1 /*****************************************************************************
2 * common.h: h264 encoder
3 *****************************************************************************
4 * Copyright (C) 2003-2008 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.
22 *****************************************************************************/
27 /****************************************************************************
29 ****************************************************************************/
30 #define X264_MIN(a,b) ( (a)<(b) ? (a) : (b) )
31 #define X264_MAX(a,b) ( (a)>(b) ? (a) : (b) )
32 #define X264_MIN3(a,b,c) X264_MIN((a),X264_MIN((b),(c)))
33 #define X264_MAX3(a,b,c) X264_MAX((a),X264_MAX((b),(c)))
34 #define X264_MIN4(a,b,c,d) X264_MIN((a),X264_MIN3((b),(c),(d)))
35 #define X264_MAX4(a,b,c,d) X264_MAX((a),X264_MAX3((b),(c),(d)))
36 #define XCHG(type,a,b) do{ type t = a; a = b; b = t; } while(0)
37 #define IS_DISPOSABLE(type) ( type == X264_TYPE_B )
38 #define FIX8(f) ((int)(f*(1<<8)+.5))
40 #define CHECKED_MALLOC( var, size )\
42 var = x264_malloc( size );\
46 #define CHECKED_MALLOCZERO( var, size )\
48 CHECKED_MALLOC( var, size );\
49 memset( var, 0, size );\
52 #define X264_BFRAME_MAX 16
53 #define X264_THREAD_MAX 128
54 #define X264_PCM_COST (386*8)
55 #define X264_LOOKAHEAD_MAX 250
56 // arbitrary, but low because SATD scores are 1/4 normal
57 #define X264_LOOKAHEAD_QP 12
59 // number of pixels (per thread) in progress at any given time.
60 // 16 for the macroblock in progress + 3 for deblocking + 3 for motion compensation filter + 2 for extra safety
61 #define X264_THREAD_HEIGHT 24
63 /* WEIGHTP_FAKE is set when mb_tree & psy are enabled, but normal weightp is disabled
64 * (such as in baseline). It checks for fades in lookahead and adjusts qp accordingly
65 * to increase quality. Defined as (-1) so that if(i_weighted_pred > 0) is true only when
66 * real weights are being used. */
68 #define X264_WEIGHTP_FAKE (-1)
70 /****************************************************************************
72 ****************************************************************************/
81 /* Unions for type-punning.
82 * Mn: load or store n bits, aligned, native-endian
83 * CPn: copy n bits, aligned, native-endian
84 * we don't use memcpy for CPn because memcpy's args aren't assumed to be aligned */
85 typedef union { uint16_t i; uint8_t c[2]; } MAY_ALIAS x264_union16_t;
86 typedef union { uint32_t i; uint16_t b[2]; uint8_t c[4]; } MAY_ALIAS x264_union32_t;
87 typedef union { uint64_t i; uint32_t a[2]; uint16_t b[4]; uint8_t c[8]; } MAY_ALIAS x264_union64_t;
88 #define M16(src) (((x264_union16_t*)(src))->i)
89 #define M32(src) (((x264_union32_t*)(src))->i)
90 #define M64(src) (((x264_union64_t*)(src))->i)
91 #define CP16(dst,src) M16(dst) = M16(src)
92 #define CP32(dst,src) M32(dst) = M32(src)
93 #define CP64(dst,src) M64(dst) = M64(src)
106 /****************************************************************************
108 ****************************************************************************/
109 /* x264_malloc : will do or emulate a memalign
110 * you have to use x264_free for buffers allocated with x264_malloc */
111 void *x264_malloc( int );
112 void x264_free( void * );
114 /* x264_slurp_file: malloc space for the whole file and read it */
115 char *x264_slurp_file( const char *filename );
117 /* mdate: return the current date in microsecond */
118 int64_t x264_mdate( void );
120 /* x264_param2string: return a (malloced) string containing most of
121 * the encoding options */
