hwaccel: OS X Video Decoder Acceleration (VDA) support.
[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"
0da71265
MN
30#include "dsputil.h"
31#include "avcodec.h"
32#include "mpegvideo.h"
26b4fe82 33#include "h264.h"
0da71265 34#include "h264data.h"
188d3c51 35#include "h264_mvpred.h"
0da71265 36#include "golomb.h"
199436b9 37#include "mathops.h"
626464fb 38#include "rectangle.h"
6a9c8594 39#include "thread.h"
369122dd 40#include "vdpau_internal.h"
cfa5a81e 41#include "libavutil/avassert.h"
0da71265 42
e5017ab8
LA
43#include "cabac.h"
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
MN
64/**
65 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
66 */
2bedc0e8
MN
67int ff_h264_check_intra4x4_pred_mode(H264Context *h){
68 MpegEncContext * const s = &h->s;
69 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
70 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
71 int i;
72
73 if(!(h->top_samples_available&0x8000)){
74 for(i=0; i<4; i++){
75 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
76 if(status<0){
77 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);
78 return -1;
79 } else if(status){
80 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
81 }
82 }
83 }
84
85 if((h->left_samples_available&0x8888)!=0x8888){
86 static const int mask[4]={0x8000,0x2000,0x80,0x20};
87 for(i=0; i<4; i++){
88 if(!(h->left_samples_available&mask[i])){
89 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
90 if(status<0){
91 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);
92 return -1;
93 } else if(status){
94 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
95 }
96 }
97 }
98 }
99
100 return 0;
101} //FIXME cleanup like ff_h264_check_intra_pred_mode
102
103/**
104 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
105 */
903d58f6 106int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
0da71265
MN
107 MpegEncContext * const s = &h->s;
108 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
109 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
115329f1 110
43ff0714 111 if(mode > 6U) {
5175b937 112 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 113 return -1;
5175b937 114 }
115329f1 115
0da71265
MN
116 if(!(h->top_samples_available&0x8000)){
117 mode= top[ mode ];
118 if(mode<0){
9b879566 119 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
120 return -1;
121 }
122 }
115329f1 123
d1d10e91 124 if((h->left_samples_available&0x8080) != 0x8080){
0da71265 125 mode= left[ mode ];
d1d10e91
MN
126 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
127 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
128 }
0da71265 129 if(mode<0){
9b879566 130 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 131 return -1;
115329f1 132 }
0da71265
MN
133 }
134
135 return mode;
136}
137
1790a5e9 138const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
0da71265
MN
139 int i, si, di;
140 uint8_t *dst;
24456882 141 int bufidx;
0da71265 142
bb270c08 143// src[0]&0x80; //forbidden bit
0da71265
MN
144 h->nal_ref_idc= src[0]>>5;
145 h->nal_unit_type= src[0]&0x1F;
146
147 src++; length--;
e08715d3 148
b250f9c6
AJ
149#if HAVE_FAST_UNALIGNED
150# if HAVE_FAST_64BIT
e08715d3
MN
151# define RS 7
152 for(i=0; i+1<length; i+=9){
19769ece 153 if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
e08715d3
MN
154# else
155# define RS 3
156 for(i=0; i+1<length; i+=5){
19769ece 157 if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
e08715d3
MN
158# endif
159 continue;
160 if(i>0 && !src[i]) i--;
161 while(src[i]) i++;
162#else
163# define RS 0
0da71265
MN
164 for(i=0; i+1<length; i+=2){
165 if(src[i]) continue;
166 if(i>0 && src[i-1]==0) i--;
e08715d3 167#endif
0da71265
MN
168 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
169 if(src[i+2]!=3){
170 /* startcode, so we must be past the end */
171 length=i;
172 }
173 break;
174 }
abb27cfb 175 i-= RS;
0da71265
MN
176 }
177
178 if(i>=length-1){ //no escaped 0
179 *dst_length= length;
180 *consumed= length+1; //+1 for the header
115329f1 181 return src;
0da71265
MN
182 }
183
24456882 184 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
238ef6da 185 av_fast_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
24456882 186 dst= h->rbsp_buffer[bufidx];
0da71265 187
ac658be5
FOL
188 if (dst == NULL){
189 return NULL;
190 }
191
3b66c4c5 192//printf("decoding esc\n");
593af7cd
MN
193 memcpy(dst, src, i);
194 si=di=i;
195 while(si+2<length){
0da71265 196 //remove escapes (very rare 1:2^22)
593af7cd
MN
197 if(src[si+2]>3){
198 dst[di++]= src[si++];
199 dst[di++]= src[si++];
200 }else if(src[si]==0 && src[si+1]==0){
0da71265
MN
201 if(src[si+2]==3){ //escape
202 dst[di++]= 0;
203 dst[di++]= 0;
204 si+=3;
c8470cc1 205 continue;
0da71265 206 }else //next start code
593af7cd 207 goto nsc;
0da71265
MN
208 }
209
210 dst[di++]= src[si++];
211 }
593af7cd
MN
212 while(si<length)
213 dst[di++]= src[si++];
214nsc:
0da71265 215
d4369630
AS
216 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
217
0da71265
MN
218 *dst_length= di;
219 *consumed= si + 1;//+1 for the header
90b5b51e 220//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
0da71265
MN
221 return dst;
222}
223
85297319
DEP
224/**
225 * Identify the exact end of the bitstream
226 * @return the length of the trailing, or 0 if damaged
227 */
228static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
0da71265
MN
229 int v= *src;
230 int r;
231
a9c9a240 232 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
0da71265
MN
233
234 for(r=1; r<9; r++){
235 if(v&1) return r;
236 v>>=1;
237 }
238 return 0;
239}
240
6a9c8594
AS
241static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
242 int y_offset, int list){
243 int raw_my= h->mv_cache[list][ scan8[n] ][1];
244 int filter_height= (raw_my&3) ? 2 : 0;
245 int full_my= (raw_my>>2) + y_offset;
246 int top = full_my - filter_height, bottom = full_my + height + filter_height;
247
248 return FFMAX(abs(top), bottom);
249}
250
251static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
252 int y_offset, int list0, int list1, int *nrefs){
253 MpegEncContext * const s = &h->s;
254 int my;
255
256 y_offset += 16*(s->mb_y >> MB_FIELD);
257
258 if(list0){
259 int ref_n = h->ref_cache[0][ scan8[n] ];
260 Picture *ref= &h->ref_list[0][ref_n];
261
262 // Error resilience puts the current picture in the ref list.
263 // Don't try to wait on these as it will cause a deadlock.
264 // Fields can wait on each other, though.
657ccb5a
DB
265 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
266 (ref->f.reference & 3) != s->picture_structure) {
6a9c8594
AS
267 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
268 if (refs[0][ref_n] < 0) nrefs[0] += 1;
269 refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
270 }
271 }
272
273 if(list1){
274 int ref_n = h->ref_cache[1][ scan8[n] ];
275 Picture *ref= &h->ref_list[1][ref_n];
276
657ccb5a
DB
277 if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
278 (ref->f.reference & 3) != s->picture_structure) {
6a9c8594
AS
279 my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
280 if (refs[1][ref_n] < 0) nrefs[1] += 1;
281 refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
282 }
283 }
284}
285
286/**
287 * Wait until all reference frames are available for MC operations.
288 *
289 * @param h the H264 context
290 */
291static void await_references(H264Context *h){
292 MpegEncContext * const s = &h->s;
293 const int mb_xy= h->mb_xy;
657ccb5a 294 const int mb_type = s->current_picture.f.mb_type[mb_xy];
6a9c8594
AS
295 int refs[2][48];
296 int nrefs[2] = {0};
297 int ref, list;
298
299 memset(refs, -1, sizeof(refs));
300
301 if(IS_16X16(mb_type)){
302 get_lowest_part_y(h, refs, 0, 16, 0,
303 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
304 }else if(IS_16X8(mb_type)){
305 get_lowest_part_y(h, refs, 0, 8, 0,
306 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
307 get_lowest_part_y(h, refs, 8, 8, 8,
308 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
309 }else if(IS_8X16(mb_type)){
310 get_lowest_part_y(h, refs, 0, 16, 0,
311 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
312 get_lowest_part_y(h, refs, 4, 16, 0,
313 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
314 }else{
315 int i;
316
317 assert(IS_8X8(mb_type));
318
319 for(i=0; i<4; i++){
320 const int sub_mb_type= h->sub_mb_type[i];
321 const int n= 4*i;
322 int y_offset= (i&2)<<2;
323
324 if(IS_SUB_8X8(sub_mb_type)){
325 get_lowest_part_y(h, refs, n , 8, y_offset,
326 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
327 }else if(IS_SUB_8X4(sub_mb_type)){
328 get_lowest_part_y(h, refs, n , 4, y_offset,
329 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
330 get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
331 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
332 }else if(IS_SUB_4X8(sub_mb_type)){
333 get_lowest_part_y(h, refs, n , 8, y_offset,
334 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
335 get_lowest_part_y(h, refs, n+1, 8, y_offset,
336 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
337 }else{
338 int j;
339 assert(IS_SUB_4X4(sub_mb_type));
340 for(j=0; j<4; j++){
341 int sub_y_offset= y_offset + 2*(j&2);
342 get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
343 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
344 }
345 }
346 }
347 }
348
349 for(list=h->list_count-1; list>=0; list--){
350 for(ref=0; ref<48 && nrefs[list]; ref++){
351 int row = refs[list][ref];
352 if(row >= 0){
353 Picture *ref_pic = &h->ref_list[list][ref];
657ccb5a 354 int ref_field = ref_pic->f.reference - 1;
6a9c8594
AS
355 int ref_field_picture = ref_pic->field_picture;
356 int pic_height = 16*s->mb_height >> ref_field_picture;
357
358 row <<= MB_MBAFF;
359 nrefs[list]--;
360
361 if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
362 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
363 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) , pic_height-1), 0);
364 }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
365 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field , pic_height-1), 0);
366 }else if(FIELD_PICTURE){
367 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
368 }else{
369 ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
370 }
371 }
372 }
373 }
374}
375
66c6b5e2
RB
376#if 0
377/**
378 * DCT transforms the 16 dc values.
