Mercurial > libavcodec.hg

comparison h264_direct.c @ 11136:a0c52e951a29 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Pack MVs together from the begin for spatial direct, this simplifies the code and is a bit faster (5-10 cpu cycles depending on what is meassured).
author michael
date Sat, 13 Feb 2010 02:20:56 +0000
parents f85c5cb2c402
children 510950eafba5
comparison
equal deleted inserted replaced
11135:af73d215879d 11136:a0c52e951a29
214 } 214 }
215 } 215 }
216 216
217 if(h->direct_spatial_mv_pred){ 217 if(h->direct_spatial_mv_pred){
218 int ref[2]; 218 int ref[2];
219 int mv[2][2]; 219 int mv[2];
220 int list; 220 int list;
221 221
222 /* ref = min(neighbors) */ 222 /* ref = min(neighbors) */
223 for(list=0; list<2; list++){ 223 for(list=0; list<2; list++){
224 int left_ref = h->ref_cache[list][scan8[0] - 1]; 224 int left_ref = h->ref_cache[list][scan8[0] - 1];
235 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; 235 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
236 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; 236 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
237 237
238 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]); 238 int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
239 if(match_count > 1){ //most common 239 if(match_count > 1){ //most common
240 mv[list][0]= mid_pred(A[0], B[0], C[0]); 240 mv[list]= (mid_pred(A[0], B[0], C[0])&0xFFFF)
241 mv[list][1]= mid_pred(A[1], B[1], C[1]); 241 +(mid_pred(A[1], B[1], C[1])<<16);
242 }else { 242 }else {
243 assert(match_count==1); 243 assert(match_count==1);
244 if(left_ref==ref[list]){ 244 if(left_ref==ref[list]){
245 mv[list][0]= A[0]; 245 mv[list]= *(uint32_t*)A;
246 mv[list][1]= A[1];
247 }else if(top_ref==ref[list]){ 246 }else if(top_ref==ref[list]){
248 mv[list][0]= B[0]; 247 mv[list]= *(uint32_t*)B;
249 mv[list][1]= B[1];
250 }else{ 248 }else{
251 mv[list][0]= C[0]; 249 mv[list]= *(uint32_t*)C;
252 mv[list][1]= C[1];
253 } 250 }
254 } 251 }
255 }else{ 252 }else{
256 int mask= ~(MB_TYPE_L0 << (2*list)); 253 int mask= ~(MB_TYPE_L0 << (2*list));
257 mv[list][0] = mv[list][1] = 0; 254 mv[list] = 0;
258 ref[list] = -1; 255 ref[list] = -1;
259 if(!is_b8x8) 256 if(!is_b8x8)
260 *mb_type &= mask; 257 *mb_type &= mask;
261 sub_mb_type &= mask; 258 sub_mb_type &= mask;
262 } 259 }
286 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref 283 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
287 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1) 284 && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
288 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){ 285 || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
289 a=b=0; 286 a=b=0;
290 if(ref[0] > 0) 287 if(ref[0] > 0)
291 a= pack16to32(mv[0][0],mv[0][1]); 288 a= mv[0];
292 if(ref[1] > 0) 289 if(ref[1] > 0)
293 b= pack16to32(mv[1][0],mv[1][1]); 290 b= mv[1];
294 n++; 291 n++;
295 }else{ 292 }else{
296 a= pack16to32(mv[0][0],mv[0][1]); 293 a= mv[0];
297 b= pack16to32(mv[1][0],mv[1][1]); 294 b= mv[1];
298 } 295 }
299 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4); 296 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
300 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4); 297 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
301 } 298 }
302 if(!is_b8x8 && !(n&3)) 299 if(!is_b8x8 && !(n&3))
310 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1) 307 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
311 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1 308 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
312 && h->x264_build>33U))){ 309 && h->x264_build>33U))){
313 a=b=0; 310 a=b=0;
314 if(ref[0] > 0) 311 if(ref[0] > 0)
315 a= pack16to32(mv[0][0],mv[0][1]); 312 a= mv[0];
316 if(ref[1] > 0) 313 if(ref[1] > 0)
317 b= pack16to32(mv[1][0],mv[1][1]); 314 b= mv[1];
318 }else{ 315 }else{
319 a= pack16to32(mv[0][0],mv[0][1]); 316 a= mv[0];
320 b= pack16to32(mv[1][0],mv[1][1]); 317 b= mv[1];
321 } 318 }
322 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4); 319 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
323 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4); 320 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
324 }else{ 321 }else{
325 int n=0; 322 int n=0;
329 326
330 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) 327 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
331 continue; 328 continue;
332 h->sub_mb_type[i8] = sub_mb_type; 329 h->sub_mb_type[i8] = sub_mb_type;
333 330
334 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4); 331 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
335 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4); 332 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
336 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); 333 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
337 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); 334 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
338 335
339 /* col_zero_flag */ 336 /* col_zero_flag */
340 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[x8 + y8*b8_stride] == 0 337 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[x8 + y8*b8_stride] == 0