1 /*****************************************************************************
2 * common.h: misc common functions
3 *****************************************************************************
4 * Copyright (C) 2003-2014 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
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., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
23 * This program is also available under a commercial proprietary license.
24 * For more information, contact us at licensing@x264.com.
25 *****************************************************************************/
30 /****************************************************************************
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 #define ARRAY_ELEMS(a) ((sizeof(a))/(sizeof(a[0])))
45 #define CHECKED_MALLOC( var, size )\
47 var = x264_malloc( size );\
51 #define CHECKED_MALLOCZERO( var, size )\
53 CHECKED_MALLOC( var, size );\
54 memset( var, 0, size );\
57 /* Macros for merging multiple allocations into a single large malloc, for improved
58 * use with huge pages. */
60 /* Needs to be enough to contain any set of buffers that use combined allocations */
61 #define PREALLOC_BUF_SIZE 1024
63 #define PREALLOC_INIT\
64 int prealloc_idx = 0;\
65 size_t prealloc_size = 0;\
66 uint8_t **preallocs[PREALLOC_BUF_SIZE];
68 #define PREALLOC( var, size )\
70 var = (void*)prealloc_size;\
71 preallocs[prealloc_idx++] = (uint8_t**)&var;\
72 prealloc_size += ALIGN(size, NATIVE_ALIGN);\
75 #define PREALLOC_END( ptr )\
77 CHECKED_MALLOC( ptr, prealloc_size );\
78 while( prealloc_idx-- )\
79 *preallocs[prealloc_idx] += (intptr_t)ptr;\
82 #define ARRAY_SIZE(array) (sizeof(array)/sizeof(array[0]))
84 #define X264_BFRAME_MAX 16
85 #define X264_REF_MAX 16
86 #define X264_THREAD_MAX 128
87 #define X264_LOOKAHEAD_THREAD_MAX 16
88 #define X264_PCM_COST (FRAME_SIZE(256*BIT_DEPTH)+16)
89 #define X264_LOOKAHEAD_MAX 250
90 #define QP_BD_OFFSET (6*(BIT_DEPTH-8))
91 #define QP_MAX_SPEC (51+QP_BD_OFFSET)
92 #define QP_MAX (QP_MAX_SPEC+18)
93 #define QP_MAX_MAX (51+2*6+18)
94 #define PIXEL_MAX ((1 << BIT_DEPTH)-1)
95 // arbitrary, but low because SATD scores are 1/4 normal
96 #define X264_LOOKAHEAD_QP (12+QP_BD_OFFSET)
97 #define SPEC_QP(x) X264_MIN((x), QP_MAX_SPEC)
99 // number of pixels (per thread) in progress at any given time.
100 // 16 for the macroblock in progress + 3 for deblocking + 3 for motion compensation filter + 2 for extra safety
101 #define X264_THREAD_HEIGHT 24
103 /* WEIGHTP_FAKE is set when mb_tree & psy are enabled, but normal weightp is disabled
104 * (such as in baseline). It checks for fades in lookahead and adjusts qp accordingly
105 * to increase quality. Defined as (-1) so that if(i_weighted_pred > 0) is true only when
106 * real weights are being used. */
108 #define X264_WEIGHTP_FAKE (-1)
110 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
111 #define FILLER_OVERHEAD (NALU_OVERHEAD+1)
112 #define SEI_OVERHEAD (NALU_OVERHEAD - (h->param.b_annexb && !h->param.i_avcintra_class && (h->out.i_nal-1)))
114 /****************************************************************************
116 ****************************************************************************/
126 # define MB_INTERLACED h->mb.b_interlaced
127 # define SLICE_MBAFF h->sh.b_mbaff
128 # define PARAM_INTERLACED h->param.b_interlaced
130 # define MB_INTERLACED 0
131 # define SLICE_MBAFF 0
132 # define PARAM_INTERLACED 0
136 # define CHROMA_H_SHIFT (CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422)
137 # define CHROMA_V_SHIFT (CHROMA_FORMAT == CHROMA_420)
139 # define CHROMA_FORMAT h->sps->i_chroma_format_idc
140 # define CHROMA_H_SHIFT h->mb.chroma_h_shift
141 # define CHROMA_V_SHIFT h->mb.chroma_v_shift
144 #define CHROMA_SIZE(s) ((s)>>(CHROMA_H_SHIFT+CHROMA_V_SHIFT))
145 #define FRAME_SIZE(s) ((s)+2*CHROMA_SIZE(s))
146 #define CHROMA444 (CHROMA_FORMAT == CHROMA_444)
148 /* Unions for type-punning.
