Support muxing of Zork PCM audio into RIFF formats.
[libav.git] / libavcodec / h264.c
CommitLineData
0da71265
MN
1/*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
b78e7197
DB
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg 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 *
b78e7197 12 * FFmpeg 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
b78e7197 18 * License along with FFmpeg; 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
MN
22/**
23 * @file h264.c
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
0da71265
MN
28#include "dsputil.h"
29#include "avcodec.h"
30#include "mpegvideo.h"
26b4fe82 31#include "h264.h"
0da71265 32#include "h264data.h"
26b4fe82 33#include "h264_parser.h"
0da71265 34#include "golomb.h"
626464fb 35#include "rectangle.h"
0da71265 36
e5017ab8 37#include "cabac.h"
52cb7981
JD
38#ifdef ARCH_X86
39#include "i386/h264_i386.h"
40#endif
e5017ab8 41
2848ce84 42//#undef NDEBUG
0da71265
MN
43#include <assert.h>
44
2ddcf84b
JD
45/**
46 * Value of Picture.reference when Picture is not a reference picture, but
47 * is held for delayed output.
48 */
49#define DELAYED_PIC_REF 4
50
0da71265 51static VLC coeff_token_vlc[4];
910e3668
AC
52static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
53static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
54
0da71265 55static VLC chroma_dc_coeff_token_vlc;
910e3668
AC
56static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
57static const int chroma_dc_coeff_token_vlc_table_size = 256;
0da71265
MN
58
59static VLC total_zeros_vlc[15];
910e3668
AC
60static VLC_TYPE total_zeros_vlc_tables[15][512][2];
61static const int total_zeros_vlc_tables_size = 512;
62
0da71265 63static VLC chroma_dc_total_zeros_vlc[3];
910e3668
AC
64static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
65static const int chroma_dc_total_zeros_vlc_tables_size = 8;
0da71265
MN
66
67static VLC run_vlc[6];
910e3668
AC
68static VLC_TYPE run_vlc_tables[6][8][2];
69static const int run_vlc_tables_size = 8;
70
0da71265 71static VLC run7_vlc;
910e3668
AC
72static VLC_TYPE run7_vlc_table[96][2];
73static const int run7_vlc_table_size = 96;
0da71265 74
8b82a956
MN
75static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
76static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
6ba71fc4 77static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
3e20143e 78static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
9c0e4624 79static Picture * remove_long(H264Context *h, int i, int ref_mask);
8b82a956 80
849f1035 81static av_always_inline uint32_t pack16to32(int a, int b){
377ec888
MN
82#ifdef WORDS_BIGENDIAN
83 return (b&0xFFFF) + (a<<16);
84#else
85 return (a&0xFFFF) + (b<<16);
86#endif
87}
88
acd8d10f
PI
89const uint8_t ff_rem6[52]={
900, 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,
91};
92
93const uint8_t ff_div6[52]={
940, 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,
95};
96
143d7f14
PK
97static const int left_block_options[4][8]={
98 {0,1,2,3,7,10,8,11},
99 {2,2,3,3,8,11,8,11},
100 {0,0,1,1,7,10,7,10},
101 {0,2,0,2,7,10,7,10}
102};
acd8d10f 103
70abb407 104static void fill_caches(H264Context *h, int mb_type, int for_deblock){
0da71265 105 MpegEncContext * const s = &h->s;
64514ee8 106 const int mb_xy= h->mb_xy;
0da71265
MN
107 int topleft_xy, top_xy, topright_xy, left_xy[2];
108 int topleft_type, top_type, topright_type, left_type[2];
143d7f14 109 int * left_block;
02f7695b 110 int topleft_partition= -1;
0da71265
MN
111 int i;
112
36e097bc
JD
113 top_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
114
717b1733 115 //FIXME deblocking could skip the intra and nnz parts.
36e097bc 116 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
e2e5894a
LM
117 return;
118
2cab6401
DB
119 /* Wow, what a mess, why didn't they simplify the interlacing & intra
120 * stuff, I can't imagine that these complex rules are worth it. */
115329f1 121
6867a90b
LLL
122 topleft_xy = top_xy - 1;
123 topright_xy= top_xy + 1;
124 left_xy[1] = left_xy[0] = mb_xy-1;
143d7f14 125 left_block = left_block_options[0];
5d18eaad 126 if(FRAME_MBAFF){
6867a90b
LLL
127 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
128 const int top_pair_xy = pair_xy - s->mb_stride;
129 const int topleft_pair_xy = top_pair_xy - 1;
130 const int topright_pair_xy = top_pair_xy + 1;
131 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
132 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
133 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
134 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
135 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
136 const int bottom = (s->mb_y & 1);
a9c9a240 137 tprintf(s->avctx, "fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
6867a90b
LLL
138 if (bottom
139 ? !curr_mb_frame_flag // bottom macroblock
140 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
141 ) {
142 top_xy -= s->mb_stride;
143 }
144 if (bottom
145 ? !curr_mb_frame_flag // bottom macroblock
146 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
147 ) {
148 topleft_xy -= s->mb_stride;
02f7695b
LM
149 } else if(bottom && curr_mb_frame_flag && !left_mb_frame_flag) {
150 topleft_xy += s->mb_stride;
1412060e 151 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
02f7695b 152 topleft_partition = 0;
6867a90b
LLL
153 }
154 if (bottom
155 ? !curr_mb_frame_flag // bottom macroblock
156 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
157 ) {
158 topright_xy -= s->mb_stride;
159 }
160 if (left_mb_frame_flag != curr_mb_frame_flag) {
161 left_xy[1] = left_xy[0] = pair_xy - 1;
162 if (curr_mb_frame_flag) {
163 if (bottom) {
143d7f14 164 left_block = left_block_options[1];
6867a90b 165 } else {
143d7f14 166 left_block= left_block_options[2];
6867a90b
LLL
167 }
168 } else {
169 left_xy[1] += s->mb_stride;
143d7f14 170 left_block = left_block_options[3];
6867a90b
LLL
171 }
172 }
0da71265
MN
173 }
174
826de46e
LLL
175 h->top_mb_xy = top_xy;
176 h->left_mb_xy[0] = left_xy[0];
177 h->left_mb_xy[1] = left_xy[1];
6ba71fc4 178 if(for_deblock){
717b1733
LM
179 topleft_type = 0;
180 topright_type = 0;
46f2f05f 181 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
46f2f05f
MN
182 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
183 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
5d18eaad
LM
184
185 if(FRAME_MBAFF && !IS_INTRA(mb_type)){
186 int list;
187 int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
188 for(i=0; i<16; i++)
189 h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
3425501d 190 for(list=0; list<h->list_count; list++){
5d18eaad
LM
191 if(USES_LIST(mb_type,list)){
192 uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
193 uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
191e8ca7 194 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
5d18eaad
LM
195 for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
196 dst[0] = src[0];
197 dst[1] = src[1];
198 dst[2] = src[2];
199 dst[3] = src[3];
200 }
201 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
202 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
203 ref += h->b8_stride;
204 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
205 *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
206 }else{
207 fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
208 fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
209 }
210 }
211 }
46f2f05f
MN
212 }else{
213 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
214 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
215 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
216 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
217 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
218 }
0da71265
MN
219
220 if(IS_INTRA(mb_type)){
115329f1
DB
221 h->topleft_samples_available=
222 h->top_samples_available=
0da71265
MN
223 h->left_samples_available= 0xFFFF;
224 h->topright_samples_available= 0xEEEA;
225
226 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
227 h->topleft_samples_available= 0xB3FF;
228 h->top_samples_available= 0x33FF;
229 h->topright_samples_available= 0x26EA;
230 }
231 for(i=0; i<2; i++){
232 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
233 h->topleft_samples_available&= 0xDF5F;
234 h->left_samples_available&= 0x5F5F;
235 }
236 }
115329f1 237
0da71265
MN
238 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
239 h->topleft_samples_available&= 0x7FFF;
115329f1 240
0da71265
MN
241 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
242 h->topright_samples_available&= 0xFBFF;
115329f1 243
0da71265
MN
244 if(IS_INTRA4x4(mb_type)){
245 if(IS_INTRA4x4(top_type)){
246 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
247 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
248 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
249 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
250 }else{
251 int pred;
6fbcaaa0 252 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
0da71265 253 pred= -1;
6fbcaaa0
LLL
254 else{
255 pred= 2;
0da71265
MN
256 }
257 h->intra4x4_pred_mode_cache[4+8*0]=
258 h->intra4x4_pred_mode_cache[5+8*0]=
259 h->intra4x4_pred_mode_cache[6+8*0]=
260 h->intra4x4_pred_mode_cache[7+8*0]= pred;
261 }
262 for(i=0; i<2; i++){
263 if(IS_INTRA4x4(left_type[i])){
264 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
265 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
266 }else{
267 int pred;
6fbcaaa0 268 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
0da71265 269 pred= -1;
6fbcaaa0
LLL
270 else{
271 pred= 2;
0da71265
MN
272 }
273 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
274 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
275 }
276 }
277 }
278 }
115329f1
DB
279
280
0da71265 281/*
115329f1
DB
2820 . T T. T T T T
2831 L . .L . . . .
2842 L . .L . . . .
2853 . T TL . . . .
2864 L . .L . . . .
2875 L . .. . . . .
0da71265 288*/
1412060e 289//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
0da71265 290 if(top_type){
6867a90b
LLL
291 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
292 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
293 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
53c05b1e 294 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
115329f1 295
6867a90b 296 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
53c05b1e 297 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
115329f1 298
6867a90b 299 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
53c05b1e 300 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
115329f1 301
0da71265 302 }else{
115329f1 303 h->non_zero_count_cache[4+8*0]=
0da71265
MN
304 h->non_zero_count_cache[5+8*0]=
305 h->non_zero_count_cache[6+8*0]=
306 h->non_zero_count_cache[7+8*0]=
115329f1 307
0da71265
MN
308 h->non_zero_count_cache[1+8*0]=
309 h->non_zero_count_cache[2+8*0]=
115329f1 310
0da71265 311 h->non_zero_count_cache[1+8*3]=
3981c385 312 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
115329f1 313
0da71265 314 }
826de46e 315
6867a90b
LLL
316 for (i=0; i<2; i++) {
317 if(left_type[i]){
318 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
319 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
320 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
321 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
6867a90b 322 }else{
115329f1
DB
323 h->non_zero_count_cache[3+8*1 + 2*8*i]=
324 h->non_zero_count_cache[3+8*2 + 2*8*i]=
325 h->non_zero_count_cache[0+8*1 + 8*i]=
6867a90b 326 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
826de46e
LLL
327 }
328 }
329
330 if( h->pps.cabac ) {
331 // top_cbp
332 if(top_type) {
333 h->top_cbp = h->cbp_table[top_xy];
334 } else if(IS_INTRA(mb_type)) {
335 h->top_cbp = 0x1C0;
336 } else {
337 h->top_cbp = 0;
338 }
339 // left_cbp
340 if (left_type[0]) {
341 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
342 } else if(IS_INTRA(mb_type)) {
343 h->left_cbp = 0x1C0;
344 } else {
345 h->left_cbp = 0;
346 }
347 if (left_type[0]) {
348 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
349 }
350 if (left_type[1]) {
351 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
6867a90b 352 }
0da71265 353 }
6867a90b 354
0da71265 355#if 1
e2e5894a 356 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
0da71265 357 int list;
3425501d 358 for(list=0; list<h->list_count; list++){
e2e5894a 359 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
0da71265
MN
360 /*if(!h->mv_cache_clean[list]){
361 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
362 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
363 h->mv_cache_clean[list]= 1;
364 }*/
5ad984c9 365 continue;
0da71265
MN
366 }
367 h->mv_cache_clean[list]= 0;
115329f1 368
53b19144 369 if(USES_LIST(top_type, list)){
0da71265
MN
370 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
371 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
372 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
373 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
374 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
375 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
376 h->ref_cache[list][scan8[0] + 0 - 1*8]=
377 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
378 h->ref_cache[list][scan8[0] + 2 - 1*8]=
379 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
380 }else{
115329f1
DB
381 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
382 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
383 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
0da71265
MN
384 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
385 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
386 }
387
4672503d
LM
388 for(i=0; i<2; i++){
389 int cache_idx = scan8[0] - 1 + i*2*8;
390 if(USES_LIST(left_type[i], list)){
391 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
392 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
393 *(uint32_t*)h->mv_cache[list][cache_idx ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
394 *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
395 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
396 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
397 }else{
398 *(uint32_t*)h->mv_cache [list][cache_idx ]=
399 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
400 h->ref_cache[list][cache_idx ]=
401 h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
402 }
0da71265
MN
403 }
404
ae08a563 405 if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
46f2f05f
MN
406 continue;
407
53b19144 408 if(USES_LIST(topleft_type, list)){
02f7695b
LM
409 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
410 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
e2e5894a
LM
411 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
412 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
413 }else{
414 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
415 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
416 }
115329f1 417
53b19144 418 if(USES_LIST(topright_type, list)){
e2e5894a
LM
419 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
420 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
421 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
422 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
423 }else{
424 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
425 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
426 }
e2e5894a 427
ae08a563 428 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
717b1733 429 continue;
115329f1
DB
430
431 h->ref_cache[list][scan8[5 ]+1] =
432 h->ref_cache[list][scan8[7 ]+1] =
3b66c4c5 433 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
115329f1 434 h->ref_cache[list][scan8[4 ]] =
0da71265
MN
435 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
436 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
437 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
3b66c4c5 438 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
0da71265
MN
439 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
440 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
9e528114
LA
441
442 if( h->pps.cabac ) {
443 /* XXX beurk, Load mvd */
53b19144 444 if(USES_LIST(top_type, list)){
9e528114
LA
445 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
446 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
447 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
448 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
449 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
450 }else{
115329f1
DB
451 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
452 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
453 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
9e528114
LA
454 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
455 }
53b19144 456 if(USES_LIST(left_type[0], list)){
9e528114
LA
457 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
458 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
459 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
460 }else{
461 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
462 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
463 }
53b19144 464 if(USES_LIST(left_type[1], list)){
9e528114
LA
465 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
466 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
467 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
468 }else{
469 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
470 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
471 }
472 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
473 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
3b66c4c5 474 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
9e528114
LA
475 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
476 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
5ad984c9 477
9f5c1037 478 if(h->slice_type_nos == FF_B_TYPE){
5ad984c9
LM
479 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
480
481 if(IS_DIRECT(top_type)){
482 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
483 }else if(IS_8X8(top_type)){
484 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
485 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
486 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
487 }else{
488 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
489 }
115329f1 490
5d18eaad
LM
491 if(IS_DIRECT(left_type[0]))
492 h->direct_cache[scan8[0] - 1 + 0*8]= 1;
493 else if(IS_8X8(left_type[0]))
494 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
495 else
496 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
497
498 if(IS_DIRECT(left_type[1]))
5ad984c9 499 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
5d18eaad
LM
500 else if(IS_8X8(left_type[1]))
501 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
502 else
5ad984c9 503 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
5d18eaad
LM
504 }
505 }
506
507 if(FRAME_MBAFF){
508#define MAP_MVS\
509 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
510 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
511 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
512 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
513 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
514 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
515 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
516 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
517 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
518 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
519 if(MB_FIELD){
520#define MAP_F2F(idx, mb_type)\
521 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
522 h->ref_cache[list][idx] <<= 1;\
523 h->mv_cache[list][idx][1] /= 2;\
524 h->mvd_cache[list][idx][1] /= 2;\
525 }
526 MAP_MVS
527#undef MAP_F2F
528 }else{
529#define MAP_F2F(idx, mb_type)\
530 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
531 h->ref_cache[list][idx] >>= 1;\
532 h->mv_cache[list][idx][1] <<= 1;\
533 h->mvd_cache[list][idx][1] <<= 1;\
5ad984c9 534 }
5d18eaad
LM
535 MAP_MVS
536#undef MAP_F2F
5ad984c9 537 }
9e528114 538 }
0da71265 539 }
0da71265
MN
540 }
541#endif
43efd19a
LM
542
543 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
0da71265
MN
544}
545
546static inline void write_back_intra_pred_mode(H264Context *h){
64514ee8 547 const int mb_xy= h->mb_xy;
0da71265
MN
548
549 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
550 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
551 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
552 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
553 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
554 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
555 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
556}
557
558/**
559 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
560 */
561static inline int check_intra4x4_pred_mode(H264Context *h){
562 MpegEncContext * const s = &h->s;
563 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
564 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
565 int i;
115329f1 566
0da71265
MN
567 if(!(h->top_samples_available&0x8000)){
568 for(i=0; i<4; i++){
569 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
570 if(status<0){
9b879566 571 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);
0da71265
MN
572 return -1;
573 } else if(status){
574 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
575 }
576 }
577 }
115329f1 578
0da71265
MN
579 if(!(h->left_samples_available&0x8000)){
580 for(i=0; i<4; i++){
581 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
582 if(status<0){
9b879566 583 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);
0da71265
MN
584 return -1;
585 } else if(status){
586 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
587 }
588 }
589 }
590
591 return 0;
592} //FIXME cleanup like next
593
594/**
595 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
596 */
597static inline int check_intra_pred_mode(H264Context *h, int mode){
598 MpegEncContext * const s = &h->s;
599 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
600 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
115329f1 601
43ff0714 602 if(mode > 6U) {
5175b937 603 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 604 return -1;
5175b937 605 }
115329f1 606
0da71265
MN
607 if(!(h->top_samples_available&0x8000)){
608 mode= top[ mode ];
609 if(mode<0){
9b879566 610 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
611 return -1;
612 }
613 }
115329f1 614
0da71265
MN
615 if(!(h->left_samples_available&0x8000)){
616 mode= left[ mode ];
617 if(mode<0){
9b879566 618 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 619 return -1;
115329f1 620 }
0da71265
MN
621 }
622
623 return mode;
624}
625
626/**
627 * gets the predicted intra4x4 prediction mode.
