1 /*****************************************************************************
2 * common.h: misc common functions
3 *****************************************************************************
4 * Copyright (C) 2003-2013 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 #define X264_BFRAME_MAX 16
58 #define X264_REF_MAX 16
59 #define X264_THREAD_MAX 128
60 #define X264_LOOKAHEAD_THREAD_MAX 16
61 #define X264_PCM_COST (FRAME_SIZE(256*BIT_DEPTH)+16)
62 #define X264_LOOKAHEAD_MAX 250
63 #define QP_BD_OFFSET (6*(BIT_DEPTH-8))
64 #define QP_MAX_SPEC (51+QP_BD_OFFSET)
65 #define QP_MAX (QP_MAX_SPEC+18)
66 #define QP_MAX_MAX (51+2*6+18)
67 #define PIXEL_MAX ((1 << BIT_DEPTH)-1)
68 // arbitrary, but low because SATD scores are 1/4 normal
69 #define X264_LOOKAHEAD_QP (12+QP_BD_OFFSET)
70 #define SPEC_QP(x) X264_MIN((x), QP_MAX_SPEC)
72 // number of pixels (per thread) in progress at any given time.
73 // 16 for the macroblock in progress + 3 for deblocking + 3 for motion compensation filter + 2 for extra safety
74 #define X264_THREAD_HEIGHT 24
76 /* WEIGHTP_FAKE is set when mb_tree & psy are enabled, but normal weightp is disabled
77 * (such as in baseline). It checks for fades in lookahead and adjusts qp accordingly
78 * to increase quality. Defined as (-1) so that if(i_weighted_pred > 0) is true only when
79 * real weights are being used. */
81 #define X264_WEIGHTP_FAKE (-1)
83 #define NALU_OVERHEAD 5 // startcode + NAL type costs 5 bytes per frame
84 #define FILLER_OVERHEAD (NALU_OVERHEAD+1)
86 /****************************************************************************
88 ****************************************************************************/
98 # define MB_INTERLACED h->mb.b_interlaced
99 # define SLICE_MBAFF h->sh.b_mbaff
100 # define PARAM_INTERLACED h->param.b_interlaced
102 # define MB_INTERLACED 0
103 # define SLICE_MBAFF 0
104 # define PARAM_INTERLACED 0
108 # define CHROMA_H_SHIFT (CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422)
109 # define CHROMA_V_SHIFT (CHROMA_FORMAT == CHROMA_420)
111 # define CHROMA_FORMAT h->sps->i_chroma_format_idc
112 # define CHROMA_H_SHIFT h->mb.chroma_h_shift
113 # define CHROMA_V_SHIFT h->mb.chroma_v_shift
116 #define CHROMA_SIZE(s) ((s)>>(CHROMA_H_SHIFT+CHROMA_V_SHIFT))
117 #define FRAME_SIZE(s) ((s)+2*CHROMA_SIZE(s))
118 #define CHROMA444 (CHROMA_FORMAT == CHROMA_444)
120 /* Unions for type-punning.
121 * Mn: load or store n bits, aligned, native-endian
122 * CPn: copy n bits, aligned, native-endian
123 * we don't use memcpy for CPn because memcpy's args aren't assumed to be aligned */
124 typedef union { uint16_t i; uint8_t c[2]; } MAY_ALIAS x264_union16_t;
125 typedef union { uint32_t i; uint16_t b[2]; uint8_t c[4]; } MAY_ALIAS x264_union32_t;
126 typedef union { uint64_t i; uint32_t a[2]; uint16_t b[4]; uint8_t c[8]; } MAY_ALIAS x264_union64_t;
127 typedef struct { uint64_t i[2]; } x264_uint128_t;
128 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;
129 #define M16(src) (((x264_union16_t*)(src))->i)
130 #define M32(src) (((x264_union32_t*)(src))->i)
131 #define M64(src) (((x264_union64_t*)(src))->i)
132 #define M128(src) (((x264_union128_t*)(src))->i)
133 #define M128_ZERO ((x264_uint128_t){{0,0}})
134 #define CP16(dst,src) M16(dst) = M16(src)
135 #define CP32(dst,src) M32(dst) = M32(src)
136 #define CP64(dst,src) M64(dst) = M64(src)
137 #define CP128(dst,src) M128(dst) = M128(src)
140 typedef uint16_t pixel;
141 typedef uint64_t pixel4;
142 typedef int32_t dctcoef;
143 typedef uint32_t udctcoef;
145 # define PIXEL_SPLAT_X4(x) ((x)*0x0001000100010001ULL)
146 # define MPIXEL_X4(src) M64(src)
148 typedef uint8_t pixel;
149 typedef uint32_t pixel4;
150 typedef int16_t dctcoef;
151 typedef uint16_t udctcoef;
153 # define PIXEL_SPLAT_X4(x) ((x)*0x01010101U)
154 # define MPIXEL_X4(src) M32(src)
157 #define BIT_DEPTH X264_BIT_DEPTH
159 #define CPPIXEL_X4(dst,src) MPIXEL_X4(dst) = MPIXEL_X4(src)
161 #define X264_SCAN8_LUMA_SIZE (5*8)
162 #define X264_SCAN8_SIZE (X264_SCAN8_LUMA_SIZE*3)
163 #define X264_SCAN8_0 (4+1*8)
165 /* Scan8 organization:
182 * DY/DU/DV are for luma/chroma DC.
