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];
41 if( (idx&3) >= 2 + (i_width&1) || i_refc == -2 )
43 i_refc = h->mb.cache.ref[i_list][i8 - 8 - 1];
44 mv_c = h->mb.cache.mv[i_list][i8 - 8 - 1];
47 if( h->mb.i_partition == D_16x8 )
53 *(uint32_t*)mvp = *(uint32_t*)mv_b;
61 *(uint32_t*)mvp = *(uint32_t*)mv_a;
66 else if( h->mb.i_partition == D_8x16 )
72 *(uint32_t*)mvp = *(uint32_t*)mv_a;
80 *(uint32_t*)mvp = *(uint32_t*)mv_c;
86 if( i_refa == i_ref ) i_count++;
87 if( i_refb == i_ref ) i_count++;
88 if( i_refc == i_ref ) i_count++;
93 x264_median_mv( mvp, mv_a, mv_b, mv_c );
95 else if( i_count == 1 )
98 *(uint32_t*)mvp = *(uint32_t*)mv_a;
99 else if( i_refb == i_ref )
100 *(uint32_t*)mvp = *(uint32_t*)mv_b;
102 *(uint32_t*)mvp = *(uint32_t*)mv_c;
104 else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
105 *(uint32_t*)mvp = *(uint32_t*)mv_a;
110 void x264_mb_predict_mv_16x16( x264_t *h, int i_list, int i_ref, int16_t mvp[2] )
112 int i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
113 int16_t *mv_a = h->mb.cache.mv[i_list][X264_SCAN8_0 - 1];
114 int i_refb = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8];
115 int16_t *mv_b = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8];
116 int i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
117 int16_t *mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 + 4];
123 i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 - 1];
124 mv_c = h->mb.cache.mv[i_list][X264_SCAN8_0 - 8 - 1];
127 if( i_refa == i_ref ) i_count++;
128 if( i_refb == i_ref ) i_count++;
129 if( i_refc == i_ref ) i_count++;
134 x264_median_mv( mvp, mv_a, mv_b, mv_c );
136 else if( i_count == 1 )
138 if( i_refa == i_ref )
139 *(uint32_t*)mvp = *(uint32_t*)mv_a;
140 else if( i_refb == i_ref )
141 *(uint32_t*)mvp = *(uint32_t*)mv_b;
143 *(uint32_t*)mvp = *(uint32_t*)mv_c;
145 else if( i_refb == -2 && i_refc == -2 && i_refa != -2 )
146 *(uint32_t*)mvp = *(uint32_t*)mv_a;
152 void x264_mb_predict_mv_pskip( x264_t *h, int16_t mv[2] )
154 int i_refa = h->mb.cache.ref[0][X264_SCAN8_0 - 1];
155 int i_refb = h->mb.cache.ref[0][X264_SCAN8_0 - 8];
156 int16_t *mv_a = h->mb.cache.mv[0][X264_SCAN8_0 - 1];
157 int16_t *mv_b = h->mb.cache.mv[0][X264_SCAN8_0 - 8];
159 if( i_refa == -2 || i_refb == -2 ||
160 !( i_refa | *(uint32_t*)mv_a ) ||
161 !( i_refb | *(uint32_t*)mv_b ) )
167 x264_mb_predict_mv_16x16( h, 0, 0, mv );
171 static int x264_mb_predict_mv_direct16x16_temporal( x264_t *h )
173 int i_mb_4x4 = 16 * h->mb.i_mb_stride * h->mb.i_mb_y + 4 * h->mb.i_mb_x;
174 int i_mb_8x8 = 4 * h->mb.i_mb_stride * h->mb.i_mb_y + 2 * h->mb.i_mb_x;
176 const int type_col = h->fref1[0]->mb_type[h->mb.i_mb_xy];
178 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
180 if( IS_INTRA( type_col ) )
182 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
183 x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, 0 );
184 x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, 0 );
188 for( i8 = 0; i8 < 4; i8++ )
192 const int i_part_8x8 = i_mb_8x8 + x8 + y8 * h->mb.i_b8_stride;
193 const int i_ref = map_col_to_list0(h->fref1[0]->ref[0][i_part_8x8]);
197 const int dist_scale_factor = h->mb.dist_scale_factor[i_ref][0];
198 const int16_t *mv_col = h->fref1[0]->mv[0][ i_mb_4x4 + 3*x8 + 3*y8 * h->mb.i_b4_stride];
199 const int l0x = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
200 const int l0y = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
201 if( h->param.i_threads > 1 && (l0y > h->mb.mv_max_spel[1] || l0y-mv_col[1] > h->mb.mv_max_spel[1]) )
203 x264_macroblock_cache_ref( h, 2*x8, 2*y8, 2, 2, 0, i_ref );
204 x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, pack16to32_mask(l0x, l0y) );
205 x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_col[1]) );
209 /* the collocated ref isn't in the current list0 */
210 /* FIXME: we might still be able to use direct_8x8 on some partitions */
211 /* FIXME: with B-pyramid + extensive ref list reordering
212 * (not currently used), we would also have to check
213 * l1mv1 like in spatial mode */
221 static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
224 ALIGNED_8( int16_t mv[2][2] );
227 const int8_t *l1ref0 = &h->fref1[0]->ref[0][ h->mb.i_b8_xy ];
228 const int8_t *l1ref1 = &h->fref1[0]->ref[1][ h->mb.i_b8_xy ];
229 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->fref1[0]->mv[0][ h->mb.i_b4_xy ];
230 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->fref1[0]->mv[1][ h->mb.i_b4_xy ];
231 const int type_col = h->fref1[0]->mb_type[ h->mb.i_mb_xy ];
233 for( i_list=0; i_list<2; i_list++ )
235 int i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
236 int i_refb = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8];
237 int i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 + 4];
239 i_refc = h->mb.cache.ref[i_list][X264_SCAN8_0 - 8 - 1];
241 ref[i_list] = i_refa;
242 if( ref[i_list] < 0 || ( i_refb < ref[i_list] && i_refb >= 0 ))
243 ref[i_list] = i_refb;
244 if( ref[i_list] < 0 || ( i_refc < ref[i_list] && i_refc >= 0 ))
245 ref[i_list] = i_refc;
246 if( ref[i_list] < 0 )
250 if( ref[0] < 0 && ref[1] < 0 )
252 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
253 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
254 x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, 0 );
255 x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, 0 );
260 x264_mb_predict_mv_16x16( h, 0, ref[0], mv[0] );
267 x264_mb_predict_mv_16x16( h, 1, ref[1], mv[1] );
274 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, ref[0] );