628 */
629static inline int pred_intra_mode(H264Context *h, int n){
630 const int index8= scan8[n];
631 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
632 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
633 const int min= FFMIN(left, top);
634
a9c9a240 635 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
0da71265
MN
636
637 if(min<0) return DC_PRED;
638 else return min;
639}
640
641static inline void write_back_non_zero_count(H264Context *h){
64514ee8 642 const int mb_xy= h->mb_xy;
0da71265 643
6867a90b
LLL
644 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
645 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
646 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
53c05b1e 647 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
6867a90b
LLL
648 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
649 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
650 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
115329f1 651
6867a90b 652 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
53c05b1e 653 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
6867a90b 654 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
53c05b1e 655
6867a90b 656 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
53c05b1e 657 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
6867a90b 658 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
5d18eaad
LM
659
660 if(FRAME_MBAFF){
661 // store all luma nnzs, for deblocking
662 int v = 0, i;
663 for(i=0; i<16; i++)
664 v += (!!h->non_zero_count_cache[scan8[i]]) << i;
665 *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
666 }
0da71265
MN
667}
668
669/**
1412060e 670 * gets the predicted number of non-zero coefficients.
0da71265
MN
671 * @param n block index
672 */
673static inline int pred_non_zero_count(H264Context *h, int n){
674 const int index8= scan8[n];
675 const int left= h->non_zero_count_cache[index8 - 1];
676 const int top = h->non_zero_count_cache[index8 - 8];
677 int i= left + top;
115329f1 678
0da71265
MN
679 if(i<64) i= (i+1)>>1;
680
a9c9a240 681 tprintf(h->s.avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
0da71265
MN
682
683 return i&31;
684}
685
1924f3ce
MN
686static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
687 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
a9c9a240 688 MpegEncContext *s = &h->s;
1924f3ce 689
5d18eaad
LM
690 /* there is no consistent mapping of mvs to neighboring locations that will
691 * make mbaff happy, so we can't move all this logic to fill_caches */
692 if(FRAME_MBAFF){
191e8ca7 693 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
5d18eaad
LM
694 const int16_t *mv;
695 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
696 *C = h->mv_cache[list][scan8[0]-2];
697
698 if(!MB_FIELD
699 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
700 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
701 if(IS_INTERLACED(mb_types[topright_xy])){
702#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
703 const int x4 = X4, y4 = Y4;\
704 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
02f7695b 705 if(!USES_LIST(mb_type,list))\
5d18eaad
LM
706 return LIST_NOT_USED;\
707 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
708 h->mv_cache[list][scan8[0]-2][0] = mv[0];\
709 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
710 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
711
712 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
713 }
714 }
715 if(topright_ref == PART_NOT_AVAILABLE
716 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
717 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
718 if(!MB_FIELD
719 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
720 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
721 }
722 if(MB_FIELD
723 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
724 && i >= scan8[0]+8){
1412060e 725 // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
02f7695b 726 SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
5d18eaad
LM
727 }
728 }
729#undef SET_DIAG_MV
730 }
731
1924f3ce
MN
732 if(topright_ref != PART_NOT_AVAILABLE){
733 *C= h->mv_cache[list][ i - 8 + part_width ];
734 return topright_ref;
735 }else{
a9c9a240 736 tprintf(s->avctx, "topright MV not available\n");
95c26348 737
1924f3ce
MN
738 *C= h->mv_cache[list][ i - 8 - 1 ];
739 return h->ref_cache[list][ i - 8 - 1 ];
740 }
741}
742
0da71265
MN
743/**
744 * gets the predicted MV.
745 * @param n the block index
746 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
747 * @param mx the x component of the predicted motion vector
748 * @param my the y component of the predicted motion vector
749 */
750static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
0da71265
MN
751 const int index8= scan8[n];
752 const int top_ref= h->ref_cache[list][ index8 - 8 ];
0da71265
MN
753 const int left_ref= h->ref_cache[list][ index8 - 1 ];
754 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
755 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1924f3ce
MN
756 const int16_t * C;
757 int diagonal_ref, match_count;
758
0da71265 759 assert(part_width==1 || part_width==2 || part_width==4);
1924f3ce 760
0da71265 761/* mv_cache
115329f1 762 B . . A T T T T
0da71265
MN
763 U . . L . . , .
764 U . . L . . . .
765 U . . L . . , .
766 . . . L . . . .
767*/
1924f3ce
MN
768
769 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
770 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
a9c9a240 771 tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
1924f3ce
MN
772 if(match_count > 1){ //most common
773 *mx= mid_pred(A[0], B[0], C[0]);
774 *my= mid_pred(A[1], B[1], C[1]);
775 }else if(match_count==1){
776 if(left_ref==ref){
777 *mx= A[0];
115329f1 778 *my= A[1];
1924f3ce
MN
779 }else if(top_ref==ref){
780 *mx= B[0];
115329f1 781 *my= B[1];
0da71265 782 }else{
1924f3ce 783 *mx= C[0];
115329f1 784 *my= C[1];
0da71265
MN
785 }
786 }else{
1924f3ce 787 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
0da71265 788 *mx= A[0];
115329f1 789 *my= A[1];
0da71265 790 }else{
1924f3ce
MN
791 *mx= mid_pred(A[0], B[0], C[0]);
792 *my= mid_pred(A[1], B[1], C[1]);
0da71265 793 }
0da71265 794 }
115329f1 795
a9c9a240 796 tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
797}
798
799/**
800 * gets the directionally predicted 16x8 MV.
801 * @param n the block index
802 * @param mx the x component of the predicted motion vector
803 * @param my the y component of the predicted motion vector
804 */
805static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
806 if(n==0){
807 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
808 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
809
a9c9a240 810 tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
115329f1 811
0da71265
MN
812 if(top_ref == ref){
813 *mx= B[0];
814 *my= B[1];
815 return;
816 }
817 }else{
818 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
819 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
115329f1 820
a9c9a240 821 tprintf(h->s.avctx, "pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
822
823 if(left_ref == ref){
824 *mx= A[0];
825 *my= A[1];
826 return;
827 }
828 }
829
830 //RARE
831 pred_motion(h, n, 4, list, ref, mx, my);
832}
833
834/**
835 * gets the directionally predicted 8x16 MV.