149 * Mn: load or store n bits, aligned, native-endian
150 * CPn: copy n bits, aligned, native-endian
151 * we don't use memcpy for CPn because memcpy's args aren't assumed to be aligned */
152 typedef union { uint16_t i; uint8_t c[2]; } MAY_ALIAS x264_union16_t;
153 typedef union { uint32_t i; uint16_t b[2]; uint8_t c[4]; } MAY_ALIAS x264_union32_t;
154 typedef union { uint64_t i; uint32_t a[2]; uint16_t b[4]; uint8_t c[8]; } MAY_ALIAS x264_union64_t;
155 typedef struct { uint64_t i[2]; } x264_uint128_t;
156 typedef 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;
157 #define M16(src) (((x264_union16_t*)(src))->i)
158 #define M32(src) (((x264_union32_t*)(src))->i)
159 #define M64(src) (((x264_union64_t*)(src))->i)
160 #define M128(src) (((x264_union128_t*)(src))->i)
161 #define M128_ZERO ((x264_uint128_t){{0,0}})
162 #define CP16(dst,src) M16(dst) = M16(src)
163 #define CP32(dst,src) M32(dst) = M32(src)
164 #define CP64(dst,src) M64(dst) = M64(src)
165 #define CP128(dst,src) M128(dst) = M128(src)
168 typedef uint16_t pixel;
169 typedef uint64_t pixel4;
170 typedef int32_t dctcoef;
171 typedef uint32_t udctcoef;
173 # define PIXEL_SPLAT_X4(x) ((x)*0x0001000100010001ULL)
174 # define MPIXEL_X4(src) M64(src)
176 typedef uint8_t pixel;
177 typedef uint32_t pixel4;
178 typedef int16_t dctcoef;
179 typedef uint16_t udctcoef;
181 # define PIXEL_SPLAT_X4(x) ((x)*0x01010101U)
182 # define MPIXEL_X4(src) M32(src)
185 #define BIT_DEPTH X264_BIT_DEPTH
187 #define CPPIXEL_X4(dst,src) MPIXEL_X4(dst) = MPIXEL_X4(src)
189 #define X264_SCAN8_LUMA_SIZE (5*8)
190 #define X264_SCAN8_SIZE (X264_SCAN8_LUMA_SIZE*3)
191 #define X264_SCAN8_0 (4+1*8)
193 /* Scan8 organization:
210 * DY/DU/DV are for luma/chroma DC.
216 static const uint8_t x264_scan8[16*3 + 3] =
218 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8,
219 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8,
220 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8,
221 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8,
222 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8,
223 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8,
224 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8,
225 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8,
226 4+11*8, 5+11*8, 4+12*8, 5+12*8,
227 6+11*8, 7+11*8, 6+12*8, 7+12*8,
228 4+13*8, 5+13*8, 4+14*8, 5+14*8,
229 6+13*8, 7+13*8, 6+14*8, 7+14*8,
230 0+ 0*8, 0+ 5*8, 0+10*8
238 #include "bitstream.h"
247 #include "threadpool.h"
249 /****************************************************************************
251 ****************************************************************************/
252 /* x264_malloc : will do or emulate a memalign
253 * you have to use x264_free for buffers allocated with x264_malloc */
254 void *x264_malloc( int );
255 void x264_free( void * );
257 /* x264_slurp_file: malloc space for the whole file and read it */
258 char *x264_slurp_file( const char *filename );
260 /* mdate: return the current date in microsecond */
261 int64_t x264_mdate( void );
