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( int 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( int *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( int16_t, dct4x4[4][4], 16 );
90 if( h->mb.b_lossless )
92 h->zigzagf.sub_4x4( h->dct.block[idx].luma4x4, 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.block[idx].luma4x4, 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( int16_t, dct8x8[8][8], 16 );
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( int16_t, dct4x4[16+1][4][4], 16 );
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_4x4ac( h->dct.block[i].residual_ac, p_src+oe, p_dst+od );
146 dct4x4[0][block_idx_x[i]][block_idx_y[i]] = p_src[oe] - p_dst[od];
147 p_dst[od] = p_src[oe];
149 h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct4x4[0] );
153 h->dctf.sub16x16_dct( &dct4x4[1], p_src, p_dst );
154 for( i = 0; i < 16; i++ )
157 dct4x4[0][block_idx_y[i]][block_idx_x[i]] = dct4x4[1+i][0][0];
159 /* quant/scan/dequant */
160 if( h->mb.b_trellis )
161 x264_quant_4x4_trellis( h, dct4x4[1+i], CQM_4IY, i_qscale, DCT_LUMA_AC, 1 );
163 h->quantf.quant_4x4( dct4x4[1+i], h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
165 h->zigzagf.scan_4x4ac( h->dct.block[i].residual_ac, dct4x4[1+i] );
166 h->quantf.dequant_4x4( dct4x4[1+i], h->dequant4_mf[CQM_4IY], i_qscale );
169 h->dctf.dct4x4dc( dct4x4[0] );
170 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 );
171 h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct4x4[0] );
173 /* output samples to fdec */
174 h->dctf.idct4x4dc( dct4x4[0] );
175 x264_mb_dequant_4x4_dc( dct4x4[0], h->dequant4_mf[CQM_4IY], i_qscale ); /* XXX not inversed */
177 /* calculate dct coeffs */
178 for( i = 0; i < 16; i++ )
181 dct4x4[1+i][0][0] = dct4x4[0][block_idx_y[i]][block_idx_x[i]];
183 /* put pixels to fdec */
184 h->dctf.add16x16_idct( p_dst, &dct4x4[1] );
187 void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qscale )
190 int b_decimate = b_inter && (h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate);
192 for( ch = 0; ch < 2; ch++ )
194 uint8_t *p_src = h->mb.pic.p_fenc[1+ch];
195 uint8_t *p_dst = h->mb.pic.p_fdec[1+ch];
196 int i_decimate_score = 0;
198 DECLARE_ALIGNED( int16_t, dct2x2[2][2] , 16 );
199 DECLARE_ALIGNED( int16_t, dct4x4[4][4][4], 16 );
201 if( h->mb.b_lossless )
203 for( i = 0; i < 4; i++ )
205 int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
206 int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
207 h->zigzagf.sub_4x4ac( h->dct.block[16+i+ch*4].residual_ac, p_src+oe, p_dst+od );
208 h->dct.chroma_dc[ch][i] = p_src[oe] - p_dst[od];
209 p_dst[od] = p_src[oe];
214 h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
215 /* calculate dct coeffs */
216 for( i = 0; i < 4; i++ )
219 dct2x2[block_idx_y[i]][block_idx_x[i]] = dct4x4[i][0][0];
221 /* no trellis; it doesn't seem to help chroma noticeably */
222 h->quantf.quant_4x4( dct4x4[i], h->quant4_mf[CQM_4IC+b_inter][i_qscale], h->quant4_bias[CQM_4IC+b_inter][i_qscale] );
223 h->zigzagf.scan_4x4ac( h->dct.block[16+i+ch*4].residual_ac, dct4x4[i] );
