1 /*****************************************************************************
2 * cavlc.c: cavlc bitstream writing
3 *****************************************************************************
4 * Copyright (C) 2003-2014 x264 project
5 *
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
8 * Fiona Glaser <fiona@x264.com>
9 *
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.
14 *
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.
19 *
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 *
24 * This program is also available under a commercial proprietary license.
25 * For more information, contact us at licensing@x264.com.
26 *****************************************************************************/
27
28 #include "common/common.h"
29 #include "macroblock.h"
30
31 #ifndef RDO_SKIP_BS
32 #define RDO_SKIP_BS 0
33 #endif
34
35 /* [400,420][inter,intra] */
36 static const uint8_t cbp_to_golomb[2][2][48] =
37 {
38 {{ 0, 1, 2, 5, 3, 6, 14, 10, 4, 15, 7, 11, 8, 12, 13, 9 },
39 { 1, 10, 11, 6, 12, 7, 14, 2, 13, 15, 8, 3, 9, 4, 5, 0 }},
40 {{ 0, 2, 3, 7, 4, 8, 17, 13, 5, 18, 9, 14, 10, 15, 16, 11,
41 1, 32, 33, 36, 34, 37, 44, 40, 35, 45, 38, 41, 39, 42, 43, 19,
42 6, 24, 25, 20, 26, 21, 46, 28, 27, 47, 22, 29, 23, 30, 31, 12 },
43 { 3, 29, 30, 17, 31, 18, 37, 8, 32, 38, 19, 9, 20, 10, 11, 2,
44 16, 33, 34, 21, 35, 22, 39, 4, 36, 40, 23, 5, 24, 6, 7, 1,
45 41, 42, 43, 25, 44, 26, 46, 12, 45, 47, 27, 13, 28, 14, 15, 0 }}
46 };
47
48 static const uint8_t mb_type_b_to_golomb[3][9]=
49 {
50 { 4, 8, 12, 10, 6, 14, 16, 18, 20 }, /* D_16x8 */
51 { 5, 9, 13, 11, 7, 15, 17, 19, 21 }, /* D_8x16 */
52 { 1, -1, -1, -1, 2, -1, -1, -1, 3 } /* D_16x16 */
53 };
54
55 static const uint8_t subpartition_p_to_golomb[4]=
56 {
57 3, 1, 2, 0
58 };
59
60 static const uint8_t subpartition_b_to_golomb[13]=
61 {
62 10, 4, 5, 1, 11, 6, 7, 2, 12, 8, 9, 3, 0
63 };
64
65 #define bs_write_vlc(s,v) bs_write( s, (v).i_size, (v).i_bits )
66
67 /****************************************************************************
68 * x264_cavlc_block_residual:
69 ****************************************************************************/
x264_cavlc_block_residual_escape(x264_t * h,int i_suffix_length,int level)70 static inline int x264_cavlc_block_residual_escape( x264_t *h, int i_suffix_length, int level )
71 {
72 bs_t *s = &h->out.bs;
73 static const uint16_t next_suffix[7] = { 0, 3, 6, 12, 24, 48, 0xffff };
74 int i_level_prefix = 15;
75 int mask = level >> 31;
76 int abs_level = (level^mask)-mask;
77 int i_level_code = abs_level*2-mask-2;
78 if( ( i_level_code >> i_suffix_length ) < 15 )
79 {
80 bs_write( s, (i_level_code >> i_suffix_length) + 1 + i_suffix_length,
81 (1<<i_suffix_length) + (i_level_code & ((1<<i_suffix_length)-1)) );
82 }
83 else
84 {
85 i_level_code -= 15 << i_suffix_length;
86 if( i_suffix_length == 0 )
87 i_level_code -= 15;
88
89 /* If the prefix size exceeds 15, High Profile is required. */
90 if( i_level_code >= 1<<12 )
91 {
92 if( h->sps->i_profile_idc >= PROFILE_HIGH )
93 {
94 while( i_level_code > 1<<(i_level_prefix-3) )
95 {
96 i_level_code -= 1<<(i_level_prefix-3);
97 i_level_prefix++;
98 }
99 }
100 else
101 {
102 #if RDO_SKIP_BS
103 /* Weight highly against overflows. */
