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 FIX8(f) ((int)(f*(1<<8)+.5))
39 #define CHECKED_MALLOC( var, size )\
41 var = x264_malloc( size );\
45 #define CHECKED_MALLOCZERO( var, size )\
47 CHECKED_MALLOC( var, size );\
48 memset( var, 0, size );\
51 #define X264_BFRAME_MAX 16
52 #define X264_THREAD_MAX 128
53 #define X264_PCM_COST (386*8)
54 #define X264_LOOKAHEAD_MAX 250
56 // number of pixels (per thread) in progress at any given time.
57 // 16 for the macroblock in progress + 3 for deblocking + 3 for motion compensation filter + 2 for extra safety
58 #define X264_THREAD_HEIGHT 24
60 /****************************************************************************
62 ****************************************************************************/
81 /****************************************************************************
83 ****************************************************************************/
84 /* x264_malloc : will do or emulate a memalign
85 * you have to use x264_free for buffers allocated with x264_malloc */
86 void *x264_malloc( int );
87 void x264_free( void * );
89 /* x264_slurp_file: malloc space for the whole file and read it */
90 char *x264_slurp_file( const char *filename );
92 /* mdate: return the current date in microsecond */
93 int64_t x264_mdate( void );
95 /* x264_param2string: return a (malloced) string containing most of
96 * the encoding options */
97 char *x264_param2string( x264_param_t *p, int b_res );
100 void x264_log( x264_t *h, int i_level, const char *psz_fmt, ... );
102 void x264_reduce_fraction( int *n, int *d );
103 void x264_init_vlc_tables();
105 static inline uint8_t x264_clip_uint8( int x )
107 return x&(~255) ? (-x)>>31 : x;
110 static inline int x264_clip3( int v, int i_min, int i_max )
112 return ( (v < i_min) ? i_min : (v > i_max) ? i_max : v );
115 static inline double x264_clip3f( double v, double f_min, double f_max )
117 return ( (v < f_min) ? f_min : (v > f_max) ? f_max : v );
120 static inline int x264_median( int a, int b, int c )
122 int t = (a-b)&((a-b)>>31);
125 b -= (b-c)&((b-c)>>31);
126 b += (a-b)&((a-b)>>31);
130 static inline void x264_median_mv( int16_t *dst, int16_t *a, int16_t *b, int16_t *c )
132 dst[0] = x264_median( a[0], b[0], c[0] );
133 dst[1] = x264_median( a[1], b[1], c[1] );
136 static inline int x264_predictor_difference( int16_t (*mvc)[2], intptr_t i_mvc )
139 for( i = 0; i < i_mvc-1; i++ )
141 sum += abs( mvc[i][0] - mvc[i+1][0] )
142 + abs( mvc[i][1] - mvc[i+1][1] );
147 static inline uint32_t x264_cabac_amvd_sum( int16_t *mvdleft, int16_t *mvdtop )
149 int amvd0 = abs(mvdleft[0]) + abs(mvdtop[0]);
150 int amvd1 = abs(mvdleft[1]) + abs(mvdtop[1]);
151 amvd0 = (amvd0 > 2) + (amvd0 > 32);
152 amvd1 = (amvd1 > 2) + (amvd1 > 32);
153 return amvd0 + (amvd1<<16);
156 extern const uint8_t x264_exp2_lut[64];
157 extern const float x264_log2_lut[128];
158 extern const float x264_log2_lz_lut[32];
160 /* Not a general-purpose function; multiplies input by -1/6 to convert
162 static ALWAYS_INLINE int x264_exp2fix8( float x )
164 if( x >= 512.f/6.f ) return 0;
165 if( x <= -512.f/6.f ) return 0xffff;
166 int i = x*(-64.f/6.f) + 512;
167 return (x264_exp2_lut[i&63]+256) << (i>>6) >> 8;
170 static ALWAYS_INLINE float x264_log2( uint32_t x )
172 int lz = x264_clz( x );
173 return x264_log2_lut[(x<<lz>>24)&0x7f] + x264_log2_lz_lut[lz];
176 /****************************************************************************
178 ****************************************************************************/
188 static const char slice_type_to_char[] = { 'P', 'B', 'I', 'S', 'S' };