227 i_decimate_score += x264_mb_decimate_score( h->dct.block[16+i+ch*4].residual_ac, 15 );
231 h->dctf.dct2x2dc( dct2x2 );
232 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 );
233 zigzag_scan_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );
235 /* output samples to fdec */
236 h->dctf.idct2x2dc( dct2x2 );
237 x264_mb_dequant_2x2_dc( dct2x2, h->dequant4_mf[CQM_4IC + b_inter], i_qscale ); /* XXX not inversed */
239 if( b_decimate && i_decimate_score < 7 )
241 /* Near null chroma 8x8 block so make it null (bits saving) */
242 memset( &h->dct.block[16+ch*4], 0, 4 * sizeof( *h->dct.block ) );
243 if( !array_non_zero( dct2x2 ) )
245 memset( dct4x4, 0, sizeof( dct4x4 ) );
249 for( i = 0; i < 4; i++ )
250 h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IC + b_inter], i_qscale );
253 for( i = 0; i < 4; i++ )
254 dct4x4[i][0][0] = dct2x2[0][i];
255 h->dctf.add8x8_idct( p_dst, dct4x4 );
258 /* coded block pattern */
259 h->mb.i_cbp_chroma = 0;
260 for( i = 0; i < 8; i++ )
262 int nz = array_non_zero_count( h->dct.block[16+i].residual_ac, 15 );
263 h->mb.cache.non_zero_count[x264_scan8[16+i]] = nz;
264 h->mb.i_cbp_chroma |= nz;
266 if( h->mb.i_cbp_chroma )
267 h->mb.i_cbp_chroma = 2; /* dc+ac (we can't do only ac) */
268 else if( array_non_zero( h->dct.chroma_dc ) )
269 h->mb.i_cbp_chroma = 1; /* dc only */
272 static void x264_macroblock_encode_skip( x264_t *h )
275 h->mb.i_cbp_luma = 0x00;
276 h->mb.i_cbp_chroma = 0x00;
278 for( i = 0; i < 16+8; i++ )
280 h->mb.cache.non_zero_count[x264_scan8[i]] = 0;
284 h->mb.cbp[h->mb.i_mb_xy] = 0;
287 /*****************************************************************************
288 * x264_macroblock_encode_pskip:
289 * Encode an already marked skip block
290 *****************************************************************************/
291 void x264_macroblock_encode_pskip( x264_t *h )
293 const int mvx = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][0],
294 h->mb.mv_min[0], h->mb.mv_max[0] );
295 const int mvy = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][1],
296 h->mb.mv_min[1], h->mb.mv_max[1] );
298 /* Motion compensation XXX probably unneeded */
299 h->mc.mc_luma( h->mb.pic.p_fdec[0], FDEC_STRIDE,
300 h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
304 h->mc.mc_chroma( h->mb.pic.p_fdec[1], FDEC_STRIDE,
305 h->mb.pic.p_fref[0][0][4], h->mb.pic.i_stride[1],
308 h->mc.mc_chroma( h->mb.pic.p_fdec[2], FDEC_STRIDE,
309 h->mb.pic.p_fref[0][0][5], h->mb.pic.i_stride[2],
312 x264_macroblock_encode_skip( h );
315 /*****************************************************************************
316 * x264_macroblock_encode:
317 *****************************************************************************/
318 void x264_macroblock_encode( x264_t *h )
321 int i_qp = h->mb.i_qp;
322 int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
323 int b_force_no_skip = 0;
327 && h->mb.i_mb_xy == h->sh.i_first_mb + h->mb.i_mb_stride
328 && IS_SKIP(h->mb.type[h->sh.i_first_mb]) )
330 /* The first skip is predicted to be a frame mb pair.