836 * @param n the block index
837 * @param mx the x component of the predicted motion vector
838 * @param my the y component of the predicted motion vector
839 */
840static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
841 if(n==0){
842 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
843 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
115329f1 844
a9c9a240 845 tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
846
847 if(left_ref == ref){
848 *mx= A[0];
849 *my= A[1];
850 return;
851 }
852 }else{
1924f3ce
MN
853 const int16_t * C;
854 int diagonal_ref;
855
856 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
115329f1 857
a9c9a240 858 tprintf(h->s.avctx, "pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265 859
115329f1 860 if(diagonal_ref == ref){
0da71265
MN
861 *mx= C[0];
862 *my= C[1];
863 return;
864 }
0da71265
MN
865 }
866
867 //RARE
868 pred_motion(h, n, 2, list, ref, mx, my);
869}
870
871static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
0da71265
MN
872 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
873 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
874
a9c9a240 875 tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
0da71265
MN
876
877 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
878 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
879 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
115329f1 880
0da71265
MN
881 *mx = *my = 0;
882 return;
883 }
115329f1 884
0da71265
MN
885 pred_motion(h, 0, 4, 0, 0, mx, my);
886
887 return;
888}
889
5ad984c9 890static inline void direct_dist_scale_factor(H264Context * const h){
2879c75f
MN
891 MpegEncContext * const s = &h->s;
892 const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
5ad984c9
LM
893 const int poc1 = h->ref_list[1][0].poc;
894 int i;
895 for(i=0; i<h->ref_count[0]; i++){
896 int poc0 = h->ref_list[0][i].poc;
f66e4f5f 897 int td = av_clip(poc1 - poc0, -128, 127);
4c0d57bd 898 if(td == 0 || h->ref_list[0][i].long_ref){
5ad984c9
LM
899 h->dist_scale_factor[i] = 256;
900 }else{
f66e4f5f 901 int tb = av_clip(poc - poc0, -128, 127);
c26abfa5 902 int tx = (16384 + (FFABS(td) >> 1)) / td;
f66e4f5f 903 h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
5ad984c9
LM
904 }
905 }
5d18eaad
LM
906 if(FRAME_MBAFF){
907 for(i=0; i<h->ref_count[0]; i++){
908 h->dist_scale_factor_field[2*i] =
909 h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
910 }
911 }
5ad984c9 912}
2f944356
LM
913static inline void direct_ref_list_init(H264Context * const h){
914 MpegEncContext * const s = &h->s;
915 Picture * const ref1 = &h->ref_list[1][0];
916 Picture * const cur = s->current_picture_ptr;
917 int list, i, j;
2879c75f 918 int sidx= s->picture_structure&1;
45260d4f 919 int ref1sidx= ref1->reference&1;
2f944356 920 for(list=0; list<2; list++){
2879c75f 921 cur->ref_count[sidx][list] = h->ref_count[list];
2f944356 922 for(j=0; j<h->ref_count[list]; j++)
42de393d 923 cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
2f944356 924 }
7762cc3d
MN
925 if(s->picture_structure == PICT_FRAME){
926 memcpy(cur->ref_count[0], cur->ref_count[1], sizeof(cur->ref_count[0]));
927 memcpy(cur->ref_poc [0], cur->ref_poc [1], sizeof(cur->ref_poc [0]));
928 }
9701840b 929 if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
2f944356
LM
930 return;
931 for(list=0; list<2; list++){
45260d4f 932 for(i=0; i<ref1->ref_count[ref1sidx][list]; i++){
42de393d
MN
933 int poc = ref1->ref_poc[ref1sidx][list][i];
934 if(((poc&3) == 3) != (s->picture_structure == PICT_FRAME))
935 poc= (poc&~3) + s->picture_structure;
171c4076 936 h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
2f944356 937 for(j=0; j<h->ref_count[list]; j++)
42de393d 938 if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
2f944356
LM
939 h->map_col_to_list0[list][i] = j;
940 break;
941 }
942 }
943 }
5d18eaad
LM
944 if(FRAME_MBAFF){
945 for(list=0; list<2; list++){
45260d4f 946 for(i=0; i<ref1->ref_count[ref1sidx][list]; i++){
5d18eaad
LM
947 j = h->map_col_to_list0[list][i];
948 h->map_col_to_list0_field[list][2*i] = 2*j;
949 h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
950 }
951 }
952 }
2f944356 953}
5ad984c9
LM
954
955static inline void pred_direct_motion(H264Context * const h, int *mb_type){
956 MpegEncContext * const s = &h->s;
71041842
MN
957 const int fieldoff= (s->picture_structure & h->ref_list[1][0].reference) ? 0 : (3-2*s->picture_structure);
958 const int mb_xy = h->mb_xy + s->mb_stride*fieldoff;
959 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride + 2*h->b8_stride*fieldoff;
960 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h-> b_stride + 4*h-> b_stride*fieldoff;
5ad984c9
LM
961 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
962 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
4866bd2b 963 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
5ad984c9 964 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
2f944356 965 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
5ad984c9 966 const int is_b8x8 = IS_8X8(*mb_type);
88e7a4d1 967 unsigned int sub_mb_type;
5ad984c9
LM
968 int i8, i4;
969
5d18eaad 970#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
5ad984c9
LM
971 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
972 /* FIXME save sub mb types from previous frames (or derive from MVs)
973 * so we know exactly what block size to use */
974 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
3622988f 975 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5d18eaad 976 }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
5ad984c9
LM
977 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
978 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
979 }else{
980 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
3622988f 981 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5ad984c9
LM
982 }
983 if(!is_b8x8)
984 *mb_type |= MB_TYPE_DIRECT2;
5d18eaad
LM
985 if(MB_FIELD)
986 *mb_type |= MB_TYPE_INTERLACED;
5ad984c9 987
a9c9a240 988 tprintf(s->avctx, "mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
115329f1 989
5ad984c9
LM
990 if(h->direct_spatial_mv_pred){
991 int ref[2];
992 int mv[2][2];
993 int list;
994
5d18eaad
LM
995 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
996
5ad984c9
LM
997 /* ref = min(neighbors) */
998 for(list=0; list<2; list++){
999 int refa = h->ref_cache[list][scan8[0] - 1];
1000 int refb = h->ref_cache[list][scan8[0] - 8];
1001 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
9bec77fe 1002 if(refc == PART_NOT_AVAILABLE)
5ad984c9 1003 refc = h->ref_cache[list][scan8[0] - 8 - 1];
29d05ebc 1004 ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
5ad984c9
LM
1005 if(ref[list] < 0)
1006 ref[list] = -1;
1007 }
1008
1009 if(ref[0] < 0 && ref[1] < 0){
1010 ref[0] = ref[1] = 0;
1011 mv[0][0] = mv[0][1] =
1012 mv[1][0] = mv[1][1] = 0;
1013 }else{
1014 for(list=0; list<2; list++){
1015 if(ref[list] >= 0)
1016 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1017 else
1018 mv[list][0] = mv[list][1] = 0;
1019 }
1020 }
1021
1022 if(ref[1] < 0){
50b3ab0f
LM
1023 if(!is_b8x8)
1024 *mb_type &= ~MB_TYPE_L1;
1025 sub_mb_type &= ~MB_TYPE_L1;
5ad984c9 1026 }else if(ref[0] < 0){
50b3ab0f
LM
1027 if(!is_b8x8)
1028 *mb_type &= ~MB_TYPE_L0;
1029 sub_mb_type &= ~MB_TYPE_L0;
5ad984c9
LM
1030 }
1031
50b3ab0f
LM
1032 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1033 int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1034 int mb_types_col[2];
1035 int b8_stride = h->b8_stride;
1036 int b4_stride = h->b_stride;
1037
1038 *mb_type = (*mb_type & ~MB_TYPE_16x16) | MB_TYPE_8x8;
1039
1040 if(IS_INTERLACED(*mb_type)){
1041 mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1042 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1043 if(s->mb_y&1){
1044 l1ref0 -= 2*b8_stride;
1045 l1ref1 -= 2*b8_stride;
1046 l1mv0 -= 4*b4_stride;
1047 l1mv1 -= 4*b4_stride;
1048 }
1049 b8_stride *= 3;
1050 b4_stride *= 6;
1051 }else{
1052 int cur_poc = s->current_picture_ptr->poc;
1053 int *col_poc = h->ref_list[1]->field_poc;
1054 int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1055 int dy = 2*col_parity - (s->mb_y&1);
1056 mb_types_col[0] =
1057 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy + col_parity*s->mb_stride];
1058 l1ref0 += dy*b8_stride;
1059 l1ref1 += dy*b8_stride;
1060 l1mv0 += 2*dy*b4_stride;
1061 l1mv1 += 2*dy*b4_stride;
1062 b8_stride = 0;
1063 }
1064
1065 for(i8=0; i8<4; i8++){
1066 int x8 = i8&1;
1067 int y8 = i8>>1;
1068 int xy8 = x8+y8*b8_stride;
1069 int xy4 = 3*x8+y8*b4_stride;
1070 int a=0, b=0;
1071
1072 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1073 continue;
1074 h->sub_mb_type[i8] = sub_mb_type;
1075
1076 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1077 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1078 if(!IS_INTRA(mb_types_col[y8])
1079 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1080 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1081 if(ref[0] > 0)
1082 a= pack16to32(mv[0][0],mv[0][1]);
1083 if(ref[1] > 0)
1084 b= pack16to32(mv[1][0],mv[1][1]);
1085 }else{
1086 a= pack16to32(mv[0][0],mv[0][1]);
1087 b= pack16to32(mv[1][0],mv[1][1]);
1088 }
1089 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1090 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1091 }
1092 }else if(IS_16X16(*mb_type)){
d19f5acb
MN
1093 int a=0, b=0;
1094
cec93959
LM
1095 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1096 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
115329f1 1097 if(!IS_INTRA(mb_type_col)
c26abfa5
DB
1098 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1099 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
bf4e3bd2 1100 && (h->x264_build>33 || !h->x264_build)))){
5ad984c9 1101 if(ref[0] > 0)
d19f5acb 1102 a= pack16to32(mv[0][0],mv[0][1]);
5ad984c9 1103 if(ref[1] > 0)
d19f5acb 1104 b= pack16to32(mv[1][0],mv[1][1]);
5ad984c9 1105 }else{
d19f5acb
MN
1106 a= pack16to32(mv[0][0],mv[0][1]);
1107 b= pack16to32(mv[1][0],mv[1][1]);
5ad984c9 1108 }
d19f5acb
MN
1109 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1110 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
5ad984c9
LM
1111 }else{
1112 for(i8=0; i8<4; i8++){
1113 const int x8 = i8&1;
1114 const int y8 = i8>>1;
115329f1 1115
5ad984c9
LM
1116 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1117 continue;
1118 h->sub_mb_type[i8] = sub_mb_type;
115329f1 1119
5ad984c9
LM
1120 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1121 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
cec93959
LM
1122 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1123 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
115329f1 1124
5ad984c9 1125 /* col_zero_flag */
115329f1
DB
1126 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1127 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
bf4e3bd2 1128 && (h->x264_build>33 || !h->x264_build)))){
4866bd2b 1129 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
f1f17e54
LM
1130 if(IS_SUB_8X8(sub_mb_type)){
1131 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
c26abfa5 1132 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
f1f17e54
LM
1133 if(ref[0] == 0)
1134 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1135 if(ref[1] == 0)
1136 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1137 }
1138 }else
5ad984c9 1139 for(i4=0; i4<4; i4++){
4866bd2b 1140 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
c26abfa5 1141 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
5ad984c9
LM
1142 if(ref[0] == 0)
1143 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1144 if(ref[1] == 0)
1145 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1146 }
1147 }
1148 }
1149 }
1150 }
1151 }else{ /* direct temporal mv pred */
5d18eaad
LM
1152 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1153 const int *dist_scale_factor = h->dist_scale_factor;
1154
301e1057 1155 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
5d18eaad
LM
1156 map_col_to_list0[0] = h->map_col_to_list0_field[0];
1157 map_col_to_list0[1] = h->map_col_to_list0_field[1];
1158 dist_scale_factor = h->dist_scale_factor_field;
1159 }
1160 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1161 /* FIXME assumes direct_8x8_inference == 1 */
1162 const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1163 int mb_types_col[2];
8354477e
MN
1164 int b8_stride = h->b8_stride;
1165 int b4_stride = h->b_stride;
5d18eaad 1166 int y_shift;
6c86c44b 1167 int ref_shift;
5d18eaad
LM
1168
1169 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1170 | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1171 | (*mb_type & MB_TYPE_INTERLACED);
1172 sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1173
1174 if(IS_INTERLACED(*mb_type)){
1175 /* frame to field scaling */
1176 mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1177 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1178 if(s->mb_y&1){
1179 l1ref0 -= 2*h->b8_stride;
1180 l1ref1 -= 2*h->b8_stride;
1181 l1mv0 -= 4*h->b_stride;
1182 l1mv1 -= 4*h->b_stride;
1183 }
1184 y_shift = 0;
6c86c44b 1185 ref_shift= FRAME_MBAFF ? 0 : 1;
5d18eaad
LM
1186
1187 if( (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1188 && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1189 && !is_b8x8)
1190 *mb_type |= MB_TYPE_16x8;
1191 else
1192 *mb_type |= MB_TYPE_8x8;
8354477e
MN
1193 b8_stride *= 3;
1194 b4_stride *= 6;
5d18eaad 1195 }else{
f0538030
MN
1196 int cur_poc = s->current_picture_ptr->poc;
1197 int *col_poc = h->ref_list[1]->field_poc;
1198 int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1199 int dy = 2*col_parity - (s->mb_y&1);
5d18eaad 1200 mb_types_col[0] =
f0538030 1201 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy + col_parity*s->mb_stride];
5d18eaad
LM
1202 l1ref0 += dy*h->b8_stride;
1203 l1ref1 += dy*h->b8_stride;
1204 l1mv0 += 2*dy*h->b_stride;
1205 l1mv1 += 2*dy*h->b_stride;
1206 y_shift = 2;
6c86c44b 1207 ref_shift= FRAME_MBAFF ? 2 : 1;
5d18eaad
LM
1208
1209 if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1210 && !is_b8x8)
1211 *mb_type |= MB_TYPE_16x16;
1212 else
1213 *mb_type |= MB_TYPE_8x8;
8354477e 1214 b8_stride = 0;
5d18eaad
LM
1215 }
1216
1217 for(i8=0; i8<4; i8++){
1218 const int x8 = i8&1;
1219 const int y8 = i8>>1;
1220 int ref0, scale;
1221 const int16_t (*l1mv)[2]= l1mv0;
1222
1223 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1224 continue;
1225 h->sub_mb_type[i8] = sub_mb_type;
1226
1227 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1228 if(IS_INTRA(mb_types_col[y8])){
1229 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1230 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1231 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1232 continue;
1233 }
1234
8354477e 1235 ref0 = l1ref0[x8 + y8*b8_stride];
5d18eaad 1236 if(ref0 >= 0)
6c86c44b 1237 ref0 = map_col_to_list0[0][ref0*2>>ref_shift];
5d18eaad 1238 else{
8354477e 1239 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride]*2>>ref_shift];
5d18eaad
LM
1240 l1mv= l1mv1;
1241 }
1242 scale = dist_scale_factor[ref0];
1243 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1244
1245 {
8354477e 1246 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
5d18eaad
LM
1247 int my_col = (mv_col[1]<<y_shift)/2;
1248 int mx = (scale * mv_col[0] + 128) >> 8;
1249 int my = (scale * my_col + 128) >> 8;
1250 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1251 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1252 }
1253 }
1254 return;
1255 }
5d18eaad
LM
1256
1257 /* one-to-one mv scaling */
1258
5ad984c9 1259 if(IS_16X16(*mb_type)){
fda51641
MN
1260 int ref, mv0, mv1;
1261
5ad984c9
LM
1262 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1263 if(IS_INTRA(mb_type_col)){
fda51641 1264 ref=mv0=mv1=0;
5ad984c9 1265 }else{
5d18eaad
LM
1266 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1267 : map_col_to_list0[1][l1ref1[0]];
1268 const int scale = dist_scale_factor[ref0];
8583bef8 1269 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
5ad984c9 1270 int mv_l0[2];
5d18eaad
LM
1271 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1272 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
fda51641
MN
1273 ref= ref0;
1274 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1275 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
5ad984c9 1276 }
fda51641
MN
1277 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1278 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1279 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
5ad984c9
LM
1280 }else{
1281 for(i8=0; i8<4; i8++){
1282 const int x8 = i8&1;
1283 const int y8 = i8>>1;
5d18eaad 1284 int ref0, scale;
bf4e3bd2 1285 const int16_t (*l1mv)[2]= l1mv0;
8583bef8 1286
5ad984c9
LM
1287 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1288 continue;
1289 h->sub_mb_type[i8] = sub_mb_type;
5d18eaad 1290 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
5ad984c9
LM
1291 if(IS_INTRA(mb_type_col)){
1292 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
5ad984c9
LM
1293 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1294 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1295 continue;
1296 }
115329f1 1297
5ad984c9 1298 ref0 = l1ref0[x8 + y8*h->b8_stride];
2f944356 1299 if(ref0 >= 0)
5d18eaad 1300 ref0 = map_col_to_list0[0][ref0];
8583bef8 1301 else{
5d18eaad 1302 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
8583bef8
MN
1303 l1mv= l1mv1;
1304 }
5d18eaad 1305 scale = dist_scale_factor[ref0];
115329f1 1306
5ad984c9 1307 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
f1f17e54
LM
1308 if(IS_SUB_8X8(sub_mb_type)){
1309 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
5d18eaad
LM
1310 int mx = (scale * mv_col[0] + 128) >> 8;
1311 int my = (scale * mv_col[1] + 128) >> 8;
f1f17e54
LM
1312 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1313 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1314 }else
5ad984c9 1315 for(i4=0; i4<4; i4++){
8583bef8 1316 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
5ad984c9 1317 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
5d18eaad
LM
1318 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1319 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
5ad984c9
LM
1320 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1321 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1322 }
1323 }
1324 }
1325 }
1326}
1327
0da71265
MN
1328static inline void write_back_motion(H264Context *h, int mb_type){
1329 MpegEncContext * const s = &h->s;
0da71265
MN
1330 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1331 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1332 int list;
1333
2ea39252
LM
1334 if(!USES_LIST(mb_type, 0))
1335 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1336
3425501d 1337 for(list=0; list<h->list_count; list++){
0da71265 1338 int y;
53b19144 1339 if(!USES_LIST(mb_type, list))
5ad984c9 1340 continue;
115329f1 1341
0da71265
MN
1342 for(y=0; y<4; y++){
1343 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1344 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1345 }
9e528114 1346 if( h->pps.cabac ) {
e6e77eb6
LM
1347 if(IS_SKIP(mb_type))
1348 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1349 else
9e528114
LA
1350 for(y=0; y<4; y++){
1351 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1352 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1353 }
1354 }
53b19144
LM
1355
1356 {
191e8ca7 1357 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
53b19144
LM
1358 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1359 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1360 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1361 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
0da71265
MN
1362 }
1363 }
115329f1 1364
9f5c1037 1365 if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
5ad984c9 1366 if(IS_8X8(mb_type)){
53b19144
LM
1367 uint8_t *direct_table = &h->direct_table[b8_xy];
1368 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1369 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1370 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
5ad984c9
LM
1371 }
1372 }
0da71265
MN
1373}
1374
1375/**
1376 * Decodes a network abstraction layer unit.
