5339ed017eb064ef0b08025fcc683f53b77d6650
[libav.git] / libavcodec / h264.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file libavcodec/h264.c
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #include "dsputil.h"
29 #include "avcodec.h"
30 #include "mpegvideo.h"
31 #include "h264.h"
32 #include "h264data.h"
33 #include "h264_parser.h"
34 #include "golomb.h"
35 #include "mathops.h"
36 #include "rectangle.h"
37 #include "vdpau_internal.h"
38
39 #include "cabac.h"
40 #if ARCH_X86
41 #include "x86/h264_i386.h"
42 #endif
43
44 //#undef NDEBUG
45 #include <assert.h>
46
47 /**
48 * Value of Picture.reference when Picture is not a reference picture, but
49 * is held for delayed output.
50 */
51 #define DELAYED_PIC_REF 4
52
53 static VLC coeff_token_vlc[4];
54 static VLC_TYPE coeff_token_vlc_tables[520+332+280+256][2];
55 static const int coeff_token_vlc_tables_size[4]={520,332,280,256};
56
57 static VLC chroma_dc_coeff_token_vlc;
58 static VLC_TYPE chroma_dc_coeff_token_vlc_table[256][2];
59 static const int chroma_dc_coeff_token_vlc_table_size = 256;
60
61 static VLC total_zeros_vlc[15];
62 static VLC_TYPE total_zeros_vlc_tables[15][512][2];
63 static const int total_zeros_vlc_tables_size = 512;
64
65 static VLC chroma_dc_total_zeros_vlc[3];
66 static VLC_TYPE chroma_dc_total_zeros_vlc_tables[3][8][2];
67 static const int chroma_dc_total_zeros_vlc_tables_size = 8;
68
69 static VLC run_vlc[6];
70 static VLC_TYPE run_vlc_tables[6][8][2];
71 static const int run_vlc_tables_size = 8;
72
73 static VLC run7_vlc;
74 static VLC_TYPE run7_vlc_table[96][2];
75 static const int run7_vlc_table_size = 96;
76
77 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
78 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
79 static 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);
80 static 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);
81 static Picture * remove_long(H264Context *h, int i, int ref_mask);
82
83 static av_always_inline uint32_t pack16to32(int a, int b){
84 #ifdef WORDS_BIGENDIAN
85 return (b&0xFFFF) + (a<<16);
86 #else
87 return (a&0xFFFF) + (b<<16);
88 #endif
89 }
90
91 static const uint8_t rem6[52]={
92 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
93 };
94
95 static const uint8_t div6[52]={
96 0, 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,
97 };
98
99 static const uint8_t left_block_options[4][8]={
100 {0,1,2,3,7,10,8,11},
101 {2,2,3,3,8,11,8,11},
102 {0,0,1,1,7,10,7,10},
103 {0,2,0,2,7,10,7,10}
104 };
105
106 #define LEVEL_TAB_BITS 8
107 static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
108
109 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
110 MpegEncContext * const s = &h->s;
111 const int mb_xy= h->mb_xy;
112 int topleft_xy, top_xy, topright_xy, left_xy[2];
113 int topleft_type, top_type, topright_type, left_type[2];
114 const uint8_t * left_block;
115 int topleft_partition= -1;
116 int i;
117
118 top_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
119
120 //FIXME deblocking could skip the intra and nnz parts.
121 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
122 return;
123
124 /* Wow, what a mess, why didn't they simplify the interlacing & intra
125 * stuff, I can't imagine that these complex rules are worth it. */
126
127 topleft_xy = top_xy - 1;
128 topright_xy= top_xy + 1;
129 left_xy[1] = left_xy[0] = mb_xy-1;
130 left_block = left_block_options[0];
131 if(FRAME_MBAFF){
132 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
133 const int top_pair_xy = pair_xy - s->mb_stride;
134 const int topleft_pair_xy = top_pair_xy - 1;
135 const int topright_pair_xy = top_pair_xy + 1;
136 const int topleft_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
137 const int top_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
138 const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
139 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
140 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
141 const int bottom = (s->mb_y & 1);
142 tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
143
144 if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
145 top_xy -= s->mb_stride;
146 }
147 if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
148 topleft_xy -= s->mb_stride;
149 } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
150 topleft_xy += s->mb_stride;
151 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
152 topleft_partition = 0;
153 }
154 if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
155 topright_xy -= s->mb_stride;
156 }
157 if (left_mb_field_flag != curr_mb_field_flag) {
158 left_xy[1] = left_xy[0] = pair_xy - 1;
159 if (curr_mb_field_flag) {
160 left_xy[1] += s->mb_stride;
161 left_block = left_block_options[3];
162 } else {
163 left_block= left_block_options[2 - bottom];
164 }
165 }
166 }
167
168 h->top_mb_xy = top_xy;
169 h->left_mb_xy[0] = left_xy[0];
170 h->left_mb_xy[1] = left_xy[1];
171 if(for_deblock){
172 topleft_type = 0;
173 topright_type = 0;
174 top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
175 left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
176 left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
177
178 if(MB_MBAFF && !IS_INTRA(mb_type)){
179 int list;
180 for(list=0; list<h->list_count; list++){
181 //These values where changed for ease of performing MC, we need to change them back
182 //FIXME maybe we can make MC and loop filter use the same values or prevent
183 //the MC code from changing ref_cache and rather use a temporary array.
184 if(USES_LIST(mb_type,list)){
185 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
186 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
187 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
188 ref += h->b8_stride;
189 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
190 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
191 }
192 }
193 }
194 }else{
195 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
196 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
197 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
198 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
199 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
200
201 if(IS_INTRA(mb_type)){
202 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
203 h->topleft_samples_available=
204 h->top_samples_available=
205 h->left_samples_available= 0xFFFF;
206 h->topright_samples_available= 0xEEEA;
207
208 if(!(top_type & type_mask)){
209 h->topleft_samples_available= 0xB3FF;
210 h->top_samples_available= 0x33FF;
211 h->topright_samples_available= 0x26EA;
212 }
213 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
214 if(IS_INTERLACED(mb_type)){
215 if(!(left_type[0] & type_mask)){
216 h->topleft_samples_available&= 0xDFFF;
217 h->left_samples_available&= 0x5FFF;
218 }
219 if(!(left_type[1] & type_mask)){
220 h->topleft_samples_available&= 0xFF5F;
221 h->left_samples_available&= 0xFF5F;
222 }
223 }else{
224 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
225 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
226 assert(left_xy[0] == left_xy[1]);
227 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
228 h->topleft_samples_available&= 0xDF5F;
229 h->left_samples_available&= 0x5F5F;
230 }
231 }
232 }else{
233 if(!(left_type[0] & type_mask)){
234 h->topleft_samples_available&= 0xDF5F;
235 h->left_samples_available&= 0x5F5F;
236 }
237 }
238
239 if(!(topleft_type & type_mask))
240 h->topleft_samples_available&= 0x7FFF;
241
242 if(!(topright_type & type_mask))
243 h->topright_samples_available&= 0xFBFF;
244
245 if(IS_INTRA4x4(mb_type)){
246 if(IS_INTRA4x4(top_type)){
247 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
248 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
249 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
250 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
251 }else{
252 int pred;
253 if(!(top_type & type_mask))
254 pred= -1;
255 else{
256 pred= 2;
257 }
258 h->intra4x4_pred_mode_cache[4+8*0]=
259 h->intra4x4_pred_mode_cache[5+8*0]=
260 h->intra4x4_pred_mode_cache[6+8*0]=
261 h->intra4x4_pred_mode_cache[7+8*0]= pred;
262 }
263 for(i=0; i<2; i++){
264 if(IS_INTRA4x4(left_type[i])){
265 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
266 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
267 }else{
268 int pred;
269 if(!(left_type[i] & type_mask))
270 pred= -1;
271 else{
272 pred= 2;
273 }
274 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
275 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
276 }
277 }
278 }
279 }
280 }
281
282
283 /*
284 0 . T T. T T T T
285 1 L . .L . . . .
286 2 L . .L . . . .
287 3 . T TL . . . .
288 4 L . .L . . . .
289 5 L . .. . . . .
