1 /*****************************************************************************
2 * macroblock.c: h264 encoder library
3 *****************************************************************************
4 * Copyright (C) 2003 Laurent Aimar
5 * $Id: macroblock.c,v 1.1 2004/06/03 19:27:08 fenrir Exp $
7 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
22 *****************************************************************************/
24 #include "common/common.h"
25 #include "macroblock.h"
28 #define ZIG(i,y,x) level[i] = dct[x][y];
29 static inline void zigzag_scan_2x2_dc( int16_t level[4], int16_t dct[2][2] )
39 * x264_mb_decimate_score: given dct coeffs it returns a score to see if we could empty this dct coeffs
40 * to 0 (low score means set it to null)
41 * Used in inter macroblock (luma and chroma)
42 * luma: for a 8x8 block: if score < 4 -> null
43 * for the complete mb: if score < 6 -> null
44 * chroma: for the complete mb: if score < 7 -> null
46 static int x264_mb_decimate_score( int16_t *dct, int i_max )
48 static const int i_ds_table4[16] = {
49 3,2,2,1,1,1,0,0,0,0,0,0,0,0,0,0 };
50 static const int i_ds_table8[64] = {
51 3,3,3,3,2,2,2,2,2,2,2,2,1,1,1,1,
52 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,
53 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
54 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 };
56 const int *ds_table = (i_max == 64) ? i_ds_table8 : i_ds_table4;
60 while( idx >= 0 && dct[idx] == 0 )
67 if( abs( dct[idx--] ) > 1 )
71 while( idx >= 0 && dct[idx] == 0 )
76 i_score += ds_table[i_run];
82 void x264_mb_encode_i4x4( x264_t *h, int idx, int i_qscale )
84 int x = 4 * block_idx_x[idx];
85 int y = 4 * block_idx_y[idx];
86 uint8_t *p_src = &h->mb.pic.p_fenc[0][x+y*FENC_STRIDE];
87 uint8_t *p_dst = &h->mb.pic.p_fdec[0][x+y*FDEC_STRIDE];
88 DECLARE_ALIGNED_16( int16_t dct4x4[4][4] );
90 if( h->mb.b_lossless )
92 h->zigzagf.sub_4x4( h->dct.luma4x4[idx], p_src, p_dst );
96 h->dctf.sub4x4_dct( dct4x4, p_src, p_dst );
99 x264_quant_4x4_trellis( h, dct4x4, CQM_4IY, i_qscale, DCT_LUMA_4x4, 1 );
101 h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
103 h->zigzagf.scan_4x4( h->dct.luma4x4[idx], dct4x4 );
104 h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4IY], i_qscale );
106 /* output samples to fdec */
107 h->dctf.add4x4_idct( p_dst, dct4x4 );
110 void x264_mb_encode_i8x8( x264_t *h, int idx, int i_qscale )
113 int y = 8 * (idx>>1);
114 uint8_t *p_src = &h->mb.pic.p_fenc[0][x+y*FENC_STRIDE];
115 uint8_t *p_dst = &h->mb.pic.p_fdec[0][x+y*FDEC_STRIDE];
116 DECLARE_ALIGNED_16( int16_t dct8x8[8][8] );
118 h->dctf.sub8x8_dct8( dct8x8, p_src, p_dst );
120 if( h->mb.b_trellis )
121 x264_quant_8x8_trellis( h, dct8x8, CQM_8IY, i_qscale, 1 );
123 h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8IY][i_qscale], h->quant8_bias[CQM_8IY][i_qscale] );
125 h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8 );
126 h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8IY], i_qscale );
127 h->dctf.add8x8_idct8( p_dst, dct8x8 );
130 static void x264_mb_encode_i16x16( x264_t *h, int i_qscale )
132 uint8_t *p_src = h->mb.pic.p_fenc[0];
133 uint8_t *p_dst = h->mb.pic.p_fdec[0];
135 DECLARE_ALIGNED_16( int16_t dct4x4[16+1][4][4] );
139 if( h->mb.b_lossless )
141 for( i = 0; i < 16; i++ )
143 int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
