1 /*****************************************************************************
2 * macroblock.c: h264 encoder library
3 *****************************************************************************
4 * Copyright (C) 2003-2008 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
8 * Fiona Glaser <fiona@x264.com>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
23 *****************************************************************************/
26 #include "encoder/me.h"
28 void x264_mb_predict_mv( x264_t *h, int i_list, int idx, int i_width, int16_t mvp[2] )
30 const int i8 = x264_scan8[idx];
31 const int i_ref= h->mb.cache.ref[i_list][i8];
32 int i_refa = h->mb.cache.ref[i_list][i8 - 1];
33 int16_t *mv_a = h->mb.cache.mv[i_list][i8 - 1];
34 int i_refb = h->mb.cache.ref[i_list][i8 - 8];
35 int16_t *mv_b = h->mb.cache.mv[i_list][i8 - 8];
36 int i_refc = h->mb.cache.ref[i_list][i8 - 8 + i_width];
37 int16_t *mv_c = h->mb.cache.mv[i_list][i8 - 8 + i_width];
39 if( (idx&3) >= 2 + (i_width&1) || i_refc == -2 )
41 i_refc = h->mb.cache.ref[i_list][i8 - 8 - 1];
42 mv_c = h->mb.cache.mv[i_list][i8 - 8 - 1];
45 if( h->mb.i_partition == D_16x8 )
64 else if( h->mb.i_partition == D_8x16 )
84 int i_count = (i_refa == i_ref) + (i_refb == i_ref) + (i_refc == i_ref);
89 x264_median_mv( mvp, mv_a, mv_b, mv_c );
91 else if( i_count == 1 )
95 else if( i_refb == i_ref )
100 else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
106 void x264_mb_predict_mv_16x16( x264_t *h, int i_list, int i_ref, int16_t mvp[2] )
108 int i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
109 int16_t *mv_a = h->mb.cache.mv[i_list][X264_SCAN8_0 - 1];
110 int i_refb = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8];
111 int16_t *mv_b = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8];
112 int i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
113 int16_t *mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 + 4];
116 i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 - 1];
117 mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 - 1];
120 int i_count = (i_refa == i_ref) + (i_refb == i_ref) + (i_refc == i_ref);
125 x264_median_mv( mvp, mv_a, mv_b, mv_c );
127 else if( i_count == 1 )
129 if( i_refa == i_ref )
131 else if( i_refb == i_ref )
136 else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
143 void x264_mb_predict_mv_pskip( x264_t *h, int16_t mv[2] )
145 int i_refa = h->mb.cache.ref[0][X264_SCAN8_0 - 1];
146 int i_refb = h->mb.cache.ref[0][X264_SCAN8_0 - 8];
147 int16_t *mv_a = h->mb.cache.mv[0][X264_SCAN8_0 - 1];
148 int16_t *mv_b = h->mb.cache.mv[0][X264_SCAN8_0 - 8];
150 if( i_refa == -2 || i_refb == -2 ||
151 !( i_refa | M32( mv_a ) ) ||
152 !( i_refb | M32( mv_b ) ) )
158 x264_mb_predict_mv_16x16( h, 0, 0, mv );
162 static int x264_mb_predict_mv_direct16x16_temporal( x264_t *h )
164 int i_mb_4x4 = 16 * h->mb.i_mb_stride * h->mb.i_mb_y + 4 * h->mb.i_mb_x;
165 int i_mb_8x8 = 4 * h->mb.i_mb_stride * h->mb.i_mb_y + 2 * h->mb.i_mb_x;
167 const int type_col = h->fref1[0]->mb_type[h->mb.i_mb_xy];
168 const int partition_col = h->fref1[0]->mb_partition[h->mb.i_mb_xy];
170 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
172 h->mb.i_partition = partition_col;
174 if( IS_INTRA( type_col ) )
176 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
177 x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, 0 );
178 x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, 0 );
182 /* Don't do any checks other than the ones we have to, based
183 * on the size of the colocated partitions.
184 * Depends on the enum order: D_8x8, D_16x8, D_8x16, D_16x16 */
185 int max_i8 = (D_16x16 - partition_col) + 1;
186 int step = (partition_col == D_16x8) + 1;
187 int width = 4 >> ((D_16x16 - partition_col)&1);
188 int height = 4 >> ((D_16x16 - partition_col)>>1);
190 for( i8 = 0; i8 < max_i8; i8 += step )
194 const int i_part_8x8 = i_mb_8x8 + x8 + y8 * h->mb.i_b8_stride;
195 const int i_ref1_ref = h->fref1[0]->ref[0][i_part_8x8];
196 const int i_ref = (map_col_to_list0(i_ref1_ref>>h->sh.b_mbaff) << h->sh.b_mbaff) + (i_ref1_ref&h->sh.b_mbaff);
200 const int dist_scale_factor = h->mb.dist_scale_factor[i_ref][0];
201 const int16_t *mv_col = h->fref1[0]->mv[0][i_mb_4x4 + 3*x8 + 3*y8 * h->mb.i_b4_stride];
202 const int l0x = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
203 const int l0y = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
204 if( h->param.i_threads > 1 && (l0y > h->mb.mv_max_spel[1] || l0y-mv_col[1] > h->mb.mv_max_spel[1]) )
206 x264_macroblock_cache_ref( h, 2*x8, 2*y8, width, height, 0, i_ref );
207 x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 0, pack16to32_mask(l0x, l0y) );
208 x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_col[1]) );
212 /* the collocated ref isn't in the current list0 */
213 /* FIXME: we might still be able to use direct_8x8 on some partitions */
214 /* FIXME: with B-pyramid + extensive ref list reordering
215 * (not currently used), we would also have to check
216 * l1mv1 like in spatial mode */
224 static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
227 ALIGNED_ARRAY_8( int16_t, mv,[2],[2] );
228 int i_list, i8, i_ref;
229 const int8_t *l1ref0 = &h->fref1[0]->ref[0][h->mb.i_b8_xy];
230 const int8_t *l1ref1 = &h->fref1[0]->ref[1][h->mb.i_b8_xy];
231 const int16_t (*l1mv[2])[2] = { (const int16_t (*)[2]) &h->fref1[0]->mv[0][h->mb.i_b4_xy],
232 (const int16_t (*)[2]) &h->fref1[0]->mv[1][h->mb.i_b4_xy] };
233 const int type_col = h->fref1[0]->mb_type[h->mb.i_mb_xy];
234 const int partition_col = h->fref1[0]->mb_partition[h->mb.i_mb_xy];
236 h->mb.i_partition = partition_col;
238 for( i_list = 0; i_list < 2; i_list++ )
240 int i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
241 int16_t *mv_a = h->mb.cache.mv[i_list][X264_SCAN8_0 - 1];
242 int i_refb = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8];
243 int16_t *mv_b = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8];
244 int i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
245 int16_t *mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 + 4];
248 i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 - 1];
249 mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 - 1];
252 i_ref = X264_MIN3( (unsigned)i_refa, (unsigned)i_refb, (unsigned)i_refc );
256 M32( mv[i_list] ) = 0;
260 /* Same as x264_mb_predict_mv_16x16, but simplified to eliminate cases
261 * not relevant to spatial direct. */
262 int i_count = (i_refa == i_ref) + (i_refb == i_ref) + (i_refc == i_ref);
265 x264_median_mv( mv[i_list], mv_a, mv_b, mv_c );
268 if( i_refa == i_ref )
269 CP32( mv[i_list], mv_a );
270 else if( i_refb == i_ref )
271 CP32( mv[i_list], mv_b );
273 CP32( mv[i_list], mv_c );
277 x264_macroblock_cache_ref( h, 0, 0, 4, 4, i_list, i_ref );
278 x264_macroblock_cache_mv_ptr( h, 0, 0, 4, 4, i_list, mv[i_list] );
282 if( (M16( ref ) & 0x8080) == 0x8080 ) /* if( ref[0] < 0 && ref[1] < 0 ) */
284 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
285 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
289 if( h->param.i_threads > 1
290 && ( mv[0][1] > h->mb.mv_max_spel[1]
291 || mv[1][1] > h->mb.mv_max_spel[1] ) )
294 fprintf(stderr, "direct_spatial: (%d,%d) (%d,%d) > %d \n",
295 mv[0][0], mv[0][1], mv[1][0], mv[1][1],
296 h->mb.mv_max_spel[1]);
301 if( !M64( mv ) || IS_INTRA( type_col ) || (ref[0]&&ref[1]) )
304 /* Don't do any checks other than the ones we have to, based
305 * on the size of the colocated partitions.