1377 * @param consumed is the number of bytes used as input
1378 * @param length is the length of the array
3b66c4c5 1379 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
115329f1 1380 * @returns decoded bytes, might be src+1 if no escapes
0da71265 1381 */
30317501 1382static const uint8_t *decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
0da71265
MN
1383 int i, si, di;
1384 uint8_t *dst;
24456882 1385 int bufidx;
0da71265 1386
bb270c08 1387// src[0]&0x80; //forbidden bit
0da71265
MN
1388 h->nal_ref_idc= src[0]>>5;
1389 h->nal_unit_type= src[0]&0x1F;
1390
1391 src++; length--;
115329f1 1392#if 0
0da71265
MN
1393 for(i=0; i<length; i++)
1394 printf("%2X ", src[i]);
1395#endif
1396 for(i=0; i+1<length; i+=2){
1397 if(src[i]) continue;
1398 if(i>0 && src[i-1]==0) i--;
1399 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1400 if(src[i+2]!=3){
1401 /* startcode, so we must be past the end */
1402 length=i;
1403 }
1404 break;
1405 }
1406 }
1407
1408 if(i>=length-1){ //no escaped 0
1409 *dst_length= length;
1410 *consumed= length+1; //+1 for the header
115329f1 1411 return src;
0da71265
MN
1412 }
1413
24456882
1414 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1415 h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length);
1416 dst= h->rbsp_buffer[bufidx];
0da71265 1417
ac658be5
FOL
1418 if (dst == NULL){
1419 return NULL;
1420 }
1421
3b66c4c5 1422//printf("decoding esc\n");
0da71265 1423 si=di=0;
115329f1 1424 while(si<length){
0da71265
MN
1425 //remove escapes (very rare 1:2^22)
1426 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1427 if(src[si+2]==3){ //escape
1428 dst[di++]= 0;
1429 dst[di++]= 0;
1430 si+=3;
c8470cc1 1431 continue;
0da71265
MN
1432 }else //next start code
1433 break;
1434 }
1435
1436 dst[di++]= src[si++];
1437 }
1438
1439 *dst_length= di;
1440 *consumed= si + 1;//+1 for the header
90b5b51e 1441//FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
0da71265
MN
1442 return dst;
1443}
1444
0da71265
MN
1445/**
1446 * identifies the exact end of the bitstream
1447 * @return the length of the trailing, or 0 if damaged
1448 */
30317501 1449static int decode_rbsp_trailing(H264Context *h, const uint8_t *src){
0da71265
MN
1450 int v= *src;
1451 int r;
1452
a9c9a240 1453 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
0da71265
MN
1454
1455 for(r=1; r<9; r++){
1456 if(v&1) return r;
1457 v>>=1;
1458 }
1459 return 0;
1460}
1461
1462/**
1412060e 1463 * IDCT transforms the 16 dc values and dequantizes them.
0da71265
MN
1464 * @param qp quantization parameter
1465 */
239ea04c 1466static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1467#define stride 16
1468 int i;
1469 int temp[16]; //FIXME check if this is a good idea
1470 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1471 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1472
1473//memset(block, 64, 2*256);
1474//return;
1475 for(i=0; i<4; i++){
1476 const int offset= y_offset[i];
1477 const int z0= block[offset+stride*0] + block[offset+stride*4];
1478 const int z1= block[offset+stride*0] - block[offset+stride*4];
1479 const int z2= block[offset+stride*1] - block[offset+stride*5];
1480 const int z3= block[offset+stride*1] + block[offset+stride*5];
1481
1482 temp[4*i+0]= z0+z3;
1483 temp[4*i+1]= z1+z2;
1484 temp[4*i+2]= z1-z2;
1485 temp[4*i+3]= z0-z3;
1486 }
1487
1488 for(i=0; i<4; i++){
1489 const int offset= x_offset[i];
1490 const int z0= temp[4*0+i] + temp[4*2+i];
1491 const int z1= temp[4*0+i] - temp[4*2+i];
1492 const int z2= temp[4*1+i] - temp[4*3+i];
1493 const int z3= temp[4*1+i] + temp[4*3+i];
1494
1412060e 1495 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
239ea04c
LM
1496 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1497 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1498 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
0da71265
MN
1499 }
1500}
1501
e5017ab8 1502#if 0
0da71265 1503/**
1412060e 1504 * DCT transforms the 16 dc values.
0da71265
MN
1505 * @param qp quantization parameter ??? FIXME
1506 */
1507static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1508// const int qmul= dequant_coeff[qp][0];
1509 int i;
1510 int temp[16]; //FIXME check if this is a good idea
1511 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1512 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1513
1514 for(i=0; i<4; i++){
1515 const int offset= y_offset[i];
1516 const int z0= block[offset+stride*0] + block[offset+stride*4];
1517 const int z1= block[offset+stride*0] - block[offset+stride*4];
1518 const int z2= block[offset+stride*1] - block[offset+stride*5];
1519 const int z3= block[offset+stride*1] + block[offset+stride*5];
1520
1521 temp[4*i+0]= z0+z3;
1522 temp[4*i+1]= z1+z2;
1523 temp[4*i+2]= z1-z2;
1524 temp[4*i+3]= z0-z3;
1525 }
1526
1527 for(i=0; i<4; i++){
1528 const int offset= x_offset[i];
1529 const int z0= temp[4*0+i] + temp[4*2+i];
1530 const int z1= temp[4*0+i] - temp[4*2+i];
1531 const int z2= temp[4*1+i] - temp[4*3+i];
1532 const int z3= temp[4*1+i] + temp[4*3+i];
1533
1534 block[stride*0 +offset]= (z0 + z3)>>1;
1535 block[stride*2 +offset]= (z1 + z2)>>1;
1536 block[stride*8 +offset]= (z1 - z2)>>1;
1537 block[stride*10+offset]= (z0 - z3)>>1;
1538 }
1539}
e5017ab8
LA
1540#endif
1541
0da71265
MN
1542#undef xStride
1543#undef stride
1544
239ea04c 1545static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1546 const int stride= 16*2;
1547 const int xStride= 16;
1548 int a,b,c,d,e;
1549
1550 a= block[stride*0 + xStride*0];
1551 b= block[stride*0 + xStride*1];
1552 c= block[stride*1 + xStride*0];
1553 d= block[stride*1 + xStride*1];
1554
1555 e= a-b;
1556 a= a+b;
1557 b= c-d;
1558 c= c+d;
1559
239ea04c
LM
1560 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1561 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1562 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1563 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
0da71265
MN
1564}
1565
e5017ab8 1566#if 0
0da71265
MN
1567static void chroma_dc_dct_c(DCTELEM *block){
1568 const int stride= 16*2;
1569 const int xStride= 16;
1570 int a,b,c,d,e;
1571
1572 a= block[stride*0 + xStride*0];
1573 b= block[stride*0 + xStride*1];
1574 c= block[stride*1 + xStride*0];
1575 d= block[stride*1 + xStride*1];
1576
1577 e= a-b;
1578 a= a+b;
1579 b= c-d;
1580 c= c+d;
1581
1582 block[stride*0 + xStride*0]= (a+c);
1583 block[stride*0 + xStride*1]= (e+b);
1584 block[stride*1 + xStride*0]= (a-c);
1585 block[stride*1 + xStride*1]= (e-b);
1586}
e5017ab8 1587#endif
0da71265
MN
1588
1589/**
1590 * gets the chroma qp.
1591 */
4691a77d 1592static inline int get_chroma_qp(H264Context *h, int t, int qscale){
5a78bfbd 1593 return h->pps.chroma_qp_table[t][qscale];
0da71265
MN
1594}
1595
2cab6401 1596//FIXME need to check that this does not overflow signed 32 bit for low qp, I am not sure, it's very close
0afd2a92
DB
1597//FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1598static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
0da71265
MN
1599 int i;
1600 const int * const quant_table= quant_coeff[qscale];
1601 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1602 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1603 const unsigned int threshold2= (threshold1<<1);
1604 int last_non_zero;
1605
0afd2a92 1606 if(separate_dc){
0da71265
MN
1607 if(qscale<=18){
1608 //avoid overflows
1609 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1610 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1611 const unsigned int dc_threshold2= (dc_threshold1<<1);
1612
1613 int level= block[0]*quant_coeff[qscale+18][0];
1614 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1615 if(level>0){
1616 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1617 block[0]= level;
1618 }else{
1619 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1620 block[0]= -level;
1621 }
1622// last_non_zero = i;
1623 }else{
1624 block[0]=0;
1625 }
1626 }else{
1627 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1628 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1629 const unsigned int dc_threshold2= (dc_threshold1<<1);
1630
1631 int level= block[0]*quant_table[0];
1632 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1633 if(level>0){
1634 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1635 block[0]= level;
1636 }else{
1637 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1638 block[0]= -level;
1639 }
1640// last_non_zero = i;
1641 }else{
1642 block[0]=0;
1643 }
1644 }
1645 last_non_zero= 0;
1646 i=1;
1647 }else{
1648 last_non_zero= -1;
1649 i=0;
1650 }
1651
1652 for(; i<16; i++){
1653 const int j= scantable[i];
1654 int level= block[j]*quant_table[j];
1655
1656// if( bias+level >= (1<<(QMAT_SHIFT - 3))
1657// || bias-level >= (1<<(QMAT_SHIFT - 3))){
1658 if(((unsigned)(level+threshold1))>threshold2){
1659 if(level>0){
1660 level= (bias + level)>>QUANT_SHIFT;
1661 block[j]= level;
1662 }else{
1663 level= (bias - level)>>QUANT_SHIFT;
1664 block[j]= -level;
1665 }
1666 last_non_zero = i;
1667 }else{
1668 block[j]=0;
1669 }
1670 }
1671
1672 return last_non_zero;
1673}
1674
0da71265
MN
1675static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1676 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1677 int src_x_offset, int src_y_offset,
1678 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1679 MpegEncContext * const s = &h->s;
1680 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
5d18eaad 1681 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
0da71265 1682 const int luma_xy= (mx&3) + ((my&3)<<2);
5d18eaad
LM
1683 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1684 uint8_t * src_cb, * src_cr;
1685 int extra_width= h->emu_edge_width;
1686 int extra_height= h->emu_edge_height;
0da71265
MN
1687 int emu=0;
1688 const int full_mx= mx>>2;
1689 const int full_my= my>>2;
fbd312fd 1690 const int pic_width = 16*s->mb_width;
0d43dd8c 1691 const int pic_height = 16*s->mb_height >> MB_FIELD;
115329f1 1692
1412060e 1693 if(!pic->data[0]) //FIXME this is unacceptable, some sensible error concealment must be done for missing reference frames
171c4076 1694 return;
115329f1 1695
0da71265
MN
1696 if(mx&7) extra_width -= 3;
1697 if(my&7) extra_height -= 3;
115329f1
DB
1698
1699 if( full_mx < 0-extra_width
1700 || full_my < 0-extra_height
1701 || full_mx + 16/*FIXME*/ > pic_width + extra_width
fbd312fd 1702 || full_my + 16/*FIXME*/ > pic_height + extra_height){
5d18eaad
LM
1703 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
1704 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
0da71265
MN
1705 emu=1;
1706 }
115329f1 1707
5d18eaad 1708 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
0da71265 1709 if(!square){
5d18eaad 1710 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
0da71265 1711 }
115329f1 1712
87352549 1713 if(ENABLE_GRAY && s->flags&CODEC_FLAG_GRAY) return;
115329f1 1714
0d43dd8c 1715 if(MB_FIELD){
5d18eaad 1716 // chroma offset when predicting from a field of opposite parity
2143b118 1717 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
5d18eaad
LM
1718 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1719 }
1720 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1721 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1722
0da71265 1723 if(emu){
5d18eaad 1724 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
0da71265
MN
1725 src_cb= s->edge_emu_buffer;
1726 }
5d18eaad 1727 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
0da71265
MN
1728
1729 if(emu){
5d18eaad 1730 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
0da71265
MN
1731 src_cr= s->edge_emu_buffer;
1732 }
5d18eaad 1733 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
0da71265
MN
1734}
1735
9f2d1b4f 1736static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
0da71265
MN
1737 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1738 int x_offset, int y_offset,
1739 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1740 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1741 int list0, int list1){
1742 MpegEncContext * const s = &h->s;
1743 qpel_mc_func *qpix_op= qpix_put;
1744 h264_chroma_mc_func chroma_op= chroma_put;
115329f1 1745
5d18eaad
LM
1746 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1747 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1748 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
0da71265 1749 x_offset += 8*s->mb_x;
0d43dd8c 1750 y_offset += 8*(s->mb_y >> MB_FIELD);
115329f1 1751
0da71265 1752 if(list0){
1924f3ce 1753 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
0da71265
MN
1754 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1755 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1756 qpix_op, chroma_op);
1757
1758 qpix_op= qpix_avg;
1759 chroma_op= chroma_avg;
1760 }
1761
1762 if(list1){
1924f3ce 1763 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
0da71265
MN
1764 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1765 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1766 qpix_op, chroma_op);
1767 }
1768}
1769
9f2d1b4f
LM
1770static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1771 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1772 int x_offset, int y_offset,
1773 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1774 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1775 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1776 int list0, int list1){
1777 MpegEncContext * const s = &h->s;
1778
5d18eaad
LM
1779 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1780 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1781 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
9f2d1b4f 1782 x_offset += 8*s->mb_x;
0d43dd8c 1783 y_offset += 8*(s->mb_y >> MB_FIELD);
115329f1 1784
9f2d1b4f
LM
1785 if(list0 && list1){
1786 /* don't optimize for luma-only case, since B-frames usually
1787 * use implicit weights => chroma too. */
1788 uint8_t *tmp_cb = s->obmc_scratchpad;
5d18eaad
LM
1789 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1790 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
9f2d1b4f
LM
1791 int refn0 = h->ref_cache[0][ scan8[n] ];
1792 int refn1 = h->ref_cache[1][ scan8[n] ];
1793
1794 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1795 dest_y, dest_cb, dest_cr,
1796 x_offset, y_offset, qpix_put, chroma_put);
1797 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1798 tmp_y, tmp_cb, tmp_cr,
1799 x_offset, y_offset, qpix_put, chroma_put);
1800
1801 if(h->use_weight == 2){