290 */
291 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
292 if(top_type){
293 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
294 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
295 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
296 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
297
298 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
299 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
300
301 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
302 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
303
304 }else{
305 h->non_zero_count_cache[4+8*0]=
306 h->non_zero_count_cache[5+8*0]=
307 h->non_zero_count_cache[6+8*0]=
308 h->non_zero_count_cache[7+8*0]=
309
310 h->non_zero_count_cache[1+8*0]=
311 h->non_zero_count_cache[2+8*0]=
312
313 h->non_zero_count_cache[1+8*3]=
314 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
315
316 }
317
318 for (i=0; i<2; i++) {
319 if(left_type[i]){
320 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
321 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
322 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
323 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
324 }else{
325 h->non_zero_count_cache[3+8*1 + 2*8*i]=
326 h->non_zero_count_cache[3+8*2 + 2*8*i]=
327 h->non_zero_count_cache[0+8*1 + 8*i]=
328 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
329 }
330 }
331
332 if( h->pps.cabac ) {
333 // top_cbp
334 if(top_type) {
335 h->top_cbp = h->cbp_table[top_xy];
336 } else if(IS_INTRA(mb_type)) {
337 h->top_cbp = 0x1C0;
338 } else {
339 h->top_cbp = 0;
340 }
341 // left_cbp
342 if (left_type[0]) {
343 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
344 } else if(IS_INTRA(mb_type)) {
345 h->left_cbp = 0x1C0;
346 } else {
347 h->left_cbp = 0;
348 }
349 if (left_type[0]) {
350 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
351 }
352 if (left_type[1]) {
353 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
354 }
355 }
356
357 #if 1
358 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
359 int list;
360 for(list=0; list<h->list_count; list++){
361 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
362 /*if(!h->mv_cache_clean[list]){
363 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
364 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
365 h->mv_cache_clean[list]= 1;
366 }*/
367 continue;
368 }
369 h->mv_cache_clean[list]= 0;
370
371 if(USES_LIST(top_type, list)){
372 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
373 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
374 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
375 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
376 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
377 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
378 h->ref_cache[list][scan8[0] + 0 - 1*8]=
379 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
380 h->ref_cache[list][scan8[0] + 2 - 1*8]=
381 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
382 }else{
383 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
384 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
385 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
386 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
387 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
388 }
389
390 for(i=0; i<2; i++){
391 int cache_idx = scan8[0] - 1 + i*2*8;
392 if(USES_LIST(left_type[i], list)){
393 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
394 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
395 *(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]];
396 *(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]];
397 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
398 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
399 }else{
400 *(uint32_t*)h->mv_cache [list][cache_idx ]=
401 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
402 h->ref_cache[list][cache_idx ]=
403 h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
404 }
405 }
406
407 if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
408 continue;
409
410 if(USES_LIST(topleft_type, list)){
411 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
412 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
413 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
414 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
415 }else{
416 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
417 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
418 }
419
420 if(USES_LIST(topright_type, list)){
421 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
422 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
423 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
424 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
425 }else{
426 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
427 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
428 }
429
430 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
431 continue;
432
433 h->ref_cache[list][scan8[5 ]+1] =
434 h->ref_cache[list][scan8[7 ]+1] =
435 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
436 h->ref_cache[list][scan8[4 ]] =
437 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
438 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
439 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
440 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
441 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
442 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
443
444 if( h->pps.cabac ) {
445 /* XXX beurk, Load mvd */
446 if(USES_LIST(top_type, list)){
447 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
448 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
449 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
450 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
451 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
452 }else{
453 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
454 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
455 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
456 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
457 }
458 if(USES_LIST(left_type[0], list)){
459 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
460 *(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]];
461 *(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]];
462 }else{
463 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
464 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
465 }
466 if(USES_LIST(left_type[1], list)){
467 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
468 *(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]];
469 *(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]];
470 }else{
471 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
472 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
473 }
474 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
475 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
476 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
477 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
478 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
479
480 if(h->slice_type_nos == FF_B_TYPE){
481 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
482
483 if(IS_DIRECT(top_type)){
484 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
485 }else if(IS_8X8(top_type)){
486 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
487 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
488 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
489 }else{
490 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
491 }
492
493 if(IS_DIRECT(left_type[0]))
494 h->direct_cache[scan8[0] - 1 + 0*8]= 1;
495 else if(IS_8X8(left_type[0]))
496 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)];
497 else
498 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
499
500 if(IS_DIRECT(left_type[1]))
501 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
502 else if(IS_8X8(left_type[1]))
503 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)];
504 else
505 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
506 }
507 }
508
509 if(FRAME_MBAFF){
510 #define MAP_MVS\
511 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
512 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
513 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
514 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
515 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
516 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
517 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
518 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
519 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
520 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
521 if(MB_FIELD){
522 #define MAP_F2F(idx, mb_type)\
523 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
524 h->ref_cache[list][idx] <<= 1;\
525 h->mv_cache[list][idx][1] /= 2;\
526 h->mvd_cache[list][idx][1] /= 2;\
527 }
528 MAP_MVS
529 #undef MAP_F2F
530 }else{
531 #define MAP_F2F(idx, mb_type)\
532 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
533 h->ref_cache[list][idx] >>= 1;\
534 h->mv_cache[list][idx][1] <<= 1;\
535 h->mvd_cache[list][idx][1] <<= 1;\
536 }
537 MAP_MVS
538 #undef MAP_F2F
539 }
540 }
541 }
542 }
543 #endif
544
545 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
546 }
547
548 static inline void write_back_intra_pred_mode(H264Context *h){
549 const int mb_xy= h->mb_xy;
550
551 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
552 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
553 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
554 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
555 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
556 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
557 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
558 }
559
560 /**
561 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
562 */
563 static inline int check_intra4x4_pred_mode(H264Context *h){
564 MpegEncContext * const s = &h->s;
565 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
566 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
567 int i;
568
569 if(!(h->top_samples_available&0x8000)){
570 for(i=0; i<4; i++){
571 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
572 if(status<0){
573 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);
574 return -1;
575 } else if(status){
576 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
577 }
578 }
579 }
580
581 if((h->left_samples_available&0x8888)!=0x8888){
582 static const int mask[4]={0x8000,0x2000,0x80,0x20};
583 for(i=0; i<4; i++){
584 if(!(h->left_samples_available&mask[i])){
585 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
586 if(status<0){
587 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);
588 return -1;
589 } else if(status){
590 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
591 }
592 }
593 }
594 }
595
596 return 0;
597 } //FIXME cleanup like next
598
599 /**
600 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
601 */
602 static inline int check_intra_pred_mode(H264Context *h, int mode){
603 MpegEncContext * const s = &h->s;
604 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
605 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
606
607 if(mode > 6U) {
608 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);
609 return -1;
610 }
611
612 if(!(h->top_samples_available&0x8000)){
613 mode= top[ mode ];
614 if(mode<0){
615 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);
616 return -1;
617 }
618 }
619
620 if((h->left_samples_available&0x8080) != 0x8080){
621 mode= left[ mode ];
622 if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
623 mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
624 }
625 if(mode<0){
626 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);
627 return -1;
628 }
629 }
630
631 return mode;
632 }
633
634 /**
635 * gets the predicted intra4x4 prediction mode.
636 */
637 static inline int pred_intra_mode(H264Context *h, int n){
638 const int index8= scan8[n];
639 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
640 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
641 const int min= FFMIN(left, top);
642
643 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
644
645 if(min<0) return DC_PRED;
646 else return min;
647 }
648
649 static inline void write_back_non_zero_count(H264Context *h){
650 const int mb_xy= h->mb_xy;
651
652 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
653 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
654 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
655 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
656 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
657 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
658 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
659
660 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
661 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
662 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
663
664 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
665 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
666 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
667 }
668
669 /**
670 * gets the predicted number of non-zero coefficients.
671 * @param n block index
672 */
673 static 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;
678
679 if(i<64) i= (i+1)>>1;
680
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);
682
683 return i&31;
684 }
685
686 static 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 ];
688 MpegEncContext *s = &h->s;
689
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){
693 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
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];\
705 if(!USES_LIST(mb_type,list))\
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){
725 // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
726 SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
727 }
728 }
729 #undef SET_DIAG_MV
730 }
731
732 if(topright_ref != PART_NOT_AVAILABLE){
733 *C= h->mv_cache[list][ i - 8 + part_width ];
734 return topright_ref;
735 }else{
736 tprintf(s->avctx, "topright MV not available\n");
737
738 *C= h->mv_cache[list][ i - 8 - 1 ];
739 return h->ref_cache[list][ i - 8 - 1 ];
740 }
741 }
742
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 */
750 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
751 const int index8= scan8[n];
752 const int top_ref= h->ref_cache[list][ index8 - 8 ];
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 ];
756 const int16_t * C;
757 int diagonal_ref, match_count;
758
759 assert(part_width==1 || part_width==2 || part_width==4);
760
761 /* mv_cache
762 B . . A T T T T
763 U . . L . . , .
764 U . . L . . . .
765 U . . L . . , .
766 . . . L . . . .