306 * Depends on the enum order: D_8x8, D_16x8, D_8x16, D_16x16 */
307 int max_i8 = (D_16x16 - partition_col) + 1;
308 int step = (partition_col == D_16x8) + 1;
309 int width = 4 >> ((D_16x16 - partition_col)&1);
310 int height = 4 >> ((D_16x16 - partition_col)>>1);
313 for( i8 = 0; i8 < max_i8; i8 += step )
316 const int y8 = i8>>1;
317 const int o8 = x8 + y8 * h->mb.i_b8_stride;
318 const int o4 = 3*(x8 + y8 * h->mb.i_b4_stride);
320 if( l1ref0[o8] == 0 )
322 else if( l1ref0[o8] < 0 && l1ref1[o8] == 0 )
327 if( abs( l1mv[idx][o4][0] ) <= 1 && abs( l1mv[idx][o4][1] ) <= 1 )
329 if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 0, 0 );
330 if( ref[1] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 1, 0 );
337 int x264_mb_predict_mv_direct16x16( x264_t *h, int *b_changed )
340 if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_NONE )
342 else if( h->sh.b_direct_spatial_mv_pred )
343 b_available = x264_mb_predict_mv_direct16x16_spatial( h );
345 b_available = x264_mb_predict_mv_direct16x16_temporal( h );
347 if( b_changed != NULL && b_available )
351 changed = M32( h->mb.cache.direct_mv[0][0] ) ^ M32( h->mb.cache.mv[0][x264_scan8[0]] );
352 changed |= M32( h->mb.cache.direct_mv[1][0] ) ^ M32( h->mb.cache.mv[1][x264_scan8[0]] );
353 changed |= h->mb.cache.direct_ref[0][0] ^ h->mb.cache.ref[0][x264_scan8[0]];
354 changed |= h->mb.cache.direct_ref[1][0] ^ h->mb.cache.ref[1][x264_scan8[0]];
355 if( !changed && h->mb.i_partition != D_16x16 )
357 changed |= M32( h->mb.cache.direct_mv[0][3] ) ^ M32( h->mb.cache.mv[0][x264_scan8[12]] );
358 changed |= M32( h->mb.cache.direct_mv[1][3] ) ^ M32( h->mb.cache.mv[1][x264_scan8[12]] );
359 changed |= h->mb.cache.direct_ref[0][3] ^ h->mb.cache.ref[0][x264_scan8[12]];
360 changed |= h->mb.cache.direct_ref[1][3] ^ h->mb.cache.ref[1][x264_scan8[12]];
362 if( !changed && h->mb.i_partition == D_8x8 )
364 changed |= M32( h->mb.cache.direct_mv[0][1] ) ^ M32( h->mb.cache.mv[0][x264_scan8[4]] );
365 changed |= M32( h->mb.cache.direct_mv[1][1] ) ^ M32( h->mb.cache.mv[1][x264_scan8[4]] );
366 changed |= M32( h->mb.cache.direct_mv[0][2] ) ^ M32( h->mb.cache.mv[0][x264_scan8[8]] );
367 changed |= M32( h->mb.cache.direct_mv[1][2] ) ^ M32( h->mb.cache.mv[1][x264_scan8[8]] );
368 changed |= h->mb.cache.direct_ref[0][1] ^ h->mb.cache.ref[0][x264_scan8[4]];
369 changed |= h->mb.cache.direct_ref[1][1] ^ h->mb.cache.ref[1][x264_scan8[4]];
370 changed |= h->mb.cache.direct_ref[0][2] ^ h->mb.cache.ref[0][x264_scan8[8]];
371 changed |= h->mb.cache.direct_ref[1][2] ^ h->mb.cache.ref[1][x264_scan8[8]];
373 *b_changed = changed;
382 for( l = 0; l < 2; l++ )
384 CP32( h->mb.cache.direct_mv[l][0], h->mb.cache.mv[l][x264_scan8[ 0]] );
385 CP32( h->mb.cache.direct_mv[l][1], h->mb.cache.mv[l][x264_scan8[ 4]] );
386 CP32( h->mb.cache.direct_mv[l][2], h->mb.cache.mv[l][x264_scan8[ 8]] );
387 CP32( h->mb.cache.direct_mv[l][3], h->mb.cache.mv[l][x264_scan8[12]] );
388 h->mb.cache.direct_ref[l][0] = h->mb.cache.ref[l][x264_scan8[ 0]];
389 h->mb.cache.direct_ref[l][1] = h->mb.cache.ref[l][x264_scan8[ 4]];
390 h->mb.cache.direct_ref[l][2] = h->mb.cache.ref[l][x264_scan8[ 8]];
391 h->mb.cache.direct_ref[l][3] = h->mb.cache.ref[l][x264_scan8[12]];
392 h->mb.cache.direct_partition = h->mb.i_partition;
399 /* This just improves encoder performance, it's not part of the spec */
400 void x264_mb_predict_mv_ref16x16( x264_t *h, int i_list, int i_ref, int16_t mvc[9][2], int *i_mvc )
402 int16_t (*mvr)[2] = h->mb.mvr[i_list][i_ref];
405 #define SET_MVP(mvp)\
407 CP32( mvc[i], mvp ); \
412 if( h->sh.i_type == SLICE_TYPE_B
413 && h->mb.cache.ref[i_list][x264_scan8[12]] == i_ref )
415 SET_MVP( h->mb.cache.mv[i_list][x264_scan8[12]] );
418 if( i_ref == 0 && h->frames.b_have_lowres )
420 int16_t (*lowres_mv)[2] = i_list ? h->fenc->lowres_mvs[1][h->fref1[0]->i_frame-h->fenc->i_frame-1]
421 : h->fenc->lowres_mvs[0][h->fenc->i_frame-h->fref0[0]->i_frame-1];
422 if( lowres_mv[0][0] != 0x7fff )
424 M32( mvc[i] ) = (M32( lowres_mv[h->mb.i_mb_xy] )*2)&0xfffeffff;
429 /* spatial predictors */
430 if( h->mb.i_neighbour & MB_LEFT )
432 int i_mb_l = h->mb.i_mb_xy - 1;
433 SET_MVP( mvr[i_mb_l] );
435 if( h->mb.i_neighbour & MB_TOP )
437 int i_mb_t = h->mb.i_mb_top_xy;
438 SET_MVP( mvr[i_mb_t] );
440 if( h->mb.i_neighbour & MB_TOPLEFT )
441 SET_MVP( mvr[i_mb_t-1] );
442 if( h->mb.i_mb_x < h->mb.i_mb_stride - 1 )
443 SET_MVP( mvr[i_mb_t+1] );
447 /* temporal predictors */
448 if( h->fref0[0]->i_ref[0] > 0 )
450 x264_frame_t *l0 = h->fref0[0];
451 x264_frame_t **fref = i_list ? h->fref1 : h->fref0;
452 int field = h->mb.i_mb_y&1;
453 int curpoc = h->fdec->i_poc + field*h->sh.i_delta_poc_bottom;
454 int refpoc = fref[i_ref>>h->sh.b_mbaff]->i_poc;
455 if( h->sh.b_mbaff && field^(i_ref&1) )
456 refpoc += h->sh.i_delta_poc_bottom;
458 #define SET_TMVP(dx, dy) { \
459 int i_b4 = h->mb.i_b4_xy + dx*4 + dy*4*h->mb.i_b4_stride; \
460 int i_b8 = h->mb.i_b8_xy + dx*2 + dy*2*h->mb.i_b8_stride; \
461 int ref_col = l0->ref[0][i_b8]; \
464 int scale = (curpoc - refpoc) * l0->inv_ref_poc[h->mb.b_interlaced&field][ref_col];\
465 mvc[i][0] = (l0->mv[0][i_b4][0]*scale + 128) >> 8;\
466 mvc[i][1] = (l0->mv[0][i_b4][1]*scale + 128) >> 8;\
472 if( h->mb.i_mb_x < h->sps->i_mb_width-1 )
474 if( h->mb.i_mb_y < h->sps->i_mb_height-1 )
482 /* Set up a lookup table for delta pocs to reduce an IDIV to an IMUL */
483 static void setup_inverse_delta_pocs( x264_t *h )
486 for( field = 0; field <= h->sh.b_mbaff; field++ )
488 int curpoc = h->fdec->i_poc + field*h->sh.i_delta_poc_bottom;
489 for( i = 0; i < (h->i_ref0<<h->sh.b_mbaff); i++ )
491 int refpoc = h->fref0[i>>h->sh.b_mbaff]->i_poc;
492 if( h->sh.b_mbaff && field^(i&1) )
493 refpoc += h->sh.i_delta_poc_bottom;
494 int delta = curpoc - refpoc;
496 h->fdec->inv_ref_poc[field][i] = (256 + delta/2) / delta;