188 static const uint8_t x264_scan8[16*3 + 3] =
190 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8,
191 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8,
192 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8,
193 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8,
194 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8,
195 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8,
196 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8,
197 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8,
198 4+11*8, 5+11*8, 4+12*8, 5+12*8,
199 6+11*8, 7+11*8, 6+12*8, 7+12*8,
200 4+13*8, 5+13*8, 4+14*8, 5+14*8,
201 6+13*8, 7+13*8, 6+14*8, 7+14*8,
202 0+ 0*8, 0+ 5*8, 0+10*8
206 #include "bitstream.h"
216 #include "threadpool.h"
218 /****************************************************************************
220 ****************************************************************************/
221 /* x264_malloc : will do or emulate a memalign
222 * you have to use x264_free for buffers allocated with x264_malloc */
223 void *x264_malloc( int );
224 void x264_free( void * );
226 /* x264_slurp_file: malloc space for the whole file and read it */
227 char *x264_slurp_file( const char *filename );
229 /* mdate: return the current date in microsecond */
230 int64_t x264_mdate( void );
232 /* x264_param2string: return a (malloced) string containing most of
233 * the encoding options */
234 char *x264_param2string( x264_param_t *p, int b_res );
237 void x264_log( x264_t *h, int i_level, const char *psz_fmt, ... );
239 void x264_reduce_fraction( uint32_t *n, uint32_t *d );
240 void x264_reduce_fraction64( uint64_t *n, uint64_t *d );
241 void x264_cavlc_init( x264_t *h );
242 void x264_cabac_init( x264_t *h );
244 static ALWAYS_INLINE pixel x264_clip_pixel( int x )
246 return ( (x & ~PIXEL_MAX) ? (-x)>>31 & PIXEL_MAX : x );
249 static ALWAYS_INLINE int x264_clip3( int v, int i_min, int i_max )
251 return ( (v < i_min) ? i_min : (v > i_max) ? i_max : v );
254 static ALWAYS_INLINE double x264_clip3f( double v, double f_min, double f_max )
256 return ( (v < f_min) ? f_min : (v > f_max) ? f_max : v );
259 static ALWAYS_INLINE int x264_median( int a, int b, int c )
261 int t = (a-b)&((a-b)>>31);
264 b -= (b-c)&((b-c)>>31);
265 b += (a-b)&((a-b)>>31);
269 static ALWAYS_INLINE void x264_median_mv( int16_t *dst, int16_t *a, int16_t *b, int16_t *c )
271 dst[0] = x264_median( a[0], b[0], c[0] );
272 dst[1] = x264_median( a[1], b[1], c[1] );
275 static ALWAYS_INLINE int x264_predictor_difference( int16_t (*mvc)[2], intptr_t i_mvc )
278 for( int i = 0; i < i_mvc-1; i++ )
280 sum += abs( mvc[i][0] - mvc[i+1][0] )
281 + abs( mvc[i][1] - mvc[i+1][1] );
286 static ALWAYS_INLINE uint16_t x264_cabac_mvd_sum( uint8_t *mvdleft, uint8_t *mvdtop )
288 int amvd0 = abs(mvdleft[0]) + abs(mvdtop[0]);
289 int amvd1 = abs(mvdleft[1]) + abs(mvdtop[1]);
290 amvd0 = (amvd0 > 2) + (amvd0 > 32);
291 amvd1 = (amvd1 > 2) + (amvd1 > 32);
292 return amvd0 + (amvd1<<8);
295 extern const uint8_t x264_exp2_lut[64];
296 extern const float x264_log2_lut[128];
297 extern const float x264_log2_lz_lut[32];
299 /* Not a general-purpose function; multiplies input by -1/6 to convert
301 static ALWAYS_INLINE int x264_exp2fix8( float x )
