/*****************************************************************************
- * common.h: h264 encoder
+ * common.h: misc common functions
*****************************************************************************
- * Copyright (C) 2003-2008 x264 project
+ * Copyright (C) 2003-2011 x264 project
*
* Authors: Laurent Aimar <fenrir@via.ecp.fr>
* Loren Merritt <lorenm@u.washington.edu>
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
+ *
+ * This program is also available under a commercial proprietary license.
+ * For more information, contact us at licensing@x264.com.
*****************************************************************************/
#ifndef X264_COMMON_H
} while( 0 )
#define X264_BFRAME_MAX 16
+#define X264_REF_MAX 16
#define X264_THREAD_MAX 128
-#define X264_PCM_COST (386*8)
+#define X264_PCM_COST (384*BIT_DEPTH+16)
#define X264_LOOKAHEAD_MAX 250
+#define QP_BD_OFFSET (6*(BIT_DEPTH-8))
+#define QP_MAX_SPEC (51+QP_BD_OFFSET)
+#define QP_MAX (QP_MAX_SPEC+18)
+#define QP_MAX_MAX (51+2*6+18)
+#define PIXEL_MAX ((1 << BIT_DEPTH)-1)
// arbitrary, but low because SATD scores are 1/4 normal
-#define X264_LOOKAHEAD_QP 12
+#define X264_LOOKAHEAD_QP (12+QP_BD_OFFSET)
+#define SPEC_QP(x) X264_MIN((x), QP_MAX_SPEC)
// number of pixels (per thread) in progress at any given time.
// 16 for the macroblock in progress + 3 for deblocking + 3 for motion compensation filter + 2 for extra safety
#define CP64(dst,src) M64(dst) = M64(src)
#define CP128(dst,src) M128(dst) = M128(src)
-typedef uint8_t pixel;
-typedef uint32_t pixel4;
-typedef int16_t dctcoef;
+#if HIGH_BIT_DEPTH
+ typedef uint16_t pixel;
+ typedef uint64_t pixel4;
+ typedef int32_t dctcoef;
+ typedef uint32_t udctcoef;
+
+# define PIXEL_SPLAT_X4(x) ((x)*0x0001000100010001ULL)
+# define MPIXEL_X4(src) M64(src)
+#else
+ typedef uint8_t pixel;
+ typedef uint32_t pixel4;
+ typedef int16_t dctcoef;
+ typedef uint16_t udctcoef;
+
+# define PIXEL_SPLAT_X4(x) ((x)*0x01010101U)
+# define MPIXEL_X4(src) M32(src)
+#endif
+
+#define BIT_DEPTH X264_BIT_DEPTH
-#define PIXEL_SPLAT_X4(x) ((x)*0x01010101U)
-#define MPIXEL_X4(src) M32(src)
-#define CPPIXEL_X4(dst,src) CP32(dst,src)
-#define CPPIXEL_X8(dst,src) CP64(dst,src)
-#define MDCT_X2(dct) M32(dct)
-#define CPDCT_X2(dst,src) CP32(dst,src)
-#define CPDCT_X4(dst,src) CP64(dst,src)
+#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_0 (4+1*8)
-static const int x264_scan8[16+2*4+3] =
+static const unsigned x264_scan8[16+2*4+3] =
{
/* Luma */
4+1*8, 5+1*8, 4+2*8, 5+2*8,
#include "cabac.h"
#include "quant.h"
#include "cpu.h"
+#include "threadpool.h"
/****************************************************************************
* General functions
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_init_vlc_tables();
+void x264_reduce_fraction64( uint64_t *n, uint64_t *d );
+void x264_cavlc_init( void );
+void x264_cabac_init( void );
static ALWAYS_INLINE pixel x264_clip_pixel( int x )
{
- return x&(~255) ? (-x)>>31 : x;
+ return ( (x & ~PIXEL_MAX) ? (-x)>>31 & PIXEL_MAX : x );
}
static ALWAYS_INLINE int x264_clip3( int v, int i_min, int i_max )
SLICE_TYPE_P = 0,
SLICE_TYPE_B = 1,
SLICE_TYPE_I = 2,
- SLICE_TYPE_SP = 3,
- SLICE_TYPE_SI = 4
};
-static const char slice_type_to_char[] = { 'P', 'B', 'I', 'S', 'S' };
+static const char slice_type_to_char[] = { 'P', 'B', 'I' };
enum sei_payload_type_e
{
SEI_USER_DATA_REGISTERED = 4,
SEI_USER_DATA_UNREGISTERED = 5,
SEI_RECOVERY_POINT = 6,
+ SEI_DEC_REF_PIC_MARKING = 7,
+ SEI_FRAME_PACKING = 45,
};
typedef struct
int i_num_ref_idx_l0_active;
int i_num_ref_idx_l1_active;
- int b_ref_pic_list_reordering_l0;
- int b_ref_pic_list_reordering_l1;
+ int b_ref_pic_list_reordering[2];
