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 *****************************************************************************/
27 #include "common/common.h"
28 #include "macroblock.h"
31 #define ZIG(i,y,x) level[i] = dct[x][y];
32 static inline void zigzag_scan_2x2_dc( int level[4], int16_t dct[2][2] )
42 * x264_mb_decimate_score: given dct coeffs it returns a score to see if we could empty this dct coeffs
43 * to 0 (low score means set it to null)
44 * Used in inter macroblock (luma and chroma)
45 * luma: for a 8x8 block: if score < 4 -> null
46 * for the complete mb: if score < 6 -> null
47 * chroma: for the complete mb: if score < 7 -> null
49 static int x264_mb_decimate_score( int *dct, int i_max )
51 static const int i_ds_table4[16] = {
52 3,2,2,1,1,1,0,0,0,0,0,0,0,0,0,0 };
53 static const int i_ds_table8[64] = {
54 3,3,3,3,2,2,2,2,2,2,2,2,1,1,1,1,
55 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,
56 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
57 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 };
59 const int *ds_table = (i_max == 64) ? i_ds_table8 : i_ds_table4;
63 while( idx >= 0 && dct[idx] == 0 )
70 if( abs( dct[idx--] ) > 1 )
74 while( idx >= 0 && dct[idx] == 0 )
79 i_score += ds_table[i_run];
85 void x264_mb_encode_i4x4( x264_t *h, int idx, int i_qscale )
87 int x = 4 * block_idx_x[idx];
88 int y = 4 * block_idx_y[idx];
89 uint8_t *p_src = &h->mb.pic.p_fenc[0][x+y*FENC_STRIDE];
90 uint8_t *p_dst = &h->mb.pic.p_fdec[0][x+y*FDEC_STRIDE];
91 DECLARE_ALIGNED( int16_t, dct4x4[4][4], 16 );
93 if( h->mb.b_lossless )
95 h->zigzagf.sub_4x4( h->dct.block[idx].luma4x4, p_src, p_dst );
99 h->dctf.sub4x4_dct( dct4x4, p_src, p_dst );
101 if( h->mb.b_trellis )
102 x264_quant_4x4_trellis( h, dct4x4, CQM_4IY, i_qscale, DCT_LUMA_4x4, 1 );
104 h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
106 h->zigzagf.scan_4x4( h->dct.block[idx].luma4x4, dct4x4 );
107 h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4IY], i_qscale );
109 /* output samples to fdec */
110 h->dctf.add4x4_idct( p_dst, dct4x4 );
113 void x264_mb_encode_i8x8( x264_t *h, int idx, int i_qscale )
116 int y = 8 * (idx>>1);
117 uint8_t *p_src = &h->mb.pic.p_fenc[0][x+y*FENC_STRIDE];
118 uint8_t *p_dst = &h->mb.pic.p_fdec[0][x+y*FDEC_STRIDE];
119 DECLARE_ALIGNED( int16_t, dct8x8[8][8], 16 );
121 h->dctf.sub8x8_dct8( dct8x8, p_src, p_dst );
123 if( h->mb.b_trellis )
124 x264_quant_8x8_trellis( h, dct8x8, CQM_8IY, i_qscale, 1 );
126 h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8IY][i_qscale], h->quant8_bias[CQM_8IY][i_qscale] );
128 h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8 );
129 h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8IY], i_qscale );
130 h->dctf.add8x8_idct8( p_dst, dct8x8 );
133 static void x264_mb_encode_i16x16( x264_t *h, int i_qscale )
135 uint8_t *p_src = h->mb.pic.p_fenc[0];
136 uint8_t *p_dst = h->mb.pic.p_fdec[0];
138 DECLARE_ALIGNED( int16_t, dct4x4[16+1][4][4], 16 );
142 if( h->mb.b_lossless )
144 for( i = 0; i < 16; i++ )
146 int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
147 int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
