2 * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
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.
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.
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
23 * @file libavcodec/h264_direct.c
24 * H.264 / AVC / MPEG4 part10 direct mb/block decoding.
25 * @author Michael Niedermayer <michaelni@gmx.at>
31 #include "mpegvideo.h"
33 #include "rectangle.h"
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){
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);
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;
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);
63 for(i=0; i<h->ref_count[0]; i++){
64 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
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;
76 /* bogus; fills in for missing frames */
77 memset(map[list], 0, sizeof(map[list]));
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];
85 else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
86 poc= (poc&~3) + rfield + 1;
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;
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;
106 int sidx= (s->picture_structure&1)^1;
107 int ref1sidx= (ref1->reference&1)^1;
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);
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]));
120 cur->mbaff= FRAME_MBAFF;
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);
132 if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
135 for(list=0; list<2; list++){
136 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
138 for(field=0; field<2; field++)
139 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
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;
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;;
158 assert(h->ref_list[1][0].reference&3);
160 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
163 /* ref = min(neighbors) */
164 for(list=0; list<2; list++){
165 int left_ref = h->ref_cache[list][scan8[0] - 1];
166 int top_ref = h->ref_cache[list][scan8[0] - 8];
167 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
168 const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
169 if(refc == PART_NOT_AVAILABLE){
170 refc = h->ref_cache[list][scan8[0] - 8 - 1];
171 C = h-> mv_cache[list][scan8[0] - 8 - 1];
173 ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
175 //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
176 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
177 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
179 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
180 if(match_count > 1){ //most common
181 mv[list]= (mid_pred(A[0], B[0], C[0])&0xFFFF)
182 +(mid_pred(A[1], B[1], C[1])<<16);
184 assert(match_count==1);
185 if(left_ref==ref[list]){
186 mv[list]= *(uint32_t*)A;
187 }else if(top_ref==ref[list]){
188 mv[list]= *(uint32_t*)B;
190 mv[list]= *(uint32_t*)C;
194 int mask= ~(MB_TYPE_L0 << (2*list));
202 if(ref[0] < 0 && ref[1] < 0){
205 *mb_type |= MB_TYPE_L0L1;
206 sub_mb_type |= MB_TYPE_L0L1;
209 if(!(is_b8x8|mv[0]|mv[1])){
210 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
211 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
212 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
213 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
214 *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;
218 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
219 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL
220 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
223 mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
226 }else{ // AFL/AFR/FR/FL -> AFR/FR
227 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
228 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
229 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
230 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
234 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
235 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
236 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
238 *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
240 *mb_type |= MB_TYPE_8x8;
242 }else{ // AFR/FR -> AFR/FR
245 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
247 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
248 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
249 *mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
250 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
251 *mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
253 if(!h->sps.direct_8x8_inference_flag){
254 /* FIXME save sub mb types from previous frames (or derive from MVs)
255 * so we know exactly what block size to use */
256 sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
258 *mb_type |= MB_TYPE_8x8;
263 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
264 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
265 l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
266 l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
269 l1ref0 += h->b8_stride;
270 l1ref1 += h->b8_stride;
271 l1mv0 += 2*b4_stride;
272 l1mv1 += 2*b4_stride;
277 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
279 for(i8=0; i8<4; i8++){
282 int xy8 = x8+y8*b8_stride;
283 int xy4 = 3*x8+y8*b4_stride;
286 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
288 h->sub_mb_type[i8] = sub_mb_type;
290 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
291 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
292 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
293 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
294 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
305 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
306 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
308 if(!is_b8x8 && !(n&3))
309 *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;
310 }else if(IS_16X16(*mb_type)){
313 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
314 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
315 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
316 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
317 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
318 && h->x264_build>33U))){
328 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
329 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
332 for(i8=0; i8<4; i8++){
334 const int y8 = i8>>1;
336 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
338 h->sub_mb_type[i8] = sub_mb_type;
340 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
341 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
342 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
343 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
346 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[x8 + y8*b8_stride] == 0
347 || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
