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 /****************************************************************************
65 ****************************************************************************/
84 /****************************************************************************
86 ****************************************************************************/
87 /* x264_malloc : will do or emulate a memalign
88 * you have to use x264_free for buffers allocated with x264_malloc */
89 void *x264_malloc( int );
90 void x264_free( void * );
92 /* x264_slurp_file: malloc space for the whole file and read it */
93 char *x264_slurp_file( const char *filename );
95 /* mdate: return the current date in microsecond */
96 int64_t x264_mdate( void );
98 /* x264_param2string: return a (malloced) string containing most of
99 * the encoding options */
100 char *x264_param2string( x264_param_t *p, int b_res );
102 int x264_nal_encode( uint8_t *dst, int b_annexb, x264_nal_t *nal );
105 void x264_log( x264_t *h, int i_level, const char *psz_fmt, ... );
107 void x264_reduce_fraction( int *n, int *d );
108 void x264_init_vlc_tables();
110 static inline uint8_t x264_clip_uint8( int x )
112 return x&(~255) ? (-x)>>31 : x;
115 static inline int x264_clip3( int v, int i_min, int i_max )
117 return ( (v < i_min) ? i_min : (v > i_max) ? i_max : v );
120 static inline double x264_clip3f( double v, double f_min, double f_max )
122 return ( (v < f_min) ? f_min : (v > f_max) ? f_max : v );
125 static inline int x264_median( int a, int b, int c )
127 int t = (a-b)&((a-b)>>31);
130 b -= (b-c)&((b-c)>>31);
131 b += (a-b)&((a-b)>>31);
135 static inline void x264_median_mv( int16_t *dst, int16_t *a, int16_t *b, int16_t *c )
137 dst[0] = x264_median( a[0], b[0], c[0] );
138 dst[1] = x264_median( a[1], b[1], c[1] );
141 static inline int x264_predictor_difference( int16_t (*mvc)[2], intptr_t i_mvc )
144 for( i = 0; i < i_mvc-1; i++ )
146 sum += abs( mvc[i][0] - mvc[i+1][0] )
147 + abs( mvc[i][1] - mvc[i+1][1] );
152 static inline uint32_t x264_cabac_amvd_sum( int16_t *mvdleft, int16_t *mvdtop )
154 int amvd0 = abs(mvdleft[0]) + abs(mvdtop[0]);
155 int amvd1 = abs(mvdleft[1]) + abs(mvdtop[1]);
156 amvd0 = (amvd0 > 2) + (amvd0 > 32);
157 amvd1 = (amvd1 > 2) + (amvd1 > 32);
158 return amvd0 + (amvd1<<16);
161 extern const uint8_t x264_exp2_lut[64];
162 extern const float x264_log2_lut[128];
163 extern const float x264_log2_lz_lut[32];
165 /* Not a general-purpose function; multiplies input by -1/6 to convert
167 static ALWAYS_INLINE int x264_exp2fix8( float x )
169 int i = x*(-64.f/6.f) + 512.5f;
170 if( i < 0 ) return 0;
171 if( i > 1023 ) return 0xffff;
172 return (x264_exp2_lut[i&63]+256) << (i>>6) >> 8;
175 static ALWAYS_INLINE float x264_log2( uint32_t x )
177 int lz = x264_clz( x );
178 return x264_log2_lut[(x<<lz>>24)&0x7f] + x264_log2_lz_lut[lz];
181 /****************************************************************************
183 ****************************************************************************/
193 static const char slice_type_to_char[] = { 'P', 'B', 'I', 'S', 'S' };
212 int i_idr_pic_id; /* -1 if nal_type != 5 */
215 int i_delta_poc_bottom;
218 int i_redundant_pic_cnt;
220 int b_direct_spatial_mv_pred;
222 int b_num_ref_idx_override;
223 int i_num_ref_idx_l0_active;
224 int i_num_ref_idx_l1_active;
226 int b_ref_pic_list_reordering_l0;
227 int b_ref_pic_list_reordering_l1;
232 } ref_pic_list_order[2][16];
234 int i_mmco_remove_from_end;
235 int i_mmco_command_count;
236 struct /* struct for future expansion */
238 int i_difference_of_pic_nums;