148 h->zigzagf.sub_4x4ac( h->dct.block[i].residual_ac, p_src+oe, p_dst+od );
149 dct4x4[0][block_idx_x[i]][block_idx_y[i]] = p_src[oe] - p_dst[od];
150 p_dst[od] = p_src[oe];
152 h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct4x4[0] );
156 h->dctf.sub16x16_dct( &dct4x4[1], p_src, p_dst );
157 for( i = 0; i < 16; i++ )
160 dct4x4[0][block_idx_y[i]][block_idx_x[i]] = dct4x4[1+i][0][0];
162 /* quant/scan/dequant */
163 if( h->mb.b_trellis )
164 x264_quant_4x4_trellis( h, dct4x4[1+i], CQM_4IY, i_qscale, DCT_LUMA_AC, 1 );
166 h->quantf.quant_4x4( dct4x4[1+i], h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
168 h->zigzagf.scan_4x4ac( h->dct.block[i].residual_ac, dct4x4[1+i] );
169 h->quantf.dequant_4x4( dct4x4[1+i], h->dequant4_mf[CQM_4IY], i_qscale );
172 h->dctf.dct4x4dc( dct4x4[0] );
173 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 );
174 h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct4x4[0] );
176 /* output samples to fdec */
177 h->dctf.idct4x4dc( dct4x4[0] );
178 x264_mb_dequant_4x4_dc( dct4x4[0], h->dequant4_mf[CQM_4IY], i_qscale ); /* XXX not inversed */
180 /* calculate dct coeffs */
181 for( i = 0; i < 16; i++ )
184 dct4x4[1+i][0][0] = dct4x4[0][block_idx_y[i]][block_idx_x[i]];
186 /* put pixels to fdec */
187 h->dctf.add16x16_idct( p_dst, &dct4x4[1] );
190 void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qscale )
193 int b_decimate = b_inter && (h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate);
195 for( ch = 0; ch < 2; ch++ )
197 uint8_t *p_src = h->mb.pic.p_fenc[1+ch];
198 uint8_t *p_dst = h->mb.pic.p_fdec[1+ch];
199 int i_decimate_score = 0;
201 DECLARE_ALIGNED( int16_t, dct2x2[2][2] , 16 );
202 DECLARE_ALIGNED( int16_t, dct4x4[4][4][4], 16 );
204 if( h->mb.b_lossless )
206 for( i = 0; i < 4; i++ )
208 int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
209 int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
210 h->zigzagf.sub_4x4ac( h->dct.block[16+i+ch*4].residual_ac, p_src+oe, p_dst+od );
211 h->dct.chroma_dc[ch][i] = p_src[oe] - p_dst[od];
212 p_dst[od] = p_src[oe];
217 h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
218 /* calculate dct coeffs */
219 for( i = 0; i < 4; i++ )
222 dct2x2[block_idx_y[i]][block_idx_x[i]] = dct4x4[i][0][0];
224 /* no trellis; it doesn't seem to help chroma noticeably */
225 h->quantf.quant_4x4( dct4x4[i], h->quant4_mf[CQM_4IC+b_inter][i_qscale], h->quant4_bias[CQM_4IC+b_inter][i_qscale] );
226 h->zigzagf.scan_4x4ac( h->dct.block[16+i+ch*4].residual_ac, dct4x4[i] );
230 i_decimate_score += x264_mb_decimate_score( h->dct.block[16+i+ch*4].residual_ac, 15 );
234 h->dctf.dct2x2dc( dct2x2 );
235 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 );
236 zigzag_scan_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );
238 /* output samples to fdec */
239 h->dctf.idct2x2dc( dct2x2 );
240 x264_mb_dequant_2x2_dc( dct2x2, h->dequant4_mf[CQM_4IC + b_inter], i_qscale ); /* XXX not inversed */
242 if( b_decimate && i_decimate_score < 7 )
244 /* Near null chroma 8x8 block so make it null (bits saving) */
245 memset( &h->dct.block[16+ch*4], 0, 4 * sizeof( *h->dct.block ) );
246 if( !array_non_zero( dct2x2 ) )
248 memset( dct4x4, 0, sizeof( dct4x4 ) );