275 x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, ref[1] );
276 x264_macroblock_cache_mv_ptr( h, 0, 0, 4, 4, 0, mv[0] );
277 x264_macroblock_cache_mv_ptr( h, 0, 0, 4, 4, 1, mv[1] );
279 if( h->param.i_threads > 1
280 && ( mv[0][1] > h->mb.mv_max_spel[1]
281 || mv[1][1] > h->mb.mv_max_spel[1] ) )
284 fprintf(stderr, "direct_spatial: (%d,%d) (%d,%d) > %d \n",
285 mv[0][0], mv[0][1], mv[1][0], mv[1][1],
286 h->mb.mv_max_spel[1]);
291 if( IS_INTRA( type_col ) || (ref[0]&&ref[1]) )
295 for( i8=0; i8<4; i8++ )
299 const int o8 = x8 + y8 * h->mb.i_b8_stride;
300 const int o4 = 3*(x8 + y8 * h->mb.i_b4_stride);
301 if( l1ref0[o8] == 0 )
303 if( abs( l1mv0[o4][0] ) <= 1 && abs( l1mv0[o4][1] ) <= 1 )
305 if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, 0 );
306 if( ref[1] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, 0 );
309 else if( l1ref0[o8] < 0 && l1ref1[o8] == 0 )
311 if( abs( l1mv1[o4][0] ) <= 1 && abs( l1mv1[o4][1] ) <= 1 )
313 if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 0, 0 );
314 if( ref[1] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, 1, 0 );
322 int x264_mb_predict_mv_direct16x16( x264_t *h, int *b_changed )
325 if( h->param.analyse.i_direct_mv_pred == X264_DIRECT_PRED_NONE )
327 else if( h->sh.b_direct_spatial_mv_pred )
328 b_available = x264_mb_predict_mv_direct16x16_spatial( h );
330 b_available = x264_mb_predict_mv_direct16x16_temporal( h );
332 if( b_changed != NULL && b_available )
334 int type_col = h->fref1[0]->mb_type[ h->mb.i_mb_xy ];
335 if( IS_INTRA(type_col) || type_col == P_SKIP )
337 *b_changed = h->mb.cache.direct_ref[0][0] != h->mb.cache.ref[0][X264_SCAN8_0]
338 || h->mb.cache.direct_ref[1][0] != h->mb.cache.ref[1][X264_SCAN8_0]
339 || *(uint32_t*)h->mb.cache.direct_mv[0][X264_SCAN8_0] != *(uint32_t*)h->mb.cache.mv[0][X264_SCAN8_0]
340 || *(uint32_t*)h->mb.cache.direct_mv[1][X264_SCAN8_0] != *(uint32_t*)h->mb.cache.mv[1][X264_SCAN8_0];
346 for( l = 0; l < 2; l++ )
347 for( i = 0; i < 4; i++ )
348 *b_changed |= h->mb.cache.direct_ref[l][i] != h->mb.cache.ref[l][x264_scan8[i*4]];
349 *b_changed = *b_changed || memcmp(h->mb.cache.direct_mv, h->mb.cache.mv, sizeof(h->mb.cache.mv));
359 for( l = 0; l < 2; l++ )
360 for( i = 0; i < 4; i++ )
361 h->mb.cache.direct_ref[l][i] = h->mb.cache.ref[l][x264_scan8[i*4]];
362 h->mc.memcpy_aligned(h->mb.cache.direct_mv, h->mb.cache.mv, sizeof(h->mb.cache.mv));
368 void x264_mb_load_mv_direct8x8( x264_t *h, int idx )
370 const int x = 2*(idx%2);
371 const int y = 2*(idx/2);
372 x264_macroblock_cache_ref( h, x, y, 2, 2, 0, h->mb.cache.direct_ref[0][idx] );
373 x264_macroblock_cache_ref( h, x, y, 2, 2, 1, h->mb.cache.direct_ref[1][idx] );
374 *(uint64_t*)h->mb.cache.mv[0][x264_scan8[idx*4]] =
375 *(uint64_t*)h->mb.cache.direct_mv[0][x264_scan8[idx*4]];
376 *(uint64_t*)h->mb.cache.mv[0][x264_scan8[idx*4]+8] =
377 *(uint64_t*)h->mb.cache.direct_mv[0][x264_scan8[idx*4]+8];
378 *(uint64_t*)h->mb.cache.mv[1][x264_scan8[idx*4]] =
379 *(uint64_t*)h->mb.cache.direct_mv[1][x264_scan8[idx*4]];
380 *(uint64_t*)h->mb.cache.mv[1][x264_scan8[idx*4]+8] =
381 *(uint64_t*)h->mb.cache.direct_mv[1][x264_scan8[idx*4]+8];
384 /* This just improves encoder performance, it's not part of the spec */
385 void x264_mb_predict_mv_ref16x16( x264_t *h, int i_list, int i_ref, int16_t mvc[9][2], int *i_mvc )
387 int16_t (*mvr)[2] = h->mb.mvr[i_list][i_ref];
390 #define SET_MVP(mvp) { \
391 *(uint32_t*)mvc[i] = *(uint32_t*)mvp; \
396 if( h->sh.i_type == SLICE_TYPE_B
397 && h->mb.cache.ref[i_list][x264_scan8[12]] == i_ref )
399 SET_MVP( h->mb.cache.mv[i_list][x264_scan8[12]] );
402 if( i_ref == 0 && h->frames.b_have_lowres )
404 int16_t (*lowres_mv)[2] = i_list ? h->fenc->lowres_mvs[1][h->fref1[0]->i_frame-h->fenc->i_frame-1]
405 : h->fenc->lowres_mvs[0][h->fenc->i_frame-h->fref0[0]->i_frame-1];
406 if( lowres_mv[0][0] != 0x7fff ) *(uint32_t*)mvc[i++] = (*(uint32_t*)lowres_mv[h->mb.i_mb_xy]*2)&0xfffeffff;
409 /* spatial predictors */
410 if( h->mb.i_neighbour & MB_LEFT )
412 int i_mb_l = h->mb.i_mb_xy - 1;
413 /* skip MBs didn't go through the whole search process, so mvr is undefined */
414 if( !IS_SKIP( h->mb.type[i_mb_l] ) )
415 SET_MVP( mvr[i_mb_l] );
417 if( h->mb.i_neighbour & MB_TOP )
419 int i_mb_t = h->mb.i_mb_top_xy;
420 if( !IS_SKIP( h->mb.type[i_mb_t] ) )
421 SET_MVP( mvr[i_mb_t] );
423 if( h->mb.i_neighbour & MB_TOPLEFT && !IS_SKIP( h->mb.type[i_mb_t - 1] ) )
424 SET_MVP( mvr[i_mb_t-1] );
425 if( h->mb.i_mb_x < h->mb.i_mb_stride - 1 && !IS_SKIP( h->mb.type[i_mb_t + 1] ) )
426 SET_MVP( mvr[i_mb_t+1] );
430 /* temporal predictors */
431 /* FIXME temporal scaling w/ interlace */
432 if( h->fref0[0]->i_ref[0] > 0 && !h->sh.b_mbaff )
434 x264_frame_t *l0 = h->fref0[0];
436 #define SET_TMVP(dx, dy) { \
437 int i_b4 = h->mb.i_b4_xy + dx*4 + dy*4*h->mb.i_b4_stride; \
438 int i_b8 = h->mb.i_b8_xy + dx*2 + dy*2*h->mb.i_b8_stride; \
439 int ref_col = l0->ref[0][i_b8]; \
442 int scale = (h->fdec->i_poc - h->fdec->ref_poc[0][i_ref]) * l0->inv_ref_poc[ref_col];\
443 mvc[i][0] = (l0->mv[0][i_b4][0]*scale + 128) >> 8;\
444 mvc[i][1] = (l0->mv[0][i_b4][1]*scale + 128) >> 8;\
450 if( h->mb.i_mb_x < h->sps->i_mb_width-1 )
452 if( h->mb.i_mb_y < h->sps->i_mb_height-1 )
460 /* Set up a lookup table for delta pocs to reduce an IDIV to an IMUL */
461 static void setup_inverse_delta_pocs( x264_t *h )
464 for( i = 0; i < h->i_ref0; i++ )
466 int delta = h->fdec->i_poc - h->fref0[i]->i_poc;
467 h->fdec->inv_ref_poc[i] = (256 + delta/2) / delta;
471 static NOINLINE void x264_mb_mc_0xywh( x264_t *h, int x, int y, int width, int height )
