X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=common%2Fcommon.h;h=12e5763bb1bd6593e1672e65af97ec46074dfe59;hb=fa1e2b746d95575b5c5b8e49fcfcad3ded9a5420;hp=64947ed1dc4cdb03061e5ae3f313faea0c1d38e3;hpb=94b9141609d17ebbeb3184a8a5fc0660725a4cf2;p=x264 diff --git a/common/common.h b/common/common.h index 64947ed1..12e5763b 100644 --- a/common/common.h +++ b/common/common.h @@ -1,7 +1,7 @@ /***************************************************************************** * common.h: misc common functions ***************************************************************************** - * Copyright (C) 2003-2011 x264 project + * Copyright (C) 2003-2013 x264 project * * Authors: Laurent Aimar * Loren Merritt @@ -40,6 +40,7 @@ #define IS_DISPOSABLE(type) ( type == X264_TYPE_B ) #define FIX8(f) ((int)(f*(1<<8)+.5)) #define ALIGN(x,a) (((x)+((a)-1))&~((a)-1)) +#define ARRAY_ELEMS(a) ((sizeof(a))/(sizeof(a[0]))) #define CHECKED_MALLOC( var, size )\ do {\ @@ -53,10 +54,38 @@ do {\ memset( var, 0, size );\ } while( 0 ) +/* Macros for merging multiple allocations into a single large malloc, for improved + * use with huge pages. */ + +/* Needs to be enough to contain any set of buffers that use combined allocations */ +#define PREALLOC_BUF_SIZE 1024 + +#define PREALLOC_INIT\ + int prealloc_idx = 0;\ + size_t prealloc_size = 0;\ + uint8_t **preallocs[PREALLOC_BUF_SIZE]; + +#define PREALLOC( var, size )\ +do {\ + var = (void*)prealloc_size;\ + preallocs[prealloc_idx++] = (uint8_t**)&var;\ + prealloc_size += ALIGN(size, NATIVE_ALIGN);\ +} while(0) + +#define PREALLOC_END( ptr )\ +do {\ + CHECKED_MALLOC( ptr, prealloc_size );\ + while( prealloc_idx-- )\ + *preallocs[prealloc_idx] += (intptr_t)ptr;\ +} while(0) + +#define ARRAY_SIZE(array) (sizeof(array)/sizeof(array[0])) + #define X264_BFRAME_MAX 16 #define X264_REF_MAX 16 #define X264_THREAD_MAX 128 -#define X264_PCM_COST (384*BIT_DEPTH+16) +#define X264_LOOKAHEAD_THREAD_MAX 16 +#define X264_PCM_COST (FRAME_SIZE(256*BIT_DEPTH)+16) #define X264_LOOKAHEAD_MAX 250 #define QP_BD_OFFSET (6*(BIT_DEPTH-8)) #define QP_MAX_SPEC (51+QP_BD_OFFSET) @@ -92,6 +121,29 @@ do {\ #include #include +#if HAVE_INTERLACED +# define MB_INTERLACED h->mb.b_interlaced +# define SLICE_MBAFF h->sh.b_mbaff +# define PARAM_INTERLACED h->param.b_interlaced +#else +# define MB_INTERLACED 0 +# define SLICE_MBAFF 0 +# define PARAM_INTERLACED 0 +#endif + +#ifdef CHROMA_FORMAT +# define CHROMA_H_SHIFT (CHROMA_FORMAT == CHROMA_420 || CHROMA_FORMAT == CHROMA_422) +# define CHROMA_V_SHIFT (CHROMA_FORMAT == CHROMA_420) +#else +# define CHROMA_FORMAT h->sps->i_chroma_format_idc +# define CHROMA_H_SHIFT h->mb.chroma_h_shift +# define CHROMA_V_SHIFT h->mb.chroma_v_shift +#endif + +#define CHROMA_SIZE(s) ((s)>>(CHROMA_H_SHIFT+CHROMA_V_SHIFT)) +#define FRAME_SIZE(s) ((s)+2*CHROMA_SIZE(s)) +#define CHROMA444 (CHROMA_FORMAT == CHROMA_444) + /* Unions for