104 s->i_bits_encoded += 2000;
105 #else
106 /* We've had an overflow; note it down and re-encode the MB later. */
107 h->mb.b_overflow = 1;
108 #endif
109 }
110 }
111 bs_write( s, i_level_prefix + 1, 1 );
112 bs_write( s, i_level_prefix - 3, i_level_code & ((1<<(i_level_prefix-3))-1) );
113 }
114 if( i_suffix_length == 0 )
115 i_suffix_length++;
116 if( abs_level > next_suffix[i_suffix_length] )
117 i_suffix_length++;
118 return i_suffix_length;
119 }
120
x264_cavlc_block_residual_internal(x264_t * h,int ctx_block_cat,dctcoef * l,int nC)121 static int x264_cavlc_block_residual_internal( x264_t *h, int ctx_block_cat, dctcoef *l, int nC )
122 {
123 bs_t *s = &h->out.bs;
124 static const uint8_t ctz_index[8] = {3,0,1,0,2,0,1,0};
125 static const uint8_t count_cat[14] = {16, 15, 16, 0, 15, 64, 16, 15, 16, 64, 16, 15, 16, 64};
126 x264_run_level_t runlevel;
127 int i_total, i_trailing, i_total_zero, i_suffix_length;
128 unsigned int i_sign;
129 int zero_run_code;
130
131 /* level and run and total */
132 i_total = h->quantf.coeff_level_run[ctx_block_cat]( l, &runlevel );
133 x264_prefetch( &x264_run_before[runlevel.mask] );
134 i_total_zero = runlevel.last + 1 - i_total;
135
136 /* branchless i_trailing calculation */
137 runlevel.level[i_total+0] = 2;
138 runlevel.level[i_total+1] = 2;
139 i_trailing = ((((runlevel.level[0]+1) | (1-runlevel.level[0])) >> 31) & 1) // abs(runlevel.level[0])>1
140 | ((((runlevel.level[1]+1) | (1-runlevel.level[1])) >> 31) & 2)
141 | ((((runlevel.level[2]+1) | (1-runlevel.level[2])) >> 31) & 4);
142 i_trailing = ctz_index[i_trailing];
143 i_sign = ((runlevel.level[2] >> 31) & 1)
144 | ((runlevel.level[1] >> 31) & 2)
145 | ((runlevel.level[0] >> 31) & 4);
146 i_sign >>= 3-i_trailing;
147
148 /* total/trailing */
149 bs_write_vlc( s, x264_coeff_token[nC][i_total-1][i_trailing] );
150
151 i_suffix_length = i_total > 10 && i_trailing < 3;
152 bs_write( s, i_trailing, i_sign );
153
154 if( i_trailing < i_total )
155 {
156 int val = runlevel.level[i_trailing];
157 int val_original = runlevel.level[i_trailing]+LEVEL_TABLE_SIZE/2;
158 int i;
159
160 val -= ((val>>31)|1) & -(i_trailing < 3); /* as runlevel.level[i] can't be 1 for the first one if i_trailing < 3 */
161 val += LEVEL_TABLE_SIZE/2;
162
163 if( (unsigned)val_original < LEVEL_TABLE_SIZE )
164 {
165 bs_write_vlc( s, x264_level_token[i_suffix_length][val] );
166 i_suffix_length = x264_level_token[i_suffix_length][val_original].i_next;
167 }
168 else
169 i_suffix_length = x264_cavlc_block_residual_escape( h, i_suffix_length, val-LEVEL_TABLE_SIZE/2 );
170 for( i = i_trailing+1; i < i_total; i++ )
171 {
172 val = runlevel.level[i] + LEVEL_TABLE_SIZE/2;
173 if( (unsigned)val < LEVEL_TABLE_SIZE )
174 {
175 bs_write_vlc( s, x264_level_token[i_suffix_length][val] );
176 i_suffix_length = x264_level_token[i_suffix_length][val].i_next;
177 }
178 else
179 i_suffix_length = x264_cavlc_block_residual_escape( h, i_suffix_length, val-LEVEL_TABLE_SIZE/2 );
180 }
181 }
182
183 if( ctx_block_cat == DCT_CHROMA_DC )
184 {
185 if( i_total < 8>>CHROMA_V_SHIFT )
186 {
187 vlc_t total_zeros = CHROMA_FORMAT == CHROMA_420 ? x264_total_zeros_2x2_dc[i_total-1][i_total_zero]
188 : x264_total_zeros_2x4_dc[i_total-1][i_total_zero];
