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 Libav.
7 * Libav 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 * Libav 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 Libav; 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;
92 map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
93 if(rfield == field || !interl)
94 map[list][old_ref] = cur_ref;
102 void ff_h264_direct_ref_list_init(H264Context * const h){
103 MpegEncContext * const s = &h->s;
104 Picture * const ref1 = &h->ref_list[1][0];
105 Picture * const cur = s->current_picture_ptr;
107 int sidx= (s->picture_structure&1)^1;
108 int ref1sidx = (ref1->f.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].f.reference & 3);
116 if(s->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;
124 if(s->picture_structure == PICT_FRAME){
125 int cur_poc = s->current_picture_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 (!(s->picture_structure & h->ref_list[1][0].f.reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity
130 h->col_fieldoff = 2 * h->ref_list[1][0].f.reference - 3;
133 if (cur->f.pict_type != 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->f.reference - 1;
147 int ref_field_picture = ref->field_picture;
148 int ref_height = 16*h->s.mb_height >> ref_field_picture;
150 if(!HAVE_PTHREADS || !(h->s.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((AVFrame*)ref, FFMIN(16*mb_y >> ref_field_picture, ref_height-1),
157 ref_field_picture && ref_field);
160 static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
161 MpegEncContext * const s = &h->s;
163 int b4_stride = h->b_stride;
164 int mb_xy = h->mb_xy, mb_y = s->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], s->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);
213 int mask= ~(MB_TYPE_L0 << (2*list));
221 if(ref[0] < 0 && ref[1] < 0){
224 *mb_type |= MB_TYPE_L0L1;
225 sub_mb_type |= MB_TYPE_L0L1;
228 if(!(is_b8x8|mv[0]|mv[1])){
229 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
230 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
231 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
232 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
233 *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;
237 if (IS_INTERLACED(h->ref_list[1][0].f.mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
238 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
239 mb_y = (s->mb_y&~1) + h->col_parity;
240 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
243 mb_y += h->col_fieldoff;
244 mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
247 }else{ // AFL/AFR/FR/FL -> AFR/FR
248 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
250 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
251 mb_type_col[0] = h->ref_list[1][0].f.mb_type[mb_xy];
252 mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy + s->mb_stride];
253 b8_stride = 2+4*s->mb_stride;
256 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
257 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
258 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
260 *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
262 *mb_type |= MB_TYPE_8x8;
264 }else{ // AFR/FR -> AFR/FR
267 mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy];
269 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
270 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
271 *mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
272 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
273 *mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
275 if(!h->sps.direct_8x8_inference_flag){
276 /* FIXME save sub mb types from previous frames (or derive from MVs)
277 * so we know exactly what block size to use */
278 sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
280 *mb_type |= MB_TYPE_8x8;
285 await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
287 l1mv0 = &h->ref_list[1][0].f.motion_val[0][h->mb2b_xy [mb_xy]];
288 l1mv1 = &h->ref_list[1][0].f.motion_val[1][h->mb2b_xy [mb_xy]];
289 l1ref0 = &h->ref_list[1][0].f.ref_index [0][4 * mb_xy];
290 l1ref1 = &h->ref_list[1][0].f.ref_index [1][4 * mb_xy];
295 l1mv0 += 2*b4_stride;
296 l1mv1 += 2*b4_stride;
301 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
303 for(i8=0; i8<4; i8++){
306 int xy8 = x8+y8*b8_stride;
307 int xy4 = 3*x8+y8*b4_stride;
310 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
312 h->sub_mb_type[i8] = sub_mb_type;
314 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
315 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
316 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
317 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
318 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
329 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
330 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
332 if(!is_b8x8 && !(n&3))
333 *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;
334 }else if(IS_16X16(*mb_type)){
337 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
338 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
339 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
340 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
341 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
342 && h->x264_build>33U))){
352 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
353 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
356 for(i8=0; i8<4; i8++){
358 const int y8 = i8>>1;
360 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
362 h->sub_mb_type[i8] = sub_mb_type;
364 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
365 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
366 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
367 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
369 assert(b8_stride==2);
371 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[i8] == 0
372 || (l1ref0[i8] < 0 && l1ref1[i8] == 0
373 && h->x264_build>33U))){
374 const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
375 if(IS_SUB_8X8(sub_mb_type)){
376 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
377 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
379 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
381 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
386 for(i4=0; i4<4; i4++){
387 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
388 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
390 AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
