
1219
1220 h->thread_context[0] = h;
1221
1222 // frame_start may not be called for the next thread (if it's decoding a bottom field)
1223 // so this has to be allocated here
c90b9442 1224 h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
6a9c8594
AS
1225
1226 s->dsp.clear_blocks(h->mb);
c90b9442 1227 s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
6a9c8594
AS
1228 }
1229
1230 //extradata/NAL handling
1231 h->is_avc = h1->is_avc;
1232
1233 //SPS/PPS
1234 copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1235 h->sps = h1->sps;
1236 copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1237 h->pps = h1->pps;
1238
1239 //Dequantization matrices
1240 //FIXME these are big - can they be only copied when PPS changes?
1241 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1242
1243 for(i=0; i<6; i++)
1244 h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1245
c90b9442 1246 for(i=0; i<6; i++)
6a9c8594
AS
1247 h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1248
1249 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1250
1251 //POC timing
1252 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1253
1254 //reference lists
1255 copy_fields(h, h1, ref_count, list_count);
1256 copy_fields(h, h1, ref_list, intra_gb);
1257 copy_fields(h, h1, short_ref, cabac_init_idc);
1258
1259 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1260 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1261 copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1262
1263 h->last_slice_type = h1->last_slice_type;
1264
1265 if(!s->current_picture_ptr) return 0;
1266
1267 if(!s->dropable) {
12fe7594 1268 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6a9c8594
AS
1269 h->prev_poc_msb = h->poc_msb;
1270 h->prev_poc_lsb = h->poc_lsb;
1271 }
1272 h->prev_frame_num_offset= h->frame_num_offset;
1273 h->prev_frame_num = h->frame_num;
1274 h->outputed_poc = h->next_outputed_poc;
1275
12fe7594 1276 return err;
6a9c8594
AS
1277}
1278
903d58f6 1279int ff_h264_frame_start(H264Context *h){
0da71265
MN
1280 MpegEncContext * const s = &h->s;
1281 int i;
6e3ef511 1282 const int pixel_shift = h->pixel_shift;
0da71265 1283
efd29844 1284 if(ff_MPV_frame_start(s, s->avctx) < 0)
af8aa846 1285 return -1;
0da71265 1286 ff_er_frame_start(s);
3a22d7fa 1287 /*
efd29844 1288 * ff_MPV_frame_start uses pict_type to derive key_frame.
3a22d7fa 1289 * This is incorrect for H.264; IDR markings must be used.
1412060e 1290 * Zero here; IDR markings per slice in frame or fields are ORed in later.
3a22d7fa
JD
1291 * See decode_nal_units().
1292 */
657ccb5a 1293 s->current_picture_ptr->f.key_frame = 0;
c173a088 1294 s->current_picture_ptr->mmco_reset= 0;
0da71265
MN
1295
1296 assert(s->linesize && s->uvlinesize);
1297
1298 for(i=0; i<16; i++){
6e3ef511 1299 h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
c90b9442 1300 h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265 1301 }
c90b9442 1302 for(i=0; i<16; i++){
0da71265 1303 h->block_offset[16+i]=
c90b9442
JGG
1304 h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1305 h->block_offset[48+16+i]=
1306 h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
1307 }
1308
934b0821
LM
1309 /* can't be in alloc_tables because linesize isn't known there.
1310 * FIXME: redo bipred weight to not require extra buffer? */
881a5e04 1311 for(i = 0; i < s->slice_context_count; i++)
0f016023 1312 if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
c90b9442 1313 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
5d18eaad 1314
2ce1c2e0 1315 /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
5820b90d 1316 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
934b0821 1317
657ccb5a 1318// s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
28bb9eb2 1319
1412060e 1320 // We mark the current picture as non-reference after allocating it, so
28bb9eb2 1321 // that if we break out due to an error it can be released automatically
efd29844 1322 // in the next ff_MPV_frame_start().
28bb9eb2
MN
1323 // SVQ3 as well as most other codecs have only last/next/current and thus
1324 // get released even with set reference, besides SVQ3 and others do not
1325 // mark frames as reference later "naturally".
1326 if(s->codec_id != CODEC_ID_SVQ3)
657ccb5a 1327 s->current_picture_ptr->f.reference = 0;
357282c6
MN
1328
1329 s->current_picture_ptr->field_poc[0]=
1330 s->current_picture_ptr->field_poc[1]= INT_MAX;
6a9c8594
AS
1331
1332 h->next_output_pic = NULL;
1333
5118c6c7 1334 assert(s->current_picture_ptr->long_ref==0);
357282c6 1335
af8aa846 1336 return 0;
0da71265
MN
1337}
1338
6a9c8594
AS
1339/**
1340 * Run setup operations that must be run after slice header decoding.
1341 * This includes finding the next displayed frame.
1342 *
1343 * @param h h264 master context
1344 * @param setup_finished enough NALs have been read that we can call
1345 * ff_thread_finish_setup()
1346 */
1347static void decode_postinit(H264Context *h, int setup_finished){
1348 MpegEncContext * const s = &h->s;
1349 Picture *out = s->current_picture_ptr;
1350 Picture *cur = s->current_picture_ptr;
1351 int i, pics, out_of_order, out_idx;
adedd840 1352 int invalid = 0, cnt = 0;
6a9c8594 1353
657ccb5a
DB
1354 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1355 s->current_picture_ptr->f.pict_type = s->pict_type;
6a9c8594
AS
1356
1357 if (h->next_output_pic) return;
1358
1359 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1360 //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1361 //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1362 //to find this yet, so we assume the worst for now.
1363 //if (setup_finished)
1364 // ff_thread_finish_setup(s->avctx);
1365 return;
1366 }
1367
657ccb5a
DB
1368 cur->f.interlaced_frame = 0;
1369 cur->f.repeat_pict = 0;
6a9c8594
AS
1370
1371 /* Signal interlacing information externally. */
1372 /* Prioritize picture timing SEI information over used decoding process if it exists. */
1373
1374 if(h->sps.pic_struct_present_flag){
1375 switch (h->sei_pic_struct)
1376 {
1377 case SEI_PIC_STRUCT_FRAME:
1378 break;
1379 case SEI_PIC_STRUCT_TOP_FIELD:
1380 case SEI_PIC_STRUCT_BOTTOM_FIELD:
657ccb5a 1381 cur->f.interlaced_frame = 1;
6a9c8594
AS
1382 break;
1383 case SEI_PIC_STRUCT_TOP_BOTTOM:
1384 case SEI_PIC_STRUCT_BOTTOM_TOP:
1385 if (FIELD_OR_MBAFF_PICTURE)
657ccb5a 1386 cur->f.interlaced_frame = 1;
6a9c8594
AS
1387 else
1388 // try to flag soft telecine progressive
657ccb5a 1389 cur->f.interlaced_frame = h->prev_interlaced_frame;
6a9c8594
AS
1390 break;
1391 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1392 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1393 // Signal the possibility of telecined film externally (pic_struct 5,6)
1394 // From these hints, let the applications decide if they apply deinterlacing.
657ccb5a 1395 cur->f.repeat_pict = 1;
6a9c8594
AS
1396 break;
1397 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1398 // Force progressive here, as doubling interlaced frame is a bad idea.
657ccb5a 1399 cur->f.repeat_pict = 2;
6a9c8594
AS
1400 break;
1401 case SEI_PIC_STRUCT_FRAME_TRIPLING:
657ccb5a 1402 cur->f.repeat_pict = 4;
6a9c8594
AS
1403 break;
1404 }
1405
1406 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
657ccb5a 1407 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
6a9c8594
AS
1408 }else{
1409 /* Derive interlacing flag from used decoding process. */
657ccb5a 1410 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
6a9c8594 1411 }
657ccb5a 1412 h->prev_interlaced_frame = cur->f.interlaced_frame;
6a9c8594
AS
1413
1414 if (cur->field_poc[0] != cur->field_poc[1]){
1415 /* Derive top_field_first from field pocs. */
657ccb5a 1416 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
6a9c8594 1417 }else{
657ccb5a 1418 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
6a9c8594
AS
1419 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1420 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1421 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
657ccb5a 1422 cur->f.top_field_first = 1;
6a9c8594 1423 else
657ccb5a 1424 cur->f.top_field_first = 0;
6a9c8594
AS
1425 }else{
1426 /* Most likely progressive */
657ccb5a 1427 cur->f.top_field_first = 0;
6a9c8594
AS
1428 }
1429 }
1430
1431 //FIXME do something with unavailable reference frames
1432
1433 /* Sort B-frames into display order */
1434
1435 if(h->sps.bitstream_restriction_flag
1436 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1437 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1438 s->low_delay = 0;
1439 }
1440
1441 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1442 && !h->sps.bitstream_restriction_flag){
2574f08d 1443 s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
6a9c8594
AS
1444 s->low_delay= 0;
1445 }
1446
1447 pics = 0;
1448 while(h->delayed_pic[pics]) pics++;
1449
1450 assert(pics <= MAX_DELAYED_PIC_COUNT);
1451
1452 h->delayed_pic[pics++] = cur;
657ccb5a
DB
1453 if (cur->f.reference == 0)
1454 cur->f.reference = DELAYED_PIC_REF;
6a9c8594 1455
adedd840
RB
1456 /* Frame reordering. This code takes pictures from coding order and sorts
1457 * them by their incremental POC value into display order. It supports POC
1458 * gaps, MMCO reset codes and random resets.