242 int i_cabac_init_idc;
249 /* deblocking filter */
250 int i_disable_deblocking_filter_idc;
251 int i_alpha_c0_offset;
254 } x264_slice_header_t;
256 typedef struct x264_lookahead_t
258 volatile uint8_t b_exit_thread;
259 uint8_t b_thread_active;
260 uint8_t b_analyse_keyframe;
262 int i_slicetype_length;
263 x264_frame_t *last_nonb;
264 x264_synch_frame_list_t ifbuf;
265 x264_synch_frame_list_t next;
266 x264_synch_frame_list_t ofbuf;
271 #define X264_SCAN8_SIZE (6*8)
272 #define X264_SCAN8_0 (4+1*8)
274 static const int x264_scan8[16+2*4+3] =
277 4+1*8, 5+1*8, 4+2*8, 5+2*8,
278 6+1*8, 7+1*8, 6+2*8, 7+2*8,
279 4+3*8, 5+3*8, 4+4*8, 5+4*8,
280 6+3*8, 7+3*8, 6+4*8, 7+4*8,
306 typedef struct x264_ratecontrol_t x264_ratecontrol_t;
310 /* encoder parameters */
313 x264_t *thread[X264_THREAD_MAX+1];
314 x264_pthread_t thread_handle;
316 int i_thread_phase; /* which thread to use for the next frame */
318 /* bitstream output */
322 int i_nals_allocated;
324 int i_bitstream; /* size of p_bitstream */
325 uint8_t *p_bitstream; /* will hold data for all nal */
332 /**** thread synchronization starts here ****/
334 /* frame number/poc */
337 int i_frame_offset; /* decoding only */
338 int i_frame_num; /* decoding only */
339 int i_poc_msb; /* decoding only */
340 int i_poc_lsb; /* decoding only */
341 int i_poc; /* decoding only */
343 int i_thread_num; /* threads only */
344 int i_nal_type; /* threads only */
345 int i_nal_ref_idc; /* threads only */
347 /* We use only one SPS and one PPS */
348 x264_sps_t sps_array[1];
350 x264_pps_t pps_array[1];
354 /* quantization matrix for decoding, [cqm][qp%6][coef] */
355 int (*dequant4_mf[4])[16]; /* [4][6][16] */
356 int (*dequant8_mf[2])[64]; /* [2][6][64] */
357 /* quantization matrix for trellis, [cqm][qp][coef] */
358 int (*unquant4_mf[4])[16]; /* [4][52][16] */
359 int (*unquant8_mf[2])[64]; /* [2][52][64] */
360 /* quantization matrix for deadzone */
361 uint16_t (*quant4_mf[4])[16]; /* [4][52][16] */
362 uint16_t (*quant8_mf[2])[64]; /* [2][52][64] */
363 uint16_t (*quant4_bias[4])[16]; /* [4][52][16] */
364 uint16_t (*quant8_bias[2])[64]; /* [2][52][64] */
366 /* mv/ref cost arrays. Indexed by lambda instead of
367 * qp because, due to rounding, some quantizers share
368 * lambdas. This saves memory. */
369 uint16_t *cost_mv[92];
370 uint16_t *cost_mv_fpel[92][4];
372 const uint8_t *chroma_qp_table; /* includes both the nonlinear luma->chroma mapping and chroma_qp_offset */
374 ALIGNED_16( uint32_t nr_residual_sum[2][64] );
375 ALIGNED_16( uint16_t nr_offset[2][64] );
376 uint32_t nr_count[2];
379 x264_slice_header_t sh;
386 /* Frames to be encoded (whose types have been decided) */
387 x264_frame_t **current;
388 /* Unused frames: 0 = fenc, 1 = fdec */
389 x264_frame_t **unused[2];
391 /* frames used for reference + sentinels */
392 x264_frame_t *reference[16+2];
394 int i_last_idr; /* Frame number of the last IDR */
396 int i_input; /* Number of input frames already accepted */
398 int i_max_dpb; /* Number of frames allocated in the decoded picture buffer */
401 int i_delay; /* Number of frames buffered for B reordering */
402 int b_have_lowres; /* Whether 1/2 resolution luma planes are being used */
403 int b_have_sub8x8_esa;
406 /* current frame being encoded */
409 /* frame being reconstructed */
412 /* references lists */
414 x264_frame_t *fref0[16+3]; /* ref list 0 */
416 x264_frame_t *fref1[16+3]; /* ref list 1 */
417 int b_ref_reorder[2];
421 /* Current MB DCT coeffs */
424 ALIGNED_16( int16_t luma16x16_dc[16] );
425 ALIGNED_16( int16_t chroma_dc[2][4] );
426 // FIXME share memory?