1802 int weight0 = h->implicit_weight[refn0][refn1];
1803 int weight1 = 64 - weight0;
5d18eaad
LM
1804 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
1805 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1806 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
9f2d1b4f 1807 }else{
5d18eaad 1808 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
115329f1 1809 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
e8b56208 1810 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
5d18eaad 1811 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
115329f1 1812 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
e8b56208 1813 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
5d18eaad 1814 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
115329f1 1815 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
e8b56208 1816 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
9f2d1b4f
LM
1817 }
1818 }else{
1819 int list = list1 ? 1 : 0;
1820 int refn = h->ref_cache[list][ scan8[n] ];
1821 Picture *ref= &h->ref_list[list][refn];
1822 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1823 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1824 qpix_put, chroma_put);
1825
5d18eaad 1826 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
9f2d1b4f
LM
1827 h->luma_weight[list][refn], h->luma_offset[list][refn]);
1828 if(h->use_weight_chroma){
5d18eaad 1829 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
9f2d1b4f 1830 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
5d18eaad 1831 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
9f2d1b4f
LM
1832 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1833 }
1834 }
1835}
1836
1837static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1838 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1839 int x_offset, int y_offset,
1840 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1841 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
115329f1 1842 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
9f2d1b4f
LM
1843 int list0, int list1){
1844 if((h->use_weight==2 && list0 && list1
1845 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1846 || h->use_weight==1)
1847 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1848 x_offset, y_offset, qpix_put, chroma_put,
1849 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1850 else
1851 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1852 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1853}
1854
513fbd8e
LM
1855static inline void prefetch_motion(H264Context *h, int list){
1856 /* fetch pixels for estimated mv 4 macroblocks ahead
1857 * optimized for 64byte cache lines */
1858 MpegEncContext * const s = &h->s;
1859 const int refn = h->ref_cache[list][scan8[0]];
1860 if(refn >= 0){
1861 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1862 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1863 uint8_t **src= h->ref_list[list][refn].data;
5d18eaad 1864 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
513fbd8e
LM
1865 s->dsp.prefetch(src[0]+off, s->linesize, 4);
1866 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1867 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1868 }
1869}
1870
0da71265
MN
1871static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1872 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f
LM
1873 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1874 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
0da71265 1875 MpegEncContext * const s = &h->s;
64514ee8 1876 const int mb_xy= h->mb_xy;
0da71265 1877 const int mb_type= s->current_picture.mb_type[mb_xy];
115329f1 1878
0da71265 1879 assert(IS_INTER(mb_type));
115329f1 1880
513fbd8e
LM
1881 prefetch_motion(h, 0);
1882
0da71265
MN
1883 if(IS_16X16(mb_type)){
1884 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1885 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
9f2d1b4f 1886 &weight_op[0], &weight_avg[0],
0da71265
MN
1887 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1888 }else if(IS_16X8(mb_type)){
1889 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1890 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 1891 &weight_op[1], &weight_avg[1],
0da71265
MN
1892 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1893 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1894 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 1895 &weight_op[1], &weight_avg[1],
0da71265
MN
1896 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1897 }else if(IS_8X16(mb_type)){
5d18eaad 1898 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 1899 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 1900 &weight_op[2], &weight_avg[2],
0da71265 1901 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
5d18eaad 1902 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
0da71265 1903 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 1904 &weight_op[2], &weight_avg[2],
0da71265
MN
1905 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1906 }else{
1907 int i;
115329f1 1908
0da71265
MN
1909 assert(IS_8X8(mb_type));
1910
1911 for(i=0; i<4; i++){
1912 const int sub_mb_type= h->sub_mb_type[i];
1913 const int n= 4*i;
1914 int x_offset= (i&1)<<2;
1915 int y_offset= (i&2)<<1;
1916
1917 if(IS_SUB_8X8(sub_mb_type)){
1918 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1919 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 1920 &weight_op[3], &weight_avg[3],
0da71265
MN
1921 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1922 }else if(IS_SUB_8X4(sub_mb_type)){
1923 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1924 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 1925 &weight_op[4], &weight_avg[4],
0da71265
MN
1926 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1927 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1928 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 1929 &weight_op[4], &weight_avg[4],
0da71265
MN
1930 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1931 }else if(IS_SUB_4X8(sub_mb_type)){
5d18eaad 1932 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 1933 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 1934 &weight_op[5], &weight_avg[5],
0da71265 1935 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
5d18eaad 1936 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
0da71265 1937 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 1938 &weight_op[5], &weight_avg[5],
0da71265
MN
1939 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1940 }else{
1941 int j;
1942 assert(IS_SUB_4X4(sub_mb_type));
1943 for(j=0; j<4; j++){
1944 int sub_x_offset= x_offset + 2*(j&1);
1945 int sub_y_offset= y_offset + (j&2);
1946 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1947 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 1948 &weight_op[6], &weight_avg[6],
0da71265
MN
1949 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1950 }
1951 }
1952 }
1953 }
513fbd8e
LM
1954
1955 prefetch_motion(h, 1);
0da71265
MN
1956}
1957
98a6fff9 1958static av_cold void decode_init_vlc(void){
0da71265
MN
1959 static int done = 0;
1960
1961 if (!done) {
1962 int i;
910e3668 1963 int offset;
0da71265
MN
1964 done = 1;
1965
910e3668
AC
1966 chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1967 chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
115329f1 1968 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
0da71265 1969 &chroma_dc_coeff_token_len [0], 1, 1,
910e3668
AC
1970 &chroma_dc_coeff_token_bits[0], 1, 1,
1971 INIT_VLC_USE_NEW_STATIC);
0da71265 1972
910e3668 1973 offset = 0;
0da71265 1974 for(i=0; i<4; i++){
910e3668
AC
1975 coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1976 coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
115329f1 1977 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
0da71265 1978 &coeff_token_len [i][0], 1, 1,
910e3668
AC
1979 &coeff_token_bits[i][0], 1, 1,
1980 INIT_VLC_USE_NEW_STATIC);
1981 offset += coeff_token_vlc_tables_size[i];
0da71265 1982 }
910e3668
AC
1983 /*
1984 * This is a one time safety check to make sure that
1985 * the packed static coeff_token_vlc table sizes
1986 * were initialized correctly.
1987 */
1988 assert(offset == sizeof(coeff_token_vlc_tables)/(sizeof(VLC_TYPE)*2));
0da71265
MN
1989
1990 for(i=0; i<3; i++){
910e3668
AC
1991 chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1992 chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1993 init_vlc(&chroma_dc_total_zeros_vlc[i],
1994 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
0da71265 1995 &chroma_dc_total_zeros_len [i][0], 1, 1,
910e3668
AC
1996 &chroma_dc_total_zeros_bits[i][0], 1, 1,
1997 INIT_VLC_USE_NEW_STATIC);
0da71265
MN
1998 }
1999 for(i=0; i<15; i++){
910e3668
AC
2000 total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
2001 total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
2002 init_vlc(&total_zeros_vlc[i],
2003 TOTAL_ZEROS_VLC_BITS, 16,
0da71265 2004 &total_zeros_len [i][0], 1, 1,
910e3668
AC
2005 &total_zeros_bits[i][0], 1, 1,
2006 INIT_VLC_USE_NEW_STATIC);
0da71265
MN
2007 }
2008
2009 for(i=0; i<6; i++){
910e3668
AC
2010 run_vlc[i].table = run_vlc_tables[i];
2011 run_vlc[i].table_allocated = run_vlc_tables_size;
2012 init_vlc(&run_vlc[i],
2013 RUN_VLC_BITS, 7,
0da71265 2014 &run_len [i][0], 1, 1,
910e3668
AC
2015 &run_bits[i][0], 1, 1,
2016 INIT_VLC_USE_NEW_STATIC);
0da71265 2017 }
910e3668
AC
2018 run7_vlc.table = run7_vlc_table,
2019 run7_vlc.table_allocated = run7_vlc_table_size;
115329f1 2020 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
0da71265 2021 &run_len [6][0], 1, 1,
910e3668
AC
2022 &run_bits[6][0], 1, 1,
2023 INIT_VLC_USE_NEW_STATIC);
0da71265
MN
2024 }
2025}
2026
0da71265 2027static void free_tables(H264Context *h){
7978debd 2028 int i;
afebe2f7 2029 H264Context *hx;
0da71265 2030 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
2031 av_freep(&h->chroma_pred_mode_table);
2032 av_freep(&h->cbp_table);
9e528114
LA
2033 av_freep(&h->mvd_table[0]);
2034 av_freep(&h->mvd_table[1]);
5ad984c9 2035 av_freep(&h->direct_table);
0da71265
MN
2036 av_freep(&h->non_zero_count);
2037 av_freep(&h->slice_table_base);
2038 h->slice_table= NULL;
e5017ab8 2039
0da71265
MN
2040 av_freep(&h->mb2b_xy);
2041 av_freep(&h->mb2b8_xy);
9f2d1b4f 2042
7978debd
2043 for(i = 0; i < MAX_SPS_COUNT; i++)
2044 av_freep(h->sps_buffers + i);
2045
2046 for(i = 0; i < MAX_PPS_COUNT; i++)
2047 av_freep(h->pps_buffers + i);
afebe2f7
2048
2049 for(i = 0; i < h->s.avctx->thread_count; i++) {
2050 hx = h->thread_context[i];
2051 if(!hx) continue;
2052 av_freep(&hx->top_borders[1]);
2053 av_freep(&hx->top_borders[0]);
2054 av_freep(&hx->s.obmc_scratchpad);
afebe2f7 2055 }
0da71265
MN
2056}
2057
239ea04c
LM
2058static void init_dequant8_coeff_table(H264Context *h){
2059 int i,q,x;
548a1c8a 2060 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
239ea04c
LM
2061 h->dequant8_coeff[0] = h->dequant8_buffer[0];
2062 h->dequant8_coeff[1] = h->dequant8_buffer[1];
2063
2064 for(i=0; i<2; i++ ){
2065 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2066 h->dequant8_coeff[1] = h->dequant8_buffer[0];
2067 break;
2068 }
2069
2070 for(q=0; q<52; q++){
acd8d10f
PI
2071 int shift = ff_div6[q];
2072 int idx = ff_rem6[q];
239ea04c 2073 for(x=0; x<64; x++)
548a1c8a
LM
2074 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2075 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2076 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
2077 }
2078 }
2079}
2080
2081static void init_dequant4_coeff_table(H264Context *h){
2082 int i,j,q,x;
ab2e3e2c 2083 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
239ea04c
LM
2084 for(i=0; i<6; i++ ){
2085 h->dequant4_coeff[i] = h->dequant4_buffer[i];
2086 for(j=0; j<i; j++){
2087 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2088 h->dequant4_coeff[i] = h->dequant4_buffer[j];
2089 break;
2090 }
2091 }
2092 if(j<i)
2093 continue;
2094
2095 for(q=0; q<52; q++){
acd8d10f
PI
2096 int shift = ff_div6[q] + 2;
2097 int idx = ff_rem6[q];
239ea04c 2098 for(x=0; x<16; x++)
ab2e3e2c
LM
2099 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2100 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
239ea04c
LM
2101 h->pps.scaling_matrix4[i][x]) << shift;
2102 }
2103 }
2104}
2105
2106static void init_dequant_tables(H264Context *h){
2107 int i,x;
2108 init_dequant4_coeff_table(h);
2109 if(h->pps.transform_8x8_mode)
2110 init_dequant8_coeff_table(h);
2111 if(h->sps.transform_bypass){
2112 for(i=0; i<6; i++)
2113 for(x=0; x<16; x++)
2114 h->dequant4_coeff[i][0][x] = 1<<6;
2115 if(h->pps.transform_8x8_mode)
2116 for(i=0; i<2; i++)
2117 for(x=0; x<64; x++)
2118 h->dequant8_coeff[i][0][x] = 1<<6;
2119 }
2120}
2121
2122
0da71265
MN
2123/**
2124 * allocates tables.
3b66c4c5 2125 * needs width/height
0da71265
MN
2126 */
2127static int alloc_tables(H264Context *h){
2128 MpegEncContext * const s = &h->s;
7bc9090a 2129 const int big_mb_num= s->mb_stride * (s->mb_height+1);
239ea04c 2130 int x,y;
0da71265
MN
2131
2132 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
e5017ab8 2133
53c05b1e 2134 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
5d18eaad 2135 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
5d0e4cb8 2136 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
0da71265 2137
7526ade2
MN
2138 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2139 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2140 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2141 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
e5017ab8 2142
5d18eaad
LM
2143 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(uint8_t));
2144 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
0da71265 2145
a55f20bd
LM
2146 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2147 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
0da71265
MN
2148 for(y=0; y<s->mb_height; y++){
2149 for(x=0; x<s->mb_width; x++){
7bc9090a 2150 const int mb_xy= x + y*s->mb_stride;
0da71265
MN
2151 const int b_xy = 4*x + 4*y*h->b_stride;
2152 const int b8_xy= 2*x + 2*y*h->b8_stride;
115329f1 2153
0da71265
MN
2154 h->mb2b_xy [mb_xy]= b_xy;
2155 h->mb2b8_xy[mb_xy]= b8_xy;
2156 }
2157 }
9f2d1b4f 2158
9c6221ae
GV
2159 s->obmc_scratchpad = NULL;
2160
56edbd81
LM
2161 if(!h->dequant4_coeff[0])
2162 init_dequant_tables(h);
2163
0da71265
MN
2164 return 0;
2165fail:
2166 free_tables(h);
2167 return -1;
2168}
2169
afebe2f7
2170/**
2171 * Mimic alloc_tables(), but for every context thread.
