FFmpeg: libavcodec/h264.h Source File

00001 /*
00002  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
00003  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
00004  *
00005  * This file is part of FFmpeg.
00006  *
00007  * FFmpeg is free software; you can redistribute it and/or
00008  * modify it under the terms of the GNU Lesser General Public
00009  * License as published by the Free Software Foundation; either
00010  * version 2.1 of the License, or (at your option) any later version.
00011  *
00012  * FFmpeg is distributed in the hope that it will be useful,
00013  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00014  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00015  * Lesser General Public License for more details.
00016  *
00017  * You should have received a copy of the GNU Lesser General Public
00018  * License along with FFmpeg; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00028 #ifndef AVCODEC_H264_H
00029 #define AVCODEC_H264_H
00030 
00031 #include "libavutil/intreadwrite.h"
00032 #include "dsputil.h"
00033 #include "cabac.h"
00034 #include "mpegvideo.h"
00035 #include "h264dsp.h"
00036 #include "h264pred.h"
00037 #include "rectangle.h"
00038 
00039 #define interlaced_dct interlaced_dct_is_a_bad_name
00040 #define mb_intra       mb_intra_is_not_initialized_see_mb_type
00041 
00042 #define MAX_SPS_COUNT          32
00043 #define MAX_PPS_COUNT         256
00044 
00045 #define MAX_MMCO_COUNT         66
00046 
00047 #define MAX_DELAYED_PIC_COUNT  16
00048 
00049 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
00050 
00051 /* Compiling in interlaced support reduces the speed
00052  * of progressive decoding by about 2%. */
00053 #define ALLOW_INTERLACE
00054 
00055 #define FMO 0
00056 
00061 #define MAX_SLICES 16
00062 
00063 #ifdef ALLOW_INTERLACE
00064 #define MB_MBAFF    h->mb_mbaff
00065 #define MB_FIELD    h->mb_field_decoding_flag
00066 #define FRAME_MBAFF h->mb_aff_frame
00067 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
00068 #define LEFT_MBS 2
00069 #define LTOP     0
00070 #define LBOT     1
00071 #define LEFT(i)  (i)
00072 #else
00073 #define MB_MBAFF      0
00074 #define MB_FIELD      0
00075 #define FRAME_MBAFF   0
00076 #define FIELD_PICTURE 0
00077 #undef  IS_INTERLACED
00078 #define IS_INTERLACED(mb_type) 0
00079 #define LEFT_MBS 1
00080 #define LTOP     0
00081 #define LBOT     0
00082 #define LEFT(i)  0
00083 #endif
00084 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
00085 
00086 #ifndef CABAC
00087 #define CABAC h->pps.cabac
00088 #endif
00089 
00090 #define CHROMA    (h->sps.chroma_format_idc)
00091 #define CHROMA422 (h->sps.chroma_format_idc == 2)
00092 #define CHROMA444 (h->sps.chroma_format_idc == 3)
00093 
00094 #define EXTENDED_SAR       255
00095 
00096 #define MB_TYPE_REF0       MB_TYPE_ACPRED // dirty but it fits in 16 bit
00097 #define MB_TYPE_8x8DCT     0x01000000
00098 #define IS_REF0(a)         ((a) & MB_TYPE_REF0)
00099 #define IS_8x8DCT(a)       ((a) & MB_TYPE_8x8DCT)
00100 
00105 #define DELAYED_PIC_REF 4
00106 
00107 #define QP_MAX_NUM (51 + 6*6)           // The maximum supported qp
00108 
00109 /* NAL unit types */
00110 enum {
00111     NAL_SLICE = 1,
00112     NAL_DPA,
00113     NAL_DPB,
00114     NAL_DPC,
00115     NAL_IDR_SLICE,
00116     NAL_SEI,
00117     NAL_SPS,
00118     NAL_PPS,
00119     NAL_AUD,
00120     NAL_END_SEQUENCE,
00121     NAL_END_STREAM,
00122     NAL_FILLER_DATA,
00123     NAL_SPS_EXT,
00124     NAL_AUXILIARY_SLICE = 19
00125 };
00126 
00130 typedef enum {
00131     SEI_BUFFERING_PERIOD            = 0,   
00132     SEI_TYPE_PIC_TIMING             = 1,   
00133     SEI_TYPE_USER_DATA_UNREGISTERED = 5,   
00134     SEI_TYPE_RECOVERY_POINT         = 6    
00135 } SEI_Type;
00136 
00140 typedef enum {
00141     SEI_PIC_STRUCT_FRAME             = 0, 
00142     SEI_PIC_STRUCT_TOP_FIELD         = 1, 
00143     SEI_PIC_STRUCT_BOTTOM_FIELD      = 2, 
00144     SEI_PIC_STRUCT_TOP_BOTTOM        = 3, 
00145     SEI_PIC_STRUCT_BOTTOM_TOP        = 4, 
