· 7 years ago · Apr 08, 2018, 01:24 AM
1/*****************************************************************************
2 * common.h: misc common functions
3 *****************************************************************************
4 * Copyright (C) 2003-2011 x264 project
5 *
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
8 *
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.
13 *
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.
18 *
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., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
22 *
23 * This program is also available under a commercial proprietary license.
24 * For more information, contact us at licensing@x264.com.
25 *****************************************************************************/
26
27#ifndef X264_COMMON_H
28#define X264_COMMON_H
29
30/****************************************************************************
31 * Macros
32 ****************************************************************************/
33#define X264_MIN(a,b) ( (a)<(b) ? (a) : (b) )
34#define X264_MAX(a,b) ( (a)>(b) ? (a) : (b) )
35#define X264_MIN3(a,b,c) X264_MIN((a),X264_MIN((b),(c)))
36#define X264_MAX3(a,b,c) X264_MAX((a),X264_MAX((b),(c)))
37#define X264_MIN4(a,b,c,d) X264_MIN((a),X264_MIN3((b),(c),(d)))
38#define X264_MAX4(a,b,c,d) X264_MAX((a),X264_MAX3((b),(c),(d)))
39#define XCHG(type,a,b) do{ type t = a; a = b; b = t; } while(0)
40#define IS_DISPOSABLE(type) ( type == X264_TYPE_B )
41#define FIX8(f) ((int)(f*(1<<8)+.5))
42#define ALIGN(x,a) (((x)+((a)-1))&~((a)-1))
43
44#define CHECKED_MALLOC( var, size )\
45do {\
46 var = x264_malloc( size );\
47 if( !var )\
48 goto fail;\
49} while( 0 )
50#define CHECKED_MALLOCZERO( var, size )\
51do {\
52 CHECKED_MALLOC( var, size );\
53 memset( var, 0, size );\
54} while( 0 )
55
56#define X264_BFRAME_MAX 16
57#define X264_REF_MAX 16
58#define X264_THREAD_MAX 128
59#define X264_PCM_COST (FRAME_SIZE(256*BIT_DEPTH)+16)
60#define X264_LOOKAHEAD_MAX 250
61#define QP_BD_OFFSET (6*(BIT_DEPTH-8))
62#define QP_MAX_SPEC_H264 (51+QP_BD_OFFSET)
63#define QP_MAX_SPEC_MPEG2 31
64#define QP_MAX_SPEC (MPEG2 ? QP_MAX_SPEC_MPEG2 : QP_MAX_SPEC_H264)
65#define QP_MAX (QP_MAX_SPEC_H264+18)
66#define QP_MAX_MAX (51+2*6+18)
67#define LAMBDA_MAX (91 << (BIT_DEPTH-8))
68#define PIXEL_MAX ((1 << BIT_DEPTH)-1)
69// arbitrary, but low because SATD scores are 1/4 normal
70#define X264_LOOKAHEAD_QP (12+QP_BD_OFFSET)
71#define SPEC_QP(x) X264_MIN((x), QP_MAX_SPEC)
72
73// number of pixels (per thread) in progress at any given time.
74// 16 for the macroblock in progress + 3 for deblocking + 3 for motion compensation filter + 2 for extra safety
75#define X264_THREAD_HEIGHT 24
76
77/* WEIGHTP_FAKE is set when mb_tree & psy are enabled, but normal weightp is disabled
78 * (such as in baseline). It checks for fades in lookahead and adjusts qp accordingly
79 * to increase quality. Defined as (-1) so that if(i_weighted_pred > 0) is true only when
80 * real weights are being used. */
81
82#define X264_WEIGHTP_FAKE (-1)
83
84#define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
85#define FILLER_OVERHEAD (NALU_OVERHEAD+1)
86#define STRUCTURE_OVERHEAD 4 // startcode
87
88#define LOG2_16(x) (31 - x264_clz((x)|1))
89
90/****************************************************************************
91 * Includes
92 ****************************************************************************/
93#include "osdep.h"
94#include <stdarg.h>
95#include <stddef.h>
96#include <stdlib.h>
97#include <string.h>
98#include <assert.h>
99#include <limits.h>
100
101#if HAVE_INTERLACED
102# define MB_INTERLACED h->mb.b_interlaced
103# define SLICE_MBAFF h->sh.b_mbaff
104# define PARAM_INTERLACED h->param.b_interlaced
105#else
106# define MB_INTERLACED 0
107# define SLICE_MBAFF 0
108# define PARAM_INTERLACED 0
109#endif
110
111/* MPEG-2 Support */
112#if HAVE_MPEG2
113# define MPEG2 h->param.b_mpeg2
114#else
115# define MPEG2 0
116#endif
117
118#ifdef CHROMA_FORMAT
119# define CHROMA_H_SHIFT (CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422)
120# define CHROMA_V_SHIFT (CHROMA_FORMAT == CHROMA_420)
121#else
122# define CHROMA_FORMAT h->sps->i_chroma_format_idc
123# define CHROMA_H_SHIFT h->mb.chroma_h_shift
124# define CHROMA_V_SHIFT h->mb.chroma_v_shift
125#endif
126
127#define CHROMA_SIZE(s) ((s)>>(CHROMA_H_SHIFT+CHROMA_V_SHIFT))
128#define FRAME_SIZE(s) ((s)+2*CHROMA_SIZE(s))
129#define CHROMA444 (CHROMA_FORMAT == CHROMA_444)
130
131/* Unions for type-punning.
