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
24 * H.264 / AVC / MPEG4 part10 direct mb/block decoding.
25 * @author Michael Niedermayer <michaelni@gmx.at>
30 #include "mpegvideo.h"
32 #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 const int poc = h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD];
53 const int poc1 = h->ref_list[1][0].poc;
57 for (field = 0; field < 2; field++){
58 const int poc = h->cur_pic_ptr->field_poc[field];
59 const int poc1 = h->ref_list[1][0].field_poc[field];
60 for (i = 0; i < 2 * h->ref_count[0]; i++)
61 h->dist_scale_factor_field[field][i^field] =
62 get_scale_factor(h, poc, poc1, i+16);
65 for (i = 0; i < h->ref_count[0]; i++){
66 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
70 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
71 Picture * const ref1 = &h->ref_list[1][0];
72 int j, old_ref, rfield;
73 int start= mbafi ? 16 : 0;
74 int end = mbafi ? 16+2*h->ref_count[0] : h->ref_count[0];
75 int interl= mbafi || h->picture_structure != PICT_FRAME;
77 /* bogus; fills in for missing frames */
78 memset(map[list], 0, sizeof(map[list]));
80 for(rfield=0; rfield<2; rfield++){
81 for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
82 int poc = ref1->ref_poc[colfield][list][old_ref];
86 else if( interl && (poc&3) == 3) // FIXME: store all MBAFF references so this is not needed
87 poc= (poc&~3) + rfield + 1;
89 for(j=start; j<end; j++){
90 if (4 * h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference & 3) == poc) {
91 int cur_ref= mbafi ? (j-16)^field : j;
93 map[list][2 * old_ref + (rfield^field) + 16] = cur_ref;
94 if(rfield == field || !interl)
95 map[list][old_ref] = cur_ref;
103 void ff_h264_direct_ref_list_init(H264Context * const h){
104 Picture * const ref1 = &h->ref_list[1][0];
105 Picture * const cur = h->cur_pic_ptr;
107 int sidx= (h->picture_structure&1)^1;
108 int ref1sidx = (ref1->reference&1)^1;
110 for(list=0; list<2; list++){
111 cur->ref_count[sidx][list] = h->ref_count[list];
112 for(j=0; j<h->ref_count[list]; j++)
113 cur->ref_poc[sidx][list][j] = 4 * h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference & 3);
116 if(h->picture_structure == PICT_FRAME){
117 memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
118 memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0]));
121 cur->mbaff = FRAME_MBAFF(h);
124 if(h->picture_structure == PICT_FRAME){
125 int cur_poc = h->cur_pic_ptr->poc;
126 int *col_poc = h->ref_list[1]->field_poc;
127 h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
128 ref1sidx=sidx= h->col_parity;
129 } else if (!(h->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity
130 h->col_fieldoff = 2 * h->ref_list[1][0].reference - 3;
133 if (h->slice_type_nos != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred)
136 for(list=0; list<2; list++){
137 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
139 for(field=0; field<2; field++)
140 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
144 static void await_reference_mb_row(H264Context * const h, Picture *ref, int mb_y)
146 int ref_field = ref->reference - 1;
147 int ref_field_picture = ref->field_picture;
148 int ref_height = 16*h->mb_height >> ref_field_picture;
150 if(!HAVE_THREADS || !(h->avctx->active_thread_type&FF_THREAD_FRAME))
153 //FIXME it can be safe to access mb stuff
154 //even if pixels aren't deblocked yet
156 ff_thread_await_progress(&ref->tf,
157 FFMIN(16 * mb_y >> ref_field_picture, ref_height - 1),
158 ref_field_picture && ref_field);
161 static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
163 int b4_stride = h->b_stride;
164 int mb_xy = h->mb_xy, mb_y = h->mb_y;
166 const int16_t (*l1mv0)[2], (*l1mv1)[2];
167 const int8_t *l1ref0, *l1ref1;
168 const int is_b8x8 = IS_8X8(*mb_type);
169 unsigned int sub_mb_type= MB_TYPE_L0L1;
175 assert(h->ref_list[1][0].reference & 3);
177 await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type));
179 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
182 /* ref = min(neighbors) */
183 for(list=0; list<2; list++){
184 int left_ref = h->ref_cache[list][scan8[0] - 1];
185 int top_ref = h->ref_cache[list][scan8[0] - 8];
186 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
187 const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
188 if(refc == PART_NOT_AVAILABLE){
189 refc = h->ref_cache[list][scan8[0] - 8 - 1];
190 C = h-> mv_cache[list][scan8[0] - 8 - 1];
192 ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
194 //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
195 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
196 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
198 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
199 if(match_count > 1){ //most common
200 mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
201 mid_pred(A[1], B[1], C[1]) );
203 assert(match_count==1);
204 if(left_ref==ref[list]){
205 mv[list]= AV_RN32A(A);
206 }else if(top_ref==ref[list]){
207 mv[list]= AV_RN32A(B);
209 mv[list]= AV_RN32A(C);
212 av_assert2(ref[list] < (h->ref_count[list] << !!FRAME_MBAFF(h)));
214 int mask= ~(MB_TYPE_L0 << (2*list));
222 if(ref[0] < 0 && ref[1] < 0){
225 *mb_type |= MB_TYPE_L0L1;
226 sub_mb_type |= MB_TYPE_L0L1;
229 if(!(is_b8x8|mv[0]|mv[1])){
230 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
231 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
232 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
233 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