379 * @param qp quantization parameter ??? FIXME
380 */
381static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
382// const int qmul= dequant_coeff[qp][0];
383 int i;
384 int temp[16]; //FIXME check if this is a good idea
385 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
386 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
387
388 for(i=0; i<4; i++){
389 const int offset= y_offset[i];
390 const int z0= block[offset+stride*0] + block[offset+stride*4];
391 const int z1= block[offset+stride*0] - block[offset+stride*4];
392 const int z2= block[offset+stride*1] - block[offset+stride*5];
393 const int z3= block[offset+stride*1] + block[offset+stride*5];
394
395 temp[4*i+0]= z0+z3;
396 temp[4*i+1]= z1+z2;
397 temp[4*i+2]= z1-z2;
398 temp[4*i+3]= z0-z3;
399 }
400
401 for(i=0; i<4; i++){
402 const int offset= x_offset[i];
403 const int z0= temp[4*0+i] + temp[4*2+i];
404 const int z1= temp[4*0+i] - temp[4*2+i];
405 const int z2= temp[4*1+i] - temp[4*3+i];
406 const int z3= temp[4*1+i] + temp[4*3+i];
407
408 block[stride*0 +offset]= (z0 + z3)>>1;
409 block[stride*2 +offset]= (z1 + z2)>>1;
410 block[stride*8 +offset]= (z1 - z2)>>1;
411 block[stride*10+offset]= (z0 - z3)>>1;
412 }
413}
414#endif
415
416#undef xStride
417#undef stride
418
66c6b5e2
RB
419#if 0
420static void chroma_dc_dct_c(DCTELEM *block){
421 const int stride= 16*2;
422 const int xStride= 16;
423 int a,b,c,d,e;
424
425 a= block[stride*0 + xStride*0];
426 b= block[stride*0 + xStride*1];
427 c= block[stride*1 + xStride*0];
428 d= block[stride*1 + xStride*1];
429
430 e= a-b;
431 a= a+b;
432 b= c-d;
433 c= c+d;
434
435 block[stride*0 + xStride*0]= (a+c);
436 block[stride*0 + xStride*1]= (e+b);
437 block[stride*1 + xStride*0]= (a-c);
438 block[stride*1 + xStride*1]= (e-b);
439}
440#endif
441
27209bb1
RB
442static av_always_inline void
443mc_dir_part(H264Context *h, Picture *pic, int n, int square,
444 int height, int delta, int list,
445 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
446 int src_x_offset, int src_y_offset,
447 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
448 int pixel_shift, int chroma_idc)
05fb63f5 449{
0da71265
MN
450 MpegEncContext * const s = &h->s;
451 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
5d18eaad 452 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
0da71265 453 const int luma_xy= (mx&3) + ((my&3)<<2);
c90b9442 454 int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
657ccb5a 455 uint8_t * src_y = pic->f.data[0] + offset;
5d18eaad
LM
456 uint8_t * src_cb, * src_cr;
457 int extra_width= h->emu_edge_width;
458 int extra_height= h->emu_edge_height;
0da71265
MN
459 int emu=0;
460 const int full_mx= mx>>2;
461 const int full_my= my>>2;
fbd312fd 462 const int pic_width = 16*s->mb_width;
0d43dd8c 463 const int pic_height = 16*s->mb_height >> MB_FIELD;
229d263c 464 int ysh;
115329f1 465
0da71265
MN
466 if(mx&7) extra_width -= 3;
467 if(my&7) extra_height -= 3;
115329f1
DB
468
469 if( full_mx < 0-extra_width
470 || full_my < 0-extra_height
471 || full_mx + 16/*FIXME*/ > pic_width + extra_width
fbd312fd 472 || full_my + 16/*FIXME*/ > pic_height + extra_height){
229d263c
RB
473 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
474 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
6e3ef511 475 src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
0da71265
MN
476 emu=1;
477 }
115329f1 478
5d18eaad 479 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
0da71265 480 if(!square){
5d18eaad 481 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
0da71265 482 }
115329f1 483
7b442ad9
JGG
484 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
485
05fb63f5 486 if(chroma_idc == 3 /* yuv444 */){
657ccb5a 487 src_cb = pic->f.data[1] + offset;
c90b9442
JGG
488 if(emu){
489 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
490 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
491 src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
492 }
493 qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
494 if(!square){
495 qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
496 }
497
657ccb5a 498 src_cr = pic->f.data[2] + offset;
c90b9442
JGG
499 if(emu){
500 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
501 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
502 src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
503 }
504 qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
505 if(!square){
506 qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
507 }
508 return;
509 }
510
05fb63f5
RB
511 ysh = 3 - (chroma_idc == 2 /* yuv422 */);
512 if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
5d18eaad 513 // chroma offset when predicting from a field of opposite parity
657ccb5a 514 my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
5d18eaad
LM
515 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
516 }
229d263c
RB
517
518 src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
519 src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
5d18eaad 520
0da71265 521 if(emu){
05fb63f5
RB
522 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
523 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
524 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
0da71265
MN
525 src_cb= s->edge_emu_buffer;
526 }
05fb63f5
RB
527 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
528 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
0da71265
MN
529
530 if(emu){
05fb63f5
RB
531 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
532 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
533 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
0da71265
MN
534 src_cr= s->edge_emu_buffer;
535 }
05fb63f5
RB
536 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
537 mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
0da71265
MN
538}
539
27209bb1
RB
540static av_always_inline void
541mc_part_std(H264Context *h, int n, int square, int height, int delta,
542 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
543 int x_offset, int y_offset,
544 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
545 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
546 int list0, int list1, int pixel_shift, int chroma_idc)
05fb63f5 547{
0da71265
MN
548 MpegEncContext * const s = &h->s;
549 qpel_mc_func *qpix_op= qpix_put;
550 h264_chroma_mc_func chroma_op= chroma_put;
115329f1 551
c90b9442 552 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
05fb63f5 553 if (chroma_idc == 3 /* yuv444 */) {
c90b9442
JGG
554 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
555 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
05fb63f5 556 } else if (chroma_idc == 2 /* yuv422 */) {
229d263c
RB
557 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
558 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
05fb63f5 559 } else /* yuv420 */ {
c90b9442
JGG
560 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
561 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
562 }
0da71265 563 x_offset += 8*s->mb_x;
0d43dd8c 564 y_offset += 8*(s->mb_y >> MB_FIELD);
115329f1 565
0da71265 566 if(list0){
1924f3ce 567 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
c2d33742 568 mc_dir_part(h, ref, n, square, height, delta, 0,
0da71265 569 dest_y, dest_cb, dest_cr, x_offset, y_offset,
05fb63f5 570 qpix_op, chroma_op, pixel_shift, chroma_idc);
0da71265
MN
571
572 qpix_op= qpix_avg;
573 chroma_op= chroma_avg;
574 }
575
576 if(list1){
1924f3ce 577 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
c2d33742 578 mc_dir_part(h, ref, n, square, height, delta, 1,
0da71265 579 dest_y, dest_cb, dest_cr, x_offset, y_offset,
05fb63f5 580 qpix_op, chroma_op, pixel_shift, chroma_idc);
0da71265
MN
581 }
582}
583
27209bb1
RB
584static av_always_inline void
585mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
586 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
587 int x_offset, int y_offset,
588 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
589 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
590 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
591 int list0, int list1, int pixel_shift, int chroma_idc){
9f2d1b4f 592 MpegEncContext * const s = &h->s;
c2d33742 593 int chroma_height;
9f2d1b4f 594
c90b9442 595 dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
05fb63f5 596 if (chroma_idc == 3 /* yuv444 */) {
c2d33742 597 chroma_height = height;
c90b9442
JGG
598 chroma_weight_avg = luma_weight_avg;
599 chroma_weight_op = luma_weight_op;
600 dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
601 dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
05fb63f5 602 } else if (chroma_idc == 2 /* yuv422 */) {
c2d33742 603 chroma_height = height;
229d263c
RB
604 dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
605 dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
05fb63f5 606 } else /* yuv420 */ {
c2d33742 607 chroma_height = height >> 1;
c90b9442
JGG
608 dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
609 dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
610 }
9f2d1b4f 611 x_offset += 8*s->mb_x;
0d43dd8c 612 y_offset += 8*(s->mb_y >> MB_FIELD);
115329f1 613
9f2d1b4f
LM
614 if(list0 && list1){
615 /* don't optimize for luma-only case, since B-frames usually
616 * use implicit weights => chroma too. */
617 uint8_t *tmp_cb = s->obmc_scratchpad;
c90b9442
JGG
618 uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
619 uint8_t *tmp_y = s->obmc_scratchpad + 16*h->mb_uvlinesize;
9f2d1b4f
LM
620 int refn0 = h->ref_cache[0][ scan8[n] ];
621 int refn1 = h->ref_cache[1][ scan8[n] ];
622
c2d33742 623 mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
9f2d1b4f 624 dest_y, dest_cb, dest_cr,
05fb63f5
RB
625 x_offset, y_offset, qpix_put, chroma_put,
626 pixel_shift, chroma_idc);
c2d33742 627 mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
9f2d1b4f 628 tmp_y, tmp_cb, tmp_cr,
05fb63f5
RB
629 x_offset, y_offset, qpix_put, chroma_put,
630 pixel_shift, chroma_idc);
9f2d1b4f
LM
631
632 if(h->use_weight == 2){
1052b76f 633 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
9f2d1b4f 634 int weight1 = 64 - weight0;
c2d33742
RB
635 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize,
636 height, 5, weight0, weight1, 0);
637 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
638 chroma_height, 5, weight0, weight1, 0);
639 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
640 chroma_height, 5, weight0, weight1, 0);
9f2d1b4f 641 }else{
c2d33742 642 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
3d9137c8
MN
643 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
644 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
c2d33742 645 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
3d9137c8
MN
646 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
647 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
c2d33742 648 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
3d9137c8
MN
649 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
650 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
9f2d1b4f
LM
651 }
652 }else{
653 int list = list1 ? 1 : 0;
654 int refn = h->ref_cache[list][ scan8[n] ];
655 Picture *ref= &h->ref_list[list][refn];
c2d33742 656 mc_dir_part(h, ref, n, square, height, delta, list,
9f2d1b4f 657 dest_y, dest_cb, dest_cr, x_offset, y_offset,
05fb63f5 658 qpix_put, chroma_put, pixel_shift, chroma_idc);
9f2d1b4f 659
c2d33742 660 luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
3d9137c8 661 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
9f2d1b4f 662 if(h->use_weight_chroma){
c2d33742 663 chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
3d9137c8 664 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
c2d33742 665 chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
3d9137c8 666 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
9f2d1b4f
LM
667 }
668 }
669}
670
27209bb1
RB
671static av_always_inline void
672mc_part(H264Context *h, int n, int square, int height, int delta,
673 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
674 int x_offset, int y_offset,
675 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
676 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
677 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
678 int list0, int list1, int pixel_shift, int chroma_idc)
05fb63f5 679{
9f2d1b4f 680 if((h->use_weight==2 && list0 && list1
1052b76f 681 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
9f2d1b4f 682 || h->use_weight==1)
c2d33742 683 mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
9f2d1b4f 684 x_offset, y_offset, qpix_put, chroma_put,
c2d33742 685 weight_op[0], weight_op[1], weight_avg[0],
05fb63f5 686 weight_avg[1], list0, list1, pixel_shift, chroma_idc);
9f2d1b4f 687 else
c2d33742 688 mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
6e3ef511 689 x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
05fb63f5 690 chroma_avg, list0, list1, pixel_shift, chroma_idc);
9f2d1b4f
LM
691}
692
27209bb1
RB
693static av_always_inline void
694prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
695{
513fbd8e
LM
696 /* fetch pixels for estimated mv 4 macroblocks ahead
697 * optimized for 64byte cache lines */
698 MpegEncContext * const s = &h->s;
699 const int refn = h->ref_cache[list][scan8[0]];
700 if(refn >= 0){
701 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
702 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
657ccb5a 703 uint8_t **src = h->ref_list[list][refn].f.data;
6e3ef511 704 int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
513fbd8e 705 s->dsp.prefetch(src[0]+off, s->linesize, 4);
05fb63f5 706 if (chroma_idc == 3 /* yuv444 */) {
c90b9442
JGG
707 s->dsp.prefetch(src[1]+off, s->linesize, 4);
708 s->dsp.prefetch(src[2]+off, s->linesize, 4);
709 }else{
710 off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
711 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
712 }
513fbd8e
LM
713 }
714}
715
6e3ef511 716static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
0da71265 717 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f 718 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
6e3ef511 719 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
05fb63f5
RB
720 int pixel_shift, int chroma_idc)
721{
0da71265 722 MpegEncContext * const s = &h->s;
64514ee8 723 const int mb_xy= h->mb_xy;
657ccb5a 724 const int mb_type = s->current_picture.f.mb_type[mb_xy];
115329f1 725
0da71265 726 assert(IS_INTER(mb_type));
115329f1 727
27237d52 728 if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
6a9c8594 729 await_references(h);
05fb63f5 730 prefetch_motion(h, 0, pixel_shift, chroma_idc);
513fbd8e 731
0da71265 732 if(IS_16X16(mb_type)){
c2d33742 733 mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 734 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
7231ccf4 735 weight_op, weight_avg,
6e3ef511 736 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
05fb63f5 737 pixel_shift, chroma_idc);
0da71265 738 }else if(IS_16X8(mb_type)){
c2d33742 739 mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 740 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
c2d33742 741 weight_op, weight_avg,
6e3ef511 742 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
05fb63f5 743 pixel_shift, chroma_idc);
c2d33742 744 mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
0da71265 745 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
c2d33742 746 weight_op, weight_avg,
6e3ef511 747 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
05fb63f5 748 pixel_shift, chroma_idc);
0da71265 749 }else if(IS_8X16(mb_type)){
c2d33742 750 mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 751 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
c2d33742 752 &weight_op[1], &weight_avg[1],
6e3ef511 753 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
05fb63f5 754 pixel_shift, chroma_idc);
c2d33742 755 mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
0da71265 756 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
c2d33742 757 &weight_op[1], &weight_avg[1],
6e3ef511 758 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
05fb63f5 759 pixel_shift, chroma_idc);
0da71265
MN
760 }else{
761 int i;
115329f1 762
0da71265
MN
763 assert(IS_8X8(mb_type));
764
765 for(i=0; i<4; i++){
766 const int sub_mb_type= h->sub_mb_type[i];
767 const int n= 4*i;
768 int x_offset= (i&1)<<2;
769 int y_offset= (i&2)<<1;
770
771 if(IS_SUB_8X8(sub_mb_type)){
c2d33742 772 mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 773 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
c2d33742 774 &weight_op[1], &weight_avg[1],
6e3ef511 775 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 776 pixel_shift, chroma_idc);
0da71265 777 }else if(IS_SUB_8X4(sub_mb_type)){
c2d33742 778 mc_part(h, n , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 779 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
c2d33742 780 &weight_op[1], &weight_avg[1],
6e3ef511 781 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 782 pixel_shift, chroma_idc);
c2d33742 783 mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
0da71265 784 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
c2d33742 785 &weight_op[1], &weight_avg[1],
6e3ef511 786 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 787 pixel_shift, chroma_idc);
0da71265 788 }else if(IS_SUB_4X8(sub_mb_type)){
c2d33742 789 mc_part(h, n , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 790 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
c2d33742 791 &weight_op[2], &weight_avg[2],
6e3ef511 792 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 793 pixel_shift, chroma_idc);
c2d33742 794 mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
0da71265 795 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
c2d33742 796 &weight_op[2], &weight_avg[2],
6e3ef511 797 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 798 pixel_shift, chroma_idc);
0da71265
MN
799 }else{
800 int j;
801 assert(IS_SUB_4X4(sub_mb_type));
802 for(j=0; j<4; j++){
803 int sub_x_offset= x_offset + 2*(j&1);
804 int sub_y_offset= y_offset + (j&2);
c2d33742 805 mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
0da71265 806 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
c2d33742 807 &weight_op[2], &weight_avg[2],
6e3ef511 808 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
05fb63f5 809 pixel_shift, chroma_idc);
0da71265
MN
810 }
811 }
812 }
813 }
513fbd8e 814
05fb63f5
RB
815 prefetch_motion(h, 1, pixel_shift, chroma_idc);
816}
817
818static av_always_inline void
819hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
820 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
821 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
822 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
823 int pixel_shift)
824{
825 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
826 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
827}
828
829static av_always_inline void
830hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
831 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
832 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
833 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
834 int pixel_shift)
835{
836 hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
837 qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
0da71265
MN
838}
839
91078926 840static void free_tables(H264Context *h, int free_rbsp){
7978debd 841 int i;
afebe2f7 842 H264Context *hx;
6a9c8594 843
0da71265 844 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
845 av_freep(&h->chroma_pred_mode_table);
846 av_freep(&h->cbp_table);
9e528114
LA
847 av_freep(&h->mvd_table[0]);
848 av_freep(&h->mvd_table[1]);
5ad984c9 849 av_freep(&h->direct_table);
0da71265
MN
850 av_freep(&h->non_zero_count);
851 av_freep(&h->slice_table_base);
852 h->slice_table= NULL;
c988f975 853 av_freep(&h->list_counts);
e5017ab8 854
0da71265 855 av_freep(&h->mb2b_xy);
d43c1922 856 av_freep(&h->mb2br_xy);
9f2d1b4f 857
6752dd5a 858 for(i = 0; i < MAX_THREADS; i++) {
afebe2f7
859 hx = h->thread_context[i];
860 if(!hx) continue;
861 av_freep(&hx->top_borders[1]);
862 av_freep(&hx->top_borders[0]);
863 av_freep(&hx->s.obmc_scratchpad);
91078926 864 if (free_rbsp){
fcb7e535
RB
865 av_freep(&hx->rbsp_buffer[1]);
866 av_freep(&hx->rbsp_buffer[0]);
867 hx->rbsp_buffer_size[0] = 0;
868 hx->rbsp_buffer_size[1] = 0;
91078926 869 }
d2d5e067 870 if (i) av_freep(&h->thread_context[i]);
afebe2f7 871 }
0da71265
MN
872}
873
239ea04c 874static void init_dequant8_coeff_table(H264Context *h){
c90b9442 875 int i,j,q,x;
fcc0224e 876 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
239ea04c 877
c90b9442
JGG
878 for(i=0; i<6; i++ ){
879 h->dequant8_coeff[i] = h->dequant8_buffer[i];
880 for(j=0; j<i; j++){
881 if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
882 h->dequant8_coeff[i] = h->dequant8_buffer[j];
883 break;
884 }
239ea04c 885 }
c90b9442
JGG
886 if(j<i)
887 continue;
239ea04c 888
fcc0224e 889 for(q=0; q<max_qp+1; q++){
d9ec210b
DP
890 int shift = div6[q];
891 int idx = rem6[q];
239ea04c 892 for(x=0; x<64; x++)
ca32f7f2 893 h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
548a1c8a
LM
894 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
895 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
896 }
897 }
898}
899
900static void init_dequant4_coeff_table(H264Context *h){
901 int i,j,q,x;
fcc0224e 902 const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
239ea04c
LM
903 for(i=0; i<6; i++ ){
904 h->dequant4_coeff[i] = h->dequant4_buffer[i];
905 for(j=0; j<i; j++){
906 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
907 h->dequant4_coeff[i] = h->dequant4_buffer[j];
908 break;
909 }
910 }
911 if(j<i)
912 continue;
913
fcc0224e 914 for(q=0; q<max_qp+1; q++){
d9ec210b
DP
915 int shift = div6[q] + 2;
916 int idx = rem6[q];
239ea04c 917 for(x=0; x<16; x++)
ca32f7f2 918 h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
ab2e3e2c 919 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
239ea04c
LM
920 h->pps.scaling_matrix4[i][x]) << shift;
921 }
922 }
923}
924
925static void init_dequant_tables(H264Context *h){
926 int i,x;
927 init_dequant4_coeff_table(h);
928 if(h->pps.transform_8x8_mode)
929 init_dequant8_coeff_table(h);
930 if(h->sps.transform_bypass){
931 for(i=0; i<6; i++)
932 for(x=0; x<16; x++)
933 h->dequant4_coeff[i][0][x] = 1<<6;
934 if(h->pps.transform_8x8_mode)
c90b9442 935 for(i=0; i<6; i++)
239ea04c
LM
936 for(x=0; x<64; x++)
937 h->dequant8_coeff[i][0][x] = 1<<6;
938 }
939}
940
941
903d58f6 942int ff_h264_alloc_tables(H264Context *h){
0da71265 943 MpegEncContext * const s = &h->s;
7bc9090a 944 const int big_mb_num= s->mb_stride * (s->mb_height+1);
145061a1 945 const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
239ea04c 946 int x,y;
0da71265 947
145061a1 948 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
e5017ab8 949
c90b9442 950 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 48 * sizeof(uint8_t), fail)
d31dbec3
RP
951 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
952 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
0da71265 953
d31dbec3 954 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
145061a1
MN
955 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
956 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
36b54927 957 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
c988f975 958 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
e5017ab8 959
b735aeea 960 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
5d18eaad 961 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
0da71265 962
d31dbec3 963 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
d43c1922 964 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
0da71265
MN
965 for(y=0; y<s->mb_height; y++){
966 for(x=0; x<s->mb_width; x++){
7bc9090a 967 const int mb_xy= x + y*s->mb_stride;
0da71265 968 const int b_xy = 4*x + 4*y*h->b_stride;
115329f1 969
0da71265 970 h->mb2b_xy [mb_xy]= b_xy;
e1c88a21 971 h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
0da71265
MN
972 }
973 }
9f2d1b4f 974
9c6221ae
GV
975 s->obmc_scratchpad = NULL;
976
56edbd81
LM
977 if(!h->dequant4_coeff[0])
978 init_dequant_tables(h);
979
0da71265
MN
980 return 0;
981fail:
91078926 982 free_tables(h, 1);
0da71265
MN
983 return -1;
984}
985
afebe2f7
986/**
987 * Mimic alloc_tables(), but for every context thread.