144 int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
145 h->zigzagf.sub_4x4( h->dct.luma4x4[i], p_src+oe, p_dst+od );
146 dct4x4[0][block_idx_x[i]][block_idx_y[i]] = h->dct.luma4x4[i][0];
148 h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct4x4[0] );
152 h->dctf.sub16x16_dct( &dct4x4[1], p_src, p_dst );
153 for( i = 0; i < 16; i++ )
156 dct4x4[0][block_idx_y[i]][block_idx_x[i]] = dct4x4[1+i][0][0];
158 /* quant/scan/dequant */
159 if( h->mb.b_trellis )
160 x264_quant_4x4_trellis( h, dct4x4[1+i], CQM_4IY, i_qscale, DCT_LUMA_AC, 1 );
162 h->quantf.quant_4x4( dct4x4[1+i], h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
164 h->zigzagf.scan_4x4( h->dct.luma4x4[i], dct4x4[1+i] );
165 h->quantf.dequant_4x4( dct4x4[1+i], h->dequant4_mf[CQM_4IY], i_qscale );
168 h->dctf.dct4x4dc( dct4x4[0] );
169 h->quantf.quant_4x4_dc( dct4x4[0], h->quant4_mf[CQM_4IY][i_qscale][0]>>1, h->quant4_bias[CQM_4IY][i_qscale][0]<<1 );
170 h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct4x4[0] );
172 /* output samples to fdec */
173 h->dctf.idct4x4dc( dct4x4[0] );
174 x264_mb_dequant_4x4_dc( dct4x4[0], h->dequant4_mf[CQM_4IY], i_qscale ); /* XXX not inversed */
176 /* calculate dct coeffs */
177 for( i = 0; i < 16; i++ )
180 dct4x4[1+i][0][0] = dct4x4[0][block_idx_y[i]][block_idx_x[i]];
182 /* put pixels to fdec */
183 h->dctf.add16x16_idct( p_dst, &dct4x4[1] );
186 void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qscale )
189 int b_decimate = b_inter && (h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate);
191 for( ch = 0; ch < 2; ch++ )
193 uint8_t *p_src = h->mb.pic.p_fenc[1+ch];
194 uint8_t *p_dst = h->mb.pic.p_fdec[1+ch];
195 int i_decimate_score = 0;
197 DECLARE_ALIGNED_16( int16_t dct2x2[2][2] );
198 DECLARE_ALIGNED_16( int16_t dct4x4[4][4][4] );
200 if( h->mb.b_lossless )
202 for( i = 0; i < 4; i++ )
204 int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
205 int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
206 h->zigzagf.sub_4x4( h->dct.luma4x4[16+i+ch*4], p_src+oe, p_dst+od );
207 h->dct.chroma_dc[ch][i] = h->dct.luma4x4[16+i+ch*4][0];
212 h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
213 /* calculate dct coeffs */
214 for( i = 0; i < 4; i++ )
217 dct2x2[block_idx_y[i]][block_idx_x[i]] = dct4x4[i][0][0];
219 /* no trellis; it doesn't seem to help chroma noticeably */
220 h->quantf.quant_4x4( dct4x4[i], h->quant4_mf[CQM_4IC+b_inter][i_qscale], h->quant4_bias[CQM_4IC+b_inter][i_qscale] );
221 h->zigzagf.scan_4x4( h->dct.luma4x4[16+i+ch*4], dct4x4[i] );
225 i_decimate_score += x264_mb_decimate_score( h->dct.luma4x4[16+i+ch*4]+1, 15 );
229 h->dctf.dct2x2dc( dct2x2 );
230 h->quantf.quant_2x2_dc( dct2x2, h->quant4_mf[CQM_4IC+b_inter][i_qscale][0]>>1, h->quant4_bias[CQM_4IC+b_inter][i_qscale][0]<<1 );
231 zigzag_scan_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );
233 /* output samples to fdec */
234 h->dctf.idct2x2dc( dct2x2 );
235 x264_mb_dequant_2x2_dc( dct2x2, h->dequant4_mf[CQM_4IC + b_inter], i_qscale ); /* XXX not inversed */
237 if( b_decimate && i_decimate_score < 7 )
239 /* Near null chroma 8x8 block so make it null (bits saving) */
240 memset( &h->dct.luma4x4[16+ch*4], 0, 4 * sizeof( *h->dct.luma4x4 ) );
241 if( !array_non_zero( dct2x2 ) )
243 memset( dct4x4, 0, sizeof( dct4x4 ) );
247 for( i = 0; i < 4; i++ )
248 h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IC + b_inter], i_qscale );