263 /* x264_param2string: return a (malloced) string containing most of
264 * the encoding options */
265 char *x264_param2string( x264_param_t *p, int b_res );
268 void x264_log( x264_t *h, int i_level, const char *psz_fmt, ... );
270 void x264_reduce_fraction( uint32_t *n, uint32_t *d );
271 void x264_reduce_fraction64( uint64_t *n, uint64_t *d );
272 void x264_cavlc_init( x264_t *h );
273 void x264_cabac_init( x264_t *h );
275 static ALWAYS_INLINE pixel x264_clip_pixel( int x )
277 return ( (x & ~PIXEL_MAX) ? (-x)>>31 & PIXEL_MAX : x );
280 static ALWAYS_INLINE int x264_clip3( int v, int i_min, int i_max )
282 return ( (v < i_min) ? i_min : (v > i_max) ? i_max : v );
285 static ALWAYS_INLINE double x264_clip3f( double v, double f_min, double f_max )
287 return ( (v < f_min) ? f_min : (v > f_max) ? f_max : v );
290 static ALWAYS_INLINE int x264_median( int a, int b, int c )
292 int t = (a-b)&((a-b)>>31);
295 b -= (b-c)&((b-c)>>31);
296 b += (a-b)&((a-b)>>31);
300 static ALWAYS_INLINE void x264_median_mv( int16_t *dst, int16_t *a, int16_t *b, int16_t *c )
302 dst[0] = x264_median( a[0], b[0], c[0] );
303 dst[1] = x264_median( a[1], b[1], c[1] );
306 static ALWAYS_INLINE int x264_predictor_difference( int16_t (*mvc)[2], intptr_t i_mvc )
309 for( int i = 0; i < i_mvc-1; i++ )
311 sum += abs( mvc[i][0] - mvc[i+1][0] )
312 + abs( mvc[i][1] - mvc[i+1][1] );
317 static ALWAYS_INLINE uint16_t x264_cabac_mvd_sum( uint8_t *mvdleft, uint8_t *mvdtop )
319 int amvd0 = abs(mvdleft[0]) + abs(mvdtop[0]);
320 int amvd1 = abs(mvdleft[1]) + abs(mvdtop[1]);
321 amvd0 = (amvd0 > 2) + (amvd0 > 32);
322 amvd1 = (amvd1 > 2) + (amvd1 > 32);
323 return amvd0 + (amvd1<<8);
326 extern const uint8_t x264_exp2_lut[64];
327 extern const float x264_log2_lut[128];
328 extern const float x264_log2_lz_lut[32];
330 /* Not a general-purpose function; multiplies input by -1/6 to convert
332 static ALWAYS_INLINE int x264_exp2fix8( float x )
334 int i = x*(-64.f/6.f) + 512.5f;
335 if( i < 0 ) return 0;
336 if( i > 1023 ) return 0xffff;
337 return (x264_exp2_lut[i&63]+256) << (i>>6) >> 8;
340 static ALWAYS_INLINE float x264_log2( uint32_t x )
342 int lz = x264_clz( x );
343 return x264_log2_lut[(x<<lz>>24)&0x7f] + x264_log2_lz_lut[lz];
346 /****************************************************************************
348 ****************************************************************************/
356 static const char slice_type_to_char[] = { 'P', 'B', 'I' };
358 enum sei_payload_type_e
360 SEI_BUFFERING_PERIOD = 0,
362 SEI_PAN_SCAN_RECT = 2,
364 SEI_USER_DATA_REGISTERED = 4,
365 SEI_USER_DATA_UNREGISTERED = 5,
366 SEI_RECOVERY_POINT = 6,
367 SEI_DEC_REF_PIC_MARKING = 7,
368 SEI_FRAME_PACKING = 45,
388 int i_idr_pic_id; /* -1 if nal_type != 5 */
391 int i_delta_poc_bottom;
394 int i_redundant_pic_cnt;
396 int b_direct_spatial_mv_pred;
398 int b_num_ref_idx_override;
399 int i_num_ref_idx_l0_active;
400 int i_num_ref_idx_l1_active;
402 int b_ref_pic_list_reordering[2];
407 } ref_pic_list_order[2][X264_REF_MAX];
409 /* P-frame weighting */