132 * Mn: load or store n bits, aligned, native-endian
133 * CPn: copy n bits, aligned, native-endian
134 * we don't use memcpy for CPn because memcpy's args aren't assumed to be aligned */
135typedef union { uint16_t i; uint8_t c[2]; } MAY_ALIAS x264_union16_t;
136typedef union { uint32_t i; uint16_t b[2]; uint8_t c[4]; } MAY_ALIAS x264_union32_t;
137typedef union { uint64_t i; uint32_t a[2]; uint16_t b[4]; uint8_t c[8]; } MAY_ALIAS x264_union64_t;
138typedef struct { uint64_t i[2]; } x264_uint128_t;
139typedef union { x264_uint128_t i; uint64_t a[2]; uint32_t b[4]; uint16_t c[8]; uint8_t d[16]; } MAY_ALIAS x264_union128_t;
140#define M16(src) (((x264_union16_t*)(src))->i)
141#define M32(src) (((x264_union32_t*)(src))->i)
142#define M64(src) (((x264_union64_t*)(src))->i)
143#define M128(src) (((x264_union128_t*)(src))->i)
144#define M128_ZERO ((x264_uint128_t){{0,0}})
145#define CP16(dst,src) M16(dst) = M16(src)
146#define CP32(dst,src) M32(dst) = M32(src)
147#define CP64(dst,src) M64(dst) = M64(src)
148#define CP128(dst,src) M128(dst) = M128(src)
149
150#if HIGH_BIT_DEPTH
151 typedef uint16_t pixel;
152 typedef uint64_t pixel4;
153 typedef int32_t dctcoef;
154 typedef uint32_t udctcoef;
155
156# define PIXEL_SPLAT_X4(x) ((x)*0x0001000100010001ULL)
157# define MPIXEL_X4(src) M64(src)
158#else
159 typedef uint8_t pixel;
160 typedef uint32_t pixel4;
161 typedef int16_t dctcoef;
162 typedef uint16_t udctcoef;
163
164# define PIXEL_SPLAT_X4(x) ((x)*0x01010101U)
165# define MPIXEL_X4(src) M32(src)
166#endif
167
168#define BIT_DEPTH X264_BIT_DEPTH
169
170#define CPPIXEL_X4(dst,src) MPIXEL_X4(dst) = MPIXEL_X4(src)
171
172#define X264_SCAN8_LUMA_SIZE (5*8)
173#define X264_SCAN8_SIZE (X264_SCAN8_LUMA_SIZE*3)
174#define X264_SCAN8_0 (4+1*8)
175
176/* Scan8 organization:
177 * 0 1 2 3 4 5 6 7
178 * 0 DY y y y y y
179 * 1 y Y Y Y Y
180 * 2 y Y Y Y Y
181 * 3 y Y Y Y Y
182 * 4 y Y Y Y Y
183 * 5 DU u u u u u
184 * 6 u U U U U
185 * 7 u U U U U
186 * 8 u U U U U
187 * 9 u U U U U
188 * 10 DV v v v v v
189 * 11 v V V V V
190 * 12 v V V V V
191 * 13 v V V V V
192 * 14 v V V V V
193 * DY/DU/DV are for luma/chroma DC.
194 */
195
196#define LUMA_DC 48
197#define CHROMA_DC 49
198
199static const uint8_t x264_scan8[16*3 + 3] =
200{
201 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8,
202 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8,
203 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8,
204 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8,
205 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8,
206 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8,
207 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8,
208 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8,
209 4+11*8, 5+11*8, 4+12*8, 5+12*8,
210 6+11*8, 7+11*8, 6+12*8, 7+12*8,
211 4+13*8, 5+13*8, 4+14*8, 5+14*8,
212 6+13*8, 7+13*8, 6+14*8, 7+14*8,
213 0+ 0*8, 0+ 5*8, 0+10*8
214};
215
216#include "x264.h"
217#include "bitstream.h"
218#include "set.h"
219#include "predict.h"
220#include "pixel.h"
221#include "mc.h"
222#include "frame.h"
223#include "dct.h"
224#include "cabac.h"
225#include "quant.h"
226#include "cpu.h"
227#include "threadpool.h"
228
229/****************************************************************************
230 * General functions
231 ****************************************************************************/
232/* x264_malloc : will do or emulate a memalign
233 * you have to use x264_free for buffers allocated with x264_malloc */
234void *x264_malloc( int );
235void x264_free( void * );
236
237/* x264_slurp_file: malloc space for the whole file and read it */
238char *x264_slurp_file( const char *filename );
239
240/* mdate: return the current date in microsecond */
