bf944a1f393af55d8b307735884868b946ee15c0
[libav.git] / libavcodec / h264_direct.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
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_direct.c
24 * H.264 / AVC / MPEG4 part10 direct mb/block decoding.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #include "internal.h"
29 #include "dsputil.h"
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "h264.h"
33 #include "rectangle.h"
34
35 //#undef NDEBUG
36 #include <assert.h>
37
38
39 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
40 int poc0 = h->ref_list[0][i].poc;
41 int td = av_clip(poc1 - poc0, -128, 127);
42 if(td == 0 || h->ref_list[0][i].long_ref){
43 return 256;
44 }else{
45 int tb = av_clip(poc - poc0, -128, 127);
46 int tx = (16384 + (FFABS(td) >> 1)) / td;
47 return av_clip((tb*tx + 32) >> 6, -1024, 1023);
48 }
49 }
50
51 void ff_h264_direct_dist_scale_factor(H264Context * const h){
52 MpegEncContext * const s = &h->s;
53 const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
54 const int poc1 = h->ref_list[1][0].poc;
55 int i, field;
56 for(field=0; field<2; field++){
57 const int poc = h->s.current_picture_ptr->field_poc[field];
58 const int poc1 = h->ref_list[1][0].field_poc[field];
59 for(i=0; i < 2*h->ref_count[0]; i++)
60 h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
61 }
62
63 for(i=0; i<h->ref_count[0]; i++){
64 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
65 }
66 }
67
68 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
69 MpegEncContext * const s = &h->s;
70 Picture * const ref1 = &h->ref_list[1][0];
71 int j, old_ref, rfield;
72 int start= mbafi ? 16 : 0;
73 int end = mbafi ? 16+2*h->ref_count[0] : h->ref_count[0];
74 int interl= mbafi || s->picture_structure != PICT_FRAME;
75
76 /* bogus; fills in for missing frames */
77 memset(map[list], 0, sizeof(map[list]));
78
79 for(rfield=0; rfield<2; rfield++){
80 for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
81 int poc = ref1->ref_poc[colfield][list][old_ref];
82
83 if (!interl)
84 poc |= 3;
85 else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
86 poc= (poc&~3) + rfield + 1;
87
88 for(j=start; j<end; j++){
89 if(4*h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference&3) == poc){
90 int cur_ref= mbafi ? (j-16)^field : j;
91 map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
92 if(rfield == field || !interl)
93 map[list][old_ref] = cur_ref;
94 break;
95 }
96 }
97 }
98 }
99 }
100
101 void ff_h264_direct_ref_list_init(H264Context * const h){
102 MpegEncContext * const s = &h->s;
103 Picture * const ref1 = &h->ref_list[1][0];
104 Picture * const cur = s->current_picture_ptr;
105 int list, j, field;
106 int sidx= (s->picture_structure&1)^1;
107 int ref1sidx= (ref1->reference&1)^1;
108
109 for(list=0; list<2; list++){
110 cur->ref_count[sidx][list] = h->ref_count[list];
111 for(j=0; j<h->ref_count[list]; j++)
112 cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
113 }
114
115 if(s->picture_structure == PICT_FRAME){
116 memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
117 memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0]));
118 }
119
120 cur->mbaff= FRAME_MBAFF;
121
122 h->col_fieldoff= 0;
123 if(s->picture_structure == PICT_FRAME){
124 int cur_poc = s->current_picture_ptr->poc;
125 int *col_poc = h->ref_list[1]->field_poc;
126 h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
127 ref1sidx=sidx= h->col_parity;
128 }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){ // FL -> FL & differ parity
129 h->col_fieldoff= s->mb_stride*(2*(h->ref_list[1][0].reference) - 3);
130 }
131
132 if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
133 return;
134
135 for(list=0; list<2; list++){
136 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
137 if(FRAME_MBAFF)
138 for(field=0; field<2; field++)
139 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
140 }
141 }
142
143 static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
144 MpegEncContext * const s = &h->s;
145 int b8_stride = h->b8_stride;
146 int b4_stride = h->b_stride;
147 int mb_xy = h->mb_xy;
148 int mb_type_col[2];
149 const int16_t (*l1mv0)[2], (*l1mv1)[2];
150 const int8_t *l1ref0, *l1ref1;
151 const int is_b8x8 = IS_8X8(*mb_type);
152 unsigned int sub_mb_type= MB_TYPE_L0L1;;
153 int i8, i4;
154 int ref[2];
155 int mv[2];
156 int list;
157
158 assert(h->ref_list[1][0].reference&3);
159
160 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
161
162 *mb_type |= MB_TYPE_L0L1;
163
164 /* ref = min(neighbors) */
165 for(list=0; list<2; list++){
166 int left_ref = h->ref_cache[list][scan8[0] - 1];
167 int top_ref = h->ref_cache[list][scan8[0] - 8];
168 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
169 const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
170 if(refc == PART_NOT_AVAILABLE){
171 refc = h->ref_cache[list][scan8[0] - 8 - 1];
172 C = h-> mv_cache[list][scan8[0] - 8 - 1];
173 }
174 ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
175 if(ref[list] >= 0){
176 //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
177 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
178 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
179
180 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
181 if(match_count > 1){ //most common
182 mv[list]= (mid_pred(A[0], B[0], C[0])&0xFFFF)
183 +(mid_pred(A[1], B[1], C[1])<<16);
184 }else {
185 assert(match_count==1);
186 if(left_ref==ref[list]){
187 mv[list]= *(uint32_t*)A;
188 }else if(top_ref==ref[list]){
189 mv[list]= *(uint32_t*)B;
190 }else{
191 mv[list]= *(uint32_t*)C;
192 }
193 }
194 }else{
195 int mask= ~(MB_TYPE_L0 << (2*list));
196 mv[list] = 0;
197 ref[list] = -1;
198 if(!is_b8x8)
199 *mb_type &= mask;
200 sub_mb_type &= mask;
201 }
202 }
203 if(ref[0] < 0 && ref[1] < 0){
204 ref[0] = ref[1] = 0;
205 if(!is_b8x8)
206 *mb_type |= MB_TYPE_L0L1;
207 sub_mb_type |= MB_TYPE_L0L1;
208 }
209
210 if(!(is_b8x8|mv[0]|mv[1])){
211 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
212 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
213 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
214 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