1459 * A "display group" can start either with a IDR frame (f.key_frame = 1),
1460 * and/or can be closed down with a MMCO reset code. In sequences where
1461 * there is no delay, we can't detect that (since the frame was already
1462 * output to the user), so we also set h->mmco_reset to detect the MMCO
1463 * reset code.
1464 * FIXME: if we detect insufficient delays (as per s->avctx->has_b_frames),
1465 * we increase the delay between input and output. All frames affected by
1466 * the lag (e.g. those that should have been output before another frame
1467 * that we already returned to the user) will be dropped. This is a bug
1468 * that we will fix later. */
1469 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
1470 cnt += out->poc < h->last_pocs[i];
1471 invalid += out->poc == INT_MIN;
1472 }
1473 if (!h->mmco_reset && !cur->f.key_frame && cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
1474 h->mmco_reset = 2;
1475 if (pics > 1)
1476 h->delayed_pic[pics - 2]->mmco_reset = 2;
1477 }
1478 if (h->mmco_reset || cur->f.key_frame) {
1479 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1480 h->last_pocs[i] = INT_MIN;
1481 cnt = 0;
1482 invalid = MAX_DELAYED_PIC_COUNT;
1483 }
6a9c8594
AS
1484 out = h->delayed_pic[0];
1485 out_idx = 0;
adedd840
RB
1486 for (i = 1; i < MAX_DELAYED_PIC_COUNT && h->delayed_pic[i] &&
1487 !h->delayed_pic[i-1]->mmco_reset && !h->delayed_pic[i]->f.key_frame; i++)
1488 {
6a9c8594
AS
1489 if(h->delayed_pic[i]->poc < out->poc){
1490 out = h->delayed_pic[i];
1491 out_idx = i;
1492 }
adedd840
RB
1493 }
1494 if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
1495 h->next_outputed_poc = INT_MIN;
1496 out_of_order = !out->f.key_frame && !h->mmco_reset && (out->poc < h->next_outputed_poc);
6a9c8594
AS
1497
1498 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1499 { }
ea2bb12e
RB
1500 else if (out_of_order && pics-1 == s->avctx->has_b_frames &&
1501 s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
ea2bb12e
RB
1502 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
1503 s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
ea2bb12e
RB
1504 }
1505 s->low_delay = 0;
1506 } else if (s->low_delay &&
1507 ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2) ||
1508 cur->f.pict_type == AV_PICTURE_TYPE_B)) {
6a9c8594
AS
1509 s->low_delay = 0;
1510 s->avctx->has_b_frames++;
1511 }
1512
adedd840 1513 if(pics > s->avctx->has_b_frames){
657ccb5a 1514 out->f.reference &= ~DELAYED_PIC_REF;
6a9c8594
AS
1515 out->owner2 = s; // for frame threading, the owner must be the second field's thread
1516 // or else the first thread can release the picture and reuse it unsafely
1517 for(i=out_idx; h->delayed_pic[i]; i++)
1518 h->delayed_pic[i] = h->delayed_pic[i+1];
1519 }
ea2bb12e 1520 memmove(h->last_pocs, &h->last_pocs[1], sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
adedd840 1521 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
6a9c8594
AS
1522 if(!out_of_order && pics > s->avctx->has_b_frames){
1523 h->next_output_pic = out;
adedd840
RB
1524 if (out->mmco_reset) {
1525 if (out_idx > 0) {
1526 h->next_outputed_poc = out->poc;
1527 h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
1528 } else {
1529 h->next_outputed_poc = INT_MIN;
1530 }
1531 } else {
0b4c3232
RB
1532 if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f.key_frame) {
1533 h->next_outputed_poc = INT_MIN;
1534 } else {
1535 h->next_outputed_poc = out->poc;
1536 }
adedd840
RB
1537 }
1538 h->mmco_reset = 0;
6a9c8594
AS
1539 }else{
1540 av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1541 }
1542
1543 if (setup_finished)
1544 ff_thread_finish_setup(s->avctx);
1545}
1546
76741b0e
BC
1547static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1548 uint8_t *src_cb, uint8_t *src_cr,
1549 int linesize, int uvlinesize, int simple)
1550{
53c05b1e 1551 MpegEncContext * const s = &h->s;
0b69d625 1552 uint8_t *top_border;
5f7f9719 1553 int top_idx = 1;
6e3ef511 1554 const int pixel_shift = h->pixel_shift;
76741b0e
BC
1555 int chroma444 = CHROMA444;
1556 int chroma422 = CHROMA422;
115329f1 1557
53c05b1e
MN
1558 src_y -= linesize;
1559 src_cb -= uvlinesize;
1560 src_cr -= uvlinesize;
1561
5f7f9719
MN
1562 if(!simple && FRAME_MBAFF){
1563 if(s->mb_y&1){
5f7f9719 1564 if(!MB_MBAFF){
0b69d625
AS
1565 top_border = h->top_borders[0][s->mb_x];
1566 AV_COPY128(top_border, src_y + 15*linesize);
6e3ef511
OA
1567 if (pixel_shift)
1568 AV_COPY128(top_border+16, src_y+15*linesize+16);
49fb20cb 1569 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c90b9442
JGG
1570 if(chroma444){
1571 if (pixel_shift){
1572 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1573 AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1574 AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1575 AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1576 } else {
1577 AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1578 AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1579 }
76741b0e
BC
1580 } else if(chroma422) {
1581 if (pixel_shift) {
1582 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1583 AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1584 } else {
1585 AV_COPY64(top_border+16, src_cb + 15*uvlinesize);
1586 AV_COPY64(top_border+24, src_cr + 15*uvlinesize);
1587 }
6e3ef511 1588 } else {
c90b9442
JGG
1589 if (pixel_shift) {
1590 AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1591 AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1592 } else {
1593 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1594 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1595 }
6e3ef511 1596 }
5f7f9719
MN
1597 }
1598 }
c988f975
MN
1599 }else if(MB_MBAFF){
1600 top_idx = 0;
1601 }else
1602 return;
5f7f9719
MN
1603 }
1604
0b69d625 1605 top_border = h->top_borders[top_idx][s->mb_x];
3b66c4c5 1606 // There are two lines saved, the line above the the top macroblock of a pair,
6867a90b 1607 // and the line above the bottom macroblock
0b69d625 1608 AV_COPY128(top_border, src_y + 16*linesize);
6e3ef511
OA
1609 if (pixel_shift)
1610 AV_COPY128(top_border+16, src_y+16*linesize+16);
53c05b1e 1611
49fb20cb 1612 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c90b9442
JGG
1613 if(chroma444){
1614 if (pixel_shift){
1615 AV_COPY128(top_border+32, src_cb + 16*linesize);
1616 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1617 AV_COPY128(top_border+64, src_cr + 16*linesize);
1618 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1619 } else {
1620 AV_COPY128(top_border+16, src_cb + 16*linesize);
1621 AV_COPY128(top_border+32, src_cr + 16*linesize);
1622 }
76741b0e
BC
1623 } else if(chroma422) {
1624 if (pixel_shift) {
1625 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1626 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1627 } else {
1628 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1629 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1630 }
6e3ef511 1631 } else {
c90b9442
JGG
1632 if (pixel_shift) {
1633 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1634 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1635 } else {
1636 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1637 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1638 }
6e3ef511 1639 }
53c05b1e
MN
1640 }
1641}
1642
bbdd52ed 1643static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
6e3ef511
OA
1644 uint8_t *src_cb, uint8_t *src_cr,
1645 int linesize, int uvlinesize,
c90b9442
JGG
1646 int xchg, int chroma444,
1647 int simple, int pixel_shift){
53c05b1e 1648 MpegEncContext * const s = &h->s;
4e987f82 1649 int deblock_topleft;
b69378e2 1650 int deblock_top;
5f7f9719 1651 int top_idx = 1;
1e4f1c56
AS
1652 uint8_t *top_border_m1;
1653 uint8_t *top_border;
5f7f9719
MN
1654
1655 if(!simple && FRAME_MBAFF){
1656 if(s->mb_y&1){
c988f975
MN
1657 if(!MB_MBAFF)
1658 return;
5f7f9719 1659 }else{
5f7f9719
MN
1660 top_idx = MB_MBAFF ? 0 : 1;
1661 }
5f7f9719 1662 }
b69378e2
1663
1664 if(h->deblocking_filter == 2) {
4e987f82
RB
1665 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1666 deblock_top = h->top_type;
b69378e2 1667 } else {
4e987f82
RB
1668 deblock_topleft = (s->mb_x > 0);
1669 deblock_top = (s->mb_y > !!MB_FIELD);
b69378e2 1670 }
53c05b1e 1671
6e3ef511
OA
1672 src_y -= linesize + 1 + pixel_shift;
1673 src_cb -= uvlinesize + 1 + pixel_shift;
1674 src_cr -= uvlinesize + 1 + pixel_shift;
53c05b1e 1675
1e4f1c56
AS
1676 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1677 top_border = h->top_borders[top_idx][s->mb_x];
1678
0b69d625 1679#define XCHG(a,b,xchg)\
6e3ef511
OA
1680 if (pixel_shift) {\
1681 if (xchg) {\
1682 AV_SWAP64(b+0,a+0);\
1683 AV_SWAP64(b+8,a+8);\
1684 } else {\