501 static NOINLINE void x264_mb_mc_0xywh( x264_t *h, int x, int y, int width, int height )
503 const int i8 = x264_scan8[0]+x+8*y;
504 const int i_ref = h->mb.cache.ref[0][i8];
505 const int mvx = x264_clip3( h->mb.cache.mv[0][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] ) + 4*4*x;
506 int mvy = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
508 h->mc.mc_luma( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
509 h->mb.pic.p_fref[0][i_ref], h->mb.pic.i_stride[0],
510 mvx, mvy, 4*width, 4*height, &h->sh.weight[i_ref][0] );
512 // chroma is offset if MCing from a field of opposite parity
513 if( h->mb.b_interlaced & i_ref )
514 mvy += (h->mb.i_mb_y & 1)*4 - 2;
516 h->mc.mc_chroma( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
517 h->mb.pic.p_fref[0][i_ref][4], h->mb.pic.i_stride[1],
518 mvx, mvy, 2*width, 2*height );
520 if( h->sh.weight[i_ref][1].weightfn )
521 h->sh.weight[i_ref][1].weightfn[width>>1]( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
522 &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
523 &h->sh.weight[i_ref][1], height*2 );
525 h->mc.mc_chroma( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
526 h->mb.pic.p_fref[0][i_ref][5], h->mb.pic.i_stride[2],
527 mvx, mvy, 2*width, 2*height );
529 if( h->sh.weight[i_ref][2].weightfn )
530 h->sh.weight[i_ref][2].weightfn[width>>1]( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
531 &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
532 &h->sh.weight[i_ref][2],height*2 );
535 static NOINLINE void x264_mb_mc_1xywh( x264_t *h, int x, int y, int width, int height )
537 const int i8 = x264_scan8[0]+x+8*y;
538 const int i_ref = h->mb.cache.ref[1][i8];
539 const int mvx = x264_clip3( h->mb.cache.mv[1][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] ) + 4*4*x;
540 int mvy = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
542 h->mc.mc_luma( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
543 h->mb.pic.p_fref[1][i_ref], h->mb.pic.i_stride[0],
544 mvx, mvy, 4*width, 4*height, weight_none );
546 if( h->mb.b_interlaced & i_ref )
547 mvy += (h->mb.i_mb_y & 1)*4 - 2;
549 h->mc.mc_chroma( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
550 h->mb.pic.p_fref[1][i_ref][4], h->mb.pic.i_stride[1],
551 mvx, mvy, 2*width, 2*height );
553 h->mc.mc_chroma( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
554 h->mb.pic.p_fref[1][i_ref][5], h->mb.pic.i_stride[2],
555 mvx, mvy, 2*width, 2*height );
558 static NOINLINE void x264_mb_mc_01xywh( x264_t *h, int x, int y, int width, int height )
560 const int i8 = x264_scan8[0]+x+8*y;
561 const int i_ref0 = h->mb.cache.ref[0][i8];
562 const int i_ref1 = h->mb.cache.ref[1][i8];
563 const int weight = h->mb.bipred_weight[i_ref0][i_ref1];
564 const int mvx0 = x264_clip3( h->mb.cache.mv[0][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] ) + 4*4*x;
565 const int mvx1 = x264_clip3( h->mb.cache.mv[1][i8][0], h->mb.mv_min[0], h->mb.mv_max[0] ) + 4*4*x;
566 int mvy0 = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
567 int mvy1 = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
568 int i_mode = x264_size2pixel[height][width];
569 int i_stride0 = 16, i_stride1 = 16;
570 ALIGNED_ARRAY_16( uint8_t, tmp0,[16*16] );
571 ALIGNED_ARRAY_16( uint8_t, tmp1,[16*16] );
572 uint8_t *src0, *src1;
574 src0 = h->mc.get_ref( tmp0, &i_stride0, h->mb.pic.p_fref[0][i_ref0], h->mb.pic.i_stride[0],
575 mvx0, mvy0, 4*width, 4*height, weight_none );
576 src1 = h->mc.get_ref( tmp1, &i_stride1, h->mb.pic.p_fref[1][i_ref1], h->mb.pic.i_stride[0],
577 mvx1, mvy1, 4*width, 4*height, weight_none );
578 h->mc.avg[i_mode]( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
579 src0, i_stride0, src1, i_stride1, weight );
581 if( h->mb.b_interlaced & i_ref0 )
582 mvy0 += (h->mb.i_mb_y & 1)*4 - 2;
583 if( h->mb.b_interlaced & i_ref1 )
584 mvy1 += (h->mb.i_mb_y & 1)*4 - 2;
586 h->mc.mc_chroma( tmp0, 16, h->mb.pic.p_fref[0][i_ref0][4], h->mb.pic.i_stride[1],
587 mvx0, mvy0, 2*width, 2*height );
588 h->mc.mc_chroma( tmp1, 16, h->mb.pic.p_fref[1][i_ref1][4], h->mb.pic.i_stride[1],
589 mvx1, mvy1, 2*width, 2*height );
590 h->mc.avg[i_mode+3]( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp0, 16, tmp1, 16, weight );
591 h->mc.mc_chroma( tmp0, 16, h->mb.pic.p_fref[0][i_ref0][5], h->mb.pic.i_stride[2],
592 mvx0, mvy0, 2*width, 2*height );
593 h->mc.mc_chroma( tmp1, 16, h->mb.pic.p_fref[1][i_ref1][5], h->mb.pic.i_stride[2],
594 mvx1, mvy1, 2*width, 2*height );
595 h->mc.avg[i_mode+3]( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE, tmp0, 16, tmp1, 16, weight );
598 void x264_mb_mc_8x8( x264_t *h, int i8 )
600 const int x = 2*(i8&1);
601 const int y = 2*(i8>>1);
603 if( h->sh.i_type == SLICE_TYPE_P )
605 switch( h->mb.i_sub_partition[i8] )
608 x264_mb_mc_0xywh( h, x, y, 2, 2 );
611 x264_mb_mc_0xywh( h, x, y+0, 2, 1 );
612 x264_mb_mc_0xywh( h, x, y+1, 2, 1 );
615 x264_mb_mc_0xywh( h, x+0, y, 1, 2 );
616 x264_mb_mc_0xywh( h, x+1, y, 1, 2 );
619 x264_mb_mc_0xywh( h, x+0, y+0, 1, 1 );
620 x264_mb_mc_0xywh( h, x+1, y+0, 1, 1 );
621 x264_mb_mc_0xywh( h, x+0, y+1, 1, 1 );
622 x264_mb_mc_0xywh( h, x+1, y+1, 1, 1 );
628 const int i8 = x264_scan8[0] + x + 8*y;
630 if( h->mb.cache.ref[0][i8] >= 0 )
631 if( h->mb.cache.ref[1][i8] >= 0 )
632 x264_mb_mc_01xywh( h, x, y, 2, 2 );
634 x264_mb_mc_0xywh( h, x, y, 2, 2 );
636 x264_mb_mc_1xywh( h, x, y, 2, 2 );
640 void x264_mb_mc( x264_t *h )
642 if( h->mb.i_partition == D_8x8 )
645 for( i = 0; i < 4; i++ )
646 x264_mb_mc_8x8( h, i );
650 const int ref0a = h->mb.cache.ref[0][x264_scan8[ 0]];
651 const int ref0b = h->mb.cache.ref[0][x264_scan8[12]];
652 const int ref1a = h->mb.cache.ref[1][x264_scan8[ 0]];
653 const int ref1b = h->mb.cache.ref[1][x264_scan8[12]];
655 if( h->mb.i_partition == D_16x16 )