988 */
145061a1
MN
989static void clone_tables(H264Context *dst, H264Context *src, int i){
990 MpegEncContext * const s = &src->s;
991 dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
afebe2f7
992 dst->non_zero_count = src->non_zero_count;
993 dst->slice_table = src->slice_table;
994 dst->cbp_table = src->cbp_table;
995 dst->mb2b_xy = src->mb2b_xy;
d43c1922 996 dst->mb2br_xy = src->mb2br_xy;
afebe2f7 997 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
145061a1
MN
998 dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
999 dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
afebe2f7 1000 dst->direct_table = src->direct_table;
fb823b77 1001 dst->list_counts = src->list_counts;
afebe2f7 1002
afebe2f7 1003 dst->s.obmc_scratchpad = NULL;
76741b0e 1004 ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
afebe2f7
1005}
1006
1007/**
1008 * Init context
1009 * Allocate buffers which are not shared amongst multiple threads.
1010 */
1011static int context_init(H264Context *h){
c90b9442
JGG
1012 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1013 FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
afebe2f7 1014
145061a1
MN
1015 h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1016 h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1017
afebe2f7
1018 return 0;
1019fail:
1020 return -1; // free_tables will clean up for us
1021}
1022
9855b2e3
MN
1023static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1024
98a6fff9 1025static av_cold void common_init(H264Context *h){
0da71265 1026 MpegEncContext * const s = &h->s;
0da71265
MN
1027
1028 s->width = s->avctx->width;
1029 s->height = s->avctx->height;
1030 s->codec_id= s->avctx->codec->id;
115329f1 1031
76741b0e
BC
1032 ff_h264dsp_init(&h->h264dsp, 8, 1);
1033 ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
0da71265 1034
239ea04c 1035 h->dequant_coeff_pps= -1;
9a41c2c7 1036 s->unrestricted_mv=1;
56edbd81 1037
a5805aa9
MN
1038 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1039
56edbd81
LM
1040 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1041 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
0da71265
MN
1042}
1043
05e95319
HC
1044int ff_h264_decode_extradata(H264Context *h)
1045{
1046 AVCodecContext *avctx = h->s.avctx;
9855b2e3 1047
0c17beba 1048 if(avctx->extradata[0] == 1){
9855b2e3
MN
1049 int i, cnt, nalsize;
1050 unsigned char *p = avctx->extradata;
1051
1052 h->is_avc = 1;
1053
1054 if(avctx->extradata_size < 7) {
1055 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1056 return -1;
1057 }
1058 /* sps and pps in the avcC always have length coded with 2 bytes,
1059 so put a fake nal_length_size = 2 while parsing them */
1060 h->nal_length_size = 2;
1061 // Decode sps from avcC
1062 cnt = *(p+5) & 0x1f; // Number of sps
1063 p += 6;
1064 for (i = 0; i < cnt; i++) {
1065 nalsize = AV_RB16(p) + 2;
d1186ff7
LA
1066 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1067 return -1;
9855b2e3
MN
1068 if(decode_nal_units(h, p, nalsize) < 0) {
1069 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1070 return -1;
1071 }
1072 p += nalsize;
1073 }
1074 // Decode pps from avcC
1075 cnt = *(p++); // Number of pps
1076 for (i = 0; i < cnt; i++) {
1077 nalsize = AV_RB16(p) + 2;
d1186ff7
LA
1078 if (p - avctx->extradata + nalsize > avctx->extradata_size)
1079 return -1;
108f318d 1080 if (decode_nal_units(h, p, nalsize) < 0) {
9855b2e3
MN
1081 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1082 return -1;
1083 }
1084 p += nalsize;
1085 }
1086 // Now store right nal length size, that will be use to parse all other nals
0c17beba 1087 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
9855b2e3
MN
1088 } else {
1089 h->is_avc = 0;
05e95319 1090 if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
9855b2e3
MN
1091 return -1;
1092 }
05e95319
HC
1093 return 0;
1094}
1095
1096av_cold int ff_h264_decode_init(AVCodecContext *avctx){
1097 H264Context *h= avctx->priv_data;
1098 MpegEncContext * const s = &h->s;
ea2bb12e 1099 int i;
05e95319
HC
1100
1101 MPV_decode_defaults(s);
1102
1103 s->avctx = avctx;
1104 common_init(h);
1105
1106 s->out_format = FMT_H264;
1107 s->workaround_bugs= avctx->workaround_bugs;
1108
1109 // set defaults
1110// s->decode_mb= ff_h263_decode_mb;
1111 s->quarter_sample = 1;
1112 if(!avctx->has_b_frames)
1113 s->low_delay= 1;
1114
1115 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1116
1117 ff_h264_decode_init_vlc();
1118
6e3ef511 1119 h->pixel_shift = 0;
19a0729b 1120 h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
6e3ef511 1121
05e95319 1122 h->thread_context[0] = h;
6a9c8594 1123 h->outputed_poc = h->next_outputed_poc = INT_MIN;
ea2bb12e
RB
1124 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1125 h->last_pocs[i] = INT_MIN;
05e95319
HC
1126 h->prev_poc_msb= 1<<16;
1127 h->x264_build = -1;
1128 ff_h264_reset_sei(h);
1129 if(avctx->codec_id == CODEC_ID_H264){
1130 if(avctx->ticks_per_frame == 1){
1131 s->avctx->time_base.den *=2;
1132 }
1133 avctx->ticks_per_frame = 2;
1134 }
1135
1136 if(avctx->extradata_size > 0 && avctx->extradata &&
1137 ff_h264_decode_extradata(h))
1138 return -1;
1139
db8cb47d
MN
1140 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1141 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1142 s->low_delay = 0;
1143 }
9855b2e3 1144
0da71265
MN
1145 return 0;
1146}
1147
6a9c8594
AS
1148#define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1149static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1150{
1151 int i;
1152
1153 for (i=0; i<count; i++){
1154 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1155 IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1156 !from[i]));
1157 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1158 }
1159}
1160
1161static void copy_parameter_set(void **to, void **from, int count, int size)
1162{
1163 int i;
1164
1165 for (i=0; i<count; i++){
1166 if (to[i] && !from[i]) av_freep(&to[i]);
1167 else if (from[i] && !to[i]) to[i] = av_malloc(size);
1168
1169 if (from[i]) memcpy(to[i], from[i], size);
1170 }
1171}
1172
1173static int decode_init_thread_copy(AVCodecContext *avctx){
1174 H264Context *h= avctx->priv_data;
1175
1176 if (!avctx->is_copy) return 0;
1177 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1178 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1179
1180 return 0;
1181}
1182
1183#define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1184static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
1185 H264Context *h= dst->priv_data, *h1= src->priv_data;
1186 MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1187 int inited = s->context_initialized, err;
1188 int i;
1189
1190 if(dst == src || !s1->context_initialized) return 0;
1191
1192 err = ff_mpeg_update_thread_context(dst, src);
1193 if(err) return err;
1194
1195 //FIXME handle width/height changing
1196 if(!inited){
1197 for(i = 0; i < MAX_SPS_COUNT; i++)
1198 av_freep(h->sps_buffers + i);
1199
1200 for(i = 0; i < MAX_PPS_COUNT; i++)
1201 av_freep(h->pps_buffers + i);
1202
1203 memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1204 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1205 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
bac3ab13
DB
1206 if (ff_h264_alloc_tables(h) < 0) {
1207 av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1208 return AVERROR(ENOMEM);
1209 }
6a9c8594
AS
1210 context_init(h);
1211
1212 for(i=0; i<2; i++){
1213 h->rbsp_buffer[i] = NULL;
1214 h->rbsp_buffer_size[i] = 0;
1215 }
1216
1217 h->thread_context[0] = h;
1218
1219 // frame_start may not be called for the next thread (if it's decoding a bottom field)
1220 // so this has to be allocated here
c90b9442 1221 h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
6a9c8594
AS
1222
1223 s->dsp.clear_blocks(h->mb);
c90b9442 1224 s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
6a9c8594
AS
1225 }
1226
1227 //extradata/NAL handling
1228 h->is_avc = h1->is_avc;
1229
1230 //SPS/PPS
1231 copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1232 h->sps = h1->sps;
1233 copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1234 h->pps = h1->pps;
1235
1236 //Dequantization matrices
1237 //FIXME these are big - can they be only copied when PPS changes?
1238 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1239
1240 for(i=0; i<6; i++)
1241 h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1242
c90b9442 1243 for(i=0; i<6; i++)
6a9c8594
AS
1244 h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1245
1246 h->dequant_coeff_pps = h1->dequant_coeff_pps;
1247
1248 //POC timing
1249 copy_fields(h, h1, poc_lsb, redundant_pic_count);
1250
1251 //reference lists
1252 copy_fields(h, h1, ref_count, list_count);
1253 copy_fields(h, h1, ref_list, intra_gb);
1254 copy_fields(h, h1, short_ref, cabac_init_idc);
1255
1256 copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1257 copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1258 copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1259
1260 h->last_slice_type = h1->last_slice_type;
1261
1262 if(!s->current_picture_ptr) return 0;
1263
1264 if(!s->dropable) {
12fe7594 1265 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6a9c8594
AS
1266 h->prev_poc_msb = h->poc_msb;
1267 h->prev_poc_lsb = h->poc_lsb;
1268 }
1269 h->prev_frame_num_offset= h->frame_num_offset;
1270 h->prev_frame_num = h->frame_num;
1271 h->outputed_poc = h->next_outputed_poc;
1272
12fe7594 1273 return err;
6a9c8594
AS
1274}
1275
903d58f6 1276int ff_h264_frame_start(H264Context *h){
0da71265
MN
1277 MpegEncContext * const s = &h->s;
1278 int i;
6e3ef511 1279 const int pixel_shift = h->pixel_shift;
6a9c8594 1280 int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
0da71265 1281
af8aa846
MN
1282 if(MPV_frame_start(s, s->avctx) < 0)
1283 return -1;
0da71265 1284 ff_er_frame_start(s);
3a22d7fa
JD
1285 /*
1286 * MPV_frame_start uses pict_type to derive key_frame.
1287 * This is incorrect for H.264; IDR markings must be used.
1412060e 1288 * Zero here; IDR markings per slice in frame or fields are ORed in later.
3a22d7fa
JD
1289 * See decode_nal_units().
1290 */
657ccb5a 1291 s->current_picture_ptr->f.key_frame = 0;
c173a088 1292 s->current_picture_ptr->mmco_reset= 0;
0da71265
MN
1293
1294 assert(s->linesize && s->uvlinesize);
1295
1296 for(i=0; i<16; i++){
6e3ef511 1297 h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
c90b9442 1298 h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265 1299 }
c90b9442 1300 for(i=0; i<16; i++){
0da71265 1301 h->block_offset[16+i]=
c90b9442
JGG
1302 h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1303 h->block_offset[48+16+i]=
1304 h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
1305 }
1306
934b0821
LM
1307 /* can't be in alloc_tables because linesize isn't known there.