331 * We don't yet support the aff part of mbaff, so force it to non-skip
332 * so that we can pick the aff flag. */
334 if( IS_SKIP(h->mb.i_type) )
336 if( h->mb.i_type == P_SKIP )
338 else if( h->mb.i_type == B_SKIP )
339 h->mb.i_type = B_DIRECT;
343 if( h->mb.i_type == P_SKIP )
346 x264_macroblock_encode_pskip( h );
349 if( h->mb.i_type == B_SKIP )
351 /* XXX motion compensation is probably unneeded */
353 x264_macroblock_encode_skip( h );
357 if( h->mb.i_type == I_16x16 )
359 const int i_mode = h->mb.i_intra16x16_pred_mode;
360 h->mb.b_transform_8x8 = 0;
361 /* do the right prediction */
362 h->predict_16x16[i_mode]( h->mb.pic.p_fdec[0] );
364 /* encode the 16x16 macroblock */
365 x264_mb_encode_i16x16( h, i_qp );
367 else if( h->mb.i_type == I_8x8 )
369 DECLARE_ALIGNED( uint8_t, edge[33], 8 );
370 h->mb.b_transform_8x8 = 1;
371 for( i = 0; i < 4; i++ )
373 uint8_t *p_dst = &h->mb.pic.p_fdec[0][8 * (i&1) + 8 * (i>>1) * FDEC_STRIDE];
374 int i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[4*i]];
376 x264_predict_8x8_filter( p_dst, edge, h->mb.i_neighbour8[i], x264_pred_i4x4_neighbors[i_mode] );
377 h->predict_8x8[i_mode]( p_dst, edge );
378 x264_mb_encode_i8x8( h, i, i_qp );
381 else if( h->mb.i_type == I_4x4 )
383 h->mb.b_transform_8x8 = 0;
384 for( i = 0; i < 16; i++ )
386 uint8_t *p_dst = &h->mb.pic.p_fdec[0][4 * block_idx_x[i] + 4 * block_idx_y[i] * FDEC_STRIDE];
387 int i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[i]];
389 if( (h->mb.i_neighbour4[i] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
390 /* emulate missing topright samples */
391 *(uint32_t*) &p_dst[4-FDEC_STRIDE] = p_dst[3-FDEC_STRIDE] * 0x01010101U;
393 h->predict_4x4[i_mode]( p_dst );
394 x264_mb_encode_i4x4( h, i, i_qp );
400 int i_decimate_mb = 0;
402 /* Motion compensation */
405 if( h->mb.b_lossless )
407 for( i4x4 = 0; i4x4 < 16; i4x4++ )
409 int x = 4*block_idx_x[i4x4];
410 int y = 4*block_idx_y[i4x4];
411 h->zigzagf.sub_4x4( h->dct.block[i4x4].luma4x4,
412 h->mb.pic.p_fenc[0]+x+y*FENC_STRIDE,
413 h->mb.pic.p_fdec[0]+x+y*FDEC_STRIDE );
416 else if( h->mb.b_transform_8x8 )
418 DECLARE_ALIGNED( int16_t, dct8x8[4][8][8], 16 );
419 int nnz8x8[4] = {1,1,1,1};
420 b_decimate &= !h->mb.b_trellis; // 8x8 trellis is inherently optimal decimation
421 h->dctf.sub16x16_dct8( dct8x8, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
423 for( idx = 0; idx < 4; idx++ )
425 if( h->mb.b_noise_reduction )
426 x264_denoise_dct( h, (int16_t*)dct8x8[idx] );
427 if( h->mb.b_trellis )
428 x264_quant_8x8_trellis( h, dct8x8[idx], CQM_8PY, i_qp, 0 );
430 h->quantf.quant_8x8( dct8x8[idx], h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
432 h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8[idx] );
436 int i_decimate_8x8 = x264_mb_decimate_score( h->dct.luma8x8[idx], 64 );
437 i_decimate_mb += i_decimate_8x8;
438 if( i_decimate_8x8 < 4 )
440 memset( h->dct.luma8x8[idx], 0, sizeof( h->dct.luma8x8[idx] ) );
441 memset( dct8x8[idx], 0, sizeof( dct8x8[idx] ) );
446 nnz8x8[idx] = array_non_zero( dct8x8[idx] );
449 if( i_decimate_mb < 6 && b_decimate )
450 memset( h->dct.luma8x8, 0, sizeof( h->dct.luma8x8 ) );
453 for( idx = 0; idx < 4; idx++ )
456 h->quantf.dequant_8x8( dct8x8[idx], h->dequant8_mf[CQM_8PY], i_qp );
457 h->dctf.add8x8_idct8( &h->mb.pic.p_fdec[0][(idx&1)*8 + (idx>>1)*8*FDEC_STRIDE], dct8x8[idx] );
463 DECLARE_ALIGNED( int16_t, dct4x4[16][4][4], 16 );
464 int nnz8x8[4] = {1,1,1,1};