189 bs_write_vlc( s, total_zeros );
190 }
191 }
192 else if( (uint8_t)i_total < count_cat[ctx_block_cat] )
193 bs_write_vlc( s, x264_total_zeros[i_total-1][i_total_zero] );
194
195 zero_run_code = x264_run_before[runlevel.mask];
196 bs_write( s, zero_run_code&0x1f, zero_run_code>>5 );
197
198 return i_total;
199 }
200
201 static const uint8_t ct_index[17] = {0,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,3};
202
203 #define x264_cavlc_block_residual(h,cat,idx,l)\
204 {\
205 int nC = cat == DCT_CHROMA_DC ? 5 - CHROMA_V_SHIFT\
206 : ct_index[x264_mb_predict_non_zero_code( h, cat == DCT_LUMA_DC ? (idx - LUMA_DC)*16 : idx )];\
207 uint8_t *nnz = &h->mb.cache.non_zero_count[x264_scan8[idx]];\
208 if( !*nnz )\
209 bs_write_vlc( &h->out.bs, x264_coeff0_token[nC] );\
210 else\
211 *nnz = x264_cavlc_block_residual_internal(h,cat,l,nC);\
212 }
213
x264_cavlc_qp_delta(x264_t * h)214 static void x264_cavlc_qp_delta( x264_t *h )
215 {
216 bs_t *s = &h->out.bs;
217 int i_dqp = h->mb.i_qp - h->mb.i_last_qp;
218
219 /* Avoid writing a delta quant if we have an empty i16x16 block, e.g. in a completely
220 * flat background area. Don't do this if it would raise the quantizer, since that could
221 * cause unexpected deblocking artifacts. */
222 if( h->mb.i_type == I_16x16 && !(h->mb.i_cbp_luma | h->mb.i_cbp_chroma)
223 && !h->mb.cache.non_zero_count[x264_scan8[LUMA_DC]]
224 && !h->mb.cache.non_zero_count[x264_scan8[CHROMA_DC+0]]
225 && !h->mb.cache.non_zero_count[x264_scan8[CHROMA_DC+1]]
226 && h->mb.i_qp > h->mb.i_last_qp )
227 {
228 #if !RDO_SKIP_BS
229 h->mb.i_qp = h->mb.i_last_qp;
230 #endif
231 i_dqp = 0;
232 }
233
234 if( i_dqp )
235 {
236 if( i_dqp < -(QP_MAX_SPEC+1)/2 )
237 i_dqp += QP_MAX_SPEC+1;
238 else if( i_dqp > QP_MAX_SPEC/2 )
239 i_dqp -= QP_MAX_SPEC+1;
240 }
241 bs_write_se( s, i_dqp );
242 }
243
x264_cavlc_mvd(x264_t * h,int i_list,int idx,int width)244 static void x264_cavlc_mvd( x264_t *h, int i_list, int idx, int width )
245 {
246 bs_t *s = &h->out.bs;
247 ALIGNED_4( int16_t mvp[2] );
248 x264_mb_predict_mv( h, i_list, idx, width, mvp );
249 bs_write_se( s, h->mb.cache.mv[i_list][x264_scan8[idx]][0] - mvp[0] );
250 bs_write_se( s, h->mb.cache.mv[i_list][x264_scan8[idx]][1] - mvp[1] );
251 }
252
x264_cavlc_8x8_mvd(x264_t * h,int i)253 static inline void x264_cavlc_8x8_mvd( x264_t *h, int i )
254 {
255 switch( h->mb.i_sub_partition[i] )
256 {
257 case D_L0_8x8:
258 x264_cavlc_mvd( h, 0, 4*i, 2 );
259 break;
260 case D_L0_8x4:
261 x264_cavlc_mvd( h, 0, 4*i+0, 2 );
262 x264_cavlc_mvd( h, 0, 4*i+2, 2 );
263 break;
264 case D_L0_4x8:
265 x264_cavlc_mvd( h, 0, 4*i+0, 1 );
266 x264_cavlc_mvd( h, 0, 4*i+1, 1 );
267 break;
268 case D_L0_4x4:
269 x264_cavlc_mvd( h, 0, 4*i+0, 1 );
270 x264_cavlc_mvd( h, 0, 4*i+1, 1 );
271 x264_cavlc_mvd( h, 0, 4*i+2, 1 );
272 x264_cavlc_mvd( h, 0, 4*i+3, 1 );
273 break;
274 }
275 }
276
x264_cavlc_macroblock_luma_residual(x264_t * h,int plane_count)277 static ALWAYS_INLINE void x264_cavlc_macroblock_luma_residual( x264_t *h, int plane_count )
278 {
279 int p;
280
281 if( h->mb.b_transform_8x8 )
282 {
283 /* shuffle 8x8 dct coeffs into 4x4 lists */
284 for( p = 0; p < plane_count; p++ ) {
285 int i8;
286 for( i8 = 0; i8 < 4; i8++ )
287 if( h->mb.cache.non_zero_count[x264_scan8[p*16+i8*4]] )