767 */
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);
771 tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
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];
778 *my= A[1];
779 }else if(top_ref==ref){
780 *mx= B[0];
781 *my= B[1];
782 }else{
783 *mx= C[0];
784 *my= C[1];
785 }
786 }else{
787 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
788 *mx= A[0];
789 *my= A[1];
790 }else{
791 *mx= mid_pred(A[0], B[0], C[0]);
792 *my= mid_pred(A[1], B[1], C[1]);
793 }
794 }
795
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);
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 */
805 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
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
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);
811
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 ];
820
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);
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 */
840 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
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 ];
844
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);
846
847 if(left_ref == ref){
848 *mx= A[0];
849 *my= A[1];
850 return;
851 }
852 }else{
853 const int16_t * C;
854 int diagonal_ref;
855
856 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
857
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);
859
860 if(diagonal_ref == ref){
861 *mx= C[0];
862 *my= C[1];
863 return;
864 }
865 }
866
867 //RARE
868 pred_motion(h, n, 2, list, ref, mx, my);
869 }
870
871 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
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
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);
876
877 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
878 || !( top_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ])
879 || !(left_ref | *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ])){
880
881 *mx = *my = 0;
882 return;
883 }
884
885 pred_motion(h, 0, 4, 0, 0, mx, my);
886
887 return;
888 }
889
890 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
891 int poc0 = h->ref_list[0][i].poc;
892 int td = av_clip(poc1 - poc0, -128, 127);
893 if(td == 0 || h->ref_list[0][i].long_ref){
894 return 256;
895 }else{
896 int tb = av_clip(poc - poc0, -128, 127);
897 int tx = (16384 + (FFABS(td) >> 1)) / td;
898 return av_clip((tb*tx + 32) >> 6, -1024, 1023);
899 }
900 }
901
902 static inline void direct_dist_scale_factor(H264Context * const h){
903 MpegEncContext * const s = &h->s;
904 const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
905 const int poc1 = h->ref_list[1][0].poc;
906 int i, field;
907 for(field=0; field<2; field++){
908 const int poc = h->s.current_picture_ptr->field_poc[field];
909 const int poc1 = h->ref_list[1][0].field_poc[field];
910 for(i=0; i < 2*h->ref_count[0]; i++)
911 h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
912 }
913
914 for(i=0; i<h->ref_count[0]; i++){
915 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
916 }
917 }
918
919 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
920 MpegEncContext * const s = &h->s;
921 Picture * const ref1 = &h->ref_list[1][0];
922 int j, old_ref, rfield;
923 int start= mbafi ? 16 : 0;
924 int end = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];
925 int interl= mbafi || s->picture_structure != PICT_FRAME;
926
927 /* bogus; fills in for missing frames */
928 memset(map[list], 0, sizeof(map[list]));
929
930 for(rfield=0; rfield<2; rfield++){
931 for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
932 int poc = ref1->ref_poc[colfield][list][old_ref];
933
934 if (!interl)
935 poc |= 3;
936 else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
937 poc= (poc&~3) + rfield + 1;
938
939 for(j=start; j<end; j++){
940 if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){
941 int cur_ref= mbafi ? (j-16)^field : j;
942 map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
943 if(rfield == field)
944 map[list][old_ref] = cur_ref;
945 break;
946 }
947 }
948 }
949 }
950 }
951
952 static inline void direct_ref_list_init(H264Context * const h){
953 MpegEncContext * const s = &h->s;
954 Picture * const ref1 = &h->ref_list[1][0];
955 Picture * const cur = s->current_picture_ptr;
956 int list, j, field;
957 int sidx= (s->picture_structure&1)^1;
958 int ref1sidx= (ref1->reference&1)^1;
959
960 for(list=0; list<2; list++){
961 cur->ref_count[sidx][list] = h->ref_count[list];
962 for(j=0; j<h->ref_count[list]; j++)
963 cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
964 }
965
966 if(s->picture_structure == PICT_FRAME){
967 memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
968 memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0]));
969 }
970
971 cur->mbaff= FRAME_MBAFF;
972
973 if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
974 return;
975
976 for(list=0; list<2; list++){
977 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
978 for(field=0; field<2; field++)
979 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
980 }
981 }
982
983 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
984 MpegEncContext * const s = &h->s;
985 int b8_stride = h->b8_stride;
986 int b4_stride = h->b_stride;
987 int mb_xy = h->mb_xy;
988 int mb_type_col[2];
989 const int16_t (*l1mv0)[2], (*l1mv1)[2];
990 const int8_t *l1ref0, *l1ref1;
991 const int is_b8x8 = IS_8X8(*mb_type);
992 unsigned int sub_mb_type;
993 int i8, i4;
994
995 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
996
997 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
998 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL
999 int cur_poc = s->current_picture_ptr->poc;
1000 int *col_poc = h->ref_list[1]->field_poc;
1001 int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
1002 mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;
1003 b8_stride = 0;
1004 }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity
1005 int fieldoff= 2*(h->ref_list[1][0].reference)-3;
1006 mb_xy += s->mb_stride*fieldoff;
1007 }
1008 goto single_col;
1009 }else{ // AFL/AFR/FR/FL -> AFR/FR
1010 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
1011 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
1012 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
1013 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
1014 b8_stride *= 3;
1015 b4_stride *= 6;
1016 //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag
1017 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
1018 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
1019 && !is_b8x8){
1020 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1021 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
1022 }else{
1023 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1024 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1025 }
1026 }else{ // AFR/FR -> AFR/FR
1027 single_col:
1028 mb_type_col[0] =
1029 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
1030 if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
1031 /* FIXME save sub mb types from previous frames (or derive from MVs)
1032 * so we know exactly what block size to use */
1033 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1034 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1035 }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
1036 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1037 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1038 }else{
1039 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1040 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
1041 }
1042 }
1043 }
1044
1045 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
1046 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
1047 l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
1048 l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
1049 if(!b8_stride){
1050 if(s->mb_y&1){
1051 l1ref0 += h->b8_stride;
1052 l1ref1 += h->b8_stride;
1053 l1mv0 += 2*b4_stride;
1054 l1mv1 += 2*b4_stride;
1055 }
1056 }
1057
1058 if(h->direct_spatial_mv_pred){
1059 int ref[2];
1060 int mv[2][2];
1061 int list;
1062
1063 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1064
1065 /* ref = min(neighbors) */
1066 for(list=0; list<2; list++){
1067 int refa = h->ref_cache[list][scan8[0] - 1];
1068 int refb = h->ref_cache[list][scan8[0] - 8];
1069 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1070 if(refc == PART_NOT_AVAILABLE)
1071 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1072 ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);
1073 if(ref[list] < 0)
1074 ref[list] = -1;
1075 }
1076
1077 if(ref[0] < 0 && ref[1] < 0){
1078 ref[0] = ref[1] = 0;
1079 mv[0][0] = mv[0][1] =
1080 mv[1][0] = mv[1][1] = 0;
1081 }else{
1082 for(list=0; list<2; list++){
1083 if(ref[list] >= 0)
1084 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1085 else
1086 mv[list][0] = mv[list][1] = 0;
1087 }
1088 }
1089
1090 if(ref[1] < 0){
1091 if(!is_b8x8)
1092 *mb_type &= ~MB_TYPE_L1;
1093 sub_mb_type &= ~MB_TYPE_L1;
1094 }else if(ref[0] < 0){
1095 if(!is_b8x8)
1096 *mb_type &= ~MB_TYPE_L0;
1097 sub_mb_type &= ~MB_TYPE_L0;
1098 }
1099
1100 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1101 for(i8=0; i8<4; i8++){
1102 int x8 = i8&1;
1103 int y8 = i8>>1;
1104 int xy8 = x8+y8*b8_stride;
1105 int xy4 = 3*x8+y8*b4_stride;
1106 int a=0, b=0;
1107
1108 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1109 continue;
1110 h->sub_mb_type[i8] = sub_mb_type;
1111
1112 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1113 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1114 if(!IS_INTRA(mb_type_col[y8])
1115 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
1116 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
1117 if(ref[0] > 0)
1118 a= pack16to32(mv[0][0],mv[0][1]);
1119 if(ref[1] > 0)
1120 b= pack16to32(mv[1][0],mv[1][1]);
1121 }else{
1122 a= pack16to32(mv[0][0],mv[0][1]);
1123 b= pack16to32(mv[1][0],mv[1][1]);
1124 }
1125 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
1126 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
1127 }
1128 }else if(IS_16X16(*mb_type)){
1129 int a=0, b=0;
1130
1131 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1132 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1133 if(!IS_INTRA(mb_type_col[0])
1134 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1135 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