2172 */
2173static void clone_tables(H264Context *dst, H264Context *src){
2174 dst->intra4x4_pred_mode = src->intra4x4_pred_mode;
2175 dst->non_zero_count = src->non_zero_count;
2176 dst->slice_table = src->slice_table;
2177 dst->cbp_table = src->cbp_table;
2178 dst->mb2b_xy = src->mb2b_xy;
2179 dst->mb2b8_xy = src->mb2b8_xy;
2180 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
2181 dst->mvd_table[0] = src->mvd_table[0];
2182 dst->mvd_table[1] = src->mvd_table[1];
2183 dst->direct_table = src->direct_table;
2184
afebe2f7
2185 dst->s.obmc_scratchpad = NULL;
2186 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
afebe2f7
2187}
2188
2189/**
2190 * Init context
2191 * Allocate buffers which are not shared amongst multiple threads.
2192 */
2193static int context_init(H264Context *h){
afebe2f7
2194 CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2195 CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2196
afebe2f7
2197 return 0;
2198fail:
2199 return -1; // free_tables will clean up for us
2200}
2201
98a6fff9 2202static av_cold void common_init(H264Context *h){
0da71265 2203 MpegEncContext * const s = &h->s;
0da71265
MN
2204
2205 s->width = s->avctx->width;
2206 s->height = s->avctx->height;
2207 s->codec_id= s->avctx->codec->id;
115329f1 2208
c92a30bb 2209 ff_h264_pred_init(&h->hpc, s->codec_id);
0da71265 2210
239ea04c 2211 h->dequant_coeff_pps= -1;
9a41c2c7 2212 s->unrestricted_mv=1;
0da71265 2213 s->decode=1; //FIXME
56edbd81
LM
2214
2215 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2216 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
0da71265
MN
2217}
2218
98a6fff9 2219static av_cold int decode_init(AVCodecContext *avctx){
0da71265
MN
2220 H264Context *h= avctx->priv_data;
2221 MpegEncContext * const s = &h->s;
2222
3edcacde 2223 MPV_decode_defaults(s);
115329f1 2224
0da71265
MN
2225 s->avctx = avctx;
2226 common_init(h);
2227
2228 s->out_format = FMT_H264;
2229 s->workaround_bugs= avctx->workaround_bugs;
2230
2231 // set defaults
0da71265 2232// s->decode_mb= ff_h263_decode_mb;
9a5a05d0 2233 s->quarter_sample = 1;
0da71265 2234 s->low_delay= 1;
7a9dba3c
MN
2235
2236 if(avctx->codec_id == CODEC_ID_SVQ3)
2237 avctx->pix_fmt= PIX_FMT_YUVJ420P;
2238 else
1d42f410 2239 avctx->pix_fmt= PIX_FMT_YUV420P;
0da71265 2240
c2212338 2241 decode_init_vlc();
115329f1 2242
26165f99
MR
2243 if(avctx->extradata_size > 0 && avctx->extradata &&
2244 *(char *)avctx->extradata == 1){
4770b1b4
RT
2245 h->is_avc = 1;
2246 h->got_avcC = 0;
26165f99
MR
2247 } else {
2248 h->is_avc = 0;
4770b1b4
RT
2249 }
2250
afebe2f7 2251 h->thread_context[0] = h;
18c7be65 2252 h->outputed_poc = INT_MIN;
0da71265
MN
2253 return 0;
2254}
2255
af8aa846 2256static int frame_start(H264Context *h){
0da71265
MN
2257 MpegEncContext * const s = &h->s;
2258 int i;
2259
af8aa846
MN
2260 if(MPV_frame_start(s, s->avctx) < 0)
2261 return -1;
0da71265 2262 ff_er_frame_start(s);
3a22d7fa
JD
2263 /*
2264 * MPV_frame_start uses pict_type to derive key_frame.
2265 * This is incorrect for H.264; IDR markings must be used.
1412060e 2266 * Zero here; IDR markings per slice in frame or fields are ORed in later.
3a22d7fa
JD
2267 * See decode_nal_units().
2268 */
2269 s->current_picture_ptr->key_frame= 0;
0da71265
MN
2270
2271 assert(s->linesize && s->uvlinesize);
2272
2273 for(i=0; i<16; i++){
2274 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
6867a90b 2275 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
2276 }
2277 for(i=0; i<4; i++){
2278 h->block_offset[16+i]=
2279 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
6867a90b
LLL
2280 h->block_offset[24+16+i]=
2281 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
2282 }
2283
934b0821
LM
2284 /* can't be in alloc_tables because linesize isn't known there.
2285 * FIXME: redo bipred weight to not require extra buffer? */
afebe2f7
2286 for(i = 0; i < s->avctx->thread_count; i++)
2287 if(!h->thread_context[i]->s.obmc_scratchpad)
2288 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
5d18eaad
LM
2289
2290 /* some macroblocks will be accessed before they're available */
afebe2f7 2291 if(FRAME_MBAFF || s->avctx->thread_count > 1)
5d18eaad 2292 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
934b0821 2293
0da71265 2294// s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
28bb9eb2 2295
1412060e 2296 // We mark the current picture as non-reference after allocating it, so
28bb9eb2
MN
2297 // that if we break out due to an error it can be released automatically
2298 // in the next MPV_frame_start().
2299 // SVQ3 as well as most other codecs have only last/next/current and thus
2300 // get released even with set reference, besides SVQ3 and others do not
2301 // mark frames as reference later "naturally".
2302 if(s->codec_id != CODEC_ID_SVQ3)
2303 s->current_picture_ptr->reference= 0;
357282c6
MN
2304
2305 s->current_picture_ptr->field_poc[0]=
2306 s->current_picture_ptr->field_poc[1]= INT_MAX;
5118c6c7 2307 assert(s->current_picture_ptr->long_ref==0);
357282c6 2308
af8aa846 2309 return 0;
0da71265
MN
2310}
2311
93cc10fa 2312static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
53c05b1e
MN
2313 MpegEncContext * const s = &h->s;
2314 int i;
115329f1 2315
53c05b1e
MN
2316 src_y -= linesize;
2317 src_cb -= uvlinesize;
2318 src_cr -= uvlinesize;
2319
3b66c4c5 2320 // There are two lines saved, the line above the the top macroblock of a pair,
6867a90b 2321 // and the line above the bottom macroblock
6ba71fc4 2322 h->left_border[0]= h->top_borders[0][s->mb_x][15];
53c05b1e
MN
2323 for(i=1; i<17; i++){
2324 h->left_border[i]= src_y[15+i* linesize];
2325 }
115329f1 2326
6ba71fc4
LLL
2327 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2328 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
53c05b1e 2329
87352549 2330 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6ba71fc4
LLL
2331 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
2332 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
53c05b1e
MN
2333 for(i=1; i<9; i++){
2334 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2335 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2336 }
6ba71fc4
LLL
2337 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2338 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
53c05b1e
MN
2339 }
2340}
2341
93cc10fa 2342static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
53c05b1e
MN
2343 MpegEncContext * const s = &h->s;
2344 int temp8, i;
2345 uint64_t temp64;
b69378e2
2346 int deblock_left;
2347 int deblock_top;
2348 int mb_xy;
2349
2350 if(h->deblocking_filter == 2) {
64514ee8 2351 mb_xy = h->mb_xy;
b69378e2
2352 deblock_left = h->slice_table[mb_xy] == h->slice_table[mb_xy - 1];
2353 deblock_top = h->slice_table[mb_xy] == h->slice_table[h->top_mb_xy];
2354 } else {
2355 deblock_left = (s->mb_x > 0);
2356 deblock_top = (s->mb_y > 0);
2357 }
53c05b1e
MN
2358
2359 src_y -= linesize + 1;
2360 src_cb -= uvlinesize + 1;
2361 src_cr -= uvlinesize + 1;
2362
2363#define XCHG(a,b,t,xchg)\
2364t= a;\
2365if(xchg)\
2366 a= b;\
2367b= t;
d89dc06a
LM
2368
2369 if(deblock_left){
2370 for(i = !deblock_top; i<17; i++){
2371 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2372 }
2373 }
2374
2375 if(deblock_top){
6ba71fc4
LLL
2376 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2377 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
cad4368a 2378 if(s->mb_x+1 < s->mb_width){
43efd19a
LM
2379 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2380 }
53c05b1e 2381 }
53c05b1e 2382
87352549 2383 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
d89dc06a
LM
2384 if(deblock_left){
2385 for(i = !deblock_top; i<9; i++){
2386 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2387 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2388 }
2389 }
2390 if(deblock_top){
6ba71fc4
LLL
2391 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2392 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2393 }
2394 }
2395}
2396
2397static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2398 MpegEncContext * const s = &h->s;
2399 int i;
115329f1 2400
6ba71fc4
LLL
2401 src_y -= 2 * linesize;
2402 src_cb -= 2 * uvlinesize;
2403 src_cr -= 2 * uvlinesize;
2404
3b66c4c5 2405 // There are two lines saved, the line above the the top macroblock of a pair,
6ba71fc4
LLL
2406 // and the line above the bottom macroblock
2407 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2408 h->left_border[1]= h->top_borders[1][s->mb_x][15];
2409 for(i=2; i<34; i++){
2410 h->left_border[i]= src_y[15+i* linesize];
2411 }
115329f1 2412
6ba71fc4
LLL
2413 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
2414 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2415 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
2416 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2417
87352549 2418 if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6ba71fc4
LLL
2419 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
2420 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
2421 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
2422 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2423 for(i=2; i<18; i++){
2424 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
2425 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2426 }
2427 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2428 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2429 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2430 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2431 }
2432}
2433
2434static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2435 MpegEncContext * const s = &h->s;
2436 int temp8, i;
2437 uint64_t temp64;
2438 int deblock_left = (s->mb_x > 0);
5d18eaad 2439 int deblock_top = (s->mb_y > 1);
6ba71fc4 2440
a9c9a240 2441 tprintf(s->avctx, "xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
6ba71fc4
LLL
2442
2443 src_y -= 2 * linesize + 1;
2444 src_cb -= 2 * uvlinesize + 1;
2445 src_cr -= 2 * uvlinesize + 1;
2446
2447#define XCHG(a,b,t,xchg)\
2448t= a;\
2449if(xchg)\
2450 a= b;\
2451b= t;
2452
2453 if(deblock_left){
2454 for(i = (!deblock_top)<<1; i<34; i++){
2455 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2456 }
2457 }
2458
2459 if(deblock_top){
2460 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2461 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2462 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2463 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
5d18eaad
LM
2464 if(s->mb_x+1 < s->mb_width){
2465 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2466 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);
2467 }
6ba71fc4
LLL
2468 }
2469
87352549 2470 if(!ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
6ba71fc4
LLL
2471 if(deblock_left){
2472 for(i = (!deblock_top) << 1; i<18; i++){
2473 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
2474 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2475 }
2476 }
2477 if(deblock_top){
2478 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2479 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2480 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2481 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
53c05b1e 2482 }
53c05b1e
MN
2483 }
2484}
2485
5a6a6cc7 2486static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
0da71265
MN
2487 MpegEncContext * const s = &h->s;
2488 const int mb_x= s->mb_x;
2489 const int mb_y= s->mb_y;
64514ee8 2490 const int mb_xy= h->mb_xy;
0da71265
MN
2491 const int mb_type= s->current_picture.mb_type[mb_xy];
2492 uint8_t *dest_y, *dest_cb, *dest_cr;
2493 int linesize, uvlinesize /*dct_offset*/;
2494 int i;
6867a90b 2495 int *block_offset = &h->block_offset[0];
6ba71fc4 2496 const unsigned int bottom = mb_y & 1;
bd91fee3 2497 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass), is_h264 = (simple || s->codec_id == CODEC_ID_H264);
36940eca 2498 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15 2499 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
0da71265 2500
0da71265
MN
2501 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2502 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2503 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2504
a957c27b
LM
2505 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2506 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2507
bd91fee3 2508 if (!simple && MB_FIELD) {
5d18eaad
LM
2509 linesize = h->mb_linesize = s->linesize * 2;
2510 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
6867a90b 2511 block_offset = &h->block_offset[24];
1412060e 2512 if(mb_y&1){ //FIXME move out of this function?
0da71265 2513 dest_y -= s->linesize*15;
6867a90b
LLL
2514 dest_cb-= s->uvlinesize*7;
2515 dest_cr-= s->uvlinesize*7;
0da71265 2516 }
5d18eaad
LM
2517 if(FRAME_MBAFF) {
2518 int list;
3425501d 2519 for(list=0; list<h->list_count; list++){
5d18eaad
LM
2520 if(!USES_LIST(mb_type, list))
2521 continue;
2522 if(IS_16X16(mb_type)){
2523 int8_t *ref = &h->ref_cache[list][scan8[0]];
1710856c 2524 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
5d18eaad
LM
2525 }else{
2526 for(i=0; i<16; i+=4){
2527 //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?