288 h->zigzagf.interleave_8x8_cavlc( h->dct.luma4x4[p*16+i8*4], h->dct.luma8x8[p*4+i8],
289 &h->mb.cache.non_zero_count[x264_scan8[p*16+i8*4]] );
290 }
291 }
292
293 for( p = 0; p < plane_count; p++ ) {
294 int i8;
295 int msk;
296 int skip;
297
298 FOREACH_BIT( i8, 0, h->mb.i_cbp_luma ) {
299 int i4;
300 for( i4 = 0; i4 < 4; i4++ )
301 x264_cavlc_block_residual( h, DCT_LUMA_4x4, i4+i8*4+p*16, h->dct.luma4x4[i4+i8*4+p*16] );
302 }
303 }
304 }
305
x264_cavlc_partition_luma_residual(x264_t * h,int i8,int p)306 static ALWAYS_INLINE void x264_cavlc_partition_luma_residual( x264_t *h, int i8, int p )
307 {
308 if( h->mb.b_transform_8x8 && h->mb.cache.non_zero_count[x264_scan8[i8*4]] )
309 h->zigzagf.interleave_8x8_cavlc( h->dct.luma4x4[i8*4+p*16], h->dct.luma8x8[i8+p*4],
310 &h->mb.cache.non_zero_count[x264_scan8[i8*4+p*16]] );
311
312 if( h->mb.i_cbp_luma & (1 << i8) ) {
313 int i4;
314 for( i4 = 0; i4 < 4; i4++ )
315 x264_cavlc_block_residual( h, DCT_LUMA_4x4, i4+i8*4+p*16, h->dct.luma4x4[i4+i8*4+p*16] );
316 }
317 }
318
x264_cavlc_mb_header_i(x264_t * h,int i_mb_type,int i_mb_i_offset,int chroma)319 static void x264_cavlc_mb_header_i( x264_t *h, int i_mb_type, int i_mb_i_offset, int chroma )
320 {
321 bs_t *s = &h->out.bs;
322 if( i_mb_type == I_16x16 )
323 {
324 bs_write_ue( s, i_mb_i_offset + 1 + x264_mb_pred_mode16x16_fix[h->mb.i_intra16x16_pred_mode] +
325 h->mb.i_cbp_chroma * 4 + ( h->mb.i_cbp_luma == 0 ? 0 : 12 ) );
326 }
327 else //if( i_mb_type == I_4x4 || i_mb_type == I_8x8 )
328 {
329 int di = i_mb_type == I_8x8 ? 4 : 1;
330 int i;
331
332 bs_write_ue( s, i_mb_i_offset + 0 );
333 if( h->pps->b_transform_8x8_mode )
334 bs_write1( s, h->mb.b_transform_8x8 );
335
336 /* Prediction: Luma */
337 for( i = 0; i < 16; i += di )
338 {
339 int i_pred = x264_mb_predict_intra4x4_mode( h, i );
340 int i_mode = x264_mb_pred_mode4x4_fix( h->mb.cache.intra4x4_pred_mode[x264_scan8[i]] );
341
342 if( i_pred == i_mode )
343 bs_write1( s, 1 ); /* b_prev_intra4x4_pred_mode */
344 else
345 bs_write( s, 4, i_mode - (i_mode > i_pred) );
346 }
347
348 }
349 if( chroma )
350 bs_write_ue( s, x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode] );
351 }
352
x264_cavlc_mb_header_p(x264_t * h,int i_mb_type,int chroma)353 static ALWAYS_INLINE void x264_cavlc_mb_header_p( x264_t *h, int i_mb_type, int chroma )
354 {
355 bs_t *s = &h->out.bs;
356 if( i_mb_type == P_L0 )
357 {
358 if( h->mb.i_partition == D_16x16 )
359 {
360 bs_write1( s, 1 );
361
362 if( h->mb.pic.i_fref[0] > 1 )
363 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[0]] );
364 x264_cavlc_mvd( h, 0, 0, 4 );
365 }
366 else if( h->mb.i_partition == D_16x8 )
367 {
368 bs_write_ue( s, 1 );
369 if( h->mb.pic.i_fref[0] > 1 )
370 {
371 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[0]] );
372 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[8]] );
373 }
374 x264_cavlc_mvd( h, 0, 0, 4 );
375 x264_cavlc_mvd( h, 0, 8, 4 );
376 }
377 else if( h->mb.i_partition == D_8x16 )
378 {
379 bs_write_ue( s, 2 );
380 if( h->mb.pic.i_fref[0] > 1 )
381 {
382 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[0]] );
383 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[4]] );
384 }
385 x264_cavlc_mvd( h, 0, 0, 2 );