473 const int i8 = x264_scan8[0]+x+8*y;
474 const int i_ref = h->mb.cache.ref[0][i8];
475 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;
476 int mvy = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
478 h->mc.mc_luma( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
479 h->mb.pic.p_fref[0][i_ref], h->mb.pic.i_stride[0],
480 mvx, mvy, 4*width, 4*height );
482 // chroma is offset if MCing from a field of opposite parity
483 if( h->mb.b_interlaced & i_ref )
484 mvy += (h->mb.i_mb_y & 1)*4 - 2;
486 h->mc.mc_chroma( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
487 h->mb.pic.p_fref[0][i_ref][4], h->mb.pic.i_stride[1],
488 mvx, mvy, 2*width, 2*height );
490 h->mc.mc_chroma( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
491 h->mb.pic.p_fref[0][i_ref][5], h->mb.pic.i_stride[2],
492 mvx, mvy, 2*width, 2*height );
494 static NOINLINE void x264_mb_mc_1xywh( x264_t *h, int x, int y, int width, int height )
496 const int i8 = x264_scan8[0]+x+8*y;
497 const int i_ref = h->mb.cache.ref[1][i8];
498 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;
499 int mvy = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
501 h->mc.mc_luma( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
502 h->mb.pic.p_fref[1][i_ref], h->mb.pic.i_stride[0],
503 mvx, mvy, 4*width, 4*height );
505 if( h->mb.b_interlaced & i_ref )
506 mvy += (h->mb.i_mb_y & 1)*4 - 2;
508 h->mc.mc_chroma( &h->mb.pic.p_fdec[1][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
509 h->mb.pic.p_fref[1][i_ref][4], h->mb.pic.i_stride[1],
510 mvx, mvy, 2*width, 2*height );
512 h->mc.mc_chroma( &h->mb.pic.p_fdec[2][2*y*FDEC_STRIDE+2*x], FDEC_STRIDE,
513 h->mb.pic.p_fref[1][i_ref][5], h->mb.pic.i_stride[2],
514 mvx, mvy, 2*width, 2*height );
517 static NOINLINE void x264_mb_mc_01xywh( x264_t *h, int x, int y, int width, int height )
519 const int i8 = x264_scan8[0]+x+8*y;
520 const int i_ref0 = h->mb.cache.ref[0][i8];
521 const int i_ref1 = h->mb.cache.ref[1][i8];
522 const int weight = h->mb.bipred_weight[i_ref0][i_ref1];
523 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;
524 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;
525 int mvy0 = x264_clip3( h->mb.cache.mv[0][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
526 int mvy1 = x264_clip3( h->mb.cache.mv[1][i8][1], h->mb.mv_min[1], h->mb.mv_max[1] ) + 4*4*y;
527 int i_mode = x264_size2pixel[height][width];
528 int i_stride0 = 16, i_stride1 = 16;
529 ALIGNED_ARRAY_16( uint8_t, tmp0,[16*16] );
530 ALIGNED_ARRAY_16( uint8_t, tmp1,[16*16] );
531 uint8_t *src0, *src1;
533 src0 = h->mc.get_ref( tmp0, &i_stride0, h->mb.pic.p_fref[0][i_ref0], h->mb.pic.i_stride[0],
534 mvx0, mvy0, 4*width, 4*height );
535 src1 = h->mc.get_ref( tmp1, &i_stride1, h->mb.pic.p_fref[1][i_ref1], h->mb.pic.i_stride[0],
536 mvx1, mvy1, 4*width, 4*height );
537 h->mc.avg[i_mode]( &h->mb.pic.p_fdec[0][4*y*FDEC_STRIDE+4*x], FDEC_STRIDE,
538 src0, i_stride0, src1, i_stride1, weight );
540 if( h->mb.b_interlaced & i_ref0 )
541 mvy0 += (h->mb.i_mb_y & 1)*4 - 2;
542 if( h->mb.b_interlaced & i_ref1 )
543 mvy1 += (h->mb.i_mb_y & 1)*4 - 2;
545 h->mc.mc_chroma( tmp0, 16, h->mb.pic.p_fref[0][i_ref0][4], h->mb.pic.i_stride[1],
546 mvx0, mvy0, 2*width, 2*height );
547 h->mc.mc_chroma( tmp1, 16, h->mb.pic.p_fref[1][i_ref1][4], h->mb.pic.i_stride[1],
548 mvx1, mvy1, 2*width, 2*height );
549 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 );
550 h->mc.mc_chroma( tmp0, 16, h->mb.pic.p_fref[0][i_ref0][5], h->mb.pic.i_stride[2],
551 mvx0, mvy0, 2*width, 2*height );
552 h->mc.mc_chroma( tmp1, 16, h->mb.pic.p_fref[1][i_ref1][5], h->mb.pic.i_stride[2],
553 mvx1, mvy1, 2*width, 2*height );
554 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 );
557 static void x264_mb_mc_direct8x8( x264_t *h, int x, int y )
559 const int i8 = x264_scan8[0] + x + 8*y;
561 if( h->mb.cache.ref[0][i8] >= 0 )
562 if( h->mb.cache.ref[1][i8] >= 0 )
563 x264_mb_mc_01xywh( h, x, y, 2, 2 );
565 x264_mb_mc_0xywh( h, x, y, 2, 2 );
567 x264_mb_mc_1xywh( h, x, y, 2, 2 );
570 void x264_mb_mc_8x8( x264_t *h, int i8 )
572 const int x = 2*(i8&1);
573 const int y = 2*(i8>>1);
574 switch( h->mb.i_sub_partition[i8] )
577 x264_mb_mc_0xywh( h, x, y, 2, 2 );
580 x264_mb_mc_0xywh( h, x, y+0, 2, 1 );
581 x264_mb_mc_0xywh( h, x, y+1, 2, 1 );
584 x264_mb_mc_0xywh( h, x+0, y, 1, 2 );
585 x264_mb_mc_0xywh( h, x+1, y, 1, 2 );
588 x264_mb_mc_0xywh( h, x+0, y+0, 1, 1 );
589 x264_mb_mc_0xywh( h, x+1, y+0, 1, 1 );
590 x264_mb_mc_0xywh( h, x+0, y+1, 1, 1 );
591 x264_mb_mc_0xywh( h, x+1, y+1, 1, 1 );
594 x264_mb_mc_1xywh( h, x, y, 2, 2 );
597 x264_mb_mc_01xywh( h, x, y, 2, 2 );
600 x264_mb_mc_direct8x8( h, x, y );
605 void x264_mb_mc( x264_t *h )
607 if( h->mb.i_type == P_L0 )
609 if( h->mb.i_partition == D_16x16 )
611 x264_mb_mc_0xywh( h, 0, 0, 4, 4 );
613 else if( h->mb.i_partition == D_16x8 )
615 x264_mb_mc_0xywh( h, 0, 0, 4, 2 );
616 x264_mb_mc_0xywh( h, 0, 2, 4, 2 );
618 else if( h->mb.i_partition == D_8x16 )
620 x264_mb_mc_0xywh( h, 0, 0, 2, 4 );
621 x264_mb_mc_0xywh( h, 2, 0, 2, 4 );
624 else if( h->mb.i_type == P_8x8 || h->mb.i_type == B_8x8 )
627 for( i = 0; i < 4; i++ )
628 x264_mb_mc_8x8( h, i );
630 else if( h->mb.i_type == B_SKIP || h->mb.i_type == B_DIRECT )
632 x264_mb_mc_direct8x8( h, 0, 0 );
633 x264_mb_mc_direct8x8( h, 2, 0 );
634 x264_mb_mc_direct8x8( h, 0, 2 );
635 x264_mb_mc_direct8x8( h, 2, 2 );
639 const uint8_t *b_list0 = x264_mb_type_list_table[h->mb.i_type][0];
640 const uint8_t *b_list1 = x264_mb_type_list_table[h->mb.i_type][1];
642 if( h->mb.i_partition == D_16x16 )