411 x264_weight_t weight[X264_REF_MAX*2][3];
413 int i_mmco_remove_from_end;
414 int i_mmco_command_count;
415 struct /* struct for future expansion */
417 int i_difference_of_pic_nums;
419 } mmco[X264_REF_MAX];
421 int i_cabac_init_idc;
428 /* deblocking filter */
429 int i_disable_deblocking_filter_idc;
430 int i_alpha_c0_offset;
433 } x264_slice_header_t;
435 typedef struct x264_lookahead_t
437 volatile uint8_t b_exit_thread;
438 uint8_t b_thread_active;
439 uint8_t b_analyse_keyframe;
441 int i_slicetype_length;
442 x264_frame_t *last_nonb;
443 x264_pthread_t thread_handle;
444 x264_sync_frame_list_t ifbuf;
445 x264_sync_frame_list_t next;
446 x264_sync_frame_list_t ofbuf;
449 typedef struct x264_ratecontrol_t x264_ratecontrol_t;
451 typedef struct x264_left_table_t
455 uint8_t nnz_chroma[4];
460 /* Current frame stats */
463 /* MV bits (MV+Ref+Block Type) */
465 /* Texture bits (DCT coefs) */
474 int i_mb_count_8x8dct[2];
475 int i_mb_count_ref[2][X264_REF_MAX*2];
476 int i_mb_partition[17];
478 int i_mb_pred_mode[4][13];
480 /* Adaptive direct mv pred */
481 int i_direct_score[2];
490 /* encoder parameters */
493 x264_t *thread[X264_THREAD_MAX+1];
494 x264_t *lookahead_thread[X264_LOOKAHEAD_THREAD_MAX];
496 int i_thread_phase; /* which thread to use for the next frame */
497 int i_thread_idx; /* which thread this is */
498 int i_threadslice_start; /* first row in this thread slice */
499 int i_threadslice_end; /* row after the end of this thread slice */
500 int i_threadslice_pass; /* which pass of encoding we are on */
501 x264_threadpool_t *threadpool;
502 x264_threadpool_t *lookaheadpool;
503 x264_pthread_mutex_t mutex;
504 x264_pthread_cond_t cv;
506 /* bitstream output */
510 int i_nals_allocated;
512 int i_bitstream; /* size of p_bitstream */
513 uint8_t *p_bitstream; /* will hold data for all nal */
523 /**** thread synchronization starts here ****/
525 /* frame number/poc */
529 int i_thread_frames; /* Number of different frames being encoded by threads;
530 * 1 when sliced-threads is on. */
534 int64_t i_disp_fields; /* Number of displayed fields (both coded and implied via pic_struct) */
535 int i_disp_fields_last_frame;
536 int64_t i_prev_duration; /* Duration of previous frame */
537 int64_t i_coded_fields; /* Number of coded fields (both coded and implied via pic_struct) */
538 int64_t i_cpb_delay; /* Equal to number of fields preceding this field
539 * since last buffering_period SEI */
540 int64_t i_coded_fields_lookahead; /* Use separate counters for lookahead */
541 int64_t i_cpb_delay_lookahead;
543 int64_t i_cpb_delay_pir_offset;
544 int64_t i_cpb_delay_pir_offset_next;
546 int b_queued_intra_refresh;
547 int64_t i_last_idr_pts;
551 /* quantization matrix for decoding, [cqm][qp%6][coef] */
552 int (*dequant4_mf[4])[16]; /* [4][6][16] */
553 int (*dequant8_mf[4])[64]; /* [4][6][64] */
554 /* quantization matrix for trellis, [cqm][qp][coef] */
555 int (*unquant4_mf[4])[16]; /* [4][QP_MAX_SPEC+1][16] */
556 int (*unquant8_mf[4])[64]; /* [4][QP_MAX_SPEC+1][64] */
557 /* quantization matrix for deadzone */