386 x264_cavlc_mvd( h, 0, 4, 2 );
387 }
388 }
389 else if( i_mb_type == P_8x8 )
390 {
391 int b_sub_ref;
392 int i2;
393
394 if( (h->mb.cache.ref[0][x264_scan8[0]] | h->mb.cache.ref[0][x264_scan8[ 4]] |
395 h->mb.cache.ref[0][x264_scan8[8]] | h->mb.cache.ref[0][x264_scan8[12]]) == 0 )
396 {
397 bs_write_ue( s, 4 );
398 b_sub_ref = 0;
399 }
400 else
401 {
402 bs_write_ue( s, 3 );
403 b_sub_ref = 1;
404 }
405
406 /* sub mb type */
407 if( h->param.analyse.inter & X264_ANALYSE_PSUB8x8 ) {
408 int i;
409 for( i = 0; i < 4; i++ )
410 bs_write_ue( s, subpartition_p_to_golomb[ h->mb.i_sub_partition[i] ] );
411 } else
412 bs_write( s, 4, 0xf );
413
414 /* ref0 */
415 if( b_sub_ref )
416 {
417 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[0]] );
418 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[4]] );
419 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[8]] );
420 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[12]] );
421 }
422
423 for( i2 = 0; i2 < 4; i2++ )
424 x264_cavlc_8x8_mvd( h, i2 );
425 }
426 else //if( IS_INTRA( i_mb_type ) )
427 x264_cavlc_mb_header_i( h, i_mb_type, 5, chroma );
428 }
429
x264_cavlc_mb_header_b(x264_t * h,int i_mb_type,int chroma)430 static ALWAYS_INLINE void x264_cavlc_mb_header_b( x264_t *h, int i_mb_type, int chroma )
431 {
432 bs_t *s = &h->out.bs;
433 if( i_mb_type == B_8x8 )
434 {
435 int i2;
436 bs_write_ue( s, 22 );
437
438 /* sub mb type */
439 for( i2 = 0; i2 < 4; i2++ )
440 bs_write_ue( s, subpartition_b_to_golomb[ h->mb.i_sub_partition[i2] ] );
441
442 /* ref */
443 if( h->mb.pic.i_fref[0] > 1 ) {
444 int i;
445 for( i = 0; i < 4; i++ )
446 if( x264_mb_partition_listX_table[0][ h->mb.i_sub_partition[i] ] )
447 bs_write_te( s, h->mb.pic.i_fref[0] - 1, h->mb.cache.ref[0][x264_scan8[i*4]] );
448 }
449 if( h->mb.pic.i_fref[1] > 1 ) {
450 for( i2 = 0; i2 < 4; i2++ )
451 if( x264_mb_partition_listX_table[1][ h->mb.i_sub_partition[i2] ] )
452 bs_write_te( s, h->mb.pic.i_fref[1] - 1, h->mb.cache.ref[1][x264_scan8[i2*4]] );
453 }
454 /* mvd */
455 for( i2 = 0; i2 < 4; i2++ ) {
456 if( x264_mb_partition_listX_table[0][ h->mb.i_sub_partition[i2] ] )
457 x264_cavlc_mvd( h, 0, 4*i2, 2 );
458 }
459 for( i2 = 0; i2 < 4; i2++ )
460 if( x264_mb_partition_listX_table[1][ h->mb.i_sub_partition[i2] ] )
461 x264_cavlc_mvd( h, 1, 4*i2, 2 );
462 }
463 else if( i_mb_type >= B_L0_L0 && i_mb_type <= B_BI_BI )
464 {
465 /* All B mode */
466 /* Motion Vector */
467 const uint8_t (*b_list)[2] = x264_mb_type_list_table[i_mb_type];
468 const int i_ref0_max = h->mb.pic.i_fref[0] - 1;
469 const int i_ref1_max = h->mb.pic.i_fref[1] - 1;
470
471 bs_write_ue( s, mb_type_b_to_golomb[ h->mb.i_partition - D_16x8 ][ i_mb_type - B_L0_L0 ] );
472 if( h->mb.i_partition == D_16x16 )
473 {
474 if( i_ref0_max && b_list[0][0] ) bs_write_te( s, i_ref0_max, h->mb.cache.ref[0][x264_scan8[0]] );
475 if( i_ref1_max && b_list[1][0] ) bs_write_te( s, i_ref1_max, h->mb.cache.ref[1][x264_scan8[0]] );
476 if( b_list[0][0] ) x264_cavlc_mvd( h, 0, 0, 4 );
477 if( b_list[1][0] ) x264_cavlc_mvd( h, 1, 0, 4 );
478 }
479 else
480 {
481 if( i_ref0_max && b_list[0][0] ) bs_write_te( s, i_ref0_max, h->mb.cache.ref[0][x264_scan8[ 0]] );