303 int i = x*(-64.f/6.f) + 512.5f;
304 if( i < 0 ) return 0;
305 if( i > 1023 ) return 0xffff;
306 return (x264_exp2_lut[i&63]+256) << (i>>6) >> 8;
309 static ALWAYS_INLINE float x264_log2( uint32_t x )
311 int lz = x264_clz( x );
312 return x264_log2_lut[(x<<lz>>24)&0x7f] + x264_log2_lz_lut[lz];
315 /****************************************************************************
317 ****************************************************************************/
325 static const char slice_type_to_char[] = { 'P', 'B', 'I' };
327 enum sei_payload_type_e
329 SEI_BUFFERING_PERIOD = 0,
331 SEI_PAN_SCAN_RECT = 2,
333 SEI_USER_DATA_REGISTERED = 4,
334 SEI_USER_DATA_UNREGISTERED = 5,
335 SEI_RECOVERY_POINT = 6,
336 SEI_DEC_REF_PIC_MARKING = 7,
337 SEI_FRAME_PACKING = 45,
357 int i_idr_pic_id; /* -1 if nal_type != 5 */
360 int i_delta_poc_bottom;
363 int i_redundant_pic_cnt;
365 int b_direct_spatial_mv_pred;
367 int b_num_ref_idx_override;
368 int i_num_ref_idx_l0_active;
369 int i_num_ref_idx_l1_active;
371 int b_ref_pic_list_reordering[2];
376 } ref_pic_list_order[2][X264_REF_MAX];
378 /* P-frame weighting */
380 x264_weight_t weight[X264_REF_MAX*2][3];
382 int i_mmco_remove_from_end;
383 int i_mmco_command_count;
384 struct /* struct for future expansion */
386 int i_difference_of_pic_nums;
388 } mmco[X264_REF_MAX];
390 int i_cabac_init_idc;
397 /* deblocking filter */
398 int i_disable_deblocking_filter_idc;
399 int i_alpha_c0_offset;
402 } x264_slice_header_t;
404 typedef struct x264_lookahead_t
406 volatile uint8_t b_exit_thread;
407 uint8_t b_thread_active;
408 uint8_t b_analyse_keyframe;
410 int i_slicetype_length;
411 x264_frame_t *last_nonb;
412 x264_pthread_t thread_handle;
413 x264_sync_frame_list_t ifbuf;
414 x264_sync_frame_list_t next;
415 x264_sync_frame_list_t ofbuf;
418 typedef struct x264_ratecontrol_t x264_ratecontrol_t;
420 typedef struct x264_left_table_t
424 uint8_t nnz_chroma[4];
429 /* Current frame stats */
432 /* MV bits (MV+Ref+Block Type) */
434 /* Texture bits (DCT coefs) */
443 int i_mb_count_8x8dct[2];
444 int i_mb_count_ref[2][X264_REF_MAX*2];
445 int i_mb_partition[17];
447 int i_mb_pred_mode[4][13];
449 /* Adaptive direct mv pred */
450 int i_direct_score[2];
459 /* encoder parameters */
462 x264_t *thread[X264_THREAD_MAX+1];
463 x264_t *lookahead_thread[X264_LOOKAHEAD_THREAD_MAX];
465 int i_thread_phase; /* which thread to use for the next frame */
466 int i_thread_idx; /* which thread this is */
467 int i_threadslice_start; /* first row in this thread slice */
468 int i_threadslice_end; /* row after the end of this thread slice */
469 int i_threadslice_pass; /* which pass of encoding we are on */
470 x264_threadpool_t *threadpool;
471 x264_threadpool_t *lookaheadpool;
472 x264_pthread_mutex_t mutex;
473 x264_pthread_cond_t cv;
475 /* bitstream output */
479 int i_nals_allocated;
481 int i_bitstream; /* size of p_bitstream */
482 uint8_t *p_bitstream; /* will hold data for all nal */
489 /**** thread synchronization starts here ****/
491 /* frame number/poc */
495 int i_thread_frames; /* Number of different frames being encoded by threads;
496 * 1 when sliced-threads is on. */
500 int64_t i_disp_fields; /* Number of displayed fields (both coded and implied via pic_struct) */