1685 AV_COPY128(b,a); \
1686 }\
1687 } else \
0b69d625
AS
1688if (xchg) AV_SWAP64(b,a);\
1689else AV_COPY64(b,a);
d89dc06a 1690
d89dc06a 1691 if(deblock_top){
4e987f82 1692 if(deblock_topleft){
6e3ef511 1693 XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
c988f975 1694 }
6e3ef511
OA
1695 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1696 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
cad4368a 1697 if(s->mb_x+1 < s->mb_width){
6e3ef511 1698 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
43efd19a 1699 }
53c05b1e 1700 }
49fb20cb 1701 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c90b9442 1702 if(chroma444){
4e987f82 1703 if(deblock_topleft){
c90b9442
JGG
1704 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1705 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1706 }
1707 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1708 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1709 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1710 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1711 if(s->mb_x+1 < s->mb_width){
1712 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1713 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1714 }
1715 } else {
1716 if(deblock_top){
1717 if(deblock_topleft){
1718 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1719 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1720 }
1721 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1722 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
c988f975 1723 }
53c05b1e 1724 }
53c05b1e
MN
1725 }
1726}
1727
6e3ef511
OA
1728static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1729 if (high_bit_depth) {
1730 return AV_RN32A(((int32_t*)mb) + index);
1731 } else
1732 return AV_RN16A(mb + index);
1733}
1734
1735static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1736 if (high_bit_depth) {
1737 AV_WN32A(((int32_t*)mb) + index, value);
1738 } else
1739 AV_WN16A(mb + index, value);
1740}
1741
c90b9442
JGG
1742static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1743 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1744{
1745 MpegEncContext * const s = &h->s;
1746 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1747 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1748 int i;
1749 int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1750 block_offset += 16*p;
1751 if(IS_INTRA4x4(mb_type)){
1752 if(simple || !s->encoding){
1753 if(IS_8x8DCT(mb_type)){
1754 if(transform_bypass){
1755 idct_dc_add =
1756 idct_add = s->dsp.add_pixels8;
1757 }else{
1758 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1759 idct_add = h->h264dsp.h264_idct8_add;
1760 }
1761 for(i=0; i<16; i+=4){
1762 uint8_t * const ptr= dest_y + block_offset[i];
1763 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1764 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1765 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1766 }else{
1767 const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1768 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1769 (h->topright_samples_available<<i)&0x4000, linesize);
1770 if(nnz){
1771 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1772 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1773 else
1774 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1775 }
1776 }
1777 }
1778 }else{
1779 if(transform_bypass){
1780 idct_dc_add =
1781 idct_add = s->dsp.add_pixels4;
1782 }else{
1783 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1784 idct_add = h->h264dsp.h264_idct_add;
1785 }
1786 for(i=0; i<16; i++){
1787 uint8_t * const ptr= dest_y + block_offset[i];
1788 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1789
1790 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1791 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1792 }else{
1793 uint8_t *topright;
1794 int nnz, tr;
1795 uint64_t tr_high;
1796 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1797 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
6371ce4b 1798 assert(s->mb_y || linesize <= block_offset[i]);
c90b9442
JGG
1799 if(!topright_avail){
1800 if (pixel_shift) {
1801 tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1802 topright= (uint8_t*) &tr_high;
1803 } else {
4d1418cd 1804 tr= ptr[3 - linesize]*0x01010101u;
c90b9442
JGG
1805 topright= (uint8_t*) &tr;
1806 }
1807 }else
1808 topright= ptr + (4 << pixel_shift) - linesize;
1809 }else
1810 topright= NULL;
1811
1812 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1813 nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1814 if(nnz){
1815 if(is_h264){
1816 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1817 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1818 else
1819 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
301fb921 1820 } else if (CONFIG_SVQ3_DECODER)
c90b9442
JGG
1821 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1822 }
1823 }
1824 }
1825 }
1826 }
1827 }else{
1828 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1829 if(is_h264){
1830 if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1831 if(!transform_bypass)
1832 h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1833 else{
1834 static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1835 8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1836 for(i = 0; i < 16; i++)
fdb5314e 1837 dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
c90b9442
JGG
1838 }
1839 }
301fb921 1840 } else if (CONFIG_SVQ3_DECODER)
c90b9442
JGG
1841 ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1842 }
1843}
1844
1845static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1846 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1847{
1848 MpegEncContext * const s = &h->s;
1849 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1850 int i;
1851 block_offset += 16*p;
1852 if(!IS_INTRA4x4(mb_type)){
1853 if(is_h264){
1854 if(IS_INTRA16x16(mb_type)){
1855 if(transform_bypass){
1856 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1857 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1858 }else{
1859 for(i=0; i<16; i++){
3b79f2e2 1860 if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
c90b9442
JGG
1861 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1862 }
1863 }
1864 }else{
1865 h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1866 }
1867 }else if(h->cbp&15){
1868 if(transform_bypass){
1869 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1870 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1871 for(i=0; i<16; i+=di){
1872 if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1873 idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1874 }
1875 }
1876 }else{
1877 if(IS_8x8DCT(mb_type)){
1878 h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1879 }else{
1880 h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1881 }
1882 }
1883 }
301fb921 1884 } else if (CONFIG_SVQ3_DECODER) {
c90b9442
JGG
1885 for(i=0; i<16; i++){
1886 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1887 uint8_t * const ptr= dest_y + block_offset[i];
1888 ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1889 }
1890 }
1891 }
1892 }
1893}
1894
05fb63f5
RB
1895static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1896{
0da71265
MN
1897 MpegEncContext * const s = &h->s;
1898 const int mb_x= s->mb_x;
1899 const int mb_y= s->mb_y;
64514ee8 1900 const int mb_xy= h->mb_xy;
657ccb5a 1901 const int mb_type = s->current_picture.f.mb_type[mb_xy];
0da71265
MN
1902 uint8_t *dest_y, *dest_cb, *dest_cr;
1903 int linesize, uvlinesize /*dct_offset*/;
c90b9442 1904 int i, j;
6867a90b 1905 int *block_offset = &h->block_offset[0];
41e4055b 1906 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
8b6871ed 1907 /* is_h264 should always be true if SVQ3 is disabled. */
49fb20cb 1908 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
36940eca 1909 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
76741b0e 1910 const int block_h = 16 >> s->chroma_y_shift;
05fb63f5 1911 const int chroma422 = CHROMA422;
0da71265 1912
657ccb5a 1913 dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
76741b0e
BC
1914 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1915 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
0da71265 1916
6e3ef511
OA
1917 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1918 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
a957c27b 1919
c988f975
MN
1920 h->list_counts[mb_xy]= h->list_count;
1921
bd91fee3 1922 if (!simple && MB_FIELD) {
5d18eaad
LM
1923 linesize = h->mb_linesize = s->linesize * 2;
1924 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
c90b9442 1925 block_offset = &h->block_offset[48];
1412060e 1926 if(mb_y&1){ //FIXME move out of this function?