658 if( ref1a >= 0 ) x264_mb_mc_01xywh( h, 0, 0, 4, 4 );
659 else x264_mb_mc_0xywh ( h, 0, 0, 4, 4 );
660 else x264_mb_mc_1xywh ( h, 0, 0, 4, 4 );
662 else if( h->mb.i_partition == D_16x8 )
665 if( ref1a >= 0 ) x264_mb_mc_01xywh( h, 0, 0, 4, 2 );
666 else x264_mb_mc_0xywh ( h, 0, 0, 4, 2 );
667 else x264_mb_mc_1xywh ( h, 0, 0, 4, 2 );
670 if( ref1b >= 0 ) x264_mb_mc_01xywh( h, 0, 2, 4, 2 );
671 else x264_mb_mc_0xywh ( h, 0, 2, 4, 2 );
672 else x264_mb_mc_1xywh ( h, 0, 2, 4, 2 );
674 else if( h->mb.i_partition == D_8x16 )
677 if( ref1a >= 0 ) x264_mb_mc_01xywh( h, 0, 0, 2, 4 );
678 else x264_mb_mc_0xywh ( h, 0, 0, 2, 4 );
679 else x264_mb_mc_1xywh ( h, 0, 0, 2, 4 );
682 if( ref1b >= 0 ) x264_mb_mc_01xywh( h, 2, 0, 2, 4 );
683 else x264_mb_mc_0xywh ( h, 2, 0, 2, 4 );
684 else x264_mb_mc_1xywh ( h, 2, 0, 2, 4 );
689 int x264_macroblock_cache_init( x264_t *h )
692 int i_mb_count = h->mb.i_mb_count;
694 h->mb.i_mb_stride = h->sps->i_mb_width;
695 h->mb.i_b8_stride = h->sps->i_mb_width * 2;
696 h->mb.i_b4_stride = h->sps->i_mb_width * 4;
698 h->mb.b_interlaced = h->param.b_interlaced;
700 CHECKED_MALLOC( h->mb.qp, i_mb_count * sizeof(int8_t) );
701 CHECKED_MALLOC( h->mb.cbp, i_mb_count * sizeof(int16_t) );
702 CHECKED_MALLOC( h->mb.skipbp, i_mb_count * sizeof(int8_t) );
703 CHECKED_MALLOC( h->mb.mb_transform_size, i_mb_count * sizeof(int8_t) );
705 /* 0 -> 3 top(4), 4 -> 6 : left(3) */
706 CHECKED_MALLOC( h->mb.intra4x4_pred_mode, i_mb_count * 8 * sizeof(int8_t) );
709 CHECKED_MALLOC( h->mb.non_zero_count, i_mb_count * 24 * sizeof(uint8_t) );
711 if( h->param.b_cabac )
713 CHECKED_MALLOC( h->mb.chroma_pred_mode, i_mb_count * sizeof(int8_t) );
714 CHECKED_MALLOC( h->mb.mvd[0], 2*8 * i_mb_count * sizeof(uint8_t) );
715 CHECKED_MALLOC( h->mb.mvd[1], 2*8 * i_mb_count * sizeof(uint8_t) );
720 int i_refs = X264_MIN(16, (i ? 1 + !!h->param.i_bframe_pyramid : h->param.i_frame_reference) ) << h->param.b_interlaced;
721 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
722 i_refs = X264_MIN(16, i_refs + 2); //smart weights add two duplicate frames
723 else if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_BLIND )
724 i_refs = X264_MIN(16, i_refs + 1); //blind weights add one duplicate frame
726 for( j=0; j < i_refs; j++ )
727 CHECKED_MALLOC( h->mb.mvr[i][j], 2 * i_mb_count * sizeof(int16_t) );
730 if( h->param.analyse.i_weighted_pred )
732 int i_padv = PADV << h->param.b_interlaced;
733 #define ALIGN(x,a) (((x)+((a)-1))&~((a)-1))
734 int align = h->param.cpu&X264_CPU_CACHELINE_64 ? 64 : h->param.cpu&X264_CPU_CACHELINE_32 ? 32 : 16;
735 int i_stride, luma_plane_size;
738 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_FAKE )
740 // only need buffer for lookahead
741 if( !h->param.i_sync_lookahead || h == h->thread[h->param.i_threads] )
743 // Fake analysis only works on lowres
744 i_stride = ALIGN( h->sps->i_mb_width*8 + 2*PADH, align );
745 luma_plane_size = i_stride * (h->sps->i_mb_height*8+2*i_padv);
746 // Only need 1 buffer for analysis
754 i_stride = ALIGN( h->sps->i_mb_width*16 + 2*PADH, align );
755 luma_plane_size = i_stride * (h->sps->i_mb_height*16+2*i_padv);
757 if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART )
758 //SMART can weight one ref and one offset -1
761 //blind only has one weighted copy (offset -1)
765 for( i = 0; i < numweightbuf; i++ )
766 CHECKED_MALLOC( h->mb.p_weight_buf[i], luma_plane_size );
770 for( i=0; i<=h->param.b_interlaced; i++ )
773 /* shouldn't really be initialized, just silences a valgrind false-positive in predict_8x8_filter_mmx */
774 CHECKED_MALLOCZERO( h->mb.intra_border_backup[i][j], (h->sps->i_mb_width*16+32)>>!!j );
775 h->mb.intra_border_backup[i][j] += 8;
781 void x264_macroblock_cache_end( x264_t *h )
784 for( i=0; i<=h->param.b_interlaced; i++ )
786 x264_free( h->mb.intra_border_backup[i][j] - 8 );
788 for( j=0; j<32; j++ )
789 x264_free( h->mb.mvr[i][j] );
790 for( i=0; i<16; i++ )
791 x264_free( h->mb.p_weight_buf[i] );
793 if( h->param.b_cabac )
795 x264_free( h->mb.chroma_pred_mode );
796 x264_free( h->mb.mvd[0] );
797 x264_free( h->mb.mvd[1] );
799 x264_free( h->mb.intra4x4_pred_mode );
800 x264_free( h->mb.non_zero_count );
801 x264_free( h->mb.mb_transform_size );
802 x264_free( h->mb.skipbp );
803 x264_free( h->mb.cbp );
804 x264_free( h->mb.qp );
806 void x264_macroblock_slice_init( x264_t *h )
810 h->mb.mv[0] = h->fdec->mv[0];
811 h->mb.mv[1] = h->fdec->mv[1];
812 h->mb.ref[0] = h->fdec->ref[0];
813 h->mb.ref[1] = h->fdec->ref[1];
814 h->mb.type = h->fdec->mb_type;
815 h->mb.partition = h->fdec->mb_partition;
817 h->fdec->i_ref[0] = h->i_ref0;
818 h->fdec->i_ref[1] = h->i_ref1;
819 for( i = 0; i < h->i_ref0; i++ )
820 h->fdec->ref_poc[0][i] = h->fref0[i]->i_poc;
821 if( h->sh.i_type == SLICE_TYPE_B )
823 for( i = 0; i < h->i_ref1; i++ )
824 h->fdec->ref_poc[1][i] = h->fref1[i]->i_poc;
826 map_col_to_list0(-1) = -1;
827 map_col_to_list0(-2) = -2;
828 for( i = 0; i < h->fref1[0]->i_ref[0]; i++ )
830 int poc = h->fref1[0]->ref_poc[0][i];
831 map_col_to_list0(i) = -2;
832 for( j = 0; j < h->i_ref0; j++ )
833 if( h->fref0[j]->i_poc == poc )
835 map_col_to_list0(i) = j;
840 if( h->sh.i_type == SLICE_TYPE_P )
841 memset( h->mb.cache.skip, 0, X264_SCAN8_SIZE * sizeof( int8_t ) );
843 /* init with not available (for top right idx=7,15) */
844 memset( h->mb.cache.ref[0], -2, X264_SCAN8_SIZE * sizeof( int8_t ) );
845 memset( h->mb.cache.ref[1], -2, X264_SCAN8_SIZE * sizeof( int8_t ) );
847 setup_inverse_delta_pocs( h );
849 h->mb.i_neighbour4[6] =
850 h->mb.i_neighbour4[9] =
851 h->mb.i_neighbour4[12] =
852 h->mb.i_neighbour4[14] = MB_LEFT|MB_TOP|MB_TOPLEFT|MB_TOPRIGHT;
853 h->mb.i_neighbour4[3] =
854 h->mb.i_neighbour4[7] =
855 h->mb.i_neighbour4[11] =
856 h->mb.i_neighbour4[13] =
857 h->mb.i_neighbour4[15] =