struct
{
int idc;
int arg;
- } ref_pic_list_order[2][16];
+ } ref_pic_list_order[2][X264_REF_MAX];
/* P-frame weighting */
- x264_weight_t weight[32][3];
+ x264_weight_t weight[X264_REF_MAX*2][3];
int i_mmco_remove_from_end;
int i_mmco_command_count;
{
int i_difference_of_pic_nums;
int i_poc;
- } mmco[16];
+ } mmco[X264_REF_MAX];
int i_cabac_init_idc;
int i_last_keyframe;
int i_slicetype_length;
x264_frame_t *last_nonb;
- x264_synch_frame_list_t ifbuf;
- x264_synch_frame_list_t next;
- x264_synch_frame_list_t ofbuf;
+ x264_pthread_t thread_handle;
+ x264_sync_frame_list_t ifbuf;
+ x264_sync_frame_list_t next;
+ x264_sync_frame_list_t ofbuf;
} x264_lookahead_t;
typedef struct x264_ratecontrol_t x264_ratecontrol_t;
+typedef struct x264_left_table_t
+{
+ uint8_t intra[4];
+ uint8_t nnz[4];
+ uint8_t nnz_chroma[4];
+ uint8_t mv[4];
+ uint8_t ref[4];
+} x264_left_table_t;
+
struct x264_t
{
/* encoder parameters */
x264_param_t param;
x264_t *thread[X264_THREAD_MAX+1];
- x264_pthread_t thread_handle;
int b_thread_active;
int i_thread_phase; /* which thread to use for the next frame */
int i_threadslice_start; /* first row in this thread slice */
int i_threadslice_end; /* row after the end of this thread slice */
+ x264_threadpool_t *threadpool;
/* bitstream output */
struct
uint8_t *nal_buffer;
int nal_buffer_size;
+ x264_sps_t *sps;
+ x264_pps_t *pps;
+
/**** thread synchronization starts here ****/
/* frame number/poc */
int i_nal_type;
int i_nal_ref_idc;
- int i_disp_fields; /* Number of displayed fields (both coded and implied via pic_struct) */
+ int64_t i_disp_fields; /* Number of displayed fields (both coded and implied via pic_struct) */
int i_disp_fields_last_frame;
- int i_prev_duration; /* Duration of previous frame */
- int i_coded_fields; /* Number of coded fields (both coded and implied via pic_struct) */
- int i_cpb_delay; /* Equal to number of fields preceding this field
+ int64_t i_prev_duration; /* Duration of previous frame */
+ int64_t i_coded_fields; /* Number of coded fields (both coded and implied via pic_struct) */
+ int64_t i_cpb_delay; /* Equal to number of fields preceding this field
* since last buffering_period SEI */
- int i_coded_fields_lookahead; /* Use separate counters for lookahead */
- int i_cpb_delay_lookahead;
+ int64_t i_coded_fields_lookahead; /* Use separate counters for lookahead */
+ int64_t i_cpb_delay_lookahead;
+
+ int64_t i_cpb_delay_pir_offset;
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_sps_t *sps;
x264_pps_t pps_array[1];
- x264_pps_t *pps;
int i_idr_pic_id;
- /* Timebase multiplier for DTS compression */
- int i_dts_compress_multiplier;
-
/* 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 (*unquant4_mf[4])[16]; /* [4][52][16] */
int (*unquant8_mf[2])[64]; /* [2][52][64] */
/* quantization matrix for deadzone */
- uint16_t (*quant4_mf[4])[16]; /* [4][52][16] */
- uint16_t (*quant8_mf[2])[64]; /* [2][52][64] */
- uint16_t (*quant4_bias[4])[16]; /* [4][52][16] */
- uint16_t (*quant8_bias[2])[64]; /* [2][52][64] */
+ udctcoef (*quant4_mf[4])[16]; /* [4][52][16] */
+ udctcoef (*quant8_mf[2])[64]; /* [2][52][64] */
+ udctcoef (*quant4_bias[4])[16]; /* [4][52][16] */
+ udctcoef (*quant8_bias[2])[64]; /* [2][52][64] */
+ udctcoef (*nr_offset_emergency)[3][64];
- /* mv/ref cost arrays. Indexed by lambda instead of
- * qp because, due to rounding, some quantizers share
- * lambdas. This saves memory. */
- uint16_t *cost_mv[92];
- uint16_t *cost_mv_fpel[92][4];
+ /* mv/ref cost arrays. */
+ uint16_t *cost_mv[QP_MAX+1];
+ uint16_t *cost_mv_fpel[QP_MAX+1][4];
const uint8_t *chroma_qp_table; /* includes both the nonlinear luma->chroma mapping and chroma_qp_offset */
