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 "mpegutils.h"
32 #include "rectangle.h"
38 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
39 int poc0 = h->ref_list[0][i].poc;
40 int td = av_clip(poc1 - poc0, -128, 127);
41 if(td == 0 || h->ref_list[0][i].long_ref){
44 int tb = av_clip(poc - poc0, -128, 127);
45 int tx = (16384 + (FFABS(td) >> 1)) / td;
46 return av_clip((tb*tx + 32) >> 6, -1024, 1023);
50 void ff_h264_direct_dist_scale_factor(H264Context * const h){
51 const int poc = h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD];
52 const int poc1 = h->ref_list[1][0].poc;
56 for (field = 0; field < 2; field++){
57 const int poc = h->cur_pic_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] =
61 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 H264Picture * 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 || h->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 is not 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;
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 H264Picture * const ref1 = &h->ref_list[1][0];
104 H264Picture * const cur = h->cur_pic_ptr;
106 int sidx= (h->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(h->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(h);
123 if(h->picture_structure == PICT_FRAME){
124 int cur_poc = h->cur_pic_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 (!(h->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity
129 h->col_fieldoff = 2 * h->ref_list[1][0].reference - 3;
132 if (h->slice_type_nos != AV_PICTURE_TYPE_B || 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 await_reference_mb_row(H264Context * const h, H264Picture *ref, int mb_y)
145 int ref_field = ref->reference - 1;
146 int ref_field_picture = ref->field_picture;
147 int ref_height = 16*h->mb_height >> ref_field_picture;
149 if(!HAVE_THREADS || !(h->avctx->active_thread_type&FF_THREAD_FRAME))
152 //FIXME it can be safe to access mb stuff
153 //even if pixels aren't deblocked yet
155 ff_thread_await_progress(&ref->tf,
156 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){
162 int b4_stride = h->b_stride;
163 int mb_xy = h->mb_xy, mb_y = h->mb_y;
165 const int16_t (*l1mv0)[2], (*l1mv1)[2];
166 const int8_t *l1ref0, *l1ref1;
167 const int is_b8x8 = IS_8X8(*mb_type);
168 unsigned int sub_mb_type= MB_TYPE_L0L1;
174 assert(h->ref_list[1][0].reference & 3);
176 await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type));
178 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
181 /* ref = min(neighbors) */
182 for(list=0; list<2; list++){
183 int left_ref = h->ref_cache[list][scan8[0] - 1];
184 int top_ref = h->ref_cache[list][scan8[0] - 8];
185 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
186 const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
187 if(refc == PART_NOT_AVAILABLE){
188 refc = h->ref_cache[list][scan8[0] - 8 - 1];
189 C = h-> mv_cache[list][scan8[0] - 8 - 1];
191 ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
193 //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
194 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
195 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
197 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
198 if(match_count > 1){ //most common
199 mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
200 mid_pred(A[1], B[1], C[1]) );
202 assert(match_count==1);
203 if(left_ref==ref[list]){
204 mv[list]= AV_RN32A(A);
205 }else if(top_ref==ref[list]){
206 mv[list]= AV_RN32A(B);
208 mv[list]= AV_RN32A(C);
212 int mask= ~(MB_TYPE_L0 << (2*list));
220 if(ref[0] < 0 && ref[1] < 0){
223 *mb_type |= MB_TYPE_L0L1;
224 sub_mb_type |= MB_TYPE_L0L1;
227 if(!(is_b8x8|mv[0]|mv[1])){
228 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
229 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
230 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
231 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
232 *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;
236 if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
237 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
238 mb_y = (h->mb_y&~1) + h->col_parity;
239 mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride;
242 mb_y += h->col_fieldoff;
243 mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
246 }else{ // AFL/AFR/FR/FL -> AFR/FR
247 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
249 mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride;
250 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
251 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
252 b8_stride = 2+4*h->mb_stride;
254 if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
255 mb_type_col[0] &= ~MB_TYPE_INTERLACED;
256 mb_type_col[1] &= ~MB_TYPE_INTERLACED;
259 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
260 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
261 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
263 *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
265 *mb_type |= MB_TYPE_8x8;
267 }else{ // AFR/FR -> AFR/FR
270 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
272 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
273 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
274 *mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
275 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
276 *mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
278 if(!h->sps.direct_8x8_inference_flag){
279 /* FIXME save sub mb types from previous frames (or derive from MVs)
280 * so we know exactly what block size to use */
281 sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
283 *mb_type |= MB_TYPE_8x8;
288 await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
290 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
291 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
292 l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy];
293 l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy];
298 l1mv0 += 2*b4_stride;
299 l1mv1 += 2*b4_stride;
304 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
306 for(i8=0; i8<4; i8++){
309 int xy8 = x8+y8*b8_stride;
310 int xy4 = 3*x8+y8*b4_stride;
313 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
315 h->sub_mb_type[i8] = sub_mb_type;
317 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
318 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
319 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
320 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
321 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
332 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
333 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
335 if(!is_b8x8 && !(n&3))
336 *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;
337 }else if(IS_16X16(*mb_type)){
340 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
341 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
342 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
343 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
344 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
345 && h->x264_build>33U))){
355 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
356 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
359 for(i8=0; i8<4; i8++){
361 const int y8 = i8>>1;
363 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
365 h->sub_mb_type[i8] = sub_mb_type;
367 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
368 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
369 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
370 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
372 assert(b8_stride==2);
374 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[i8] == 0
375 || (l1ref0[i8] < 0 && l1ref1[i8] == 0
376 && h->x264_build>33U))){
377 const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
378 if(IS_SUB_8X8(sub_mb_type)){
379 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
380 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
382 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
384 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
389 for(i4=0; i4<4; i4++){
390 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
391 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
393 AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
395 AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
400 h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
405 if(!is_b8x8 && !(n&15))
406 *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;
410 static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
412 int b4_stride = h->b_stride;
413 int mb_xy = h->mb_xy, mb_y = h->mb_y;
415 const int16_t (*l1mv0)[2], (*l1mv1)[2];
416 const int8_t *l1ref0, *l1ref1;
417 const int is_b8x8 = IS_8X8(*mb_type);
418 unsigned int sub_mb_type;
421 assert(h->ref_list[1][0].reference & 3);
423 await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type));
425 if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
426 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
427 mb_y = (h->mb_y&~1) + h->col_parity;
428 mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride;
431 mb_y += h->col_fieldoff;
432 mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
435 }else{ // AFL/AFR/FR/FL -> AFR/FR
436 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
438 mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride;
439 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
440 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
441 b8_stride = 2+4*h->mb_stride;
443 if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
444 mb_type_col[0] &= ~MB_TYPE_INTERLACED;
445 mb_type_col[1] &= ~MB_TYPE_INTERLACED;
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].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].motion_val[0][h->mb2b_xy [mb_xy]];
481 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
482 l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy];
483 l1ref1 = &h->ref_list[1][0].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(h) && IS_INTERLACED(*mb_type)) {
499 map_col_to_list0[0] = h->map_col_to_list0_field[h->mb_y&1][0];
500 map_col_to_list0[1] = h->map_col_to_list0_field[h->mb_y&1][1];
501 dist_scale_factor =h->dist_scale_factor_field[h->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);