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>
31 #include "mpegvideo.h"
33 #include "rectangle.h"
40 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
41 int poc0 = h->ref_list[0][i].poc;
42 int td = av_clip(poc1 - poc0, -128, 127);
43 if(td == 0 || h->ref_list[0][i].long_ref){
46 int tb = av_clip(poc - poc0, -128, 127);
47 int tx = (16384 + (FFABS(td) >> 1)) / td;
48 return av_clip((tb*tx + 32) >> 6, -1024, 1023);
52 void ff_h264_direct_dist_scale_factor(H264Context * const h){
53 MpegEncContext * const s = &h->s;
54 const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
55 const int poc1 = h->ref_list[1][0].poc;
57 for(field=0; field<2; field++){
58 const int poc = h->s.current_picture_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] = get_scale_factor(h, poc, poc1, i+16);
64 for(i=0; i<h->ref_count[0]; i++){
65 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
69 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
70 MpegEncContext * const s = &h->s;
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 || s->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 isnt 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].f.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 MpegEncContext * const s = &h->s;
105 Picture * const ref1 = &h->ref_list[1][0];
106 Picture * const cur = s->current_picture_ptr;
108 int sidx= (s->picture_structure&1)^1;
109 int ref1sidx = (ref1->f.reference&1)^1;
111 for(list=0; list<2; list++){
112 cur->ref_count[sidx][list] = h->ref_count[list];
113 for(j=0; j<h->ref_count[list]; j++)
114 cur->ref_poc[sidx][list][j] = 4 * h->ref_list[list][j].frame_num + (h->ref_list[list][j].f.reference & 3);
117 if(s->picture_structure == PICT_FRAME){
118 memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
119 memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0]));
122 cur->mbaff= FRAME_MBAFF;
125 if(s->picture_structure == PICT_FRAME){
126 int cur_poc = s->current_picture_ptr->poc;
127 int *col_poc = h->ref_list[1]->field_poc;
128 h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
129 ref1sidx=sidx= h->col_parity;
130 } else if (!(s->picture_structure & h->ref_list[1][0].f.reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity
131 h->col_fieldoff = 2 * h->ref_list[1][0].f.reference - 3;
134 if (cur->f.pict_type != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred)
137 for(list=0; list<2; list++){
138 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
140 for(field=0; field<2; field++)
141 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
145 static void await_reference_mb_row(H264Context * const h, Picture *ref, int mb_y)
147 int ref_field = ref->f.reference - 1;
148 int ref_field_picture = ref->field_picture;
149 int ref_height = 16*h->s.mb_height >> ref_field_picture;
151 if(!HAVE_THREADS || !(h->s.avctx->active_thread_type&FF_THREAD_FRAME))
154 //FIXME it can be safe to access mb stuff
155 //even if pixels aren't deblocked yet
157 ff_thread_await_progress(&ref->f,
158 FFMIN(16 * mb_y >> ref_field_picture, ref_height - 1),
159 ref_field_picture && ref_field);
162 static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
163 MpegEncContext * const s = &h->s;
165 int b4_stride = h->b_stride;
166 int mb_xy = h->mb_xy, mb_y = s->mb_y;
168 const int16_t (*l1mv0)[2], (*l1mv1)[2];
169 const int8_t *l1ref0, *l1ref1;
170 const int is_b8x8 = IS_8X8(*mb_type);
171 unsigned int sub_mb_type= MB_TYPE_L0L1;
177 assert(h->ref_list[1][0].f.reference & 3);
179 await_reference_mb_row(h, &h->ref_list[1][0], s->mb_y + !!IS_INTERLACED(*mb_type));
181 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
184 /* ref = min(neighbors) */
185 for(list=0; list<2; list++){
186 int left_ref = h->ref_cache[list][scan8[0] - 1];
187 int top_ref = h->ref_cache[list][scan8[0] - 8];
188 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
189 const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
190 if(refc == PART_NOT_AVAILABLE){
191 refc = h->ref_cache[list][scan8[0] - 8 - 1];
192 C = h-> mv_cache[list][scan8[0] - 8 - 1];
194 ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
196 //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
197 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
198 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
200 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
201 if(match_count > 1){ //most common
202 mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
203 mid_pred(A[1], B[1], C[1]) );
205 assert(match_count==1);
206 if(left_ref==ref[list]){
207 mv[list]= AV_RN32A(A);
208 }else if(top_ref==ref[list]){
209 mv[list]= AV_RN32A(B);
211 mv[list]= AV_RN32A(C);
215 int mask= ~(MB_TYPE_L0 << (2*list));
223 if(ref[0] < 0 && ref[1] < 0){
226 *mb_type |= MB_TYPE_L0L1;
227 sub_mb_type |= MB_TYPE_L0L1;
230 if(!(is_b8x8|mv[0]|mv[1])){
231 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
232 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
233 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
234 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
235 *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;
239 if (IS_INTERLACED(h->ref_list[1][0].f.mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
240 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
241 mb_y = (s->mb_y&~1) + h->col_parity;