/* Slice header */
x264_slice_header_t sh;
+ /* Slice header backup, for SEI_DEC_REF_PIC_MARKING */
+ int b_sh_backup;
+ x264_slice_header_t sh_backup;
+
/* cabac context */
x264_cabac_t cabac;
x264_frame_t **blank_unused;
/* frames used for reference + sentinels */
- x264_frame_t *reference[16+2];
+ x264_frame_t *reference[X264_REF_MAX+2];
- int i_last_keyframe; /* Frame number of the last keyframe */
+ int i_last_keyframe; /* Frame number of the last keyframe */
+ int i_last_idr; /* Frame number of the last IDR (not RP)*/
+ int i_poc_last_open_gop; /* Poc of the I frame of the last open-gop. The value
+ * is only assigned during the period between that
+ * I frame and the next P or I frame, else -1 */
int i_input; /* Number of input frames already accepted */
int i_delay; /* Number of frames buffered for B reordering */
int i_bframe_delay;
int64_t i_bframe_delay_time;
- int64_t i_init_delta;
+ int64_t i_first_pts;
int64_t i_prev_reordered_pts[2];
int64_t i_largest_pts;
int64_t i_second_largest_pts;
x264_frame_t *fdec;
/* references lists */
- int i_ref0;
- x264_frame_t *fref0[16+3]; /* ref list 0 */
- int i_ref1;
- x264_frame_t *fref1[16+3]; /* ref list 1 */
+ int i_ref[2];
+ x264_frame_t *fref[2][X264_REF_MAX+3];
+ x264_frame_t *fref_nearest[2];
int b_ref_reorder[2];
/* hrd */
/* MB table and cache for current frame/mb */
struct
{
+ int i_mb_width;
+ int i_mb_height;
int i_mb_count; /* number of mbs in a frame */
/* Strides */
int i_mb_stride;
int i_b8_stride;
int i_b4_stride;
+ int left_b8[2];
+ int left_b4[2];
/* Current index */
int i_mb_x;
int i_psy_trellis; /* Psy trellis strength--fixed point value*/
int b_interlaced;
+ int b_adaptive_mbaff; /* MBAFF+subme 0 requires non-adaptive MBAFF i.e. all field mbs */
/* Allowed qpel MV range to stay within the picture + emulated edge pixels */
int mv_min[2];
int mv_max[2];
+ int mv_miny_row[3]; /* 0 == top progressive, 1 == bot progressive, 2 == interlaced */
+ int mv_maxy_row[3];
/* Subpel MV range for motion search.
* same mv_min/max but includes levels' i_mv_range. */
int mv_min_spel[2];
int mv_max_spel[2];
+ 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];
+ int mv_miny_fpel_row[3];
+ int mv_maxy_fpel_row[3];
/* neighboring MBs */
unsigned int i_neighbour;
unsigned int i_neighbour_intra; /* for constrained intra pred */
unsigned int i_neighbour_frame; /* ignoring slice boundaries */
int i_mb_type_top;
- int i_mb_type_left;
+ int i_mb_type_left[2];
int i_mb_type_topleft;
int i_mb_type_topright;
int i_mb_prev_xy;
- int i_mb_left_xy;
+ int i_mb_left_xy[2];
int i_mb_top_xy;
int i_mb_topleft_xy;
int i_mb_topright_xy;
+ int i_mb_top_y;
+ int i_mb_topleft_y;
+ int i_mb_topright_y;
+ x264_left_table_t *left_index_table;
+ int i_mb_top_mbpair_xy;
+ int topleft_partition;
+ int b_allow_skip;
+ int field_decoding_flag;
/**** thread synchronization ends here ****/
/* subsequent variables are either thread-local or constant,
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 */
int8_t *ref[2]; /* mb ref. set to -1 if non used (intra or Lx only) */
- int16_t (*mvr[2][32])[2]; /* 16x16 mv for each possible ref */
+ int16_t (*mvr[2][X264_REF_MAX*2])[2];/* 16x16 mv for each possible ref */
int8_t *skipbp; /* block pattern for SKIP or DIRECT (sub)mbs. B-frames + cabac only */
int8_t *mb_transform_size; /* transform_size_8x8_flag of each mb */
uint16_t *slice_table; /* sh->first_mb of the slice that the indexed mb is part of
* NOTE: this will fail on resolutions above 2^16 MBs... */
+ uint8_t *field;
/* buffer for weighted versions of the reference frames */
- pixel *p_weight_buf[16];
+ pixel *p_weight_buf[X264_REF_MAX];
/* current value */