215 *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
216 return;
217 }
218
219 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
220 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL
221 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
222 b8_stride = 0;
223 }else{
224 mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
225 }
226 goto single_col;
227 }else{ // AFL/AFR/FR/FL -> AFR/FR
228 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
229 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
230 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
231 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
232 b8_stride *= 3;
233 b4_stride *= 6;
234
235 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
236 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
237 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
238 && !is_b8x8){
239 *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
240 }else{
241 *mb_type |= MB_TYPE_8x8;
242 }
243 }else{ // AFR/FR -> AFR/FR
244 single_col:
245 mb_type_col[0] =
246 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
247
248 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
249 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
250 *mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
251 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
252 *mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
253 }else{
254 if(!h->sps.direct_8x8_inference_flag){
255 /* FIXME save sub mb types from previous frames (or derive from MVs)
256 * so we know exactly what block size to use */
257 sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
258 }
259 *mb_type |= MB_TYPE_8x8;
260 }
261 }
262 }
263
264 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
265 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
266 l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
267 l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
268 if(!b8_stride){
269 if(s->mb_y&1){
270 l1ref0 += h->b8_stride;
271 l1ref1 += h->b8_stride;
272 l1mv0 += 2*b4_stride;
273 l1mv1 += 2*b4_stride;
274 }
275 }
276
277
278 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
279 int n=0;
280 for(i8=0; i8<4; i8++){
281 int x8 = i8&1;
282 int y8 = i8>>1;
283 int xy8 = x8+y8*b8_stride;
284 int xy4 = 3*x8+y8*b4_stride;
285 int a,b;
286
287 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
288 continue;
289 h->sub_mb_type[i8] = sub_mb_type;
290
291 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
292 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
293 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
294 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
295 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
296 a=b=0;
297 if(ref[0] > 0)
298 a= mv[0];
299 if(ref[1] > 0)
300 b= mv[1];
301 n++;
302 }else{
303 a= mv[0];
304 b= mv[1];
305 }
306 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
307 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
308 }
309 if(!is_b8x8 && !(n&3))
310 *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
311 }else if(IS_16X16(*mb_type)){
312 int a,b;
313
314 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
315 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
316 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
317 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
318 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
319 && h->x264_build>33U))){
320 a=b=0;
321 if(ref[0] > 0)
322 a= mv[0];
323 if(ref[1] > 0)
324 b= mv[1];
325 }else{
326 a= mv[0];
327 b= mv[1];
328 }
329 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
330 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
331 }else{
332 int n=0;
333 for(i8=0; i8<4; i8++){
334 const int x8 = i8&1;
335 const int y8 = i8>>1;
336
337 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
338 continue;
339 h->sub_mb_type[i8] = sub_mb_type;
340
341 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
342 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
343 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
344 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
345
346 /* col_zero_flag */
347 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[x8 + y8*b8_stride] == 0
348 || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
349 && h->x264_build>33U))){
350 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
351 if(IS_SUB_8X8(sub_mb_type)){
352 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
353 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
354 if(ref[0] == 0)
355 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
356 if(ref[1] == 0)
357 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
358 n+=4;
359 }
360 }else{
361 int m=0;
362 for(i4=0; i4<4; i4++){
363 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
364 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
365 if(ref[0] == 0)
366 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
367 if(ref[1] == 0)
368 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
369 m++;
370 }
371 }
372 if(!(m&3))
373 h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
374 n+=m;
375 }
376 }
377 }
378 if(!is_b8x8 && !(n&15))
379 *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
380 }
381 }
382
383 static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
384 MpegEncContext * const s = &h->s;
385 int b8_stride = h->b8_stride;
386 int b4_stride = h->b_stride;
387 int mb_xy = h->mb_xy;
388 int mb_type_col[2];
389 const int16_t (*l1mv0)[2], (*l1mv1)[2];
390 const int8_t *l1ref0, *l1ref1;
391 const int is_b8x8 = IS_8X8(*mb_type);
392 unsigned int sub_mb_type;
393 int i8, i4;
394
395 assert(h->ref_list[1][0].reference&3);
396
397 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
398 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL
399 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
400 b8_stride = 0;
401 }else{
402 mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
403 }
404 goto single_col;
405 }else{ // AFL/AFR/FR/FL -> AFR/FR
406 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
407 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