1136 && (h->x264_build>33 || !h->x264_build)))){
1137 if(ref[0] > 0)
1138 a= pack16to32(mv[0][0],mv[0][1]);
1139 if(ref[1] > 0)
1140 b= pack16to32(mv[1][0],mv[1][1]);
1141 }else{
1142 a= pack16to32(mv[0][0],mv[0][1]);
1143 b= pack16to32(mv[1][0],mv[1][1]);
1144 }
1145 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1146 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
1147 }else{
1148 for(i8=0; i8<4; i8++){
1149 const int x8 = i8&1;
1150 const int y8 = i8>>1;
1151
1152 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1153 continue;
1154 h->sub_mb_type[i8] = sub_mb_type;
1155
1156 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1157 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1158 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1159 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1160
1161 /* col_zero_flag */
1162 if(!IS_INTRA(mb_type_col[0]) && ( l1ref0[x8 + y8*b8_stride] == 0
1163 || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
1164 && (h->x264_build>33 || !h->x264_build)))){
1165 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
1166 if(IS_SUB_8X8(sub_mb_type)){
1167 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1168 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1169 if(ref[0] == 0)
1170 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1171 if(ref[1] == 0)
1172 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1173 }
1174 }else
1175 for(i4=0; i4<4; i4++){
1176 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1177 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
1178 if(ref[0] == 0)
1179 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1180 if(ref[1] == 0)
1181 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1182 }
1183 }
1184 }
1185 }
1186 }
1187 }else{ /* direct temporal mv pred */
1188 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1189 const int *dist_scale_factor = h->dist_scale_factor;
1190 int ref_offset= 0;
1191
1192 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
1193 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
1194 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
1195 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
1196 }
1197 if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
1198 ref_offset += 16;
1199
1200 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
1201 /* FIXME assumes direct_8x8_inference == 1 */
1202 int y_shift = 2*!IS_INTERLACED(*mb_type);
1203
1204 for(i8=0; i8<4; i8++){
1205 const int x8 = i8&1;
1206 const int y8 = i8>>1;
1207 int ref0, scale;
1208 const int16_t (*l1mv)[2]= l1mv0;
1209
1210 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1211 continue;
1212 h->sub_mb_type[i8] = sub_mb_type;
1213
1214 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1215 if(IS_INTRA(mb_type_col[y8])){
1216 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1217 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1218 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1219 continue;
1220 }
1221
1222 ref0 = l1ref0[x8 + y8*b8_stride];
1223 if(ref0 >= 0)
1224 ref0 = map_col_to_list0[0][ref0 + ref_offset];
1225 else{
1226 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1227 l1mv= l1mv1;
1228 }
1229 scale = dist_scale_factor[ref0];
1230 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1231
1232 {
1233 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
1234 int my_col = (mv_col[1]<<y_shift)/2;
1235 int mx = (scale * mv_col[0] + 128) >> 8;
1236 int my = (scale * my_col + 128) >> 8;
1237 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1238 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1239 }
1240 }
1241 return;
1242 }
1243
1244 /* one-to-one mv scaling */
1245
1246 if(IS_16X16(*mb_type)){
1247 int ref, mv0, mv1;
1248
1249 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1250 if(IS_INTRA(mb_type_col[0])){
1251 ref=mv0=mv1=0;
1252 }else{
1253 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
1254 : map_col_to_list0[1][l1ref1[0] + ref_offset];
1255 const int scale = dist_scale_factor[ref0];
1256 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1257 int mv_l0[2];
1258 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1259 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1260 ref= ref0;
1261 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1262 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1263 }
1264 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1265 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1266 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
1267 }else{
1268 for(i8=0; i8<4; i8++){
1269 const int x8 = i8&1;
1270 const int y8 = i8>>1;
1271 int ref0, scale;
1272 const int16_t (*l1mv)[2]= l1mv0;
1273
1274 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1275 continue;
1276 h->sub_mb_type[i8] = sub_mb_type;
1277 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1278 if(IS_INTRA(mb_type_col[0])){
1279 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1280 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1281 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1282 continue;
1283 }
1284
1285 ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;
1286 if(ref0 >= 0)
1287 ref0 = map_col_to_list0[0][ref0];
1288 else{
1289 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
1290 l1mv= l1mv1;
1291 }
1292 scale = dist_scale_factor[ref0];
1293
1294 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1295 if(IS_SUB_8X8(sub_mb_type)){
1296 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
1297 int mx = (scale * mv_col[0] + 128) >> 8;
1298 int my = (scale * mv_col[1] + 128) >> 8;
1299 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1300 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1301 }else
1302 for(i4=0; i4<4; i4++){
1303 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
1304 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1305 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1306 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
1307 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1308 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1309 }
1310 }
1311 }
1312 }
1313 }
1314
1315 static inline void write_back_motion(H264Context *h, int mb_type){
1316 MpegEncContext * const s = &h->s;
1317 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1318 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1319 int list;
1320
1321 if(!USES_LIST(mb_type, 0))
1322 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1323
1324 for(list=0; list<h->list_count; list++){
1325 int y;
1326 if(!USES_LIST(mb_type, list))
1327 continue;
1328
1329 for(y=0; y<4; y++){
1330 *(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];
1331 *(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];
1332 }
1333 if( h->pps.cabac ) {
1334 if(IS_SKIP(mb_type))
1335 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1336 else
1337 for(y=0; y<4; y++){
1338 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1339 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1340 }
1341 }
1342
1343 {
1344 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1345 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1346 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1347 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1348 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1349 }
1350 }
1351
1352 if(h->slice_type_nos == FF_B_TYPE && h->pps.cabac){
1353 if(IS_8X8(mb_type)){
1354 uint8_t *direct_table = &h->direct_table[b8_xy];
1355 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1356 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1357 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1358 }
1359 }
1360 }
1361
1362 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
1363 int i, si, di;
1364 uint8_t *dst;
1365 int bufidx;
1366
1367 // src[0]&0x80; //forbidden bit
1368 h->nal_ref_idc= src[0]>>5;
1369 h->nal_unit_type= src[0]&0x1F;
1370
1371 src++; length--;
1372 #if 0
1373 for(i=0; i<length; i++)
1374 printf("%2X ", src[i]);
1375 #endif
1376
1377 #if HAVE_FAST_UNALIGNED
1378 # if HAVE_FAST_64BIT
1379 # define RS 7
1380 for(i=0; i+1<length; i+=9){
1381 if(!((~*(const uint64_t*)(src+i) & (*(const uint64_t*)(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
1382 # else
1383 # define RS 3
1384 for(i=0; i+1<length; i+=5){
1385 if(!((~*(const uint32_t*)(src+i) & (*(const uint32_t*)(src+i) - 0x01000101U)) & 0x80008080U))
1386 # endif
1387 continue;
1388 if(i>0 && !src[i]) i--;
1389 while(src[i]) i++;
1390 #else
1391 # define RS 0
1392 for(i=0; i+1<length; i+=2){
1393 if(src[i]) continue;
1394 if(i>0 && src[i-1]==0) i--;
1395 #endif
1396 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1397 if(src[i+2]!=3){
1398 /* startcode, so we must be past the end */
1399 length=i;
1400 }
1401 break;
1402 }
1403 i-= RS;
1404 }
1405
1406 if(i>=length-1){ //no escaped 0
1407 *dst_length= length;
1408 *consumed= length+1; //+1 for the header
1409 return src;
1410 }
1411
1412 bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
1413 h->rbsp_buffer[bufidx]= av_fast_realloc(h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+FF_INPUT_BUFFER_PADDING_SIZE);
1414 dst= h->rbsp_buffer[bufidx];
1415
1416 if (dst == NULL){
1417 return NULL;
1418 }
1419
1420 //printf("decoding esc\n");
1421 memcpy(dst, src, i);
1422 si=di=i;
1423 while(si+2<length){
1424 //remove escapes (very rare 1:2^22)
1425 if(src[si+2]>3){
1426 dst[di++]= src[si++];
1427 dst[di++]= src[si++];
1428 }else if(src[si]==0 && src[si+1]==0){
1429 if(src[si+2]==3){ //escape
1430 dst[di++]= 0;
1431 dst[di++]= 0;
1432 si+=3;
1433 continue;
1434 }else //next start code
1435 goto nsc;
1436 }
1437
1438 dst[di++]= src[si++];
1439 }
1440 while(si<length)
1441 dst[di++]= src[si++];
1442 nsc:
1443
1444 memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
1445
1446 *dst_length= di;
1447 *consumed= si + 1;//+1 for the header
1448 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
1449 return dst;
1450 }
1451
1452 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
1453 int v= *src;
1454 int r;
1455
1456 tprintf(h->s.avctx, "rbsp trailing %X\n", v);
1457
1458 for(r=1; r<9; r++){
1459 if(v&1) return r;
1460 v>>=1;
1461 }
1462 return 0;
1463 }
1464
1465 /**
1466 * IDCT transforms the 16 dc values and dequantizes them.