252 for( i = 0; i < 4; i++ )
253 h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IC + b_inter], i_qscale );
256 for( i = 0; i < 4; i++ )
257 dct4x4[i][0][0] = dct2x2[0][i];
258 h->dctf.add8x8_idct( p_dst, dct4x4 );
261 /* coded block pattern */
262 h->mb.i_cbp_chroma = 0;
263 for( i = 0; i < 8; i++ )
265 int nz = array_non_zero_count( h->dct.block[16+i].residual_ac, 15 );
266 h->mb.cache.non_zero_count[x264_scan8[16+i]] = nz;
267 h->mb.i_cbp_chroma |= nz;
269 if( h->mb.i_cbp_chroma )
270 h->mb.i_cbp_chroma = 2; /* dc+ac (we can't do only ac) */
271 else if( array_non_zero( h->dct.chroma_dc ) )
272 h->mb.i_cbp_chroma = 1; /* dc only */
275 static void x264_macroblock_encode_skip( x264_t *h )
278 h->mb.i_cbp_luma = 0x00;
279 h->mb.i_cbp_chroma = 0x00;
281 for( i = 0; i < 16+8; i++ )
283 h->mb.cache.non_zero_count[x264_scan8[i]] = 0;
287 h->mb.cbp[h->mb.i_mb_xy] = 0;
290 /*****************************************************************************
291 * x264_macroblock_encode_pskip:
292 * Encode an already marked skip block
293 *****************************************************************************/
294 void x264_macroblock_encode_pskip( x264_t *h )
296 const int mvx = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][0],
297 h->mb.mv_min[0], h->mb.mv_max[0] );
298 const int mvy = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][1],
299 h->mb.mv_min[1], h->mb.mv_max[1] );
301 /* Motion compensation XXX probably unneeded */
302 h->mc.mc_luma( h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
303 h->mb.pic.p_fdec[0], FDEC_STRIDE,
307 h->mc.mc_chroma( h->mb.pic.p_fref[0][0][4], h->mb.pic.i_stride[1],
308 h->mb.pic.p_fdec[1], FDEC_STRIDE,
311 h->mc.mc_chroma( h->mb.pic.p_fref[0][0][5], h->mb.pic.i_stride[2],
312 h->mb.pic.p_fdec[2], FDEC_STRIDE,
315 x264_macroblock_encode_skip( h );
318 /*****************************************************************************
319 * x264_macroblock_encode:
320 *****************************************************************************/
321 void x264_macroblock_encode( x264_t *h )
324 int i_qp = h->mb.i_qp;
325 int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
326 int b_force_no_skip = 0;
330 && h->mb.i_mb_xy == h->sh.i_first_mb + h->mb.i_mb_stride
331 && IS_SKIP(h->mb.type[h->sh.i_first_mb]) )
333 /* The first skip is predicted to be a frame mb pair.
334 * We don't yet support the aff part of mbaff, so force it to non-skip
335 * so that we can pick the aff flag. */
337 if( IS_SKIP(h->mb.i_type) )
339 if( h->mb.i_type == P_SKIP )
341 else if( h->mb.i_type == B_SKIP )
342 h->mb.i_type = B_DIRECT;
346 if( h->mb.i_type == P_SKIP )
349 x264_macroblock_encode_pskip( h );
352 if( h->mb.i_type == B_SKIP )
354 /* XXX motion compensation is probably unneeded */
356 x264_macroblock_encode_skip( h );
360 if( h->mb.i_type == I_16x16 )
362 const int i_mode = h->mb.i_intra16x16_pred_mode;
363 h->mb.b_transform_8x8 = 0;
364 /* do the right prediction */
365 h->predict_16x16[i_mode]( h->mb.pic.p_fdec[0] );
367 /* encode the 16x16 macroblock */
368 x264_mb_encode_i16x16( h, i_qp );
370 else if( h->mb.i_type == I_8x8 )
372 DECLARE_ALIGNED( uint8_t, edge[33], 8 );