427 ALIGNED_16( int16_t luma8x8[4][64] );
428 ALIGNED_16( int16_t luma4x4[16+8][16] );
431 /* MB table and cache for current frame/mb */
434 int i_mb_count; /* number of mbs in a frame */
448 /* Search parameters */
453 int b_noise_reduction;
454 int i_psy_rd; /* Psy RD strength--fixed point value*/
455 int i_psy_trellis; /* Psy trellis strength--fixed point value*/
459 /* Allowed qpel MV range to stay within the picture + emulated edge pixels */
462 /* Subpel MV range for motion search.
463 * same mv_min/max but includes levels' i_mv_range. */
466 /* Fullpel MV range for motion search */
470 /* neighboring MBs */
471 unsigned int i_neighbour;
472 unsigned int i_neighbour8[4]; /* neighbours of each 8x8 or 4x4 block that are available */
473 unsigned int i_neighbour4[16]; /* at the time the block is coded */
474 unsigned int i_neighbour_intra; /* for constrained intra pred */
477 int i_mb_type_topleft;
478 int i_mb_type_topright;
482 /**** thread synchronization ends here ****/
483 /* subsequent variables are either thread-local or constant,
484 * and won't be copied from one thread to another */
487 int8_t *type; /* mb type */
488 int8_t *qp; /* mb qp */
489 int16_t *cbp; /* mb cbp: 0x0?: luma, 0x?0: chroma, 0x100: luma dc, 0x0200 and 0x0400: chroma dc (all set for PCM)*/
490 int8_t (*intra4x4_pred_mode)[8]; /* intra4x4 pred mode. for non I4x4 set to I_PRED_4x4_DC(2) */
491 /* actually has only 7 entries; set to 8 for write-combining optimizations */
492 uint8_t (*non_zero_count)[16+4+4]; /* nzc. for I_PCM set to 16 */
493 int8_t *chroma_pred_mode; /* chroma_pred_mode. cabac only. for non intra I_PRED_CHROMA_DC(0) */
494 int16_t (*mv[2])[2]; /* mb mv. set to 0 for intra mb */
495 int16_t (*mvd[2])[2]; /* mb mv difference with predict. set to 0 if intra. cabac only */
496 int8_t *ref[2]; /* mb ref. set to -1 if non used (intra or Lx only) */
497 int16_t (*mvr[2][32])[2]; /* 16x16 mv for each possible ref */
498 int8_t *skipbp; /* block pattern for SKIP or DIRECT (sub)mbs. B-frames + cabac only */
499 int8_t *mb_transform_size; /* transform_size_8x8_flag of each mb */
500 uint8_t *intra_border_backup[2][3]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
501 uint8_t (*nnz_backup)[16]; /* when using cavlc + 8x8dct, the deblocker uses a modified nnz */
506 ALIGNED_4( uint8_t i_sub_partition[4] );
512 int i_intra16x16_pred_mode;
513 int i_chroma_pred_mode;
515 /* skip flags for i4x4 and i8x8
516 * 0 = encode as normal.
517 * 1 (non-RD only) = the DCT is still in h->dct, restore fdec and skip reconstruction.
518 * 2 (RD only) = the DCT has since been overwritten by RD; restore that too. */
520 /* skip flag for motion compensation */
521 /* if we've already done MC, we don't need to do it again */
523 /* set to true if we are re-encoding a macroblock. */
528 /* space for p_fenc and p_fdec */
529 #define FENC_STRIDE 16
530 #define FDEC_STRIDE 32
531 ALIGNED_16( uint8_t fenc_buf[24*FENC_STRIDE] );
532 ALIGNED_16( uint8_t fdec_buf[27*FDEC_STRIDE] );
534 /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */
535 ALIGNED_16( uint8_t i4x4_fdec_buf[16*16] );
536 ALIGNED_16( uint8_t i8x8_fdec_buf[16*16] );
537 ALIGNED_16( int16_t i8x8_dct_buf[3][64] );
538 ALIGNED_16( int16_t i4x4_dct_buf[15][16] );
539 uint32_t i4x4_nnz_buf[4];
540 uint32_t i8x8_nnz_buf[4];
544 /* Psy trellis DCT data */
545 ALIGNED_16( int16_t fenc_dct8[4][64] );
546 ALIGNED_16( int16_t fenc_dct4[16][16] );
548 /* Psy RD SATD scores */