1467 * @param qp quantization parameter
1468 */
1469 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1470 #define stride 16
1471 int i;
1472 int temp[16]; //FIXME check if this is a good idea
1473 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1474 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1475
1476 //memset(block, 64, 2*256);
1477 //return;
1478 for(i=0; i<4; i++){
1479 const int offset= y_offset[i];
1480 const int z0= block[offset+stride*0] + block[offset+stride*4];
1481 const int z1= block[offset+stride*0] - block[offset+stride*4];
1482 const int z2= block[offset+stride*1] - block[offset+stride*5];
1483 const int z3= block[offset+stride*1] + block[offset+stride*5];
1484
1485 temp[4*i+0]= z0+z3;
1486 temp[4*i+1]= z1+z2;
1487 temp[4*i+2]= z1-z2;
1488 temp[4*i+3]= z0-z3;
1489 }
1490
1491 for(i=0; i<4; i++){
1492 const int offset= x_offset[i];
1493 const int z0= temp[4*0+i] + temp[4*2+i];
1494 const int z1= temp[4*0+i] - temp[4*2+i];
1495 const int z2= temp[4*1+i] - temp[4*3+i];
1496 const int z3= temp[4*1+i] + temp[4*3+i];
1497
1498 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
1499 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1500 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1501 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1502 }
1503 }
1504
1505 #if 0
1506 /**
1507 * DCT transforms the 16 dc values.
1508 * @param qp quantization parameter ??? FIXME
1509 */
1510 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1511 // const int qmul= dequant_coeff[qp][0];
1512 int i;
1513 int temp[16]; //FIXME check if this is a good idea
1514 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1515 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1516
1517 for(i=0; i<4; i++){
1518 const int offset= y_offset[i];
1519 const int z0= block[offset+stride*0] + block[offset+stride*4];
1520 const int z1= block[offset+stride*0] - block[offset+stride*4];
1521 const int z2= block[offset+stride*1] - block[offset+stride*5];
1522 const int z3= block[offset+stride*1] + block[offset+stride*5];
1523
1524 temp[4*i+0]= z0+z3;
1525 temp[4*i+1]= z1+z2;
1526 temp[4*i+2]= z1-z2;
1527 temp[4*i+3]= z0-z3;
1528 }
1529
1530 for(i=0; i<4; i++){
1531 const int offset= x_offset[i];
1532 const int z0= temp[4*0+i] + temp[4*2+i];
1533 const int z1= temp[4*0+i] - temp[4*2+i];
1534 const int z2= temp[4*1+i] - temp[4*3+i];
1535 const int z3= temp[4*1+i] + temp[4*3+i];
1536
1537 block[stride*0 +offset]= (z0 + z3)>>1;
1538 block[stride*2 +offset]= (z1 + z2)>>1;
1539 block[stride*8 +offset]= (z1 - z2)>>1;
1540 block[stride*10+offset]= (z0 - z3)>>1;
1541 }
1542 }
1543 #endif
1544
1545 #undef xStride
1546 #undef stride
1547
1548 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1549 const int stride= 16*2;
1550 const int xStride= 16;
1551 int a,b,c,d,e;
1552
1553 a= block[stride*0 + xStride*0];
1554 b= block[stride*0 + xStride*1];
1555 c= block[stride*1 + xStride*0];
1556 d= block[stride*1 + xStride*1];
1557
1558 e= a-b;
1559 a= a+b;
1560 b= c-d;
1561 c= c+d;
1562
1563 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1564 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1565 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1566 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1567 }
1568
1569 #if 0
1570 static void chroma_dc_dct_c(DCTELEM *block){
1571 const int stride= 16*2;
1572 const int xStride= 16;
1573 int a,b,c,d,e;
1574
1575 a= block[stride*0 + xStride*0];
1576 b= block[stride*0 + xStride*1];
1577 c= block[stride*1 + xStride*0];
1578 d= block[stride*1 + xStride*1];
1579
1580 e= a-b;
1581 a= a+b;
1582 b= c-d;
1583 c= c+d;
1584
1585 block[stride*0 + xStride*0]= (a+c);
1586 block[stride*0 + xStride*1]= (e+b);
1587 block[stride*1 + xStride*0]= (a-c);
1588 block[stride*1 + xStride*1]= (e-b);
1589 }
1590 #endif
1591
1592 /**
1593 * gets the chroma qp.
1594 */
1595 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
1596 return h->pps.chroma_qp_table[t][qscale];
1597 }
1598
1599 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1600 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1601 int src_x_offset, int src_y_offset,
1602 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1603 MpegEncContext * const s = &h->s;
1604 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1605 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1606 const int luma_xy= (mx&3) + ((my&3)<<2);
1607 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
1608 uint8_t * src_cb, * src_cr;
1609 int extra_width= h->emu_edge_width;
1610 int extra_height= h->emu_edge_height;
1611 int emu=0;
1612 const int full_mx= mx>>2;
1613 const int full_my= my>>2;
1614 const int pic_width = 16*s->mb_width;
1615 const int pic_height = 16*s->mb_height >> MB_FIELD;
1616
1617 if(mx&7) extra_width -= 3;
1618 if(my&7) extra_height -= 3;
1619
1620 if( full_mx < 0-extra_width
1621 || full_my < 0-extra_height
1622 || full_mx + 16/*FIXME*/ > pic_width + extra_width
1623 || full_my + 16/*FIXME*/ > pic_height + extra_height){
1624 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);
1625 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
1626 emu=1;
1627 }
1628
1629 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
1630 if(!square){
1631 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
1632 }
1633
1634 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
1635
1636 if(MB_FIELD){
1637 // chroma offset when predicting from a field of opposite parity
1638 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
1639 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
1640 }
1641 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1642 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
1643
1644 if(emu){
1645 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);
1646 src_cb= s->edge_emu_buffer;
1647 }
1648 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1649
1650 if(emu){
1651 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);
1652 src_cr= s->edge_emu_buffer;
1653 }
1654 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
1655 }
1656
1657 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
1658 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1659 int x_offset, int y_offset,
1660 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1661 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1662 int list0, int list1){
1663 MpegEncContext * const s = &h->s;
1664 qpel_mc_func *qpix_op= qpix_put;
1665 h264_chroma_mc_func chroma_op= chroma_put;
1666
1667 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1668 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1669 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1670 x_offset += 8*s->mb_x;
1671 y_offset += 8*(s->mb_y >> MB_FIELD);
1672
1673 if(list0){
1674 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1675 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1676 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1677 qpix_op, chroma_op);
1678
1679 qpix_op= qpix_avg;
1680 chroma_op= chroma_avg;
1681 }
1682
1683 if(list1){
1684 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1685 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1686 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1687 qpix_op, chroma_op);
1688 }
1689 }
1690
1691 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
1692 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1693 int x_offset, int y_offset,
1694 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1695 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
1696 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
1697 int list0, int list1){
1698 MpegEncContext * const s = &h->s;
1699
1700 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
1701 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
1702 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
1703 x_offset += 8*s->mb_x;
1704 y_offset += 8*(s->mb_y >> MB_FIELD);
1705
1706 if(list0 && list1){
1707 /* don't optimize for luma-only case, since B-frames usually
1708 * use implicit weights => chroma too. */
1709 uint8_t *tmp_cb = s->obmc_scratchpad;
1710 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
1711 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
1712 int refn0 = h->ref_cache[0][ scan8[n] ];
1713 int refn1 = h->ref_cache[1][ scan8[n] ];
1714
1715 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
1716 dest_y, dest_cb, dest_cr,
1717 x_offset, y_offset, qpix_put, chroma_put);
1718 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
1719 tmp_y, tmp_cb, tmp_cr,
1720 x_offset, y_offset, qpix_put, chroma_put);
1721
1722 if(h->use_weight == 2){
1723 int weight0 = h->implicit_weight[refn0][refn1];
1724 int weight1 = 64 - weight0;
1725 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
1726 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
1727 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
1728 }else{
1729 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
1730 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
1731 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
1732 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1733 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
1734 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
1735 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1736 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
1737 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
1738 }
1739 }else{
1740 int list = list1 ? 1 : 0;
1741 int refn = h->ref_cache[list][ scan8[n] ];
1742 Picture *ref= &h->ref_list[list][refn];
1743 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
1744 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1745 qpix_put, chroma_put);
1746
1747 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
1748 h->luma_weight[list][refn], h->luma_offset[list][refn]);
1749 if(h->use_weight_chroma){
1750 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1751 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
1752 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
1753 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
1754 }
1755 }
1756 }
1757
1758 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1759 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1760 int x_offset, int y_offset,
1761 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1762 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1763 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
1764 int list0, int list1){
1765 if((h->use_weight==2 && list0 && list1