644 if( b_list0[0] && b_list1[0] ) x264_mb_mc_01xywh( h, 0, 0, 4, 4 );
645 else if( b_list0[0] ) x264_mb_mc_0xywh ( h, 0, 0, 4, 4 );
646 else if( b_list1[0] ) x264_mb_mc_1xywh ( h, 0, 0, 4, 4 );
648 else if( h->mb.i_partition == D_16x8 )
650 if( b_list0[0] && b_list1[0] ) x264_mb_mc_01xywh( h, 0, 0, 4, 2 );
651 else if( b_list0[0] ) x264_mb_mc_0xywh ( h, 0, 0, 4, 2 );
652 else if( b_list1[0] ) x264_mb_mc_1xywh ( h, 0, 0, 4, 2 );
654 if( b_list0[1] && b_list1[1] ) x264_mb_mc_01xywh( h, 0, 2, 4, 2 );
655 else if( b_list0[1] ) x264_mb_mc_0xywh ( h, 0, 2, 4, 2 );
656 else if( b_list1[1] ) x264_mb_mc_1xywh ( h, 0, 2, 4, 2 );
658 else if( h->mb.i_partition == D_8x16 )
660 if( b_list0[0] && b_list1[0] ) x264_mb_mc_01xywh( h, 0, 0, 2, 4 );
661 else if( b_list0[0] ) x264_mb_mc_0xywh ( h, 0, 0, 2, 4 );
662 else if( b_list1[0] ) x264_mb_mc_1xywh ( h, 0, 0, 2, 4 );
664 if( b_list0[1] && b_list1[1] ) x264_mb_mc_01xywh( h, 2, 0, 2, 4 );
665 else if( b_list0[1] ) x264_mb_mc_0xywh ( h, 2, 0, 2, 4 );
666 else if( b_list1[1] ) x264_mb_mc_1xywh ( h, 2, 0, 2, 4 );
671 int x264_macroblock_cache_init( x264_t *h )
674 int i_mb_count = h->mb.i_mb_count;
676 h->mb.i_mb_stride = h->sps->i_mb_width;
677 h->mb.i_b8_stride = h->sps->i_mb_width * 2;
678 h->mb.i_b4_stride = h->sps->i_mb_width * 4;
680 h->mb.b_interlaced = h->param.b_interlaced;
682 CHECKED_MALLOC( h->mb.qp, i_mb_count * sizeof(int8_t) );
683 CHECKED_MALLOC( h->mb.cbp, i_mb_count * sizeof(int16_t) );
684 CHECKED_MALLOC( h->mb.skipbp, i_mb_count * sizeof(int8_t) );
685 CHECKED_MALLOC( h->mb.mb_transform_size, i_mb_count * sizeof(int8_t) );
687 /* 0 -> 3 top(4), 4 -> 6 : left(3) */
688 CHECKED_MALLOC( h->mb.intra4x4_pred_mode, i_mb_count * 8 * sizeof(int8_t) );
691 CHECKED_MALLOC( h->mb.non_zero_count, i_mb_count * 24 * sizeof(uint8_t) );
692 CHECKED_MALLOC( h->mb.nnz_backup, h->sps->i_mb_width * 4 * 16 * sizeof(uint8_t) );
694 if( h->param.b_cabac )
696 CHECKED_MALLOC( h->mb.chroma_pred_mode, i_mb_count * sizeof(int8_t) );
697 CHECKED_MALLOC( h->mb.mvd[0], 2*16 * i_mb_count * sizeof(int16_t) );
698 CHECKED_MALLOC( h->mb.mvd[1], 2*16 * i_mb_count * sizeof(int16_t) );
703 int i_refs = X264_MIN(16, (i ? 1 + !!h->param.i_bframe_pyramid : h->param.i_frame_reference) ) << h->param.b_interlaced;
704 for( j=0; j < i_refs; j++ )
705 CHECKED_MALLOC( h->mb.mvr[i][j], 2 * i_mb_count * sizeof(int16_t) );
708 for( i=0; i<=h->param.b_interlaced; i++ )
711 /* shouldn't really be initialized, just silences a valgrind false-positive in predict_8x8_filter_mmx */
712 CHECKED_MALLOCZERO( h->mb.intra_border_backup[i][j], (h->sps->i_mb_width*16+32)>>!!j );
713 h->mb.intra_border_backup[i][j] += 8;
716 /* init with not available (for top right idx=7,15) */
717 memset( h->mb.cache.ref[0], -2, X264_SCAN8_SIZE * sizeof( int8_t ) );
718 memset( h->mb.cache.ref[1], -2, X264_SCAN8_SIZE * sizeof( int8_t ) );
730 h->mb.pic.p_fenc[0] = h->mb.pic.fenc_buf;
731 h->mb.pic.p_fenc[1] = h->mb.pic.fenc_buf + 16*FENC_STRIDE;
732 h->mb.pic.p_fenc[2] = h->mb.pic.fenc_buf + 16*FENC_STRIDE + 8;
733 h->mb.pic.p_fdec[0] = h->mb.pic.fdec_buf + 2*FDEC_STRIDE;
734 h->mb.pic.p_fdec[1] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE;
735 h->mb.pic.p_fdec[2] = h->mb.pic.fdec_buf + 19*FDEC_STRIDE + 16;
737 h->mb.i_neighbour4[6] =
738 h->mb.i_neighbour4[9] =
739 h->mb.i_neighbour4[12] =
740 h->mb.i_neighbour4[14] = MB_LEFT|MB_TOP|MB_TOPLEFT|MB_TOPRIGHT;
741 h->mb.i_neighbour4[3] =
742 h->mb.i_neighbour4[7] =
743 h->mb.i_neighbour4[11] =
744 h->mb.i_neighbour4[13] =
745 h->mb.i_neighbour4[15] =
746 h->mb.i_neighbour8[3] = MB_LEFT|MB_TOP|MB_TOPLEFT;
748 int buf_hpel = (h->param.i_width+48) * sizeof(int16_t);
749 int buf_ssim = h->param.analyse.b_ssim * 8 * (h->param.i_width/4+3) * sizeof(int);
750 int me_range = X264_MIN(h->param.analyse.i_me_range, h->param.analyse.i_mv_range);
751 int buf_tesa = (h->param.analyse.i_me_method >= X264_ME_ESA) *
752 ((me_range*2+18) * sizeof(int16_t) + (me_range+4) * (me_range+1) * 4 * sizeof(mvsad_t));
753 int buf_mbtree = h->param.rc.b_mb_tree * ((h->sps->i_mb_width+3)&~3) * sizeof(int);
754 CHECKED_MALLOC( h->scratch_buffer, X264_MAX4( buf_hpel, buf_ssim, buf_tesa, buf_mbtree ) );
759 void x264_macroblock_cache_end( x264_t *h )
762 for( i=0; i<=h->param.b_interlaced; i++ )
764 x264_free( h->mb.intra_border_backup[i][j] - 8 );
766 for( j=0; j<32; j++ )
767 x264_free( h->mb.mvr[i][j] );
768 if( h->param.b_cabac )
770 x264_free( h->mb.chroma_pred_mode );
771 x264_free( h->mb.mvd[0] );
772 x264_free( h->mb.mvd[1] );
774 x264_free( h->mb.intra4x4_pred_mode );
775 x264_free( h->mb.non_zero_count );
776 x264_free( h->mb.nnz_backup );
777 x264_free( h->mb.mb_transform_size );
778 x264_free( h->mb.skipbp );
779 x264_free( h->mb.cbp );
780 x264_free( h->mb.qp );
781 x264_free( h->scratch_buffer );
783 void x264_macroblock_slice_init( x264_t *h )
787 h->mb.mv[0] = h->fdec->mv[0];
788 h->mb.mv[1] = h->fdec->mv[1];
789 h->mb.ref[0] = h->fdec->ref[0];
790 h->mb.ref[1] = h->fdec->ref[1];
791 h->mb.type = h->fdec->mb_type;
793 h->fdec->i_ref[0] = h->i_ref0;
794 h->fdec->i_ref[1] = h->i_ref1;
795 for( i = 0; i < h->i_ref0; i++ )
796 h->fdec->ref_poc[0][i] = h->fref0[i]->i_poc;
797 if( h->sh.i_type == SLICE_TYPE_B )
799 for( i = 0; i < h->i_ref1; i++ )
800 h->fdec->ref_poc[1][i] = h->fref1[i]->i_poc;
802 map_col_to_list0(-1) = -1;
803 map_col_to_list0(-2) = -2;
804 for( i = 0; i < h->fref1[0]->i_ref[0]; i++ )
806 int poc = h->fref1[0]->ref_poc[0][i];
807 map_col_to_list0(i) = -2;
808 for( j = 0; j < h->i_ref0; j++ )
809 if( h->fref0[j]->i_poc == poc )