554 /* pointer over mb of the frame to be compressed */
556 /* pointer to the actual source frame, not a block copy */
557 uint8_t *p_fenc_plane[3];
559 /* pointer over mb of the frame to be reconstructed */
562 /* pointer over mb of the references */
564 uint8_t *p_fref[2][32][4+2]; /* last: lN, lH, lV, lHV, cU, cV */
565 uint16_t *p_integral[2][16];
574 /* real intra4x4_pred_mode if I_4X4 or I_8X8, I_PRED_4x4_DC if mb available, -1 if not */
575 int8_t intra4x4_pred_mode[X264_SCAN8_SIZE];
577 /* i_non_zero_count if available else 0x80 */
578 uint8_t non_zero_count[X264_SCAN8_SIZE];
580 /* -1 if unused, -2 if unavailable */
581 ALIGNED_4( int8_t ref[2][X264_SCAN8_SIZE] );
583 /* 0 if not available */
584 ALIGNED_16( int16_t mv[2][X264_SCAN8_SIZE][2] );
585 ALIGNED_8( int16_t mvd[2][X264_SCAN8_SIZE][2] );
587 /* 1 if SKIP or DIRECT. set only for B-frames + CABAC */
588 ALIGNED_4( int8_t skip[X264_SCAN8_SIZE] );
590 ALIGNED_16( int16_t direct_mv[2][X264_SCAN8_SIZE][2] );
591 ALIGNED_4( int8_t direct_ref[2][X264_SCAN8_SIZE] );
592 ALIGNED_4( int16_t pskip_mv[2] );
594 /* number of neighbors (top and left) that used 8x8 dct */
595 int i_neighbour_transform_size;
596 int i_neighbour_interlaced;
604 int i_qp; /* current qp */
606 int i_last_qp; /* last qp */
607 int i_last_dqp; /* last delta qp */
608 int b_variable_qp; /* whether qp is allowed to vary per macroblock */
610 int b_direct_auto_read; /* take stats for --direct auto from the 2pass log */
611 int b_direct_auto_write; /* analyse direct modes, to use and/or save */
614 int i_trellis_lambda2[2][2]; /* [luma,chroma][inter,intra] */
616 int i_chroma_lambda2_offset;
618 /* B_direct and weighted prediction */
619 int16_t dist_scale_factor[16][2];
620 int16_t bipred_weight[32][4];
621 /* maps fref1[0]'s ref indices into the current list0 */
622 int8_t map_col_to_list0_buf[2]; // for negative indices
623 int8_t map_col_to_list0[16];
626 /* rate control encoding only */
627 x264_ratecontrol_t *rc;
632 /* Current frame stats */
635 /* MV bits (MV+Ref+Block Type) */
637 /* Texture bits (DCT coefs) */
646 int i_mb_count_8x8dct[2];
647 int i_mb_count_ref[2][32];
648 int i_mb_partition[17];
650 int i_mb_pred_mode[3][13];
651 /* Adaptive direct mv pred */
652 int i_direct_score[2];
658 /* Cumulated stats */
661 int i_frame_count[5];
662 int64_t i_frame_size[5];
663 double f_frame_qp[5];
664 int i_consecutive_bframes[X264_BFRAME_MAX+1];
666 int64_t i_ssd_global[5];
667 double f_psnr_average[5];
668 double f_psnr_mean_y[5];
669 double f_psnr_mean_u[5];
670 double f_psnr_mean_v[5];
671 double f_ssim_mean_y[5];
673 int64_t i_mb_count[5][19];
674 int64_t i_mb_partition[2][17];
675 int64_t i_mb_count_8x8dct[2];
676 int64_t i_mb_count_ref[2][2][32];
678 int64_t i_mb_pred_mode[3][13];
680 int i_direct_score[2];
681 int i_direct_frames[2];
685 void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */
687 /* CPU functions dependents */
688 x264_predict_t predict_16x16[4+3];
689 x264_predict_t predict_8x8c[4+3];
690 x264_predict8x8_t predict_8x8[9+3];
691 x264_predict_t predict_4x4[9+3];
692 x264_predict_8x8_filter_t predict_8x8_filter;
694 x264_pixel_function_t pixf;
695 x264_mc_functions_t mc;
696 x264_dct_function_t dctf;
697 x264_zigzag_function_t zigzagf;
698 x264_quant_function_t quantf;
699 x264_deblock_function_t loopf;
702 struct visualize_t *visualize;
704 x264_lookahead_t *lookahead;
707 // included at the end because it needs x264_t
708 #include "macroblock.h"
711 #include "x86/util.h"