501 int i_disp_fields_last_frame;
502 int64_t i_prev_duration; /* Duration of previous frame */
503 int64_t i_coded_fields; /* Number of coded fields (both coded and implied via pic_struct) */
504 int64_t i_cpb_delay; /* Equal to number of fields preceding this field
505 * since last buffering_period SEI */
506 int64_t i_coded_fields_lookahead; /* Use separate counters for lookahead */
507 int64_t i_cpb_delay_lookahead;
509 int64_t i_cpb_delay_pir_offset;
510 int64_t i_cpb_delay_pir_offset_next;
512 int b_queued_intra_refresh;
513 int64_t i_last_idr_pts;
517 /* quantization matrix for decoding, [cqm][qp%6][coef] */
518 int (*dequant4_mf[4])[16]; /* [4][6][16] */
519 int (*dequant8_mf[4])[64]; /* [4][6][64] */
520 /* quantization matrix for trellis, [cqm][qp][coef] */
521 int (*unquant4_mf[4])[16]; /* [4][52][16] */
522 int (*unquant8_mf[4])[64]; /* [4][52][64] */
523 /* quantization matrix for deadzone */
524 udctcoef (*quant4_mf[4])[16]; /* [4][52][16] */
525 udctcoef (*quant8_mf[4])[64]; /* [4][52][64] */
526 udctcoef (*quant4_bias[4])[16]; /* [4][52][16] */
527 udctcoef (*quant8_bias[4])[64]; /* [4][52][64] */
528 udctcoef (*quant4_bias0[4])[16]; /* [4][52][16] */
529 udctcoef (*quant8_bias0[4])[64]; /* [4][52][64] */
530 udctcoef (*nr_offset_emergency)[4][64];
532 /* mv/ref cost arrays. */
533 uint16_t *cost_mv[QP_MAX+1];
534 uint16_t *cost_mv_fpel[QP_MAX+1][4];
536 const uint8_t *chroma_qp_table; /* includes both the nonlinear luma->chroma mapping and chroma_qp_offset */
539 x264_slice_header_t sh;
545 /* Slice header backup, for SEI_DEC_REF_PIC_MARKING */
547 x264_slice_header_t sh_backup;
554 /* Frames to be encoded (whose types have been decided) */
555 x264_frame_t **current;
556 /* Unused frames: 0 = fenc, 1 = fdec */
557 x264_frame_t **unused[2];
559 /* Unused blank frames (for duplicates) */
560 x264_frame_t **blank_unused;
562 /* frames used for reference + sentinels */
563 x264_frame_t *reference[X264_REF_MAX+2];
565 int i_last_keyframe; /* Frame number of the last keyframe */
566 int i_last_idr; /* Frame number of the last IDR (not RP)*/
567 int i_poc_last_open_gop; /* Poc of the I frame of the last open-gop. The value
568 * is only assigned during the period between that
569 * I frame and the next P or I frame, else -1 */
571 int i_input; /* Number of input frames already accepted */
573 int i_max_dpb; /* Number of frames allocated in the decoded picture buffer */
576 int i_delay; /* Number of frames buffered for B reordering */
578 int64_t i_bframe_delay_time;
580 int64_t i_prev_reordered_pts[2];
581 int64_t i_largest_pts;
582 int64_t i_second_largest_pts;
583 int b_have_lowres; /* Whether 1/2 resolution luma planes are being used */
584 int b_have_sub8x8_esa;
587 /* current frame being encoded */
590 /* frame being reconstructed */
593 /* references lists */
595 x264_frame_t *fref[2][X264_REF_MAX+3];
596 x264_frame_t *fref_nearest[2];
597 int b_ref_reorder[2];
600 int initial_cpb_removal_delay;
601 int initial_cpb_removal_delay_offset;
602 int64_t i_reordered_pts_delay;
604 /* Current MB DCT coeffs */
607 ALIGNED_16( dctcoef luma16x16_dc[3][16] );
608 ALIGNED_16( dctcoef chroma_dc[2][8] );
609 // FIXME share memory?