373 h->mb.b_transform_8x8 = 1;
374 for( i = 0; i < 4; i++ )
376 uint8_t *p_dst = &h->mb.pic.p_fdec[0][8 * (i&1) + 8 * (i>>1) * FDEC_STRIDE];
377 int i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[4*i]];
379 x264_predict_8x8_filter( p_dst, edge, h->mb.i_neighbour8[i], x264_pred_i4x4_neighbors[i_mode] );
380 h->predict_8x8[i_mode]( p_dst, edge );
381 x264_mb_encode_i8x8( h, i, i_qp );
384 else if( h->mb.i_type == I_4x4 )
386 h->mb.b_transform_8x8 = 0;
387 for( i = 0; i < 16; i++ )
389 uint8_t *p_dst = &h->mb.pic.p_fdec[0][4 * block_idx_x[i] + 4 * block_idx_y[i] * FDEC_STRIDE];
390 int i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[i]];
392 if( (h->mb.i_neighbour4[i] & (MB_TOPRIGHT|MB_TOP)) == MB_TOP )
393 /* emulate missing topright samples */
394 *(uint32_t*) &p_dst[4-FDEC_STRIDE] = p_dst[3-FDEC_STRIDE] * 0x01010101U;
396 h->predict_4x4[i_mode]( p_dst );
397 x264_mb_encode_i4x4( h, i, i_qp );
403 int i_decimate_mb = 0;
405 /* Motion compensation */
408 if( h->mb.b_lossless )
410 for( i4x4 = 0; i4x4 < 16; i4x4++ )
412 int x = 4*block_idx_x[i4x4];
413 int y = 4*block_idx_y[i4x4];
414 h->zigzagf.sub_4x4( h->dct.block[i4x4].luma4x4,
415 h->mb.pic.p_fenc[0]+x+y*FENC_STRIDE,
416 h->mb.pic.p_fdec[0]+x+y*FDEC_STRIDE );
419 else if( h->mb.b_transform_8x8 )
421 DECLARE_ALIGNED( int16_t, dct8x8[4][8][8], 16 );
422 int nnz8x8[4] = {1,1,1,1};
423 b_decimate &= !h->mb.b_trellis; // 8x8 trellis is inherently optimal decimation
424 h->dctf.sub16x16_dct8( dct8x8, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
426 for( idx = 0; idx < 4; idx++ )
428 if( h->mb.b_noise_reduction )
429 x264_denoise_dct( h, (int16_t*)dct8x8[idx] );
430 if( h->mb.b_trellis )
431 x264_quant_8x8_trellis( h, dct8x8[idx], CQM_8PY, i_qp, 0 );
433 h->quantf.quant_8x8( dct8x8[idx], h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
435 h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8[idx] );
439 int i_decimate_8x8 = x264_mb_decimate_score( h->dct.luma8x8[idx], 64 );
440 i_decimate_mb += i_decimate_8x8;
441 if( i_decimate_8x8 < 4 )
443 memset( h->dct.luma8x8[idx], 0, sizeof( h->dct.luma8x8[idx] ) );
444 memset( dct8x8[idx], 0, sizeof( dct8x8[idx] ) );
449 nnz8x8[idx] = array_non_zero( dct8x8[idx] );
452 if( i_decimate_mb < 6 && b_decimate )
453 memset( h->dct.luma8x8, 0, sizeof( h->dct.luma8x8 ) );
456 for( idx = 0; idx < 4; idx++ )
459 h->quantf.dequant_8x8( dct8x8[idx], h->dequant8_mf[CQM_8PY], i_qp );
460 h->dctf.add8x8_idct8( &h->mb.pic.p_fdec[0][(idx&1)*8 + (idx>>1)*8*FDEC_STRIDE], dct8x8[idx] );
466 DECLARE_ALIGNED( int16_t, dct4x4[16][4][4], 16 );
467 int nnz8x8[4] = {1,1,1,1};
468 h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
470 for( i8x8 = 0; i8x8 < 4; i8x8++ )
474 /* encode one 4x4 block */
476 for( i4x4 = 0; i4x4 < 4; i4x4++ )
478 idx = i8x8 * 4 + i4x4;
480 if( h->mb.b_noise_reduction )
481 x264_denoise_dct( h, (int16_t*)dct4x4[idx] );
482 if( h->mb.b_trellis )
483 x264_quant_4x4_trellis( h, dct4x4[idx], CQM_4PY, i_qp, DCT_LUMA_4x4, 0 );
485 h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
487 h->zigzagf.scan_4x4( h->dct.block[idx].luma4x4, dct4x4[idx] );
490 i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[idx].luma4x4, 16 );