234 *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;
238 if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
239 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
240 mb_y = (h->mb_y&~1) + h->col_parity;
241 mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride;
244 mb_y += h->col_fieldoff;
245 mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
248 }else{ // AFL/AFR/FR/FL -> AFR/FR
249 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
251 mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride;
252 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
253 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
254 b8_stride = 2+4*h->mb_stride;
256 if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
257 mb_type_col[0] &= ~MB_TYPE_INTERLACED;
258 mb_type_col[1] &= ~MB_TYPE_INTERLACED;
261 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
262 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
263 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
265 *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
267 *mb_type |= MB_TYPE_8x8;
269 }else{ // AFR/FR -> AFR/FR
272 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
274 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
275 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
276 *mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
277 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
278 *mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
280 if(!h->sps.direct_8x8_inference_flag){
281 /* FIXME save sub mb types from previous frames (or derive from MVs)
282 * so we know exactly what block size to use */
283 sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
285 *mb_type |= MB_TYPE_8x8;
290 await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
292 l1mv0 = (void*)&h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
293 l1mv1 = (void*)&h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
294 l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy];
295 l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy];
300 l1mv0 += 2*b4_stride;
301 l1mv1 += 2*b4_stride;
306 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
308 for(i8=0; i8<4; i8++){
311 int xy8 = x8+y8*b8_stride;
312 int xy4 = 3*x8+y8*b4_stride;
315 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
317 h->sub_mb_type[i8] = sub_mb_type;
319 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
320 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
321 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
322 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
323 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
334 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
335 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
337 if(!is_b8x8 && !(n&3))
338 *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;
339 }else if(IS_16X16(*mb_type)){
342 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
343 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
344 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
345 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
346 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
347 && h->x264_build>33U))){
357 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
358 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
361 for(i8=0; i8<4; i8++){
363 const int y8 = i8>>1;
365 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
367 h->sub_mb_type[i8] = sub_mb_type;
369 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
370 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
371 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
372 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
374 assert(b8_stride==2);
376 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[i8] == 0
377 || (l1ref0[i8] < 0 && l1ref1[i8] == 0
378 && h->x264_build>33U))){
379 const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
380 if(IS_SUB_8X8(sub_mb_type)){
381 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
382 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
384 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
386 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
391 for(i4=0; i4<4; i4++){
392 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
393 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
395 AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
397 AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
402 h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
407 if(!is_b8x8 && !(n&15))
408 *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;
412 static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
414 int b4_stride = h->b_stride;
415 int mb_xy = h->mb_xy, mb_y = h->mb_y;
417 const int16_t (*l1mv0)[2], (*l1mv1)[2];
418 const int8_t *l1ref0, *l1ref1;
419 const int is_b8x8 = IS_8X8(*mb_type);
420 unsigned int sub_mb_type;
423 assert(h->ref_list[1][0].reference & 3);
425 await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type));
427 if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
428 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
429 mb_y = (h->mb_y&~1) + h->col_parity;
430 mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride;
433 mb_y += h->col_fieldoff;
434 mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
437 }else{ // AFL/AFR/FR/FL -> AFR/FR
438 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
440 mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride;