122 char *x264_param2string( x264_param_t *p, int b_res );
124 int x264_nal_encode( uint8_t *dst, int b_annexb, x264_nal_t *nal );
127 void x264_log( x264_t *h, int i_level, const char *psz_fmt, ... );
129 void x264_reduce_fraction( int *n, int *d );
130 void x264_init_vlc_tables();
132 static inline uint8_t x264_clip_uint8( int x )
134 return x&(~255) ? (-x)>>31 : x;
137 static inline int x264_clip3( int v, int i_min, int i_max )
139 return ( (v < i_min) ? i_min : (v > i_max) ? i_max : v );
142 static inline double x264_clip3f( double v, double f_min, double f_max )
144 return ( (v < f_min) ? f_min : (v > f_max) ? f_max : v );
147 static inline int x264_median( int a, int b, int c )
149 int t = (a-b)&((a-b)>>31);
152 b -= (b-c)&((b-c)>>31);
153 b += (a-b)&((a-b)>>31);
157 static inline void x264_median_mv( int16_t *dst, int16_t *a, int16_t *b, int16_t *c )
159 dst[0] = x264_median( a[0], b[0], c[0] );
160 dst[1] = x264_median( a[1], b[1], c[1] );
163 static inline int x264_predictor_difference( int16_t (*mvc)[2], intptr_t i_mvc )
166 for( i = 0; i < i_mvc-1; i++ )
168 sum += abs( mvc[i][0] - mvc[i+1][0] )
169 + abs( mvc[i][1] - mvc[i+1][1] );
174 static inline uint32_t x264_cabac_amvd_sum( int16_t *mvdleft, int16_t *mvdtop )
176 int amvd0 = abs(mvdleft[0]) + abs(mvdtop[0]);
177 int amvd1 = abs(mvdleft[1]) + abs(mvdtop[1]);
178 amvd0 = (amvd0 > 2) + (amvd0 > 32);
179 amvd1 = (amvd1 > 2) + (amvd1 > 32);
180 return amvd0 + (amvd1<<16);
183 extern const uint8_t x264_exp2_lut[64];
184 extern const float x264_log2_lut[128];
185 extern const float x264_log2_lz_lut[32];
187 /* Not a general-purpose function; multiplies input by -1/6 to convert
189 static ALWAYS_INLINE int x264_exp2fix8( float x )
191 int i = x*(-64.f/6.f) + 512.5f;
192 if( i < 0 ) return 0;
193 if( i > 1023 ) return 0xffff;
194 return (x264_exp2_lut[i&63]+256) << (i>>6) >> 8;
197 static ALWAYS_INLINE float x264_log2( uint32_t x )
199 int lz = x264_clz( x );
200 return x264_log2_lut[(x<<lz>>24)&0x7f] + x264_log2_lz_lut[lz];
203 /****************************************************************************
205 ****************************************************************************/
215 static const char slice_type_to_char[] = { 'P', 'B', 'I', 'S', 'S' };
234 int i_idr_pic_id; /* -1 if nal_type != 5 */
237 int i_delta_poc_bottom;
240 int i_redundant_pic_cnt;
242 int b_direct_spatial_mv_pred;
244 int b_num_ref_idx_override;
245 int i_num_ref_idx_l0_active;
246 int i_num_ref_idx_l1_active;
248 int b_ref_pic_list_reordering_l0;
249 int b_ref_pic_list_reordering_l1;
254 } ref_pic_list_order[2][16];
256 /* P-frame weighting */
257 x264_weight_t weight[32][3];
259 int i_mmco_remove_from_end;
260 int i_mmco_command_count;
261 struct /* struct for future expansion */
263 int i_difference_of_pic_nums;
267 int i_cabac_init_idc;
274 /* deblocking filter */
275 int i_disable_deblocking_filter_idc;
276 int i_alpha_c0_offset;
279 } x264_slice_header_t;
281 typedef struct x264_lookahead_t
283 volatile uint8_t b_exit_thread;
284 uint8_t b_thread_active;
285 uint8_t b_analyse_keyframe;
287 int i_slicetype_length;
288 x264_frame_t *last_nonb;
289 x264_synch_frame_list_t ifbuf;
290 x264_synch_frame_list_t next;
291 x264_synch_frame_list_t ofbuf;
296 #define X264_SCAN8_SIZE (6*8)
297 #define X264_SCAN8_0 (4+1*8)
299 static const int x264_scan8[16+2*4+3] =
302 4+1*8, 5+1*8, 4+2*8, 5+2*8,
303 6+1*8, 7+1*8, 6+2*8, 7+2*8,
304 4+3*8, 5+3*8, 4+4*8, 5+4*8,
305 6+3*8, 7+3*8, 6+4*8, 7+4*8,
331 typedef struct x264_ratecontrol_t x264_ratecontrol_t;