493 /* decimate this 8x8 block */
494 i_decimate_mb += i_decimate_8x8;
495 if( i_decimate_8x8 < 4 && b_decimate )
497 memset( &dct4x4[i8x8*4], 0, 4 * sizeof( *dct4x4 ) );
498 memset( &h->dct.block[i8x8*4], 0, 4 * sizeof( *h->dct.block ) );
503 if( i_decimate_mb < 6 && b_decimate )
504 memset( h->dct.block, 0, 16 * sizeof( *h->dct.block ) );
507 for( i8x8 = 0; i8x8 < 4; i8x8++ )
510 for( i = 0; i < 4; i++ )
511 h->quantf.dequant_4x4( dct4x4[i8x8*4+i], h->dequant4_mf[CQM_4PY], i_qp );
512 h->dctf.add8x8_idct( &h->mb.pic.p_fdec[0][(i8x8&1)*8 + (i8x8>>1)*8*FDEC_STRIDE], &dct4x4[i8x8*4] );
519 if( IS_INTRA( h->mb.i_type ) )
521 const int i_mode = h->mb.i_chroma_pred_mode;
522 h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[1] );
523 h->predict_8x8c[i_mode]( h->mb.pic.p_fdec[2] );
526 /* encode the 8x8 blocks */
527 x264_mb_encode_8x8_chroma( h, !IS_INTRA( h->mb.i_type ), h->mb.i_chroma_qp );
529 /* coded block pattern and non_zero_count */
530 h->mb.i_cbp_luma = 0x00;
531 if( h->mb.i_type == I_16x16 )
533 for( i = 0; i < 16; i++ )
535 const int nz = array_non_zero_count( h->dct.block[i].residual_ac, 15 );
536 h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
538 h->mb.i_cbp_luma = 0x0f;
541 else if( h->mb.b_transform_8x8 )
543 /* coded_block_flag is enough for CABAC.
544 * the full non_zero_count is done only in CAVLC. */
545 for( i = 0; i < 4; i++ )
547 const int nz = array_non_zero( h->dct.luma8x8[i] );
549 for( j = 0; j < 4; j++ )
550 h->mb.cache.non_zero_count[x264_scan8[4*i+j]] = nz;
552 h->mb.i_cbp_luma |= 1 << i;
557 for( i = 0; i < 16; i++ )
559 const int nz = array_non_zero_count( h->dct.block[i].luma4x4, 16 );
560 h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
562 h->mb.i_cbp_luma |= 1 << (i/4);
566 if( h->param.b_cabac )
568 i_cbp_dc = ( h->mb.i_type == I_16x16 && array_non_zero( h->dct.luma16x16_dc ) )
569 | array_non_zero( h->dct.chroma_dc[0] ) << 1
570 | array_non_zero( h->dct.chroma_dc[1] ) << 2;
574 h->mb.cbp[h->mb.i_mb_xy] = (i_cbp_dc << 8) | (h->mb.i_cbp_chroma << 4) | h->mb.i_cbp_luma;
577 * XXX: in the me perhaps we should take x264_mb_predict_mv_pskip into account
578 * (if multiple mv give same result)*/
579 if( !b_force_no_skip )
581 if( h->mb.i_type == P_L0 && h->mb.i_partition == D_16x16 &&
582 h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma == 0x00 &&
583 h->mb.cache.mv[0][x264_scan8[0]][0] == h->mb.cache.pskip_mv[0] &&
584 h->mb.cache.mv[0][x264_scan8[0]][1] == h->mb.cache.pskip_mv[1] &&
585 h->mb.cache.ref[0][x264_scan8[0]] == 0 )
587 h->mb.i_type = P_SKIP;
590 /* Check for B_SKIP */
591 if( h->mb.i_type == B_DIRECT &&
592 h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma== 0x00 )
594 h->mb.i_type = B_SKIP;
599 /*****************************************************************************
600 * x264_macroblock_probe_skip:
601 * Check if the current MB could be encoded as a [PB]_SKIP (it supposes you use
603 *****************************************************************************/
604 int x264_macroblock_probe_skip( x264_t *h, const int b_bidir )
606 DECLARE_ALIGNED( int16_t, dct4x4[16][4][4], 16 );
607 DECLARE_ALIGNED( int16_t, dct2x2[2][2], 16 );