207 int i_idr_pic_id; /* -1 if nal_type != 5 */
210 int i_delta_poc_bottom;
213 int i_redundant_pic_cnt;
215 int b_direct_spatial_mv_pred;
217 int b_num_ref_idx_override;
218 int i_num_ref_idx_l0_active;
219 int i_num_ref_idx_l1_active;
221 int b_ref_pic_list_reordering_l0;
222 int b_ref_pic_list_reordering_l1;
226 } ref_pic_list_order[2][16];
228 int i_cabac_init_idc;
235 /* deblocking filter */
236 int i_disable_deblocking_filter_idc;
237 int i_alpha_c0_offset;
240 } x264_slice_header_t;
244 #define X264_SCAN8_SIZE (6*8)
245 #define X264_SCAN8_0 (4+1*8)
247 static const int x264_scan8[16+2*4+3] =
250 4+1*8, 5+1*8, 4+2*8, 5+2*8,
251 6+1*8, 7+1*8, 6+2*8, 7+2*8,
252 4+3*8, 5+3*8, 4+4*8, 5+4*8,
253 6+3*8, 7+3*8, 6+4*8, 7+4*8,
279 typedef struct x264_ratecontrol_t x264_ratecontrol_t;
283 /* encoder parameters */
286 x264_t *thread[X264_THREAD_MAX];
287 x264_pthread_t thread_handle;
289 int i_thread_phase; /* which thread to use for the next frame */
291 /* bitstream output */
295 int i_nals_allocated;
297 int i_bitstream; /* size of p_bitstream */
298 uint8_t *p_bitstream; /* will hold data for all nal */
303 /**** thread synchronization starts here ****/
305 /* frame number/poc */
308 int i_frame_offset; /* decoding only */
309 int i_frame_num; /* decoding only */
310 int i_poc_msb; /* decoding only */
311 int i_poc_lsb; /* decoding only */
312 int i_poc; /* decoding only */
314 int i_thread_num; /* threads only */
315 int i_nal_type; /* threads only */
316 int i_nal_ref_idc; /* threads only */
318 /* We use only one SPS and one PPS */
319 x264_sps_t sps_array[1];
321 x264_pps_t pps_array[1];
325 /* quantization matrix for decoding, [cqm][qp%6][coef_y][coef_x] */
326 int (*dequant4_mf[4])[4][4]; /* [4][6][4][4] */
327 int (*dequant8_mf[2])[8][8]; /* [2][6][8][8] */
328 /* quantization matrix for trellis, [cqm][qp][coef] */
329 int (*unquant4_mf[4])[16]; /* [4][52][16] */
330 int (*unquant8_mf[2])[64]; /* [2][52][64] */
331 /* quantization matrix for deadzone */
332 uint16_t (*quant4_mf[4])[16]; /* [4][52][16] */
333 uint16_t (*quant8_mf[2])[64]; /* [2][52][64] */
334 uint16_t (*quant4_bias[4])[16]; /* [4][52][16] */
335 uint16_t (*quant8_bias[2])[64]; /* [2][52][64] */
337 const uint8_t *chroma_qp_table; /* includes both the nonlinear luma->chroma mapping and chroma_qp_offset */
339 ALIGNED_16( uint32_t nr_residual_sum[2][64] );
340 ALIGNED_16( uint16_t nr_offset[2][64] );
341 uint32_t nr_count[2];
344 x264_slice_header_t sh;
351 /* Frames to be encoded (whose types have been decided) */
352 x264_frame_t *current[X264_LOOKAHEAD_MAX+3];
353 /* Temporary buffer (frames types not yet decided) */
354 x264_frame_t *next[X264_LOOKAHEAD_MAX+3];
356 x264_frame_t *unused[X264_LOOKAHEAD_MAX + X264_THREAD_MAX*2 + 16+4];
357 /* For adaptive B decision */
358 x264_frame_t *last_nonb;
360 /* frames used for reference + sentinels */
361 x264_frame_t *reference[16+2];
363 int i_last_idr; /* Frame number of the last IDR */
365 int i_input; /* Number of input frames already accepted */
367 int i_max_dpb; /* Number of frames allocated in the decoded picture buffer */
370 int i_delay; /* Number of frames buffered for B reordering */
371 int b_have_lowres; /* Whether 1/2 resolution luma planes are being used */
372 int b_have_sub8x8_esa;
375 /* current frame being encoded */
378 /* frame being reconstructed */
381 /* references lists */
383 x264_frame_t *fref0[16+3]; /* ref list 0 */
385 x264_frame_t *fref1[16+3]; /* ref list 1 */
386 int b_ref_reorder[2];
390 /* Current MB DCT coeffs */
393 ALIGNED_16( int16_t luma16x16_dc[16] );
394 ALIGNED_16( int16_t chroma_dc[2][4] );
395 // FIXME share memory?