type-punning. * Mn: load or store n bits, aligned, native-endian * CPn: copy n bits, aligned, native-endian @@ -133,43 +185,55 @@ typedef union { x264_uint128_t i; uint64_t a[2]; uint32_t b[4]; uint16_t c[8]; u #define CPPIXEL_X4(dst,src) MPIXEL_X4(dst) = MPIXEL_X4(src) -#define X264_SCAN8_SIZE (6*8) #define X264_SCAN8_LUMA_SIZE (5*8) +#define X264_SCAN8_SIZE (X264_SCAN8_LUMA_SIZE*3) #define X264_SCAN8_0 (4+1*8) -static const unsigned x264_scan8[16+2*4+3] = +/* Scan8 organization: + * 0 1 2 3 4 5 6 7 + * 0 DY y y y y y + * 1 y Y Y Y Y + * 2 y Y Y Y Y + * 3 y Y Y Y Y + * 4 y Y Y Y Y + * 5 DU u u u u u + * 6 u U U U U + * 7 u U U U U + * 8 u U U U U + * 9 u U U U U + * 10 DV v v v v v + * 11 v V V V V + * 12 v V V V V + * 13 v V V V V + * 14 v V V V V + * DY/DU/DV are for luma/chroma DC. + */ + +#define LUMA_DC 48 +#define CHROMA_DC 49 + +static const uint8_t x264_scan8[16*3 + 3] = { - /* Luma */ - 4+1*8, 5+1*8, 4+2*8, 5+2*8, - 6+1*8, 7+1*8, 6+2*8, 7+2*8, - 4+3*8, 5+3*8, 4+4*8, 5+4*8, - 6+3*8, 7+3*8, 6+4*8, 7+4*8, - - /* Cb */ - 1+1*8, 2+1*8, - 1+2*8, 2+2*8, - - /* Cr */ - 1+4*8, 2+4*8, - 1+5*8, 2+5*8, - - /* Luma DC */ - 4+5*8, - - /* Chroma DC */ - 6+5*8, 7+5*8 + 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8, + 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8, + 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8, + 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8, + 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8, + 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8, + 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8, + 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8, + 4+11*8, 5+11*8, 4+12*8, 5+12*8, + 6+11*8, 7+11*8, 6+12*8, 7+12*8, + 4+13*8, 5+13*8, 4+14*8, 5+14*8, + 6+13*8, 7+13*8, 6+14*8, 7+14*8, + 0+ 0*8, 0+ 5*8, 0+10*8 }; -/* - 0 1 2 3 4 5 6 7 - 0 - 1 B B L L L L - 2 B B L L L L - 3 L L L L - 4 R R L L L L - 5 R R Dy DuDv -*/ #include "x264.h" +#if HAVE_OPENCL +#include "opencl.h" +#endif +#include "cabac.h" #include "bitstream.h" #include "set.h" #include "predict.h" @@ -177,7 +241,6 @@ static const unsigned x264_scan8[16+2*4+3] = #include "mc.h" #include "frame.h" #include "dct.h" -#include "cabac.h" #include "quant.h" #include "cpu.h" #include "threadpool.h" @@ -205,8 +268,8 @@ void x264_log( x264_t *h, int i_level, const char *psz_fmt, ... ); void x264_reduce_fraction( uint32_t *n, uint32_t *d ); void x264_reduce_fraction64( uint64_t *n, uint64_t *d ); -void x264_cavlc_init( void ); -void x264_cabac_init( void ); +void x264_cavlc_init( x264_t *h ); +void x264_cabac_init( x264_t *h ); static ALWAYS_INLINE pixel x264_clip_pixel( int x ) { @@ -259,17 +322,6 @@ static ALWAYS_INLINE uint16_t x264_cabac_mvd_sum( uint8_t *mvdleft, uint8_t *mvd return amvd0 + (amvd1<<8); } -static void ALWAYS_INLINE x264_predictor_roundclip( int16_t (*dst)[2], int16_t (*mvc)[2], int i_mvc, int mv_x_min, int mv_x_max, int