858 h->mb.i_neighbour8[3] = MB_LEFT|MB_TOP|MB_TOPLEFT;
861 void x264_macroblock_thread_init( x264_t *h )
863 h->mb.i_me_method = h->param.analyse.i_me_method;
864 h->mb.i_subpel_refine = h->param.analyse.i_subpel_refine;
865 if( h->sh.i_type == SLICE_TYPE_B && (h->mb.i_subpel_refine == 6 || h->mb.i_subpel_refine == 8) )
866 h->mb.i_subpel_refine--;
867 h->mb.b_chroma_me = h->param.analyse.b_chroma_me && h->sh.i_type == SLICE_TYPE_P
868 && h->mb.i_subpel_refine >= 5;
869 h->mb.b_dct_decimate = h->sh.i_type == SLICE_TYPE_B ||
870 (h->param.analyse.b_dct_decimate && h->sh.i_type != SLICE_TYPE_I);
883 h->mb.pic.p_fenc[0] = h->mb.pic.fenc_buf;
884 h->mb.pic.p_fenc[1] = h->mb.pic.fenc_buf + 16*FENC_STRIDE;
885 h->mb.pic.p_fenc[2] = h->mb.pic.fenc_buf + 16*FENC_STRIDE + 8;
886 h->mb.pic.p_fdec[0] = h->mb.pic.fdec_buf + 2*FDEC_STRIDE;
887 h->mb.pic.p_fdec[1] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE;
888 h->mb.pic.p_fdec[2] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE + 16;
891 void x264_prefetch_fenc( x264_t *h, x264_frame_t *fenc, int i_mb_x, int i_mb_y )
893 int stride_y = fenc->i_stride[0];
894 int stride_uv = fenc->i_stride[1];
895 int off_y = 16 * (i_mb_x + i_mb_y * stride_y);
896 int off_uv = 8 * (i_mb_x + i_mb_y * stride_uv);
897 h->mc.prefetch_fenc( fenc->plane[0]+off_y, stride_y,
898 fenc->plane[1+(i_mb_x&1)]+off_uv, stride_uv, i_mb_x );
901 static NOINLINE void copy_column8( uint8_t *dst, uint8_t *src )
903 // input pointers are offset by 4 rows because that's faster (smaller instruction size on x86)
905 for( i = -4; i < 4; i++ )
906 dst[i*FDEC_STRIDE] = src[i*FDEC_STRIDE];
909 static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int i_mb_x, int i_mb_y, int i)
911 const int w = (i == 0 ? 16 : 8);
912 const int i_stride = h->fdec->i_stride[!!i];
913 const int i_stride2 = i_stride << h->mb.b_interlaced;
914 const int i_pix_offset = h->mb.b_interlaced
915 ? w * (i_mb_x + (i_mb_y&~1) * i_stride) + (i_mb_y&1) * i_stride
916 : w * (i_mb_x + i_mb_y * i_stride);
917 const uint8_t *plane_fdec = &h->fdec->plane[i][i_pix_offset];
918 const uint8_t *intra_fdec = h->param.b_sliced_threads ? plane_fdec-i_stride2 :
919 &h->mb.intra_border_backup[i_mb_y & h->sh.b_mbaff][i][i_mb_x*16>>!!i];
920 int ref_pix_offset[2] = { i_pix_offset, i_pix_offset };
921 x264_frame_t **fref[2] = { h->fref0, h->fref1 };
923 if( h->mb.b_interlaced )
924 ref_pix_offset[1] += (1-2*(i_mb_y&1)) * i_stride;
925 h->mb.pic.i_stride[i] = i_stride2;
926 h->mb.pic.p_fenc_plane[i] = &h->fenc->plane[i][i_pix_offset];
927 h->mc.copy[i?PIXEL_8x8:PIXEL_16x16]( h->mb.pic.p_fenc[i], FENC_STRIDE,
928 h->mb.pic.p_fenc_plane[i], i_stride2, w );
930 memcpy( &h->mb.pic.p_fdec[i][-1-FDEC_STRIDE], intra_fdec-1, w*3/2+1 );
932 memset( &h->mb.pic.p_fdec[i][-1-FDEC_STRIDE], 0, w*3/2+1 );
933 if( h->mb.b_interlaced || h->mb.b_reencode_mb )
934 for( j = 0; j < w; j++ )
935 h->mb.pic.p_fdec[i][-1+j*FDEC_STRIDE] = plane_fdec[-1+j*i_stride2];
936 for( j = 0; j < h->mb.pic.i_fref[0]; j++ )
938 h->mb.pic.p_fref[0][j][i==0 ? 0:i+3] = &fref[0][j >> h->mb.b_interlaced]->plane[i][ref_pix_offset[j&1]];
941 for( k = 1; k < 4; k++ )
942 h->mb.pic.p_fref[0][j][k] = &fref[0][j >> h->mb.b_interlaced]->filtered[k][ref_pix_offset[j&1]];
943 if( h->sh.weight[j][0].weightfn )
944 h->mb.pic.p_fref_w[j] = &h->fenc->weighted[j >> h->mb.b_interlaced][ref_pix_offset[j&1]];
946 h->mb.pic.p_fref_w[j] = h->mb.pic.p_fref[0][j][0];
949 if( h->sh.i_type == SLICE_TYPE_B )
950 for( j = 0; j < h->mb.pic.i_fref[1]; j++ )
952 h->mb.pic.p_fref[1][j][i==0 ? 0:i+3] = &fref[1][j >> h->mb.b_interlaced]->plane[i][ref_pix_offset[j&1]];
954 for( k = 1; k < 4; k++ )
955 h->mb.pic.p_fref[1][j][k] = &fref[1][j >> h->mb.b_interlaced]->filtered[k][ref_pix_offset[j&1]];
959 void x264_macroblock_cache_load( x264_t *h, int i_mb_x, int i_mb_y )
961 int i_mb_xy = i_mb_y * h->mb.i_mb_stride + i_mb_x;
962 int i_mb_4x4 = 4*(i_mb_y * h->mb.i_b4_stride + i_mb_x);
963 int i_mb_8x8 = 2*(i_mb_y * h->mb.i_b8_stride + i_mb_x);
964 int i_top_y = i_mb_y - (1 << h->mb.b_interlaced);
965 int i_top_xy = i_top_y * h->mb.i_mb_stride + i_mb_x;
966 int i_top_4x4 = (4*i_top_y+3) * h->mb.i_b4_stride + 4*i_mb_x;
967 int i_top_8x8 = (2*i_top_y+1) * h->mb.i_b8_stride + 2*i_mb_x;
969 int i_top_type = -1; /* gcc warn */
975 h->mb.i_mb_x = i_mb_x;
976 h->mb.i_mb_y = i_mb_y;
977 h->mb.i_mb_xy = i_mb_xy;
978 h->mb.i_b8_xy = i_mb_8x8;
979 h->mb.i_b4_xy = i_mb_4x4;
980 h->mb.i_mb_top_xy = i_top_xy;
981 h->mb.i_neighbour = 0;
982 h->mb.i_neighbour_intra = 0;
985 if( i_top_xy >= h->sh.i_first_mb )
987 h->mb.i_mb_type_top =
988 i_top_type = h->mb.type[i_top_xy];
989 h->mb.cache.i_cbp_top = h->mb.cbp[i_top_xy];
991 h->mb.i_neighbour |= MB_TOP;
993 if( !h->param.b_constrained_intra || IS_INTRA( i_top_type ) )
994 h->mb.i_neighbour_intra |= MB_TOP;
997 CP32( &h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8], &h->mb.intra4x4_pred_mode[i_top_xy][0] );
999 /* load non_zero_count */
1000 CP32( &h->mb.cache.non_zero_count[x264_scan8[0] - 8], &h->mb.non_zero_count[i_top_xy][12] );
1001 /* shift because x264_scan8[16] is misaligned */
1002 M32( &h->mb.cache.non_zero_count[x264_scan8[16+0] - 9] ) = M16( &h->mb.non_zero_count[i_top_xy][18] ) << 8;
1003 M32( &h->mb.cache.non_zero_count[x264_scan8[16+4] - 9] ) = M16( &h->mb.non_zero_count[i_top_xy][22] ) << 8;
1007 h->mb.i_mb_type_top = -1;
1008 h->mb.cache.i_cbp_top = -1;
1011 M32( &h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8] ) = 0xFFFFFFFFU;
1013 /* load non_zero_count */
1014 M32( &h->mb.cache.non_zero_count[x264_scan8[ 0] - 8] ) = 0x80808080U;
1015 M32( &h->mb.cache.non_zero_count[x264_scan8[16+0] - 9] ) = 0x80808080U;
1016 M32( &h->mb.cache.non_zero_count[x264_scan8[16+4] - 9] ) = 0x80808080U;
1019 if( i_mb_x > 0 && i_mb_xy > h->sh.i_first_mb )
1021 i_left_xy = i_mb_xy - 1;