2528 int ref = h->ref_cache[list][scan8[i]];
2529 if(ref >= 0)
1710856c 2530 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
5d18eaad
LM
2531 }
2532 }
2533 }
2534 }
0da71265 2535 } else {
5d18eaad
LM
2536 linesize = h->mb_linesize = s->linesize;
2537 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
0da71265
MN
2538// dct_offset = s->linesize * 16;
2539 }
115329f1 2540
ef9d1d15
LM
2541 if(transform_bypass){
2542 idct_dc_add =
2543 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
2544 }else if(IS_8x8DCT(mb_type)){
2545 idct_dc_add = s->dsp.h264_idct8_dc_add;
2546 idct_add = s->dsp.h264_idct8_add;
2547 }else{
2548 idct_dc_add = s->dsp.h264_idct_dc_add;
2549 idct_add = s->dsp.h264_idct_add;
2550 }
0da71265 2551
bd91fee3 2552 if(!simple && FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)
5d18eaad
LM
2553 && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){
2554 int mbt_y = mb_y&~1;
2555 uint8_t *top_y = s->current_picture.data[0] + (mbt_y * 16* s->linesize ) + mb_x * 16;
2556 uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2557 uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;
2558 xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);
2559 }
2560
bd91fee3 2561 if (!simple && IS_INTRA_PCM(mb_type)) {
c1708e8d
MN
2562 for (i=0; i<16; i++) {
2563 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
6fbcaaa0 2564 }
c1708e8d
MN
2565 for (i=0; i<8; i++) {
2566 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
2567 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
6fbcaaa0 2568 }
e7e09b49
LLL
2569 } else {
2570 if(IS_INTRA(mb_type)){
bd91fee3 2571 if(h->deblocking_filter && (simple || !FRAME_MBAFF))
93cc10fa 2572 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
53c05b1e 2573
87352549 2574 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
c92a30bb
KS
2575 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2576 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
e7e09b49 2577 }
0da71265 2578
e7e09b49 2579 if(IS_INTRA4x4(mb_type)){
bd91fee3 2580 if(simple || !s->encoding){
43efd19a
LM
2581 if(IS_8x8DCT(mb_type)){
2582 for(i=0; i<16; i+=4){
2583 uint8_t * const ptr= dest_y + block_offset[i];
2584 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
ef9d1d15 2585 const int nnz = h->non_zero_count_cache[ scan8[i] ];
c92a30bb 2586 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
4672503d 2587 (h->topright_samples_available<<i)&0x4000, linesize);
ef9d1d15
LM
2588 if(nnz){
2589 if(nnz == 1 && h->mb[i*16])
2590 idct_dc_add(ptr, h->mb + i*16, linesize);
2591 else
2592 idct_add(ptr, h->mb + i*16, linesize);
2593 }
43efd19a
LM
2594 }
2595 }else
e7e09b49 2596 for(i=0; i<16; i++){
6867a90b 2597 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
2598 uint8_t *topright;
2599 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
ef9d1d15 2600 int nnz, tr;
e7e09b49
LLL
2601
2602 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2603 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
6867a90b 2604 assert(mb_y || linesize <= block_offset[i]);
e7e09b49
LLL
2605 if(!topright_avail){
2606 tr= ptr[3 - linesize]*0x01010101;
2607 topright= (uint8_t*) &tr;
115329f1 2608 }else
e7e09b49 2609 topright= ptr + 4 - linesize;
a9799653 2610 }else
e7e09b49
LLL
2611 topright= NULL;
2612
c92a30bb 2613 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
ef9d1d15
LM
2614 nnz = h->non_zero_count_cache[ scan8[i] ];
2615 if(nnz){
bd91fee3 2616 if(is_h264){
ef9d1d15
LM
2617 if(nnz == 1 && h->mb[i*16])
2618 idct_dc_add(ptr, h->mb + i*16, linesize);
2619 else
2620 idct_add(ptr, h->mb + i*16, linesize);
2621 }else
e7e09b49
LLL
2622 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2623 }
8b82a956 2624 }
0da71265 2625 }
e7e09b49 2626 }else{
c92a30bb 2627 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
bd91fee3 2628 if(is_h264){
36940eca 2629 if(!transform_bypass)
93f0c0a4 2630 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[0][s->qscale][0]);
36940eca 2631 }else
e7e09b49 2632 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
0da71265 2633 }
bd91fee3 2634 if(h->deblocking_filter && (simple || !FRAME_MBAFF))
93cc10fa 2635 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
bd91fee3 2636 }else if(is_h264){
e7e09b49 2637 hl_motion(h, dest_y, dest_cb, dest_cr,
2833fc46
LM
2638 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
2639 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
e7e09b49 2640 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
0da71265 2641 }
e7e09b49
LLL
2642
2643
2644 if(!IS_INTRA4x4(mb_type)){
bd91fee3 2645 if(is_h264){
ef9d1d15
LM
2646 if(IS_INTRA16x16(mb_type)){
2647 for(i=0; i<16; i++){
2648 if(h->non_zero_count_cache[ scan8[i] ])
2649 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2650 else if(h->mb[i*16])
2651 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2652 }
2653 }else{
2654 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
2655 for(i=0; i<16; i+=di){
2656 int nnz = h->non_zero_count_cache[ scan8[i] ];
2657 if(nnz){
2658 if(nnz==1 && h->mb[i*16])
2659 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2660 else
2661 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
2662 }
e7e09b49 2663 }
4704097a 2664 }
e7e09b49
LLL
2665 }else{
2666 for(i=0; i<16; i++){
2667 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
6867a90b 2668 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
2669 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2670 }
4704097a 2671 }
0da71265
MN
2672 }
2673 }
0da71265 2674
87352549 2675 if(simple || !ENABLE_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
ef9d1d15
LM
2676 uint8_t *dest[2] = {dest_cb, dest_cr};
2677 if(transform_bypass){
2678 idct_add = idct_dc_add = s->dsp.add_pixels4;
2679 }else{
2680 idct_add = s->dsp.h264_idct_add;
2681 idct_dc_add = s->dsp.h264_idct_dc_add;
4691a77d
2682 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp[0], h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
2683 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp[1], h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
36940eca 2684 }
bd91fee3 2685 if(is_h264){
ef9d1d15
LM
2686 for(i=16; i<16+8; i++){
2687 if(h->non_zero_count_cache[ scan8[i] ])
2688 idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
2689 else if(h->mb[i*16])
2690 idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
4704097a 2691 }
e7e09b49 2692 }else{
ef9d1d15 2693 for(i=16; i<16+8; i++){
e7e09b49 2694 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
ef9d1d15 2695 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
e7e09b49
LLL
2696 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2697 }
4704097a 2698 }
0da71265
MN
2699 }
2700 }
2701 }
53c05b1e 2702 if(h->deblocking_filter) {
bd91fee3 2703 if (!simple && FRAME_MBAFF) {
5d18eaad
LM
2704 //FIXME try deblocking one mb at a time?
2705 // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border
6ba71fc4
LLL
2706 const int mb_y = s->mb_y - 1;
2707 uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
2708 const int mb_xy= mb_x + mb_y*s->mb_stride;
2709 const int mb_type_top = s->current_picture.mb_type[mb_xy];
2710 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
6ba71fc4
LLL
2711 if (!bottom) return;
2712 pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2713 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2714 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2715
5d18eaad
LM
2716 if(IS_INTRA(mb_type_top | mb_type_bottom))
2717 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
2718
6ba71fc4 2719 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
5d18eaad 2720 // deblock a pair
115329f1 2721 // top
64514ee8 2722 s->mb_y--; h->mb_xy -= s->mb_stride;
a9c9a240 2723 tprintf(h->s.avctx, "call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
3b66c4c5 2724 fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
4691a77d
2725 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2726 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
6ba71fc4 2727 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
6ba71fc4 2728 // bottom
64514ee8 2729 s->mb_y++; h->mb_xy += s->mb_stride;
a9c9a240 2730 tprintf(h->s.avctx, "call mbaff filter_mb\n");
3b66c4c5 2731 fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
4691a77d
2732 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
2733 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy+s->mb_stride]);
6ba71fc4 2734 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
6ba71fc4 2735 } else {
a9c9a240 2736 tprintf(h->s.avctx, "call filter_mb\n");
93cc10fa 2737 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, simple);
3b66c4c5 2738 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
a82688b0
MN
2739 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
2740 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
3e20143e 2741 filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
6ba71fc4 2742 }
53c05b1e 2743 }
0da71265
MN
2744}
2745
0da71265 2746/**
bd91fee3
AS
2747 * Process a macroblock; this case avoids checks for expensive uncommon cases.
2748 */
2749static void hl_decode_mb_simple(H264Context *h){
2750 hl_decode_mb_internal(h, 1);
2751}
2752
2753/**
2754 * Process a macroblock; this handles edge cases, such as interlacing.
2755 */
2756static void av_noinline hl_decode_mb_complex(H264Context *h){
2757 hl_decode_mb_internal(h, 0);
2758}
2759
2760static void hl_decode_mb(H264Context *h){
2761 MpegEncContext * const s = &h->s;
64514ee8 2762 const int mb_xy= h->mb_xy;
bd91fee3 2763 const int mb_type= s->current_picture.mb_type[mb_xy];
58cc7dd9
AS
2764 int is_complex = FRAME_MBAFF || MB_FIELD || IS_INTRA_PCM(mb_type) || s->codec_id != CODEC_ID_H264 ||
2765 (ENABLE_GRAY && (s->flags&CODEC_FLAG_GRAY)) || (ENABLE_H264_ENCODER && s->encoding) || ENABLE_SMALL;
bd91fee3 2766
fedec603 2767 if(ENABLE_H264_ENCODER && !s->decode)
bd91fee3
AS
2768 return;
2769
2770 if (is_complex)
2771 hl_decode_mb_complex(h);
2772 else hl_decode_mb_simple(h);
2773}
2774
2143b118 2775static void pic_as_field(Picture *pic, const int parity){
11cc1d8c
JD
2776 int i;
2777 for (i = 0; i < 4; ++i) {
2143b118 2778 if (parity == PICT_BOTTOM_FIELD)
11cc1d8c 2779 pic->data[i] += pic->linesize[i];
2143b118 2780 pic->reference = parity;
11cc1d8c
JD
2781 pic->linesize[i] *= 2;
2782 }
2879c75f 2783 pic->poc= pic->field_poc[parity == PICT_BOTTOM_FIELD];
11cc1d8c
JD
2784}
2785
2786static int split_field_copy(Picture *dest, Picture *src,
2787 int parity, int id_add){
2788 int match = !!(src->reference & parity);
2789
2790 if (match) {
2791 *dest = *src;
d4f7d838 2792 if(parity != PICT_FRAME){
b3e93fd4
MN
2793 pic_as_field(dest, parity);
2794 dest->pic_id *= 2;
2795 dest->pic_id += id_add;
d4f7d838 2796 }
11cc1d8c
JD
2797 }
2798
2799 return match;
2800}
2801
d4f7d838
MN
2802static int build_def_list(Picture *def, Picture **in, int len, int is_long, int sel){
2803 int i[2]={0};
2804 int index=0;
11cc1d8c 2805
d4f7d838
MN
2806 while(i[0]<len || i[1]<len){
2807 while(i[0]<len && !(in[ i[0] ] && (in[ i[0] ]->reference & sel)))
2808 i[0]++;
2809 while(i[1]<len && !(in[ i[1] ] && (in[ i[1] ]->reference & (sel^3))))
2810 i[1]++;
2811 if(i[0] < len){
2812 in[ i[0] ]->pic_id= is_long ? i[0] : in[ i[0] ]->frame_num;
2813 split_field_copy(&def[index++], in[ i[0]++ ], sel , 1);
2814 }
2815 if(i[1] < len){
2816 in[ i[1] ]->pic_id= is_long ? i[1] : in[ i[1] ]->frame_num;
2817 split_field_copy(&def[index++], in[ i[1]++ ], sel^3, 0);
11cc1d8c
JD
2818 }
2819 }
2820
d4f7d838 2821 return index;
11cc1d8c
JD
2822}
2823
d4f7d838
MN
2824static int add_sorted(Picture **sorted, Picture **src, int len, int limit, int dir){
2825 int i, best_poc;
2826 int out_i= 0;
11cc1d8c 2827
d4f7d838
MN
2828 for(;;){
2829 best_poc= dir ? INT_MIN : INT_MAX;
11cc1d8c 2830
d4f7d838
MN
2831 for(i=0; i<len; i++){
2832 const int poc= src[i]->poc;
2833 if(((poc > limit) ^ dir) && ((poc < best_poc) ^ dir)){
2834 best_poc= poc;
2835 sorted[out_i]= src[i];
2836 }
2837 }
2838 if(best_poc == (dir ? INT_MIN : INT_MAX))
2839 break;
2840 limit= sorted[out_i++]->poc - dir;
2841 }
2842 return out_i;
11cc1d8c
JD
2843}
2844
bd91fee3 2845/**
0da71265
MN
2846 * fills the default_ref_list.
2847 */
2848static int fill_default_ref_list(H264Context *h){
2849 MpegEncContext * const s = &h->s;
d4f7d838 2850 int i, len;
115329f1 2851
9f5c1037 2852 if(h->slice_type_nos==FF_B_TYPE){
d4f7d838
MN
2853 Picture *sorted[32];
2854 int cur_poc, list;
2855 int lens[2];
11cc1d8c 2856
d4f7d838
MN
2857 if(FIELD_PICTURE)
2858 cur_poc= s->current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
2859 else
2860 cur_poc= s->current_picture_ptr->poc;
086acdd5 2861
d4f7d838
MN
2862 for(list= 0; list<2; list++){
2863 len= add_sorted(sorted , h->short_ref, h->short_ref_count, cur_poc, 1^list);
2864 len+=add_sorted(sorted+len, h->short_ref, h->short_ref_count, cur_poc, 0^list);
2865 assert(len<=32);
2866 len= build_def_list(h->default_ref_list[list] , sorted , len, 0, s->picture_structure);
2867 len+=build_def_list(h->default_ref_list[list]+len, h->long_ref, 16 , 1, s->picture_structure);
2868 assert(len<=32);
086acdd5 2869
d4f7d838
MN
2870 if(len < h->ref_count[list])
2871 memset(&h->default_ref_list[list][len], 0, sizeof(Picture)*(h->ref_count[list] - len));
2872 lens[list]= len;
086acdd5
JD
2873 }
2874
d4f7d838
MN
2875 if(lens[0] == lens[1] && lens[1] > 1){
2876 for(i=0; h->default_ref_list[0][i].data[0] == h->default_ref_list[1][i].data[0] && i<lens[0]; i++);
2877 if(i == lens[0])
2878 FFSWAP(Picture, h->default_ref_list[1][0], h->default_ref_list[1][1]);
086acdd5 2879 }
086acdd5 2880 }else{
d4f7d838
MN
2881 len = build_def_list(h->default_ref_list[0] , h->short_ref, h->short_ref_count, 0, s->picture_structure);
2882 len+= build_def_list(h->default_ref_list[0]+len, h-> long_ref, 16 , 1, s->picture_structure);
2883 assert(len <= 32);
2884 if(len < h->ref_count[0])
2885 memset(&h->default_ref_list[0][len], 0, sizeof(Picture)*(h->ref_count[0] - len));
0da71265 2886 }
827c91bf
LLL
2887#ifdef TRACE
2888 for (i=0; i<h->ref_count[0]; i++) {
a9c9a240 2889 tprintf(h->s.avctx, "List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
827c91bf 2890 }
9f5c1037 2891 if(h->slice_type_nos==FF_B_TYPE){
827c91bf 2892 for (i=0; i<h->ref_count[1]; i++) {
ffbc5e04 2893 tprintf(h->s.avctx, "List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[1][i].data[0]);
827c91bf
LLL
2894 }
2895 }
2896#endif
0da71265
MN
2897 return 0;
2898}
2899
827c91bf
LLL
2900static void print_short_term(H264Context *h);
2901static void print_long_term(H264Context *h);
2902
949da388
JD
2903/**
2904 * Extract structure information about the picture described by pic_num in
2905 * the current decoding context (frame or field). Note that pic_num is
2906 * picture number without wrapping (so, 0<=pic_num<max_pic_num).