mv_y_min, int mv_y_max ) -{ - for( int i = 0; i < i_mvc; i++ ) - { - int mx = (mvc[i][0] + 2) >> 2; - int my = (mvc[i][1] + 2) >> 2; - dst[i][0] = x264_clip3( mx, mv_x_min, mv_x_max ); - dst[i][1] = x264_clip3( my, mv_y_min, mv_y_max ); - } -} - extern const uint8_t x264_exp2_lut[64]; extern const float x264_log2_lut[128]; extern const float x264_log2_lz_lut[32]; @@ -354,6 +406,7 @@ typedef struct } ref_pic_list_order[2][X264_REF_MAX]; /* P-frame weighting */ + int b_weighted_pred; x264_weight_t weight[X264_REF_MAX*2][3]; int i_mmco_remove_from_end; @@ -403,17 +456,51 @@ typedef struct x264_left_table_t uint8_t ref[4]; } x264_left_table_t; +/* Current frame stats */ +typedef struct +{ + /* MV bits (MV+Ref+Block Type) */ + int i_mv_bits; + /* Texture bits (DCT coefs) */ + int i_tex_bits; + /* ? */ + int i_misc_bits; + /* MB type counts */ + int i_mb_count[19]; + int i_mb_count_i; + int i_mb_count_p; + int i_mb_count_skip; + int i_mb_count_8x8dct[2]; + int i_mb_count_ref[2][X264_REF_MAX*2]; + int i_mb_partition[17]; + int i_mb_cbp[6]; + int i_mb_pred_mode[4][13]; + int i_mb_field[3]; + /* Adaptive direct mv pred */ + int i_direct_score[2]; + /* Metrics */ + int64_t i_ssd[3]; + double f_ssim; + int i_ssim_cnt; +} x264_frame_stat_t; + struct x264_t { /* encoder parameters */ x264_param_t param; x264_t *thread[X264_THREAD_MAX+1]; + x264_t *lookahead_thread[X264_LOOKAHEAD_THREAD_MAX]; int b_thread_active; int i_thread_phase; /* which thread to use for the next frame */ + int i_thread_idx; /* which thread this is */ int i_threadslice_start; /* first row in this thread slice */ int i_threadslice_end; /* row after the end of this thread slice */ + int i_threadslice_pass; /* which pass of encoding we are on */ x264_threadpool_t *threadpool; + x264_threadpool_t *lookaheadpool; + x264_pthread_mutex_t mutex; + x264_pthread_cond_t cv; /* bitstream output */ struct @@ -429,9 +516,6 @@ struct x264_t uint8_t *nal_buffer; int nal_buffer_size; - x264_sps_t *sps; - x264_pps_t *pps; - /**** thread synchronization starts here ****/ /* frame number/poc */ @@ -453,27 +537,27 @@ struct x264_t int64_t i_cpb_delay_lookahead; int64_t i_cpb_delay_pir_offset; + int64_t i_cpb_delay_pir_offset_next; int b_queued_intra_refresh; int64_t i_last_idr_pts; - /* We use only one SPS and one PPS */ - x264_sps_t sps_array[1]; - x264_pps_t pps_array[1]; int i_idr_pic_id; /* quantization matrix for decoding, [cqm][qp%6][coef] */ int (*dequant4_mf[4])[16]; /* [4][6][16] */ - int (*dequant8_mf[2])[64]; /* [2][6][64] */ + int (*dequant8_mf[4])[64]; /* [4][6][64] */ /* quantization matrix for trellis, [cqm][qp][coef] */ int (*unquant4_mf[4])[16]; /* [4][52][16] */ - int (*unquant8_mf[2])[64]; /* [2][52][64] */ + int (*unquant8_mf[4])[64]; /* [4][52][64] */ /* quantization matrix for deadzone */ udctcoef (*quant4_mf[4])[16]; /* [4][52][16] */ - udctcoef (*quant8_mf[2])[64]; /* [2][52][64] */ + udctcoef (*quant8_mf[4])[64]; /* [4][52][64] */ udctcoef (*quant4_bias[4])[16]; /* [4][52][16] */ - udctcoef (*quant8_bias[2])[64]; /* [2][52][64] */ - udctcoef (*nr_offset_emergency)[3][64]; + udctcoef (*quant8_bias[4])[64]; /* [4][52][64] */ + udctcoef (*quant4_bias0[4])[16]; /* [4][52][16] */ + udctcoef (*quant8_bias0[4])[64]; /* [4][52][64] */ + udctcoef (*nr_offset_emergency)[4][64]; /* mv/ref cost arrays. */ uint16_t *cost_mv[QP_MAX+1]; @@ -484,6 +568,10 @@ struct x264_t /* Slice header */ x264_slice_header_t sh; + /* SPS / PPS */ + x264_sps_t sps[1]; + x264_pps_t pps[1]; + /* Slice header backup, for SEI_DEC_REF_PIC_MARKING */ int b_sh_backup; x264_slice_header_t sh_backup; @@ -546,11 +634,11 @@ struct x264_t /* Current MB DCT coeffs */ struct { - ALIGNED_16( dctcoef luma16x16_dc[16] ); - ALIGNED_16( dctcoef chroma_dc[2][4] ); + ALIGNED_N( dctcoef luma16x16_dc[3][16] ); + ALIGNED_16( dctcoef chroma_dc[2][8] ); // FIXME share memory? - ALIGNED_16( dctcoef luma8x8[4][64] ); - ALIGNED_16( dctcoef luma4x4[16+8][16] ); + ALIGNED_N( dctcoef luma8x8[12][64] ); + ALIGNED_N( dctcoef luma4x4[16*3][16] ); } dct; /* MB table and cache for current frame/mb */ @@ -560,6 +648,10 @@ struct x264_t int i_mb_height; int i_mb_count; /* number of mbs in a frame */ + /* Chroma subsampling */ + int chroma_h_shift; + int chroma_v_shift; + /* Strides */ int i_mb_stride; int i_b8_stride; @@ -599,8 +691,7 @@ struct x264_t int mv_miny_spel_row[3]; int mv_maxy_spel_row[3]; /* Fullpel MV range for motion search */ - int mv_min_fpel[2]; - int mv_max_fpel[2]; + ALIGNED_8( int16_t mv_limit_fpel[2][2] ); /* min_x, min_y, max_x, max_y */ int mv_miny_fpel_row[3]; int mv_maxy_fpel_row[3]; @@ -622,7 +713,7 @@ struct x264_t int i_mb_top_y; int i_mb_topleft_y; int i_mb_topright_y; - x264_left_table_t *left_index_table; + const x264_left_table_t *left_index_table; int i_mb_top_mbpair_xy; int topleft_partition; int b_allow_skip; @@ -633,13 +724,14 @@ struct x264_t * and won't be copied from one thread to another */ /* mb table */ + uint8_t *base; /* base pointer for all malloced data in this mb */ int8_t *type; /* mb type */ uint8_t *partition; /* mb partition */ int8_t *qp; /* mb qp */ int16_t *cbp; /* mb cbp: 0x0?: luma, 0x?0: chroma, 0x100: luma dc, 0x0200 and 0x0400: chroma dc (all set for PCM)*/ int8_t (*intra4x4_pred_mode)[8]; /* intra4x4 pred mode. for non I4x4 set to I_PRED_4x4_DC(2) */ /* actually has only 7 entries; set to 8 for write-combining optimizations */ - uint8_t (*non_zero_count)[16+4+4]; /* nzc. for I_PCM set to 16 */ + uint8_t (*non_zero_count)[16*3]; /* nzc. for I_PCM set to 16 */ int8_t *chroma_pred_mode; /* chroma_pred_mode. cabac only. for non intra I_PRED_CHROMA_DC(0) */ int16_t (*mv[2])[2]; /* mb mv. set to 0 for intra mb */ 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 */ @@ -685,8 +777,8 @@ struct x264_t /* space for p_fenc and p_fdec */ #define FENC_STRIDE 16 #define FDEC_STRIDE 32 - ALIGNED_16( pixel fenc_buf[24*FENC_STRIDE] ); - ALIGNED_16( pixel fdec_buf[27*FDEC_STRIDE] ); + ALIGNED_16( pixel fenc_buf[48*FENC_STRIDE] ); + ALIGNED_N( pixel fdec_buf[52*FDEC_STRIDE] ); /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */ ALIGNED_16( pixel i4x4_fdec_buf[16*16] ); @@ -703,20 +795,21 @@ struct x264_t ALIGNED_16( dctcoef fenc_dct4[16][16] ); /* Psy RD SATD/SA8D scores cache */ - ALIGNED_16( uint64_t fenc_hadamard_cache[9] ); - ALIGNED_16( uint32_t fenc_satd_cache[32] ); + ALIGNED_N( uint64_t fenc_hadamard_cache[9] ); + ALIGNED_N( uint32_t fenc_satd_cache[32] ); /* pointer over mb of the frame to be compressed */ pixel *p_fenc[3]; /* y,u,v */ /* pointer to the actual source frame, not a block copy */ - pixel *p_fenc_plane[2]; /* y,uv */ + pixel *p_fenc_plane[3]; /* pointer over mb of the frame to be reconstructed */ pixel *p_fdec[3]; /* pointer over mb of the references */ int i_fref[2]; - pixel *p_fref[2][X264_REF_MAX*2][4+1]; /* last: yN, yH, yV, yHV, uv */ + /* [12]: yN, yH, yV, yHV, (NV12 ? uv : I444 ? (uN, uH, uV, uHV, vN, ...)) */ + pixel *p_fref[2][X264_REF_MAX*2][12]; pixel *p_fref_w[X264_REF_MAX*2]; /* weighted fullpel luma */ uint16_t *p_integral[2][X264_REF_MAX]; @@ -759,6 +852,9 @@ struct x264_t /* extra data required for mbaff in mv prediction */ int16_t topright_mv[2][3][2]; int8_t topright_ref[2][3]; + + /* current mb deblock strength */ + uint8_t (*deblock_strength)[8][4]; } cache; /* */ @@ -777,9 +873,9 @@ struct x264_t int i_chroma_lambda2_offset; /* B_direct and weighted prediction */ - int16_t dist_scale_factor_buf[2][X264_REF_MAX*2][4]; + int16_t dist_scale_factor_buf[2][2][X264_REF_MAX*2][4]; int16_t (*dist_scale_factor)[4]; - int8_t bipred_weight_buf[2][X264_REF_MAX*2][4]; + int8_t bipred_weight_buf[2][2][X264_REF_MAX*2][4]; int8_t (*bipred_weight)[4]; /* maps fref1[0]'s ref indices into the current list0 */ #define map_col_to_list0(col) h->mb.map_col_to_list0[(col)+2] @@ -796,30 +892,7 @@ struct x264_t struct { /* Current frame stats */ - struct - { - /* MV bits (MV+Ref+Block Type) */ - int i_mv_bits; - /* Texture bits (DCT coefs) */ - int i_tex_bits; - /* ? */ - int i_misc_bits; - /* MB type counts */ - int i_mb_count[19]; - int i_mb_count_i; - int i_mb_count_p; - int i_mb_count_skip; - int i_mb_count_8x8dct[2]; - int i_mb_count_ref[2][X264_REF_MAX*2]; - int i_mb_partition[17]; - int i_mb_cbp[6]; - int i_mb_pred_mode[4][13]; - /* Adaptive direct mv pred */ - int i_direct_score[2]; - /* Metrics */ - int64_t i_ssd[3]; - double f_ssim; - } frame; + x264_frame_stat_t frame; /* Cumulated stats */ @@ -843,6 +916,7 @@ struct x264_t int64_t i_mb_count_ref[2][2][X264_REF_MAX*2]; int64_t i_mb_cbp[6]; int64_t i_mb_pred_mode[4][13]; + int64_t i_mb_field[3]; /* */ int i_direct_score[2]; int i_direct_frames[2]; @@ -851,33 +925,40 @@ struct x264_t } stat; - /* 0 = luma 