392 AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
397 h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
402 if(!is_b8x8 && !(n&15))
403 *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;
407 static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
408 MpegEncContext * const s = &h->s;
410 int b4_stride = h->b_stride;
411 int mb_xy = h->mb_xy, mb_y = s->mb_y;
413 const int16_t (*l1mv0)[2], (*l1mv1)[2];
414 const int8_t *l1ref0, *l1ref1;
415 const int is_b8x8 = IS_8X8(*mb_type);
416 unsigned int sub_mb_type;
419 assert(h->ref_list[1][0].reference&3);
421 await_reference_mb_row(h, &h->ref_list[1][0], s->mb_y + !!IS_INTERLACED(*mb_type));
423 if (IS_INTERLACED(h->ref_list[1][0].f.mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
424 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
425 mb_y = (s->mb_y&~1) + h->col_parity;
426 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
429 mb_y += h->col_fieldoff;
430 mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
433 }else{ // AFL/AFR/FR/FL -> AFR/FR
434 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
436 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
437 mb_type_col[0] = h->ref_list[1][0].f.mb_type[mb_xy];
438 mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy + s->mb_stride];
439 b8_stride = 2+4*s->mb_stride;
442 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
444 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
445 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
447 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
449 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
451 }else{ // AFR/FR -> AFR/FR
454 mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy];
456 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
457 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
458 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
459 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
460 *mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
462 if(!h->sps.direct_8x8_inference_flag){
463 /* FIXME save sub mb types from previous frames (or derive from MVs)
464 * so we know exactly what block size to use */
465 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
467 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
472 await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
474 l1mv0 = &h->ref_list[1][0].f.motion_val[0][h->mb2b_xy [mb_xy]];
475 l1mv1 = &h->ref_list[1][0].f.motion_val[1][h->mb2b_xy [mb_xy]];
476 l1ref0 = &h->ref_list[1][0].f.ref_index [0][4 * mb_xy];
477 l1ref1 = &h->ref_list[1][0].f.ref_index [1][4 * mb_xy];
482 l1mv0 += 2*b4_stride;
483 l1mv1 += 2*b4_stride;
488 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
489 const int *dist_scale_factor = h->dist_scale_factor;
492 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
493 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
494 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
495 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
497 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
499 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
500 int y_shift = 2*!IS_INTERLACED(*mb_type);
501 assert(h->sps.direct_8x8_inference_flag);
503 for(i8=0; i8<4; i8++){
505 const int y8 = i8>>1;
507 const int16_t (*l1mv)[2]= l1mv0;
509 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
511 h->sub_mb_type[i8] = sub_mb_type;
513 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
514 if(IS_INTRA(mb_type_col[y8])){
515 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
516 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
517 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
521 ref0 = l1ref0[x8 + y8*b8_stride];
523 ref0 = map_col_to_list0[0][ref0 + ref_offset];
525 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
528 scale = dist_scale_factor[ref0];
529 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
532 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
533 int my_col = (mv_col[1]<<y_shift)/2;
534 int mx = (scale * mv_col[0] + 128) >> 8;
535 int my = (scale * my_col + 128) >> 8;
536 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
537 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
543 /* one-to-one mv scaling */
545 if(IS_16X16(*mb_type)){
548 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
549 if(IS_INTRA(mb_type_col[0])){
552 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
553 : map_col_to_list0[1][l1ref1[0] + ref_offset];
554 const int scale = dist_scale_factor[ref0];
555 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
557 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
558 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
560 mv0= pack16to32(mv_l0[0],mv_l0[1]);
561 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
563 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
564 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
565 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
567 for(i8=0; i8<4; i8++){
569 const int y8 = i8>>1;
571 const int16_t (*l1mv)[2]= l1mv0;
573 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
575 h->sub_mb_type[i8] = sub_mb_type;
576 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
577 if(IS_INTRA(mb_type_col[0])){
578 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
579 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
580 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
584 assert(b8_stride == 2);
587 ref0 = map_col_to_list0[0][ref0 + ref_offset];
589 ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
592 scale = dist_scale_factor[ref0];
594 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
595 if(IS_SUB_8X8(sub_mb_type)){
596 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
597 int mx = (scale * mv_col[0] + 128) >> 8;
598 int my = (scale * mv_col[1] + 128) >> 8;
599 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
600 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
602 for(i4=0; i4<4; i4++){
603 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
604 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
605 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
606 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
607 AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
608 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
615 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
616 if(h->direct_spatial_mv_pred){
617 pred_spatial_direct_motion(h, mb_type);
619 pred_temp_direct_motion(h, mb_type);