241int64_t x264_mdate( void );
242
243/* x264_param2string: return a (malloced) string containing most of
244 * the encoding options */
245char *x264_param2string( x264_param_t *p, int b_res );
246
247/* log */
248void x264_log( x264_t *h, int i_level, const char *psz_fmt, ... );
249
250void x264_reduce_fraction( uint32_t *n, uint32_t *d );
251void x264_reduce_fraction64( uint64_t *n, uint64_t *d );
252void x264_cavlc_init( void );
253void x264_cabac_init( x264_t *h );
254
255static ALWAYS_INLINE pixel x264_clip_pixel( int x )
256{
257 return ( (x & ~PIXEL_MAX) ? (-x)>>31 & PIXEL_MAX : x );
258}
259
260static ALWAYS_INLINE int x264_clip3( int v, int i_min, int i_max )
261{
262 return ( (v < i_min) ? i_min : (v > i_max) ? i_max : v );
263}
264
265static ALWAYS_INLINE double x264_clip3f( double v, double f_min, double f_max )
266{
267 return ( (v < f_min) ? f_min : (v > f_max) ? f_max : v );
268}
269
270static ALWAYS_INLINE int x264_median( int a, int b, int c )
271{
272 int t = (a-b)&((a-b)>>31);
273 a -= t;
274 b += t;
275 b -= (b-c)&((b-c)>>31);
276 b += (a-b)&((a-b)>>31);
277 return b;
278}
279
280static ALWAYS_INLINE void x264_median_mv( int16_t *dst, int16_t *a, int16_t *b, int16_t *c )
281{
282 dst[0] = x264_median( a[0], b[0], c[0] );
283 dst[1] = x264_median( a[1], b[1], c[1] );
284}
285
286static ALWAYS_INLINE int x264_predictor_difference( int16_t (*mvc)[2], intptr_t i_mvc )
287{
288 int sum = 0;
289 for( int i = 0; i < i_mvc-1; i++ )
290 {
291 sum += abs( mvc[i][0] - mvc[i+1][0] )
292 + abs( mvc[i][1] - mvc[i+1][1] );
293 }
294 return sum;
295}
296
297static ALWAYS_INLINE uint16_t x264_cabac_mvd_sum( uint8_t *mvdleft, uint8_t *mvdtop )
298{
299 int amvd0 = abs(mvdleft[0]) + abs(mvdtop[0]);
300 int amvd1 = abs(mvdleft[1]) + abs(mvdtop[1]);
301 amvd0 = (amvd0 > 2) + (amvd0 > 32);
302 amvd1 = (amvd1 > 2) + (amvd1 > 32);
303 return amvd0 + (amvd1<<8);
304}
305
306static void ALWAYS_INLINE x264_predictor_roundclip( int16_t (*dst)[2], int16_t (*mvc)[2], int i_mvc, int mv_x_min, int mv_x_max, int mv_y_min, int mv_y_max )
307{
308 for( int i = 0; i < i_mvc; i++ )
309 {
310 int mx = (mvc[i][0] + 2) >> 2;
311 int my = (mvc[i][1] + 2) >> 2;
312 dst[i][0] = x264_clip3( mx, mv_x_min, mv_x_max );
313 dst[i][1] = x264_clip3( my, mv_y_min, mv_y_max );
314 }
315}
316
317extern const uint8_t x264_exp2_lut[64];
318extern const float x264_log2_lut[128];
319extern const float x264_log2_lz_lut[32];
320
321/* Not a general-purpose function; multiplies input by -1/6 to convert
322 * qp to qscale. */
323static ALWAYS_INLINE int x264_exp2fix8( float x )
324{
325 int i = x*(-64.f/6.f) + 512.5f;
326 if( i < 0 ) return 0;
327 if( i > 1023 ) return 0xffff;
328 return (x264_exp2_lut[i&63]+256) << (i>>6) >> 8;
329}
330
331static ALWAYS_INLINE float x264_log2( uint32_t x )
332{
333 int lz = x264_clz( x );
334 return x264_log2_lut[(x<<lz>>24)&0x7f] + x264_log2_lz_lut[lz];
335}
336
337/****************************************************************************
338 *
339 ****************************************************************************/
340enum slice_type_e
341{
342 SLICE_TYPE_P = 0,
343 SLICE_TYPE_B = 1,
344 SLICE_TYPE_I = 2,
345};
346
347static const char slice_type_to_char[] = { 'P', 'B', 'I' };
348
349enum sei_payload_type_e
350{
351 SEI_BUFFERING_PERIOD = 0,
352 SEI_PIC_TIMING = 1,
353 SEI_PAN_SCAN_RECT = 2,
354 SEI_FILLER = 3,
355 SEI_USER_DATA_REGISTERED = 4,
356 SEI_USER_DATA_UNREGISTERED = 5,
357 SEI_RECOVERY_POINT = 6,
358 SEI_DEC_REF_PIC_MARKING = 7,
359 SEI_FRAME_PACKING = 45,
360};
361
362enum mpeg2_start_code_e
363{
364 MPEG2_PICTURE_START_CODE = 0x00,
365 MPEG2_USER_DATA_START_CODE = 0xB2,
366 MPEG2_SEQ_HEADER_CODE = 0xB3,
367 MPEG2_SEQ_ERROR_CODE = 0xB4,