811 map_col_to_list0(i) = j;
816 if( h->sh.i_type == SLICE_TYPE_P )
817 memset( h->mb.cache.skip, 0, X264_SCAN8_SIZE * sizeof( int8_t ) );
819 setup_inverse_delta_pocs( h );
822 void x264_prefetch_fenc( x264_t *h, x264_frame_t *fenc, int i_mb_x, int i_mb_y )
824 int stride_y = fenc->i_stride[0];
825 int stride_uv = fenc->i_stride[1];
826 int off_y = 16 * (i_mb_x + i_mb_y * stride_y);
827 int off_uv = 8 * (i_mb_x + i_mb_y * stride_uv);
828 h->mc.prefetch_fenc( fenc->plane[0]+off_y, stride_y,
829 fenc->plane[1+(i_mb_x&1)]+off_uv, stride_uv, i_mb_x );
832 static NOINLINE void copy_column8( uint8_t *dst, uint8_t *src )
834 // input pointers are offset by 4 rows because that's faster (smaller instruction size on x86)
836 for( i = -4; i < 4; i++ )
837 dst[i*FDEC_STRIDE] = src[i*FDEC_STRIDE];
840 static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int i_mb_x, int i_mb_y, int i)
842 const int w = (i == 0 ? 16 : 8);
843 const int i_stride = h->fdec->i_stride[!!i];
844 const int i_stride2 = i_stride << h->mb.b_interlaced;
845 const int i_pix_offset = h->mb.b_interlaced
846 ? w * (i_mb_x + (i_mb_y&~1) * i_stride) + (i_mb_y&1) * i_stride
847 : w * (i_mb_x + i_mb_y * i_stride);
848 int ref_pix_offset[2] = { i_pix_offset, i_pix_offset };
849 const uint8_t *intra_fdec = &h->mb.intra_border_backup[i_mb_y & h->sh.b_mbaff][i][i_mb_x*16>>!!i];
850 x264_frame_t **fref[2] = { h->fref0, h->fref1 };
852 if( h->mb.b_interlaced )
853 ref_pix_offset[1] += (1-2*(i_mb_y&1)) * i_stride;
854 h->mb.pic.i_stride[i] = i_stride2;
855 h->mb.pic.p_fenc_plane[i] = &h->fenc->plane[i][i_pix_offset];
856 h->mc.copy[i?PIXEL_8x8:PIXEL_16x16]( h->mb.pic.p_fenc[i], FENC_STRIDE,
857 h->mb.pic.p_fenc_plane[i], i_stride2, w );
858 memcpy( &h->mb.pic.p_fdec[i][-1-FDEC_STRIDE], intra_fdec-1, w*3/2+1 );
859 if( h->mb.b_interlaced || h->mb.b_reencode_mb )
861 const uint8_t *plane_fdec = &h->fdec->plane[i][i_pix_offset];
862 for( j = 0; j < w; j++ )
863 h->mb.pic.p_fdec[i][-1+j*FDEC_STRIDE] = plane_fdec[-1+j*i_stride2];
865 for( j = 0; j < h->mb.pic.i_fref[0]; j++ )
867 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]];
869 for( k = 1; k < 4; k++ )
870 h->mb.pic.p_fref[0][j][k] = &fref[0][j >> h->mb.b_interlaced]->filtered[k][ref_pix_offset[j&1]];
872 if( h->sh.i_type == SLICE_TYPE_B )
873 for( j = 0; j < h->mb.pic.i_fref[1]; j++ )
875 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]];
877 for( k = 1; k < 4; k++ )
878 h->mb.pic.p_fref[1][j][k] = &fref[1][j >> h->mb.b_interlaced]->filtered[k][ref_pix_offset[j&1]];
882 void x264_macroblock_cache_load( x264_t *h, int i_mb_x, int i_mb_y )
884 int i_mb_xy = i_mb_y * h->mb.i_mb_stride + i_mb_x;
885 int i_mb_4x4 = 4*(i_mb_y * h->mb.i_b4_stride + i_mb_x);
886 int i_mb_8x8 = 2*(i_mb_y * h->mb.i_b8_stride + i_mb_x);
887 int i_top_y = i_mb_y - (1 << h->mb.b_interlaced);
888 int i_top_xy = i_top_y * h->mb.i_mb_stride + i_mb_x;
889 int i_top_4x4 = (4*i_top_y+3) * h->mb.i_b4_stride + 4*i_mb_x;
890 int i_top_8x8 = (2*i_top_y+1) * h->mb.i_b8_stride + 2*i_mb_x;
892 int i_top_type = -1; /* gcc warn */
898 h->mb.i_mb_x = i_mb_x;
899 h->mb.i_mb_y = i_mb_y;
900 h->mb.i_mb_xy = i_mb_xy;
901 h->mb.i_b8_xy = i_mb_8x8;
902 h->mb.i_b4_xy = i_mb_4x4;
903 h->mb.i_mb_top_xy = i_top_xy;
904 h->mb.i_neighbour = 0;
905 h->mb.i_neighbour_intra = 0;
908 if( i_top_xy >= h->sh.i_first_mb )
910 h->mb.i_mb_type_top =
911 i_top_type = h->mb.type[i_top_xy];
912 h->mb.cache.i_cbp_top = h->mb.cbp[i_top_xy];
914 h->mb.i_neighbour |= MB_TOP;
916 if( !h->param.b_constrained_intra || IS_INTRA( i_top_type ) )
917 h->mb.i_neighbour_intra |= MB_TOP;
920 *(uint32_t*)&h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8] = *(uint32_t*)&h->mb.intra4x4_pred_mode[i_top_xy][0];
922 /* load non_zero_count */
923 *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0] - 8] = *(uint32_t*)&h->mb.non_zero_count[i_top_xy][12];
924 /* shift because x264_scan8[16] is misaligned */
925 *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16] - 9] = *(uint16_t*)&h->mb.non_zero_count[i_top_xy][18] << 8;
926 *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+4] - 9] = *(uint16_t*)&h->mb.non_zero_count[i_top_xy][22] << 8;
930 h->mb.i_mb_type_top = -1;
931 h->mb.cache.i_cbp_top = -1;
934 *(uint32_t*)&h->mb.cache.intra4x4_pred_mode[x264_scan8[0] - 8] = 0xFFFFFFFFU;
936 /* load non_zero_count */
937 *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0] - 8] =
938 *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+0] - 9] =
939 *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+4] - 9] = 0x80808080U;
942 if( i_mb_x > 0 && i_mb_xy > h->sh.i_first_mb )
944 i_left_xy = i_mb_xy - 1;
945 h->mb.i_mb_type_left =
946 i_left_type = h->mb.type[i_left_xy];
947 h->mb.cache.i_cbp_left = h->mb.cbp[h->mb.i_mb_xy - 1];
949 h->mb.i_neighbour |= MB_LEFT;
951 if( !h->param.b_constrained_intra || IS_INTRA( i_left_type ) )
952 h->mb.i_neighbour_intra |= MB_LEFT;
955 h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][4];
956 h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][5];
957 h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][6];
958 h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = h->mb.intra4x4_pred_mode[i_left_xy][3];
960 /* load non_zero_count */
961 h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = h->mb.non_zero_count[i_left_xy][3];
962 h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = h->mb.non_zero_count[i_left_xy][7];
963 h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = h->mb.non_zero_count[i_left_xy][11];
964 h->mb.cache.non_zero_count[x264_scan8[10] - 1] = h->mb.non_zero_count[i_left_xy][15];