1022 h->mb.i_mb_type_left =
1023 i_left_type = h->mb.type[i_left_xy];
1024 h->mb.cache.i_cbp_left = h->mb.cbp[h->mb.i_mb_xy - 1];
1026 h->mb.i_neighbour |= MB_LEFT;
1028 if( !h->param.b_constrained_intra || IS_INTRA( i_left_type ) )
1029 h->mb.i_neighbour_intra |= MB_LEFT;
1032 h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][4];
1033 h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][5];
1034 h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][6];
1035 h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][3];
1037 /* load non_zero_count */
1038 h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = h->mb.non_zero_count[i_left_xy][3];
1039 h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = h->mb.non_zero_count[i_left_xy][7];
1040 h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = h->mb.non_zero_count[i_left_xy][11];
1041 h->mb.cache.non_zero_count[x264_scan8[10] - 1] = h->mb.non_zero_count[i_left_xy][15];
1043 h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] = h->mb.non_zero_count[i_left_xy][16+1];
1044 h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] = h->mb.non_zero_count[i_left_xy][16+3];
1046 h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] = h->mb.non_zero_count[i_left_xy][16+4+1];
1047 h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = h->mb.non_zero_count[i_left_xy][16+4+3];
1051 h->mb.i_mb_type_left = -1;
1052 h->mb.cache.i_cbp_left = -1;
1054 h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] =
1055 h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] =
1056 h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] =
1057 h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = -1;
1059 /* load non_zero_count */
1060 h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] =
1061 h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] =
1062 h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] =
1063 h->mb.cache.non_zero_count[x264_scan8[10] - 1] =
1064 h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] =
1065 h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] =
1066 h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] =
1067 h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = 0x80;
1070 if( i_mb_x < h->sps->i_mb_width - 1 && i_top_xy + 1 >= h->sh.i_first_mb )
1072 h->mb.i_neighbour |= MB_TOPRIGHT;
1073 h->mb.i_mb_type_topright = h->mb.type[ i_top_xy + 1 ];
1074 if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_topright ) )
1075 h->mb.i_neighbour_intra |= MB_TOPRIGHT;
1078 h->mb.i_mb_type_topright = -1;
1079 if( i_mb_x > 0 && i_top_xy - 1 >= h->sh.i_first_mb )
1081 h->mb.i_neighbour |= MB_TOPLEFT;
1082 h->mb.i_mb_type_topleft = h->mb.type[ i_top_xy - 1 ];
1083 if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_topleft ) )
1084 h->mb.i_neighbour_intra |= MB_TOPLEFT;
1087 h->mb.i_mb_type_topleft = -1;
1089 if( h->pps->b_transform_8x8_mode )
1091 h->mb.cache.i_neighbour_transform_size =
1092 ( i_left_type >= 0 && h->mb.mb_transform_size[i_left_xy] )
1093 + ( i_top_type >= 0 && h->mb.mb_transform_size[i_top_xy] );
1098 h->mb.pic.i_fref[0] = h->i_ref0 << h->mb.b_interlaced;
1099 h->mb.pic.i_fref[1] = h->i_ref1 << h->mb.b_interlaced;
1100 h->mb.cache.i_neighbour_interlaced =
1101 !!(h->mb.i_neighbour & MB_LEFT)
1102 + !!(h->mb.i_neighbour & MB_TOP);
1105 if( !h->mb.b_interlaced && !h->mb.b_reencode_mb )
1107 copy_column8( h->mb.pic.p_fdec[0]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+ 4*FDEC_STRIDE );
1108 copy_column8( h->mb.pic.p_fdec[0]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+12*FDEC_STRIDE );
1109 copy_column8( h->mb.pic.p_fdec[1]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[1]+ 7+ 4*FDEC_STRIDE );
1110 copy_column8( h->mb.pic.p_fdec[2]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[2]+ 7+ 4*FDEC_STRIDE );
1113 /* load picture pointers */
1114 x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 0 );
1115 x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 1 );
1116 x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 2 );
1118 if( h->fdec->integral )
1120 assert( !h->mb.b_interlaced );
1121 for( i = 0; i < h->mb.pic.i_fref[0]; i++ )
1122 h->mb.pic.p_integral[0][i] = &h->fref0[i]->integral[ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
1123 for( i = 0; i < h->mb.pic.i_fref[1]; i++ )
1124 h->mb.pic.p_integral[1][i] = &h->fref1[i]->integral[ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
1127 x264_prefetch_fenc( h, h->fenc, i_mb_x, i_mb_y );
1129 /* load ref/mv/mvd */
1130 if( h->sh.i_type != SLICE_TYPE_I )
1132 const int s8x8 = h->mb.i_b8_stride;
1133 const int s4x4 = h->mb.i_b4_stride;
1137 for( i_list = 0; i_list < (h->sh.i_type == SLICE_TYPE_B ? 2 : 1 ); i_list++ )
1140 h->mb.cache.ref[i_list][x264_scan8[5 ]+1] =
1141 h->mb.cache.ref[i_list][x264_scan8[7 ]+1] =
1142 h->mb.cache.ref[i_list][x264_scan8[13]+1] = -2;
1145 if( h->mb.i_neighbour & MB_TOPLEFT )
1147 const int i8 = x264_scan8[0] - 1 - 1*8;
1148 const int ir = i_top_8x8 - 1;
1149 const int iv = i_top_4x4 - 1;
1150 h->mb.cache.ref[i_list][i8] = h->mb.ref[i_list][ir];
1151 CP32( h->mb.cache.mv[i_list][i8], h->mb.mv[i_list][iv] );
1155 const int i8 = x264_scan8[0] - 1 - 1*8;
1156 h->mb.cache.ref[i_list][i8] = -2;
1157 M32( h->mb.cache.mv[i_list][i8] ) = 0;
1160 if( h->mb.i_neighbour & MB_TOP )
1162 const int i8 = x264_scan8[0] - 8;
1163 const int ir = i_top_8x8;
1164 const int iv = i_top_4x4;
1165 h->mb.cache.ref[i_list][i8+0] =
1166 h->mb.cache.ref[i_list][i8+1] = h->mb.ref[i_list][ir + 0];
1167 h->mb.cache.ref[i_list][i8+2] =
1168 h->mb.cache.ref[i_list][i8+3] = h->mb.ref[i_list][ir + 1];
1169 CP64( h->mb.cache.mv[i_list][i8+0], h->mb.mv[i_list][iv+0] );
1170 CP64( h->mb.cache.mv[i_list][i8+2], h->mb.mv[i_list][iv+2] );
1174 const int i8 = x264_scan8[0] - 8;