465 h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
467 for( i8x8 = 0; i8x8 < 4; i8x8++ )
471 /* encode one 4x4 block */
473 for( i4x4 = 0; i4x4 < 4; i4x4++ )
475 idx = i8x8 * 4 + i4x4;
477 if( h->mb.b_noise_reduction )
478 x264_denoise_dct( h, (int16_t*)dct4x4[idx] );
479 if( h->mb.b_trellis )
480 x264_quant_4x4_trellis( h, dct4x4[idx], CQM_4PY, i_qp, DCT_LUMA_4x4, 0 );
482 h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
484 h->zigzagf.scan_4x4( h->dct.block[idx].luma4x4, dct4x4[idx] );
487 i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[idx].luma4x4, 16 );
490 /* decimate this 8x8 block */
491 i_decimate_mb += i_decimate_8x8;
492 if( i_decimate_8x8 < 4 && b_decimate )
494 memset( &dct4x4[i8x8*4], 0, 4 * sizeof( *dct4x4 ) );
495 memset( &h->dct.block[i8x8*4], 0, 4 * sizeof( *h->dct.block ) );
500 if( i_decimate_mb < 6 && b_decimate )
501 memset( h->dct.block, 0, 16 * sizeof( *h->dct.block ) );
504 for( i8x8 = 0; i8x8 < 4; i8x8++ )
507 for( i = 0; i < 4; i++ )
508 h->quantf.dequant_4x4( dct4x4[i8x8*4+i], h->dequant4_mf[CQM_4PY], i_qp );
509 h->dctf.add8x8_idct( &h->mb.pic.p_fdec[0][(i8x8&1)*8 + (i8x8>>1)*8*FDEC_STRIDE], &dct4x4[i8x8*4] );
516 if( IS_INTRA( h->mb.i_type ) )
518 const int i_mode = h->mb.i_chroma_pred_mode;
519 h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[1] );
520 h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[2] );
523 /* encode the 8x8 blocks */
524 x264_mb_encode_8x8_chroma( h, !IS_INTRA( h->mb.i_type ), h->mb.i_chroma_qp );
526 /* coded block pattern and non_zero_count */
527 h->mb.i_cbp_luma = 0x00;
528 if( h->mb.i_type == I_16x16 )
530 for( i = 0; i < 16; i++ )
532 const int nz = array_non_zero_count( h->dct.block[i].residual_ac, 15 );
533 h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
535 h->mb.i_cbp_luma = 0x0f;
538 else if( h->mb.b_transform_8x8 )
540 /* coded_block_flag is enough for CABAC.
541 * the full non_zero_count is done only in CAVLC. */
542 for( i = 0; i < 4; i++ )
544 const int nz = array_non_zero( h->dct.luma8x8[i] );
546 for( j = 0; j < 4; j++ )
547 h->mb.cache.non_zero_count[x264_scan8[4*i+j]] = nz;
549 h->mb.i_cbp_luma |= 1 << i;
554 for( i = 0; i < 16; i++ )
556 const int nz = array_non_zero_count( h->dct.block[i].luma4x4, 16 );
557 h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
559 h->mb.i_cbp_luma |= 1 << (i/4);
563 if( h->param.b_cabac )
565 i_cbp_dc = ( h->mb.i_type == I_16x16 && array_non_zero( h->dct.luma16x16_dc ) )
566 | array_non_zero( h->dct.chroma_dc[0] ) << 1
567 | array_non_zero( h->dct.chroma_dc[1] ) << 2;
571 h->mb.cbp[h->mb.i_mb_xy] = (i_cbp_dc << 8) | (h->mb.i_cbp_chroma << 4) | h->mb.i_cbp_luma;
574 * XXX: in the me perhaps we should take x264_mb_predict_mv_pskip into account
575 * (if multiple mv give same result)*/
576 if( !b_force_no_skip )
578 if( h->mb.i_type == P_L0 && h->mb.i_partition == D_16x16 &&
579 h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma == 0x00 &&
580 h->mb.cache.mv[0][x264_scan8[0]][0] == h->mb.cache.pskip_mv[0] &&
581 h->mb.cache.mv[0][x264_scan8[0]][1] == h->mb.cache.pskip_mv[1] &&
582 h->mb.cache.ref[0][x264_scan8[0]] == 0 )
584 h->mb.i_type = P_SKIP;
587 /* Check for B_SKIP */
588 if( h->mb.i_type == B_DIRECT &&
589 h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma== 0x00 )
591 h->mb.i_type = B_SKIP;
596 /*****************************************************************************
597 * x264_macroblock_probe_skip:
598 * Check if the current MB could be encoded as a [PB]_SKIP (it supposes you use
600 *****************************************************************************/
601 int x264_macroblock_probe_skip( x264_t *h, const int b_bidir )
603 DECLARE_ALIGNED( int16_t, dct4x4[16][4][4], 16 );
604 DECLARE_ALIGNED( int16_t, dct2x2[2][2], 16 );
605 DECLARE_ALIGNED( int, dctscan[16], 16 );
607 int i_qp = h->mb.i_qp;
617 mvp[0] = x264_clip3( h->mb.cache.pskip_mv[0], h->mb.mv_min[0], h->mb.mv_max[0] );
618 mvp[1] = x264_clip3( h->mb.cache.pskip_mv[1], h->mb.mv_min[1], h->mb.mv_max[1] );
620 /* Motion compensation */
621 h->mc.mc_luma( h->mb.pic.p_fdec[0], FDEC_STRIDE,
622 h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
623 mvp[0], mvp[1], 16, 16 );
627 h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0],
628 h->mb.pic.p_fdec[0] );
630 for( i8x8 = 0, i_decimate_mb = 0; i8x8 < 4; i8x8++ )
632 /* encode one 4x4 block */
633 for( i4x4 = 0; i4x4 < 4; i4x4++ )
635 const int idx = i8x8 * 4 + i4x4;
637 h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
638 h->zigzagf.scan_4x4( dctscan, dct4x4[idx] );
640 i_decimate_mb += x264_mb_decimate_score( dctscan, 16 );
642 if( i_decimate_mb >= 6 )
651 i_qp = h->mb.i_chroma_qp;
653 for( ch = 0; ch < 2; ch++ )
655 uint8_t *p_src = h->mb.pic.p_fenc[1+ch];
656 uint8_t *p_dst = h->mb.pic.p_fdec[1+ch];
660 h->mc.mc_chroma( h->mb.pic.p_fdec[1+ch], FDEC_STRIDE,
661 h->mb.pic.p_fref[0][0][4+ch], h->mb.pic.i_stride[1+ch],
662 mvp[0], mvp[1], 8, 8 );
665 h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
667 /* calculate dct DC */
668 dct2x2[0][0] = dct4x4[0][0][0];
669 dct2x2[0][1] = dct4x4[1][0][0];
670 dct2x2[1][0] = dct4x4[2][0][0];
671 dct2x2[1][1] = dct4x4[3][0][0];
672 h->dctf.dct2x2dc( dct2x2 );
673 h->quantf.quant_2x2_dc( dct2x2, h->quant4_mf[CQM_4PC][i_qp][0]>>1, h->quant4_bias[CQM_4PC][i_qp][0]<<1 );
674 if( dct2x2[0][0] || dct2x2[0][1] || dct2x2[1][0] || dct2x2[1][1] )
680 /* calculate dct coeffs */
681 for( i4x4 = 0, i_decimate_mb = 0; i4x4 < 4; i4x4++ )
683 h->quantf.quant_4x4( dct4x4[i4x4], h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
684 h->zigzagf.scan_4x4ac( dctscan, dct4x4[i4x4] );
686 i_decimate_mb += x264_mb_decimate_score( dctscan, 15 );
687 if( i_decimate_mb >= 7 )
697 /****************************************************************************
698 * DCT-domain noise reduction / adaptive deadzone
700 ****************************************************************************/
702 void x264_noise_reduction_update( x264_t *h )
705 for( cat = 0; cat < 2; cat++ )
707 int size = cat ? 64 : 16;
708 const int *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
710 if( h->nr_count[cat] > (cat ? (1<<16) : (1<<18)) )
712 for( i = 0; i < size; i++ )
713 h->nr_residual_sum[cat][i] >>= 1;
714 h->nr_count[cat] >>= 1;
717 for( i = 0; i < size; i++ )
718 h->nr_offset[cat][i] =
719 ((uint64_t)h->param.analyse.i_noise_reduction * h->nr_count[cat]
720 + h->nr_residual_sum[cat][i]/2)
721 / ((uint64_t)h->nr_residual_sum[cat][i] * weight[i]/256 + 1);
725 void x264_denoise_dct( x264_t *h, int16_t *dct )
727 const int cat = h->mb.b_transform_8x8;
732 for( i = (cat ? 63 : 15); i >= 1; i-- )
739 h->nr_residual_sum[cat][i] += level;
740 level -= h->nr_offset[cat][i];
746 h->nr_residual_sum[cat][i] -= level;
747 level += h->nr_offset[cat][i];
756 /*****************************************************************************
757 * RD only; 4 calls to this do not make up for one macroblock_encode.