966 h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] = h->mb.non_zero_count[i_left_xy][16+1];
967 h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] = h->mb.non_zero_count[i_left_xy][16+3];
969 h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] = h->mb.non_zero_count[i_left_xy][16+4+1];
970 h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = h->mb.non_zero_count[i_left_xy][16+4+3];
974 h->mb.i_mb_type_left = -1;
975 h->mb.cache.i_cbp_left = -1;
977 h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] =
978 h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] =
979 h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] =
980 h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = -1;
982 /* load non_zero_count */
983 h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] =
984 h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] =
985 h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] =
986 h->mb.cache.non_zero_count[x264_scan8[10] - 1] =
987 h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] =
988 h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] =
989 h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] =
990 h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = 0x80;
993 if( i_mb_x < h->sps->i_mb_width - 1 && i_top_xy + 1 >= h->sh.i_first_mb )
995 h->mb.i_neighbour |= MB_TOPRIGHT;
996 h->mb.i_mb_type_topright = h->mb.type[ i_top_xy + 1 ];
997 if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_topright ) )
998 h->mb.i_neighbour_intra |= MB_TOPRIGHT;
1001 h->mb.i_mb_type_topright = -1;
1002 if( i_mb_x > 0 && i_top_xy - 1 >= h->sh.i_first_mb )
1004 h->mb.i_neighbour |= MB_TOPLEFT;
1005 h->mb.i_mb_type_topleft = h->mb.type[ i_top_xy - 1 ];
1006 if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_topleft ) )
1007 h->mb.i_neighbour_intra |= MB_TOPLEFT;
1010 h->mb.i_mb_type_topleft = -1;
1012 if( h->pps->b_transform_8x8_mode )
1014 h->mb.cache.i_neighbour_transform_size =
1015 ( i_left_type >= 0 && h->mb.mb_transform_size[i_left_xy] )
1016 + ( i_top_type >= 0 && h->mb.mb_transform_size[i_top_xy] );
1021 h->mb.pic.i_fref[0] = h->i_ref0 << h->mb.b_interlaced;
1022 h->mb.pic.i_fref[1] = h->i_ref1 << h->mb.b_interlaced;
1023 h->mb.cache.i_neighbour_interlaced =
1024 !!(h->mb.i_neighbour & MB_LEFT)
1025 + !!(h->mb.i_neighbour & MB_TOP);
1028 if( !h->mb.b_interlaced && !h->mb.b_reencode_mb )
1030 copy_column8( h->mb.pic.p_fdec[0]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+ 4*FDEC_STRIDE );
1031 copy_column8( h->mb.pic.p_fdec[0]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+12*FDEC_STRIDE );
1032 copy_column8( h->mb.pic.p_fdec[1]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[1]+ 7+ 4*FDEC_STRIDE );
1033 copy_column8( h->mb.pic.p_fdec[2]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[2]+ 7+ 4*FDEC_STRIDE );
1036 /* load picture pointers */
1037 x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 0 );
1038 x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 1 );
1039 x264_macroblock_load_pic_pointers( h, i_mb_x, i_mb_y, 2 );
1041 if( h->fdec->integral )
1043 assert( !h->mb.b_interlaced );
1044 for( i = 0; i < h->mb.pic.i_fref[0]; i++ )
1045 h->mb.pic.p_integral[0][i] = &h->fref0[i]->integral[ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
1046 for( i = 0; i < h->mb.pic.i_fref[1]; i++ )
1047 h->mb.pic.p_integral[1][i] = &h->fref1[i]->integral[ 16 * ( i_mb_x + i_mb_y * h->fdec->i_stride[0] )];
1050 x264_prefetch_fenc( h, h->fenc, i_mb_x, i_mb_y );
1052 /* load ref/mv/mvd */
1053 if( h->sh.i_type != SLICE_TYPE_I )
1055 const int s8x8 = h->mb.i_b8_stride;
1056 const int s4x4 = h->mb.i_b4_stride;
1060 for( i_list = 0; i_list < (h->sh.i_type == SLICE_TYPE_B ? 2 : 1 ); i_list++ )
1063 h->mb.cache.ref[i_list][x264_scan8[5 ]+1] =
1064 h->mb.cache.ref[i_list][x264_scan8[7 ]+1] =
1065 h->mb.cache.ref[i_list][x264_scan8[13]+1] = -2;
1068 if( h->mb.i_neighbour & MB_TOPLEFT )
1070 const int i8 = x264_scan8[0] - 1 - 1*8;
1071 const int ir = i_top_8x8 - 1;
1072 const int iv = i_top_4x4 - 1;
1073 h->mb.cache.ref[i_list][i8] = h->mb.ref[i_list][ir];
1074 *(uint32_t*)h->mb.cache.mv[i_list][i8] = *(uint32_t*)h->mb.mv[i_list][iv];
1078 const int i8 = x264_scan8[0] - 1 - 1*8;
1079 h->mb.cache.ref[i_list][i8] = -2;
1080 *(uint32_t*)h->mb.cache.mv[i_list][i8] = 0;
1083 if( h->mb.i_neighbour & MB_TOP )
1085 const int i8 = x264_scan8[0] - 8;
1086 const int ir = i_top_8x8;
1087 const int iv = i_top_4x4;
1088 h->mb.cache.ref[i_list][i8+0] =
1089 h->mb.cache.ref[i_list][i8+1] = h->mb.ref[i_list][ir + 0];
1090 h->mb.cache.ref[i_list][i8+2] =
1091 h->mb.cache.ref[i_list][i8+3] = h->mb.ref[i_list][ir + 1];
1092 *(uint64_t*)h->mb.cache.mv[i_list][i8+0] = *(uint64_t*)h->mb.mv[i_list][iv+0];
1093 *(uint64_t*)h->mb.cache.mv[i_list][i8+2] = *(uint64_t*)h->mb.mv[i_list][iv+2];
1097 const int i8 = x264_scan8[0] - 8;
1098 *(uint64_t*)h->mb.cache.mv[i_list][i8+0] = 0;
1099 *(uint64_t*)h->mb.cache.mv[i_list][i8+2] = 0;
1100 *(uint32_t*)&h->mb.cache.ref[i_list][i8] = (uint8_t)(-2) * 0x01010101U;
1103 if( h->mb.i_neighbour & MB_TOPRIGHT )
1105 const int i8 = x264_scan8[0] + 4 - 1*8;
1106 const int ir = i_top_8x8 + 2;
1107 const int iv = i_top_4x4 + 4;
1108 h->mb.cache.ref[i_list][i8] = h->mb.ref[i_list][ir];
1109 *(uint32_t*)h->mb.cache.mv[i_list][i8] = *(uint32_t*)h->mb.mv[i_list][iv];
1113 const int i8 = x264_scan8[0] + 4 - 1*8;
1114 h->mb.cache.ref[i_list][i8] = -2;
1115 *(uint32_t*)h->mb.cache.mv[i_list][i8] = 0;
1118 if( h->mb.i_neighbour & MB_LEFT )