396 ALIGNED_16( int16_t luma8x8[4][64] );
397 ALIGNED_16( int16_t luma4x4[16+8][16] );
400 /* MB table and cache for current frame/mb */
403 int i_mb_count; /* number of mbs in a frame */
417 /* Search parameters */
422 int b_noise_reduction;
423 int i_psy_rd; /* Psy RD strength--fixed point value*/
424 int i_psy_trellis; /* Psy trellis strength--fixed point value*/
428 /* Allowed qpel MV range to stay within the picture + emulated edge pixels */
431 /* Subpel MV range for motion search.
432 * same mv_min/max but includes levels' i_mv_range. */
435 /* Fullpel MV range for motion search */
439 /* neighboring MBs */
440 unsigned int i_neighbour;
441 unsigned int i_neighbour8[4]; /* neighbours of each 8x8 or 4x4 block that are available */
442 unsigned int i_neighbour4[16]; /* at the time the block is coded */
445 int i_mb_type_topleft;
446 int i_mb_type_topright;
450 /**** thread synchronization ends here ****/
451 /* subsequent variables are either thread-local or constant,
452 * and won't be copied from one thread to another */
455 int8_t *type; /* mb type */
456 int8_t *qp; /* mb qp */
457 int16_t *cbp; /* mb cbp: 0x0?: luma, 0x?0: chroma, 0x100: luma dc, 0x0200 and 0x0400: chroma dc (all set for PCM)*/
458 int8_t (*intra4x4_pred_mode)[8]; /* intra4x4 pred mode. for non I4x4 set to I_PRED_4x4_DC(2) */
459 /* actually has only 7 entries; set to 8 for write-combining optimizations */
460 uint8_t (*non_zero_count)[16+4+4]; /* nzc. for I_PCM set to 16 */
461 int8_t *chroma_pred_mode; /* chroma_pred_mode. cabac only. for non intra I_PRED_CHROMA_DC(0) */
462 int16_t (*mv[2])[2]; /* mb mv. set to 0 for intra mb */
463 int16_t (*mvd[2])[2]; /* mb mv difference with predict. set to 0 if intra. cabac only */
464 int8_t *ref[2]; /* mb ref. set to -1 if non used (intra or Lx only) */
465 int16_t (*mvr[2][32])[2]; /* 16x16 mv for each possible ref */
466 int8_t *skipbp; /* block pattern for SKIP or DIRECT (sub)mbs. B-frames + cabac only */
467 int8_t *mb_transform_size; /* transform_size_8x8_flag of each mb */
468 uint8_t *intra_border_backup[2][3]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
469 uint8_t (*nnz_backup)[16]; /* when using cavlc + 8x8dct, the deblocker uses a modified nnz */
474 ALIGNED_4( uint8_t i_sub_partition[4] );
480 int i_intra16x16_pred_mode;
481 int i_chroma_pred_mode;
483 /* skip flags for i4x4 and i8x8
484 * 0 = encode as normal.
485 * 1 (non-RD only) = the DCT is still in h->dct, restore fdec and skip reconstruction.
