Libav
|
00001 /* 00002 * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding 00003 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> 00004 * 00005 * This file is part of FFmpeg. 00006 * 00007 * FFmpeg is free software; you can redistribute it and/or 00008 * modify it under the terms of the GNU Lesser General Public 00009 * License as published by the Free Software Foundation; either 00010 * version 2.1 of the License, or (at your option) any later version. 00011 * 00012 * FFmpeg is distributed in the hope that it will be useful, 00013 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00014 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00015 * Lesser General Public License for more details. 00016 * 00017 * You should have received a copy of the GNU Lesser General Public 00018 * License along with FFmpeg; if not, write to the Free Software 00019 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00020 */ 00021 00028 #include "internal.h" 00029 #include "dsputil.h" 00030 #include "avcodec.h" 00031 #include "mpegvideo.h" 00032 #include "h264.h" 00033 #include "rectangle.h" 00034 00035 //#undef NDEBUG 00036 #include <assert.h> 00037 00038 00039 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){ 00040 int poc0 = h->ref_list[0][i].poc; 00041 int td = av_clip(poc1 - poc0, -128, 127); 00042 if(td == 0 || h->ref_list[0][i].long_ref){ 00043 return 256; 00044 }else{ 00045 int tb = av_clip(poc - poc0, -128, 127); 00046 int tx = (16384 + (FFABS(td) >> 1)) / td; 00047 return av_clip((tb*tx + 32) >> 6, -1024, 1023); 00048 } 00049 } 00050 00051 void ff_h264_direct_dist_scale_factor(H264Context * const h){ 00052 MpegEncContext * const s = &h->s; 00053 const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ]; 00054 const int poc1 = h->ref_list[1][0].poc; 00055 int i, field; 00056 for(field=0; field<2; field++){ 00057 const int poc = h->s.current_picture_ptr->field_poc[field]; 00058 const int poc1 = h->ref_list[1][0].field_poc[field]; 00059 for(i=0; i < 2*h->ref_count[0]; i++) 00060 h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16); 00061 } 00062 00063 for(i=0; i<h->ref_count[0]; i++){ 00064 h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i); 00065 } 00066 } 00067 00068 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){ 00069 MpegEncContext * const s = &h->s; 00070 Picture * const ref1 = &h->ref_list[1][0]; 00071 int j, old_ref, rfield; 00072 int start= mbafi ? 16 : 0; 00073 int end = mbafi ? 16+2*h->ref_count[0] : h->ref_count[0]; 00074 int interl= mbafi || s->picture_structure != PICT_FRAME; 00075 00076 /* bogus; fills in for missing frames */ 00077 memset(map[list], 0, sizeof(map[list])); 00078 00079 for(rfield=0; rfield<2; rfield++){ 00080 for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){ 00081 int poc = ref1->ref_poc[colfield][list][old_ref]; 00082 00083 if (!interl) 00084 poc |= 3; 00085 else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed 00086 poc= (poc&~3) + rfield + 1; 00087 00088 for(j=start; j<end; j++){ 00089 if(4*h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference&3) == poc){ 00090 int cur_ref= mbafi ? (j-16)^field : j; 00091 map[list][2*old_ref + (rfield^field) + 16] = cur_ref; 00092 if(rfield == field || !interl) 00093 map[list][old_ref] = cur_ref; 00094 break; 00095 } 00096 } 00097 } 00098 } 00099 } 00100 00101 void ff_h264_direct_ref_list_init(H264Context * const h){ 00102 MpegEncContext * const s = &h->s; 00103 Picture * const ref1 = &h->ref_list[1][0]; 00104 Picture * const cur = s->current_picture_ptr; 00105 int list, j, field; 00106 int sidx= (s->picture_structure&1)^1; 00107 int ref1sidx= (ref1->reference&1)^1; 00108 00109 for(list=0; list<2; list++){ 00110 cur->ref_count[sidx][list] = h->ref_count[list]; 00111 for(j=0; j<h->ref_count[list]; j++) 00112 cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3); 00113 } 00114 00115 if(s->picture_structure == PICT_FRAME){ 00116 memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0])); 00117 memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0])); 00118 } 00119 00120 cur->mbaff= FRAME_MBAFF; 00121 00122 h->col_fieldoff= 0; 00123 if(s->picture_structure == PICT_FRAME){ 00124 int cur_poc = s->current_picture_ptr->poc; 00125 int *col_poc = h->ref_list[1]->field_poc; 00126 h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc)); 00127 ref1sidx=sidx= h->col_parity; 00128 }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){ // FL -> FL & differ parity 00129 h->col_fieldoff= s->mb_stride*(2*(h->ref_list[1][0].reference) - 3); 00130 } 00131 00132 if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred) 00133 return; 00134 00135 for(list=0; list<2; list++){ 00136 fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0); 00137 if(FRAME_MBAFF) 00138 for(field=0; field<2; field++) 00139 fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1); 00140 } 00141 } 00142 00143 static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){ 00144 MpegEncContext * const s = &h->s; 00145 int b8_stride = 2; 00146 int b4_stride = h->b_stride; 00147 int mb_xy = h->mb_xy; 00148 int mb_type_col[2]; 00149 const int16_t (*l1mv0)[2], (*l1mv1)[2]; 00150 const int8_t *l1ref0, *l1ref1; 00151 const int is_b8x8 = IS_8X8(*mb_type); 00152 unsigned int sub_mb_type= MB_TYPE_L0L1; 00153 int i8, i4; 00154 int ref[2]; 00155 int mv[2]; 00156 int list; 00157 00158 assert(h->ref_list[1][0].reference&3); 00159 00160 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM) 00161 00162 00163 /* ref = min(neighbors) */ 00164 for(list=0; list<2; list++){ 00165 int left_ref = h->ref_cache[list][scan8[0] - 1]; 00166 int top_ref = h->ref_cache[list][scan8[0] - 8]; 00167 int refc = h->ref_cache[list][scan8[0] - 8 + 4]; 00168 const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4]; 00169 if(refc == PART_NOT_AVAILABLE){ 00170 refc = h->ref_cache[list][scan8[0] - 8 - 1]; 00171 C = h-> mv_cache[list][scan8[0] - 8 - 1]; 00172 } 00173 ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc); 00174 if(ref[list] >= 0){ 00175 //this is just pred_motion() but with the cases removed that cannot happen for direct blocks 00176 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; 00177 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; 00178 00179 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]); 00180 if(match_count > 1){ //most common 00181 mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]), 00182 mid_pred(A[1], B[1], C[1]) ); 00183 }else { 00184 assert(match_count==1); 00185 if(left_ref==ref[list]){ 00186 mv[list]= AV_RN32A(A); 00187 }else if(top_ref==ref[list]){ 00188 mv[list]= AV_RN32A(B); 00189 }else{ 00190 mv[list]= AV_RN32A(C); 00191 } 00192 } 00193 }else{ 00194 int mask= ~(MB_TYPE_L0 << (2*list)); 00195 mv[list] = 0; 00196 ref[list] = -1; 00197 if(!is_b8x8) 00198 *mb_type &= mask; 00199 sub_mb_type &= mask; 00200 } 00201 } 00202 if(ref[0] < 0 && ref[1] < 0){ 00203 ref[0] = ref[1] = 0; 00204 if(!is_b8x8) 00205 *mb_type |= MB_TYPE_L0L1; 00206 sub_mb_type |= MB_TYPE_L0L1; 00207 } 00208 00209 if(!(is_b8x8|mv[0]|mv[1])){ 00210 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1); 00211 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1); 00212 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4); 00213 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4); 00214 *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; 00215 return; 00216 } 00217 00218 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL 00219 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL 00220 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride; 00221 b8_stride = 0; 00222 }else{ 00223 mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity 00224 } 00225 goto single_col; 00226 }else{ // AFL/AFR/FR/FL -> AFR/FR 00227 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR 00228 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride; 00229 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy]; 00230 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride]; 00231 b8_stride = 2+4*s->mb_stride; 00232 b4_stride *= 6; 00233 00234 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ 00235 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) 00236 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) 00237 && !is_b8x8){ 00238 *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */ 00239 }else{ 00240 *mb_type |= MB_TYPE_8x8; 00241 } 00242 }else{ // AFR/FR -> AFR/FR 00243 single_col: 00244 mb_type_col[0] = 00245 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy]; 00246 00247 sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ 00248 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){ 00249 *mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */ 00250 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){ 00251 *mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16)); 00252 }else{ 00253 if(!h->sps.direct_8x8_inference_flag){ 00254 /* FIXME save sub mb types from previous frames (or derive from MVs) 00255 * so we know exactly what block size to use */ 00256 sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */ 00257 } 00258 *mb_type |= MB_TYPE_8x8; 00259 } 00260 } 00261 } 00262 00263 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]]; 00264 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]]; 00265 l1ref0 = &h->ref_list[1][0].ref_index [0][4*mb_xy]; 00266 l1ref1 = &h->ref_list[1][0].ref_index [1][4*mb_xy]; 00267 if(!b8_stride){ 00268 if(s->mb_y&1){ 00269 l1ref0 += 2; 00270 l1ref1 += 2; 00271 l1mv0 += 2*b4_stride; 00272 l1mv1 += 2*b4_stride; 00273 } 00274 } 00275 00276 00277 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){ 00278 int n=0; 00279 for(i8=0; i8<4; i8++){ 00280 int x8 = i8&1; 00281 int y8 = i8>>1; 00282 int xy8 = x8+y8*b8_stride; 00283 int xy4 = 3*x8+y8*b4_stride; 00284 int a,b; 00285 00286 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 00287 continue; 00288 h->sub_mb_type[i8] = sub_mb_type; 00289 00290 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); 00291 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); 00292 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref 00293 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1) 00294 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){ 00295 a=b=0; 00296 if(ref[0] > 0) 00297 a= mv[0]; 00298 if(ref[1] > 0) 00299 b= mv[1]; 00300 n++; 00301 }else{ 00302 a= mv[0]; 00303 b= mv[1]; 00304 } 00305 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4); 00306 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4); 00307 } 00308 if(!is_b8x8 && !