1308 * FIXME: redo bipred weight to not require extra buffer? */
6a9c8594 1309 for(i = 0; i < thread_count; i++)
0f016023 1310 if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
c90b9442 1311 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
5d18eaad 1312
2ce1c2e0 1313 /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
5820b90d 1314 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
934b0821 1315
657ccb5a 1316// s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
28bb9eb2 1317
1412060e 1318 // We mark the current picture as non-reference after allocating it, so
28bb9eb2
MN
1319 // that if we break out due to an error it can be released automatically
1320 // in the next MPV_frame_start().
1321 // SVQ3 as well as most other codecs have only last/next/current and thus
1322 // get released even with set reference, besides SVQ3 and others do not
1323 // mark frames as reference later "naturally".
1324 if(s->codec_id != CODEC_ID_SVQ3)
657ccb5a 1325 s->current_picture_ptr->f.reference = 0;
357282c6
MN
1326
1327 s->current_picture_ptr->field_poc[0]=
1328 s->current_picture_ptr->field_poc[1]= INT_MAX;
6a9c8594
AS
1329
1330 h->next_output_pic = NULL;
1331
5118c6c7 1332 assert(s->current_picture_ptr->long_ref==0);
357282c6 1333
af8aa846 1334 return 0;
0da71265
MN
1335}
1336
6a9c8594
AS
1337/**
1338 * Run setup operations that must be run after slice header decoding.
1339 * This includes finding the next displayed frame.
1340 *
1341 * @param h h264 master context
1342 * @param setup_finished enough NALs have been read that we can call
1343 * ff_thread_finish_setup()
1344 */
1345static void decode_postinit(H264Context *h, int setup_finished){
1346 MpegEncContext * const s = &h->s;
1347 Picture *out = s->current_picture_ptr;
1348 Picture *cur = s->current_picture_ptr;
1349 int i, pics, out_of_order, out_idx;
1350
657ccb5a
DB
1351 s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1352 s->current_picture_ptr->f.pict_type = s->pict_type;
6a9c8594
AS
1353
1354 if (h->next_output_pic) return;
1355
1356 if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1357 //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1358 //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1359 //to find this yet, so we assume the worst for now.
1360 //if (setup_finished)
1361 // ff_thread_finish_setup(s->avctx);
1362 return;
1363 }
1364
657ccb5a
DB
1365 cur->f.interlaced_frame = 0;
1366 cur->f.repeat_pict = 0;
6a9c8594
AS
1367
1368 /* Signal interlacing information externally. */
1369 /* Prioritize picture timing SEI information over used decoding process if it exists. */
1370
1371 if(h->sps.pic_struct_present_flag){
1372 switch (h->sei_pic_struct)
1373 {
1374 case SEI_PIC_STRUCT_FRAME:
1375 break;
1376 case SEI_PIC_STRUCT_TOP_FIELD:
1377 case SEI_PIC_STRUCT_BOTTOM_FIELD:
657ccb5a 1378 cur->f.interlaced_frame = 1;
6a9c8594
AS
1379 break;
1380 case SEI_PIC_STRUCT_TOP_BOTTOM:
1381 case SEI_PIC_STRUCT_BOTTOM_TOP:
1382 if (FIELD_OR_MBAFF_PICTURE)
657ccb5a 1383 cur->f.interlaced_frame = 1;
6a9c8594
AS
1384 else
1385 // try to flag soft telecine progressive
657ccb5a 1386 cur->f.interlaced_frame = h->prev_interlaced_frame;
6a9c8594
AS
1387 break;
1388 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1389 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1390 // Signal the possibility of telecined film externally (pic_struct 5,6)
1391 // From these hints, let the applications decide if they apply deinterlacing.
657ccb5a 1392 cur->f.repeat_pict = 1;
6a9c8594
AS
1393 break;
1394 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1395 // Force progressive here, as doubling interlaced frame is a bad idea.
657ccb5a 1396 cur->f.repeat_pict = 2;
6a9c8594
AS
1397 break;
1398 case SEI_PIC_STRUCT_FRAME_TRIPLING:
657ccb5a 1399 cur->f.repeat_pict = 4;
6a9c8594
AS
1400 break;
1401 }
1402
1403 if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
657ccb5a 1404 cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
6a9c8594
AS
1405 }else{
1406 /* Derive interlacing flag from used decoding process. */
657ccb5a 1407 cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
6a9c8594 1408 }
657ccb5a 1409 h->prev_interlaced_frame = cur->f.interlaced_frame;
6a9c8594
AS
1410
1411 if (cur->field_poc[0] != cur->field_poc[1]){
1412 /* Derive top_field_first from field pocs. */
657ccb5a 1413 cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
6a9c8594 1414 }else{
657ccb5a 1415 if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
6a9c8594
AS
1416 /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1417 if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
1418 || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
657ccb5a 1419 cur->f.top_field_first = 1;
6a9c8594 1420 else
657ccb5a 1421 cur->f.top_field_first = 0;
6a9c8594
AS
1422 }else{
1423 /* Most likely progressive */
657ccb5a 1424 cur->f.top_field_first = 0;
6a9c8594
AS
1425 }
1426 }
1427
1428 //FIXME do something with unavailable reference frames
1429
1430 /* Sort B-frames into display order */
1431
1432 if(h->sps.bitstream_restriction_flag
1433 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
1434 s->avctx->has_b_frames = h->sps.num_reorder_frames;
1435 s->low_delay = 0;
1436 }
1437
1438 if( s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
1439 && !h->sps.bitstream_restriction_flag){
1440 s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
1441 s->low_delay= 0;
1442 }
1443
1444 pics = 0;
1445 while(h->delayed_pic[pics]) pics++;
1446
1447 assert(pics <= MAX_DELAYED_PIC_COUNT);
1448
1449 h->delayed_pic[pics++] = cur;
657ccb5a
DB
1450 if (cur->f.reference == 0)
1451 cur->f.reference = DELAYED_PIC_REF;
6a9c8594
AS
1452
1453 out = h->delayed_pic[0];
1454 out_idx = 0;
657ccb5a 1455 for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
6a9c8594
AS
1456 if(h->delayed_pic[i]->poc < out->poc){
1457 out = h->delayed_pic[i];
1458 out_idx = i;
1459 }
657ccb5a 1460 if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
6a9c8594
AS
1461 h->next_outputed_poc= INT_MIN;
1462 out_of_order = out->poc < h->next_outputed_poc;
1463
1464 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
1465 { }
ea2bb12e
RB
1466 else if (out_of_order && pics-1 == s->avctx->has_b_frames &&
1467 s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
1468 int cnt = 0, invalid = 0;
1469 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
1470 cnt += out->poc < h->last_pocs[i];
1471 invalid += h->last_pocs[i] == INT_MIN;
1472 }
1473 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
1474 s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
1475 } else if (cnt) {
1476 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1477 h->last_pocs[i] = INT_MIN;
1478 }
1479 s->low_delay = 0;
1480 } else if (s->low_delay &&
1481 ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2) ||
1482 cur->f.pict_type == AV_PICTURE_TYPE_B)) {
6a9c8594
AS
1483 s->low_delay = 0;
1484 s->avctx->has_b_frames++;
1485 }
1486
1487 if(out_of_order || pics > s->avctx->has_b_frames){
657ccb5a 1488 out->f.reference &= ~DELAYED_PIC_REF;
6a9c8594
AS
1489 out->owner2 = s; // for frame threading, the owner must be the second field's thread
1490 // or else the first thread can release the picture and reuse it unsafely
1491 for(i=out_idx; h->delayed_pic[i]; i++)
1492 h->delayed_pic[i] = h->delayed_pic[i+1];
1493 }
ea2bb12e
RB
1494 memmove(h->last_pocs, &h->last_pocs[1], sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
1495 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = out->poc;
6a9c8594
AS
1496 if(!out_of_order && pics > s->avctx->has_b_frames){
1497 h->next_output_pic = out;
657ccb5a 1498 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
6a9c8594
AS
1499 h->next_outputed_poc = INT_MIN;
1500 } else
1501 h->next_outputed_poc = out->poc;
1502 }else{
1503 av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1504 }
1505
1506 if (setup_finished)
1507 ff_thread_finish_setup(s->avctx);
1508}
1509
76741b0e
BC
1510static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1511 uint8_t *src_cb, uint8_t *src_cr,
1512 int linesize, int uvlinesize, int simple)
1513{
53c05b1e 1514 MpegEncContext * const s = &h->s;
0b69d625 1515 uint8_t *top_border;
5f7f9719 1516 int top_idx = 1;
6e3ef511 1517 const int pixel_shift = h->pixel_shift;
76741b0e
BC
1518 int chroma444 = CHROMA444;
1519 int chroma422 = CHROMA422;
115329f1 1520
53c05b1e
MN
1521 src_y -= linesize;
1522 src_cb -= uvlinesize;
1523 src_cr -= uvlinesize;
1524
5f7f9719
MN
1525 if(!simple && FRAME_MBAFF){
1526 if(s->mb_y&1){
5f7f9719 1527 if(!MB_MBAFF){
0b69d625
AS
1528 top_border = h->top_borders[0][s->mb_x];
1529 AV_COPY128(top_border, src_y + 15*linesize);
6e3ef511
OA
1530 if (pixel_shift)
1531 AV_COPY128(top_border+16, src_y+15*linesize+16);
49fb20cb 1532 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c90b9442
JGG
1533 if(chroma444){
1534 if (pixel_shift){
1535 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1536 AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1537 AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1538 AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1539 } else {
1540 AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1541 AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1542 }
76741b0e
BC
1543 } else if(chroma422) {
1544 if (pixel_shift) {
1545 AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1546 AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1547 } else {
1548 AV_COPY64(top_border+16, src_cb + 15*uvlinesize);
1549 AV_COPY64(top_border+24, src_cr + 15*uvlinesize);
1550 }
6e3ef511 1551 } else {
c90b9442
JGG
1552 if (pixel_shift) {
1553 AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1554 AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1555 } else {
1556 AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1557 AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1558 }
6e3ef511 1559 }
5f7f9719
MN
1560 }
1561 }
c988f975
MN
1562 }else if(MB_MBAFF){
1563 top_idx = 0;
1564 }else
1565 return;
5f7f9719
MN
1566 }
1567
0b69d625 1568 top_border = h->top_borders[top_idx][s->mb_x];
3b66c4c5 1569 // There are two lines saved, the line above the the top macroblock of a pair,
6867a90b 1570 // and the line above the bottom macroblock
0b69d625 1571 AV_COPY128(top_border, src_y + 16*linesize);
6e3ef511
OA
1572 if (pixel_shift)
1573 AV_COPY128(top_border+16, src_y+16*linesize+16);
53c05b1e 1574
49fb20cb 1575 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c90b9442
JGG
1576 if(chroma444){
1577 if (pixel_shift){
1578 AV_COPY128(top_border+32, src_cb + 16*linesize);
1579 AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1580 AV_COPY128(top_border+64, src_cr + 16*linesize);
1581 AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1582 } else {
1583 AV_COPY128(top_border+16, src_cb + 16*linesize);
1584 AV_COPY128(top_border+32, src_cr + 16*linesize);
1585 }
76741b0e
BC
1586 } else if(chroma422) {
1587 if (pixel_shift) {
1588 AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1589 AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1590 } else {
1591 AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1592 AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1593 }
6e3ef511 1594 } else {
c90b9442
JGG
1595 if (pixel_shift) {
1596 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1597 AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1598 } else {
1599 AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1600 AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1601 }
6e3ef511 1602 }
53c05b1e
MN
1603 }
1604}
1605
bbdd52ed 1606static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
6e3ef511
OA
1607 uint8_t *src_cb, uint8_t *src_cr,
1608 int linesize, int uvlinesize,
c90b9442
JGG
1609 int xchg, int chroma444,
1610 int simple, int pixel_shift){
53c05b1e 1611 MpegEncContext * const s = &h->s;
4e987f82 1612 int deblock_topleft;
b69378e2 1613 int deblock_top;
5f7f9719 1614 int top_idx = 1;
1e4f1c56
AS
1615 uint8_t *top_border_m1;
1616 uint8_t *top_border;
5f7f9719
MN
1617
1618 if(!simple && FRAME_MBAFF){
1619 if(s->mb_y&1){
c988f975
MN
1620 if(!MB_MBAFF)
1621 return;
5f7f9719 1622 }else{
5f7f9719
MN
1623 top_idx = MB_MBAFF ? 0 : 1;
1624 }
5f7f9719 1625 }
b69378e2
1626
1627 if(h->deblocking_filter == 2) {
4e987f82
RB
1628 deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1629 deblock_top = h->top_type;
b69378e2 1630 } else {
4e987f82
RB
1631 deblock_topleft = (s->mb_x > 0);
1632 deblock_top = (s->mb_y > !!MB_FIELD);
b69378e2 1633 }
53c05b1e 1634
6e3ef511
OA
1635 src_y -= linesize + 1 + pixel_shift;
1636 src_cb -= uvlinesize + 1 + pixel_shift;
1637 src_cr -= uvlinesize + 1 + pixel_shift;
53c05b1e 1638
1e4f1c56
AS
1639 top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1640 top_border = h->top_borders[top_idx][s->mb_x];
1641
0b69d625 1642#define XCHG(a,b,xchg)\
6e3ef511
OA
1643 if (pixel_shift) {\
1644 if (xchg) {\
1645 AV_SWAP64(b+0,a+0);\
1646 AV_SWAP64(b+8,a+8);\
1647 } else {\
1648 AV_COPY128(b,a); \
1649 }\
1650 } else \
0b69d625
AS
1651if (xchg) AV_SWAP64(b,a);\
1652else AV_COPY64(b,a);
d89dc06a 1653
d89dc06a 1654 if(deblock_top){
4e987f82 1655 if(deblock_topleft){
6e3ef511 1656 XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
c988f975 1657 }
6e3ef511
OA
1658 XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1659 XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
cad4368a 1660 if(s->mb_x+1 < s->mb_width){
6e3ef511 1661 XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
43efd19a 1662 }
53c05b1e 1663 }
49fb20cb 1664 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c90b9442 1665 if(chroma444){
4e987f82 1666 if(deblock_topleft){
c90b9442
JGG
1667 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1668 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1669 }
1670 XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1671 XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1672 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1673 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1674 if(s->mb_x+1 < s->mb_width){
1675 XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1676 XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1677 }
1678 } else {
1679 if(deblock_top){
1680 if(deblock_topleft){
1681 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1682 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1683 }
1684 XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1685 XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