348 && h->x264_build>33U))){
349 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
350 if(IS_SUB_8X8(sub_mb_type)){
351 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
352 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
354 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
356 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
361 for(i4=0; i4<4; i4++){
362 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
363 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
365 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
367 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
372 h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
377 if(!is_b8x8 && !(n&15))
378 *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;
382 static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
383 MpegEncContext * const s = &h->s;
384 int b8_stride = h->b8_stride;
385 int b4_stride = h->b_stride;
386 int mb_xy = h->mb_xy;
388 const int16_t (*l1mv0)[2], (*l1mv1)[2];
389 const int8_t *l1ref0, *l1ref1;
390 const int is_b8x8 = IS_8X8(*mb_type);
391 unsigned int sub_mb_type;
394 assert(h->ref_list[1][0].reference&3);
396 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
397 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL
398 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
401 mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
404 }else{ // AFL/AFR/FR/FL -> AFR/FR
405 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
406 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
407 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
408 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
412 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
414 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
415 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
417 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
419 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
421 }else{ // AFR/FR -> AFR/FR
424 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
426 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
427 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
428 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
429 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
430 *mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
432 if(!h->sps.direct_8x8_inference_flag){
433 /* FIXME save sub mb types from previous frames (or derive from MVs)
434 * so we know exactly what block size to use */
435 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
437 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
442 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
443 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
444 l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
445 l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
448 l1ref0 += h->b8_stride;
449 l1ref1 += h->b8_stride;
450 l1mv0 += 2*b4_stride;
451 l1mv1 += 2*b4_stride;
456 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
457 const int *dist_scale_factor = h->dist_scale_factor;
460 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
461 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
462 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
463 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
465 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 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
468 int y_shift = 2*!IS_INTERLACED(*mb_type);
469 assert(h->sps.direct_8x8_inference_flag);
471 for(i8=0; i8<4; i8++){
473 const int y8 = i8>>1;
475 const int16_t (*l1mv)[2]= l1mv0;
477 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
479 h->sub_mb_type[i8] = sub_mb_type;
481 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
482 if(IS_INTRA(mb_type_col[y8])){
483 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
484 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
485 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
489 ref0 = l1ref0[x8 + y8*b8_stride];
491 ref0 = map_col_to_list0[0][ref0 + ref_offset];
493 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
496 scale = dist_scale_factor[ref0];
497 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
500 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
501 int my_col = (mv_col[1]<<y_shift)/2;
502 int mx = (scale * mv_col[0] + 128) >> 8;
503 int my = (scale * my_col + 128) >> 8;
504 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
505 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
511 /* one-to-one mv scaling */
513 if(IS_16X16(*mb_type)){
516 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
517 if(IS_INTRA(mb_type_col[0])){
520 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
521 : map_col_to_list0[1][l1ref1[0] + ref_offset];
522 const int scale = dist_scale_factor[ref0];
523 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
525 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
526 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
528 mv0= pack16to32(mv_l0[0],mv_l0[1]);
529 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
531 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
532 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
533 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
535 for(i8=0; i8<4; i8++){
537 const int y8 = i8>>1;
539 const int16_t (*l1mv)[2]= l1mv0;
541 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
543 h->sub_mb_type[i8] = sub_mb_type;
544 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
545 if(IS_INTRA(mb_type_col[0])){
546 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
547 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
548 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
552 ref0 = l1ref0[x8 + y8*b8_stride];
554 ref0 = map_col_to_list0[0][ref0 + ref_offset];
556 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
559 scale = dist_scale_factor[ref0];
561 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
562 if(IS_SUB_8X8(sub_mb_type)){
563 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
564 int mx = (scale * mv_col[0] + 128) >> 8;
565 int my = (scale * mv_col[1] + 128) >> 8;
566 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
567 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
569 for(i4=0; i4<4; i4++){
570 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
571 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
572 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
573 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
574 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
575 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
582 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
583 if(h->direct_spatial_mv_pred){
584 pred_spatial_direct_motion(h, mb_type);
586 pred_temp_direct_motion(h, mb_type);