1766 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
1767 || h->use_weight==1)
1768 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1769 x_offset, y_offset, qpix_put, chroma_put,
1770 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
1771 else
1772 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
1773 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
1774 }
1775
1776 static inline void prefetch_motion(H264Context *h, int list){
1777 /* fetch pixels for estimated mv 4 macroblocks ahead
1778 * optimized for 64byte cache lines */
1779 MpegEncContext * const s = &h->s;
1780 const int refn = h->ref_cache[list][scan8[0]];
1781 if(refn >= 0){
1782 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
1783 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
1784 uint8_t **src= h->ref_list[list][refn].data;
1785 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
1786 s->dsp.prefetch(src[0]+off, s->linesize, 4);
1787 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
1788 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
1789 }
1790 }
1791
1792 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1793 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1794 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
1795 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
1796 MpegEncContext * const s = &h->s;
1797 const int mb_xy= h->mb_xy;
1798 const int mb_type= s->current_picture.mb_type[mb_xy];
1799
1800 assert(IS_INTER(mb_type));
1801
1802 prefetch_motion(h, 0);
1803
1804 if(IS_16X16(mb_type)){
1805 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1806 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1807 &weight_op[0], &weight_avg[0],
1808 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1809 }else if(IS_16X8(mb_type)){
1810 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1811 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1812 &weight_op[1], &weight_avg[1],
1813 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1814 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1815 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1816 &weight_op[1], &weight_avg[1],
1817 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1818 }else if(IS_8X16(mb_type)){
1819 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
1820 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1821 &weight_op[2], &weight_avg[2],
1822 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1823 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
1824 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1825 &weight_op[2], &weight_avg[2],
1826 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1827 }else{
1828 int i;
1829
1830 assert(IS_8X8(mb_type));
1831
1832 for(i=0; i<4; i++){
1833 const int sub_mb_type= h->sub_mb_type[i];
1834 const int n= 4*i;
1835 int x_offset= (i&1)<<2;
1836 int y_offset= (i&2)<<1;
1837
1838 if(IS_SUB_8X8(sub_mb_type)){
1839 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1840 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1841 &weight_op[3], &weight_avg[3],
1842 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1843 }else if(IS_SUB_8X4(sub_mb_type)){
1844 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1845 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1846 &weight_op[4], &weight_avg[4],
1847 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1848 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1849 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1850 &weight_op[4], &weight_avg[4],
1851 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1852 }else if(IS_SUB_4X8(sub_mb_type)){
1853 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1854 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1855 &weight_op[5], &weight_avg[5],
1856 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1857 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
1858 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1859 &weight_op[5], &weight_avg[5],
1860 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1861 }else{
1862 int j;
1863 assert(IS_SUB_4X4(sub_mb_type));
1864 for(j=0; j<4; j++){
1865 int sub_x_offset= x_offset + 2*(j&1);
1866 int sub_y_offset= y_offset + (j&2);
1867 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
1868 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1869 &weight_op[6], &weight_avg[6],
1870 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1871 }
1872 }
1873 }
1874 }
1875
1876 prefetch_motion(h, 1);
1877 }
1878
1879 static av_cold void init_cavlc_level_tab(void){
1880 int suffix_length, mask;
1881 unsigned int i;
1882
1883 for(suffix_length=0; suffix_length<7; suffix_length++){
1884 for(i=0; i<(1<<LEVEL_TAB_BITS); i++){
1885 int prefix= LEVEL_TAB_BITS - av_log2(2*i);
1886 int level_code= (prefix<<suffix_length) + (i>>(LEVEL_TAB_BITS-prefix-1-suffix_length)) - (1<<suffix_length);
1887
1888 mask= -(level_code&1);
1889 level_code= (((2+level_code)>>1) ^ mask) - mask;
1890 if(prefix + 1 + suffix_length <= LEVEL_TAB_BITS){
1891 cavlc_level_tab[suffix_length][i][0]= level_code;
1892 cavlc_level_tab[suffix_length][i][1]= prefix + 1 + suffix_length;
1893 }else if(prefix + 1 <= LEVEL_TAB_BITS){
1894 cavlc_level_tab[suffix_length][i][0]= prefix+100;
1895 cavlc_level_tab[suffix_length][i][1]= prefix + 1;
1896 }else{
1897 cavlc_level_tab[suffix_length][i][0]= LEVEL_TAB_BITS+100;
1898 cavlc_level_tab[suffix_length][i][1]= LEVEL_TAB_BITS;
1899 }
1900 }
1901 }
1902 }
1903
1904 static av_cold void decode_init_vlc(void){
1905 static int done = 0;
1906
1907 if (!done) {
1908 int i;
1909 int offset;
1910 done = 1;
1911
1912 chroma_dc_coeff_token_vlc.table = chroma_dc_coeff_token_vlc_table;
1913 chroma_dc_coeff_token_vlc.table_allocated = chroma_dc_coeff_token_vlc_table_size;
1914 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
1915 &chroma_dc_coeff_token_len [0], 1, 1,
1916 &chroma_dc_coeff_token_bits[0], 1, 1,
1917 INIT_VLC_USE_NEW_STATIC);
1918
1919 offset = 0;
1920 for(i=0; i<4; i++){
1921 coeff_token_vlc[i].table = coeff_token_vlc_tables+offset;
1922 coeff_token_vlc[i].table_allocated = coeff_token_vlc_tables_size[i];
1923 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
1924 &coeff_token_len [i][0], 1, 1,
1925 &coeff_token_bits[i][0], 1, 1,
1926 INIT_VLC_USE_NEW_STATIC);
1927 offset += coeff_token_vlc_tables_size[i];
1928 }
1929 /*
1930 * This is a one time safety check to make sure that
1931 * the packed static coeff_token_vlc table sizes
1932 * were initialized correctly.
1933 */
1934 assert(offset == FF_ARRAY_ELEMS(coeff_token_vlc_tables));
1935
1936 for(i=0; i<3; i++){
1937 chroma_dc_total_zeros_vlc[i].table = chroma_dc_total_zeros_vlc_tables[i];
1938 chroma_dc_total_zeros_vlc[i].table_allocated = chroma_dc_total_zeros_vlc_tables_size;
1939 init_vlc(&chroma_dc_total_zeros_vlc[i],
1940 CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
1941 &chroma_dc_total_zeros_len [i][0], 1, 1,
1942 &chroma_dc_total_zeros_bits[i][0], 1, 1,
1943 INIT_VLC_USE_NEW_STATIC);
1944 }
1945 for(i=0; i<15; i++){
1946 total_zeros_vlc[i].table = total_zeros_vlc_tables[i];
1947 total_zeros_vlc[i].table_allocated = total_zeros_vlc_tables_size;
1948 init_vlc(&total_zeros_vlc[i],
1949 TOTAL_ZEROS_VLC_BITS, 16,
1950 &total_zeros_len [i][0], 1, 1,
1951 &total_zeros_bits[i][0], 1, 1,
1952 INIT_VLC_USE_NEW_STATIC);
1953 }
1954
1955 for(i=0; i<6; i++){
1956 run_vlc[i].table = run_vlc_tables[i];
1957 run_vlc[i].table_allocated = run_vlc_tables_size;
1958 init_vlc(&run_vlc[i],
1959 RUN_VLC_BITS, 7,
1960 &run_len [i][0], 1, 1,
1961 &run_bits[i][0], 1, 1,
1962 INIT_VLC_USE_NEW_STATIC);
1963 }
1964 run7_vlc.table = run7_vlc_table,
1965 run7_vlc.table_allocated = run7_vlc_table_size;
1966 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
1967 &run_len [6][0], 1, 1,
1968 &run_bits[6][0], 1, 1,
1969 INIT_VLC_USE_NEW_STATIC);
1970
1971 init_cavlc_level_tab();
1972 }
1973 }
1974
1975 static void free_tables(H264Context *h){
1976 int i;
1977 H264Context *hx;
1978 av_freep(&h->intra4x4_pred_mode);
1979 av_freep(&h->chroma_pred_mode_table);
1980 av_freep(&h->cbp_table);
1981 av_freep(&h->mvd_table[0]);
1982 av_freep(&h->mvd_table[1]);
1983 av_freep(&h->direct_table);
1984 av_freep(&h->non_zero_count);
1985 av_freep(&h->slice_table_base);
1986 h->slice_table= NULL;
1987
1988 av_freep(&h->mb2b_xy);
1989 av_freep(&h->mb2b8_xy);
1990
1991 for(i = 0; i < h->s.avctx->thread_count; i++) {
1992 hx = h->thread_context[i];
1993 if(!hx) continue;
1994 av_freep(&hx->top_borders[1]);
1995 av_freep(&hx->top_borders[0]);
1996 av_freep(&hx->s.obmc_scratchpad);
1997 }
1998 }
1999
2000 static void init_dequant8_coeff_table(H264Context *h){
2001 int i,q,x;
2002 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
2003 h->dequant8_coeff[0] = h->dequant8_buffer[0];
2004 h->dequant8_coeff[1] = h->dequant8_buffer[1];
2005
2006 for(i=0; i<2; i++ ){
2007 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2008 h->dequant8_coeff[1] = h->dequant8_buffer[0];
2009 break;
2010 }
2011
2012 for(q=0; q<52; q++){
2013 int shift = div6[q];
2014 int idx = rem6[q];
2015 for(x=0; x<64; x++)
2016 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2017 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2018 h->pps.scaling_matrix8[i][x]) << shift;
2019 }
2020 }
2021 }
2022
2023 static void init_dequant4_coeff_table(H264Context *h){
2024 int i,j,q,x;
2025 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
2026 for(i=0; i<6; i++ ){
2027 h->dequant4_coeff[i] = h->dequant4_buffer[i];
2028 for(j=0; j<i; j++){
2029 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2030 h->dequant4_coeff[i] = h->dequant4_buffer[j];
2031 break;
2032 }
2033 }
2034 if(j<i)
2035 continue;
2036
2037 for(q=0; q<52; q++){
2038 int shift = div6[q] + 2;
2039 int idx = rem6[q];
2040 for(x=0; x<16; x++)
2041 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2042 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
2043 h->pps.scaling_matrix4[i][x]) << shift;
2044 }
2045 }
2046 }
2047
2048 static void init_dequant_tables(H264Context *h){
2049 int i,x;
2050 init_dequant4_coeff_table(h);
2051 if(h->pps.transform_8x8_mode)
2052 init_dequant8_coeff_table(h);
2053 if(h->sps.transform_bypass){
2054 for(i=0; i<6; i++)
2055 for(x=0; x<16; x++)
2056 h->dequant4_coeff[i][0][x] = 1<<6;
2057 if(h->pps.transform_8x8_mode)
2058 for(i=0; i<2; i++)
2059 for(x=0; x<64; x++)
2060 h->dequant8_coeff[i][0][x] = 1<<6;
2061 }
2062 }
2063
2064
2065 /**
2066 * allocates tables.