558 udctcoef (*quant4_mf[4])[16]; /* [4][QP_MAX_SPEC+1][16] */
559 udctcoef (*quant8_mf[4])[64]; /* [4][QP_MAX_SPEC+1][64] */
560 udctcoef (*quant4_bias[4])[16]; /* [4][QP_MAX_SPEC+1][16] */
561 udctcoef (*quant8_bias[4])[64]; /* [4][QP_MAX_SPEC+1][64] */
562 udctcoef (*quant4_bias0[4])[16]; /* [4][QP_MAX_SPEC+1][16] */
563 udctcoef (*quant8_bias0[4])[64]; /* [4][QP_MAX_SPEC+1][64] */
564 udctcoef (*nr_offset_emergency)[4][64];
566 /* mv/ref cost arrays. */
567 uint16_t *cost_mv[QP_MAX+1];
568 uint16_t *cost_mv_fpel[QP_MAX+1][4];
570 const uint8_t *chroma_qp_table; /* includes both the nonlinear luma->chroma mapping and chroma_qp_offset */
573 x264_slice_header_t sh;
579 /* Slice header backup, for SEI_DEC_REF_PIC_MARKING */
581 x264_slice_header_t sh_backup;
588 /* Frames to be encoded (whose types have been decided) */
589 x264_frame_t **current;
590 /* Unused frames: 0 = fenc, 1 = fdec */
591 x264_frame_t **unused[2];
593 /* Unused blank frames (for duplicates) */
594 x264_frame_t **blank_unused;
596 /* frames used for reference + sentinels */
597 x264_frame_t *reference[X264_REF_MAX+2];
599 int i_last_keyframe; /* Frame number of the last keyframe */
600 int i_last_idr; /* Frame number of the last IDR (not RP)*/
601 int i_poc_last_open_gop; /* Poc of the I frame of the last open-gop. The value
602 * is only assigned during the period between that
603 * I frame and the next P or I frame, else -1 */
605 int i_input; /* Number of input frames already accepted */
607 int i_max_dpb; /* Number of frames allocated in the decoded picture buffer */
610 int i_delay; /* Number of frames buffered for B reordering */
612 int64_t i_bframe_delay_time;
614 int64_t i_prev_reordered_pts[2];
615 int64_t i_largest_pts;
616 int64_t i_second_largest_pts;
617 int b_have_lowres; /* Whether 1/2 resolution luma planes are being used */
618 int b_have_sub8x8_esa;
621 /* current frame being encoded */
624 /* frame being reconstructed */
627 /* references lists */
629 x264_frame_t *fref[2][X264_REF_MAX+3];
630 x264_frame_t *fref_nearest[2];
631 int b_ref_reorder[2];
634 int initial_cpb_removal_delay;
635 int initial_cpb_removal_delay_offset;
636 int64_t i_reordered_pts_delay;
638 /* Current MB DCT coeffs */
641 ALIGNED_N( dctcoef luma16x16_dc[3][16] );
642 ALIGNED_16( dctcoef chroma_dc[2][8] );
643 // FIXME share memory?
644 ALIGNED_N( dctcoef luma8x8[12][64] );
645 ALIGNED_N( dctcoef luma4x4[16*3][16] );
648 /* MB table and cache for current frame/mb */
653 int i_mb_count; /* number of mbs in a frame */
655 /* Chroma subsampling */
673 /* Search parameters */
678 int b_noise_reduction;
680 int i_psy_rd; /* Psy RD strength--fixed point value*/
681 int i_psy_trellis; /* Psy trellis strength--fixed point value*/
684 int b_adaptive_mbaff; /* MBAFF+subme 0 requires non-adaptive MBAFF i.e. all field mbs */
686 /* Allowed qpel MV range to stay within the picture + emulated edge pixels */
689 int mv_miny_row[3]; /* 0 == top progressive, 1 == bot progressive, 2 == interlaced */
691 /* Subpel MV range for motion search.