610 ALIGNED_16( dctcoef luma8x8[12][64] );
611 ALIGNED_16( dctcoef luma4x4[16*3][16] );
614 /* MB table and cache for current frame/mb */
619 int i_mb_count; /* number of mbs in a frame */
621 /* Chroma subsampling */
639 /* Search parameters */
644 int b_noise_reduction;
646 int i_psy_rd; /* Psy RD strength--fixed point value*/
647 int i_psy_trellis; /* Psy trellis strength--fixed point value*/
650 int b_adaptive_mbaff; /* MBAFF+subme 0 requires non-adaptive MBAFF i.e. all field mbs */
652 /* Allowed qpel MV range to stay within the picture + emulated edge pixels */
655 int mv_miny_row[3]; /* 0 == top progressive, 1 == bot progressive, 2 == interlaced */
657 /* Subpel MV range for motion search.
658 * same mv_min/max but includes levels' i_mv_range. */
661 int mv_miny_spel_row[3];
662 int mv_maxy_spel_row[3];
663 /* Fullpel MV range for motion search */
664 ALIGNED_8( int16_t mv_limit_fpel[2][2] ); /* min_x, min_y, max_x, max_y */
665 int mv_miny_fpel_row[3];
666 int mv_maxy_fpel_row[3];
668 /* neighboring MBs */
669 unsigned int i_neighbour;
670 unsigned int i_neighbour8[4]; /* neighbours of each 8x8 or 4x4 block that are available */
671 unsigned int i_neighbour4[16]; /* at the time the block is coded */
672 unsigned int i_neighbour_intra; /* for constrained intra pred */
673 unsigned int i_neighbour_frame; /* ignoring slice boundaries */
675 int i_mb_type_left[2];
676 int i_mb_type_topleft;
677 int i_mb_type_topright;
682 int i_mb_topright_xy;
686 const x264_left_table_t *left_index_table;
687 int i_mb_top_mbpair_xy;
688 int topleft_partition;
690 int field_decoding_flag;
692 /**** thread synchronization ends here ****/
693 /* subsequent variables are either thread-local or constant,
694 * and won't be copied from one thread to another */
697 int8_t *type; /* mb type */
698 uint8_t *partition; /* mb partition */
699 int8_t *qp; /* mb qp */
700 int16_t *cbp; /* mb cbp: 0x0?: luma, 0x?0: chroma, 0x100: luma dc, 0x0200 and 0x0400: chroma dc (all set for PCM)*/
701 int8_t (*intra4x4_pred_mode)[8]; /* intra4x4 pred mode. for non I4x4 set to I_PRED_4x4_DC(2) */
702 /* actually has only 7 entries; set to 8 for write-combining optimizations */
703 uint8_t (*non_zero_count)[16*3]; /* nzc. for I_PCM set to 16 */
704 int8_t *chroma_pred_mode; /* chroma_pred_mode. cabac only. for non intra I_PRED_CHROMA_DC(0) */
705 int16_t (*mv[2])[2]; /* mb mv. set to 0 for intra mb */
706 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 */
707 int8_t *ref[2]; /* mb ref. set to -1 if non used (intra or Lx only) */
708 int16_t (*mvr[2][X264_REF_MAX*2])[2];/* 16x16 mv for each possible ref */
709 int8_t *skipbp; /* block pattern for SKIP or DIRECT (sub)mbs. B-frames + cabac only */
710 int8_t *mb_transform_size; /* transform_size_8x8_flag of each mb */
711 uint16_t *slice_table; /* sh->first_mb of the slice that the indexed mb is part of
712 * NOTE: this will fail on resolutions above 2^16 MBs... */
715 /* buffer for weighted versions of the reference frames */
716 pixel *p_weight_buf[X264_REF_MAX];
721 ALIGNED_4( uint8_t i_sub_partition[4] );
727 int i_intra16x16_pred_mode;
728 int i_chroma_pred_mode;
730 /* skip flags for i4x4 and i8x8
731 * 0 = encode as normal.
732 * 1 (non-RD only) = the DCT is still in h->dct, restore fdec and skip reconstruction.