0da71265 1927 dest_y -= s->linesize*15;
76741b0e
BC
1928 dest_cb-= s->uvlinesize * (block_h - 1);
1929 dest_cr-= s->uvlinesize * (block_h - 1);
0da71265 1930 }
5d18eaad
LM
1931 if(FRAME_MBAFF) {
1932 int list;
3425501d 1933 for(list=0; list<h->list_count; list++){
5d18eaad
LM
1934 if(!USES_LIST(mb_type, list))
1935 continue;
1936 if(IS_16X16(mb_type)){
1937 int8_t *ref = &h->ref_cache[list][scan8[0]];
1710856c 1938 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
5d18eaad
LM
1939 }else{
1940 for(i=0; i<16; i+=4){
5d18eaad
LM
1941 int ref = h->ref_cache[list][scan8[i]];
1942 if(ref >= 0)
1710856c 1943 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
5d18eaad
LM
1944 }
1945 }
1946 }
1947 }
0da71265 1948 } else {
5d18eaad
LM
1949 linesize = h->mb_linesize = s->linesize;
1950 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
0da71265
MN
1951// dct_offset = s->linesize * 16;
1952 }
115329f1 1953
bd91fee3 1954 if (!simple && IS_INTRA_PCM(mb_type)) {
6e3ef511
OA
1955 if (pixel_shift) {
1956 const int bit_depth = h->sps.bit_depth_luma;
1957 int j;
1958 GetBitContext gb;
1959 init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1960
1961 for (i = 0; i < 16; i++) {
1962 uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1963 for (j = 0; j < 16; j++)
1964 tmp_y[j] = get_bits(&gb, bit_depth);
1965 }
7b442ad9 1966 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6581e161 1967 if (!h->sps.chroma_format_idc) {
229d263c 1968 for (i = 0; i < block_h; i++) {
6581e161
JD
1969 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1970 for (j = 0; j < 8; j++) {
1971 tmp_cb[j] = 1 << (bit_depth - 1);
1972 }
1973 }
229d263c 1974 for (i = 0; i < block_h; i++) {
6581e161
JD
1975 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1976 for (j = 0; j < 8; j++) {
1977 tmp_cr[j] = 1 << (bit_depth - 1);
1978 }
1979 }
1980 } else {
76741b0e 1981 for (i = 0; i < block_h; i++) {
6581e161
JD
1982 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1983 for (j = 0; j < 8; j++)
1984 tmp_cb[j] = get_bits(&gb, bit_depth);
1985 }
76741b0e 1986 for (i = 0; i < block_h; i++) {
6581e161
JD
1987 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1988 for (j = 0; j < 8; j++)
1989 tmp_cr[j] = get_bits(&gb, bit_depth);
1990 }
7b442ad9 1991 }
c177cfb4 1992 }
504811ba 1993 } else {
7b442ad9
JGG
1994 for (i=0; i<16; i++) {
1995 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1996 }
1997 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6581e161 1998 if (!h->sps.chroma_format_idc) {
229d263c 1999 for (i = 0; i < block_h; i++) {
6581e161
JD
2000 memset(dest_cb + i*uvlinesize, 128, 8);
2001 memset(dest_cr + i*uvlinesize, 128, 8);
2002 }
2003 } else {
76741b0e 2004 for (i = 0; i < block_h; i++) {
6581e161
JD
2005 memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4, 8);
2006 memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4, 8);
2007 }
7b442ad9
JGG
2008 }
2009 }
6e3ef511 2010 }
e7e09b49
LLL
2011 } else {
2012 if(IS_INTRA(mb_type)){
5f7f9719 2013 if(h->deblocking_filter)
c90b9442 2014 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
53c05b1e 2015
49fb20cb 2016 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c92a30bb
KS
2017 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2018 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
e7e09b49 2019 }
0da71265 2020
c90b9442
JGG
2021 hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2022
5f7f9719 2023 if(h->deblocking_filter)
c90b9442 2024 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
bd91fee3 2025 }else if(is_h264){
05fb63f5
RB
2026 if (chroma422) {
2027 hl_motion_422(h, dest_y, dest_cb, dest_cr,
2028 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2029 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2030 h->h264dsp.weight_h264_pixels_tab,
2031 h->h264dsp.biweight_h264_pixels_tab,
2032 pixel_shift);
2033 } else {
2034 hl_motion_420(h, dest_y, dest_cb, dest_cr,
2035 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2036 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2037 h->h264dsp.weight_h264_pixels_tab,
2038 h->h264dsp.biweight_h264_pixels_tab,
2039 pixel_shift);
2040 }
0da71265 2041 }
e7e09b49 2042
c90b9442 2043 hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
0da71265 2044
49fb20cb 2045 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
ef9d1d15
LM
2046 uint8_t *dest[2] = {dest_cb, dest_cr};
2047 if(transform_bypass){
96465b90 2048 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
c90b9442
JGG
2049 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
2050 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
96465b90 2051 }else{
c25ac15a 2052 idct_add = s->dsp.add_pixels4;
c90b9442
JGG
2053 for(j=1; j<3; j++){
2054 for(i=j*16; i<j*16+4; i++){
2055 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2056 idct_add (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
2057 }
05fb63f5 2058 if (chroma422) {
229d263c 2059 for(i=j*16+4; i<j*16+8; i++){
640d5f1c 2060 if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
229d263c
RB
2061 idct_add (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2062 }
2063 }
96465b90
MN
2064 }
2065 }
ef9d1d15 2066 }else{
aebb5d6d 2067 if(is_h264){
76741b0e 2068 int qp[2];
05fb63f5 2069 if (chroma422) {
76741b0e
BC
2070 qp[0] = h->chroma_qp[0] + 3;
2071 qp[1] = h->chroma_qp[1] + 3;
2072 } else {
2073 qp[0] = h->chroma_qp[0];
2074 qp[1] = h->chroma_qp[1];
2075 }
2e186601 2076 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
76741b0e 2077 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][qp[0]][0]);
2e186601 2078 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
76741b0e 2079 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][qp[1]][0]);
1d16a1cf
RB
2080 h->h264dsp.h264_idct_add8(dest, block_offset,
2081 h->mb, uvlinesize,
2082 h->non_zero_count_cache);
301fb921 2083 } else if (CONFIG_SVQ3_DECODER) {
c90b9442
JGG
2084 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*1, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
2085 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*2, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
2086 for(j=1; j<3; j++){
2087 for(i=j*16; i<j*16+4; i++){
2088 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2089 uint8_t * const ptr= dest[j-1] + block_offset[i];
2090 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2091 }
aebb5d6d 2092 }
e7e09b49 2093 }
4704097a 2094 }
0da71265
MN
2095 }
2096 }
2097 }
c212fb0c 2098 if(h->cbp || IS_INTRA(mb_type))
c90b9442 2099 {
c9c49387 2100 s->dsp.clear_blocks(h->mb);
c90b9442
JGG
2101 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2102 }
2103}
2104
2105static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2106 MpegEncContext * const s = &h->s;
2107 const int mb_x= s->mb_x;
2108 const int mb_y= s->mb_y;
2109 const int mb_xy= h->mb_xy;
657ccb5a 2110 const int mb_type = s->current_picture.f.mb_type[mb_xy];
c90b9442
JGG
2111 uint8_t *dest[3];
2112 int linesize;
2113 int i, j, p;
2114 int *block_offset = &h->block_offset[0];
2115 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
7b442ad9 2116 const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
c90b9442 2117
7b442ad9 2118 for (p = 0; p < plane_count; p++)
c90b9442 2119 {
657ccb5a 2120 dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
c90b9442
JGG
2121 s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2122 }
2123
2124 h->list_counts[mb_xy]= h->list_count;
2125
2126 if (!simple && MB_FIELD) {
2127 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2128 block_offset = &h->block_offset[48];
2129 if(mb_y&1) //FIXME move out of this function?
2130 for (p = 0; p < 3; p++)
2131 dest[p] -= s->linesize*15;
2132 if(FRAME_MBAFF) {
2133 int list;
2134 for(list=0; list<h->list_count; list++){
2135 if(!USES_LIST(mb_type, list))
2136 continue;
2137 if(IS_16X16(mb_type)){
2138 int8_t *ref = &h->ref_cache[list][scan8[0]];
2139 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2140 }else{
2141 for(i=0; i<16; i+=4){
2142 int ref = h->ref_cache[list][scan8[i]];
2143 if(ref >= 0)
2144 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2145 }
2146 }
2147 }
2148 }
2149 } else {
2150 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2151 }
2152
2153 if (!simple && IS_INTRA_PCM(mb_type)) {
2154 if (pixel_shift) {
2155 const int bit_depth = h->sps.bit_depth_luma;
2156 GetBitContext gb;
2157 init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2158
7b442ad9 2159 for (p = 0; p < plane_count; p++) {
c90b9442
JGG
2160 for (i = 0; i < 16; i++) {
2161 uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2162 for (j = 0; j < 16; j++)
2163 tmp[j] = get_bits(&gb, bit_depth);
2164 }
2165 }
2166 } else {
7b442ad9 2167 for (p = 0; p < plane_count; p++) {
c90b9442
JGG
2168 for (i = 0; i < 16; i++) {
2169 memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2170 }
2171 }
2172 }
2173 } else {
2174 if(IS_INTRA(mb_type)){
2175 if(h->deblocking_filter)
2176 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2177
7b442ad9 2178 for (p = 0; p < plane_count; p++)
c90b9442
JGG
2179 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2180
2181 if(h->deblocking_filter)
2182 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2183 }else{
298e52c9
JGG
2184 hl_motion(h, dest[0], dest[1], dest[2],
2185 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2186 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2187 h->h264dsp.weight_h264_pixels_tab,
05fb63f5 2188 h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
c90b9442
JGG
2189 }
2190
7b442ad9 2191 for (p = 0; p < plane_count; p++)
c90b9442
JGG
2192 hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2193 }
2194 if(h->cbp || IS_INTRA(mb_type))
2195 {
2196 s->dsp.clear_blocks(h->mb);
2197 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2198 }
0da71265
MN
2199}
2200
0da71265 2201/**
bd91fee3
AS
2202 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2203 */
6e3ef511
OA
2204#define hl_decode_mb_simple(sh, bits) \
2205static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2206 hl_decode_mb_internal(h, 1, sh); \
bd91fee3 2207}
373211d8
MR
2208hl_decode_mb_simple(0, 8)
2209hl_decode_mb_simple(1, 16)
bd91fee3
AS
2210
2211/**
2212 * Process a macroblock; this handles edge cases, such as interlacing.