408 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
409 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
410 b8_stride *= 3;
411 b4_stride *= 6;
412
413 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
414
415 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
416 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
417 && !is_b8x8){
418 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
419 }else{
420 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
421 }
422 }else{ // AFR/FR -> AFR/FR
423 single_col:
424 mb_type_col[0] =
425 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
426
427 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
428 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
429 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
430 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
431 *mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
432 }else{
433 if(!h->sps.direct_8x8_inference_flag){
434 /* FIXME save sub mb types from previous frames (or derive from MVs)
435 * so we know exactly what block size to use */
436 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
437 }
438 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
439 }
440 }
441 }
442
443 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
444 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
445 l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
446 l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
447 if(!b8_stride){
448 if(s->mb_y&1){
449 l1ref0 += h->b8_stride;
450 l1ref1 += h->b8_stride;
451 l1mv0 += 2*b4_stride;
452 l1mv1 += 2*b4_stride;
453 }
454 }
455
456 {
457 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
458 const int *dist_scale_factor = h->dist_scale_factor;
459 int ref_offset;
460
461 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
462 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
463 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
464 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
465 }
466 ref_offset = (h->ref_list[1][0].mbaff<<4) & (mb_type_col[0]>>3); //if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0])) ref_offset=16 else 0
467
468 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
469 int y_shift = 2*!IS_INTERLACED(*mb_type);
470 assert(h->sps.direct_8x8_inference_flag);
471
472 for(i8=0; i8<4; i8++){
473 const int x8 = i8&1;
474 const int y8 = i8>>1;
475 int ref0, scale;
476 const int16_t (*l1mv)[2]= l1mv0;
477
478 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
479 continue;
480 h->sub_mb_type[i8] = sub_mb_type;
481
482 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
483 if(IS_INTRA(mb_type_col[y8])){
484 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
485 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
486 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
487 continue;
488 }
489
490 ref0 = l1ref0[x8 + y8*b8_stride];
491 if(ref0 >= 0)
492 ref0 = map_col_to_list0[0][ref0 + ref_offset];
493 else{
494 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
495 l1mv= l1mv1;
496 }
497 scale = dist_scale_factor[ref0];
498 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
499
500 {
501 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
502 int my_col = (mv_col[1]<<y_shift)/2;
503 int mx = (scale * mv_col[0] + 128) >> 8;
504 int my = (scale * my_col + 128) >> 8;
505 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
506 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
507 }
508 }
509 return;
510 }
511
512 /* one-to-one mv scaling */
513
514 if(IS_16X16(*mb_type)){
515 int ref, mv0, mv1;
516
517 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
518 if(IS_INTRA(mb_type_col[0])){
519 ref=mv0=mv1=0;
520 }else{
521 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
522 : map_col_to_list0[1][l1ref1[0] + ref_offset];
523 const int scale = dist_scale_factor[ref0];
524 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
525 int mv_l0[2];
526 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
527 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
528 ref= ref0;
529 mv0= pack16to32(mv_l0[0],mv_l0[1]);
530 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
531 }
532 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
533 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
534 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
535 }else{
536 for(i8=0; i8<4; i8++){
537 const int x8 = i8&1;
538 const int y8 = i8>>1;
539 int ref0, scale;
540 const int16_t (*l1mv)[2]= l1mv0;
541
542 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
543 continue;
544 h->sub_mb_type[i8] = sub_mb_type;
545 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
546 if(IS_INTRA(mb_type_col[0])){
547 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
548 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
549 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
550 continue;
551 }
552
553 ref0 = l1ref0[x8 + y8*b8_stride];
554 if(ref0 >= 0)
555 ref0 = map_col_to_list0[0][ref0 + ref_offset];
556 else{
557 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
558 l1mv= l1mv1;
559 }
560 scale = dist_scale_factor[ref0];
561
562 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
563 if(IS_SUB_8X8(sub_mb_type)){
564 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
565 int mx = (scale * mv_col[0] + 128) >> 8;
566 int my = (scale * mv_col[1] + 128) >> 8;
567 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
568 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
569 }else
570 for(i4=0; i4<4; i4++){
571 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
572 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
573 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
574 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
575 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
576 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
577 }
578 }
579 }
580 }
581 }
582
583 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
584 if(h->direct_spatial_mv_pred){
585 pred_spatial_direct_motion(h, mb_type);
586 }else{
587 pred_temp_direct_motion(h, mb_type);
588 }
589 }