692 * same mv_min/max but includes levels' i_mv_range. */
695 int mv_miny_spel_row[3];
696 int mv_maxy_spel_row[3];
697 /* Fullpel MV range for motion search */
698 ALIGNED_8( int16_t mv_limit_fpel[2][2] ); /* min_x, min_y, max_x, max_y */
699 int mv_miny_fpel_row[3];
700 int mv_maxy_fpel_row[3];
702 /* neighboring MBs */
703 unsigned int i_neighbour;
704 unsigned int i_neighbour8[4]; /* neighbours of each 8x8 or 4x4 block that are available */
705 unsigned int i_neighbour4[16]; /* at the time the block is coded */
706 unsigned int i_neighbour_intra; /* for constrained intra pred */
707 unsigned int i_neighbour_frame; /* ignoring slice boundaries */
709 int i_mb_type_left[2];
710 int i_mb_type_topleft;
711 int i_mb_type_topright;
716 int i_mb_topright_xy;
720 const x264_left_table_t *left_index_table;
721 int i_mb_top_mbpair_xy;
722 int topleft_partition;
724 int field_decoding_flag;
726 /**** thread synchronization ends here ****/
727 /* subsequent variables are either thread-local or constant,
728 * and won't be copied from one thread to another */
731 uint8_t *base; /* base pointer for all malloced data in this mb */
732 int8_t *type; /* mb type */
733 uint8_t *partition; /* mb partition */
734 int8_t *qp; /* mb qp */
735 int16_t *cbp; /* mb cbp: 0x0?: luma, 0x?0: chroma, 0x100: luma dc, 0x0200 and 0x0400: chroma dc (all set for PCM)*/
736 int8_t (*intra4x4_pred_mode)[8]; /* intra4x4 pred mode. for non I4x4 set to I_PRED_4x4_DC(2) */
737 /* actually has only 7 entries; set to 8 for write-combining optimizations */
738 uint8_t (*non_zero_count)[16*3]; /* nzc. for I_PCM set to 16 */
739 int8_t *chroma_pred_mode; /* chroma_pred_mode. cabac only. for non intra I_PRED_CHROMA_DC(0) */
740 int16_t (*mv[2])[2]; /* mb mv. set to 0 for intra mb */
741 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 */
742 int8_t *ref[2]; /* mb ref. set to -1 if non used (intra or Lx only) */
743 int16_t (*mvr[2][X264_REF_MAX*2])[2];/* 16x16 mv for each possible ref */
744 int8_t *skipbp; /* block pattern for SKIP or DIRECT (sub)mbs. B-frames + cabac only */
745 int8_t *mb_transform_size; /* transform_size_8x8_flag of each mb */
746 uint16_t *slice_table; /* sh->first_mb of the slice that the indexed mb is part of
747 * NOTE: this will fail on resolutions above 2^16 MBs... */
750 /* buffer for weighted versions of the reference frames */
751 pixel *p_weight_buf[X264_REF_MAX];
756 ALIGNED_4( uint8_t i_sub_partition[4] );
762 int i_intra16x16_pred_mode;
763 int i_chroma_pred_mode;
765 /* skip flags for i4x4 and i8x8
766 * 0 = encode as normal.
767 * 1 (non-RD only) = the DCT is still in h->dct, restore fdec and skip reconstruction.
768 * 2 (RD only) = the DCT has since been overwritten by RD; restore that too. */
770 /* skip flag for motion compensation */
771 /* if we've already done MC, we don't need to do it again */
773 /* set to true if we are re-encoding a macroblock. */
775 int ip_offset; /* Used by PIR to offset the quantizer of intra-refresh blocks. */
777 int b_overflow; /* If CAVLC had a level code overflow during bitstream writing. */
781 /* space for p_fenc and p_fdec */
782 #define FENC_STRIDE 16
783 #define FDEC_STRIDE 32
784 ALIGNED_16( pixel fenc_buf[48*FENC_STRIDE] );
785 ALIGNED_N( pixel fdec_buf[52*FDEC_STRIDE] );
787 /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */
788 ALIGNED_16( pixel i4x4_fdec_buf[16*16] );
789 ALIGNED_16( pixel i8x8_fdec_buf[16*16] );
790 ALIGNED_16( dctcoef i8x8_dct_buf[3][64] );
791 ALIGNED_16( dctcoef i4x4_dct_buf[15][16] );
792 uint32_t i4x4_nnz_buf[4];
793 uint32_t i8x8_nnz_buf[4];
797 /* Psy trellis DCT data */
798 ALIGNED_16( dctcoef fenc_dct8[4][64] );
799 ALIGNED_16( dctcoef fenc_dct4[16][16] );
801 /* Psy RD SATD/SA8D scores cache */
802 ALIGNED_N( uint64_t fenc_hadamard_cache[9] );
803 ALIGNED_N( uint32_t fenc_satd_cache[32] );
805 /* pointer over mb of the frame to be compressed */
806 pixel *p_fenc[3]; /* y,u,v */
807 /* pointer to the actual source frame, not a block copy */
808 pixel *p_fenc_plane[3];
810 /* pointer over mb of the frame to be reconstructed */
813 /* pointer over mb of the references */
815 /* [12]: yN, yH, yV, yHV, (NV12 ? uv : I444 ? (uN, uH, uV, uHV, vN, ...)) */
816 pixel *p_fref[2][X264_REF_MAX*2][12];
817 pixel *p_fref_w[X264_REF_MAX*2]; /* weighted fullpel luma */
818 uint16_t *p_integral[2][X264_REF_MAX];
827 /* real intra4x4_pred_mode if I_4X4 or I_8X8, I_PRED_4x4_DC if mb available, -1 if not */
828 ALIGNED_8( int8_t intra4x4_pred_mode[X264_SCAN8_LUMA_SIZE] );
830 /* i_non_zero_count if available else 0x80 */
831 ALIGNED_16( uint8_t non_zero_count[X264_SCAN8_SIZE] );
833 /* -1 if unused, -2 if unavailable */
834 ALIGNED_4( int8_t ref[2][X264_SCAN8_LUMA_SIZE] );
836 /* 0 if not available */
837 ALIGNED_16( int16_t mv[2][X264_SCAN8_LUMA_SIZE][2] );
838 ALIGNED_8( uint8_t mvd[2][X264_SCAN8_LUMA_SIZE][2] );
840 /* 1 if SKIP or DIRECT. set only for B-frames + CABAC */
841 ALIGNED_4( int8_t skip[X264_SCAN8_LUMA_SIZE] );
843 ALIGNED_4( int16_t direct_mv[2][4][2] );
844 ALIGNED_4( int8_t direct_ref[2][4] );
845 int direct_partition;
846 ALIGNED_4( int16_t pskip_mv[2] );
848 /* number of neighbors (top and left) that used 8x8 dct */
849 int i_neighbour_transform_size;
850 int i_neighbour_skip;
856 /* extra data required for mbaff in mv prediction */
857 int16_t topright_mv[2][3][2];
858 int8_t topright_ref[2][3];
860 /* current mb deblock strength */
861 uint8_t (*deblock_strength)[8][4];
865 int i_qp; /* current qp */
867 int i_last_qp; /* last qp */
868 int i_last_dqp; /* last delta qp */
869 int b_variable_qp; /* whether qp is allowed to vary per macroblock */
871 int b_direct_auto_read; /* take stats for --direct auto from the 2pass log */
872 int b_direct_auto_write; /* analyse direct modes, to use and/or save */
875 int i_trellis_lambda2[2][2]; /* [luma,chroma][inter,intra] */
877 int i_chroma_lambda2_offset;
879 /* B_direct and weighted prediction */
880 int16_t dist_scale_factor_buf[2][2][X264_REF_MAX*2][4];
881 int16_t (*dist_scale_factor)[4];
882 int8_t bipred_weight_buf[2][2][X264_REF_MAX*2][4];
883 int8_t (*bipred_weight)[4];
884 /* maps fref1[0]'s ref indices into the current list0 */
885 #define map_col_to_list0(col) h->mb.map_col_to_list0[(col)+2]
886 int8_t map_col_to_list0[X264_REF_MAX+2];
887 int ref_blind_dupe; /* The index of the blind reference frame duplicate. */
888 int8_t deblock_ref_table[X264_REF_MAX*2+2];
889 #define deblock_ref_table(x) h->mb.deblock_ref_table[(x)+2]
892 /* rate control encoding only */
893 x264_ratecontrol_t *rc;
898 /* Current frame stats */
899 x264_frame_stat_t frame;
901 /* Cumulated stats */