1175 M64( h->mb.cache.mv[i_list][i8+0] ) = 0;
1176 M64( h->mb.cache.mv[i_list][i8+2] ) = 0;
1177 M32( &h->mb.cache.ref[i_list][i8] ) = (uint8_t)(-2) * 0x01010101U;
1180 if( h->mb.i_neighbour & MB_TOPRIGHT )
1182 const int i8 = x264_scan8[0] + 4 - 1*8;
1183 const int ir = i_top_8x8 + 2;
1184 const int iv = i_top_4x4 + 4;
1185 h->mb.cache.ref[i_list][i8] = h->mb.ref[i_list][ir];
1186 CP32( h->mb.cache.mv[i_list][i8], h->mb.mv[i_list][iv] );
1190 const int i8 = x264_scan8[0] + 4 - 1*8;
1191 h->mb.cache.ref[i_list][i8] = -2;
1194 if( h->mb.i_neighbour & MB_LEFT )
1196 const int i8 = x264_scan8[0] - 1;
1197 const int ir = i_mb_8x8 - 1;
1198 const int iv = i_mb_4x4 - 1;
1199 h->mb.cache.ref[i_list][i8+0*8] =
1200 h->mb.cache.ref[i_list][i8+1*8] = h->mb.ref[i_list][ir + 0*s8x8];
1201 h->mb.cache.ref[i_list][i8+2*8] =
1202 h->mb.cache.ref[i_list][i8+3*8] = h->mb.ref[i_list][ir + 1*s8x8];
1204 CP32( h->mb.cache.mv[i_list][i8+0*8], h->mb.mv[i_list][iv + 0*s4x4] );
1205 CP32( h->mb.cache.mv[i_list][i8+1*8], h->mb.mv[i_list][iv + 1*s4x4] );
1206 CP32( h->mb.cache.mv[i_list][i8+2*8], h->mb.mv[i_list][iv + 2*s4x4] );
1207 CP32( h->mb.cache.mv[i_list][i8+3*8], h->mb.mv[i_list][iv + 3*s4x4] );
1211 const int i8 = x264_scan8[0] - 1;
1212 for( i = 0; i < 4; i++ )
1214 h->mb.cache.ref[i_list][i8+i*8] = -2;
1215 M32( h->mb.cache.mv[i_list][i8+i*8] ) = 0;
1219 if( h->param.b_cabac )
1221 if( i_top_type >= 0 )
1222 CP64( h->mb.cache.mvd[i_list][x264_scan8[0]-8], h->mb.mvd[i_list][i_top_xy*8] );
1224 M64( h->mb.cache.mvd[i_list][x264_scan8[0]-8] ) = 0;
1226 if( i_left_type >= 0 )
1228 CP16( h->mb.cache.mvd[i_list][x264_scan8[0]-1+0*8], h->mb.mvd[i_list][i_left_xy*8+4] );
1229 CP16( h->mb.cache.mvd[i_list][x264_scan8[0]-1+1*8], h->mb.mvd[i_list][i_left_xy*8+5] );
1230 CP16( h->mb.cache.mvd[i_list][x264_scan8[0]-1+2*8], h->mb.mvd[i_list][i_left_xy*8+6] );
1231 CP16( h->mb.cache.mvd[i_list][x264_scan8[0]-1+3*8], h->mb.mvd[i_list][i_left_xy*8+3] );
1234 for( i = 0; i < 4; i++ )
1235 M16( h->mb.cache.mvd[i_list][x264_scan8[0]-1+i*8] ) = 0;
1240 if( h->sh.i_type == SLICE_TYPE_B )
1242 h->mb.bipred_weight = h->mb.bipred_weight_buf[h->mb.b_interlaced&(i_mb_y&1)];
1243 h->mb.dist_scale_factor = h->mb.dist_scale_factor_buf[h->mb.b_interlaced&(i_mb_y&1)];
1244 if( h->param.b_cabac )
1247 x264_macroblock_cache_skip( h, 0, 0, 4, 4, 0 );
1248 skipbp = i_left_type >= 0 ? h->mb.skipbp[i_left_xy] : 0;
1249 h->mb.cache.skip[x264_scan8[0] - 1] = skipbp & 0x2;
1250 h->mb.cache.skip[x264_scan8[8] - 1] = skipbp & 0x8;
1251 skipbp = i_top_type >= 0 ? h->mb.skipbp[i_top_xy] : 0;
1252 h->mb.cache.skip[x264_scan8[0] - 8] = skipbp & 0x4;
1253 h->mb.cache.skip[x264_scan8[4] - 8] = skipbp & 0x8;
1257 if( h->sh.i_type == SLICE_TYPE_P )
1258 x264_mb_predict_mv_pskip( h, h->mb.cache.pskip_mv );
1261 h->mb.i_neighbour4[0] =
1262 h->mb.i_neighbour8[0] = (h->mb.i_neighbour_intra & (MB_TOP|MB_LEFT|MB_TOPLEFT))
1263 | ((h->mb.i_neighbour_intra & MB_TOP) ? MB_TOPRIGHT : 0);
1264 h->mb.i_neighbour4[4] =
1265 h->mb.i_neighbour4[1] = MB_LEFT | ((h->mb.i_neighbour_intra & MB_TOP) ? (MB_TOP|MB_TOPLEFT|MB_TOPRIGHT) : 0);
1266 h->mb.i_neighbour4[2] =
1267 h->mb.i_neighbour4[8] =
1268 h->mb.i_neighbour4[10] =
1269 h->mb.i_neighbour8[2] = MB_TOP|MB_TOPRIGHT | ((h->mb.i_neighbour_intra & MB_LEFT) ? (MB_LEFT|MB_TOPLEFT) : 0);
1270 h->mb.i_neighbour4[5] =
1271 h->mb.i_neighbour8[1] = MB_LEFT | (h->mb.i_neighbour_intra & MB_TOPRIGHT)
1272 | ((h->mb.i_neighbour_intra & MB_TOP) ? MB_TOP|MB_TOPLEFT : 0);
1275 static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int i )
1278 int i_stride = h->fdec->i_stride[!!i];
1279 int i_stride2 = i_stride << h->mb.b_interlaced;
1280 int i_pix_offset = h->mb.b_interlaced
1281 ? w * (h->mb.i_mb_x + (h->mb.i_mb_y&~1) * i_stride) + (h->mb.i_mb_y&1) * i_stride
1282 : w * (h->mb.i_mb_x + h->mb.i_mb_y * i_stride);
1283 h->mc.copy[i?PIXEL_8x8:PIXEL_16x16](
1284 &h->fdec->plane[i][i_pix_offset], i_stride2,
1285 h->mb.pic.p_fdec[i], FDEC_STRIDE, w );
1288 void x264_macroblock_cache_save( x264_t *h )
1290 const int i_mb_xy = h->mb.i_mb_xy;
1291 const int i_mb_type = x264_mb_type_fix[h->mb.i_type];
1292 const int s8x8 = h->mb.i_b8_stride;
1293 const int s4x4 = h->mb.i_b4_stride;
1294 const int i_mb_4x4 = h->mb.i_b4_xy;
1295 const int i_mb_8x8 = h->mb.i_b8_xy;
1297 /* GCC pessimizes direct stores to heap-allocated 8-bit arrays due to aliasing.*/
1298 /* By only dereferencing them once, we avoid this issue. */
1299 int8_t *intra4x4_pred_mode = h->mb.intra4x4_pred_mode[i_mb_xy];
1300 uint8_t *non_zero_count = h->mb.non_zero_count[i_mb_xy];
1304 x264_macroblock_store_pic( h, 0 );
1305 x264_macroblock_store_pic( h, 1 );
1306 x264_macroblock_store_pic( h, 2 );
1308 x264_prefetch_fenc( h, h->fdec, h->mb.i_mb_x, h->mb.i_mb_y );
1310 h->mb.type[i_mb_xy] = i_mb_type;
1311 h->mb.partition[i_mb_xy] = IS_INTRA( i_mb_type ) ? D_16x16 : h->mb.i_partition;
1312 h->mb.i_mb_prev_xy = i_mb_xy;
1315 if( i_mb_type == I_4x4 )
1317 CP32( &intra4x4_pred_mode[0], &h->mb.cache.intra4x4_pred_mode[x264_scan8[10]] );
1318 M32( &intra4x4_pred_mode[4] ) = pack8to32(h->mb.cache.intra4x4_pred_mode[x264_scan8[5] ],
1319 h->mb.cache.intra4x4_pred_mode[x264_scan8[7] ],
1320 h->mb.cache.intra4x4_pred_mode[x264_scan8[13] ], 0);
1322 else if( !h->param.b_constrained_intra || IS_INTRA(i_mb_type) )
1323 M64( intra4x4_pred_mode ) = I_PRED_4x4_DC * 0x0101010101010101ULL;
1325 M64( intra4x4_pred_mode ) = (uint8_t)(-1) * 0x0101010101010101ULL;
1328 if( i_mb_type == I_PCM )
1330 h->mb.qp[i_mb_xy] = 0;
1331 h->mb.i_last_dqp = 0;
1332 h->mb.i_cbp_chroma = 2;
1333 h->mb.i_cbp_luma = 0xf;
1334 h->mb.cbp[i_mb_xy] = 0x72f; /* all set */
1335 h->mb.b_transform_8x8 = 0;
1336 memset( non_zero_count, 16, 24 );
1340 /* save non zero count */
1341 CP32( &non_zero_count[0*4], &h->mb.cache.non_zero_count[x264_scan8[0]+0*8] );