251 for( i = 0; i < 4; i++ )
252 dct4x4[i][0][0] = dct2x2[0][i];
253 h->dctf.add8x8_idct( p_dst, dct4x4 );
256 /* coded block pattern */
257 h->mb.i_cbp_chroma = 0;
258 for( i = 0; i < 8; i++ )
260 int nz = array_non_zero_count( h->dct.luma4x4[16+i]+1, 15 );
261 h->mb.cache.non_zero_count[x264_scan8[16+i]] = nz;
262 h->mb.i_cbp_chroma |= nz;
264 if( h->mb.i_cbp_chroma )
265 h->mb.i_cbp_chroma = 2; /* dc+ac (we can't do only ac) */
266 else if( array_non_zero( h->dct.chroma_dc ) )
267 h->mb.i_cbp_chroma = 1; /* dc only */
270 static void x264_macroblock_encode_skip( x264_t *h )
273 h->mb.i_cbp_luma = 0x00;
274 h->mb.i_cbp_chroma = 0x00;
276 for( i = 0; i < 16+8; i++ )
278 h->mb.cache.non_zero_count[x264_scan8[i]] = 0;
282 h->mb.cbp[h->mb.i_mb_xy] = 0;
285 /*****************************************************************************
286 * x264_macroblock_encode_pskip:
287 * Encode an already marked skip block
288 *****************************************************************************/
289 void x264_macroblock_encode_pskip( x264_t *h )
291 const int mvx = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][0],
292 h->mb.mv_min[0], h->mb.mv_max[0] );
293 const int mvy = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][1],
294 h->mb.mv_min[1], h->mb.mv_max[1] );
296 /* Motion compensation XXX probably unneeded */
297 h->mc.mc_luma( h->mb.pic.p_fdec[0], FDEC_STRIDE,
298 h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
302 h->mc.mc_chroma( h->mb.pic.p_fdec[1], FDEC_STRIDE,
303 h->mb.pic.p_fref[0][0][4], h->mb.pic.i_stride[1],
306 h->mc.mc_chroma( h->mb.pic.p_fdec[2], FDEC_STRIDE,
307 h->mb.pic.p_fref[0][0][5], h->mb.pic.i_stride[2],
310 x264_macroblock_encode_skip( h );
313 /*****************************************************************************
314 * x264_macroblock_encode:
315 *****************************************************************************/
316 void x264_macroblock_encode( x264_t *h )
319 int i_qp = h->mb.i_qp;
320 int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
321 int b_force_no_skip = 0;
325 && h->mb.i_mb_xy == h->sh.i_first_mb + h->mb.i_mb_stride
326 && IS_SKIP(h->mb.type[h->sh.i_first_mb]) )
328 /* The first skip is predicted to be a frame mb pair.
329 * We don't yet support the aff part of mbaff, so force it to non-skip
330 * so that we can pick the aff flag. */
332 if( IS_SKIP(h->mb.i_type) )
334 if( h->mb.i_type == P_SKIP )
336 else if( h->mb.i_type == B_SKIP )
337 h->mb.i_type = B_DIRECT;
341 if( h->mb.i_type == P_SKIP )
344 x264_macroblock_encode_pskip( h );
347 if( h->mb.i_type == B_SKIP )
349 /* XXX motion compensation is probably unneeded */
351 x264_macroblock_encode_skip( h );
355 if( h->mb.i_type == I_16x16 )
357 const int i_mode = h->mb.i_intra16x16_pred_mode;
358 h->mb.b_transform_8x8 = 0;
359 /* do the right prediction */
360 h->predict_16x16[i_mode]( h->mb.pic.p_fdec[0] );
362 /* encode the 16x16 macroblock */
363 x264_mb_encode_i16x16( h, i_qp );
365 else if( h->mb.i_type == I_8x8 )
367 DECLARE_ALIGNED_16( uint8_t edge[33] );
368 h->mb.b_transform_8x8 = 1;
369 /* If we already encoded 3 of the 4 i8x8 blocks, we don't have to do them again. */
370 if( h->mb.i_skip_intra )
372 h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fdec[0], FDEC_STRIDE, h->mb.pic.i8x8_fdec_buf, 16, 16 );
373 /* In RD mode, restore the now-overwritten DCT data. */