335 /* encoder parameters */
338 x264_t *thread[X264_THREAD_MAX+1];
339 x264_pthread_t thread_handle;
341 int i_thread_phase; /* which thread to use for the next frame */
342 int i_threadslice_start; /* first row in this thread slice */
343 int i_threadslice_end; /* row after the end of this thread slice */
345 /* bitstream output */
349 int i_nals_allocated;
351 int i_bitstream; /* size of p_bitstream */
352 uint8_t *p_bitstream; /* will hold data for all nal */
359 /**** thread synchronization starts here ****/
361 /* frame number/poc */
365 int i_thread_frames; /* Number of different frames being encoded by threads;
366 * 1 when sliced-threads is on. */
370 /* We use only one SPS and one PPS */
371 x264_sps_t sps_array[1];
373 x264_pps_t pps_array[1];
377 /* Timebase multiplier for DTS compression */
378 int i_dts_compress_multiplier;
380 /* quantization matrix for decoding, [cqm][qp%6][coef] */
381 int (*dequant4_mf[4])[16]; /* [4][6][16] */
382 int (*dequant8_mf[2])[64]; /* [2][6][64] */
383 /* quantization matrix for trellis, [cqm][qp][coef] */
384 int (*unquant4_mf[4])[16]; /* [4][52][16] */
385 int (*unquant8_mf[2])[64]; /* [2][52][64] */
386 /* quantization matrix for deadzone */
387 uint16_t (*quant4_mf[4])[16]; /* [4][52][16] */
388 uint16_t (*quant8_mf[2])[64]; /* [2][52][64] */
389 uint16_t (*quant4_bias[4])[16]; /* [4][52][16] */
390 uint16_t (*quant8_bias[2])[64]; /* [2][52][64] */
392 /* mv/ref cost arrays. Indexed by lambda instead of
393 * qp because, due to rounding, some quantizers share
394 * lambdas. This saves memory. */
395 uint16_t *cost_mv[92];
396 uint16_t *cost_mv_fpel[92][4];
398 const uint8_t *chroma_qp_table; /* includes both the nonlinear luma->chroma mapping and chroma_qp_offset */
400 ALIGNED_16( uint32_t nr_residual_sum[2][64] );
401 ALIGNED_16( uint16_t nr_offset[2][64] );
402 uint32_t nr_count[2];
405 x264_slice_header_t sh;
412 /* Frames to be encoded (whose types have been decided) */
413 x264_frame_t **current;
414 /* Unused frames: 0 = fenc, 1 = fdec */
415 x264_frame_t **unused[2];
417 /* Unused blank frames (for duplicates) */
418 x264_frame_t **blank_unused;
420 /* frames used for reference + sentinels */
421 x264_frame_t *reference[16+2];
423 int i_last_keyframe; /* Frame number of the last keyframe */
425 int i_input; /* Number of input frames already accepted */
427 int i_max_dpb; /* Number of frames allocated in the decoded picture buffer */
430 int i_delay; /* Number of frames buffered for B reordering */
432 int64_t i_bframe_delay_time;
433 int64_t i_init_delta;
434 int64_t i_prev_dts[2];
435 int b_have_lowres; /* Whether 1/2 resolution luma planes are being used */
436 int b_have_sub8x8_esa;
439 /* current frame being encoded */
442 /* frame being reconstructed */
445 /* references lists */
447 x264_frame_t *fref0[16+3]; /* ref list 0 */
449 x264_frame_t *fref1[16+3]; /* ref list 1 */
450 int b_ref_reorder[2];
454 /* Current MB DCT coeffs */
457 ALIGNED_16( int16_t luma16x16_dc[16] );
458 ALIGNED_16( int16_t chroma_dc[2][4] );
459 // FIXME share memory?
460 ALIGNED_16( int16_t luma8x8[4][64] );
461 ALIGNED_16( int16_t luma4x4[16+8][16] );
464 /* MB table and cache for current frame/mb */
467 int i_mb_count; /* number of mbs in a frame */
481 /* Search parameters */
486 int b_noise_reduction;
488 int i_psy_rd; /* Psy RD strength--fixed point value*/
489 int i_psy_trellis; /* Psy trellis strength--fixed point value*/
493 /* Allowed qpel MV range to stay within the picture + emulated edge pixels */
496 /* Subpel MV range for motion search.