2213 */
2214static void av_noinline hl_decode_mb_complex(H264Context *h){
6e3ef511 2215 hl_decode_mb_internal(h, 0, h->pixel_shift);
bd91fee3
AS
2216}
2217
c90b9442
JGG
2218static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2219 hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2220}
2221
2222static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2223 hl_decode_mb_444_internal(h, 1, 0);
2224}
2225
903d58f6 2226void ff_h264_hl_decode_mb(H264Context *h){
bd91fee3 2227 MpegEncContext * const s = &h->s;
64514ee8 2228 const int mb_xy= h->mb_xy;
657ccb5a 2229 const int mb_type = s->current_picture.f.mb_type[mb_xy];
49fb20cb 2230 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
bd91fee3 2231
c90b9442
JGG
2232 if (CHROMA444) {
2233 if(is_complex || h->pixel_shift)
2234 hl_decode_mb_444_complex(h);
2235 else
2236 hl_decode_mb_444_simple(h);
2237 } else if (is_complex) {
bd91fee3 2238 hl_decode_mb_complex(h);
6e3ef511
OA
2239 } else if (h->pixel_shift) {
2240 hl_decode_mb_simple_16(h);
2241 } else
2242 hl_decode_mb_simple_8(h);
bd91fee3
AS
2243}
2244
0da71265
MN
2245static int pred_weight_table(H264Context *h){
2246 MpegEncContext * const s = &h->s;
2247 int list, i;
9f2d1b4f 2248 int luma_def, chroma_def;
115329f1 2249
9f2d1b4f
LM
2250 h->use_weight= 0;
2251 h->use_weight_chroma= 0;
0da71265 2252 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
c90b9442 2253 if(h->sps.chroma_format_idc)
09fffe9b 2254 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
9f2d1b4f
LM
2255 luma_def = 1<<h->luma_log2_weight_denom;
2256 chroma_def = 1<<h->chroma_log2_weight_denom;
0da71265
MN
2257
2258 for(list=0; list<2; list++){
cb99c652
GB
2259 h->luma_weight_flag[list] = 0;
2260 h->chroma_weight_flag[list] = 0;
0da71265
MN
2261 for(i=0; i<h->ref_count[list]; i++){
2262 int luma_weight_flag, chroma_weight_flag;
115329f1 2263
0da71265
MN
2264 luma_weight_flag= get_bits1(&s->gb);
2265 if(luma_weight_flag){
3d9137c8
MN
2266 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2267 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2268 if( h->luma_weight[i][list][0] != luma_def
2269 || h->luma_weight[i][list][1] != 0) {
9f2d1b4f 2270 h->use_weight= 1;
cb99c652
GB
2271 h->luma_weight_flag[list]= 1;
2272 }
9f2d1b4f 2273 }else{
3d9137c8
MN
2274 h->luma_weight[i][list][0]= luma_def;
2275 h->luma_weight[i][list][1]= 0;
0da71265
MN
2276 }
2277
c90b9442 2278 if(h->sps.chroma_format_idc){
fef744d4
MN
2279 chroma_weight_flag= get_bits1(&s->gb);
2280 if(chroma_weight_flag){
2281 int j;
2282 for(j=0; j<2; j++){
3d9137c8
MN
2283 h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2284 h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2285 if( h->chroma_weight[i][list][j][0] != chroma_def
2286 || h->chroma_weight[i][list][j][1] != 0) {
fef744d4 2287 h->use_weight_chroma= 1;
cb99c652
GB
2288 h->chroma_weight_flag[list]= 1;
2289 }
fef744d4
MN
2290 }
2291 }else{
2292 int j;
2293 for(j=0; j<2; j++){
3d9137c8
MN
2294 h->chroma_weight[i][list][j][0]= chroma_def;
2295 h->chroma_weight[i][list][j][1]= 0;
fef744d4 2296 }
0da71265
MN
2297 }
2298 }
2299 }
975a1447 2300 if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
0da71265 2301 }
9f2d1b4f 2302 h->use_weight= h->use_weight || h->use_weight_chroma;
0da71265
MN
2303 return 0;
2304}
2305
1052b76f
MN
2306/**
2307 * Initialize implicit_weight table.
6da88bd3 2308 * @param field 0/1 initialize the weight for interlaced MBAFF
1052b76f
MN
2309 * -1 initializes the rest
2310 */
2311static void implicit_weight_table(H264Context *h, int field){
9f2d1b4f 2312 MpegEncContext * const s = &h->s;
1052b76f 2313 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
9f2d1b4f 2314
ce09f927
GB
2315 for (i = 0; i < 2; i++) {
2316 h->luma_weight_flag[i] = 0;
2317 h->chroma_weight_flag[i] = 0;
2318 }
2319
1052b76f 2320 if(field < 0){
4418aa9c
RB
2321 if (s->picture_structure == PICT_FRAME) {
2322 cur_poc = s->current_picture_ptr->poc;
2323 } else {
2324 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2325 }
1052b76f 2326 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
9f2d1b4f
LM
2327 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2328 h->use_weight= 0;
2329 h->use_weight_chroma= 0;
2330 return;
2331 }
1052b76f
MN
2332 ref_start= 0;
2333 ref_count0= h->ref_count[0];
2334 ref_count1= h->ref_count[1];
2335 }else{
2336 cur_poc = s->current_picture_ptr->field_poc[field];
2337 ref_start= 16;
2338 ref_count0= 16+2*h->ref_count[0];
2339 ref_count1= 16+2*h->ref_count[1];
2340 }
9f2d1b4f
LM
2341
2342 h->use_weight= 2;
2343 h->use_weight_chroma= 2;
2344 h->luma_log2_weight_denom= 5;
2345 h->chroma_log2_weight_denom= 5;
2346
1052b76f 2347 for(ref0=ref_start; ref0 < ref_count0; ref0++){
9f2d1b4f 2348 int poc0 = h->ref_list[0][ref0].poc;
1052b76f 2349 for(ref1=ref_start; ref1 < ref_count1; ref1++){
87cf70eb
JD
2350 int w = 32;
2351 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2352 int poc1 = h->ref_list[1][ref1].poc;
2353 int td = av_clip(poc1 - poc0, -128, 127);
2354 if(td){
2355 int tb = av_clip(cur_poc - poc0, -128, 127);
2356 int tx = (16384 + (FFABS(td) >> 1)) / td;
2357 int dist_scale_factor = (tb*tx + 32) >> 8;
2358 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2359 w = 64 - dist_scale_factor;
2360 }
1052b76f
MN
2361 }
2362 if(field<0){
2363 h->implicit_weight[ref0][ref1][0]=
2364 h->implicit_weight[ref0][ref1][1]= w;
2365 }else{
2366 h->implicit_weight[ref0][ref1][field]=w;
72f86ec0 2367 }
9f2d1b4f
LM
2368 }
2369 }
2370}
2371
8fd57a66 2372/**
5175b937 2373 * instantaneous decoder refresh.
0da71265
MN
2374 */
2375static void idr(H264Context *h){
ea6f00c4 2376 ff_h264_remove_all_refs(h);
a149c1a5 2377 h->prev_frame_num= 0;
80f8e035
MN
2378 h->prev_frame_num_offset= 0;
2379 h->prev_poc_msb=
2380 h->prev_poc_lsb= 0;
0da71265
MN
2381}
2382
7c33ad19
LM
2383/* forget old pics after a seek */
2384static void flush_dpb(AVCodecContext *avctx){
2385 H264Context *h= avctx->priv_data;
2386 int i;
64b9d48f 2387 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
285b570f 2388 if(h->delayed_pic[i])
657ccb5a 2389 h->delayed_pic[i]->f.reference = 0;
7c33ad19 2390 h->delayed_pic[i]= NULL;
285b570f 2391 }
adedd840
RB
2392 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2393 h->last_pocs[i] = INT_MIN;
6a9c8594 2394 h->outputed_poc=h->next_outputed_poc= INT_MIN;
b19d493f 2395 h->prev_interlaced_frame = 1;
7c33ad19 2396 idr(h);
ca159196 2397 if(h->s.current_picture_ptr)
657ccb5a 2398 h->s.current_picture_ptr->f.reference = 0;
12d96de3 2399 h->s.first_field= 0;
9c095463 2400 ff_h264_reset_sei(h);
e240f898 2401 ff_mpeg_flush(avctx);
7c33ad19
LM
2402}
2403
0da71265
MN
2404static int init_poc(H264Context *h){
2405 MpegEncContext * const s = &h->s;
2406 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2407 int field_poc[2];
357282c6 2408 Picture *cur = s->current_picture_ptr;
0da71265 2409
b78a6baa 2410 h->frame_num_offset= h->prev_frame_num_offset;
5710b371 2411 if(h->frame_num < h->prev_frame_num)
b78a6baa 2412 h->frame_num_offset += max_frame_num;
0da71265
MN
2413
2414 if(h->sps.poc_type==0){
2415 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2416
2417 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2418 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2419 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2420 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2421 else
2422 h->poc_msb = h->prev_poc_msb;
2423//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
115329f1 2424 field_poc[0] =
0da71265 2425 field_poc[1] = h->poc_msb + h->poc_lsb;
115329f1 2426 if(s->picture_structure == PICT_FRAME)
0da71265
MN
2427 field_poc[1] += h->delta_poc_bottom;
2428 }else if(h->sps.poc_type==1){
2429 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2430 int i;
2431
2432 if(h->sps.poc_cycle_length != 0)
2433 abs_frame_num = h->frame_num_offset + h->frame_num;
2434 else
2435 abs_frame_num = 0;
2436
2437 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2438 abs_frame_num--;
115329f1 2439
0da71265
MN
2440 expected_delta_per_poc_cycle = 0;
2441 for(i=0; i < h->sps.poc_cycle_length; i++)
2442 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2443
2444 if(abs_frame_num > 0){
2445 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2446 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2447
2448 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2449 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2450 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2451 } else
2452 expectedpoc = 0;
2453
115329f1 2454 if(h->nal_ref_idc == 0)
0da71265 2455 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
115329f1 2456
0da71265
MN
2457 field_poc[0] = expectedpoc + h->delta_poc[0];
2458 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2459
2460 if(s->picture_structure == PICT_FRAME)
2461 field_poc[1] += h->delta_poc[1];
2462 }else{
b78a6baa 2463 int poc= 2*(h->frame_num_offset + h->frame_num);
5710b371 2464
b78a6baa
MN
2465 if(!h->nal_ref_idc)
2466 poc--;
5710b371 2467
0da71265