c988f975 1686 }
53c05b1e 1687 }
53c05b1e
MN
1688 }
1689}
1690
6e3ef511
OA
1691static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1692 if (high_bit_depth) {
1693 return AV_RN32A(((int32_t*)mb) + index);
1694 } else
1695 return AV_RN16A(mb + index);
1696}
1697
1698static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1699 if (high_bit_depth) {
1700 AV_WN32A(((int32_t*)mb) + index, value);
1701 } else
1702 AV_WN16A(mb + index, value);
1703}
1704
c90b9442
JGG
1705static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1706 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1707{
1708 MpegEncContext * const s = &h->s;
1709 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1710 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1711 int i;
1712 int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1713 block_offset += 16*p;
1714 if(IS_INTRA4x4(mb_type)){
1715 if(simple || !s->encoding){
1716 if(IS_8x8DCT(mb_type)){
1717 if(transform_bypass){
1718 idct_dc_add =
1719 idct_add = s->dsp.add_pixels8;
1720 }else{
1721 idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1722 idct_add = h->h264dsp.h264_idct8_add;
1723 }
1724 for(i=0; i<16; i+=4){
1725 uint8_t * const ptr= dest_y + block_offset[i];
1726 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1727 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1728 h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1729 }else{
1730 const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1731 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1732 (h->topright_samples_available<<i)&0x4000, linesize);
1733 if(nnz){
1734 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1735 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1736 else
1737 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1738 }
1739 }
1740 }
1741 }else{
1742 if(transform_bypass){
1743 idct_dc_add =
1744 idct_add = s->dsp.add_pixels4;
1745 }else{
1746 idct_dc_add = h->h264dsp.h264_idct_dc_add;
1747 idct_add = h->h264dsp.h264_idct_add;
1748 }
1749 for(i=0; i<16; i++){
1750 uint8_t * const ptr= dest_y + block_offset[i];
1751 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1752
1753 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1754 h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1755 }else{
1756 uint8_t *topright;
1757 int nnz, tr;
1758 uint64_t tr_high;
1759 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1760 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
6371ce4b 1761 assert(s->mb_y || linesize <= block_offset[i]);
c90b9442
JGG
1762 if(!topright_avail){
1763 if (pixel_shift) {
1764 tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1765 topright= (uint8_t*) &tr_high;
1766 } else {
4d1418cd 1767 tr= ptr[3 - linesize]*0x01010101u;
c90b9442
JGG
1768 topright= (uint8_t*) &tr;
1769 }
1770 }else
1771 topright= ptr + (4 << pixel_shift) - linesize;
1772 }else
1773 topright= NULL;
1774
1775 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1776 nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1777 if(nnz){
1778 if(is_h264){
1779 if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1780 idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1781 else
1782 idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1783 }else
1784 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1785 }
1786 }
1787 }
1788 }
1789 }
1790 }else{
1791 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1792 if(is_h264){
1793 if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
1794 if(!transform_bypass)
1795 h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1796 else{
1797 static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1798 8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1799 for(i = 0; i < 16; i++)
fdb5314e 1800 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
1801 }
1802 }
1803 }else
1804 ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1805 }
1806}
1807
1808static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1809 int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1810{
1811 MpegEncContext * const s = &h->s;
1812 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1813 int i;
1814 block_offset += 16*p;
1815 if(!IS_INTRA4x4(mb_type)){
1816 if(is_h264){
1817 if(IS_INTRA16x16(mb_type)){
1818 if(transform_bypass){
1819 if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
1820 h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1821 }else{
1822 for(i=0; i<16; i++){
3b79f2e2 1823 if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
c90b9442
JGG
1824 s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1825 }
1826 }
1827 }else{
1828 h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1829 }
1830 }else if(h->cbp&15){
1831 if(transform_bypass){
1832 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1833 idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1834 for(i=0; i<16; i+=di){
1835 if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1836 idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1837 }
1838 }
1839 }else{
1840 if(IS_8x8DCT(mb_type)){
1841 h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1842 }else{
1843 h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1844 }
1845 }
1846 }
1847 }else{
1848 for(i=0; i<16; i++){
1849 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1850 uint8_t * const ptr= dest_y + block_offset[i];
1851 ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1852 }
1853 }
1854 }
1855 }
1856}
1857
05fb63f5
RB
1858static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1859{
0da71265
MN
1860 MpegEncContext * const s = &h->s;
1861 const int mb_x= s->mb_x;
1862 const int mb_y= s->mb_y;
64514ee8 1863 const int mb_xy= h->mb_xy;
657ccb5a 1864 const int mb_type = s->current_picture.f.mb_type[mb_xy];
0da71265
MN
1865 uint8_t *dest_y, *dest_cb, *dest_cr;
1866 int linesize, uvlinesize /*dct_offset*/;
c90b9442 1867 int i, j;
6867a90b 1868 int *block_offset = &h->block_offset[0];
41e4055b 1869 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
8b6871ed 1870 /* is_h264 should always be true if SVQ3 is disabled. */
49fb20cb 1871 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
36940eca 1872 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
76741b0e 1873 const int block_h = 16 >> s->chroma_y_shift;
05fb63f5 1874 const int chroma422 = CHROMA422;
0da71265 1875
657ccb5a 1876 dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
76741b0e
BC
1877 dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1878 dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
0da71265 1879
6e3ef511
OA
1880 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1881 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
a957c27b 1882
c988f975
MN
1883 h->list_counts[mb_xy]= h->list_count;
1884
bd91fee3 1885 if (!simple && MB_FIELD) {
5d18eaad
LM
1886 linesize = h->mb_linesize = s->linesize * 2;
1887 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
c90b9442 1888 block_offset = &h->block_offset[48];
1412060e 1889 if(mb_y&1){ //FIXME move out of this function?
0da71265 1890 dest_y -= s->linesize*15;
76741b0e
BC
1891 dest_cb-= s->uvlinesize * (block_h - 1);
1892 dest_cr-= s->uvlinesize * (block_h - 1);
0da71265 1893 }
5d18eaad
LM
1894 if(FRAME_MBAFF) {
1895 int list;
3425501d 1896 for(list=0; list<h->list_count; list++){
5d18eaad
LM
1897 if(!USES_LIST(mb_type, list))
1898 continue;
1899 if(IS_16X16(mb_type)){
1900 int8_t *ref = &h->ref_cache[list][scan8[0]];
1710856c 1901 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
5d18eaad
LM
1902 }else{
1903 for(i=0; i<16; i+=4){
5d18eaad
LM
1904 int ref = h->ref_cache[list][scan8[i]];
1905 if(ref >= 0)
1710856c 1906 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
5d18eaad
LM
1907 }
1908 }
1909 }
1910 }
0da71265 1911 } else {
5d18eaad
LM
1912 linesize = h->mb_linesize = s->linesize;
1913 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
0da71265
MN
1914// dct_offset = s->linesize * 16;
1915 }
115329f1 1916
bd91fee3 1917 if (!simple && IS_INTRA_PCM(mb_type)) {
6e3ef511
OA
1918 if (pixel_shift) {
1919 const int bit_depth = h->sps.bit_depth_luma;
1920 int j;
1921 GetBitContext gb;
1922 init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1923
1924 for (i = 0; i < 16; i++) {
1925 uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1926 for (j = 0; j < 16; j++)
1927 tmp_y[j] = get_bits(&gb, bit_depth);
1928 }
7b442ad9 1929 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6581e161 1930 if (!h->sps.chroma_format_idc) {
229d263c 1931 for (i = 0; i < block_h; i++) {
6581e161
JD
1932 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1933 for (j = 0; j < 8; j++) {
1934 tmp_cb[j] = 1 << (bit_depth - 1);
1935 }
1936 }
229d263c 1937 for (i = 0; i < block_h; i++) {
6581e161
JD
1938 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1939 for (j = 0; j < 8; j++) {
1940 tmp_cr[j] = 1 << (bit_depth - 1);
1941 }
1942 }
1943 } else {
76741b0e 1944 for (i = 0; i < block_h; i++) {
6581e161
JD
1945 uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1946 for (j = 0; j < 8; j++)
1947 tmp_cb[j] = get_bits(&gb, bit_depth);
1948 }
76741b0e 1949 for (i = 0; i < block_h; i++) {
6581e161
JD
1950 uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1951 for (j = 0; j < 8; j++)
1952 tmp_cr[j] = get_bits(&gb, bit_depth);
1953 }
7b442ad9 1954 }
c177cfb4 1955 }
504811ba 1956 } else {
7b442ad9
JGG
1957 for (i=0; i<16; i++) {
1958 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1959 }
1960 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6581e161 1961 if (!h->sps.chroma_format_idc) {
229d263c 1962 for (i = 0; i < block_h; i++) {
6581e161
JD
1963 memset(dest_cb + i*uvlinesize, 128, 8);
1964 memset(dest_cr + i*uvlinesize, 128, 8);
1965 }
1966 } else {
76741b0e 1967 for (i = 0; i < block_h; i++) {
6581e161
JD
1968 memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4, 8);
1969 memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4, 8);
1970 }
7b442ad9
JGG
1971 }
1972 }
6e3ef511 1973 }
e7e09b49
LLL
1974 } else {
1975 if(IS_INTRA(mb_type)){
5f7f9719 1976 if(h->deblocking_filter)
c90b9442 1977 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
53c05b1e 1978
49fb20cb 1979 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c92a30bb
KS
1980 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
1981 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
e7e09b49 1982 }
0da71265 1983
c90b9442
JGG
1984 hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
1985
5f7f9719 1986 if(h->deblocking_filter)
c90b9442 1987 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
bd91fee3 1988 }else if(is_h264){
05fb63f5
RB
1989 if (chroma422) {
1990 hl_motion_422(h, dest_y, dest_cb, dest_cr,
1991 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1992 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
1993 h->h264dsp.weight_h264_pixels_tab,
1994 h->h264dsp.biweight_h264_pixels_tab,
1995 pixel_shift);
1996 } else {
1997 hl_motion_420(h, dest_y, dest_cb, dest_cr,
1998 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
1999 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2000 h->h264dsp.weight_h264_pixels_tab,
2001 h->h264dsp.biweight_h264_pixels_tab,
2002 pixel_shift);
2003 }
0da71265 2004 }
e7e09b49 2005
c90b9442 2006 hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
0da71265 2007
49fb20cb 2008 if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
ef9d1d15
LM
2009 uint8_t *dest[2] = {dest_cb, dest_cr};
2010 if(transform_bypass){
96465b90 2011 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
c90b9442
JGG
2012 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
2013 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
96465b90 2014 }else{
c25ac15a 2015 idct_add = s->dsp.add_pixels4;
c90b9442
JGG
2016 for(j=1; j<3; j++){
2017 for(i=j*16; i<j*16+4; i++){
2018 if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2019 idct_add (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
2020 }
05fb63f5 2021 if (chroma422) {
229d263c 2022 for(i=j*16+4; i<j*16+8; i++){
640d5f1c 2023 if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
229d263c
RB
2024 idct_add (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2025 }
2026 }
96465b90
MN
2027 }
2028 }
ef9d1d15 2029 }else{
aebb5d6d 2030 if(is_h264){
76741b0e 2031 int qp[2];
05fb63f5 2032 if (chroma422) {
76741b0e
BC
2033 qp[0] = h->chroma_qp[0] + 3;
2034 qp[1] = h->chroma_qp[1] + 3;
2035 } else {
2036 qp[0] = h->chroma_qp[0];
2037 qp[1] = h->chroma_qp[1];
2038 }
2e186601 2039 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
76741b0e 2040 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 2041 if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
76741b0e 2042 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
2043 h->h264dsp.h264_idct_add8(dest, block_offset,
2044 h->mb, uvlinesize,
2045 h->non_zero_count_cache);
aebb5d6d 2046 }else{
c90b9442
JGG
2047 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]);
2048 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]);
2049 for(j=1; j<3; j++){
2050 for(i=j*16; i<j*16+4; i++){
2051 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2052 uint8_t * const ptr= dest[j-1] + block_offset[i];
2053 ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2054 }
aebb5d6d 2055 }
e7e09b49 2056 }
4704097a 2057 }
0da71265
MN
2058 }
2059 }
2060 }
c212fb0c 2061 if(h->cbp || IS_INTRA(mb_type))
c90b9442 2062 {
c9c49387 2063 s->dsp.clear_blocks(h->mb);
c90b9442
JGG
2064 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2065 }
2066}
2067
2068static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2069 MpegEncContext * const s = &h->s;
2070 const int mb_x= s->mb_x;
2071 const int mb_y= s->mb_y;
2072 const int mb_xy= h->mb_xy;
657ccb5a 2073 const int mb_type = s->current_picture.f.mb_type[mb_xy];
c90b9442
JGG
2074 uint8_t *dest[3];
2075 int linesize;
2076 int i, j, p;
2077 int *block_offset = &h->block_offset[0];
2078 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
7b442ad9 2079 const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
c90b9442 2080
7b442ad9 2081 for (p = 0; p < plane_count; p++)
c90b9442 2082 {
657ccb5a 2083 dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
c90b9442
JGG
2084 s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2085 }
2086
2087 h->list_counts[mb_xy]= h->list_count;
2088
2089 if (!simple && MB_FIELD) {
2090 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2091 block_offset = &h->block_offset[48];
2092 if(mb_y&1) //FIXME move out of this function?