482 if( i_ref0_max && b_list[0][1] ) bs_write_te( s, i_ref0_max, h->mb.cache.ref[0][x264_scan8[12]] );
483 if( i_ref1_max && b_list[1][0] ) bs_write_te( s, i_ref1_max, h->mb.cache.ref[1][x264_scan8[ 0]] );
484 if( i_ref1_max && b_list[1][1] ) bs_write_te( s, i_ref1_max, h->mb.cache.ref[1][x264_scan8[12]] );
485 if( h->mb.i_partition == D_16x8 )
486 {
487 if( b_list[0][0] ) x264_cavlc_mvd( h, 0, 0, 4 );
488 if( b_list[0][1] ) x264_cavlc_mvd( h, 0, 8, 4 );
489 if( b_list[1][0] ) x264_cavlc_mvd( h, 1, 0, 4 );
490 if( b_list[1][1] ) x264_cavlc_mvd( h, 1, 8, 4 );
491 }
492 else //if( h->mb.i_partition == D_8x16 )
493 {
494 if( b_list[0][0] ) x264_cavlc_mvd( h, 0, 0, 2 );
495 if( b_list[0][1] ) x264_cavlc_mvd( h, 0, 4, 2 );
496 if( b_list[1][0] ) x264_cavlc_mvd( h, 1, 0, 2 );
497 if( b_list[1][1] ) x264_cavlc_mvd( h, 1, 4, 2 );
498 }
499 }
500 }
501 else if( i_mb_type == B_DIRECT )
502 bs_write1( s, 1 );
503 else //if( IS_INTRA( i_mb_type ) )
504 x264_cavlc_mb_header_i( h, i_mb_type, 23, chroma );
505 }
506
507 /*****************************************************************************
508 * x264_macroblock_write:
509 *****************************************************************************/
x264_macroblock_write_cavlc(x264_t * h)510 void x264_macroblock_write_cavlc( x264_t *h )
511 {
512 bs_t *s = &h->out.bs;
513 const int i_mb_type = h->mb.i_type;
514 int plane_count = CHROMA444 ? 3 : 1;
515 int chroma = !CHROMA444;
516
517 #if RDO_SKIP_BS
518 s->i_bits_encoded = 0;
519 #else
520 const int i_mb_pos_start = bs_pos( s );
521 int i_mb_pos_tex;
522 #endif
523
524 if( SLICE_MBAFF
525 && (!(h->mb.i_mb_y & 1) || IS_SKIP(h->mb.type[h->mb.i_mb_xy - h->mb.i_mb_stride])) )
526 {
527 bs_write1( s, MB_INTERLACED );
528 #if !RDO_SKIP_BS
529 h->mb.field_decoding_flag = MB_INTERLACED;
530 #endif
531 }
532
533 #if !RDO_SKIP_BS
534 if( i_mb_type == I_PCM )
535 {
536 static const uint8_t i_offsets[3] = {5,23,0};
537 uint8_t *p_start = s->p_start;
538 int p;
539
540 bs_write_ue( s, i_offsets[h->sh.i_type] + 25 );
541 i_mb_pos_tex = bs_pos( s );
542 h->stat.frame.i_mv_bits += i_mb_pos_tex - i_mb_pos_start;
543
544 bs_align_0( s );
545
546 for( p = 0; p < plane_count; p++ ) {
547 int i;
548 for( i = 0; i < 256; i++ )
549 bs_write( s, BIT_DEPTH, h->mb.pic.p_fenc[p][i] );
550 }
551 if( chroma ) {
552 int ch;
553 for( ch = 1; ch < 3; ch++ ) {
554 int i;
555 for( i = 0; i < 16>>CHROMA_V_SHIFT; i++ ) {
556 int j;
557 for( j = 0; j < 8; j++ )
558 bs_write( s, BIT_DEPTH, h->mb.pic.p_fenc[ch][i*FENC_STRIDE+j] );
559 }
560 }
561
562 bs_init( s, s->p, s->p_end - s->p );
563 }
564 s->p_start = p_start;
565
566 h->stat.frame.i_tex_bits += bs_pos(s) - i_mb_pos_tex;
567 return;
568 }
569 #endif
570
571 if( h->sh.i_type == SLICE_TYPE_P )
572 x264_cavlc_mb_header_p( h, i_mb_type, chroma );
573 else if( h->sh.i_type == SLICE_TYPE_B )
574 x264_cavlc_mb_header_b( h, i_mb_type, chroma );
575 else //if( h->sh.i_type == SLICE_TYPE_I )
576 x264_cavlc_mb_header_i( h, i_mb_type, 0, chroma );
577
578 #if !RDO_SKIP_BS
579 i_mb_pos_tex = bs_pos( s );
580 h->stat.frame.i_mv_bits += i_mb_pos_tex - i_mb_pos_start;
581 #endif
582
583 /* Coded block pattern */
584 if( i_mb_type != I_16x16 )
585 bs_write_ue( s, cbp_to_golomb[chroma][IS_INTRA(i_mb_type)][(h->mb.i_cbp_chroma << 4)|h->mb.i_cbp_luma] );