1120 const int i8 = x264_scan8[0] - 1;
1121 const int ir = i_mb_8x8 - 1;
1122 const int iv = i_mb_4x4 - 1;
1123 h->mb.cache.ref[i_list][i8+0*8] =
1124 h->mb.cache.ref[i_list][i8+1*8] = h->mb.ref[i_list][ir + 0*s8x8];
1125 h->mb.cache.ref[i_list][i8+2*8] =
1126 h->mb.cache.ref[i_list][i8+3*8] = h->mb.ref[i_list][ir + 1*s8x8];
1128 *(uint32_t*)h->mb.cache.mv[i_list][i8+0*8] = *(uint32_t*)h->mb.mv[i_list][iv + 0*s4x4];
1129 *(uint32_t*)h->mb.cache.mv[i_list][i8+1*8] = *(uint32_t*)h->mb.mv[i_list][iv + 1*s4x4];
1130 *(uint32_t*)h->mb.cache.mv[i_list][i8+2*8] = *(uint32_t*)h->mb.mv[i_list][iv + 2*s4x4];
1131 *(uint32_t*)h->mb.cache.mv[i_list][i8+3*8] = *(uint32_t*)h->mb.mv[i_list][iv + 3*s4x4];
1135 const int i8 = x264_scan8[0] - 1;
1136 for( i = 0; i < 4; i++ )
1138 h->mb.cache.ref[i_list][i8+i*8] = -2;
1139 *(uint32_t*)h->mb.cache.mv[i_list][i8+i*8] = 0;
1143 if( h->param.b_cabac )
1145 if( i_top_type >= 0 )
1147 const int i8 = x264_scan8[0] - 8;
1148 const int iv = i_top_4x4;
1149 *(uint64_t*)h->mb.cache.mvd[i_list][i8+0] = *(uint64_t*)h->mb.mvd[i_list][iv+0];
1150 *(uint64_t*)h->mb.cache.mvd[i_list][i8+2] = *(uint64_t*)h->mb.mvd[i_list][iv+2];
1154 const int i8 = x264_scan8[0] - 8;
1155 *(uint64_t*)h->mb.cache.mvd[i_list][i8+0] =
1156 *(uint64_t*)h->mb.cache.mvd[i_list][i8+2] = 0;
1159 if( i_left_type >= 0 )
1161 const int i8 = x264_scan8[0] - 1;
1162 const int iv = i_mb_4x4 - 1;
1163 *(uint32_t*)h->mb.cache.mvd[i_list][i8+0*8] = *(uint32_t*)h->mb.mvd[i_list][iv + 0*s4x4];
1164 *(uint32_t*)h->mb.cache.mvd[i_list][i8+1*8] = *(uint32_t*)h->mb.mvd[i_list][iv + 1*s4x4];
1165 *(uint32_t*)h->mb.cache.mvd[i_list][i8+2*8] = *(uint32_t*)h->mb.mvd[i_list][iv + 2*s4x4];
1166 *(uint32_t*)h->mb.cache.mvd[i_list][i8+3*8] = *(uint32_t*)h->mb.mvd[i_list][iv + 3*s4x4];
1170 const int i8 = x264_scan8[0] - 1;
1171 for( i = 0; i < 4; i++ )
1172 *(uint32_t*)h->mb.cache.mvd[i_list][i8+i*8] = 0;
1178 if( h->sh.i_type == SLICE_TYPE_B && h->param.b_cabac )
1181 x264_macroblock_cache_skip( h, 0, 0, 4, 4, 0 );
1182 skipbp = i_left_type >= 0 ? h->mb.skipbp[i_left_xy] : 0;
1183 h->mb.cache.skip[x264_scan8[0] - 1] = skipbp & 0x2;
1184 h->mb.cache.skip[x264_scan8[8] - 1] = skipbp & 0x8;
1185 skipbp = i_top_type >= 0 ? h->mb.skipbp[i_top_xy] : 0;
1186 h->mb.cache.skip[x264_scan8[0] - 8] = skipbp & 0x4;
1187 h->mb.cache.skip[x264_scan8[4] - 8] = skipbp & 0x8;
1190 if( h->sh.i_type == SLICE_TYPE_P )
1191 x264_mb_predict_mv_pskip( h, h->mb.cache.pskip_mv );
1194 h->mb.i_neighbour4[0] =
1195 h->mb.i_neighbour8[0] = (h->mb.i_neighbour_intra & (MB_TOP|MB_LEFT|MB_TOPLEFT))
1196 | ((h->mb.i_neighbour_intra & MB_TOP) ? MB_TOPRIGHT : 0);
1197 h->mb.i_neighbour4[4] =
1198 h->mb.i_neighbour4[1] = MB_LEFT | ((h->mb.i_neighbour_intra & MB_TOP) ? (MB_TOP|MB_TOPLEFT|MB_TOPRIGHT) : 0);
1199 h->mb.i_neighbour4[2] =
1200 h->mb.i_neighbour4[8] =
1201 h->mb.i_neighbour4[10] =
1202 h->mb.i_neighbour8[2] = MB_TOP|MB_TOPRIGHT | ((h->mb.i_neighbour_intra & MB_LEFT) ? (MB_LEFT|MB_TOPLEFT) : 0);
1203 h->mb.i_neighbour4[5] =
1204 h->mb.i_neighbour8[1] = MB_LEFT | (h->mb.i_neighbour_intra & MB_TOPRIGHT)
1205 | ((h->mb.i_neighbour_intra & MB_TOP) ? MB_TOP|MB_TOPLEFT : 0);
1208 static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int i )
1211 int i_stride = h->fdec->i_stride[!!i];
1212 int i_stride2 = i_stride << h->mb.b_interlaced;
1213 int i_pix_offset = h->mb.b_interlaced
1214 ? w * (h->mb.i_mb_x + (h->mb.i_mb_y&~1) * i_stride) + (h->mb.i_mb_y&1) * i_stride
1215 : w * (h->mb.i_mb_x + h->mb.i_mb_y * i_stride);
1216 h->mc.copy[i?PIXEL_8x8:PIXEL_16x16](
1217 &h->fdec->plane[i][i_pix_offset], i_stride2,
1218 h->mb.pic.p_fdec[i], FDEC_STRIDE, w );
1221 void x264_macroblock_cache_save( x264_t *h )
1223 const int i_mb_xy = h->mb.i_mb_xy;
1224 const int i_mb_type = x264_mb_type_fix[h->mb.i_type];
1225 const int s8x8 = h->mb.i_b8_stride;
1226 const int s4x4 = h->mb.i_b4_stride;
1227 const int i_mb_4x4 = h->mb.i_b4_xy;
1228 const int i_mb_8x8 = h->mb.i_b8_xy;
1230 /* GCC pessimizes direct stores to heap-allocated 8-bit arrays due to aliasing.*/
1231 /* By only dereferencing them once, we avoid this issue. */
1232 int8_t *intra4x4_pred_mode = h->mb.intra4x4_pred_mode[i_mb_xy];
1233 uint8_t *non_zero_count = h->mb.non_zero_count[i_mb_xy];
1237 x264_macroblock_store_pic( h, 0 );
1238 x264_macroblock_store_pic( h, 1 );
1239 x264_macroblock_store_pic( h, 2 );
1241 x264_prefetch_fenc( h, h->fdec, h->mb.i_mb_x, h->mb.i_mb_y );
1243 h->mb.type[i_mb_xy] = i_mb_type;
1244 h->mb.i_mb_prev_xy = i_mb_xy;
1247 if( i_mb_type == I_4x4 )
1249 *(uint32_t*)&intra4x4_pred_mode[0] = *(uint32_t*)&h->mb.cache.intra4x4_pred_mode[x264_scan8[10] ];
1250 *(uint32_t*)&intra4x4_pred_mode[4] = pack8to32(h->mb.cache.intra4x4_pred_mode[x264_scan8[5] ],
1251 h->mb.cache.intra4x4_pred_mode[x264_scan8[7] ],
1252 h->mb.cache.intra4x4_pred_mode[x264_scan8[13] ], 0);
1254 else if( !h->param.b_constrained_intra || IS_INTRA(i_mb_type) )
1255 *(uint64_t*)intra4x4_pred_mode = I_PRED_4x4_DC * 0x0101010101010101ULL;
1257 *(uint64_t*)intra4x4_pred_mode = (uint8_t)(-1) * 0x0101010101010101ULL;
1260 if( i_mb_type == I_PCM )
1262 h->mb.qp[i_mb_xy] = 0;
1263 h->mb.i_last_dqp = 0;
1264 h->mb.i_cbp_chroma = 2;
1265 h->mb.i_cbp_luma = 0xf;
1266 h->mb.cbp[i_mb_xy] = 0x72f; /* all set */
1267 h->mb.b_transform_8x8 = 0;
1268 memset( non_zero_count, 16, 24 );
1272 /* save non zero count */
1273 *(uint32_t*)&non_zero_count[0*4] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0]+0*8];
1274 *(uint32_t*)&non_zero_count[1*4] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0]+1*8];