368 MPEG2_EXT_START_CODE = 0xB5,
369 MPEG2_SEQ_END_CODE = 0xB7,
370 MPEG2_GRP_START_CODE = 0xB8,
371};
372
373static const uint8_t structure_to_start_code[] =
374{
375 [MPEG2_SEQ_HEADER] = MPEG2_SEQ_HEADER_CODE,
376 [MPEG2_USER_DATA] = MPEG2_USER_DATA_START_CODE,
377 [MPEG2_SEQ_EXT] = MPEG2_EXT_START_CODE,
378 [MPEG2_SEQ_DISPLAY_EXT] = MPEG2_EXT_START_CODE,
379 [MPEG2_GOP_HEADER] = MPEG2_GRP_START_CODE,
380 [MPEG2_PICTURE_HEADER] = MPEG2_PICTURE_START_CODE,
381 [MPEG2_PICTURE_CODING_EXT] = MPEG2_EXT_START_CODE,
382 [MPEG2_PICTURE_DISPLAY_EXT] = MPEG2_EXT_START_CODE,
383 [MPEG2_QUANT_MATRIX_EXT] = MPEG2_EXT_START_CODE,
384 [MPEG2_COPYRIGHT_EXT] = MPEG2_EXT_START_CODE,
385};
386
387enum mpeg2_extension_id_e
388{
389 MPEG2_SEQ_EXT_ID = 1,
390 MPEG2_SEQ_DISPLAY_EXT_ID = 2,
391 MPEG2_QUANT_MATRIX_EXT_ID = 3,
392 MPEG2_COPYRIGHT_EXT_ID = 4,
393 MPEG2_PIC_DISPLAY_EXT_ID = 7,
394 MPEG2_PIC_CODING_EXT_ID = 8,
395};
396
397typedef struct
398{
399 x264_sps_t *sps;
400 x264_pps_t *pps;
401
402 int i_type;
403 int i_first_mb;
404 int i_last_mb;
405
406 int i_pps_id;
407
408 int i_frame_num;
409
410 int b_mbaff;
411 int b_field_pic;
412 int b_bottom_field;
413
414 int i_idr_pic_id; /* -1 if nal_type != 5 */
415
416 int i_poc;
417 int i_delta_poc_bottom;
418
419 int i_delta_poc[2];
420 int i_redundant_pic_cnt;
421
422 int b_direct_spatial_mv_pred;
423
424 int b_num_ref_idx_override;
425 int i_num_ref_idx_l0_active;
426 int i_num_ref_idx_l1_active;
427
428 int b_ref_pic_list_reordering[2];
429 struct
430 {
431 int idc;
432 int arg;
433 } ref_pic_list_order[2][X264_REF_MAX];
434
435 /* P-frame weighting */
436 x264_weight_t weight[X264_REF_MAX*2][3];
437
438 int i_mmco_remove_from_end;
439 int i_mmco_command_count;
440 struct /* struct for future expansion */
441 {
442 int i_difference_of_pic_nums;
443 int i_poc;
444 } mmco[X264_REF_MAX];
445
446 int i_cabac_init_idc;
447
448 int i_qp;
449 int i_qp_delta;
450 int b_sp_for_swidth;
451 int i_qs_delta;
452
453 /* deblocking filter */
454 int i_disable_deblocking_filter_idc;
455 int i_alpha_c0_offset;
456 int i_beta_offset;
457
458} x264_slice_header_t;
459
460typedef struct x264_lookahead_t
461{
462 volatile uint8_t b_exit_thread;
463 uint8_t b_thread_active;
464 uint8_t b_analyse_keyframe;
465 int i_last_keyframe;
466 int i_slicetype_length;
467 x264_frame_t *last_nonb;
468 x264_pthread_t thread_handle;
469 x264_sync_frame_list_t ifbuf;
470 x264_sync_frame_list_t next;
471 x264_sync_frame_list_t ofbuf;
472} x264_lookahead_t;
473
474typedef struct x264_ratecontrol_t x264_ratecontrol_t;
475
476typedef struct x264_left_table_t
477{
478 uint8_t intra[4];
479 uint8_t nnz[4];
480 uint8_t nnz_chroma[4];
481 uint8_t mv[4];
482 uint8_t ref[4];
483} x264_left_table_t;
484
485struct x264_t
486{
487 /* encoder parameters */
488 x264_param_t param;
489
490 x264_t *thread[X264_THREAD_MAX+1];
491 int b_thread_active;
492 int i_thread_phase; /* which thread to use for the next frame */
493 int i_threadslice_start; /* first row in this thread slice */
494 int i_threadslice_end; /* row after the end of this thread slice */
495 x264_threadpool_t *threadpool;
496
497 /* bitstream output */
498 struct
499 {
500 int i_nal;
501 int i_nals_allocated;
502 x264_nal_t *nal;
503 int i_bitstream; /* size of p_bitstream */
504 uint8_t *p_bitstream; /* will hold data for all nal */
505 bs_t bs;
506 } out;
507
508 uint8_t *nal_buffer;
509 int nal_buffer_size;
510
511 /**** thread synchronization starts here ****/
512
513 /* frame number/poc */
514 int i_frame;
515 int i_frame_num;
516
517 int i_thread_frames; /* Number of different frames being encoded by threads;
518 * 1 when sliced-threads is on. */
519 int i_nal_type;
520 int i_nal_ref_idc;
