1/* 2 * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding 3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> 4 * 5 * This file is part of FFmpeg. 6 * 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. 11 * 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. 16 * 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 20 */ 21 22/** 23 * @file 24 * H.264 / AVC / MPEG4 part10 direct mb/block decoding. 25 * @author Michael Niedermayer <michaelni@gmx.at> 26 */ 27 28#include "internal.h" 29#include "avcodec.h" 30#include "h264.h" 31#include "mpegutils.h" 32#include "rectangle.h" 33#include "thread.h" 34 35#include <assert.h> 36 37 38static 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){ 42 return 256; 43 }else{ 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); 47 } 48} 49 50void ff_h264_direct_dist_scale_factor(H264Context * const h){ 51 const int poc = h->picture_structure == PICT_FRAME ? 52 h->cur_pic_ptr->poc : 53 h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD]; 54 const int poc1 = h->ref_list[1][0].poc; 55 int i, field; 56 57 if (FRAME_MBAFF(h)) 58 for (field = 0; field < 2; field++){ 59 const int poc = h->cur_pic_ptr->field_poc[field]; 60 const int poc1 = h->ref_list[1][0].field_poc[field]; 61 for (i = 0; i < 2 * h->ref_count[0]; i++) 62 h->dist_scale_factor_field[field][i^field] = 63 get_scale_factor(h, poc, poc1, i+16); 64 } 65 66 for (i = 0; i < h->ref_count[0]; i++){ 67 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i); 68 } 69} 70 71static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){ 72 H264Picture * const ref1 = &h->ref_list[1][0]; 73 int j, old_ref, rfield; 74 int start= mbafi ? 16 : 0; 75 int end = mbafi ? 16+2*h->ref_count[0] : h->ref_count[0]; 76 int interl= mbafi || h->picture_structure != PICT_FRAME; 77 78 /* bogus; fills in for missing frames */ 79 memset(map[list], 0, sizeof(map[list])); 80 81 for(rfield=0; rfield<2; rfield++){ 82 for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){ 83 int poc = ref1->ref_poc[colfield][list][old_ref]; 84 85 if (!interl) 86 poc |= 3; 87 else if( interl && (poc&3) == 3) // FIXME: store all MBAFF references so this is not needed 88 poc= (poc&~3) + rfield + 1; 89 90 for(j=start; j<end; j++){ 91 if (4 * h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference & 3) == poc) { 92 int cur_ref= mbafi ? (j-16)^field : j; 93 if (ref1->mbaff) 94 map[list][2 * old_ref + (rfield^field) + 16] = cur_ref; 95 if(rfield == field || !interl) 96 map[list][old_ref] = cur_ref; 97 break; 98 } 99 } 100 } 101 } 102} 103 104void ff_h264_direct_ref_list_init(H264Context * const h){ 105 H264Picture * const ref1 = &h->ref_list[1][0]; 106 H264Picture * const cur = h->cur_pic_ptr; 107 int list, j, field; 108 int sidx= (h->picture_structure&1)^1; 109 int ref1sidx = (ref1->reference&1)^1; 110 111 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].reference & 3); 115 } 116 117 if(h->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])); 120 } 121 122 cur->mbaff = FRAME_MBAFF(h); 123 124 h->col_fieldoff= 0; 125 if(h->picture_structure == PICT_FRAME){ 126 int cur_poc = h->cur_pic_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 (!(h->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity 131 h->col_fieldoff = 2 * h->ref_list[1][0].reference - 3; 132 } 133 134 if (h->slice_type_nos != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred) 135 return; 136 137 for(list=0; list<2; list++){ 138 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0); 139 if (FRAME_MBAFF(h)) 140 for(field=0; field<2; field++) 141 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1); 142 } 143} 144 145static void await_reference_mb_row(H264Context * const h, H264Picture *ref, int mb_y) 146{ 147 int ref_field = ref->reference - 1; 148 int ref_field_picture = ref->field_picture; 149 int ref_height = 16*h->mb_height >> ref_field_picture; 150 151 if(!HAVE_THREADS || !