497 * same mv_min/max but includes levels' i_mv_range. */
500 /* Fullpel MV range for motion search */
504 /* neighboring MBs */
505 unsigned int i_neighbour;
506 unsigned int i_neighbour8[4]; /* neighbours of each 8x8 or 4x4 block that are available */
507 unsigned int i_neighbour4[16]; /* at the time the block is coded */
508 unsigned int i_neighbour_intra; /* for constrained intra pred */
511 int i_mb_type_topleft;
512 int i_mb_type_topright;
516 /**** thread synchronization ends here ****/
517 /* subsequent variables are either thread-local or constant,
518 * and won't be copied from one thread to another */
521 int8_t *type; /* mb type */
522 int8_t *qp; /* mb qp */
523 int16_t *cbp; /* mb cbp: 0x0?: luma, 0x?0: chroma, 0x100: luma dc, 0x0200 and 0x0400: chroma dc (all set for PCM)*/
524 int8_t (*intra4x4_pred_mode)[8]; /* intra4x4 pred mode. for non I4x4 set to I_PRED_4x4_DC(2) */
525 /* actually has only 7 entries; set to 8 for write-combining optimizations */
526 uint8_t (*non_zero_count)[16+4+4]; /* nzc. for I_PCM set to 16 */
527 int8_t *chroma_pred_mode; /* chroma_pred_mode. cabac only. for non intra I_PRED_CHROMA_DC(0) */
528 int16_t (*mv[2])[2]; /* mb mv. set to 0 for intra mb */
529 int16_t (*mvd[2])[2]; /* mb mv difference with predict. set to 0 if intra. cabac only */
530 int8_t *ref[2]; /* mb ref. set to -1 if non used (intra or Lx only) */
531 int16_t (*mvr[2][32])[2]; /* 16x16 mv for each possible ref */
532 int8_t *skipbp; /* block pattern for SKIP or DIRECT (sub)mbs. B-frames + cabac only */
533 int8_t *mb_transform_size; /* transform_size_8x8_flag of each mb */
534 uint8_t *intra_border_backup[2][3]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
536 /* buffer for weighted versions of the reference frames */
537 uint8_t *p_weight_buf[16];
542 ALIGNED_4( uint8_t i_sub_partition[4] );
548 int i_intra16x16_pred_mode;
549 int i_chroma_pred_mode;
551 /* skip flags for i4x4 and i8x8
552 * 0 = encode as normal.
553 * 1 (non-RD only) = the DCT is still in h->dct, restore fdec and skip reconstruction.
554 * 2 (RD only) = the DCT has since been overwritten by RD; restore that too. */
556 /* skip flag for motion compensation */
557 /* if we've already done MC, we don't need to do it again */
559 /* set to true if we are re-encoding a macroblock. */
561 int ip_offset; /* Used by PIR to offset the quantizer of intra-refresh blocks. */
565 /* space for p_fenc and p_fdec */
566 #define FENC_STRIDE 16
567 #define FDEC_STRIDE 32
568 ALIGNED_16( uint8_t fenc_buf[24*FENC_STRIDE] );
569 ALIGNED_16( uint8_t fdec_buf[27*FDEC_STRIDE] );
571 /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */
572 ALIGNED_16( uint8_t i4x4_fdec_buf[16*16] );
573 ALIGNED_16( uint8_t i8x8_fdec_buf[16*16] );
574 ALIGNED_16( int16_t i8x8_dct_buf[3][64] );
575 ALIGNED_16( int16_t i4x4_dct_buf[15][16] );
576 uint32_t i4x4_nnz_buf[4];
577 uint32_t i8x8_nnz_buf[4];
581 /* Psy trellis DCT data */
582 ALIGNED_16( int16_t fenc_dct8[4][64] );
583 ALIGNED_16( int16_t fenc_dct4[16][16] );
585 /* Psy RD SATD scores */
591 /* pointer over mb of the frame to be compressed */
593 /* pointer to the actual source frame, not a block copy */
594 uint8_t *p_fenc_plane[3];
596 /* pointer over mb of the frame to be reconstructed */
599 /* pointer over mb of the references */
601 uint8_t *p_fref[2][32][4+2]; /* last: lN, lH, lV, lHV, cU, cV */
602 uint8_t *p_fref_w[32]; /* weighted fullpel luma */