486 * 2 (RD only) = the DCT has since been overwritten by RD; restore that too. */
488 /* skip flag for motion compensation */
489 /* if we've already done MC, we don't need to do it again */
491 /* set to true if we are re-encoding a macroblock. */
496 /* space for p_fenc and p_fdec */
497 #define FENC_STRIDE 16
498 #define FDEC_STRIDE 32
499 ALIGNED_16( uint8_t fenc_buf[24*FENC_STRIDE] );
500 ALIGNED_16( uint8_t fdec_buf[27*FDEC_STRIDE] );
502 /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */
503 ALIGNED_16( uint8_t i4x4_fdec_buf[16*16] );
504 ALIGNED_16( uint8_t i8x8_fdec_buf[16*16] );
505 ALIGNED_16( int16_t i8x8_dct_buf[3][64] );
506 ALIGNED_16( int16_t i4x4_dct_buf[15][16] );
507 uint32_t i4x4_nnz_buf[4];
508 uint32_t i8x8_nnz_buf[4];
512 /* Psy trellis DCT data */
513 ALIGNED_16( int16_t fenc_dct8[4][64] );
514 ALIGNED_16( int16_t fenc_dct4[16][16] );
516 /* Psy RD SATD scores */
522 /* pointer over mb of the frame to be compressed */
524 /* pointer to the actual source frame, not a block copy */
525 uint8_t *p_fenc_plane[3];
527 /* pointer over mb of the frame to be reconstructed */
530 /* pointer over mb of the references */
532 uint8_t *p_fref[2][32][4+2]; /* last: lN, lH, lV, lHV, cU, cV */
533 uint16_t *p_integral[2][16];
542 /* real intra4x4_pred_mode if I_4X4 or I_8X8, I_PRED_4x4_DC if mb available, -1 if not */
543 int8_t intra4x4_pred_mode[X264_SCAN8_SIZE];
545 /* i_non_zero_count if available else 0x80 */
546 uint8_t non_zero_count[X264_SCAN8_SIZE];
548 /* -1 if unused, -2 if unavailable */
549 ALIGNED_4( int8_t ref[2][X264_SCAN8_SIZE] );
551 /* 0 if not available */
552 ALIGNED_16( int16_t mv[2][X264_SCAN8_SIZE][2] );
553 ALIGNED_8( int16_t mvd[2][X264_SCAN8_SIZE][2] );
555 /* 1 if SKIP or DIRECT. set only for B-frames + CABAC */
556 ALIGNED_4( int8_t skip[X264_SCAN8_SIZE] );
558 ALIGNED_16( int16_t direct_mv[2][X264_SCAN8_SIZE][2] );
559 ALIGNED_4( int8_t direct_ref[2][X264_SCAN8_SIZE] );
560 ALIGNED_4( int16_t pskip_mv[2] );
562 /* number of neighbors (top and left) that used 8x8 dct */
563 int i_neighbour_transform_size;
564 int i_neighbour_interlaced;
572 int i_qp; /* current qp */
574 int i_last_qp; /* last qp */
575 int i_last_dqp; /* last delta qp */
576 int b_variable_qp; /* whether qp is allowed to vary per macroblock */
578 int b_direct_auto_read; /* take stats for --direct auto from the 2pass log */
579 int b_direct_auto_write; /* analyse direct modes, to use and/or save */
582 int i_trellis_lambda2[2][2]; /* [luma,chroma][inter,intra] */
584 int i_chroma_lambda2_offset;
586 /* B_direct and weighted prediction */
587 int16_t dist_scale_factor[16][2];
588 int16_t bipred_weight[32][4];
589 /* maps fref1[0]'s ref indices into the current list0 */
590 int8_t map_col_to_list0_buf[2]; // for negative indices
591 int8_t map_col_to_list0[16];
594 /* rate control encoding only */
595 x264_ratecontrol_t *rc;
600 /* Current frame stats */
603 /* MV bits (MV+Ref+Block Type) */
605 /* Texture bits (DCT coefs) */
614 int i_mb_count_8x8dct[2];
615 int i_mb_count_ref[2][32];
616 int i_mb_partition[17];
618 /* Adaptive direct mv pred */
619 int i_direct_score[2];
625 /* Cumulated stats */
628 int i_frame_count[5];
629 int64_t i_frame_size[5];
630 double f_frame_qp[5];
631 int i_consecutive_bframes[X264_BFRAME_MAX+1];
633 int64_t i_ssd_global[5];
634 double f_psnr_average[5];
635 double f_psnr_mean_y[5];
636 double f_psnr_mean_u[5];
637 double f_psnr_mean_v[5];
638 double f_ssim_mean_y[5];
640 int64_t i_mb_count[5][19];
641 int64_t i_mb_partition[2][17];
642 int64_t i_mb_count_8x8dct[2];
643 int64_t i_mb_count_ref[2][2][32];
646 int i_direct_score[2];
647 int i_direct_frames[2];
651 void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */
653 /* CPU functions dependents */
654 x264_predict_t predict_16x16[4+3];
655 x264_predict_t predict_8x8c[4+3];
656 x264_predict8x8_t predict_8x8[9+3];
657 x264_predict_t predict_4x4[9+3];
658 x264_predict_8x8_filter_t predict_8x8_filter;
660 x264_pixel_function_t pixf;
661 x264_mc_functions_t mc;
662 x264_dct_function_t dctf;
663 x264_zigzag_function_t zigzagf;
664 x264_quant_function_t quantf;
665 x264_deblock_function_t loopf;
668 struct visualize_t *visualize;
672 // included at the end because it needs x264_t
673 #include "macroblock.h"
676 #include "x86/util.h"