374 if( h->mb.i_skip_intra == 2 )
375 h->mc.memcpy_aligned( h->dct.luma8x8, h->mb.pic.i8x8_dct_buf, sizeof(h->mb.pic.i8x8_dct_buf) );
377 for( i = h->mb.i_skip_intra ? 3 : 0 ; i < 4; i++ )
379 uint8_t *p_dst = &h->mb.pic.p_fdec[0][8 * (i&1) + 8 * (i>>1) * FDEC_STRIDE];
380 int i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[4*i]];
382 x264_predict_8x8_filter( p_dst, edge, h->mb.i_neighbour8[i], x264_pred_i4x4_neighbors[i_mode] );
383 h->predict_8x8[i_mode]( p_dst, edge );
384 x264_mb_encode_i8x8( h, i, i_qp );
387 else if( h->mb.i_type == I_4x4 )
389 h->mb.b_transform_8x8 = 0;
390 /* If we already encoded 15 of the 16 i4x4 blocks, we don't have to do them again. */
391 if( h->mb.i_skip_intra )
393 h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fdec[0], FDEC_STRIDE, h->mb.pic.i4x4_fdec_buf, 16, 16 );
394 /* In RD mode, restore the now-overwritten DCT data. */
395 if( h->mb.i_skip_intra == 2 )
396 h->mc.memcpy_aligned( h->dct.luma4x4, h->mb.pic.i4x4_dct_buf, sizeof(h->mb.pic.i4x4_dct_buf) );
398 for( i = h->mb.i_skip_intra ? 15 : 0 ; i < 16; i++ )
400 uint8_t *p_dst = &h->mb.pic.p_fdec[0][4 * block_idx_x[i] + 4 * block_idx_y[i] * FDEC_STRIDE];
401 int i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[i]];
403 if( (h->mb.i_neighbour4[i] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
404 /* emulate missing topright samples */
405 *(uint32_t*) &p_dst[4-FDEC_STRIDE] = p_dst[3-FDEC_STRIDE] * 0x01010101U;
407 h->predict_4x4[i_mode]( p_dst );
408 x264_mb_encode_i4x4( h, i, i_qp );
414 int i_decimate_mb = 0;
416 /* Motion compensation */
419 if( h->mb.b_lossless )
421 for( i4x4 = 0; i4x4 < 16; i4x4++ )
423 int x = 4*block_idx_x[i4x4];
424 int y = 4*block_idx_y[i4x4];
425 h->zigzagf.sub_4x4( h->dct.luma4x4[i4x4],
426 h->mb.pic.p_fenc[0]+x+y*FENC_STRIDE,
427 h->mb.pic.p_fdec[0]+x+y*FDEC_STRIDE );
430 else if( h->mb.b_transform_8x8 )
432 DECLARE_ALIGNED_16( int16_t dct8x8[4][8][8] );
433 int nnz8x8[4] = {1,1,1,1};
434 b_decimate &= !h->mb.b_trellis; // 8x8 trellis is inherently optimal decimation
435 h->dctf.sub16x16_dct8( dct8x8, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
437 for( idx = 0; idx < 4; idx++ )
439 if( h->mb.b_noise_reduction )
440 x264_denoise_dct( h, (int16_t*)dct8x8[idx] );
441 if( h->mb.b_trellis )
442 x264_quant_8x8_trellis( h, dct8x8[idx], CQM_8PY, i_qp, 0 );
444 h->quantf.quant_8x8( dct8x8[idx], h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
446 h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8[idx] );
450 int i_decimate_8x8 = x264_mb_decimate_score( h->dct.luma8x8[idx], 64 );
451 i_decimate_mb += i_decimate_8x8;
452 if( i_decimate_8x8 < 4 )
454 memset( h->dct.luma8x8[idx], 0, sizeof( h->dct.luma8x8[idx] ) );
455 memset( dct8x8[idx], 0, sizeof( dct8x8[idx] ) );
460 nnz8x8[idx] = array_non_zero( dct8x8[idx] );
463 if( i_decimate_mb < 6 && b_decimate )
464 memset( h->dct.luma8x8, 0, sizeof( h->dct.luma8x8 ) );
467 for( idx = 0; idx < 4; idx++ )
470 h->quantf.dequant_8x8( dct8x8[idx], h->dequant8_mf[CQM_8PY], i_qp );
471 h->dctf.add8x8_idct8( &h->mb.pic.p_fdec[0][(idx&1)*8 + (idx>>1)*8*FDEC_STRIDE], dct8x8[idx] );
477 DECLARE_ALIGNED_16( int16_t dct4x4[16][4][4] );
478 int nnz8x8[4] = {1,1,1,1};
479 h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
481 for( i8x8 = 0; i8x8 < 4; i8x8++ )
485 /* encode one 4x4 block */
487 for( i4x4 = 0; i4x4 < 4; i4x4++ )