1275 *(uint32_t*)&non_zero_count[2*4] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0]+2*8];
1276 *(uint32_t*)&non_zero_count[3*4] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[0]+3*8];
1277 *(uint16_t*)&non_zero_count[16+0*2] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+0*2]-1] >> 8;
1278 *(uint16_t*)&non_zero_count[16+1*2] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+1*2]-1] >> 8;
1279 *(uint16_t*)&non_zero_count[16+2*2] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+2*2]-1] >> 8;
1280 *(uint16_t*)&non_zero_count[16+3*2] = *(uint32_t*)&h->mb.cache.non_zero_count[x264_scan8[16+3*2]-1] >> 8;
1282 if( h->mb.i_type != I_16x16 && h->mb.i_cbp_luma == 0 && h->mb.i_cbp_chroma == 0 )
1283 h->mb.i_qp = h->mb.i_last_qp;
1284 h->mb.qp[i_mb_xy] = h->mb.i_qp;
1285 h->mb.i_last_dqp = h->mb.i_qp - h->mb.i_last_qp;
1286 h->mb.i_last_qp = h->mb.i_qp;
1289 if( h->mb.i_cbp_luma == 0 && h->mb.i_type != I_8x8 )
1290 h->mb.b_transform_8x8 = 0;
1291 h->mb.mb_transform_size[i_mb_xy] = h->mb.b_transform_8x8;
1293 if( h->sh.i_type != SLICE_TYPE_I )
1295 if( !IS_INTRA( i_mb_type ) )
1297 h->mb.ref[0][i_mb_8x8+0+0*s8x8] = h->mb.cache.ref[0][x264_scan8[0]];
1298 h->mb.ref[0][i_mb_8x8+1+0*s8x8] = h->mb.cache.ref[0][x264_scan8[4]];
1299 h->mb.ref[0][i_mb_8x8+0+1*s8x8] = h->mb.cache.ref[0][x264_scan8[8]];
1300 h->mb.ref[0][i_mb_8x8+1+1*s8x8] = h->mb.cache.ref[0][x264_scan8[12]];
1301 for( y = 0; y < 4; y++ )
1303 *(uint64_t*)h->mb.mv[0][i_mb_4x4+y*s4x4+0] = *(uint64_t*)h->mb.cache.mv[0][x264_scan8[0]+8*y+0];
1304 *(uint64_t*)h->mb.mv[0][i_mb_4x4+y*s4x4+2] = *(uint64_t*)h->mb.cache.mv[0][x264_scan8[0]+8*y+2];
1306 if( h->sh.i_type == SLICE_TYPE_B )
1308 h->mb.ref[1][i_mb_8x8+0+0*s8x8] = h->mb.cache.ref[1][x264_scan8[0]];
1309 h->mb.ref[1][i_mb_8x8+1+0*s8x8] = h->mb.cache.ref[1][x264_scan8[4]];
1310 h->mb.ref[1][i_mb_8x8+0+1*s8x8] = h->mb.cache.ref[1][x264_scan8[8]];
1311 h->mb.ref[1][i_mb_8x8+1+1*s8x8] = h->mb.cache.ref[1][x264_scan8[12]];
1312 for( y = 0; y < 4; y++ )
1314 *(uint64_t*)h->mb.mv[1][i_mb_4x4+y*s4x4+0] = *(uint64_t*)h->mb.cache.mv[1][x264_scan8[0]+8*y+0];
1315 *(uint64_t*)h->mb.mv[1][i_mb_4x4+y*s4x4+2] = *(uint64_t*)h->mb.cache.mv[1][x264_scan8[0]+8*y+2];
1322 for( i_list = 0; i_list < (h->sh.i_type == SLICE_TYPE_B ? 2 : 1 ); i_list++ )
1324 *(uint16_t*)&h->mb.ref[i_list][i_mb_8x8+0*s8x8] = (uint8_t)(-1) * 0x0101;
1325 *(uint16_t*)&h->mb.ref[i_list][i_mb_8x8+1*s8x8] = (uint8_t)(-1) * 0x0101;
1326 for( y = 0; y < 4; y++ )
1328 *(uint64_t*)h->mb.mv[i_list][i_mb_4x4+y*s4x4+0] = 0;
1329 *(uint64_t*)h->mb.mv[i_list][i_mb_4x4+y*s4x4+2] = 0;
1335 if( h->param.b_cabac )
1337 if( IS_INTRA(i_mb_type) && i_mb_type != I_PCM )
1338 h->mb.chroma_pred_mode[i_mb_xy] = x264_mb_pred_mode8x8c_fix[ h->mb.i_chroma_pred_mode ];
1340 h->mb.chroma_pred_mode[i_mb_xy] = I_PRED_CHROMA_DC;
1342 if( !IS_INTRA( i_mb_type ) && !IS_SKIP( i_mb_type ) && !IS_DIRECT( i_mb_type ) )
1344 for( y = 0; y < 4; y++ )
1346 *(uint64_t*)h->mb.mvd[0][i_mb_4x4+y*s4x4+0] = *(uint64_t*)h->mb.cache.mvd[0][x264_scan8[0]+8*y+0];
1347 *(uint64_t*)h->mb.mvd[0][i_mb_4x4+y*s4x4+2] = *(uint64_t*)h->mb.cache.mvd[0][x264_scan8[0]+8*y+2];
1349 if( h->sh.i_type == SLICE_TYPE_B )
1350 for( y = 0; y < 4; y++ )
1352 *(uint64_t*)h->mb.mvd[1][i_mb_4x4+y*s4x4+0] = *(uint64_t*)h->mb.cache.mvd[1][x264_scan8[0]+8*y+0];
1353 *(uint64_t*)h->mb.mvd[1][i_mb_4x4+y*s4x4+2] = *(uint64_t*)h->mb.cache.mvd[1][x264_scan8[0]+8*y+2];
1358 for( y = 0; y < 4; y++ )
1360 *(uint64_t*)h->mb.mvd[0][i_mb_4x4+y*s4x4+0] = 0;
1361 *(uint64_t*)h->mb.mvd[0][i_mb_4x4+y*s4x4+2] = 0;
1363 if( h->sh.i_type == SLICE_TYPE_B )
1364 for( y = 0; y < 4; y++ )
1366 *(uint64_t*)h->mb.mvd[1][i_mb_4x4+y*s4x4+0] = 0;
1367 *(uint64_t*)h->mb.mvd[1][i_mb_4x4+y*s4x4+2] = 0;
1371 if( h->sh.i_type == SLICE_TYPE_B )
1373 if( i_mb_type == B_SKIP || i_mb_type == B_DIRECT )
1374 h->mb.skipbp[i_mb_xy] = 0xf;
1375 else if( i_mb_type == B_8x8 )
1377 int skipbp = ( h->mb.i_sub_partition[0] == D_DIRECT_8x8 ) << 0;
1378 skipbp |= ( h->mb.i_sub_partition[1] == D_DIRECT_8x8 ) << 1;
1379 skipbp |= ( h->mb.i_sub_partition[2] == D_DIRECT_8x8 ) << 2;
1380 skipbp |= ( h->mb.i_sub_partition[3] == D_DIRECT_8x8 ) << 3;
1381 h->mb.skipbp[i_mb_xy] = skipbp;
1384 h->mb.skipbp[i_mb_xy] = 0;
1389 void x264_macroblock_bipred_init( x264_t *h )
1392 for( i_ref0 = 0; i_ref0 < h->i_ref0; i_ref0++ )
1394 int poc0 = h->fref0[i_ref0]->i_poc;
1395 for( i_ref1 = 0; i_ref1 < h->i_ref1; i_ref1++ )
1397 int dist_scale_factor;
1398 int poc1 = h->fref1[i_ref1]->i_poc;
1399 int td = x264_clip3( poc1 - poc0, -128, 127 );
1400 if( td == 0 /* || pic0 is a long-term ref */ )
1401 dist_scale_factor = 256;
1404 int tb = x264_clip3( h->fdec->i_poc - poc0, -128, 127 );
1405 int tx = (16384 + (abs(td) >> 1)) / td;
1406 dist_scale_factor = x264_clip3( (tb * tx + 32) >> 6, -1024, 1023 );
1408 h->mb.dist_scale_factor[i_ref0][i_ref1] = dist_scale_factor;
1410 dist_scale_factor >>= 2;
1411 if( h->param.analyse.b_weighted_bipred
1412 && dist_scale_factor >= -64
1413 && dist_scale_factor <= 128 )
1415 h->mb.bipred_weight[i_ref0][i_ref1] = 64 - dist_scale_factor;
1416 // ssse3 implementation of biweight doesn't support the extrema.
1417 // if we ever generate them, we'll have to drop that optimization.
1418 assert( dist_scale_factor >= -63 && dist_scale_factor <= 127 );
1421 h->mb.bipred_weight[i_ref0][i_ref1] = 32;
1426 for( i_ref0 = 2*h->i_ref0-1; i_ref0 >= 0; i_ref0-- )
1427 for( i_ref1 = 2*h->i_ref1-1; i_ref1 >= 0; i_ref1-- )
1428 h->mb.bipred_weight[i_ref0][i_ref1] = h->mb.bipred_weight[i_ref0>>1][i_ref1>>1];