MN
2468 field_poc[0]= poc;
2469 field_poc[1]= poc;
2470 }
115329f1 2471
357282c6 2472 if(s->picture_structure != PICT_BOTTOM_FIELD)
0da71265 2473 s->current_picture_ptr->field_poc[0]= field_poc[0];
357282c6 2474 if(s->picture_structure != PICT_TOP_FIELD)
0da71265 2475 s->current_picture_ptr->field_poc[1]= field_poc[1];
357282c6 2476 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
0da71265
MN
2477
2478 return 0;
2479}
2480
b41c1db3
2481
2482/**
2483 * initialize scan tables
2484 */
2485static void init_scan_tables(H264Context *h){
b41c1db3 2486 int i;
ca32f7f2 2487 for(i=0; i<16; i++){
b41c1db3 2488#define T(x) (x>>2) | ((x<<2) & 0xF)
ca32f7f2
JGG
2489 h->zigzag_scan[i] = T(zigzag_scan[i]);
2490 h-> field_scan[i] = T( field_scan[i]);
b41c1db3 2491#undef T
b41c1db3 2492 }
ca32f7f2 2493 for(i=0; i<64; i++){
b41c1db3 2494#define T(x) (x>>3) | ((x&7)<<3)
ca32f7f2
JGG
2495 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2496 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2497 h->field_scan8x8[i] = T(field_scan8x8[i]);
2498 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
b41c1db3 2499#undef T
b41c1db3
2500 }
2501 if(h->sps.transform_bypass){ //FIXME same ugly
2502 h->zigzag_scan_q0 = zigzag_scan;
45beb850 2503 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
b41c1db3
2504 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2505 h->field_scan_q0 = field_scan;
2506 h->field_scan8x8_q0 = field_scan8x8;
2507 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
2508 }else{
2509 h->zigzag_scan_q0 = h->zigzag_scan;
2510 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
2511 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2512 h->field_scan_q0 = h->field_scan;
2513 h->field_scan8x8_q0 = h->field_scan8x8;
2514 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
2515 }
2516}
afebe2f7 2517
12fe7594 2518static int field_end(H264Context *h, int in_setup){
256299d3
MN
2519 MpegEncContext * const s = &h->s;
2520 AVCodecContext * const avctx= s->avctx;
12fe7594 2521 int err = 0;
256299d3
MN
2522 s->mb_y= 0;
2523
6a9c8594 2524 if (!in_setup && !s->dropable)
47c0ac96
DB
2525 ff_thread_report_progress(&s->current_picture_ptr->f,
2526 (16 * s->mb_height >> FIELD_PICTURE) - 1,
2527 s->picture_structure == PICT_BOTTOM_FIELD);
256299d3
MN
2528
2529 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2530 ff_vdpau_h264_set_reference_frames(s);
2531
6a9c8594
AS
2532 if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2533 if(!s->dropable) {
12fe7594 2534 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6a9c8594
AS
2535 h->prev_poc_msb= h->poc_msb;
2536 h->prev_poc_lsb= h->poc_lsb;
2537 }
2538 h->prev_frame_num_offset= h->frame_num_offset;
2539 h->prev_frame_num= h->frame_num;
2540 h->outputed_poc = h->next_outputed_poc;
256299d3 2541 }
256299d3
MN
2542
2543 if (avctx->hwaccel) {
2544 if (avctx->hwaccel->end_frame(avctx) < 0)
2545 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2546 }
2547
2548 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2549 ff_vdpau_h264_picture_complete(s);
2550
2551 /*
2552 * FIXME: Error handling code does not seem to support interlaced
2553 * when slices span multiple rows
2554 * The ff_er_add_slice calls don't work right for bottom
2555 * fields; they cause massive erroneous error concealing
2556 * Error marking covers both fields (top and bottom).
2557 * This causes a mismatched s->error_count
2558 * and a bad error table. Further, the error count goes to
2559 * INT_MAX when called for bottom field, because mb_y is
2560 * past end by one (callers fault) and resync_mb_y != 0
2561 * causes problems for the first MB line, too.
2562 */
2563 if (!FIELD_PICTURE)
2564 ff_er_frame_end(s);
2565
efd29844 2566 ff_MPV_frame_end(s);
d225a1e2
MN
2567
2568 h->current_slice=0;
12fe7594
DB
2569
2570 return err;
256299d3
MN
2571}
2572
afebe2f7 2573/**
49bd8e4b 2574 * Replicate H264 "master" context to thread contexts.
afebe2f7
2575 */
2576static void clone_slice(H264Context *dst, H264Context *src)
2577{
2578 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2579 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2580 dst->s.current_picture = src->s.current_picture;
2581 dst->s.linesize = src->s.linesize;
2582 dst->s.uvlinesize = src->s.uvlinesize;
12d96de3 2583 dst->s.first_field = src->s.first_field;
afebe2f7
2584
2585 dst->prev_poc_msb = src->prev_poc_msb;
2586 dst->prev_poc_lsb = src->prev_poc_lsb;
2587 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2588 dst->prev_frame_num = src->prev_frame_num;
2589 dst->short_ref_count = src->short_ref_count;
2590
2591 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2592 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2593 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2594 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
50c21814
2595
2596 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2597 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
afebe2f7
2598}
2599
0da71265 2600/**
58c42af7 2601 * Compute profile from profile_idc and constraint_set?_flags.
fe9a3fbe
JG
2602 *
2603 * @param sps SPS
2604 *
2605 * @return profile as defined by FF_PROFILE_H264_*
2606 */
2607int ff_h264_get_profile(SPS *sps)
2608{
2609 int profile = sps->profile_idc;
2610
2611 switch(sps->profile_idc) {
2612 case FF_PROFILE_H264_BASELINE:
2613 // constraint_set1_flag set to 1
2614 profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2615 break;
2616 case FF_PROFILE_H264_HIGH_10:
2617 case FF_PROFILE_H264_HIGH_422:
2618 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2619 // constraint_set3_flag set to 1
2620 profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2621 break;
2622 }
2623
2624 return profile;
2625}
2626
2627/**
58c42af7 2628 * Decode a slice header.
efd29844 2629 * This will also call ff_MPV_common_init() and frame_start() as needed.
afebe2f7
2630 *
2631 * @param h h264context
2632 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2633 *
d9526386 2634 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
0da71265 2635 */
afebe2f7 2636static int decode_slice_header(H264Context *h, H264Context *h0){
0da71265 2637 MpegEncContext * const s = &h->s;
12d96de3 2638 MpegEncContext * const s0 = &h0->s;
88e7a4d1 2639 unsigned int first_mb_in_slice;
ac658be5 2640 unsigned int pps_id;
0da71265 2641 int num_ref_idx_active_override_flag;
41f5c62f 2642 unsigned int slice_type, tmp, i, j;
0bf79634 2643 int default_ref_list_done = 0;
12d96de3 2644 int last_pic_structure;
0da71265 2645
2f944356 2646 s->dropable= h->nal_ref_idc == 0;
0da71265 2647
9a0dda8b
JGG
2648 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2649 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
cf653d08
JD
2650 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2651 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2652 }else{
2653 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2654 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2655 }
2656
0da71265
MN
2657 first_mb_in_slice= get_ue_golomb(&s->gb);
2658
d225a1e2
MN
2659 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2660 if(h0->current_slice && FIELD_PICTURE){
6a9c8594 2661 field_end(h, 1);
d225a1e2
MN
2662 }
2663
afebe2f7 2664 h0->current_slice = 0;
12d96de3 2665 if (!s0->first_field)
f6e3c460 2666 s->current_picture_ptr= NULL;
66a4b2c1
MN
2667 }
2668
9963b332 2669 slice_type= get_ue_golomb_31(&s->gb);
0bf79634 2670 if(slice_type > 9){
9b879566 2671 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
5175b937 2672 return -1;
0da71265 2673 }
0bf79634
LLL
2674 if(slice_type > 4){
2675 slice_type -= 5;
0da71265
MN
2676 h->slice_type_fixed=1;
2677 }else
2678 h->slice_type_fixed=0;
115329f1 2679
ee2a957f 2680 slice_type= golomb_to_pict_type[ slice_type ];
975a1447 2681 if (slice_type == AV_PICTURE_TYPE_I
afebe2f7 2682 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
0bf79634
LLL
2683 default_ref_list_done = 1;
2684 }
2685 h->slice_type= slice_type;
e3e6f18f 2686 h->slice_type_nos= slice_type & 3;
0bf79634 2687
1412060e 2688 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
115329f1 2689
0da71265 2690 pps_id= get_ue_golomb(&s->gb);
ac658be5 2691 if(pps_id>=MAX_PPS_COUNT){
9b879566 2692 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
0da71265
MN
2693 return -1;
2694 }
afebe2f7 2695 if(!h0->pps_buffers[pps_id]) {
a0f80050 2696 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
8b92b792
MN
2697 return -1;
2698 }
afebe2f7 2699 h->pps= *h0->pps_buffers[pps_id];
8b92b792 2700
afebe2f7 2701 if(!h0->sps_buffers[h->pps.sps_id]) {
a0f80050 2702 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
8b92b792
MN
2703 return -1;
2704 }
afebe2f7 2705 h->sps = *h0->sps_buffers[h->pps.sps_id];
239ea04c 2706
fe9a3fbe 2707 s->avctx->profile = ff_h264_get_profile(&h->sps);
fa37cf0d 2708 s->avctx->level = h->sps.level_idc;
6752a3cc 2709 s->avctx->refs = h->sps.ref_frame_count;
b08e38e8 2710
0da71265 2711 s->mb_width= h->sps.mb_width;
6867a90b 2712 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