242 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
245 mb_y += h->col_fieldoff;
246 mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
249 }else{ // AFL/AFR/FR/FL -> AFR/FR
250 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
252 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
253 mb_type_col[0] = h->ref_list[1][0].f.mb_type[mb_xy];
254 mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy + s->mb_stride];
255 b8_stride = 2+4*s->mb_stride;
257 if(IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])){
258 mb_type_col[0] &= ~MB_TYPE_INTERLACED;
259 mb_type_col[1] &= ~MB_TYPE_INTERLACED;
262 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
263 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
264 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
266 *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
268 *mb_type |= MB_TYPE_8x8;
270 }else{ // AFR/FR -> AFR/FR
273 mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy];
275 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
276 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
277 *mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
278 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
279 *mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
281 if(!h->sps.direct_8x8_inference_flag){
282 /* FIXME save sub mb types from previous frames (or derive from MVs)
283 * so we know exactly what block size to use */
284 sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
286 *mb_type |= MB_TYPE_8x8;
291 await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
293 l1mv0 = &h->ref_list[1][0].f.motion_val[0][h->mb2b_xy [mb_xy]];
294 l1mv1 = &h->ref_list[1][0].f.motion_val[1][h->mb2b_xy [mb_xy]];
295 l1ref0 = &h->ref_list[1][0].f.ref_index [0][4 * mb_xy];
296 l1ref1 = &h->ref_list[1][0].f.ref_index [1][4 * mb_xy];
301 l1mv0 += 2*b4_stride;
302 l1mv1 += 2*b4_stride;
307 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
309 for(i8=0; i8<4; i8++){
312 int xy8 = x8+y8*b8_stride;
313 int xy4 = 3*x8+y8*b4_stride;
316 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
318 h->sub_mb_type[i8] = sub_mb_type;
320 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
321 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
322 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
323 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
324 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
335 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
336 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
338 if(!is_b8x8 && !(n&3))
339 *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;
340 }else if(IS_16X16(*mb_type)){
343 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
344 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
345 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
346 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
347 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
348 && h->x264_build>33U))){
358 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
359 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
362 for(i8=0; i8<4; i8++){
364 const int y8 = i8>>1;
366 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
368 h->sub_mb_type[i8] = sub_mb_type;
370 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
371 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
372 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
373 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
375 assert(b8_stride==2);
377 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[i8] == 0
378 || (l1ref0[i8] < 0 && l1ref1[i8] == 0
379 && h->x264_build>33U))){
380 const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
381 if(IS_SUB_8X8(sub_mb_type)){
382 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
383 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
385 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
387 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
392 for(i4=0; i4<4; i4++){
393 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
394 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
396 AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
398 AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
403 h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
408 if(!is_b8x8 && !(n&15))
409 *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;
413 static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
414 MpegEncContext * const s = &h->s;
416 int b4_stride = h->b_stride;
417 int mb_xy = h->mb_xy, mb_y = s->mb_y;
419 const int16_t (*l1mv0)[2], (*l1mv1)[2];
420 const int8_t *l1ref0, *l1ref1;
421 const int is_b8x8 = IS_8X8(*mb_type);
422 unsigned int sub_mb_type;
425 assert(h->ref_list[1][0].f.reference & 3);
427 await_reference_mb_row(h, &h->ref_list[1][0], s->mb_y + !!IS_INTERLACED(*mb_type));
429 if (IS_INTERLACED(h->ref_list[1][0].f.mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
430 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
431 mb_y = (s->mb_y&~1) + h->col_parity;
432 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
435 mb_y += h->col_fieldoff;
436 mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
439 }else{ // AFL/AFR/FR/FL -> AFR/FR
440 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
442 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