4x4, 1 = luma 8x8, 2 = chroma 4x4 */ + /* 0 = luma 4x4, 1 = luma 8x8, 2 = chroma 4x4, 3 = chroma 8x8 */ udctcoef (*nr_offset)[64]; uint32_t (*nr_residual_sum)[64]; uint32_t *nr_count; - ALIGNED_16( udctcoef nr_offset_denoise[3][64] ); - ALIGNED_16( uint32_t nr_residual_sum_buf[2][3][64] ); - uint32_t nr_count_buf[2][3]; + ALIGNED_N( udctcoef nr_offset_denoise[4][64] ); + ALIGNED_N( uint32_t nr_residual_sum_buf[2][4][64] ); + uint32_t nr_count_buf[2][4]; + + uint8_t luma2chroma_pixel[7]; /* Subsampled pixel size */ /* Buffers that are allocated per-thread even in sliced threads. */ void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */ - pixel *intra_border_backup[5][2]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */ + void *scratch_buffer2; /* if the first one's already in use */ + pixel *intra_border_backup[5][3]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */ /* Deblock strength values are stored for each 4x4 partition. In MBAFF * there are four extra values that need to be stored, located in [4][i]. */ uint8_t (*deblock_strength[2])[2][8][4]; /* CPU functions dependents */ x264_predict_t predict_16x16[4+3]; - x264_predict_t predict_8x8c[4+3]; x264_predict8x8_t predict_8x8[9+3]; x264_predict_t predict_4x4[9+3]; + x264_predict_t predict_chroma[4+3]; + x264_predict_t predict_8x8c[4+3]; + x264_predict_t predict_8x16c[4+3]; x264_predict_8x8_filter_t predict_8x8_filter; x264_pixel_function_t pixf; x264_mc_functions_t mc; x264_dct_function_t dctf; x264_zigzag_function_t zigzagf; + x264_zigzag_function_t zigzagf_interlaced; + x264_zigzag_function_t zigzagf_progressive; x264_quant_function_t quantf; x264_deblock_function_t loopf; x264_bitstream_function_t bsf; @@ -886,11 +967,48 @@ struct x264_t struct visualize_t *visualize; #endif x264_lookahead_t *lookahead; + +#if HAVE_OPENCL + x264_opencl_t opencl; +#endif }; // included at the end because it needs x264_t #include "macroblock.h" +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 ) +{ + int cnt = 0; + for( int i = 0; i < i_mvc; i++ ) + { + int mx = (mvc[i][0] + 2) >> 2; + int my = (mvc[i][1] + 2) >> 2; + uint32_t mv = pack16to32_mask(mx, my); + if( !mv || mv == pmv ) continue; + dst[cnt][0] = x264_clip3( mx, mv_limit[0][0], mv_limit[1][0] ); + dst[cnt][1] = x264_clip3( my, mv_limit[0][1], mv_limit[1][1] ); + cnt++; + } + return cnt; +} + +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 ) +{ + int cnt = 0; + int qpel_limit[4] = {mv_limit[0][0] << 2, mv_limit[0][1] << 2, mv_limit[1][0] << 2, mv_limit[1][1] << 2}; + for( int i = 0; i < i_mvc; i++ ) + { + uint32_t mv = M32( mvc[i] ); + int mx = mvc[i][0]; + int my = mvc[i][1]; + if( !mv || mv == pmv ) continue; + dst[cnt][0] = x264_clip3( mx, qpel_limit[0], qpel_limit[2] ); + dst[cnt][1] = x264_clip3( my, qpel_limit[1], qpel_limit[3] ); + cnt++; + } + return cnt; +} + #if ARCH_X86 || ARCH_X86_64 #include "x86/util.h" #endif