115329f1 2713
bf4665ee 2714 h->b_stride= s->mb_width*4;
0da71265 2715
76741b0e
BC
2716 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2717
c90b9442 2718 s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
0da71265 2719 if(h->sps.frame_mbs_only_flag)
76741b0e 2720 s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
0da71265 2721 else
76741b0e 2722 s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
115329f1
DB
2723
2724 if (s->context_initialized
5388f0b4
JK
2725 && ( s->width != s->avctx->width || s->height != s->avctx->height
2726 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
6a9c8594
AS
2727 if(h != h0) {
2728 av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
afebe2f7 2729 return -1; // width / height changed during parallelized decoding
6a9c8594 2730 }
91078926 2731 free_tables(h, 0);
ff7f75e1 2732 flush_dpb(s->avctx);
efd29844 2733 ff_MPV_common_end(s);
0da71265
MN
2734 }
2735 if (!s->context_initialized) {
33aec3f4
MN
2736 if (h != h0) {
2737 av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2738 return -1;
2739 }
f3bdc3da
RD
2740
2741 avcodec_set_dimensions(s->avctx, s->width, s->height);
2742 s->avctx->sample_aspect_ratio= h->sps.sar;
cfa5a81e 2743 av_assert0(s->avctx->sample_aspect_ratio.den);
f3bdc3da 2744
c4dffe7e
DC
2745 if(h->sps.video_signal_type_present_flag){
2746 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2747 if(h->sps.colour_description_present_flag){
2748 s->avctx->color_primaries = h->sps.color_primaries;
2749 s->avctx->color_trc = h->sps.color_trc;
2750 s->avctx->colorspace = h->sps.colorspace;
2751 }
2752 }
2753
f3bdc3da 2754 if(h->sps.timing_info_present_flag){
3102d180 2755 int64_t den= h->sps.time_scale;
055a6aa7 2756 if(h->x264_build < 44U)
3102d180 2757 den *= 2;
f3bdc3da 2758 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3102d180 2759 h->sps.num_units_in_tick, den, 1<<30);
f3bdc3da 2760 }
d545cf80
OA
2761
2762 switch (h->sps.bit_depth_luma) {
2763 case 9 :
db431f7e
RB
2764 if (CHROMA444) {
2765 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2766 s->avctx->pix_fmt = PIX_FMT_GBRP9;
2767 } else
2768 s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2769 } else if (CHROMA422)
76741b0e
BC
2770 s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2771 else
2772 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
d545cf80
OA
2773 break;
2774 case 10 :
db431f7e
RB
2775 if (CHROMA444) {
2776 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2777 s->avctx->pix_fmt = PIX_FMT_GBRP10;
2778 } else
2779 s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2780 } else if (CHROMA422)
76741b0e
BC
2781 s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2782 else
2783 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
d545cf80 2784 break;
0ce4fe48 2785 case 8:
c90b9442 2786 if (CHROMA444){
db431f7e
RB
2787 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2788 s->avctx->pix_fmt = PIX_FMT_GBRP;
2789 } else
2790 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
76741b0e
BC
2791 } else if (CHROMA422) {
2792 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
c90b9442
JGG
2793 }else{
2794 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2795 s->avctx->codec->pix_fmts ?
2796 s->avctx->codec->pix_fmts :
2797 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2798 hwaccel_pixfmt_list_h264_jpeg_420 :
2799 ff_hwaccel_pixfmt_list_420);
2800 }
0ce4fe48
RB
2801 break;
2802 default:
2803 av_log(s->avctx, AV_LOG_ERROR,
2804 "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2805 return AVERROR_INVALIDDATA;
d545cf80
OA
2806 }
2807
f3bdc3da
RD
2808 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2809
efd29844
MS
2810 if (ff_MPV_common_init(s) < 0) {
2811 av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
0da71265 2812 return -1;
33aec3f4 2813 }
12d96de3 2814 s->first_field = 0;
b19d493f 2815 h->prev_interlaced_frame = 1;
115329f1 2816
b41c1db3 2817 init_scan_tables(h);
bac3ab13
DB
2818 if (ff_h264_alloc_tables(h) < 0) {
2819 av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2820 return AVERROR(ENOMEM);
2821 }
0da71265 2822
6a9c8594 2823 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
33aec3f4
MN
2824 if (context_init(h) < 0) {
2825 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
afebe2f7 2826 return -1;
33aec3f4 2827 }
6a9c8594 2828 } else {
881a5e04 2829 for(i = 1; i < s->slice_context_count; i++) {
6a9c8594
AS
2830 H264Context *c;
2831 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2832 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2833 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2834 c->h264dsp = h->h264dsp;
2835 c->sps = h->sps;
2836 c->pps = h->pps;
2837 c->pixel_shift = h->pixel_shift;
2838 init_scan_tables(c);
2839 clone_tables(c, h, i);
2840 }
2841
881a5e04 2842 for(i = 0; i < s->slice_context_count; i++)
33aec3f4
MN
2843 if (context_init(h->thread_context[i]) < 0) {
2844 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
6a9c8594 2845 return -1;
33aec3f4 2846 }
6a9c8594 2847 }
0da71265
MN
2848 }
2849
0ce4fe48
RB
2850 if(h == h0 && h->dequant_coeff_pps != pps_id){
2851 h->dequant_coeff_pps = pps_id;
2852 init_dequant_tables(h);
2853 }
2854
0da71265
MN
2855 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2856
5d18eaad 2857 h->mb_mbaff = 0;
6ba71fc4 2858 h->mb_aff_frame = 0;
12d96de3 2859 last_pic_structure = s0->picture_structure;
0da71265
MN
2860 if(h->sps.frame_mbs_only_flag){
2861 s->picture_structure= PICT_FRAME;
2862 }else{
6ba71fc4 2863 if(get_bits1(&s->gb)) { //field_pic_flag
0da71265 2864 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
6ba71fc4 2865 } else {
0da71265 2866 s->picture_structure= PICT_FRAME;
6ba71fc4 2867 h->mb_aff_frame = h->sps.mb_aff;
6867a90b 2868 }
0da71265 2869 }
44e9dcf1 2870 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2ddcf84b
JD
2871
2872 if(h0->current_slice == 0){
3803af22
AS
2873 // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2874 if(h->frame_num != h->prev_frame_num) {
2875 int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2876
2877 if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2878
2879 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2880 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2881 if (unwrap_prev_frame_num < 0)
2882 unwrap_prev_frame_num += max_frame_num;
2883
2884 h->prev_frame_num = unwrap_prev_frame_num;
2885 }
2886 }
6a9c8594 2887
26b86e47
MN
2888 while(h->frame_num != h->prev_frame_num &&
2889 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
4dece8c7 2890 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
bb943bb8 2891 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
903d58f6 2892 if (ff_h264_frame_start(h) < 0)
66e6038c 2893 return -1;
26b86e47
MN
2894 h->prev_frame_num++;
2895 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2896 s->current_picture_ptr->frame_num= h->prev_frame_num;
47c0ac96
DB
2897 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
2898 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 1);
3d542120 2899 ff_generate_sliding_window_mmcos(h);
12fe7594 2900 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
9abc9873 2901 (s->avctx->err_recognition & AV_EF_EXPLODE))
12fe7594 2902 return AVERROR_INVALIDDATA;
e2983d6e
JGG
2903 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2904 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2905 * about there being no actual duplicates.
2906 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
da9cea77 2907 * concealing a lost frame, this probably isn't noticeable by comparison, but it should
e2983d6e 2908 * be fixed. */
4dece8c7
JGG
2909 if (h->short_ref_count) {
2910 if (prev) {
657ccb5a
DB
2911 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2912 (const uint8_t**)prev->f.data, prev->f.linesize,
5dd7f994 2913 s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
4dece8c7
JGG
2914 h->short_ref[0]->poc = prev->poc+2;
2915 }
2916 h->short_ref[0]->frame_num = h->prev_frame_num;
2917 }
26b86e47
MN
2918 }
2919
12d96de3
JD
2920 /* See if we have a decoded first field looking for a pair... */
2921 if (s0->first_field) {
2922 assert(s0->current_picture_ptr);
657ccb5a 2923 assert(s0->current_picture_ptr->f.data[0]);
95a06eb4 2924 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
12d96de3
JD
2925
2926 /* figure out if we have a complementary field pair */
2927 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2928 /*
2929 * Previous field is unmatched. Don't display it, but let it
2930 * remain for reference if marked as such.
2931 */
2932 s0->current_picture_ptr = NULL;
2933 s0->first_field = FIELD_PICTURE;
2934
2935 } else {
2936 if (h->nal_ref_idc &&
657ccb5a 2937 s0->current_picture_ptr->f.reference &&
12d96de3
JD
2938 s0->current_picture_ptr->frame_num != h->frame_num) {
2939 /*
2940 * This and previous field were reference, but had
2941 * different frame_nums. Consider this field first in
2942 * pair. Throw away previous field except for reference
2943 * purposes.