733 * 2 (RD only) = the DCT has since been overwritten by RD; restore that too. */
735 /* skip flag for motion compensation */
736 /* if we've already done MC, we don't need to do it again */
738 /* set to true if we are re-encoding a macroblock. */
740 int ip_offset; /* Used by PIR to offset the quantizer of intra-refresh blocks. */
742 int b_overflow; /* If CAVLC had a level code overflow during bitstream writing. */
746 /* space for p_fenc and p_fdec */
747 #define FENC_STRIDE 16
748 #define FDEC_STRIDE 32
749 ALIGNED_16( pixel fenc_buf[48*FENC_STRIDE] );
750 ALIGNED_16( pixel fdec_buf[52*FDEC_STRIDE] );
752 /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */
753 ALIGNED_16( pixel i4x4_fdec_buf[16*16] );
754 ALIGNED_16( pixel i8x8_fdec_buf[16*16] );
755 ALIGNED_16( dctcoef i8x8_dct_buf[3][64] );
756 ALIGNED_16( dctcoef i4x4_dct_buf[15][16] );
757 uint32_t i4x4_nnz_buf[4];
758 uint32_t i8x8_nnz_buf[4];
762 /* Psy trellis DCT data */
763 ALIGNED_16( dctcoef fenc_dct8[4][64] );
764 ALIGNED_16( dctcoef fenc_dct4[16][16] );
766 /* Psy RD SATD/SA8D scores cache */
767 ALIGNED_16( uint64_t fenc_hadamard_cache[9] );
768 ALIGNED_16( uint32_t fenc_satd_cache[32] );
770 /* pointer over mb of the frame to be compressed */
771 pixel *p_fenc[3]; /* y,u,v */
772 /* pointer to the actual source frame, not a block copy */
773 pixel *p_fenc_plane[3];
775 /* pointer over mb of the frame to be reconstructed */
778 /* pointer over mb of the references */
780 /* [12]: yN, yH, yV, yHV, (NV12 ? uv : I444 ? (uN, uH, uV, uHV, vN, ...)) */
781 pixel *p_fref[2][X264_REF_MAX*2][12];
782 pixel *p_fref_w[X264_REF_MAX*2]; /* weighted fullpel luma */
783 uint16_t *p_integral[2][X264_REF_MAX];
792 /* real intra4x4_pred_mode if I_4X4 or I_8X8, I_PRED_4x4_DC if mb available, -1 if not */
793 ALIGNED_8( int8_t intra4x4_pred_mode[X264_SCAN8_LUMA_SIZE] );
795 /* i_non_zero_count if available else 0x80 */
796 ALIGNED_16( uint8_t non_zero_count[X264_SCAN8_SIZE] );
798 /* -1 if unused, -2 if unavailable */
799 ALIGNED_4( int8_t ref[2][X264_SCAN8_LUMA_SIZE] );
801 /* 0 if not available */
802 ALIGNED_16( int16_t mv[2][X264_SCAN8_LUMA_SIZE][2] );
803 ALIGNED_8( uint8_t mvd[2][X264_SCAN8_LUMA_SIZE][2] );
805 /* 1 if SKIP or DIRECT. set only for B-frames + CABAC */
806 ALIGNED_4( int8_t skip[X264_SCAN8_LUMA_SIZE] );
808 ALIGNED_4( int16_t direct_mv[2][4][2] );
809 ALIGNED_4( int8_t direct_ref[2][4] );
810 int direct_partition;
811 ALIGNED_4( int16_t pskip_mv[2] );
813 /* number of neighbors (top and left) that used 8x8 dct */
814 int i_neighbour_transform_size;
815 int i_neighbour_skip;
821 /* extra data required for mbaff in mv prediction */
822 int16_t topright_mv[2][3][2];
823 int8_t topright_ref[2][3];
825 /* current mb deblock strength */
826 uint8_t (*deblock_strength)[8][4];
830 int i_qp; /* current qp */
832 int i_last_qp; /* last qp */
833 int i_last_dqp; /* last delta qp */
834 int b_variable_qp; /* whether qp is allowed to vary per macroblock */
836 int b_direct_auto_read; /* take stats for --direct auto from the 2pass log */
837 int b_direct_auto_write; /* analyse direct modes, to use and/or save */
840 int i_trellis_lambda2[2][2]; /* [luma,chroma][inter,intra] */
842 int i_chroma_lambda2_offset;
844 /* B_direct and weighted prediction */
845 int16_t dist_scale_factor_buf[2][2][X264_REF_MAX*2][4];
846 int16_t (*dist_scale_factor)[4];
847 int8_t bipred_weight_buf[2][2][X264_REF_MAX*2][4];
848 int8_t (*bipred_weight)[4];
849 /* maps fref1[0]'s ref indices into the current list0 */
850 #define map_col_to_list0(col) h->mb.map_col_to_list0[(col)+2]