int i_type;
/* set to true if we are re-encoding a macroblock. */
int b_reencode_mb;
int ip_offset; /* Used by PIR to offset the quantizer of intra-refresh blocks. */
+ int b_deblock_rdo;
+ int b_overflow; /* If CAVLC had a level code overflow during bitstream writing. */
struct
{
ALIGNED_16( uint32_t fenc_satd_cache[32] );
/* pointer over mb of the frame to be compressed */
- pixel *p_fenc[3];
+ pixel *p_fenc[3]; /* y,u,v */
/* pointer to the actual source frame, not a block copy */
- pixel *p_fenc_plane[3];
+ pixel *p_fenc_plane[2]; /* y,uv */
/* 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][32][4+2]; /* last: lN, lH, lV, lHV, cU, cV */
- pixel *p_fref_w[32]; /* weighted fullpel luma */
- uint16_t *p_integral[2][16];
+ pixel *p_fref[2][X264_REF_MAX*2][4+1]; /* last: yN, yH, yV, yHV, uv */
+ pixel *p_fref_w[X264_REF_MAX*2]; /* weighted fullpel luma */
+ uint16_t *p_integral[2][X264_REF_MAX];
/* fref stride */
int i_stride[3];
/* number of neighbors (top and left) that used 8x8 dct */
int i_neighbour_transform_size;
- int i_neighbour_interlaced;
+ int i_neighbour_skip;
/* neighbor CBPs */
int i_cbp_top;
int i_cbp_left;
+
+ /* extra data required for mbaff in mv prediction */
+ int16_t topright_mv[2][3][2];
+ int8_t topright_ref[2][3];
} cache;
/* */
int i_chroma_lambda2_offset;
/* B_direct and weighted prediction */
- int16_t dist_scale_factor_buf[2][32][4];
+ int16_t dist_scale_factor_buf[2][X264_REF_MAX*2][4];
int16_t (*dist_scale_factor)[4];
- int8_t bipred_weight_buf[2][32][4];
+ int8_t bipred_weight_buf[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]
- int8_t map_col_to_list0[18];
+ int8_t map_col_to_list0[X264_REF_MAX+2];
int ref_blind_dupe; /* The index of the blind reference frame duplicate. */
- int8_t deblock_ref_table[32+2];
+ int8_t deblock_ref_table[X264_REF_MAX*2+2];
#define deblock_ref_table(x) h->mb.deblock_ref_table[(x)+2]
} mb;
int i_mb_count_p;
int i_mb_count_skip;
int i_mb_count_8x8dct[2];
- int i_mb_count_ref[2][32];
+ 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];
/* Cumulated stats */
/* per slice info */
- int i_frame_count[5];
- int64_t i_frame_size[5];
- double f_frame_qp[5];
+ int i_frame_count[3];
+ int64_t i_frame_size[3];
+ double f_frame_qp[3];
int i_consecutive_bframes[X264_BFRAME_MAX+1];
/* */
- int64_t i_ssd_global[5];
- double f_psnr_average[5];
- double f_psnr_mean_y[5];
- double f_psnr_mean_u[5];
- double f_psnr_mean_v[5];
- double f_ssim_mean_y[5];
+ double f_ssd_global[3];
+ double f_psnr_average[3];
+ double f_psnr_mean_y[3];
+ double f_psnr_mean_u[3];
+ double f_psnr_mean_v[3];
+ double f_ssim_mean_y[3];
+ double f_frame_duration[3];
/* */
- int64_t i_mb_count[5][19];
+ int64_t i_mb_count[3][19];
int64_t i_mb_partition[2][17];
int64_t i_mb_count_8x8dct[2];
- int64_t i_mb_count_ref[2][2][32];
+ 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];
/* */
int i_direct_score[2];
int i_direct_frames[2];
/* num p-frames weighted */
- int i_wpred[3];
+ int i_wpred[2];
} stat;
- ALIGNED_16( uint32_t nr_residual_sum[2][64] );
- ALIGNED_16( uint16_t nr_offset[2][64] );
- uint32_t nr_count[2];
+ /* 0 = luma 4x4, 1 = luma 8x8, 2 = chroma 4x4 */
+ 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];
/* 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[2][3]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
- uint8_t (*deblock_strength[2])[2][4][4];
+ pixel *intra_border_backup[5][2]; /* 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];
// included at the end because it needs x264_t
#include "macroblock.h"
-#include "rectangle.h"
-#if HAVE_MMX
+#if ARCH_X86 || ARCH_X86_64
#include "x86/util.h"
#endif
+#include "rectangle.h"
+
#endif