(n&3)) 00309 *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; 00310 }else if(IS_16X16(*mb_type)){ 00311 int a,b; 00312 00313 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1); 00314 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1); 00315 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref 00316 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1) 00317 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1 00318 && h->x264_build>33U))){ 00319 a=b=0; 00320 if(ref[0] > 0) 00321 a= mv[0]; 00322 if(ref[1] > 0) 00323 b= mv[1]; 00324 }else{ 00325 a= mv[0]; 00326 b= mv[1]; 00327 } 00328 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4); 00329 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4); 00330 }else{ 00331 int n=0; 00332 for(i8=0; i8<4; i8++){ 00333 const int x8 = i8&1; 00334 const int y8 = i8>>1; 00335 00336 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 00337 continue; 00338 h->sub_mb_type[i8] = sub_mb_type; 00339 00340 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4); 00341 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4); 00342 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); 00343 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); 00344 00345 assert(b8_stride==2); 00346 /* col_zero_flag */ 00347 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[i8] == 0 00348 || (l1ref0[i8] < 0 && l1ref1[i8] == 0 00349 && h->x264_build>33U))){ 00350 const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1; 00351 if(IS_SUB_8X8(sub_mb_type)){ 00352 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride]; 00353 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ 00354 if(ref[0] == 0) 00355 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); 00356 if(ref[1] == 0) 00357 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); 00358 n+=4; 00359 } 00360 }else{ 00361 int m=0; 00362 for(i4=0; i4<4; i4++){ 00363 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride]; 00364 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ 00365 if(ref[0] == 0) 00366 AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]); 00367 if(ref[1] == 0) 00368 AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]); 00369 m++; 00370 } 00371 } 00372 if(!(m&3)) 00373 h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8; 00374 n+=m; 00375 } 00376 } 00377 } 00378 if(!is_b8x8 && !(n&15)) 00379 *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; 00380 } 00381 } 00382 00383 static void pred_temp_direct_motion(H264Context * const h, int *mb_type){ 00384 MpegEncContext * const s = &h->s; 00385 int b8_stride = 2; 00386 int b4_stride = h->b_stride; 00387 int mb_xy = h->mb_xy; 00388 int mb_type_col[2]; 00389 const int16_t (*l1mv0)[2], (*l1mv1)[2]; 00390 const int8_t *l1ref0, *l1ref1; 00391 const int is_b8x8 = IS_8X8(*mb_type); 00392 unsigned int sub_mb_type; 00393 int i8, i4; 00394 00395 assert(h->ref_list[1][0].reference&3); 00396 00397 if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL 00398 if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL 00399 mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride; 00400 b8_stride = 0; 00401 }else{ 00402 mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity 00403 } 00404 goto single_col; 00405 }else{ // AFL/AFR/FR/FL -> AFR/FR 00406 if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR 00407 mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride; 00408 mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy]; 00409 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride]; 00410 b8_stride = 2+4*s->mb_stride; 00411 b4_stride *= 6; 00412 00413 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ 00414 00415 if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) 00416 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) 00417 && !is_b8x8){ 00418 *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */ 00419 }else{ 00420 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; 00421 } 00422 }else{ // AFR/FR -> AFR/FR 00423 single_col: 00424 mb_type_col[0] = 00425 mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy]; 00426 00427 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ 00428 if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){ 00429 *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */ 00430 }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){ 00431 *mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16)); 00432 }else{ 00433 if(!h->sps.direct_8x8_inference_flag){ 00434 /* FIXME save sub mb types from previous frames (or derive from MVs) 00435 * so we know exactly what block size to use */ 00436 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */ 00437 } 00438 *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; 00439 } 00440 } 00441 } 00442 00443 l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]]; 