2093 for (p = 0; p < 3; p++)
2094 dest[p] -= s->linesize*15;
2095 if(FRAME_MBAFF) {
2096 int list;
2097 for(list=0; list<h->list_count; list++){
2098 if(!USES_LIST(mb_type, list))
2099 continue;
2100 if(IS_16X16(mb_type)){
2101 int8_t *ref = &h->ref_cache[list][scan8[0]];
2102 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2103 }else{
2104 for(i=0; i<16; i+=4){
2105 int ref = h->ref_cache[list][scan8[i]];
2106 if(ref >= 0)
2107 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2108 }
2109 }
2110 }
2111 }
2112 } else {
2113 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2114 }
2115
2116 if (!simple && IS_INTRA_PCM(mb_type)) {
2117 if (pixel_shift) {
2118 const int bit_depth = h->sps.bit_depth_luma;
2119 GetBitContext gb;
2120 init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2121
7b442ad9 2122 for (p = 0; p < plane_count; p++) {
c90b9442
JGG
2123 for (i = 0; i < 16; i++) {
2124 uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2125 for (j = 0; j < 16; j++)
2126 tmp[j] = get_bits(&gb, bit_depth);
2127 }
2128 }
2129 } else {
7b442ad9 2130 for (p = 0; p < plane_count; p++) {
c90b9442
JGG
2131 for (i = 0; i < 16; i++) {
2132 memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2133 }
2134 }
2135 }
2136 } else {
2137 if(IS_INTRA(mb_type)){
2138 if(h->deblocking_filter)
2139 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2140
7b442ad9 2141 for (p = 0; p < plane_count; p++)
c90b9442
JGG
2142 hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2143
2144 if(h->deblocking_filter)
2145 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2146 }else{
298e52c9
JGG
2147 hl_motion(h, dest[0], dest[1], dest[2],
2148 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2149 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
2150 h->h264dsp.weight_h264_pixels_tab,
05fb63f5 2151 h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
c90b9442
JGG
2152 }
2153
7b442ad9 2154 for (p = 0; p < plane_count; p++)
c90b9442
JGG
2155 hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2156 }
2157 if(h->cbp || IS_INTRA(mb_type))
2158 {
2159 s->dsp.clear_blocks(h->mb);
2160 s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2161 }
0da71265
MN
2162}
2163
0da71265 2164/**
bd91fee3
AS
2165 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2166 */
6e3ef511
OA
2167#define hl_decode_mb_simple(sh, bits) \
2168static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2169 hl_decode_mb_internal(h, 1, sh); \
bd91fee3 2170}
6e3ef511
OA
2171hl_decode_mb_simple(0, 8);
2172hl_decode_mb_simple(1, 16);
bd91fee3
AS
2173
2174/**
2175 * Process a macroblock; this handles edge cases, such as interlacing.
2176 */
2177static void av_noinline hl_decode_mb_complex(H264Context *h){
6e3ef511 2178 hl_decode_mb_internal(h, 0, h->pixel_shift);
bd91fee3
AS
2179}
2180
c90b9442
JGG
2181static void av_noinline hl_decode_mb_444_complex(H264Context *h){
2182 hl_decode_mb_444_internal(h, 0, h->pixel_shift);
2183}
2184
2185static void av_noinline hl_decode_mb_444_simple(H264Context *h){
2186 hl_decode_mb_444_internal(h, 1, 0);
2187}
2188
903d58f6 2189void ff_h264_hl_decode_mb(H264Context *h){
bd91fee3 2190 MpegEncContext * const s = &h->s;
64514ee8 2191 const int mb_xy= h->mb_xy;
657ccb5a 2192 const int mb_type = s->current_picture.f.mb_type[mb_xy];
49fb20cb 2193 int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
bd91fee3 2194
c90b9442
JGG
2195 if (CHROMA444) {
2196 if(is_complex || h->pixel_shift)
2197 hl_decode_mb_444_complex(h);
2198 else
2199 hl_decode_mb_444_simple(h);
2200 } else if (is_complex) {
bd91fee3 2201 hl_decode_mb_complex(h);
6e3ef511
OA
2202 } else if (h->pixel_shift) {
2203 hl_decode_mb_simple_16(h);
2204 } else
2205 hl_decode_mb_simple_8(h);
bd91fee3
AS
2206}
2207
0da71265
MN
2208static int pred_weight_table(H264Context *h){
2209 MpegEncContext * const s = &h->s;
2210 int list, i;
9f2d1b4f 2211 int luma_def, chroma_def;
115329f1 2212
9f2d1b4f
LM
2213 h->use_weight= 0;
2214 h->use_weight_chroma= 0;
0da71265 2215 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
c90b9442 2216 if(h->sps.chroma_format_idc)
09fffe9b 2217 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
9f2d1b4f
LM
2218 luma_def = 1<<h->luma_log2_weight_denom;
2219 chroma_def = 1<<h->chroma_log2_weight_denom;
0da71265
MN
2220
2221 for(list=0; list<2; list++){
cb99c652
GB
2222 h->luma_weight_flag[list] = 0;
2223 h->chroma_weight_flag[list] = 0;
0da71265
MN
2224 for(i=0; i<h->ref_count[list]; i++){
2225 int luma_weight_flag, chroma_weight_flag;
115329f1 2226
0da71265
MN
2227 luma_weight_flag= get_bits1(&s->gb);
2228 if(luma_weight_flag){
3d9137c8
MN
2229 h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2230 h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2231 if( h->luma_weight[i][list][0] != luma_def
2232 || h->luma_weight[i][list][1] != 0) {
9f2d1b4f 2233 h->use_weight= 1;
cb99c652
GB
2234 h->luma_weight_flag[list]= 1;
2235 }
9f2d1b4f 2236 }else{
3d9137c8
MN
2237 h->luma_weight[i][list][0]= luma_def;
2238 h->luma_weight[i][list][1]= 0;
0da71265
MN
2239 }
2240
c90b9442 2241 if(h->sps.chroma_format_idc){
fef744d4
MN
2242 chroma_weight_flag= get_bits1(&s->gb);
2243 if(chroma_weight_flag){
2244 int j;
2245 for(j=0; j<2; j++){
3d9137c8
MN
2246 h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2247 h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2248 if( h->chroma_weight[i][list][j][0] != chroma_def
2249 || h->chroma_weight[i][list][j][1] != 0) {
fef744d4 2250 h->use_weight_chroma= 1;
cb99c652
GB
2251 h->chroma_weight_flag[list]= 1;
2252 }
fef744d4
MN
2253 }
2254 }else{
2255 int j;
2256 for(j=0; j<2; j++){
3d9137c8
MN
2257 h->chroma_weight[i][list][j][0]= chroma_def;
2258 h->chroma_weight[i][list][j][1]= 0;
fef744d4 2259 }
0da71265
MN
2260 }
2261 }
2262 }
975a1447 2263 if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
0da71265 2264 }
9f2d1b4f 2265 h->use_weight= h->use_weight || h->use_weight_chroma;
0da71265
MN
2266 return 0;
2267}
2268
1052b76f
MN
2269/**
2270 * Initialize implicit_weight table.
6da88bd3 2271 * @param field 0/1 initialize the weight for interlaced MBAFF
1052b76f
MN
2272 * -1 initializes the rest
2273 */
2274static void implicit_weight_table(H264Context *h, int field){
9f2d1b4f 2275 MpegEncContext * const s = &h->s;
1052b76f 2276 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
9f2d1b4f 2277
ce09f927
GB
2278 for (i = 0; i < 2; i++) {
2279 h->luma_weight_flag[i] = 0;
2280 h->chroma_weight_flag[i] = 0;
2281 }
2282
1052b76f 2283 if(field < 0){
4418aa9c
RB
2284 if (s->picture_structure == PICT_FRAME) {
2285 cur_poc = s->current_picture_ptr->poc;
2286 } else {
2287 cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2288 }
1052b76f 2289 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
9f2d1b4f
LM
2290 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2291 h->use_weight= 0;
2292 h->use_weight_chroma= 0;
2293 return;
2294 }
1052b76f
MN
2295 ref_start= 0;
2296 ref_count0= h->ref_count[0];
2297 ref_count1= h->ref_count[1];
2298 }else{
2299 cur_poc = s->current_picture_ptr->field_poc[field];
2300 ref_start= 16;
2301 ref_count0= 16+2*h->ref_count[0];
2302 ref_count1= 16+2*h->ref_count[1];
2303 }
9f2d1b4f
LM
2304
2305 h->use_weight= 2;
2306 h->use_weight_chroma= 2;
2307 h->luma_log2_weight_denom= 5;
2308 h->chroma_log2_weight_denom= 5;
2309
1052b76f 2310 for(ref0=ref_start; ref0 < ref_count0; ref0++){
9f2d1b4f 2311 int poc0 = h->ref_list[0][ref0].poc;
1052b76f 2312 for(ref1=ref_start; ref1 < ref_count1; ref1++){
87cf70eb
JD
2313 int w = 32;
2314 if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2315 int poc1 = h->ref_list[1][ref1].poc;
2316 int td = av_clip(poc1 - poc0, -128, 127);
2317 if(td){
2318 int tb = av_clip(cur_poc - poc0, -128, 127);
2319 int tx = (16384 + (FFABS(td) >> 1)) / td;
2320 int dist_scale_factor = (tb*tx + 32) >> 8;
2321 if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2322 w = 64 - dist_scale_factor;
2323 }
1052b76f
MN
2324 }
2325 if(field<0){
2326 h->implicit_weight[ref0][ref1][0]=
2327 h->implicit_weight[ref0][ref1][1]= w;
2328 }else{
2329 h->implicit_weight[ref0][ref1][field]=w;
72f86ec0 2330 }
9f2d1b4f
LM
2331 }
2332 }
2333}
2334
8fd57a66 2335/**
5175b937 2336 * instantaneous decoder refresh.
0da71265
MN
2337 */
2338static void idr(H264Context *h){
ea6f00c4 2339 ff_h264_remove_all_refs(h);
a149c1a5 2340 h->prev_frame_num= 0;
80f8e035
MN
2341 h->prev_frame_num_offset= 0;
2342 h->prev_poc_msb=
2343 h->prev_poc_lsb= 0;
0da71265
MN
2344}
2345
7c33ad19
LM
2346/* forget old pics after a seek */
2347static void flush_dpb(AVCodecContext *avctx){
2348 H264Context *h= avctx->priv_data;
2349 int i;
64b9d48f 2350 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
285b570f 2351 if(h->delayed_pic[i])
657ccb5a 2352 h->delayed_pic[i]->f.reference = 0;
7c33ad19 2353 h->delayed_pic[i]= NULL;
285b570f 2354 }
6a9c8594 2355 h->outputed_poc=h->next_outputed_poc= INT_MIN;
b19d493f 2356 h->prev_interlaced_frame = 1;
7c33ad19 2357 idr(h);
ca159196 2358 if(h->s.current_picture_ptr)
657ccb5a 2359 h->s.current_picture_ptr->f.reference = 0;
12d96de3 2360 h->s.first_field= 0;
9c095463 2361 ff_h264_reset_sei(h);
e240f898 2362 ff_mpeg_flush(avctx);
7c33ad19
LM
2363}
2364
0da71265
MN
2365static int init_poc(H264Context *h){
2366 MpegEncContext * const s = &h->s;
2367 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2368 int field_poc[2];
357282c6 2369 Picture *cur = s->current_picture_ptr;
0da71265 2370
b78a6baa 2371 h->frame_num_offset= h->prev_frame_num_offset;
5710b371 2372 if(h->frame_num < h->prev_frame_num)
b78a6baa 2373 h->frame_num_offset += max_frame_num;
0da71265
MN
2374
2375 if(h->sps.poc_type==0){
2376 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2377
2378 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2379 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2380 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2381 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2382 else
2383 h->poc_msb = h->prev_poc_msb;
2384//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
115329f1 2385 field_poc[0] =
0da71265 2386 field_poc[1] = h->poc_msb + h->poc_lsb;
115329f1 2387 if(s->picture_structure == PICT_FRAME)
0da71265
MN
2388 field_poc[1] += h->delta_poc_bottom;
2389 }else if(h->sps.poc_type==1){
2390 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2391 int i;
2392
2393 if(h->sps.poc_cycle_length != 0)
2394 abs_frame_num = h->frame_num_offset + h->frame_num;
2395 else
2396 abs_frame_num = 0;
2397
2398 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2399 abs_frame_num--;
115329f1 2400
0da71265
MN
2401 expected_delta_per_poc_cycle = 0;
2402 for(i=0; i < h->sps.poc_cycle_length; i++)
2403 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2404
2405 if(abs_frame_num > 0){
2406 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2407 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2408
2409 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2410 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2411 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2412 } else
2413 expectedpoc = 0;
2414
115329f1 2415 if(h->nal_ref_idc == 0)
0da71265 2416 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
115329f1 2417
0da71265
MN
2418 field_poc[0] = expectedpoc + h->delta_poc[0];
2419 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2420
2421 if(s->picture_structure == PICT_FRAME)
2422 field_poc[1] += h->delta_poc[1];
2423 }else{
b78a6baa 2424 int poc= 2*(h->frame_num_offset + h->frame_num);
5710b371 2425
b78a6baa
MN
2426 if(!h->nal_ref_idc)
2427 poc--;
5710b371 2428
0da71265
MN
2429 field_poc[0]= poc;
2430 field_poc[1]= poc;
2431 }
115329f1 2432
357282c6 2433 if(s->picture_structure != PICT_BOTTOM_FIELD)
0da71265 2434 s->current_picture_ptr->field_poc[0]= field_poc[0];
357282c6 2435 if(s->picture_structure != PICT_TOP_FIELD)
0da71265 2436 s->current_picture_ptr->field_poc[1]= field_poc[1];
357282c6 2437 cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
0da71265
MN
2438
2439 return 0;
2440}
2441
b41c1db3
2442
2443/**
2444 * initialize scan tables
2445 */
2446static void init_scan_tables(H264Context *h){
b41c1db3 2447 int i;
ca32f7f2 2448 for(i=0; i<16; i++){
b41c1db3 2449#define T(x) (x>>2) | ((x<<2) & 0xF)
ca32f7f2
JGG
2450 h->zigzag_scan[i] = T(zigzag_scan[i]);
2451 h-> field_scan[i] = T( field_scan[i]);
b41c1db3 2452#undef T
b41c1db3 2453 }
ca32f7f2 2454 for(i=0; i<64; i++){
b41c1db3 2455#define T(x) (x>>3) | ((x&7)<<3)
ca32f7f2
JGG
2456 h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2457 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2458 h->field_scan8x8[i] = T(field_scan8x8[i]);
2459 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]);
b41c1db3 2460#undef T
b41c1db3
2461 }
2462 if(h->sps.transform_bypass){ //FIXME same ugly
2463 h->zigzag_scan_q0 = zigzag_scan;
45beb850 2464 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
b41c1db3
2465 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
2466 h->field_scan_q0 = field_scan;
2467 h->field_scan8x8_q0 = field_scan8x8;
2468 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
2469 }else{
2470 h->zigzag_scan_q0 = h->zigzag_scan;
2471 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
2472 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
2473 h->field_scan_q0 = h->field_scan;
2474 h->field_scan8x8_q0 = h->field_scan8x8;
2475 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
2476 }
2477}
afebe2f7 2478
12fe7594 2479static int field_end(H264Context *h, int in_setup){
256299d3
MN
2480 MpegEncContext * const s = &h->s;
2481 AVCodecContext * const avctx= s->avctx;
12fe7594 2482 int err = 0;
256299d3
MN
2483 s->mb_y= 0;
2484
6a9c8594
AS
2485 if (!in_setup && !s->dropable)
2486 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
2487 s->picture_structure==PICT_BOTTOM_FIELD);
256299d3
MN
2488
2489 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2490 ff_vdpau_h264_set_reference_frames(s);
2491
6a9c8594
AS
2492 if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2493 if(!s->dropable) {
12fe7594 2494 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6a9c8594
AS
2495 h->prev_poc_msb= h->poc_msb;
2496 h->prev_poc_lsb= h->poc_lsb;
2497 }
2498 h->prev_frame_num_offset= h->frame_num_offset;
2499 h->prev_frame_num= h->frame_num;
2500 h->outputed_poc = h->next_outputed_poc;
256299d3 2501 }
256299d3
MN
2502
2503 if (avctx->hwaccel) {
2504 if (avctx->hwaccel->end_frame(avctx) < 0)
2505 av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2506 }
2507
2508 if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2509 ff_vdpau_h264_picture_complete(s);
2510
2511 /*
2512 * FIXME: Error handling code does not seem to support interlaced
2513 * when slices span multiple rows
2514 * The ff_er_add_slice calls don't work right for bottom
2515 * fields; they cause massive erroneous error concealing
2516 * Error marking covers both fields (top and bottom).