521
522 int64_t i_disp_fields; /* Number of displayed fields (both coded and implied via pic_struct) */
523 int i_disp_fields_last_frame;
524 int64_t i_prev_duration; /* Duration of previous frame */
525 int64_t i_coded_fields; /* Number of coded fields (both coded and implied via pic_struct) */
526 int64_t i_cpb_delay; /* Equal to number of fields preceding this field
527 * since last buffering_period SEI */
528 int64_t i_coded_fields_lookahead; /* Use separate counters for lookahead */
529 int64_t i_cpb_delay_lookahead;
530
531 int64_t i_cpb_delay_pir_offset;
532
533 int b_queued_intra_refresh;
534 int64_t i_last_idr_pts;
535
536 int i_idr_pic_id;
537
538 /* quantization matrix for decoding, [cqm][qp%6][coef] */
539 int (*dequant4_mf[4])[16]; /* [4][6][16] */
540 int (*dequant8_mf[4])[64]; /* [4][6][64] */
541 /* quantization matrix for trellis, [cqm][qp][coef] */
542 int (*unquant4_mf[4])[16]; /* [4][52][16] */
543 int (*unquant8_mf[4])[64]; /* [4][52][64] */
544 /* quantization matrix for deadzone */
545 udctcoef (*quant4_mf[4])[16]; /* [4][52][16] */
546 udctcoef (*quant8_mf[4])[64]; /* [4][52][64] */
547 udctcoef (*quant4_bias[4])[16]; /* [4][52][16] */
548 udctcoef (*quant8_bias[4])[64]; /* [4][52][64] */
549 udctcoef (*nr_offset_emergency)[4][64];
550
551 /* mv/ref cost arrays. */
552 uint16_t *cost_mv[QP_MAX+1];
553 uint16_t *cost_mv_fpel[QP_MAX+1][4];
554
555 const uint8_t *chroma_qp_table; /* includes both the nonlinear luma->chroma mapping and chroma_qp_offset */
556
557 /* Slice header */
558 x264_slice_header_t sh;
559
560 /* SPS / PPS */
561 x264_sps_t sps[1];
562 x264_pps_t pps[1];
563
564 /* Slice header backup, for SEI_DEC_REF_PIC_MARKING */
565 int b_sh_backup;
566 x264_slice_header_t sh_backup;
567
568 /* cabac context */
569 x264_cabac_t cabac;
570
571 struct
572 {
573 /* Frames to be encoded (whose types have been decided) */
574 x264_frame_t **current;
575 /* Unused frames: 0 = fenc, 1 = fdec */
576 x264_frame_t **unused[2];
577
578 /* Unused blank frames (for duplicates) */
579 x264_frame_t **blank_unused;
580
581 /* frames used for reference + sentinels */
582 x264_frame_t *reference[X264_REF_MAX+2];
583
584 int i_last_keyframe; /* Frame number of the last keyframe */
585 int i_last_idr; /* Frame number of the last IDR (not RP)*/
586 int i_poc_last_open_gop; /* Poc of the I frame of the last open-gop. The value
587 * is only assigned during the period between that
588 * I frame and the next P or I frame, else -1 */
589 int i_last_temporal_ref; /* MPEG-2: Frame number of the first displayed frame in a GOP */
590
591 int i_input; /* Number of input frames already accepted */
592
593 int i_max_dpb; /* Number of frames allocated in the decoded picture buffer */
594 int i_max_ref0;
595 int i_max_ref1;
596 int i_delay; /* Number of frames buffered for B reordering */
597 int i_bframe_delay;
598 int64_t i_bframe_delay_time;
599 int64_t i_first_pts;
600 int64_t i_prev_reordered_pts[2];
601 int64_t i_largest_pts;
602 int64_t i_second_largest_pts;
603 int b_have_lowres; /* Whether 1/2 resolution luma planes are being used */
604 int b_have_sub8x8_esa;
605 } frames;
606
607 /* current frame being encoded */
608 x264_frame_t *fenc;
609
610 /* frame being reconstructed */
611 x264_frame_t *fdec;
612
613 /* references lists */
614 int i_ref[2];
615 x264_frame_t *fref[2][X264_REF_MAX+3];
616 x264_frame_t *fref_nearest[2];
617 int b_ref_reorder[2];
618
619 /* hrd */
620 int initial_cpb_removal_delay;
621 int initial_cpb_removal_delay_offset;
622 int64_t i_reordered_pts_delay;
623
624 /* Current MB DCT coeffs */
625 struct
626 {
627 ALIGNED_16( dctcoef luma16x16_dc[3][16] );
628 ALIGNED_16( dctcoef chroma_dc[2][8] );
629 // FIXME share memory?