441 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
442 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
443 b8_stride = 2+4*h->mb_stride;
445 if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
446 mb_type_col[0] &= ~MB_TYPE_INTERLACED;
447 mb_type_col[1] &= ~MB_TYPE_INTERLACED;
450 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
452 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
453 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
455 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
457 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
459 }else{ // AFR/FR -> AFR/FR
462 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
464 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
465 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
466 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
467 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
468 *mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
470 if(!h->sps.direct_8x8_inference_flag){
471 /* FIXME save sub mb types from previous frames (or derive from MVs)
472 * so we know exactly what block size to use */
473 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
475 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
480 await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
482 l1mv0 = (void*)&h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
483 l1mv1 = (void*)&h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
484 l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy];
485 l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy];
490 l1mv0 += 2*b4_stride;
491 l1mv1 += 2*b4_stride;
496 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
497 const int *dist_scale_factor = h->dist_scale_factor;
500 if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) {
501 map_col_to_list0[0] = h->map_col_to_list0_field[h->mb_y&1][0];
502 map_col_to_list0[1] = h->map_col_to_list0_field[h->mb_y&1][1];
503 dist_scale_factor =h->dist_scale_factor_field[h->mb_y&1];
505 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
507 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
508 int y_shift = 2*!IS_INTERLACED(*mb_type);
509 assert(h->sps.direct_8x8_inference_flag);
511 for(i8=0; i8<4; i8++){
513 const int y8 = i8>>1;
515 const int16_t (*l1mv)[2]= l1mv0;
517 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
519 h->sub_mb_type[i8] = sub_mb_type;
521 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
522 if(IS_INTRA(mb_type_col[y8])){
523 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
524 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
525 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
529 ref0 = l1ref0[x8 + y8*b8_stride];
531 ref0 = map_col_to_list0[0][ref0 + ref_offset];
533 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
536 scale = dist_scale_factor[ref0];
537 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
540 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
541 int my_col = (mv_col[1]<<y_shift)/2;
542 int mx = (scale * mv_col[0] + 128) >> 8;
543 int my = (scale * my_col + 128) >> 8;
544 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
545 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
551 /* one-to-one mv scaling */
553 if(IS_16X16(*mb_type)){
556 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
557 if(IS_INTRA(mb_type_col[0])){
560 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
561 : map_col_to_list0[1][l1ref1[0] + ref_offset];
562 const int scale = dist_scale_factor[ref0];
563 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
565 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
566 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
568 mv0= pack16to32(mv_l0[0],mv_l0[1]);
569 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
571 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
572 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
573 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
575 for(i8=0; i8<4; i8++){
577 const int y8 = i8>>1;
579 const int16_t (*l1mv)[2]= l1mv0;
581 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
583 h->sub_mb_type[i8] = sub_mb_type;
584 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
585 if(IS_INTRA(mb_type_col[0])){
586 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
587 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
588 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
592 assert(b8_stride == 2);
595 ref0 = map_col_to_list0[0][ref0 + ref_offset];
597 ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
600 scale = dist_scale_factor[ref0];
602 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
603 if(IS_SUB_8X8(sub_mb_type)){
604 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
605 int mx = (scale * mv_col[0] + 128) >> 8;
606 int my = (scale * mv_col[1] + 128) >> 8;
607 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
608 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
610 for(i4=0; i4<4; i4++){
611 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
612 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
613 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
614 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
615 AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
616 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
623 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
624 if(h->direct_spatial_mv_pred){
625 pred_spatial_direct_motion(h, mb_type);
627 pred_temp_direct_motion(h, mb_type);