2067 * needs width/height
2068 */
2069 static int alloc_tables(H264Context *h){
2070 MpegEncContext * const s = &h->s;
2071 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2072 int x,y;
2073
2074 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2075
2076 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2077 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base))
2078 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2079
2080 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2081 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2082 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2083 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2084
2085 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
2086 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
2087
2088 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2089 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2090 for(y=0; y<s->mb_height; y++){
2091 for(x=0; x<s->mb_width; x++){
2092 const int mb_xy= x + y*s->mb_stride;
2093 const int b_xy = 4*x + 4*y*h->b_stride;
2094 const int b8_xy= 2*x + 2*y*h->b8_stride;
2095
2096 h->mb2b_xy [mb_xy]= b_xy;
2097 h->mb2b8_xy[mb_xy]= b8_xy;
2098 }
2099 }
2100
2101 s->obmc_scratchpad = NULL;
2102
2103 if(!h->dequant4_coeff[0])
2104 init_dequant_tables(h);
2105
2106 return 0;
2107 fail:
2108 free_tables(h);
2109 return -1;
2110 }
2111
2112 /**
2113 * Mimic alloc_tables(), but for every context thread.
2114 */
2115 static void clone_tables(H264Context *dst, H264Context *src){
2116 dst->intra4x4_pred_mode = src->intra4x4_pred_mode;
2117 dst->non_zero_count = src->non_zero_count;
2118 dst->slice_table = src->slice_table;
2119 dst->cbp_table = src->cbp_table;
2120 dst->mb2b_xy = src->mb2b_xy;
2121 dst->mb2b8_xy = src->mb2b8_xy;
2122 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
2123 dst->mvd_table[0] = src->mvd_table[0];
2124 dst->mvd_table[1] = src->mvd_table[1];
2125 dst->direct_table = src->direct_table;
2126
2127 dst->s.obmc_scratchpad = NULL;
2128 ff_h264_pred_init(&dst->hpc, src->s.codec_id);
2129 }
2130
2131 /**
2132 * Init context
2133 * Allocate buffers which are not shared amongst multiple threads.
2134 */
2135 static int context_init(H264Context *h){
2136 CHECKED_ALLOCZ(h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2137 CHECKED_ALLOCZ(h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t))
2138
2139 return 0;
2140 fail:
2141 return -1; // free_tables will clean up for us
2142 }
2143
2144 static av_cold void common_init(H264Context *h){
2145 MpegEncContext * const s = &h->s;
2146
2147 s->width = s->avctx->width;
2148 s->height = s->avctx->height;
2149 s->codec_id= s->avctx->codec->id;
2150
2151 ff_h264_pred_init(&h->hpc, s->codec_id);
2152
2153 h->dequant_coeff_pps= -1;
2154 s->unrestricted_mv=1;
2155 s->decode=1; //FIXME
2156
2157 dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
2158
2159 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
2160 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
2161 }
2162
2163 static av_cold int decode_init(AVCodecContext *avctx){
2164 H264Context *h= avctx->priv_data;
2165 MpegEncContext * const s = &h->s;
2166
2167 MPV_decode_defaults(s);
2168
2169 s->avctx = avctx;
2170 common_init(h);
2171
2172 s->out_format = FMT_H264;
2173 s->workaround_bugs= avctx->workaround_bugs;
2174
2175 // set defaults
2176 // s->decode_mb= ff_h263_decode_mb;
2177 s->quarter_sample = 1;
2178 s->low_delay= 1;
2179
2180 if(avctx->codec_id == CODEC_ID_SVQ3)
2181 avctx->pix_fmt= PIX_FMT_YUVJ420P;
2182 else if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
2183 avctx->pix_fmt= PIX_FMT_VDPAU_H264;
2184 else
2185 avctx->pix_fmt= PIX_FMT_YUV420P;
2186
2187 decode_init_vlc();
2188
2189 if(avctx->extradata_size > 0 && avctx->extradata &&
2190 *(char *)avctx->extradata == 1){
2191 h->is_avc = 1;
2192 h->got_avcC = 0;
2193 } else {
2194 h->is_avc = 0;
2195 }
2196
2197 h->thread_context[0] = h;
2198 h->outputed_poc = INT_MIN;
2199 h->prev_poc_msb= 1<<16;
2200 h->sei_recovery_frame_cnt = -1;
2201 h->sei_dpb_output_delay = 0;
2202 h->sei_cpb_removal_delay = -1;
2203 h->sei_buffering_period_present = 0;
2204 return 0;
2205 }
2206
2207 static int frame_start(H264Context *h){
2208 MpegEncContext * const s = &h->s;
2209 int i;
2210
2211 if(MPV_frame_start(s, s->avctx) < 0)
2212 return -1;
2213 ff_er_frame_start(s);
2214 /*
2215 * MPV_frame_start uses pict_type to derive key_frame.
2216 * This is incorrect for H.264; IDR markings must be used.
2217 * Zero here; IDR markings per slice in frame or fields are ORed in later.
2218 * See decode_nal_units().
2219 */
2220 s->current_picture_ptr->key_frame= 0;
2221
2222 assert(s->linesize && s->uvlinesize);
2223
2224 for(i=0; i<16; i++){
2225 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2226 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2227 }
2228 for(i=0; i<4; i++){
2229 h->block_offset[16+i]=
2230 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2231 h->block_offset[24+16+i]=
2232 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2233 }
2234
2235 /* can't be in alloc_tables because linesize isn't known there.
2236 * FIXME: redo bipred weight to not require extra buffer? */
2237 for(i = 0; i < s->avctx->thread_count; i++)
2238 if(!h->thread_context[i]->s.obmc_scratchpad)
2239 h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
2240
2241 /* some macroblocks will be accessed before they're available */
2242 if(FRAME_MBAFF || s->avctx->thread_count > 1)
2243 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
2244
2245 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2246
2247 // We mark the current picture as non-reference after allocating it, so
2248 // that if we break out due to an error it can be released automatically
2249 // in the next MPV_frame_start().
2250 // SVQ3 as well as most other codecs have only last/next/current and thus
2251 // get released even with set reference, besides SVQ3 and others do not
2252 // mark frames as reference later "naturally".