630 ALIGNED_16( dctcoef luma8x8[12][64] );
631 ALIGNED_16( dctcoef luma4x4[16*3][16] );
632 ALIGNED_16( dctcoef mpeg2_8x8[8][64] );
633 } dct;
634
635 /* MB table and cache for current frame/mb */
636 struct
637 {
638 int i_mb_width;
639 int i_mb_height;
640 int i_mb_count; /* number of mbs in a frame */
641
642 /* Chroma subsampling */
643 int chroma_h_shift;
644 int chroma_v_shift;
645
646 /* Strides */
647 int i_mb_stride;
648 int i_b8_stride;
649 int i_b4_stride;
650 int left_b8[2];
651 int left_b4[2];
652
653 /* Current index */
654 int i_mb_x;
655 int i_mb_y;
656 int i_mb_xy;
657 int i_b8_xy;
658 int i_b4_xy;
659
660 /* Search parameters */
661 int i_me_method;
662 int i_subpel_refine;
663 int b_chroma_me;
664 int b_trellis;
665 int b_noise_reduction;
666 int b_dct_decimate;
667 int i_psy_rd; /* Psy RD strength--fixed point value*/
668 int i_psy_trellis; /* Psy trellis strength--fixed point value*/
669
670 int b_interlaced;
671 int b_adaptive_mbaff; /* MBAFF+subme 0 requires non-adaptive MBAFF i.e. all field mbs */
672
673 /* Allowed qpel MV range to stay within the picture + emulated edge pixels */
674 int mv_min[2];
675 int mv_max[2];
676 int mv_miny_row[3]; /* 0 == top progressive, 1 == bot progressive, 2 == interlaced */
677 int mv_maxy_row[3];
678 /* Subpel MV range for motion search.
679 * same mv_min/max but includes levels' i_mv_range. */
680 int mv_min_spel[2];
681 int mv_max_spel[2];
682 int mv_miny_spel_row[3];
683 int mv_maxy_spel_row[3];
684 /* Fullpel MV range for motion search */
685 int mv_min_fpel[2];
686 int mv_max_fpel[2];
687 int mv_miny_fpel_row[3];
688 int mv_maxy_fpel_row[3];
689
690 /* neighboring MBs */
691 unsigned int i_neighbour;
692 unsigned int i_neighbour8[4]; /* neighbours of each 8x8 or 4x4 block that are available */
693 unsigned int i_neighbour4[16]; /* at the time the block is coded */
694 unsigned int i_neighbour_intra; /* for constrained intra pred */
695 unsigned int i_neighbour_frame; /* ignoring slice boundaries */
696 int i_mb_type_top;
697 int i_mb_type_left[2];
698 int i_mb_type_topleft;
699 int i_mb_type_topright;
700 int i_mb_prev_xy;
701 int i_mb_left_xy[2];
702 int i_mb_top_xy;
703 int i_mb_topleft_xy;
704 int i_mb_topright_xy;
705 int i_mb_top_y;
706 int i_mb_topleft_y;
707 int i_mb_topright_y;
708 const x264_left_table_t *left_index_table;
709 int i_mb_top_mbpair_xy;
710 int topleft_partition;
711 int b_allow_skip;
712 int field_decoding_flag;
713
714 /**** thread synchronization ends here ****/
715 /* subsequent variables are either thread-local or constant,
716 * and won't be copied from one thread to another */
717
718 /* mb table */
719 int8_t *type; /* mb type */
720 uint8_t *partition; /* mb partition */
721 int8_t *qp; /* mb qp */
722 int16_t *cbp; /* mb cbp: 0x0?: luma, 0x?0: chroma, 0x100: luma dc, 0x0200 and 0x0400: chroma dc (all set for PCM)*/
723 int8_t (*intra4x4_pred_mode)[8]; /* intra4x4 pred mode. for non I4x4 set to I_PRED_4x4_DC(2) */
724 /* actually has only 7 entries; set to 8 for write-combining optimizations */
725 uint8_t (*non_zero_count)[16*3]; /* nzc. for I_PCM set to 16 */
726 int8_t *chroma_pred_mode; /* chroma_pred_mode. cabac only. for non intra I_PRED_CHROMA_DC(0) */
727 int16_t (*mv[2])[2]; /* mb mv. set to 0 for intra mb */
728 uint8_t (*mvd[2])[8][2]; /* absolute value of mb mv difference with predict, clipped to [0,33]. set to 0 if intra. cabac only */
729 int8_t *ref[2]; /* mb ref. set to -1 if non used (intra or Lx only) */
730 int16_t (*mvr[2][X264_REF_MAX*2])[2];/* 16x16 mv for each possible ref */
731 int8_t *skipbp; /* block pattern for SKIP or DIRECT (sub)mbs. B-frames + cabac only */
732 int8_t *mb_transform_size; /* transform_size_8x8_flag of each mb */
733 uint16_t *slice_table; /* sh->first_mb of the slice that the indexed mb is part of
734 * NOTE: this will fail on resolutions above 2^16 MBs... */
735 uint8_t *field;
736
737 /* buffer for weighted versions of the reference frames */
738 pixel *p_weight_buf[X264_REF_MAX];
739
740 /* current value */
741 int i_type;
742 int i_partition;
743 ALIGNED_4( uint8_t i_sub_partition[4] );
744 int b_transform_8x8;
745
746 /* MPEG-2 */
747 int i_quant_scale_code;
748 int i_intra_dc_predictor[8];
749 int i_dct_dc_size[8];
750 int i_dct_dc_diff[8];
751 int i_bskip_type;
752 int16_t mvp[2][2];
753 int i_cbp_chroma422;
754
755 int i_cbp_luma;
756 int i_cbp_chroma;
757
758 int i_intra16x16_pred_mode;
759 int i_chroma_pred_mode;
760
761 /* skip flags for i4x4 and i8x8
762 * 0 = encode as normal.