(h->avctx->active_thread_type&FF_THREAD_FRAME)) 152 return; 153 154 //FIXME it can be safe to access mb stuff 155 //even if pixels aren't deblocked yet 156 157 ff_thread_await_progress(&ref->tf, 158 FFMIN(16 * mb_y >> ref_field_picture, ref_height - 1), 159 ref_field_picture && ref_field); 160} 161 162static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){ 163 int b8_stride = 2; 164 int b4_stride = h->b_stride; 165 int mb_xy = h->mb_xy, mb_y = h->mb_y; 166 int mb_type_col[2]; 167 const int16_t (*l1mv0)[2], (*l1mv1)[2]; 168 const int8_t *l1ref0, *l1ref1; 169 const int is_b8x8 = IS_8X8(*mb_type); 170 unsigned int sub_mb_type= MB_TYPE_L0L1; 171 int i8, i4; 172 int ref[2]; 173 int mv[2]; 174 int list; 175 176 assert(h->ref_list[1][0].reference & 3); 177 178 await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type)); 179 180#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM) 181 182 183 /* ref = min(neighbors) */ 184 for(list=0; list<2; list++){ 185 int left_ref = h->ref_cache[list][scan8[0] - 1]; 186 int top_ref = h->ref_cache[list][scan8[0] - 8]; 187 int refc = h->ref_cache[list][scan8[0] - 8 + 4]; 188 const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4]; 189 if(refc == PART_NOT_AVAILABLE){ 190 refc = h->ref_cache[list][scan8[0] - 8 - 1]; 191 C = h-> mv_cache[list][scan8[0] - 8 - 1]; 192 } 193 ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc); 194 if(ref[list] >= 0){ 195 //this is just pred_motion() but with the cases removed that cannot happen for direct blocks 196 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; 197 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; 198 199 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]); 200 if(match_count > 1){ //most common 201 mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]), 202 mid_pred(A[1], B[1], C[1]) ); 203 }else { 204 assert(match_count==1); 205 if(left_ref==ref[list]){ 206 mv[list]= AV_RN32A(A); 207 }else if(top_ref==ref[list]){ 208 mv[list]= AV_RN32A(B); 209 }else{ 210 mv[list]= AV_RN32A(C); 211 } 212 } 213 av_assert2(ref[list] < (h->ref_count[list] << !!FRAME_MBAFF(h))); 214 }else{ 215 int mask= ~(MB_TYPE_L0 << (2*list)); 216 mv[list] = 0; 217 ref[list] = -1; 218 if(!is_b8x8) 219 *mb_type &= mask; 220 sub_mb_type &= mask; 221 } 222 } 223 if(ref[0] < 0 && ref[1] < 0){ 224 ref[0] = ref[1] = 0; 225 if(!is_b8x8) 226 *mb_type |= MB_TYPE_L0L1; 227 sub_mb_type |= MB_TYPE_L0L1; 228 } 229 230 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; 236 return; 237 } 238 239 if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL 240 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL 241 mb_y = (h->mb_y&~1) + h->col_parity; 242 mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride; 243 b8_stride = 0; 244 }else{ 245 mb_y += h->col_fieldoff; 246 mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity 247 } 248 goto single_col; 249 }else{ // AFL/AFR/FR/FL -> AFR/FR 250 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR 251 mb_y = h->mb_y&~1; 252 mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride; 253 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy]; 254 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride]; 255 b8_stride = 2+4*h->mb_stride; 256 b4_stride *= 6; 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; 260 } 261 262 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) 265 && !is_b8x8){ 266 *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */ 267 }else{ 268 *mb_type |= MB_TYPE_8x8; 269 } 270 }else{ // AFR/FR -> AFR/FR 271single_col: 272 mb_type_col[0] = 273 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy]; 274 275 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)); 280 }else{ 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 */ 285 } 286 *mb_type |= MB_TYPE_8x8; 287 } 288 } 289 } 290 291 await_reference_mb_row(h, &h->ref_list[1][0], mb_y); 292 293 l1mv0 = (void*)&h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]]; 294 l1mv1 = (void*)&h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]]; 295 l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy]; 296 l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy]; 297 if(!b8_stride){ 298 if(h->mb_y&1){ 299 l1ref0 += 2; 300 l1ref1 += 2; 301 l1mv0 += 2*b4_stride; 302 l1mv1 += 2*b4_stride; 303 } 304 } 305 306 307 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){ 308 int n=0; 309 for(i8=0; i8<4; i8++){ 310 int x8 = i8&1; 311 int y8 = i8>>1; 312 int xy8 = x8+y8*b8_stride; 313 int xy4 = 3*x8+y8*b4_stride; 314 int a,b; 315 316 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 317 continue; 318 h->sub_mb_type[i8] = sub_mb_type; 319 320 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))){ 325 a=b=0; 326 if(ref[0] > 0) 327 a= mv[0]; 328 if(ref[1] > 0) 329 b= mv[1]; 330 n++; 331 }else{ 332 a= mv[0]; 333 b= mv[1]; 334 } 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); 337 } 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)){ 341 int a,b; 342 343 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))){ 349 a=b=0; 350 if(ref[0] > 0) 351 a= mv[0]; 352 if(ref[1] > 0) 353 b= mv[1]; 354 }else{ 355 a= mv[0]; 356 b= mv[1]; 357 } 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); 360 }else{ 361 int n=0; 362 for(i8=0; i8<4; i8++){ 363 const int x8 = i8&1; 364 const int y8 = i8>>1; 365 366 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 367 continue; 368 h->sub_mb_type[i8] = sub_mb_type; 369 370 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); 374 375 assert(b8_stride==2); 376 /* col_zero_flag */ 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){ 384 if(ref[0] == 0) 385 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); 386 if(ref[1] == 0) 387 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); 388 n+=4; 389 } 390 }else{ 391 int m=0; 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){ 395 if(ref[0] == 0) 396 AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]); 397 if(ref[1] == 0) 398 AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]); 399 m++; 400 } 401 } 402 if(!(m&3)) 403 h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8; 404 n+=m; 405 } 406 } 407 } 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; 410 } 411} 412 413static void pred_temp_direct_motion(H264Context * const h, int *mb_type){ 414 int b8_stride = 2; 415 int b4_stride = h->b_stride; 416 int mb_xy = h->mb_xy, mb_y = h->mb_y; 417 int mb_type_col[2]; 418 const int16_t (*l1mv0)[2], (*l1mv1)[2]; 419 const int8_t *l1ref0, *l1ref1; 420 const int is_b8x8 = IS_8X8(*mb_type); 421 unsigned int sub_mb_type; 422 int i8, i4; 423 424 assert(h->ref_list[1][0].reference & 3); 425 426 await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type)); 427 428 if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL 429 if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL 430 mb_y = (h->mb_y&~1) + h->col_parity; 431 mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride; 432 b8_stride = 0; 433 }else{ 434 mb_y += h->col_fieldoff; 435 mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity 436 } 437 goto single_col; 438 }else{ // AFL/AFR/FR/FL -> AFR/FR 439 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR 440 mb_y = h->mb_y&~1; 441 mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride; 442 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy]; 443 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride]; 444 b8_stride = 2+4*h->mb_stride; 445 b4_stride *= 6; 446 if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) { 447 mb_type_col[0] &= ~MB_TYPE_INTERLACED; 448 mb_type_col[1] &= ~MB_TYPE_INTERLACED; 449 } 450 451 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ 452 453 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) 454 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) 455 && !is_b8x8){ 456 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */ 457 }else{ 458 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; 459 } 460 }else{ // AFR/FR -> AFR/FR 461single_col: 462 mb_type_col[0] = 463 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy]; 464 465 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ 466 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){ 467 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */ 468 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){ 469 *mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16)); 470 }else{ 471 if(!h->sps.direct_8x8_inference_flag){ 472 /* FIXME save sub mb types from previous frames (or derive from MVs) 473 * so we know exactly what block size to use */ 474 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */ 475 } 476 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; 477 } 478 } 479 } 480 481 await_reference_mb_row(h, &h->ref_list[1][0], mb_y); 482 483 l1mv0 = (void*)&h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]]; 484 l1mv1 = (void*)&h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]]; 485 l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy]; 486 l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy]; 487 if(!b8_stride){ 488 if(h->mb_y&1){ 489 l1ref0 += 2; 490 l1ref1 += 2; 491 l1mv0 += 2*b4_stride; 492 l1mv1 += 2*b4_stride; 493 } 494 } 495 496 { 497 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]}; 498 const int *dist_scale_factor = h->dist_scale_factor; 499 int ref_offset; 500 501 if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) { 502 map_col_to_list0[0] = h->map_col_to_list0_field[h->mb_y&1][0]; 503 map_col_to_list0[1] = h->map_col_to_list0_field[h->mb_y&1][1]; 504 dist_scale_factor =h->dist_scale_factor_field[h->mb_y&1]; 505 } 506 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 0 507 508 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){ 509 int y_shift = 2*!IS_INTERLACED(*mb_type); 510 assert(h->sps.direct_8x8_inference_flag); 511 512 for(i8=0; i8<4; i8++){ 513 const int x8 = i8&1; 514 const int y8 = i8>>1; 515 int ref0, scale; 516 const int16_t (*l1mv)[2]= l1mv0; 517 518 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 519 continue; 520 h->sub_mb_type[i8] = sub_mb_type; 521 522 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); 523 if(IS_INTRA(mb_type_col[y8])){ 524 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); 525 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); 526 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); 527 continue; 528 } 529 530 ref0 = l1ref0[x8 + y8*b8_stride]; 531 if(ref0 >= 0) 532 ref0 = map_col_to_list0[0][ref0 + ref_offset]; 533 else{ 534 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset]; 535 l1mv= l1mv1; 536 } 537 scale = dist_scale_factor[ref0]; 538 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); 539 540 { 541 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride]; 542 int my_col = (mv_col[1]<<y_shift)/2; 543 int mx = (scale * mv_col[0] + 128) >> 8; 544 int my = (scale * my_col + 128) >> 8; 545 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); 546 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4); 547 } 548 } 549 return; 550 } 551 552 /* one-to-one mv scaling */ 553 554 if(IS_16X16(*mb_type)){ 555 int ref, mv0, mv1; 556 557 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1); 558 if(IS_INTRA(mb_type_col[0])){ 559 ref=mv0=mv1=0; 560 }else{ 561 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset] 562 : map_col_to_list0[1][l1ref1[0] + ref_offset]; 563 const int scale = dist_scale_factor[ref0]; 564 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0]; 565 int mv_l0[2]; 566 mv_l0[0] = (scale * mv_col[0] + 128) >> 8; 567 mv_l0[1] = (scale * mv_col[1] + 128) >> 8; 568 ref= ref0; 569 mv0= pack16to32(mv_l0[0],mv_l0[1]); 570 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); 571 } 572 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1); 573 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4); 574 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4); 575 }else{ 576 for(i8=0; i8<4; i8++){ 577 const int x8 = i8&1; 578 const int y8 = i8>>1; 579 int ref0, scale; 580 const int16_t (*l1mv)[2]= l1mv0; 581 582 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 583 continue; 584 h->sub_mb_type[i8] = sub_mb_type; 585 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); 586 if(IS_INTRA(mb_type_col[0])){ 587 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); 588 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); 589 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); 590 continue; 591 } 592 593 assert(b8_stride == 2); 594 ref0 = l1ref0[i8]; 595 if(ref0 >= 0) 596 ref0 = map_col_to_list0[0][ref0 + ref_offset]; 597 else{ 598 ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset]; 599 l1mv= l1mv1; 600 } 601 scale = dist_scale_factor[ref0]; 602 603 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); 604 if(IS_SUB_8X8(sub_mb_type)){ 605 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride]; 606 int mx = (scale * mv_col[0] + 128) >> 8; 607 int my = (scale * mv_col[1] + 128) >> 8; 608 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); 609 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4); 610 }else 611 for(i4=0; i4<4; i4++){ 612 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride]; 613 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]]; 614 mv_l0[0] = (scale * mv_col[0] + 128) >> 8; 615 mv_l0[1] = (scale * mv_col[1] + 128) >> 8; 616 AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]], 617 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1])); 618 } 619 } 620 } 621 } 622} 623 624void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){ 625 if(h->direct_spatial_mv_pred){ 626 pred_spatial_direct_motion(h, mb_type); 627 }else{ 628 pred_temp_direct_motion(h, mb_type); 629 } 630} 631