00444 l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]]; 00445 l1ref0 = &h->ref_list[1][0].ref_index [0][4*mb_xy]; 00446 l1ref1 = &h->ref_list[1][0].ref_index [1][4*mb_xy]; 00447 if(!b8_stride){ 00448 if(s->mb_y&1){ 00449 l1ref0 += 2; 00450 l1ref1 += 2; 00451 l1mv0 += 2*b4_stride; 00452 l1mv1 += 2*b4_stride; 00453 } 00454 } 00455 00456 { 00457 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]}; 00458 const int *dist_scale_factor = h->dist_scale_factor; 00459 int ref_offset; 00460 00461 if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){ 00462 map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0]; 00463 map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1]; 00464 dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1]; 00465 } 00466 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 00467 00468 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){ 00469 int y_shift = 2*!IS_INTERLACED(*mb_type); 00470 assert(h->sps.direct_8x8_inference_flag); 00471 00472 for(i8=0; i8<4; i8++){ 00473 const int x8 = i8&1; 00474 const int y8 = i8>>1; 00475 int ref0, scale; 00476 const int16_t (*l1mv)[2]= l1mv0; 00477 00478 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 00479 continue; 00480 h->sub_mb_type[i8] = sub_mb_type; 00481 00482 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); 00483 if(IS_INTRA(mb_type_col[y8])){ 00484 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); 00485 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); 00486 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); 00487 continue; 00488 } 00489 00490 ref0 = l1ref0[x8 + y8*b8_stride]; 00491 if(ref0 >= 0) 00492 ref0 = map_col_to_list0[0][ref0 + ref_offset]; 00493 else{ 00494 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset]; 00495 l1mv= l1mv1; 00496 } 00497 scale = dist_scale_factor[ref0]; 00498 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); 00499 00500 { 00501 const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride]; 00502 int my_col = (mv_col[1]<<y_shift)/2; 00503 int mx = (scale * mv_col[0] + 128) >> 8; 00504 int my = (scale * my_col + 128) >> 8; 00505 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); 00506 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4); 00507 } 00508 } 00509 return; 00510 } 00511 00512 /* one-to-one mv scaling */ 00513 00514 if(IS_16X16(*mb_type)){ 00515 int ref, mv0, mv1; 00516 00517 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1); 00518 if(IS_INTRA(mb_type_col[0])){ 00519 ref=mv0=mv1=0; 00520 }else{ 00521 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset] 00522 : map_col_to_list0[1][l1ref1[0] + ref_offset]; 00523 const int scale = dist_scale_factor[ref0]; 00524 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0]; 00525 int mv_l0[2]; 00526 mv_l0[0] = (scale * mv_col[0] + 128) >> 8; 00527 mv_l0[1] = (scale * mv_col[1] + 128) >> 8; 00528 ref= ref0; 00529 mv0= pack16to32(mv_l0[0],mv_l0[1]); 00530 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); 00531 } 00532 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1); 00533 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4); 00534 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4); 00535 }else{ 00536 for(i8=0; i8<4; i8++){ 00537 const int x8 = i8&1; 00538 const int y8 = i8>>1; 00539 int ref0, scale; 00540 const int16_t (*l1mv)[2]= l1mv0; 00541 00542 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 00543 continue; 00544 h->sub_mb_type[i8] = sub_mb_type; 00545 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); 00546 if(IS_INTRA(mb_type_col[0])){ 00547 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); 00548 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); 00549 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); 00550 continue; 00551 } 00552 00553 assert(b8_stride == 2); 00554 ref0 = l1ref0[i8]; 00555 if(ref0 >= 0) 00556 ref0 = map_col_to_list0[0][ref0 + ref_offset]; 00557 else{ 00558 ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset]; 00559 l1mv= l1mv1; 00560 } 00561 scale = dist_scale_factor[ref0]; 00562 00563 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); 00564 if(IS_SUB_8X8(sub_mb_type)){ 00565 const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride]; 00566 int mx = (scale * mv_col[0] + 128) >> 8; 00567 int my = (scale * mv_col[1] + 128) >> 8; 00568 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); 00569 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4); 00570 }else 00571 for(i4=0; i4<4; i4++){ 00572 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride]; 00573 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]]; 00574 mv_l0[0] = (scale * mv_col[0] + 128) >> 8; 00575 mv_l0[1] = (scale * mv_col[1] + 128) >> 8; 00576 AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]], 00577 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1])); 00578 } 00579 } 00580 } 00581 } 00582 } 00583 00584 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){ 00585 if(h->direct_spatial_mv_pred){ 00586 pred_spatial_direct_motion(h, mb_type); 00587 }else{ 00588 pred_temp_direct_motion(h, mb_type); 00589 } 00590 }