586
587 /* transform size 8x8 flag */
588 if( x264_mb_transform_8x8_allowed( h ) && h->mb.i_cbp_luma )
589 bs_write1( s, h->mb.b_transform_8x8 );
590
591 if( i_mb_type == I_16x16 )
592 {
593 int p;
594 x264_cavlc_qp_delta( h );
595
596 /* DC Luma */
597 for( p = 0; p < plane_count; p++ )
598 {
599 x264_cavlc_block_residual( h, DCT_LUMA_DC, LUMA_DC+p, h->dct.luma16x16_dc[p] );
600
601 /* AC Luma */
602 if( h->mb.i_cbp_luma ) {
603 int i;
604 for( i = p*16; i < p*16+16; i++ )
605 x264_cavlc_block_residual( h, DCT_LUMA_AC, i, h->dct.luma4x4[i]+1 );
606 }
607 }
608 }
609 else if( h->mb.i_cbp_luma | h->mb.i_cbp_chroma )
610 {
611 x264_cavlc_qp_delta( h );
612 x264_cavlc_macroblock_luma_residual( h, plane_count );
613 }
614 if( h->mb.i_cbp_chroma )
615 {
616 /* Chroma DC residual present */
617 x264_cavlc_block_residual( h, DCT_CHROMA_DC, CHROMA_DC+0, h->dct.chroma_dc[0] );
618 x264_cavlc_block_residual( h, DCT_CHROMA_DC, CHROMA_DC+1, h->dct.chroma_dc[1] );
619 if( h->mb.i_cbp_chroma == 2 ) /* Chroma AC residual present */
620 {
621 int step = 8 << CHROMA_V_SHIFT;
622 int i;
623 int j;
624 for( i = 16; i < 3*16; i += step )
625 for( j = i; j < i+4; j++ )
626 x264_cavlc_block_residual( h, DCT_CHROMA_AC, j, h->dct.luma4x4[j]+1 );
627 }
628 }
629
630 #if !RDO_SKIP_BS
631 h->stat.frame.i_tex_bits += bs_pos(s) - i_mb_pos_tex;
632 #endif
633 }
634
635 #if RDO_SKIP_BS
636 /*****************************************************************************
637 * RD only; doesn't generate a valid bitstream
638 * doesn't write cbp or chroma dc (I don't know how much this matters)
639 * doesn't write ref (never varies between calls, so no point in doing so)
640 * only writes subpartition for p8x8, needed for sub-8x8 mode decision RDO
641 * works on all partition sizes except 16x16
642 *****************************************************************************/
x264_partition_size_cavlc(x264_t * h,int i8,int i_pixel)643 static int x264_partition_size_cavlc( x264_t *h, int i8, int i_pixel )
644 {
645 bs_t *s = &h->out.bs;
646 const int i_mb_type = h->mb.i_type;
647 int b_8x16 = h->mb.i_partition == D_8x16;
648 int plane_count = CHROMA444 ? 3 : 1;
649 int j;
650
651 if( i_mb_type == P_8x8 )
652 {
653 x264_cavlc_8x8_mvd( h, i8 );
654 bs_write_ue( s, subpartition_p_to_golomb[ h->mb.i_sub_partition[i8] ] );
655 }
656 else if( i_mb_type == P_L0 )
657 x264_cavlc_mvd( h, 0, 4*i8, 4>>b_8x16 );
658 else if( i_mb_type > B_DIRECT && i_mb_type < B_8x8 )
659 {
660 if( x264_mb_type_list_table[ i_mb_type ][0][!!i8] ) x264_cavlc_mvd( h, 0, 4*i8, 4>>b_8x16 );
661 if( x264_mb_type_list_table[ i_mb_type ][1][!!i8] ) x264_cavlc_mvd( h, 1, 4*i8, 4>>b_8x16 );
662 }
663 else //if( i_mb_type == B_8x8 )
664 {
665 if( x264_mb_partition_listX_table[0][ h->mb.i_sub_partition[i8] ] )
666 x264_cavlc_mvd( h, 0, 4*i8, 2 );
667 if( x264_mb_partition_listX_table[1][ h->mb.i_sub_partition[i8] ] )
668 x264_cavlc_mvd( h, 1, 4*i8, 2 );
669 }
670
671 for( j = (i_pixel < PIXEL_8x8); j >= 0; j-- )
672 {
673 int p;
674 for( p = 0; p < plane_count; p++ )
675 x264_cavlc_partition_luma_residual( h, i8, p );
676 if( h->mb.i_cbp_chroma )
677 {
678 if( CHROMA_FORMAT == CHROMA_422 )
679 {
680 int offset = (5*i8) & 0x09;