603 uint16_t *p_integral[2][16];
612 /* real intra4x4_pred_mode if I_4X4 or I_8X8, I_PRED_4x4_DC if mb available, -1 if not */
613 ALIGNED_8( int8_t intra4x4_pred_mode[X264_SCAN8_SIZE] );
615 /* i_non_zero_count if available else 0x80 */
616 ALIGNED_4( uint8_t non_zero_count[X264_SCAN8_SIZE] );
618 /* -1 if unused, -2 if unavailable */
619 ALIGNED_4( int8_t ref[2][X264_SCAN8_SIZE] );
621 /* 0 if not available */
622 ALIGNED_16( int16_t mv[2][X264_SCAN8_SIZE][2] );
623 ALIGNED_8( int16_t mvd[2][X264_SCAN8_SIZE][2] );
625 /* 1 if SKIP or DIRECT. set only for B-frames + CABAC */
626 ALIGNED_4( int8_t skip[X264_SCAN8_SIZE] );
628 ALIGNED_4( int16_t direct_mv[2][4][2] );
629 ALIGNED_4( int8_t direct_ref[2][4] );
630 ALIGNED_4( int16_t pskip_mv[2] );
632 /* number of neighbors (top and left) that used 8x8 dct */
633 int i_neighbour_transform_size;
634 int i_neighbour_interlaced;
642 int i_qp; /* current qp */
644 int i_last_qp; /* last qp */
645 int i_last_dqp; /* last delta qp */
646 int b_variable_qp; /* whether qp is allowed to vary per macroblock */
648 int b_direct_auto_read; /* take stats for --direct auto from the 2pass log */
649 int b_direct_auto_write; /* analyse direct modes, to use and/or save */
652 int i_trellis_lambda2[2][2]; /* [luma,chroma][inter,intra] */
654 int i_chroma_lambda2_offset;
656 /* B_direct and weighted prediction */
657 int16_t dist_scale_factor_buf[2][16][2];
658 int16_t (*dist_scale_factor)[2];
659 int8_t bipred_weight_buf[2][32][4];
660 int8_t (*bipred_weight)[4];
661 /* maps fref1[0]'s ref indices into the current list0 */
662 #define map_col_to_list0(col) h->mb.map_col_to_list0[(col)+2]
663 int8_t map_col_to_list0[18];
664 int ref_blind_dupe; /* The index of the blind reference frame duplicate. */
667 /* rate control encoding only */
668 x264_ratecontrol_t *rc;
673 /* Current frame stats */
676 /* MV bits (MV+Ref+Block Type) */
678 /* Texture bits (DCT coefs) */
687 int i_mb_count_8x8dct[2];
688 int i_mb_count_ref[2][32];
689 int i_mb_partition[17];
691 int i_mb_pred_mode[3][13];
692 /* Adaptive direct mv pred */
693 int i_direct_score[2];
699 /* Cumulated stats */
702 int i_frame_count[5];
703 int64_t i_frame_size[5];
704 double f_frame_qp[5];
705 int i_consecutive_bframes[X264_BFRAME_MAX+1];
707 int64_t i_ssd_global[5];
708 double f_psnr_average[5];
709 double f_psnr_mean_y[5];
710 double f_psnr_mean_u[5];
711 double f_psnr_mean_v[5];
712 double f_ssim_mean_y[5];
714 int64_t i_mb_count[5][19];
715 int64_t i_mb_partition[2][17];
716 int64_t i_mb_count_8x8dct[2];
717 int64_t i_mb_count_ref[2][2][32];
719 int64_t i_mb_pred_mode[3][13];
721 int i_direct_score[2];
722 int i_direct_frames[2];
723 /* num p-frames weighted */
728 void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */
730 /* CPU functions dependents */
731 x264_predict_t predict_16x16[4+3];
732 x264_predict_t predict_8x8c[4+3];
733 x264_predict8x8_t predict_8x8[9+3];
734 x264_predict_t predict_4x4[9+3];
735 x264_predict_8x8_filter_t predict_8x8_filter;
737 x264_pixel_function_t pixf;
738 x264_mc_functions_t mc;
739 x264_dct_function_t dctf;
740 x264_zigzag_function_t zigzagf;
741 x264_quant_function_t quantf;
742 x264_deblock_function_t loopf;
744 #ifdef HAVE_VISUALIZE
745 struct visualize_t *visualize;
747 x264_lookahead_t *lookahead;
750 // included at the end because it needs x264_t
751 #include "macroblock.h"
754 #include "x86/util.h"