2253 if(s->codec_id != CODEC_ID_SVQ3)
2254 s->current_picture_ptr->reference= 0;
2255
2256 s->current_picture_ptr->field_poc[0]=
2257 s->current_picture_ptr->field_poc[1]= INT_MAX;
2258 assert(s->current_picture_ptr->long_ref==0);
2259
2260 return 0;
2261 }
2262
2263 static 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){
2264 MpegEncContext * const s = &h->s;
2265 int i;
2266 int step = 1;
2267 int offset = 1;
2268 int uvoffset= 1;
2269 int top_idx = 1;
2270 int skiplast= 0;
2271
2272 src_y -= linesize;
2273 src_cb -= uvlinesize;
2274 src_cr -= uvlinesize;
2275
2276 if(!simple && FRAME_MBAFF){
2277 if(s->mb_y&1){
2278 offset = MB_MBAFF ? 1 : 17;
2279 uvoffset= MB_MBAFF ? 1 : 9;
2280 if(!MB_MBAFF){
2281 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 0)= *(uint64_t*)(src_y + 15*linesize);
2282 *(uint64_t*)(h->top_borders[0][s->mb_x]+ 8)= *(uint64_t*)(src_y +8+15*linesize);
2283 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2284 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+7*uvlinesize);
2285 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+7*uvlinesize);
2286 }
2287 }
2288 }else{
2289 if(!MB_MBAFF){
2290 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2291 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2292 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7 ];
2293 h->left_border[34+18]= h->top_borders[0][s->mb_x][16+8+7];
2294 }
2295 skiplast= 1;
2296 }
2297 offset =
2298 uvoffset=
2299 top_idx = MB_MBAFF ? 0 : 1;
2300 }
2301 step= MB_MBAFF ? 2 : 1;
2302 }
2303
2304 // There are two lines saved, the line above the the top macroblock of a pair,
2305 // and the line above the bottom macroblock
2306 h->left_border[offset]= h->top_borders[top_idx][s->mb_x][15];
2307 for(i=1; i<17 - skiplast; i++){
2308 h->left_border[offset+i*step]= src_y[15+i* linesize];
2309 }
2310
2311 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2312 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2313
2314 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2315 h->left_border[uvoffset+34 ]= h->top_borders[top_idx][s->mb_x][16+7];
2316 h->left_border[uvoffset+34+18]= h->top_borders[top_idx][s->mb_x][24+7];
2317 for(i=1; i<9 - skiplast; i++){
2318 h->left_border[uvoffset+34 +i*step]= src_cb[7+i*uvlinesize];
2319 h->left_border[uvoffset+34+18+i*step]= src_cr[7+i*uvlinesize];
2320 }
2321 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2322 *(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2323 }
2324 }
2325
2326 static 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){
2327 MpegEncContext * const s = &h->s;
2328 int temp8, i;
2329 uint64_t temp64;
2330 int deblock_left;
2331 int deblock_top;
2332 int mb_xy;
2333 int step = 1;
2334 int offset = 1;
2335 int uvoffset= 1;
2336 int top_idx = 1;
2337
2338 if(!simple && FRAME_MBAFF){
2339 if(s->mb_y&1){
2340 offset = MB_MBAFF ? 1 : 17;
2341 uvoffset= MB_MBAFF ? 1 : 9;
2342 }else{
2343 offset =
2344 uvoffset=
2345 top_idx = MB_MBAFF ? 0 : 1;
2346 }
2347 step= MB_MBAFF ? 2 : 1;
2348 }
2349
2350 if(h->deblocking_filter == 2) {
2351 mb_xy = h->mb_xy;
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 > !!MB_FIELD);
2357 }
2358
2359 src_y -= linesize + 1;
2360 src_cb -= uvlinesize + 1;
2361 src_cr -= uvlinesize + 1;
2362
2363 #define XCHG(a,b,t,xchg)\
2364 t= a;\
2365 if(xchg)\
2366 a= b;\
2367 b= t;
2368
2369 if(deblock_left){
2370 for(i = !deblock_top; i<16; i++){
2371 XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, xchg);
2372 }
2373 XCHG(h->left_border[offset+i*step], src_y [i* linesize], temp8, 1);
2374 }
2375
2376 if(deblock_top){
2377 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2378 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2379 if(s->mb_x+1 < s->mb_width){
2380 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
2381 }
2382 }
2383
2384 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2385 if(deblock_left){
2386 for(i = !deblock_top; i<8; i++){
2387 XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, xchg);
2388 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, xchg);
2389 }
2390 XCHG(h->left_border[uvoffset+34 +i*step], src_cb[i*uvlinesize], temp8, 1);
2391 XCHG(h->left_border[uvoffset+34+18+i*step], src_cr[i*uvlinesize], temp8, 1);
2392 }
2393 if(deblock_top){
2394 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2395 XCHG(*(uint64_t*)(h->top_borders[top_idx][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2396 }
2397 }
2398 }
2399
2400 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
2401 MpegEncContext * const s = &h->s;
2402 const int mb_x= s->mb_x;
2403 const int mb_y= s->mb_y;
2404 const int mb_xy= h->mb_xy;
2405 const int mb_type= s->current_picture.mb_type[mb_xy];
2406 uint8_t *dest_y, *dest_cb, *dest_cr;
2407 int linesize, uvlinesize /*dct_offset*/;
2408 int i;
2409 int *block_offset = &h->block_offset[0];
2410 const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2411 /* is_h264 should always be true if SVQ3 is disabled. */
2412 const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
2413 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
2414 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
2415
2416 dest_y = s->current_picture.data[0] + (mb_x + mb_y * s->linesize ) * 16;
2417 dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
2418 dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
2419
2420 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
2421 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
2422
2423 if (!simple && MB_FIELD) {
2424 linesize = h->mb_linesize = s->linesize * 2;
2425 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
2426 block_offset = &h->block_offset[24];
2427 if(mb_y&1){ //FIXME move out of this function?
2428 dest_y -= s->linesize*15;
2429 dest_cb-= s->uvlinesize*7;
2430 dest_cr-= s->uvlinesize*7;
2431 }
2432 if(FRAME_MBAFF) {
2433 int list;
2434 for(list=0; list<h->list_count; list++){
2435 if(!USES_LIST(mb_type, list))
2436 continue;
2437 if(IS_16X16(mb_type)){
2438 int8_t *ref = &h->ref_cache[list][scan8[0]];
2439 fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2440 }else{
2441 for(i=0; i<16; i+=4){
2442 int ref = h->ref_cache[list][scan8[i]];
2443 if(ref >= 0)
2444 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2445 }
2446 }
2447 }
2448 }
2449 } else {
2450 linesize = h->mb_linesize = s->linesize;
2451 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
2452 // dct_offset = s->linesize * 16;
2453 }
2454
2455 if (!simple && IS_INTRA_PCM(mb_type)) {
2456 for (i=0; i<16; i++) {
2457 memcpy(dest_y + i* linesize, h->mb + i*8, 16);
2458 }
2459 for (i=0; i<8; i++) {
2460 memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4, 8);
2461 memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4, 8);
2462 }
2463 } else {
2464 if(IS_INTRA(mb_type)){
2465 if(h->deblocking_filter)
2466 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
2467
2468 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2469 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2470 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2471 }
2472
2473 if(IS_INTRA4x4(mb_type)){
2474 if(simple || !s->encoding){
2475 if(IS_8x8DCT(mb_type)){
2476 if(transform_bypass){
2477 idct_dc_add =
2478 idct_add = s->dsp.add_pixels8;
2479 }else{
2480 idct_dc_add = s->dsp.h264_idct8_dc_add;
2481 idct_add = s->dsp.h264_idct8_add;
2482 }
2483 for(i=0; i<16; i+=4){
2484 uint8_t * const ptr= dest_y + block_offset[i];
2485 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2486 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2487 h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
2488 }else{
2489 const int nnz = h->non_zero_count_cache[ scan8[i] ];
2490 h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
2491 (h->topright_samples_available<<i)&0x4000, linesize);
2492 if(nnz){
2493 if(nnz == 1 && h->mb[i*16])
2494 idct_dc_add(ptr, h->mb + i*16, linesize);
2495 else
2496 idct_add (ptr, h->mb + i*16, linesize);
2497 }
2498 }
2499 }
2500 }else{
2501 if(transform_bypass){
2502 idct_dc_add =
2503 idct_add = s->dsp.add_pixels4;
2504 }else{
2505 idct_dc_add = s->dsp.h264_idct_dc_add;
2506 idct_add = s->dsp.h264_idct_add;
2507 }
2508 for(i=0; i<16; i++){
2509 uint8_t * const ptr= dest_y + block_offset[i];
2510 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2511
2512 if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
2513 h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
2514 }else{
2515 uint8_t *topright;
2516 int nnz, tr;
2517 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2518 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2519 assert(mb_y || linesize <= block_offset[i]);
2520 if(!topright_avail){
2521 tr= ptr[3 - linesize]*0x01010101;
2522 topright= (uint8_t*) &tr;
2523 }else
2524 topright= ptr + 4 - linesize;
2525 }else
2526 topright= NULL;
2527
2528 h->hpc.pred4x4[ dir ](ptr, topright, linesize);
2529 nnz = h->non_zero_count_cache[ scan8[i] ];
2530 if(nnz){
2531 if(is_h264){
2532 if(nnz == 1 && h->mb[i*16])
2533 idct_dc_add(ptr, h->mb + i*16, linesize);
2534 else
2535 idct_add (ptr, h->mb + i*16, linesize);
2536 }else
2537 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2538 }
2539 }
2540 }
2541 }
2542 }
2543 }else{
2544 h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2545 if(is_h264){
2546 if(!transform_bypass