443 mb_type_col[0] = h->ref_list[1][0].f.mb_type[mb_xy];
444 mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy + s->mb_stride];
445 b8_stride = 2+4*s->mb_stride;
448 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
450 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
451 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
453 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
455 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
457 }else{ // AFR/FR -> AFR/FR
460 mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy];
462 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
463 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
464 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
465 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
466 *mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
468 if(!h->sps.direct_8x8_inference_flag){
469 /* FIXME save sub mb types from previous frames (or derive from MVs)
470 * so we know exactly what block size to use */
471 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
473 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
478 await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
480 l1mv0 = &h->ref_list[1][0].f.motion_val[0][h->mb2b_xy [mb_xy]];
481 l1mv1 = &h->ref_list[1][0].f.motion_val[1][h->mb2b_xy [mb_xy]];
482 l1ref0 = &h->ref_list[1][0].f.ref_index [0][4 * mb_xy];
483 l1ref1 = &h->ref_list[1][0].f.ref_index [1][4 * mb_xy];
488 l1mv0 += 2*b4_stride;
489 l1mv1 += 2*b4_stride;
494 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
495 const int *dist_scale_factor = h->dist_scale_factor;
498 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
499 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
500 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
501 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
503 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
505 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
506 int y_shift = 2*!IS_INTERLACED(*mb_type);
507 assert(h->sps.direct_8x8_inference_flag);
509 for(i8=0; i8<4; i8++){
511 const int y8 = i8>>1;
513 const int16_t (*l1mv)[2]= l1mv0;
515 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
517 h->sub_mb_type[i8] = sub_mb_type;
519 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
520 if(IS_INTRA(mb_type_col[y8])){
521 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
522 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
523 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
527 ref0 = l1ref0[x8 + y8*b8_stride];
529 ref0 = map_col_to_list0[0][ref0 + ref_offset];
531 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
534 scale = dist_scale_factor[ref0];
535 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
538 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
539 int my_col = (mv_col[1]<<y_shift)/2;
540 int mx = (scale * mv_col[0] + 128) >> 8;
541 int my = (scale * my_col + 128) >> 8;
542 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
543 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
549 /* one-to-one mv scaling */
551 if(IS_16X16(*mb_type)){
554 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
555 if(IS_INTRA(mb_type_col[0])){
558 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
559 : map_col_to_list0[1][l1ref1[0] + ref_offset];
560 const int scale = dist_scale_factor[ref0];
561 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
563 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
564 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
566 mv0= pack16to32(mv_l0[0],mv_l0[1]);
567 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
569 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
570 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
571 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
573 for(i8=0; i8<4; i8++){
575 const int y8 = i8>>1;
577 const int16_t (*l1mv)[2]= l1mv0;
579 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
581 h->sub_mb_type[i8] = sub_mb_type;
582 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
583 if(IS_INTRA(mb_type_col[0])){
584 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
585 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
586 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
590 assert(b8_stride == 2);
593 ref0 = map_col_to_list0[0][ref0 + ref_offset];
595 ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
598 scale = dist_scale_factor[ref0];
600 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
601 if(IS_SUB_8X8(sub_mb_type)){
602 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
603 int mx = (scale * mv_col[0] + 128) >> 8;
604 int my = (scale * mv_col[1] + 128) >> 8;
605 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
606 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
608 for(i4=0; i4<4; i4++){
609 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
610 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
611 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
612 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
613 AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
614 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
621 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
622 if(h->direct_spatial_mv_pred){
623 pred_spatial_direct_motion(h, mb_type);
625 pred_temp_direct_motion(h, mb_type);