489 idx = i8x8 * 4 + i4x4;
491 if( h->mb.b_noise_reduction )
492 x264_denoise_dct( h, (int16_t*)dct4x4[idx] );
493 if( h->mb.b_trellis )
494 x264_quant_4x4_trellis( h, dct4x4[idx], CQM_4PY, i_qp, DCT_LUMA_4x4, 0 );
496 h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
498 h->zigzagf.scan_4x4( h->dct.luma4x4[idx], dct4x4[idx] );
501 i_decimate_8x8 += x264_mb_decimate_score( h->dct.luma4x4[idx], 16 );
504 /* decimate this 8x8 block */
505 i_decimate_mb += i_decimate_8x8;
506 if( i_decimate_8x8 < 4 && b_decimate )
508 memset( &dct4x4[i8x8*4], 0, 4 * sizeof( *dct4x4 ) );
509 memset( &h->dct.luma4x4[i8x8*4], 0, 4 * sizeof( *h->dct.luma4x4 ) );
514 if( i_decimate_mb < 6 && b_decimate )
515 memset( h->dct.luma4x4, 0, 16 * sizeof( *h->dct.luma4x4 ) );
518 for( i8x8 = 0; i8x8 < 4; i8x8++ )
521 for( i = 0; i < 4; i++ )
522 h->quantf.dequant_4x4( dct4x4[i8x8*4+i], h->dequant4_mf[CQM_4PY], i_qp );
523 h->dctf.add8x8_idct( &h->mb.pic.p_fdec[0][(i8x8&1)*8 + (i8x8>>1)*8*FDEC_STRIDE], &dct4x4[i8x8*4] );
530 if( IS_INTRA( h->mb.i_type ) )
532 const int i_mode = h->mb.i_chroma_pred_mode;
533 h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[1] );
534 h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[2] );
537 /* encode the 8x8 blocks */
538 x264_mb_encode_8x8_chroma( h, !IS_INTRA( h->mb.i_type ), h->mb.i_chroma_qp );
540 /* coded block pattern and non_zero_count */
541 h->mb.i_cbp_luma = 0x00;
542 if( h->mb.i_type == I_16x16 )
544 for( i = 0; i < 16; i++ )
546 const int nz = array_non_zero_count( h->dct.luma4x4[i]+1, 15 );
547 h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
549 h->mb.i_cbp_luma = 0x0f;
552 else if( h->mb.b_transform_8x8 )
554 /* coded_block_flag is enough for CABAC.
555 * the full non_zero_count is done only in CAVLC. */
556 for( i = 0; i < 4; i++ )
558 const int nz = array_non_zero( h->dct.luma8x8[i] );
560 for( j = 0; j < 4; j++ )
561 h->mb.cache.non_zero_count[x264_scan8[4*i+j]] = nz;
563 h->mb.i_cbp_luma |= 1 << i;
568 for( i = 0; i < 16; i++ )
570 const int nz = array_non_zero_count( h->dct.luma4x4[i], 16 );
571 h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
573 h->mb.i_cbp_luma |= 1 << (i/4);
577 if( h->param.b_cabac )
579 i_cbp_dc = ( h->mb.i_type == I_16x16 && array_non_zero( h->dct.luma16x16_dc ) )
580 | array_non_zero( h->dct.chroma_dc[0] ) << 1
581 | array_non_zero( h->dct.chroma_dc[1] ) << 2;
585 h->mb.cbp[h->mb.i_mb_xy] = (i_cbp_dc << 8) | (h->mb.i_cbp_chroma << 4) | h->mb.i_cbp_luma;
588 * XXX: in the me perhaps we should take x264_mb_predict_mv_pskip into account
589 * (if multiple mv give same result)*/
590 if( !b_force_no_skip )
592 if( h->mb.i_type == P_L0 && h->mb.i_partition == D_16x16 &&
593 h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma == 0x00 &&
594 h->mb.cache.mv[0][x264_scan8[0]][0] == h->mb.cache.pskip_mv[0] &&
595 h->mb.cache.mv[0][x264_scan8[0]][1] == h->mb.cache.pskip_mv[1] &&
596 h->mb.cache.ref[0][x264_scan8[0]] == 0 )
598 h->mb.i_type = P_SKIP;
601 /* Check for B_SKIP */
602 if( h->mb.i_type == B_DIRECT &&
603 h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma== 0x00 )
605 h->mb.i_type = B_SKIP;
610 /*****************************************************************************
611 * x264_macroblock_probe_skip:
612 * Check if the current MB could be encoded as a [PB]_SKIP (it supposes you use
614 *****************************************************************************/