608 DECLARE_ALIGNED( int, dctscan[16], 16 );
610 int i_qp = h->mb.i_qp;
620 mvp[0] = x264_clip3( h->mb.cache.pskip_mv[0], h->mb.mv_min[0], h->mb.mv_max[0] );
621 mvp[1] = x264_clip3( h->mb.cache.pskip_mv[1], h->mb.mv_min[1], h->mb.mv_max[1] );
623 /* Motion compensation */
624 h->mc.mc_luma( h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
625 h->mb.pic.p_fdec[0], FDEC_STRIDE,
626 mvp[0], mvp[1], 16, 16 );
630 h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0],
631 h->mb.pic.p_fdec[0] );
633 for( i8x8 = 0, i_decimate_mb = 0; i8x8 < 4; i8x8++ )
635 /* encode one 4x4 block */
636 for( i4x4 = 0; i4x4 < 4; i4x4++ )
638 const int idx = i8x8 * 4 + i4x4;
640 h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
641 h->zigzagf.scan_4x4( dctscan, dct4x4[idx] );
643 i_decimate_mb += x264_mb_decimate_score( dctscan, 16 );
645 if( i_decimate_mb >= 6 )
654 i_qp = h->mb.i_chroma_qp;
656 for( ch = 0; ch < 2; ch++ )
658 uint8_t *p_src = h->mb.pic.p_fenc[1+ch];
659 uint8_t *p_dst = h->mb.pic.p_fdec[1+ch];
663 h->mc.mc_chroma( h->mb.pic.p_fref[0][0][4+ch], h->mb.pic.i_stride[1+ch],
664 h->mb.pic.p_fdec[1+ch], FDEC_STRIDE,
665 mvp[0], mvp[1], 8, 8 );
668 h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
670 /* calculate dct DC */
671 dct2x2[0][0] = dct4x4[0][0][0];
672 dct2x2[0][1] = dct4x4[1][0][0];
673 dct2x2[1][0] = dct4x4[2][0][0];
674 dct2x2[1][1] = dct4x4[3][0][0];
675 h->dctf.dct2x2dc( dct2x2 );
676 h->quantf.quant_2x2_dc( dct2x2, h->quant4_mf[CQM_4PC][i_qp][0]>>1, h->quant4_bias[CQM_4PC][i_qp][0]<<1 );
677 if( dct2x2[0][0] || dct2x2[0][1] || dct2x2[1][0] || dct2x2[1][1] )
683 /* calculate dct coeffs */
684 for( i4x4 = 0, i_decimate_mb = 0; i4x4 < 4; i4x4++ )
686 h->quantf.quant_4x4( dct4x4[i4x4], h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
687 h->zigzagf.scan_4x4ac( dctscan, dct4x4[i4x4] );
689 i_decimate_mb += x264_mb_decimate_score( dctscan, 15 );
690 if( i_decimate_mb >= 7 )
700 /****************************************************************************
701 * DCT-domain noise reduction / adaptive deadzone
703 ****************************************************************************/
705 void x264_noise_reduction_update( x264_t *h )
708 for( cat = 0; cat < 2; cat++ )
710 int size = cat ? 64 : 16;
711 const int *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
713 if( h->nr_count[cat] > (cat ? (1<<16) : (1<<18)) )
715 for( i = 0; i < size; i++ )
716 h->nr_residual_sum[cat][i] >>= 1;
717 h->nr_count[cat] >>= 1;
720 for( i = 0; i < size; i++ )
721 h->nr_offset[cat][i] =
722 ((uint64_t)h->param.analyse.i_noise_reduction * h->nr_count[cat]
723 + h->nr_residual_sum[cat][i]/2)
724 / ((uint64_t)h->nr_residual_sum[cat][i] * weight[i]/256 + 1);
728 void x264_denoise_dct( x264_t *h, int16_t *dct )
730 const int cat = h->mb.b_transform_8x8;
735 for( i = (cat ? 63 : 15); i >= 1; i-- )
742 h->nr_residual_sum[cat][i] += level;
743 level -= h->nr_offset[cat][i];
749 h->nr_residual_sum[cat][i] -= level;
750 level += h->nr_offset[cat][i];
759 /*****************************************************************************
760 * RD only; 4 calls to this do not make up for one macroblock_encode.