681 x264_cavlc_block_residual( h, DCT_CHROMA_AC, 16+offset, h->dct.luma4x4[16+offset]+1 );
682 x264_cavlc_block_residual( h, DCT_CHROMA_AC, 18+offset, h->dct.luma4x4[18+offset]+1 );
683 x264_cavlc_block_residual( h, DCT_CHROMA_AC, 32+offset, h->dct.luma4x4[32+offset]+1 );
684 x264_cavlc_block_residual( h, DCT_CHROMA_AC, 34+offset, h->dct.luma4x4[34+offset]+1 );
685 }
686 else
687 {
688 x264_cavlc_block_residual( h, DCT_CHROMA_AC, 16+i8, h->dct.luma4x4[16+i8]+1 );
689 x264_cavlc_block_residual( h, DCT_CHROMA_AC, 32+i8, h->dct.luma4x4[32+i8]+1 );
690 }
691 }
692 i8 += x264_pixel_size[i_pixel].h >> 3;
693 }
694
695 return h->out.bs.i_bits_encoded;
696 }
697
x264_subpartition_size_cavlc(x264_t * h,int i4,int i_pixel)698 static int x264_subpartition_size_cavlc( x264_t *h, int i4, int i_pixel )
699 {
700 int plane_count = CHROMA444 ? 3 : 1;
701 int b_8x4 = i_pixel == PIXEL_8x4;
702 int p;
703
704 h->out.bs.i_bits_encoded = 0;
705 x264_cavlc_mvd( h, 0, i4, 1+b_8x4 );
706 for( p = 0; p < plane_count; p++ )
707 {
708 x264_cavlc_block_residual( h, DCT_LUMA_4x4, p*16+i4, h->dct.luma4x4[p*16+i4] );
709 if( i_pixel != PIXEL_4x4 )
710 x264_cavlc_block_residual( h, DCT_LUMA_4x4, p*16+i4+2-b_8x4, h->dct.luma4x4[p*16+i4+2-b_8x4] );
711 }
712
713 return h->out.bs.i_bits_encoded;
714 }
715
x264_cavlc_intra4x4_pred_size(x264_t * h,int i4,int i_mode)716 static int x264_cavlc_intra4x4_pred_size( x264_t *h, int i4, int i_mode )
717 {
718 if( x264_mb_predict_intra4x4_mode( h, i4 ) == x264_mb_pred_mode4x4_fix( i_mode ) )
719 return 1;
720 else
721 return 4;
722 }
723
x264_partition_i8x8_size_cavlc(x264_t * h,int i8,int i_mode)724 static int x264_partition_i8x8_size_cavlc( x264_t *h, int i8, int i_mode )
725 {
726 int plane_count = CHROMA444 ? 3 : 1;
727 int p;
728
729 h->out.bs.i_bits_encoded = x264_cavlc_intra4x4_pred_size( h, 4*i8, i_mode );
730 bs_write_ue( &h->out.bs, cbp_to_golomb[!CHROMA444][1][(h->mb.i_cbp_chroma << 4)|h->mb.i_cbp_luma] );
731 for( p = 0; p < plane_count; p++ )
732 x264_cavlc_partition_luma_residual( h, i8, p );
733 return h->out.bs.i_bits_encoded;
734 }
735
x264_partition_i4x4_size_cavlc(x264_t * h,int i4,int i_mode)736 static int x264_partition_i4x4_size_cavlc( x264_t *h, int i4, int i_mode )
737 {
738 int plane_count = CHROMA444 ? 3 : 1;
739 int p;
740
741 h->out.bs.i_bits_encoded = x264_cavlc_intra4x4_pred_size( h, i4, i_mode );
742 for( p = 0; p < plane_count; p++ )
743 x264_cavlc_block_residual( h, DCT_LUMA_4x4, p*16+i4, h->dct.luma4x4[p*16+i4] );
744 return h->out.bs.i_bits_encoded;
745 }
746
x264_chroma_size_cavlc(x264_t * h)747 static int x264_chroma_size_cavlc( x264_t *h )
748 {
749 h->out.bs.i_bits_encoded = bs_size_ue( x264_mb_chroma_pred_mode_fix[h->mb.i_chroma_pred_mode] );
750 if( h->mb.i_cbp_chroma )
751 {
752 x264_cavlc_block_residual( h, DCT_CHROMA_DC, CHROMA_DC+0, h->dct.chroma_dc[0] );
753 x264_cavlc_block_residual( h, DCT_CHROMA_DC, CHROMA_DC+1, h->dct.chroma_dc[1] );
754
755 if( h->mb.i_cbp_chroma == 2 )
756 {
757 int step = 8 << CHROMA_V_SHIFT;
758 int i;
759 int j;
760 for( i = 16; i < 3*16; i += step )
761 for( j = i; j < i+4; j++ )
762 x264_cavlc_block_residual( h, DCT_CHROMA_AC, j, h->dct.luma4x4[j]+1 );
763 }
764 }
765 return h->out.bs.i_bits_encoded;
766 }
767 #endif
768