763 * 1 (non-RD only) = the DCT is still in h->dct, restore fdec and skip reconstruction.
764 * 2 (RD only) = the DCT has since been overwritten by RD; restore that too. */
765 int i_skip_intra;
766 /* skip flag for motion compensation */
767 /* if we've already done MC, we don't need to do it again */
768 int b_skip_mc;
769 /* set to true if we are re-encoding a macroblock. */
770 int b_reencode_mb;
771 int ip_offset; /* Used by PIR to offset the quantizer of intra-refresh blocks. */
772 int b_deblock_rdo;
773 int b_overflow; /* If CAVLC had a level code overflow during bitstream writing. */
774
775 struct
776 {
777 /* space for p_fenc and p_fdec */
778#define FENC_STRIDE 16
779#define FDEC_STRIDE 32
780 ALIGNED_16( pixel fenc_buf[48*FENC_STRIDE] );
781 ALIGNED_16( pixel fdec_buf[52*FDEC_STRIDE] );
782
783 /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */
784 ALIGNED_16( pixel i4x4_fdec_buf[16*16] );
785 ALIGNED_16( pixel i8x8_fdec_buf[16*16] );
786 ALIGNED_16( dctcoef i8x8_dct_buf[3][64] );
787 ALIGNED_16( dctcoef i4x4_dct_buf[15][16] );
788 uint32_t i4x4_nnz_buf[4];
789 uint32_t i8x8_nnz_buf[4];
790 int i4x4_cbp;
791 int i8x8_cbp;
792
793 /* Psy trellis DCT data */
794 ALIGNED_16( dctcoef fenc_dct8[4][64] );
795 ALIGNED_16( dctcoef fenc_dct4[16][16] );
796
797 /* Psy RD SATD/SA8D scores cache */
798 ALIGNED_16( uint64_t fenc_hadamard_cache[9] );
799 ALIGNED_16( uint32_t fenc_satd_cache[32] );
800
801 /* pointer over mb of the frame to be compressed */
802 pixel *p_fenc[3]; /* y,u,v */
803 /* pointer to the actual source frame, not a block copy */
804 pixel *p_fenc_plane[3];
805
806 /* pointer over mb of the frame to be reconstructed */
807 pixel *p_fdec[3];
808
809 /* pointer over mb of the references */
810 int i_fref[2];
811 /* [12]: yN, yH, yV, yHV, (NV12 ? uv : I444 ? (uN, uH, uV, uHV, vN, ...)) */
812 pixel *p_fref[2][X264_REF_MAX*2][12];
813 pixel *p_fref_w[X264_REF_MAX*2]; /* weighted fullpel luma */
814 uint16_t *p_integral[2][X264_REF_MAX];
815
816 /* fref stride */
817 int i_stride[3];
818 } pic;
819
820 /* cache */
821 struct
822 {
823 /* real intra4x4_pred_mode if I_4X4 or I_8X8, I_PRED_4x4_DC if mb available, -1 if not */
824 ALIGNED_8( int8_t intra4x4_pred_mode[X264_SCAN8_LUMA_SIZE] );
825
826 /* i_non_zero_count if available else 0x80 */
827 ALIGNED_16( uint8_t non_zero_count[X264_SCAN8_SIZE] );
828
829 /* -1 if unused, -2 if unavailable */
830 ALIGNED_4( int8_t ref[2][X264_SCAN8_LUMA_SIZE] );
831
832 /* 0 if not available */
833 ALIGNED_16( int16_t mv[2][X264_SCAN8_LUMA_SIZE][2] );
834 ALIGNED_8( uint8_t mvd[2][X264_SCAN8_LUMA_SIZE][2] );
835
836 /* 1 if SKIP or DIRECT. set only for B-frames + CABAC */
837 ALIGNED_4( int8_t skip[X264_SCAN8_LUMA_SIZE] );
838
839 ALIGNED_4( int16_t direct_mv[2][4][2] );
840 ALIGNED_4( int8_t direct_ref[2][4] );
841 int direct_partition;
842 ALIGNED_4( int16_t pskip_mv[2] );
843
844 /* number of neighbors (top and left) that used 8x8 dct */
845 int i_neighbour_transform_size;
846 int i_neighbour_skip;
847
848 /* neighbor CBPs */
849 int i_cbp_top;
850 int i_cbp_left;
851
852 /* extra data required for mbaff in mv prediction */
853 int16_t topright_mv[2][3][2];
854 int8_t topright_ref[2][3];
855 } cache;
856
857 /* */
858 int i_qp; /* current qp */
859 int i_chroma_qp;
860 int i_last_qp; /* last qp */
861 int i_last_dqp; /* last delta qp */
862 int b_variable_qp; /* whether qp is allowed to vary per macroblock */
863 int b_lossless;
864 int b_direct_auto_read; /* take stats for --direct auto from the 2pass log */
865 int b_direct_auto_write; /* analyse direct modes, to use and/or save */
866
867 /* lambda values */
868 int i_trellis_lambda2[2][2]; /* [luma,chroma][inter,intra] */
869 int i_psy_rd_lambda;
870 int i_chroma_lambda2_offset;
871
872 /* B_direct and weighted prediction */
873 int16_t dist_scale_factor_buf[2][2][X264_REF_MAX*2][4];