2907 * @param pic_num picture number for which to extract structure information
2908 * @param structure one of PICT_XXX describing structure of picture
2909 * with pic_num
2910 * @return frame number (short term) or long term index of picture
2911 * described by pic_num
2912 */
2913static int pic_num_extract(H264Context *h, int pic_num, int *structure){
2914 MpegEncContext * const s = &h->s;
2915
2916 *structure = s->picture_structure;
2917 if(FIELD_PICTURE){
2918 if (!(pic_num & 1))
2919 /* opposite field */
2920 *structure ^= PICT_FRAME;
2921 pic_num >>= 1;
2922 }
2923
2924 return pic_num;
2925}
2926
0da71265
MN
2927static int decode_ref_pic_list_reordering(H264Context *h){
2928 MpegEncContext * const s = &h->s;
949da388 2929 int list, index, pic_structure;
115329f1 2930
827c91bf
LLL
2931 print_short_term(h);
2932 print_long_term(h);
115329f1 2933
3425501d 2934 for(list=0; list<h->list_count; list++){
0da71265
MN
2935 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2936
2937 if(get_bits1(&s->gb)){
2938 int pred= h->curr_pic_num;
0da71265
MN
2939
2940 for(index=0; ; index++){
88e7a4d1
MN
2941 unsigned int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2942 unsigned int pic_id;
0da71265 2943 int i;
2f944356 2944 Picture *ref = NULL;
115329f1
DB
2945
2946 if(reordering_of_pic_nums_idc==3)
0bc42cad 2947 break;
115329f1 2948
0da71265 2949 if(index >= h->ref_count[list]){
9b879566 2950 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
0da71265
MN
2951 return -1;
2952 }
115329f1 2953
0da71265
MN
2954 if(reordering_of_pic_nums_idc<3){
2955 if(reordering_of_pic_nums_idc<2){
88e7a4d1 2956 const unsigned int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
949da388 2957 int frame_num;
0da71265 2958
03d3cab8 2959 if(abs_diff_pic_num > h->max_pic_num){
9b879566 2960 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
0da71265
MN
2961 return -1;
2962 }
2963
2964 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2965 else pred+= abs_diff_pic_num;
2966 pred &= h->max_pic_num - 1;
115329f1 2967
949da388
JD
2968 frame_num = pic_num_extract(h, pred, &pic_structure);
2969
0d175622
MN
2970 for(i= h->short_ref_count-1; i>=0; i--){
2971 ref = h->short_ref[i];
949da388 2972 assert(ref->reference);
0d175622 2973 assert(!ref->long_ref);
6edac8e1 2974 if(
af8c5e08
MN
2975 ref->frame_num == frame_num &&
2976 (ref->reference & pic_structure)
6edac8e1 2977 )
0da71265
MN
2978 break;
2979 }
0d175622 2980 if(i>=0)
949da388 2981 ref->pic_id= pred;
0da71265 2982 }else{
949da388 2983 int long_idx;
0da71265 2984 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
949da388
JD
2985
2986 long_idx= pic_num_extract(h, pic_id, &pic_structure);
2987
2988 if(long_idx>31){
88e7a4d1
MN
2989 av_log(h->s.avctx, AV_LOG_ERROR, "long_term_pic_idx overflow\n");
2990 return -1;
2991 }
949da388
JD
2992 ref = h->long_ref[long_idx];
2993 assert(!(ref && !ref->reference));
af8c5e08 2994 if(ref && (ref->reference & pic_structure)){
ac658be5 2995 ref->pic_id= pic_id;
ac658be5
FOL
2996 assert(ref->long_ref);
2997 i=0;
2998 }else{
2999 i=-1;
3000 }
0da71265
MN
3001 }
3002
0d315f28 3003 if (i < 0) {
9b879566 3004 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
0da71265 3005 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
0d175622
MN
3006 } else {
3007 for(i=index; i+1<h->ref_count[list]; i++){
3008 if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3009 break;
21be92bf
MN
3010 }
3011 for(; i > index; i--){
3012 h->ref_list[list][i]= h->ref_list[list][i-1];
3013 }
0d175622 3014 h->ref_list[list][index]= *ref;
949da388 3015 if (FIELD_PICTURE){
2143b118 3016 pic_as_field(&h->ref_list[list][index], pic_structure);
949da388 3017 }
0da71265 3018 }
0bc42cad 3019 }else{
9b879566 3020 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
0da71265
MN
3021 return -1;
3022 }
3023 }
3024 }
0da71265 3025 }
3425501d 3026 for(list=0; list<h->list_count; list++){
6ab87211 3027 for(index= 0; index < h->ref_count[list]; index++){
79b5c776
MN
3028 if(!h->ref_list[list][index].data[0]){
3029 av_log(h->s.avctx, AV_LOG_ERROR, "Missing reference picture\n");
3030 h->ref_list[list][index]= s->current_picture; //FIXME this is not a sensible solution
3031 }
6ab87211 3032 }
6ab87211 3033 }
115329f1 3034
9f5c1037 3035 if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
5ad984c9 3036 direct_dist_scale_factor(h);
2f944356 3037 direct_ref_list_init(h);
115329f1 3038 return 0;
0da71265
MN
3039}
3040
91c58c94 3041static void fill_mbaff_ref_list(H264Context *h){
5d18eaad 3042 int list, i, j;
3425501d 3043 for(list=0; list<2; list++){ //FIXME try list_count
5d18eaad
LM
3044 for(i=0; i<h->ref_count[list]; i++){
3045 Picture *frame = &h->ref_list[list][i];
3046 Picture *field = &h->ref_list[list][16+2*i];
3047 field[0] = *frame;
3048 for(j=0; j<3; j++)
3049 field[0].linesize[j] <<= 1;
2143b118 3050 field[0].reference = PICT_TOP_FIELD;
5d18eaad
LM
3051 field[1] = field[0];
3052 for(j=0; j<3; j++)
3053 field[1].data[j] += frame->linesize[j];
2143b118 3054 field[1].reference = PICT_BOTTOM_FIELD;
5d18eaad
LM
3055
3056 h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];
3057 h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];
3058 for(j=0; j<2; j++){
3059 h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];
3060 h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];
3061 }
3062 }
3063 }
3064 for(j=0; j<h->ref_count[1]; j++){
3065 for(i=0; i<h->ref_count[0]; i++)
3066 h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];
3067 memcpy(h->implicit_weight[16+2*j], h->implicit_weight[j], sizeof(*h->implicit_weight));
3068 memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));
3069 }
3070}
3071
0da71265
MN
3072static int pred_weight_table(H264Context *h){
3073 MpegEncContext * const s = &h->s;
3074 int list, i;
9f2d1b4f 3075 int luma_def, chroma_def;
115329f1 3076
9f2d1b4f
LM
3077 h->use_weight= 0;
3078 h->use_weight_chroma= 0;
0da71265
MN
3079 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3080 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
9f2d1b4f
LM
3081 luma_def = 1<<h->luma_log2_weight_denom;
3082 chroma_def = 1<<h->chroma_log2_weight_denom;
0da71265
MN
3083
3084 for(list=0; list<2; list++){
3085 for(i=0; i<h->ref_count[list]; i++){
3086 int luma_weight_flag, chroma_weight_flag;
115329f1 3087
0da71265
MN
3088 luma_weight_flag= get_bits1(&s->gb);
3089 if(luma_weight_flag){
3090 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3091 h->luma_offset[list][i]= get_se_golomb(&s->gb);
9f2d1b4f
LM
3092 if( h->luma_weight[list][i] != luma_def
3093 || h->luma_offset[list][i] != 0)
3094 h->use_weight= 1;
3095 }else{
3096 h->luma_weight[list][i]= luma_def;
3097 h->luma_offset[list][i]= 0;
0da71265
MN
3098 }
3099
0af6967e 3100 if(CHROMA){
fef744d4
MN
3101 chroma_weight_flag= get_bits1(&s->gb);
3102 if(chroma_weight_flag){
3103 int j;
3104 for(j=0; j<2; j++){
3105 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3106 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3107 if( h->chroma_weight[list][i][j] != chroma_def
3108 || h->chroma_offset[list][i][j] != 0)
3109 h->use_weight_chroma= 1;
3110 }
3111 }else{
3112 int j;
3113 for(j=0; j<2; j++){
3114 h->chroma_weight[list][i][j]= chroma_def;
3115 h->chroma_offset[list][i][j]= 0;
3116 }
0da71265
MN
3117 }
3118 }
3119 }
9f5c1037 3120 if(h->slice_type_nos != FF_B_TYPE) break;
0da71265 3121 }
9f2d1b4f 3122 h->use_weight= h->use_weight || h->use_weight_chroma;
0da71265
MN
3123 return 0;
3124}
3125
9f2d1b4f
LM
3126static void implicit_weight_table(H264Context *h){
3127 MpegEncContext * const s = &h->s;
9f2d1b4f
LM
3128 int ref0, ref1;
3129 int cur_poc = s->current_picture_ptr->poc;
3130
3131 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3132 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3133 h->use_weight= 0;
3134 h->use_weight_chroma= 0;
3135 return;
3136 }
3137
3138 h->use_weight= 2;
3139 h->use_weight_chroma= 2;
3140 h->luma_log2_weight_denom= 5;
3141 h->chroma_log2_weight_denom= 5;
3142
9f2d1b4f
LM
3143 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3144 int poc0 = h->ref_list[0][ref0].poc;
3145 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
738386a5 3146 int poc1 = h->ref_list[1][ref1].poc;
f66e4f5f 3147 int td = av_clip(poc1 - poc0, -128, 127);
9f2d1b4f 3148 if(td){
f66e4f5f 3149 int tb = av_clip(cur_poc - poc0, -128, 127);
c26abfa5 3150 int tx = (16384 + (FFABS(td) >> 1)) / td;
f66e4f5f 3151 int dist_scale_factor = av_clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
9f2d1b4f
LM
3152 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3153 h->implicit_weight[ref0][ref1] = 32;
3154 else
3155 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3156 }else
3157 h->implicit_weight[ref0][ref1] = 32;
3158 }
3159 }
3160}
3161
8fd57a66
JD
3162/**
3163 * Mark a picture as no longer needed for reference. The refmask
3164 * argument allows unreferencing of individual fields or the whole frame.
3165 * If the picture becomes entirely unreferenced, but is being held for
3166 * display purposes, it is marked as such.
3167 * @param refmask mask of fields to unreference; the mask is bitwise
3168 * anded with the reference marking of pic
3169 * @return non-zero if pic becomes entirely unreferenced (except possibly
3170 * for display purposes) zero if one of the fields remains in
3171 * reference
3172 */
3173static inline int unreference_pic(H264Context *h, Picture *pic, int refmask){
4e4d983e 3174 int i;
8fd57a66
JD
3175 if (pic->reference &= refmask) {
3176 return 0;
3177 } else {
79f4494a
MN
3178 for(i = 0; h->delayed_pic[i]; i++)
3179 if(pic == h->delayed_pic[i]){
3180 pic->reference=DELAYED_PIC_REF;
3181 break;
3182 }
8fd57a66
JD
3183 return 1;
3184 }
4e4d983e
LM
3185}
3186
0da71265 3187/**
5175b937 3188 * instantaneous decoder refresh.
0da71265
MN
3189 */
3190static void idr(H264Context *h){
4e4d983e 3191 int i;
0da71265 3192
dc032f33 3193 for(i=0; i<16; i++){
9c0e4624 3194 remove_long(h, i, 0);
0da71265 3195 }
849b9cef 3196 assert(h->long_ref_count==0);
0da71265
MN
3197
3198 for(i=0; i<h->short_ref_count; i++){
8fd57a66 3199 unreference_pic(h, h->short_ref[i], 0);
0da71265
MN
3200 h->short_ref[i]= NULL;
3201 }
3202 h->short_ref_count=0;
a149c1a5 3203 h->prev_frame_num= 0;
80f8e035
MN
3204 h->prev_frame_num_offset= 0;
3205 h->prev_poc_msb=
3206 h->prev_poc_lsb= 0;
0da71265
MN
3207}
3208
7c33ad19
LM
3209/* forget old pics after a seek */
3210static void flush_dpb(AVCodecContext *avctx){
3211 H264Context *h= avctx->priv_data;
3212 int i;
64b9d48f 3213 for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
285b570f
LM
3214 if(h->delayed_pic[i])
3215 h->delayed_pic[i]->reference= 0;
7c33ad19 3216 h->delayed_pic[i]= NULL;
285b570f 3217 }
df8a7dff 3218 h->outputed_poc= INT_MIN;