758 * doesn't transform chroma dc.
759 *****************************************************************************/
760 void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
762 int i_qp = h->mb.i_qp;
763 uint8_t *p_fenc = h->mb.pic.p_fenc[0] + (i8&1)*8 + (i8>>1)*8*FENC_STRIDE;
764 uint8_t *p_fdec = h->mb.pic.p_fdec[0] + (i8&1)*8 + (i8>>1)*8*FDEC_STRIDE;
765 int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
769 x264_mb_mc_8x8( h, i8 );
771 if( h->mb.b_transform_8x8 )
773 DECLARE_ALIGNED( int16_t, dct8x8[8][8], 16 );
774 h->dctf.sub8x8_dct8( dct8x8, p_fenc, p_fdec );
775 h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
776 h->zigzagf.scan_8x8( h->dct.luma8x8[i8], dct8x8 );
779 nnz8x8 = 4 <= x264_mb_decimate_score( h->dct.luma8x8[i8], 64 );
781 nnz8x8 = array_non_zero( dct8x8 );
785 h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8PY], i_qp );
786 h->dctf.add8x8_idct8( p_fdec, dct8x8 );
792 DECLARE_ALIGNED( int16_t, dct4x4[4][4][4], 16 );
793 h->dctf.sub8x8_dct( dct4x4, p_fenc, p_fdec );
794 h->quantf.quant_4x4( dct4x4[0], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
795 h->quantf.quant_4x4( dct4x4[1], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
796 h->quantf.quant_4x4( dct4x4[2], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
797 h->quantf.quant_4x4( dct4x4[3], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
798 for( i4 = 0; i4 < 4; i4++ )
799 h->zigzagf.scan_4x4( h->dct.block[i8*4+i4].luma4x4, dct4x4[i4] );
803 int i_decimate_8x8 = 0;
804 for( i4 = 0; i4 < 4 && i_decimate_8x8 < 4; i4++ )
805 i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[i8*4+i4].luma4x4, 16 );
806 nnz8x8 = 4 <= i_decimate_8x8;
809 nnz8x8 = array_non_zero( dct4x4 );
813 for( i4 = 0; i4 < 4; i4++ )
814 h->quantf.dequant_4x4( dct4x4[i4], h->dequant4_mf[CQM_4PY], i_qp );
815 h->dctf.add8x8_idct( p_fdec, dct4x4 );
819 i_qp = h->mb.i_chroma_qp;
821 for( ch = 0; ch < 2; ch++ )
823 DECLARE_ALIGNED( int16_t, dct4x4[4][4], 16 );
824 p_fenc = h->mb.pic.p_fenc[1+ch] + (i8&1)*4 + (i8>>1)*4*FENC_STRIDE;
825 p_fdec = h->mb.pic.p_fdec[1+ch] + (i8&1)*4 + (i8>>1)*4*FDEC_STRIDE;
827 h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
828 h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
829 h->zigzagf.scan_4x4ac( h->dct.block[16+i8+ch*4].residual_ac, dct4x4 );
830 if( array_non_zero( dct4x4 ) )
832 h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp );
833 h->dctf.add4x4_idct( p_fdec, dct4x4 );
838 h->mb.i_cbp_luma |= (1 << i8);
840 h->mb.i_cbp_luma &= ~(1 << i8);
841 h->mb.i_cbp_chroma = 0x02;