615 int x264_macroblock_probe_skip( x264_t *h, const int b_bidir )
617 DECLARE_ALIGNED_16( int16_t dct4x4[16][4][4] );
618 DECLARE_ALIGNED_16( int16_t dct2x2[2][2] );
619 DECLARE_ALIGNED_16( int16_t dctscan[16] );
621 int i_qp = h->mb.i_qp;
631 mvp[0] = x264_clip3( h->mb.cache.pskip_mv[0], h->mb.mv_min[0], h->mb.mv_max[0] );
632 mvp[1] = x264_clip3( h->mb.cache.pskip_mv[1], h->mb.mv_min[1], h->mb.mv_max[1] );
634 /* Motion compensation */
635 h->mc.mc_luma( h->mb.pic.p_fdec[0], FDEC_STRIDE,
636 h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
637 mvp[0], mvp[1], 16, 16 );
641 h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0],
642 h->mb.pic.p_fdec[0] );
644 for( i8x8 = 0, i_decimate_mb = 0; i8x8 < 4; i8x8++ )
646 /* encode one 4x4 block */
647 for( i4x4 = 0; i4x4 < 4; i4x4++ )
649 const int idx = i8x8 * 4 + i4x4;
651 h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
652 h->zigzagf.scan_4x4( dctscan, dct4x4[idx] );
654 i_decimate_mb += x264_mb_decimate_score( dctscan, 16 );
656 if( i_decimate_mb >= 6 )
665 i_qp = h->mb.i_chroma_qp;
667 for( ch = 0; ch < 2; ch++ )
669 uint8_t *p_src = h->mb.pic.p_fenc[1+ch];
670 uint8_t *p_dst = h->mb.pic.p_fdec[1+ch];
674 h->mc.mc_chroma( h->mb.pic.p_fdec[1+ch], FDEC_STRIDE,
675 h->mb.pic.p_fref[0][0][4+ch], h->mb.pic.i_stride[1+ch],
676 mvp[0], mvp[1], 8, 8 );
679 h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
681 /* calculate dct DC */
682 dct2x2[0][0] = dct4x4[0][0][0];
683 dct2x2[0][1] = dct4x4[1][0][0];
684 dct2x2[1][0] = dct4x4[2][0][0];
685 dct2x2[1][1] = dct4x4[3][0][0];
686 h->dctf.dct2x2dc( dct2x2 );
687 h->quantf.quant_2x2_dc( dct2x2, h->quant4_mf[CQM_4PC][i_qp][0]>>1, h->quant4_bias[CQM_4PC][i_qp][0]<<1 );
688 if( dct2x2[0][0] || dct2x2[0][1] || dct2x2[1][0] || dct2x2[1][1] )
694 /* calculate dct coeffs */
695 for( i4x4 = 0, i_decimate_mb = 0; i4x4 < 4; i4x4++ )
697 h->quantf.quant_4x4( dct4x4[i4x4], h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
698 h->zigzagf.scan_4x4( dctscan, dct4x4[i4x4] );
700 i_decimate_mb += x264_mb_decimate_score( dctscan+1, 15 );
701 if( i_decimate_mb >= 7 )
711 /****************************************************************************
712 * DCT-domain noise reduction / adaptive deadzone
714 ****************************************************************************/
716 void x264_noise_reduction_update( x264_t *h )
719 for( cat = 0; cat < 2; cat++ )
721 int size = cat ? 64 : 16;
722 const uint16_t *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
724 if( h->nr_count[cat] > (cat ? (1<<16) : (1<<18)) )
726 for( i = 0; i < size; i++ )
727 h->nr_residual_sum[cat][i] >>= 1;
728 h->nr_count[cat] >>= 1;
731 for( i = 0; i < size; i++ )
732 h->nr_offset[cat][i] =
733 ((uint64_t)h->param.analyse.i_noise_reduction * h->nr_count[cat]
734 + h->nr_residual_sum[cat][i]/2)
735 / ((uint64_t)h->nr_residual_sum[cat][i] * weight[i]/256 + 1);
739 void x264_denoise_dct( x264_t *h, int16_t *dct )
741 const int cat = h->mb.b_transform_8x8;
746 for( i = (cat ? 63 : 15); i >= 1; i-- )
753 h->nr_residual_sum[cat][i] += level;
754 level -= h->nr_offset[cat][i];
760 h->nr_residual_sum[cat][i] -= level;
761 level += h->nr_offset[cat][i];
770 /*****************************************************************************
771 * RD only; 4 calls to this do not make up for one macroblock_encode.