2517 * This causes a mismatched s->error_count
2518 * and a bad error table. Further, the error count goes to
2519 * INT_MAX when called for bottom field, because mb_y is
2520 * past end by one (callers fault) and resync_mb_y != 0
2521 * causes problems for the first MB line, too.
2522 */
2523 if (!FIELD_PICTURE)
2524 ff_er_frame_end(s);
2525
2526 MPV_frame_end(s);
d225a1e2
MN
2527
2528 h->current_slice=0;
12fe7594
DB
2529
2530 return err;
256299d3
MN
2531}
2532
afebe2f7 2533/**
49bd8e4b 2534 * Replicate H264 "master" context to thread contexts.
afebe2f7
2535 */
2536static void clone_slice(H264Context *dst, H264Context *src)
2537{
2538 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2539 dst->s.current_picture_ptr = src->s.current_picture_ptr;
2540 dst->s.current_picture = src->s.current_picture;
2541 dst->s.linesize = src->s.linesize;
2542 dst->s.uvlinesize = src->s.uvlinesize;
12d96de3 2543 dst->s.first_field = src->s.first_field;
afebe2f7
2544
2545 dst->prev_poc_msb = src->prev_poc_msb;
2546 dst->prev_poc_lsb = src->prev_poc_lsb;
2547 dst->prev_frame_num_offset = src->prev_frame_num_offset;
2548 dst->prev_frame_num = src->prev_frame_num;
2549 dst->short_ref_count = src->short_ref_count;
2550
2551 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2552 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2553 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2554 memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
50c21814
2555
2556 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2557 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
afebe2f7
2558}
2559
0da71265 2560/**
fe9a3fbe
JG
2561 * computes profile from profile_idc and constraint_set?_flags
2562 *
2563 * @param sps SPS
2564 *
2565 * @return profile as defined by FF_PROFILE_H264_*
2566 */
2567int ff_h264_get_profile(SPS *sps)
2568{
2569 int profile = sps->profile_idc;
2570
2571 switch(sps->profile_idc) {
2572 case FF_PROFILE_H264_BASELINE:
2573 // constraint_set1_flag set to 1
2574 profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2575 break;
2576 case FF_PROFILE_H264_HIGH_10:
2577 case FF_PROFILE_H264_HIGH_422:
2578 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2579 // constraint_set3_flag set to 1
2580 profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2581 break;
2582 }
2583
2584 return profile;
2585}
2586
2587/**
0da71265 2588 * decodes a slice header.
9c852bcf 2589 * This will also call MPV_common_init() and frame_start() as needed.
afebe2f7
2590 *
2591 * @param h h264context
2592 * @param h0 h264 master context (differs from 'h' when doing sliced based parallel decoding)
2593 *
d9526386 2594 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
0da71265 2595 */
afebe2f7 2596static int decode_slice_header(H264Context *h, H264Context *h0){
0da71265 2597 MpegEncContext * const s = &h->s;
12d96de3 2598 MpegEncContext * const s0 = &h0->s;
88e7a4d1 2599 unsigned int first_mb_in_slice;
ac658be5 2600 unsigned int pps_id;
0da71265 2601 int num_ref_idx_active_override_flag;
41f5c62f 2602 unsigned int slice_type, tmp, i, j;
0bf79634 2603 int default_ref_list_done = 0;
12d96de3 2604 int last_pic_structure;
0da71265 2605
2f944356 2606 s->dropable= h->nal_ref_idc == 0;
0da71265 2607
9a0dda8b
JGG
2608 /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2609 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
cf653d08
JD
2610 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
2611 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
2612 }else{
2613 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;
2614 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;
2615 }
2616
0da71265
MN
2617 first_mb_in_slice= get_ue_golomb(&s->gb);
2618
d225a1e2
MN
2619 if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2620 if(h0->current_slice && FIELD_PICTURE){
6a9c8594 2621 field_end(h, 1);
d225a1e2
MN
2622 }
2623
afebe2f7 2624 h0->current_slice = 0;
12d96de3 2625 if (!s0->first_field)
f6e3c460 2626 s->current_picture_ptr= NULL;
66a4b2c1
MN
2627 }
2628
9963b332 2629 slice_type= get_ue_golomb_31(&s->gb);
0bf79634 2630 if(slice_type > 9){
9b879566 2631 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 2632 return -1;
0da71265 2633 }
0bf79634
LLL
2634 if(slice_type > 4){
2635 slice_type -= 5;
0da71265
MN
2636 h->slice_type_fixed=1;
2637 }else
2638 h->slice_type_fixed=0;
115329f1 2639
ee2a957f 2640 slice_type= golomb_to_pict_type[ slice_type ];
975a1447 2641 if (slice_type == AV_PICTURE_TYPE_I
afebe2f7 2642 || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
0bf79634
LLL
2643 default_ref_list_done = 1;
2644 }
2645 h->slice_type= slice_type;
e3e6f18f 2646 h->slice_type_nos= slice_type & 3;
0bf79634 2647
1412060e 2648 s->pict_type= h->slice_type; // to make a few old functions happy, it's wrong though
115329f1 2649
0da71265 2650 pps_id= get_ue_golomb(&s->gb);
ac658be5 2651 if(pps_id>=MAX_PPS_COUNT){
9b879566 2652 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
0da71265
MN
2653 return -1;
2654 }
afebe2f7 2655 if(!h0->pps_buffers[pps_id]) {
a0f80050 2656 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
8b92b792
MN
2657 return -1;
2658 }
afebe2f7 2659 h->pps= *h0->pps_buffers[pps_id];
8b92b792 2660
afebe2f7 2661 if(!h0->sps_buffers[h->pps.sps_id]) {
a0f80050 2662 av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
8b92b792
MN
2663 return -1;
2664 }
afebe2f7 2665 h->sps = *h0->sps_buffers[h->pps.sps_id];
239ea04c 2666
fe9a3fbe 2667 s->avctx->profile = ff_h264_get_profile(&h->sps);
fa37cf0d 2668 s->avctx->level = h->sps.level_idc;
6752a3cc 2669 s->avctx->refs = h->sps.ref_frame_count;
b08e38e8 2670
50c21814 2671 if(h == h0 && h->dequant_coeff_pps != pps_id){
50eaa857 2672 h->dequant_coeff_pps = pps_id;
239ea04c
LM
2673 init_dequant_tables(h);
2674 }
115329f1 2675
0da71265 2676 s->mb_width= h->sps.mb_width;
6867a90b 2677 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
115329f1 2678
bf4665ee 2679 h->b_stride= s->mb_width*4;
0da71265 2680
76741b0e
BC
2681 s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2682
c90b9442 2683 s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
0da71265 2684 if(h->sps.frame_mbs_only_flag)
76741b0e 2685 s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
0da71265 2686 else
76741b0e 2687 s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
115329f1
DB
2688
2689 if (s->context_initialized
5388f0b4
JK
2690 && ( s->width != s->avctx->width || s->height != s->avctx->height
2691 || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
6a9c8594
AS
2692 if(h != h0) {
2693 av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
afebe2f7 2694 return -1; // width / height changed during parallelized decoding
6a9c8594 2695 }
91078926 2696 free_tables(h, 0);
ff7f75e1 2697 flush_dpb(s->avctx);
0da71265
MN
2698 MPV_common_end(s);
2699 }
2700 if (!s->context_initialized) {
33aec3f4
MN
2701 if (h != h0) {
2702 av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2703 return -1;
2704 }
f3bdc3da
RD
2705
2706 avcodec_set_dimensions(s->avctx, s->width, s->height);
2707 s->avctx->sample_aspect_ratio= h->sps.sar;
cfa5a81e 2708 av_assert0(s->avctx->sample_aspect_ratio.den);
f3bdc3da 2709
c4dffe7e
DC
2710 if(h->sps.video_signal_type_present_flag){
2711 s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
2712 if(h->sps.colour_description_present_flag){
2713 s->avctx->color_primaries = h->sps.color_primaries;
2714 s->avctx->color_trc = h->sps.color_trc;
2715 s->avctx->colorspace = h->sps.colorspace;
2716 }
2717 }
2718
f3bdc3da 2719 if(h->sps.timing_info_present_flag){
3102d180 2720 int64_t den= h->sps.time_scale;
055a6aa7 2721 if(h->x264_build < 44U)
3102d180 2722 den *= 2;
f3bdc3da 2723 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
3102d180 2724 h->sps.num_units_in_tick, den, 1<<30);
f3bdc3da 2725 }
d545cf80
OA
2726
2727 switch (h->sps.bit_depth_luma) {
2728 case 9 :
76741b0e
BC
2729 if (CHROMA444)
2730 s->avctx->pix_fmt = PIX_FMT_YUV444P9;
2731 else if (CHROMA422)
2732 s->avctx->pix_fmt = PIX_FMT_YUV422P9;
2733 else
2734 s->avctx->pix_fmt = PIX_FMT_YUV420P9;
d545cf80
OA
2735 break;
2736 case 10 :
76741b0e
BC
2737 if (CHROMA444)
2738 s->avctx->pix_fmt = PIX_FMT_YUV444P10;
2739 else if (CHROMA422)
2740 s->avctx->pix_fmt = PIX_FMT_YUV422P10;
2741 else
2742 s->avctx->pix_fmt = PIX_FMT_YUV420P10;
d545cf80
OA
2743 break;
2744 default:
c90b9442
JGG
2745 if (CHROMA444){
2746 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
76741b0e
BC
2747 } else if (CHROMA422) {
2748 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
c90b9442
JGG
2749 }else{
2750 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2751 s->avctx->codec->pix_fmts ?
2752 s->avctx->codec->pix_fmts :
2753 s->avctx->color_range == AVCOL_RANGE_JPEG ?
2754 hwaccel_pixfmt_list_h264_jpeg_420 :
2755 ff_hwaccel_pixfmt_list_420);
2756 }
d545cf80
OA
2757 }
2758
f3bdc3da
RD
2759 s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2760
33aec3f4
MN
2761 if (MPV_common_init(s) < 0) {
2762 av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
0da71265 2763 return -1;
33aec3f4 2764 }
12d96de3 2765 s->first_field = 0;
b19d493f 2766 h->prev_interlaced_frame = 1;
115329f1 2767
b41c1db3 2768 init_scan_tables(h);
bac3ab13
DB
2769 if (ff_h264_alloc_tables(h) < 0) {
2770 av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2771 return AVERROR(ENOMEM);
2772 }
0da71265 2773
6a9c8594 2774 if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
33aec3f4
MN
2775 if (context_init(h) < 0) {
2776 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
afebe2f7 2777 return -1;
33aec3f4 2778 }
6a9c8594
AS
2779 } else {
2780 for(i = 1; i < s->avctx->thread_count; i++) {
2781 H264Context *c;
2782 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2783 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2784 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2785 c->h264dsp = h->h264dsp;
2786 c->sps = h->sps;
2787 c->pps = h->pps;
2788 c->pixel_shift = h->pixel_shift;
2789 init_scan_tables(c);
2790 clone_tables(c, h, i);
2791 }
2792
2793 for(i = 0; i < s->avctx->thread_count; i++)
33aec3f4
MN
2794 if (context_init(h->thread_context[i]) < 0) {
2795 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
6a9c8594 2796 return -1;
33aec3f4 2797 }
6a9c8594 2798 }
0da71265
MN
2799 }
2800
0da71265
MN
2801 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2802
5d18eaad 2803 h->mb_mbaff = 0;
6ba71fc4 2804 h->mb_aff_frame = 0;
12d96de3 2805 last_pic_structure = s0->picture_structure;
0da71265
MN
2806 if(h->sps.frame_mbs_only_flag){
2807 s->picture_structure= PICT_FRAME;
2808 }else{
6ba71fc4 2809 if(get_bits1(&s->gb)) { //field_pic_flag
0da71265 2810 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
6ba71fc4 2811 } else {
0da71265 2812 s->picture_structure= PICT_FRAME;
6ba71fc4 2813 h->mb_aff_frame = h->sps.mb_aff;
6867a90b 2814 }
0da71265 2815 }
44e9dcf1 2816 h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
2ddcf84b
JD
2817
2818 if(h0->current_slice == 0){
3803af22
AS
2819 // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
2820 if(h->frame_num != h->prev_frame_num) {
2821 int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
2822
2823 if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2824
2825 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2826 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2827 if (unwrap_prev_frame_num < 0)
2828 unwrap_prev_frame_num += max_frame_num;
2829
2830 h->prev_frame_num = unwrap_prev_frame_num;
2831 }
2832 }
6a9c8594 2833
26b86e47
MN
2834 while(h->frame_num != h->prev_frame_num &&
2835 h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
4dece8c7 2836 Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
bb943bb8 2837 av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
903d58f6 2838 if (ff_h264_frame_start(h) < 0)
66e6038c 2839 return -1;
26b86e47
MN
2840 h->prev_frame_num++;
2841 h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
2842 s->current_picture_ptr->frame_num= h->prev_frame_num;
6a9c8594
AS
2843 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
2844 ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
3d542120 2845 ff_generate_sliding_window_mmcos(h);
12fe7594 2846 if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
9abc9873 2847 (s->avctx->err_recognition & AV_EF_EXPLODE))
12fe7594 2848 return AVERROR_INVALIDDATA;
e2983d6e
JGG
2849 /* Error concealment: if a ref is missing, copy the previous ref in its place.