2944 */
2945 s0->first_field = 1;
2946 s0->current_picture_ptr = NULL;
2947
2948 } else {
2949 /* Second field in complementary pair */
2950 s0->first_field = 0;
2951 }
2952 }
2953
2954 } else {
2955 /* Frame or first field in a potentially complementary pair */
2956 assert(!s0->current_picture_ptr);
2957 s0->first_field = FIELD_PICTURE;
2958 }
2959
6a9c8594
AS
2960 if(!FIELD_PICTURE || s0->first_field) {
2961 if (ff_h264_frame_start(h) < 0) {
2962 s0->first_field = 0;
2963 return -1;
2964 }
2965 } else {
2966 ff_release_unused_pictures(s, 0);
12d96de3 2967 }
2ddcf84b
JD
2968 }
2969 if(h != h0)
2970 clone_slice(h, h0);
2971
2972 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2973
88e7a4d1 2974 assert(s->mb_num == s->mb_width * s->mb_height);
f3e53d9f 2975 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
88e7a4d1
MN
2976 first_mb_in_slice >= s->mb_num){
2977 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
6b53b87e
MN
2978 return -1;
2979 }
88e7a4d1 2980 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
f3e53d9f
JD
2981 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2982 if (s->picture_structure == PICT_BOTTOM_FIELD)
2983 s->resync_mb_y = s->mb_y = s->mb_y + 1;
88e7a4d1 2984 assert(s->mb_y < s->mb_height);
115329f1 2985
0da71265
MN
2986 if(s->picture_structure==PICT_FRAME){
2987 h->curr_pic_num= h->frame_num;
2988 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2989 }else{
f57e2af6 2990 h->curr_pic_num= 2*h->frame_num + 1;
0da71265
MN
2991 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2992 }
115329f1 2993
0da71265 2994 if(h->nal_unit_type == NAL_IDR_SLICE){
1df1df0b 2995 get_ue_golomb(&s->gb); /* idr_pic_id */
0da71265 2996 }
115329f1 2997
0da71265
MN
2998 if(h->sps.poc_type==0){
2999 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
115329f1 3000
0da71265
MN
3001 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3002 h->delta_poc_bottom= get_se_golomb(&s->gb);
3003 }
3004 }
115329f1 3005
0da71265
MN
3006 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3007 h->delta_poc[0]= get_se_golomb(&s->gb);
115329f1 3008
0da71265
MN
3009 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3010 h->delta_poc[1]= get_se_golomb(&s->gb);
3011 }
115329f1 3012
0da71265 3013 init_poc(h);
115329f1 3014
0da71265
MN
3015 if(h->pps.redundant_pic_cnt_present){
3016 h->redundant_pic_count= get_ue_golomb(&s->gb);
3017 }
3018
1412060e 3019 //set defaults, might be overridden a few lines later
0da71265
MN
3020 h->ref_count[0]= h->pps.ref_count[0];
3021 h->ref_count[1]= h->pps.ref_count[1];
3022
975a1447
SS
3023 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
3024 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
0da71265
MN
3025 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3026 }
3027 num_ref_idx_active_override_flag= get_bits1(&s->gb);
115329f1 3028
0da71265
MN
3029 if(num_ref_idx_active_override_flag){
3030 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
975a1447 3031 if(h->slice_type_nos==AV_PICTURE_TYPE_B)
0da71265
MN
3032 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3033
187696fa 3034 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
9b879566 3035 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
88e7a4d1 3036 h->ref_count[0]= h->ref_count[1]= 1;
0da71265
MN
3037 return -1;
3038 }
3039 }
975a1447 3040 if(h->slice_type_nos == AV_PICTURE_TYPE_B)
187696fa
MN
3041 h->list_count= 2;
3042 else
3043 h->list_count= 1;
3044 }else
3045 h->list_count= 0;
0da71265 3046
0bf79634 3047 if(!default_ref_list_done){
ea6f00c4 3048 ff_h264_fill_default_ref_list(h);
0da71265
MN
3049 }
3050
4c7a232f
LA
3051 if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
3052 h->ref_count[1]= h->ref_count[0]= 0;
806bb93f 3053 return -1;
4c7a232f 3054 }
0da71265 3055
975a1447 3056 if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
07dff5c7 3057 s->last_picture_ptr= &h->ref_list[0][0];
8d2fc163 3058 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
07dff5c7 3059 }
975a1447 3060 if(h->slice_type_nos==AV_PICTURE_TYPE_B){
07dff5c7 3061 s->next_picture_ptr= &h->ref_list[1][0];
8d2fc163 3062 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
07dff5c7
MN
3063 }
3064
975a1447
SS
3065 if( (h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P )
3066 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
0da71265 3067 pred_weight_table(h);
975a1447 3068 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
1052b76f 3069 implicit_weight_table(h, -1);
1a29c6a0 3070 }else {
9f2d1b4f 3071 h->use_weight = 0;
cb99c652
GB
3072 for (i = 0; i < 2; i++) {
3073 h->luma_weight_flag[i] = 0;
3074 h->chroma_weight_flag[i] = 0;
3075 }
3076 }
115329f1 3077
12fe7594 3078 if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
9abc9873 3079 (s->avctx->err_recognition & AV_EF_EXPLODE))
12fe7594 3080 return AVERROR_INVALIDDATA;
0da71265 3081
1052b76f 3082 if(FRAME_MBAFF){
ea6f00c4 3083 ff_h264_fill_mbaff_ref_list(h);
5d18eaad 3084
975a1447 3085 if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
1052b76f
MN
3086 implicit_weight_table(h, 0);
3087 implicit_weight_table(h, 1);
3088 }
3089 }
3090
975a1447 3091 if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
943f69a6
MN
3092 ff_h264_direct_dist_scale_factor(h);
3093 ff_h264_direct_ref_list_init(h);
8f56e219 3094
975a1447 3095 if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
9963b332 3096 tmp = get_ue_golomb_31(&s->gb);
88e7a4d1
MN
3097 if(tmp > 2){
3098 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3099 return -1;
3100 }
3101 h->cabac_init_idc= tmp;
3102 }
e5017ab8
LA
3103
3104 h->last_qscale_diff = 0;
88e7a4d1 3105 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
fcc0224e 3106 if(tmp>51+6*(h->sps.bit_depth_luma-8)){
88e7a4d1 3107 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3ebc7e04
MN
3108 return -1;
3109 }
88e7a4d1 3110 s->qscale= tmp;
4691a77d
3111 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3112 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
0da71265 3113 //FIXME qscale / qp ... stuff
975a1447 3114 if(h->slice_type == AV_PICTURE_TYPE_SP){
1df1df0b 3115 get_bits1(&s->gb); /* sp_for_switch_flag */
0da71265 3116 }
975a1447 3117 if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
1df1df0b 3118 get_se_golomb(&s->gb); /* slice_qs_delta */
0da71265
MN
3119 }
3120
53c05b1e 3121 h->deblocking_filter = 1;
0c32e19d
MN
3122 h->slice_alpha_c0_offset = 52;
3123 h->slice_beta_offset = 52;
0da71265 3124 if( h->pps.deblocking_filter_parameters_present ) {
9963b332 3125 tmp= get_ue_golomb_31(&s->gb);
88e7a4d1
MN
3126 if(tmp > 2){
3127 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3128 return -1;
3129 }
3130 h->deblocking_filter= tmp;
115329f1 3131 if(h->deblocking_filter < 2)
53c05b1e
MN
3132 h->deblocking_filter^= 1; // 1<->0
3133
3134 if( h->deblocking_filter ) {
0c32e19d
MN
3135 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3136 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
3137 if( h->slice_alpha_c0_offset > 104U
3138 || h->slice_beta_offset > 104U){
3139 av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
3140 return -1;
3141 }
0da71265 3142 }
980a82b7 3143 }
afebe2f7 3144
61858a76 3145 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
975a1447
SS
3146 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
3147 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B)
61858a76
RD
3148 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3149 h->deblocking_filter= 0;
3150
afebe2f7 3151 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
ec970c21
3152 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3153 /* Cheat slightly for speed:
5d81d641 3154 Do not bother to deblock across slices. */
ec970c21
3155 h->deblocking_filter = 2;
3156 } else {
7ae94d52
3157 h0->max_contexts = 1;
3158 if(!h0->single_decode_warning) {
3159 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3160 h0->single_decode_warning = 1;
3161 }
33aec3f4
MN
3162 if (h != h0) {
3163 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3164 return 1;
3165 }
ec970c21 3166 }
afebe2f7 3167 }
6a2176aa
JGG
3168 h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
3169 - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
3170 + 6 * (h->sps.bit_depth_luma - 8);
afebe2f7 3171
0da71265
MN
3172#if 0 //FMO
3173 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3174 slice_group_change_cycle= get_bits(&s->gb, ?);
3175#endif
3176
afebe2f7
3177 h0->last_slice_type = slice_type;
3178 h->slice_num = ++h0->current_slice;
b735aeea
MN
3179 if(h->slice_num >= MAX_SLICES){
3180 av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3181 }
5175b937 3182
c32867b5 3183 for(j=0; j<2; j++){
6d7e6b26 3184 int id_list[16];
b735aeea 3185 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
6d7e6b26
MN
3186 for(i=0; i<16; i++){
3187 id_list[i]= 60;
657ccb5a 3188 if (h->ref_list[j][i].f.data[0]) {
6d7e6b26 3189 int k;
657ccb5a 3190 uint8_t *base = h->ref_list[j][i].f.base[0];
6d7e6b26 3191 for(k=0; k<h->short_ref_count; k++)
657ccb5a 3192 if (h->short_ref[k]->f.base[0] == base) {
6d7e6b26
MN
3193 id_list[i]= k;
3194 break;