772 * doesn't transform chroma dc.
773 *****************************************************************************/
774 void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
776 int i_qp = h->mb.i_qp;
777 uint8_t *p_fenc = h->mb.pic.p_fenc[0] + (i8&1)*8 + (i8>>1)*8*FENC_STRIDE;
778 uint8_t *p_fdec = h->mb.pic.p_fdec[0] + (i8&1)*8 + (i8>>1)*8*FDEC_STRIDE;
779 int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
783 x264_mb_mc_8x8( h, i8 );
785 if( h->mb.b_transform_8x8 )
787 DECLARE_ALIGNED_16( int16_t dct8x8[8][8] );
788 h->dctf.sub8x8_dct8( dct8x8, p_fenc, p_fdec );
789 h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
790 h->zigzagf.scan_8x8( h->dct.luma8x8[i8], dct8x8 );
793 nnz8x8 = 4 <= x264_mb_decimate_score( h->dct.luma8x8[i8], 64 );
795 nnz8x8 = array_non_zero( dct8x8 );
799 h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8PY], i_qp );
800 h->dctf.add8x8_idct8( p_fdec, dct8x8 );
806 DECLARE_ALIGNED_16( int16_t dct4x4[4][4][4] );
807 h->dctf.sub8x8_dct( dct4x4, p_fenc, p_fdec );
808 h->quantf.quant_4x4( dct4x4[0], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
809 h->quantf.quant_4x4( dct4x4[1], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
810 h->quantf.quant_4x4( dct4x4[2], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
811 h->quantf.quant_4x4( dct4x4[3], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
812 for( i4 = 0; i4 < 4; i4++ )
813 h->zigzagf.scan_4x4( h->dct.luma4x4[i8*4+i4], dct4x4[i4] );
817 int i_decimate_8x8 = 0;
818 for( i4 = 0; i4 < 4 && i_decimate_8x8 < 4; i4++ )
819 i_decimate_8x8 += x264_mb_decimate_score( h->dct.luma4x4[i8*4+i4], 16 );
820 nnz8x8 = 4 <= i_decimate_8x8;
823 nnz8x8 = array_non_zero( dct4x4 );
827 for( i4 = 0; i4 < 4; i4++ )
828 h->quantf.dequant_4x4( dct4x4[i4], h->dequant4_mf[CQM_4PY], i_qp );
829 h->dctf.add8x8_idct( p_fdec, dct4x4 );
833 i_qp = h->mb.i_chroma_qp;
835 for( ch = 0; ch < 2; ch++ )
837 DECLARE_ALIGNED_16( int16_t dct4x4[4][4] );
838 p_fenc = h->mb.pic.p_fenc[1+ch] + (i8&1)*4 + (i8>>1)*4*FENC_STRIDE;
839 p_fdec = h->mb.pic.p_fdec[1+ch] + (i8&1)*4 + (i8>>1)*4*FDEC_STRIDE;
841 h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
842 h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
843 h->zigzagf.scan_4x4( h->dct.luma4x4[16+i8+ch*4], dct4x4 );
844 if( array_non_zero( dct4x4 ) )
846 h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp );
847 h->dctf.add4x4_idct( p_fdec, dct4x4 );
852 h->mb.i_cbp_luma |= (1 << i8);
854 h->mb.i_cbp_luma &= ~(1 << i8);
855 h->mb.i_cbp_chroma = 0x02;