2850 * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2851 * about there being no actual duplicates.
2852 * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
2853 * concealing a lost frame, this probably isn't noticable by comparison, but it should
2854 * be fixed. */
4dece8c7
JGG
2855 if (h->short_ref_count) {
2856 if (prev) {
657ccb5a
DB
2857 av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2858 (const uint8_t**)prev->f.data, prev->f.linesize,
5dd7f994 2859 s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
4dece8c7
JGG
2860 h->short_ref[0]->poc = prev->poc+2;
2861 }
2862 h->short_ref[0]->frame_num = h->prev_frame_num;
2863 }
26b86e47
MN
2864 }
2865
12d96de3
JD
2866 /* See if we have a decoded first field looking for a pair... */
2867 if (s0->first_field) {
2868 assert(s0->current_picture_ptr);
657ccb5a 2869 assert(s0->current_picture_ptr->f.data[0]);
95a06eb4 2870 assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
12d96de3
JD
2871
2872 /* figure out if we have a complementary field pair */
2873 if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2874 /*
2875 * Previous field is unmatched. Don't display it, but let it
2876 * remain for reference if marked as such.
2877 */
2878 s0->current_picture_ptr = NULL;
2879 s0->first_field = FIELD_PICTURE;
2880
2881 } else {
2882 if (h->nal_ref_idc &&
657ccb5a 2883 s0->current_picture_ptr->f.reference &&
12d96de3
JD
2884 s0->current_picture_ptr->frame_num != h->frame_num) {
2885 /*
2886 * This and previous field were reference, but had
2887 * different frame_nums. Consider this field first in
2888 * pair. Throw away previous field except for reference
2889 * purposes.
2890 */
2891 s0->first_field = 1;
2892 s0->current_picture_ptr = NULL;
2893
2894 } else {
2895 /* Second field in complementary pair */
2896 s0->first_field = 0;
2897 }
2898 }
2899
2900 } else {
2901 /* Frame or first field in a potentially complementary pair */
2902 assert(!s0->current_picture_ptr);
2903 s0->first_field = FIELD_PICTURE;
2904 }
2905
6a9c8594
AS
2906 if(!FIELD_PICTURE || s0->first_field) {
2907 if (ff_h264_frame_start(h) < 0) {
2908 s0->first_field = 0;
2909 return -1;
2910 }
2911 } else {
2912 ff_release_unused_pictures(s, 0);
12d96de3 2913 }
2ddcf84b
JD
2914 }
2915 if(h != h0)
2916 clone_slice(h, h0);
2917
2918 s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
2919
88e7a4d1 2920 assert(s->mb_num == s->mb_width * s->mb_height);
f3e53d9f 2921 if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
88e7a4d1
MN
2922 first_mb_in_slice >= s->mb_num){
2923 av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
6b53b87e
MN
2924 return -1;
2925 }
88e7a4d1 2926 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
f3e53d9f
JD
2927 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2928 if (s->picture_structure == PICT_BOTTOM_FIELD)
2929 s->resync_mb_y = s->mb_y = s->mb_y + 1;
88e7a4d1 2930 assert(s->mb_y < s->mb_height);
115329f1 2931
0da71265
MN
2932 if(s->picture_structure==PICT_FRAME){
2933 h->curr_pic_num= h->frame_num;
2934 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2935 }else{
f57e2af6 2936 h->curr_pic_num= 2*h->frame_num + 1;
0da71265
MN
2937 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2938 }
115329f1 2939
0da71265 2940 if(h->nal_unit_type == NAL_IDR_SLICE){
1df1df0b 2941 get_ue_golomb(&s->gb); /* idr_pic_id */
0da71265 2942 }
115329f1 2943
0da71265
MN
2944 if(h->sps.poc_type==0){
2945 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
115329f1 2946
0da71265
MN
2947 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2948 h->delta_poc_bottom= get_se_golomb(&s->gb);
2949 }
2950 }
115329f1 2951
0da71265
MN
2952 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2953 h->delta_poc[0]= get_se_golomb(&s->gb);
115329f1 2954
0da71265
MN
2955 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2956 h->delta_poc[1]= get_se_golomb(&s->gb);
2957 }
115329f1 2958
0da71265 2959 init_poc(h);
115329f1 2960
0da71265
MN
2961 if(h->pps.redundant_pic_cnt_present){
2962 h->redundant_pic_count= get_ue_golomb(&s->gb);
2963 }
2964
1412060e 2965 //set defaults, might be overridden a few lines later
0da71265
MN
2966 h->ref_count[0]= h->pps.ref_count[0];
2967 h->ref_count[1]= h->pps.ref_count[1];
2968
975a1447
SS
2969 if(h->slice_type_nos != AV_PICTURE_TYPE_I){
2970 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
0da71265
MN
2971 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2972 }
2973 num_ref_idx_active_override_flag= get_bits1(&s->gb);
115329f1 2974
0da71265
MN
2975 if(num_ref_idx_active_override_flag){
2976 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
975a1447 2977 if(h->slice_type_nos==AV_PICTURE_TYPE_B)
0da71265
MN
2978 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2979
187696fa 2980 if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
9b879566 2981 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
88e7a4d1 2982 h->ref_count[0]= h->ref_count[1]= 1;
0da71265
MN
2983 return -1;
2984 }
2985 }
975a1447 2986 if(h->slice_type_nos == AV_PICTURE_TYPE_B)
187696fa
MN
2987 h->list_count= 2;
2988 else
2989 h->list_count= 1;
2990 }else
2991 h->list_count= 0;
0da71265 2992
0bf79634 2993 if(!default_ref_list_done){
ea6f00c4 2994 ff_h264_fill_default_ref_list(h);
0da71265
MN
2995 }
2996
4c7a232f
LA
2997 if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2998 h->ref_count[1]= h->ref_count[0]= 0;
806bb93f 2999 return -1;
4c7a232f 3000 }
0da71265 3001
975a1447 3002 if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
07dff5c7 3003 s->last_picture_ptr= &h->ref_list[0][0];
8d2fc163 3004 ff_copy_picture(&s->last_picture, s->last_picture_ptr);
07dff5c7 3005 }
975a1447 3006 if(h->slice_type_nos==AV_PICTURE_TYPE_B){
07dff5c7 3007 s->next_picture_ptr= &h->ref_list[1][0];
8d2fc163 3008 ff_copy_picture(&s->next_picture, s->next_picture_ptr);
07dff5c7
MN
3009 }
3010
975a1447
SS
3011 if( (h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P )
3012 || (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
0da71265 3013 pred_weight_table(h);
975a1447 3014 else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
1052b76f 3015 implicit_weight_table(h, -1);
1a29c6a0 3016 }else {
9f2d1b4f 3017 h->use_weight = 0;
cb99c652
GB
3018 for (i = 0; i < 2; i++) {
3019 h->luma_weight_flag[i] = 0;
3020 h->chroma_weight_flag[i] = 0;
3021 }
3022 }
115329f1 3023
12fe7594 3024 if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
9abc9873 3025 (s->avctx->err_recognition & AV_EF_EXPLODE))
12fe7594 3026 return AVERROR_INVALIDDATA;
0da71265 3027
1052b76f 3028 if(FRAME_MBAFF){
ea6f00c4 3029 ff_h264_fill_mbaff_ref_list(h);
5d18eaad 3030
975a1447 3031 if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
1052b76f
MN
3032 implicit_weight_table(h, 0);
3033 implicit_weight_table(h, 1);
3034 }
3035 }
3036
975a1447 3037 if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
943f69a6
MN
3038 ff_h264_direct_dist_scale_factor(h);
3039 ff_h264_direct_ref_list_init(h);
8f56e219 3040
975a1447 3041 if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
9963b332 3042 tmp = get_ue_golomb_31(&s->gb);
88e7a4d1
MN
3043 if(tmp > 2){
3044 av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3045 return -1;
3046 }
3047 h->cabac_init_idc= tmp;
3048 }
e5017ab8
LA
3049
3050 h->last_qscale_diff = 0;
88e7a4d1 3051 tmp = h->pps.init_qp + get_se_golomb(&s->gb);
fcc0224e 3052 if(tmp>51+6*(h->sps.bit_depth_luma-8)){
88e7a4d1 3053 av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3ebc7e04
MN
3054 return -1;
3055 }
88e7a4d1 3056 s->qscale= tmp;
4691a77d
3057 h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3058 h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
0da71265 3059 //FIXME qscale / qp ... stuff
975a1447 3060 if(h->slice_type == AV_PICTURE_TYPE_SP){
1df1df0b 3061 get_bits1(&s->gb); /* sp_for_switch_flag */
0da71265 3062 }
975a1447 3063 if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
1df1df0b 3064 get_se_golomb(&s->gb); /* slice_qs_delta */
0da71265
MN
3065 }
3066
53c05b1e 3067 h->deblocking_filter = 1;
0c32e19d
MN
3068 h->slice_alpha_c0_offset = 52;
3069 h->slice_beta_offset = 52;
0da71265 3070 if( h->pps.deblocking_filter_parameters_present ) {
9963b332 3071 tmp= get_ue_golomb_31(&s->gb);
88e7a4d1
MN
3072 if(tmp > 2){
3073 av_log(s->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp);
3074 return -1;
3075 }
3076 h->deblocking_filter= tmp;
115329f1 3077 if(h->deblocking_filter < 2)
53c05b1e
MN
3078 h->deblocking_filter^= 1; // 1<->0
3079
3080 if( h->deblocking_filter ) {
0c32e19d
MN
3081 h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3082 h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
3083 if( h->slice_alpha_c0_offset > 104U
3084 || h->slice_beta_offset > 104U){
3085 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);
3086 return -1;
3087 }
0da71265 3088 }
980a82b7 3089 }
afebe2f7 3090
61858a76 3091 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL
975a1447
SS
3092 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
3093 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B)
61858a76
RD
3094 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3095 h->deblocking_filter= 0;
3096
afebe2f7 3097 if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
ec970c21
3098 if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3099 /* Cheat slightly for speed:
5d81d641 3100 Do not bother to deblock across slices. */
ec970c21
3101 h->deblocking_filter = 2;
3102 } else {
7ae94d52
3103 h0->max_contexts = 1;
3104 if(!h0->single_decode_warning) {
3105 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3106 h0->single_decode_warning = 1;
3107 }
33aec3f4
MN
3108 if (h != h0) {
3109 av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3110 return 1;
3111 }
ec970c21 3112 }
afebe2f7 3113 }
6a2176aa
JGG
3114 h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
3115 - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
3116 + 6 * (h->sps.bit_depth_luma - 8);
afebe2f7 3117
0da71265
MN
3118#if 0 //FMO
3119 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3120 slice_group_change_cycle= get_bits(&s->gb, ?);
3121#endif
3122
afebe2f7
3123 h0->last_slice_type = slice_type;
3124 h->slice_num = ++h0->current_slice;
b735aeea
MN
3125 if(h->slice_num >= MAX_SLICES){
3126 av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3127 }
5175b937 3128
c32867b5 3129 for(j=0; j<2; j++){
6d7e6b26 3130 int id_list[16];
b735aeea 3131 int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
6d7e6b26
MN
3132 for(i=0; i<16; i++){
3133 id_list[i]= 60;
657ccb5a 3134 if (h->ref_list[j][i].f.data[0]) {
6d7e6b26 3135 int k;
657ccb5a 3136 uint8_t *base = h->ref_list[j][i].f.base[0];
6d7e6b26 3137 for(k=0; k<h->short_ref_count; k++)
657ccb5a 3138 if (h->short_ref[k]->f.base[0] == base) {
6d7e6b26
MN
3139 id_list[i]= k;
3140 break;
3141 }
3142 for(k=0; k<h->long_ref_count; k++)
657ccb5a 3143 if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
6d7e6b26
MN
3144 id_list[i]= h->short_ref_count + k;
3145 break;
3146 }
3147 }
3148 }
3149
c32867b5
MN
3150 ref2frm[0]=
3151 ref2frm[1]= -1;
d50cdd82 3152 for(i=0; i<16; i++)
6d7e6b26 3153 ref2frm[i+2]= 4*id_list[i]
657ccb5a 3154 + (h->ref_list[j][i].f.reference & 3);
d50cdd82
MN
3155 ref2frm[18+0]=
3156 ref2frm[18+1]= -1;
3157 for(i=16; i<48; i++)
6d7e6b26 3158 ref2frm[i+4]= 4*id_list[(i-16)>>1]
657ccb5a 3159 + (h->ref_list[j][i].f.reference & 3);
c32867b5
MN
3160 }
3161
6a9c8594
AS
3162 //FIXME: fix draw_edges+PAFF+frame threads
3163 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
8a11a969 3164 h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
5d18eaad 3165
0da71265 3166 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
49573a87 3167 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
6867a90b
LLL
3168 h->slice_num,
3169 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
115329f1 3170 first_mb_in_slice,
6209669d 3171 av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
0da71265
MN
3172 pps_id, h->frame_num,
3173 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3174 h->ref_count[0], h->ref_count[1],
3175 s->qscale,
0c32e19d 3176 h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
9f2d1b4f 3177 h->use_weight,
4806b922 3178 h->use_weight==1 && h->use_weight_chroma ? "c" : "",
975a1447 3179 h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
0da71265
MN
3180 );
3181 }
3182
3183 return 0;
3184}
3185
0dc343d4 3186int ff_h264_get_slice_type(const H264Context *h)
75dd6938
LA
3187{
3188 switch (h->slice_type) {
975a1447
SS
3189 case AV_PICTURE_TYPE_P: return 0;
3190 case AV_PICTURE_TYPE_B: return 1;
3191 case AV_PICTURE_TYPE_I: return 2;
3192 case AV_PICTURE_TYPE_SP: return 3;
3193 case AV_PICTURE_TYPE_SI: return 4;
75dd6938
LA
3194 default: return -1;
3195 }
3196}
3197
cb546946 3198static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
556f8a06 3199 int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
cb546946
JGG
3200{
3201 int b_stride = h->b_stride;
3202 int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3203 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3204 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3205 if(USES_LIST(top_type, list)){
3206 const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
3207 const int b8_xy= 4*top_xy + 2;
3208 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_x