874 int16_t (*dist_scale_factor)[4];
875 int8_t bipred_weight_buf[2][2][X264_REF_MAX*2][4];
876 int8_t (*bipred_weight)[4];
877 /* maps fref1[0]'s ref indices into the current list0 */
878#define map_col_to_list0(col) h->mb.map_col_to_list0[(col)+2]
879 int8_t map_col_to_list0[X264_REF_MAX+2];
880 int ref_blind_dupe; /* The index of the blind reference frame duplicate. */
881 int8_t deblock_ref_table[X264_REF_MAX*2+2];
882#define deblock_ref_table(x) h->mb.deblock_ref_table[(x)+2]
883 } mb;
884
885 /* rate control encoding only */
886 x264_ratecontrol_t *rc;
887
888 /* stats */
889 struct
890 {
891 /* Current frame stats */
892 struct
893 {
894 /* MV bits (MV+Ref+Block Type) */
895 int i_mv_bits;
896 /* Texture bits (DCT coefs) */
897 int i_tex_bits;
898 /* ? */
899 int i_misc_bits;
900 /* MB type counts */
901 int i_mb_count[19];
902 int i_mb_count_i;
903 int i_mb_count_p;
904 int i_mb_count_skip;
905 int i_mb_count_8x8dct[2];
906 int i_mb_count_ref[2][X264_REF_MAX*2];
907 int i_mb_partition[17];
908 int i_mb_cbp[6];
909 int i_mb_pred_mode[4][13];
910 int i_mb_field[3];
911 /* Adaptive direct mv pred */
912 int i_direct_score[2];
913 /* Metrics */
914 int64_t i_ssd[3];
915 double f_ssim;
916 int i_ssim_cnt;
917 } frame;
918
919 /* Cumulated stats */
920
921 /* per slice info */
922 int i_frame_count[3];
923 int64_t i_frame_size[3];
924 double f_frame_qp[3];
925 int i_consecutive_bframes[X264_BFRAME_MAX+1];
926 /* */
927 double f_ssd_global[3];
928 double f_psnr_average[3];
929 double f_psnr_mean_y[3];
930 double f_psnr_mean_u[3];
931 double f_psnr_mean_v[3];
932 double f_ssim_mean_y[3];
933 double f_frame_duration[3];
934 /* */
935 int64_t i_mb_count[3][19];
936 int64_t i_mb_partition[2][17];
937 int64_t i_mb_count_8x8dct[2];
938 int64_t i_mb_count_ref[2][2][X264_REF_MAX*2];
939 int64_t i_mb_cbp[6];
940 int64_t i_mb_pred_mode[4][13];
941 int64_t i_mb_field[3];
942 /* */
943 int i_direct_score[2];
944 int i_direct_frames[2];
945 /* num p-frames weighted */
946 int i_wpred[2];
947
948 } stat;
949
950 /* 0 = luma 4x4, 1 = luma 8x8, 2 = chroma 4x4, 3 = chroma 8x8 */
951 udctcoef (*nr_offset)[64];
952 uint32_t (*nr_residual_sum)[64];
953 uint32_t *nr_count;
954
955 ALIGNED_16( udctcoef nr_offset_denoise[4][64] );
956 ALIGNED_16( uint32_t nr_residual_sum_buf[2][4][64] );
957 uint32_t nr_count_buf[2][4];
958
959 uint8_t luma2chroma_pixel[7]; /* Subsampled pixel size */
960
961 /* Buffers that are allocated per-thread even in sliced threads. */
962 void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */
963 pixel *intra_border_backup[5][3]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
964 /* Deblock strength values are stored for each 4x4 partition. In MBAFF
965 * there are four extra values that need to be stored, located in [4][i]. */
966 uint8_t (*deblock_strength[2])[2][8][4];
967
968 /* CPU functions dependents */
969 x264_predict_t predict_16x16[4+3];
970 x264_predict8x8_t predict_8x8[9+3];
971 x264_predict_t predict_4x4[9+3];
972 x264_predict_t predict_chroma[4+3];
973 x264_predict_t predict_8x8c[4+3];
974 x264_predict_t predict_8x16c[4+3];
975 x264_predict_8x8_filter_t predict_8x8_filter;
976
977 x264_predict_mpeg2_t predict_8x8_mpeg2;
978
979 x264_pixel_function_t pixf;
980 x264_mc_functions_t mc;
981 x264_dct_function_t dctf;
982 x264_zigzag_function_t zigzagf;
983 x264_zigzag_function_t zigzagf_interlaced;
984 x264_zigzag_function_t zigzagf_progressive;
985 x264_quant_function_t quantf;
986 x264_deblock_function_t loopf;
987 x264_bitstream_function_t bsf;
988
989#if HAVE_VISUALIZE
990 struct visualize_t *visualize;
991#endif
992 x264_lookahead_t *lookahead;
993};
994
995// included at the end because it needs x264_t
996#include "macroblock.h"
997
998#if ARCH_X86 || ARCH_X86_64
999#include "x86/util.h"
1000#endif
1001
1002#include "rectangle.h"
1003
1004#endif