904 int i_frame_count[3];
905 int64_t i_frame_size[3];
906 double f_frame_qp[3];
907 int i_consecutive_bframes[X264_BFRAME_MAX+1];
909 double f_ssd_global[3];
910 double f_psnr_average[3];
911 double f_psnr_mean_y[3];
912 double f_psnr_mean_u[3];
913 double f_psnr_mean_v[3];
914 double f_ssim_mean_y[3];
915 double f_frame_duration[3];
917 int64_t i_mb_count[3][19];
918 int64_t i_mb_partition[2][17];
919 int64_t i_mb_count_8x8dct[2];
920 int64_t i_mb_count_ref[2][2][X264_REF_MAX*2];
922 int64_t i_mb_pred_mode[4][13];
923 int64_t i_mb_field[3];
925 int i_direct_score[2];
926 int i_direct_frames[2];
927 /* num p-frames weighted */
932 /* 0 = luma 4x4, 1 = luma 8x8, 2 = chroma 4x4, 3 = chroma 8x8 */
933 udctcoef (*nr_offset)[64];
934 uint32_t (*nr_residual_sum)[64];
937 ALIGNED_N( udctcoef nr_offset_denoise[4][64] );
938 ALIGNED_N( uint32_t nr_residual_sum_buf[2][4][64] );
939 uint32_t nr_count_buf[2][4];
941 uint8_t luma2chroma_pixel[7]; /* Subsampled pixel size */
943 /* Buffers that are allocated per-thread even in sliced threads. */
944 void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */
945 void *scratch_buffer2; /* if the first one's already in use */
946 pixel *intra_border_backup[5][3]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
947 /* Deblock strength values are stored for each 4x4 partition. In MBAFF
948 * there are four extra values that need to be stored, located in [4][i]. */
949 uint8_t (*deblock_strength[2])[2][8][4];
951 /* CPU functions dependents */
952 x264_predict_t predict_16x16[4+3];
953 x264_predict8x8_t predict_8x8[9+3];
954 x264_predict_t predict_4x4[9+3];
955 x264_predict_t predict_chroma[4+3];
956 x264_predict_t predict_8x8c[4+3];
957 x264_predict_t predict_8x16c[4+3];
958 x264_predict_8x8_filter_t predict_8x8_filter;
960 x264_pixel_function_t pixf;
961 x264_mc_functions_t mc;
962 x264_dct_function_t dctf;
963 x264_zigzag_function_t zigzagf;
964 x264_zigzag_function_t zigzagf_interlaced;
965 x264_zigzag_function_t zigzagf_progressive;
966 x264_quant_function_t quantf;
967 x264_deblock_function_t loopf;
968 x264_bitstream_function_t bsf;
970 x264_lookahead_t *lookahead;
973 x264_opencl_t opencl;
977 // included at the end because it needs x264_t
978 #include "macroblock.h"
980 static int ALWAYS_INLINE x264_predictor_roundclip( int16_t (*dst)[2], int16_t (*mvc)[2], int i_mvc, int16_t mv_limit[2][2], uint32_t pmv )
983 for( int i = 0; i < i_mvc; i++ )
985 int mx = (mvc[i][0] + 2) >> 2;
986 int my = (mvc[i][1] + 2) >> 2;
987 uint32_t mv = pack16to32_mask(mx, my);
988 if( !mv || mv == pmv ) continue;
989 dst[cnt][0] = x264_clip3( mx, mv_limit[0][0], mv_limit[1][0] );
990 dst[cnt][1] = x264_clip3( my, mv_limit[0][1], mv_limit[1][1] );
996 static int ALWAYS_INLINE x264_predictor_clip( int16_t (*dst)[2], int16_t (*mvc)[2], int i_mvc, int16_t mv_limit[2][2], uint32_t pmv )
999 int qpel_limit[4] = {mv_limit[0][0] << 2, mv_limit[0][1] << 2, mv_limit[1][0] << 2, mv_limit[1][1] << 2};
1000 for( int i = 0; i < i_mvc; i++ )
1002 uint32_t mv = M32( mvc[i] );
1005 if( !mv || mv == pmv ) continue;
1006 dst[cnt][0] = x264_clip3( mx, qpel_limit[0], qpel_limit[2] );
1007 dst[cnt][1] = x264_clip3( my, qpel_limit[1], qpel_limit[3] );
1013 #if ARCH_X86 || ARCH_X86_64
1014 #include "x86/util.h"
1017 #include "rectangle.h"