761 * doesn't transform chroma dc.
762 *****************************************************************************/
763 void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
765 int i_qp = h->mb.i_qp;
766 uint8_t *p_fenc = h->mb.pic.p_fenc[0] + (i8&1)*8 + (i8>>1)*8*FENC_STRIDE;
767 uint8_t *p_fdec = h->mb.pic.p_fdec[0] + (i8&1)*8 + (i8>>1)*8*FDEC_STRIDE;
768 int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
772 x264_mb_mc_8x8( h, i8 );
774 if( h->mb.b_transform_8x8 )
776 DECLARE_ALIGNED( int16_t, dct8x8[8][8], 16 );
777 h->dctf.sub8x8_dct8( dct8x8, p_fenc, p_fdec );
778 h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
779 h->zigzagf.scan_8x8( h->dct.luma8x8[i8], dct8x8 );
782 nnz8x8 = 4 <= x264_mb_decimate_score( h->dct.luma8x8[i8], 64 );
784 nnz8x8 = array_non_zero( dct8x8 );
788 h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8PY], i_qp );
789 h->dctf.add8x8_idct8( p_fdec, dct8x8 );
795 DECLARE_ALIGNED( int16_t, dct4x4[4][4][4], 16 );
796 h->dctf.sub8x8_dct( dct4x4, p_fenc, p_fdec );
797 h->quantf.quant_4x4( dct4x4[0], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
798 h->quantf.quant_4x4( dct4x4[1], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
799 h->quantf.quant_4x4( dct4x4[2], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
800 h->quantf.quant_4x4( dct4x4[3], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
801 for( i4 = 0; i4 < 4; i4++ )
802 h->zigzagf.scan_4x4( h->dct.block[i8*4+i4].luma4x4, dct4x4[i4] );
806 int i_decimate_8x8 = 0;
807 for( i4 = 0; i4 < 4 && i_decimate_8x8 < 4; i4++ )
808 i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[i8*4+i4].luma4x4, 16 );
809 nnz8x8 = 4 <= i_decimate_8x8;
812 nnz8x8 = array_non_zero( dct4x4 );
816 for( i4 = 0; i4 < 4; i4++ )
817 h->quantf.dequant_4x4( dct4x4[i4], h->dequant4_mf[CQM_4PY], i_qp );
818 h->dctf.add8x8_idct( p_fdec, dct4x4 );
822 i_qp = h->mb.i_chroma_qp;
824 for( ch = 0; ch < 2; ch++ )
826 DECLARE_ALIGNED( int16_t, dct4x4[4][4], 16 );
827 p_fenc = h->mb.pic.p_fenc[1+ch] + (i8&1)*4 + (i8>>1)*4*FENC_STRIDE;
828 p_fdec = h->mb.pic.p_fdec[1+ch] + (i8&1)*4 + (i8>>1)*4*FDEC_STRIDE;
830 h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
831 h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
832 h->zigzagf.scan_4x4ac( h->dct.block[16+i8+ch*4].residual_ac, dct4x4 );
833 if( array_non_zero( dct4x4 ) )
835 h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp );
836 h->dctf.add4x4_idct( p_fdec, dct4x4 );
841 h->mb.i_cbp_luma |= (1 << i8);
843 h->mb.i_cbp_luma &= ~(1 << i8);
844 h->mb.i_cbp_chroma = 0x02;