851 int8_t map_col_to_list0[X264_REF_MAX+2];
852 int ref_blind_dupe; /* The index of the blind reference frame duplicate. */
853 int8_t deblock_ref_table[X264_REF_MAX*2+2];
854 #define deblock_ref_table(x) h->mb.deblock_ref_table[(x)+2]
857 /* rate control encoding only */
858 x264_ratecontrol_t *rc;
863 /* Current frame stats */
864 x264_frame_stat_t frame;
866 /* Cumulated stats */
869 int i_frame_count[3];
870 int64_t i_frame_size[3];
871 double f_frame_qp[3];
872 int i_consecutive_bframes[X264_BFRAME_MAX+1];
874 double f_ssd_global[3];
875 double f_psnr_average[3];
876 double f_psnr_mean_y[3];
877 double f_psnr_mean_u[3];
878 double f_psnr_mean_v[3];
879 double f_ssim_mean_y[3];
880 double f_frame_duration[3];
882 int64_t i_mb_count[3][19];
883 int64_t i_mb_partition[2][17];
884 int64_t i_mb_count_8x8dct[2];
885 int64_t i_mb_count_ref[2][2][X264_REF_MAX*2];
887 int64_t i_mb_pred_mode[4][13];
888 int64_t i_mb_field[3];
890 int i_direct_score[2];
891 int i_direct_frames[2];
892 /* num p-frames weighted */
897 /* 0 = luma 4x4, 1 = luma 8x8, 2 = chroma 4x4, 3 = chroma 8x8 */
898 udctcoef (*nr_offset)[64];
899 uint32_t (*nr_residual_sum)[64];
902 ALIGNED_16( udctcoef nr_offset_denoise[4][64] );
903 ALIGNED_16( uint32_t nr_residual_sum_buf[2][4][64] );
904 uint32_t nr_count_buf[2][4];
906 uint8_t luma2chroma_pixel[7]; /* Subsampled pixel size */
908 /* Buffers that are allocated per-thread even in sliced threads. */
909 void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */
910 void *scratch_buffer2; /* if the first one's already in use */
911 pixel *intra_border_backup[5][3]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
912 /* Deblock strength values are stored for each 4x4 partition. In MBAFF
913 * there are four extra values that need to be stored, located in [4][i]. */
914 uint8_t (*deblock_strength[2])[2][8][4];
916 /* CPU functions dependents */
917 x264_predict_t predict_16x16[4+3];
918 x264_predict8x8_t predict_8x8[9+3];
919 x264_predict_t predict_4x4[9+3];
920 x264_predict_t predict_chroma[4+3];
921 x264_predict_t predict_8x8c[4+3];
922 x264_predict_t predict_8x16c[4+3];
923 x264_predict_8x8_filter_t predict_8x8_filter;
925 x264_pixel_function_t pixf;
926 x264_mc_functions_t mc;
927 x264_dct_function_t dctf;
928 x264_zigzag_function_t zigzagf;
929 x264_zigzag_function_t zigzagf_interlaced;
930 x264_zigzag_function_t zigzagf_progressive;
931 x264_quant_function_t quantf;
932 x264_deblock_function_t loopf;
933 x264_bitstream_function_t bsf;
936 struct visualize_t *visualize;
938 x264_lookahead_t *lookahead;
941 // included at the end because it needs x264_t
942 #include "macroblock.h"
944 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 )
947 for( int i = 0; i < i_mvc; i++ )
949 int mx = (mvc[i][0] + 2) >> 2;
950 int my = (mvc[i][1] + 2) >> 2;
951 uint32_t mv = pack16to32_mask(mx, my);
952 if( !mv || mv == pmv ) continue;
953 dst[cnt][0] = x264_clip3( mx, mv_limit[0][0], mv_limit[1][0] );
954 dst[cnt][1] = x264_clip3( my, mv_limit[0][1], mv_limit[1][1] );
960 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 )
963 int qpel_limit[4] = {mv_limit[0][0] << 2, mv_limit[0][1] << 2, mv_limit[1][0] << 2, mv_limit[1][1] << 2};
964 for( int i = 0; i < i_mvc; i++ )
966 uint32_t mv = M32( mvc[i] );
969 if( !mv || mv == pmv ) continue;
970 dst[cnt][0] = x264_clip3( mx, qpel_limit[0], qpel_limit[2] );
971 dst[cnt][1] = x264_clip3( my, qpel_limit[1], qpel_limit[3] );
977 #if ARCH_X86 || ARCH_X86_64
978 #include "x86/util.h"
981 #include "rectangle.h"