1342 CP32( &non_zero_count[1*4], &h->mb.cache.non_zero_count[x264_scan8[0]+1*8] );
1343 CP32( &non_zero_count[2*4], &h->mb.cache.non_zero_count[x264_scan8[0]+2*8] );
1344 CP32( &non_zero_count[3*4], &h->mb.cache.non_zero_count[x264_scan8[0]+3*8] );
1345 M16( &non_zero_count[16+0*2] ) = M32( &h->mb.cache.non_zero_count[x264_scan8[16+0*2]-1] ) >> 8;
1346 M16( &non_zero_count[16+1*2] ) = M32( &h->mb.cache.non_zero_count[x264_scan8[16+1*2]-1] ) >> 8;
1347 M16( &non_zero_count[16+2*2] ) = M32( &h->mb.cache.non_zero_count[x264_scan8[16+2*2]-1] ) >> 8;
1348 M16( &non_zero_count[16+3*2] ) = M32( &h->mb.cache.non_zero_count[x264_scan8[16+3*2]-1] ) >> 8;
1350 if( h->mb.i_type != I_16x16 && h->mb.i_cbp_luma == 0 && h->mb.i_cbp_chroma == 0 )
1351 h->mb.i_qp = h->mb.i_last_qp;
1352 h->mb.qp[i_mb_xy] = h->mb.i_qp;
1353 h->mb.i_last_dqp = h->mb.i_qp - h->mb.i_last_qp;
1354 h->mb.i_last_qp = h->mb.i_qp;
1357 if( h->mb.i_cbp_luma == 0 && h->mb.i_type != I_8x8 )
1358 h->mb.b_transform_8x8 = 0;
1359 h->mb.mb_transform_size[i_mb_xy] = h->mb.b_transform_8x8;
1361 if( h->sh.i_type != SLICE_TYPE_I )
1363 if( !IS_INTRA( i_mb_type ) )
1365 h->mb.ref[0][i_mb_8x8+0+0*s8x8] = h->mb.cache.ref[0][x264_scan8[0]];
1366 h->mb.ref[0][i_mb_8x8+1+0*s8x8] = h->mb.cache.ref[0][x264_scan8[4]];
1367 h->mb.ref[0][i_mb_8x8+0+1*s8x8] = h->mb.cache.ref[0][x264_scan8[8]];
1368 h->mb.ref[0][i_mb_8x8+1+1*s8x8] = h->mb.cache.ref[0][x264_scan8[12]];
1369 for( y = 0; y < 4; y++ )
1371 CP64( h->mb.mv[0][i_mb_4x4+y*s4x4+0], h->mb.cache.mv[0][x264_scan8[0]+8*y+0] );
1372 CP64( h->mb.mv[0][i_mb_4x4+y*s4x4+2], h->mb.cache.mv[0][x264_scan8[0]+8*y+2] );
1374 if( h->sh.i_type == SLICE_TYPE_B )
1376 h->mb.ref[1][i_mb_8x8+0+0*s8x8] = h->mb.cache.ref[1][x264_scan8[0]];
1377 h->mb.ref[1][i_mb_8x8+1+0*s8x8] = h->mb.cache.ref[1][x264_scan8[4]];
1378 h->mb.ref[1][i_mb_8x8+0+1*s8x8] = h->mb.cache.ref[1][x264_scan8[8]];
1379 h->mb.ref[1][i_mb_8x8+1+1*s8x8] = h->mb.cache.ref[1][x264_scan8[12]];
1380 for( y = 0; y < 4; y++ )
1382 CP64( h->mb.mv[1][i_mb_4x4+y*s4x4+0], h->mb.cache.mv[1][x264_scan8[0]+8*y+0] );
1383 CP64( h->mb.mv[1][i_mb_4x4+y*s4x4+2], h->mb.cache.mv[1][x264_scan8[0]+8*y+2] );
1390 for( i_list = 0; i_list < (h->sh.i_type == SLICE_TYPE_B ? 2 : 1 ); i_list++ )
1392 M16( &h->mb.ref[i_list][i_mb_8x8+0*s8x8] ) = (uint8_t)(-1) * 0x0101;
1393 M16( &h->mb.ref[i_list][i_mb_8x8+1*s8x8] ) = (uint8_t)(-1) * 0x0101;
1394 for( y = 0; y < 4; y++ )
1396 M64( h->mb.mv[i_list][i_mb_4x4+y*s4x4+0] ) = 0;
1397 M64( h->mb.mv[i_list][i_mb_4x4+y*s4x4+2] ) = 0;
1403 if( h->param.b_cabac )
1405 if( IS_INTRA(i_mb_type) && i_mb_type != I_PCM )
1406 h->mb.chroma_pred_mode[i_mb_xy] = x264_mb_pred_mode8x8c_fix[ h->mb.i_chroma_pred_mode ];
1408 h->mb.chroma_pred_mode[i_mb_xy] = I_PRED_CHROMA_DC;
1410 if( !IS_INTRA( i_mb_type ) && !IS_SKIP( i_mb_type ) && !IS_DIRECT( i_mb_type ) )
1412 CP64( h->mb.mvd[0][i_mb_xy*8+0], h->mb.cache.mvd[0][x264_scan8[0]+8*3] );
1413 CP16( h->mb.mvd[0][i_mb_xy*8+4], h->mb.cache.mvd[0][x264_scan8[0]+8*0+3] );
1414 CP16( h->mb.mvd[0][i_mb_xy*8+5], h->mb.cache.mvd[0][x264_scan8[0]+8*1+3] );
1415 CP16( h->mb.mvd[0][i_mb_xy*8+6], h->mb.cache.mvd[0][x264_scan8[0]+8*2+3] );
1416 if( h->sh.i_type == SLICE_TYPE_B )
1418 CP64( h->mb.mvd[1][i_mb_xy*8+0], h->mb.cache.mvd[1][x264_scan8[0]+8*3] );
1419 CP16( h->mb.mvd[1][i_mb_xy*8+4], h->mb.cache.mvd[1][x264_scan8[0]+8*0+3] );
1420 CP16( h->mb.mvd[1][i_mb_xy*8+5], h->mb.cache.mvd[1][x264_scan8[0]+8*1+3] );
1421 CP16( h->mb.mvd[1][i_mb_xy*8+6], h->mb.cache.mvd[1][x264_scan8[0]+8*2+3] );
1426 M64( h->mb.mvd[0][i_mb_xy*8+0] ) = 0;
1427 M64( h->mb.mvd[0][i_mb_xy*8+4] ) = 0;
1428 if( h->sh.i_type == SLICE_TYPE_B )
1430 M64( h->mb.mvd[1][i_mb_xy*8+0] ) = 0;
1431 M64( h->mb.mvd[1][i_mb_xy*8+4] ) = 0;
1435 if( h->sh.i_type == SLICE_TYPE_B )
1437 if( i_mb_type == B_SKIP || i_mb_type == B_DIRECT )
1438 h->mb.skipbp[i_mb_xy] = 0xf;
1439 else if( i_mb_type == B_8x8 )
1441 int skipbp = ( h->mb.i_sub_partition[0] == D_DIRECT_8x8 ) << 0;
1442 skipbp |= ( h->mb.i_sub_partition[1] == D_DIRECT_8x8 ) << 1;
1443 skipbp |= ( h->mb.i_sub_partition[2] == D_DIRECT_8x8 ) << 2;
1444 skipbp |= ( h->mb.i_sub_partition[3] == D_DIRECT_8x8 ) << 3;
1445 h->mb.skipbp[i_mb_xy] = skipbp;
1448 h->mb.skipbp[i_mb_xy] = 0;
1454 void x264_macroblock_bipred_init( x264_t *h )
1456 int i_ref0, i_ref1, field;
1457 for( field = 0; field <= h->sh.b_mbaff; field++ )
1458 for( i_ref0 = 0; i_ref0 < (h->i_ref0<<h->sh.b_mbaff); i_ref0++ )
1460 int poc0 = h->fref0[i_ref0>>h->sh.b_mbaff]->i_poc;
1461 if( h->sh.b_mbaff && field^(i_ref0&1) )
1462 poc0 += h->sh.i_delta_poc_bottom;
1463 for( i_ref1 = 0; i_ref1 < (h->i_ref1<<h->sh.b_mbaff); i_ref1++ )
1465 int dist_scale_factor;
1466 int poc1 = h->fref1[i_ref1>>h->sh.b_mbaff]->i_poc;
1467 if( h->sh.b_mbaff && field^(i_ref1&1) )
1468 poc1 += h->sh.i_delta_poc_bottom;
1469 int cur_poc = h->fdec->i_poc + field*h->sh.i_delta_poc_bottom;
1470 int td = x264_clip3( poc1 - poc0, -128, 127 );
1471 if( td == 0 /* || pic0 is a long-term ref */ )
1472 dist_scale_factor = 256;
1475 int tb = x264_clip3( cur_poc - poc0, -128, 127 );
1476 int tx = (16384 + (abs(td) >> 1)) / td;
1477 dist_scale_factor = x264_clip3( (tb * tx + 32) >> 6, -1024, 1023 );
1480 h->mb.dist_scale_factor_buf[field][i_ref0][i_ref1] = dist_scale_factor;
1482 dist_scale_factor >>= 2;
1483 if( h->param.analyse.b_weighted_bipred
1484 && dist_scale_factor >= -64
1485 && dist_scale_factor <= 128 )
1487 h->mb.bipred_weight_buf[field][i_ref0][i_ref1] = 64 - dist_scale_factor;
1488 // ssse3 implementation of biweight doesn't support the extrema.
1489 // if we ever generate them, we'll have to drop that optimization.
1490 assert( dist_scale_factor >= -63 && dist_scale_factor <= 127 );
1493 h->mb.bipred_weight_buf[field][i_ref0][i_ref1] = 32;