00146     SEI_PIC_STRUCT_TOP_BOTTOM_TOP    = 5, 
00147     SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, 
00148     SEI_PIC_STRUCT_FRAME_DOUBLING    = 7, 
00149     SEI_PIC_STRUCT_FRAME_TRIPLING    = 8  
00150 } SEI_PicStructType;
00151 
00155 typedef struct SPS {
00156     int profile_idc;
00157     int level_idc;
00158     int chroma_format_idc;
00159     int transform_bypass;              
00160     int log2_max_frame_num;            
00161     int poc_type;                      
00162     int log2_max_poc_lsb;              
00163     int delta_pic_order_always_zero_flag;
00164     int offset_for_non_ref_pic;
00165     int offset_for_top_to_bottom_field;
00166     int poc_cycle_length;              
00167     int ref_frame_count;               
00168     int gaps_in_frame_num_allowed_flag;
00169     int mb_width;                      
00170     int mb_height;                     
00171     int frame_mbs_only_flag;
00172     int mb_aff;                        
00173     int direct_8x8_inference_flag;
00174     int crop;                          
00175     unsigned int crop_left;            
00176     unsigned int crop_right;           
00177     unsigned int crop_top;             
00178     unsigned int crop_bottom;          
00179     int vui_parameters_present_flag;
00180     AVRational sar;
00181     int video_signal_type_present_flag;
00182     int full_range;
00183     int colour_description_present_flag;
00184     enum AVColorPrimaries color_primaries;
00185     enum AVColorTransferCharacteristic color_trc;
00186     enum AVColorSpace colorspace;
00187     int timing_info_present_flag;
00188     uint32_t num_units_in_tick;
00189     uint32_t time_scale;
00190     int fixed_frame_rate_flag;
00191     short offset_for_ref_frame[256]; // FIXME dyn aloc?
00192     int bitstream_restriction_flag;
00193     int num_reorder_frames;
00194     int scaling_matrix_present;
00195     uint8_t scaling_matrix4[6][16];
00196     uint8_t scaling_matrix8[6][64];
00197     int nal_hrd_parameters_present_flag;
00198     int vcl_hrd_parameters_present_flag;
00199     int pic_struct_present_flag;
00200     int time_offset_length;
00201     int cpb_cnt;                          
00202     int initial_cpb_removal_delay_length; 
00203     int cpb_removal_delay_length;         
00204     int dpb_output_delay_length;          
00205     int bit_depth_luma;                   
00206     int bit_depth_chroma;                 
00207     int residual_color_transform_flag;    
00208     int constraint_set_flags;             
00209 } SPS;
00210 
00214 typedef struct PPS {
00215     unsigned int sps_id;
00216     int cabac;                  
00217     int pic_order_present;      
00218     int slice_group_count;      
00219     int mb_slice_group_map_type;
00220     unsigned int ref_count[2];  
00221     int weighted_pred;          
00222     int weighted_bipred_idc;
00223     int init_qp;                
00224     int init_qs;                
00225     int chroma_qp_index_offset[2];
00226     int deblocking_filter_parameters_present; 
00227     int constrained_intra_pred;     
00228     int redundant_pic_cnt_present;  
00229     int transform_8x8_mode;         
00230     uint8_t scaling_matrix4[6][16];
00231     uint8_t scaling_matrix8[6][64];
00232     uint8_t chroma_qp_table[2][QP_MAX_NUM+1];  
00233     int chroma_qp_diff;
00234 } PPS;
00235 
00239 typedef enum MMCOOpcode {
00240     MMCO_END = 0,
00241     MMCO_SHORT2UNUSED,
00242     MMCO_LONG2UNUSED,
00243     MMCO_SHORT2LONG,
00244     MMCO_SET_MAX_LONG,
00245     MMCO_RESET,
00246     MMCO_LONG,
00247 } MMCOOpcode;
00248 
00252 typedef struct MMCO {
00253     MMCOOpcode opcode;
00254     int short_pic_num;  
00255     int long_arg;       
00256 } MMCO;
00257 
00261 typedef struct H264Context {
00262     MpegEncContext s;
00263     H264DSPContext h264dsp;
00264     int pixel_shift;    
00265     int chroma_qp[2];   // QPc
00266 
00267     int qp_thresh;      
00268 
00269     int prev_mb_skipped;
00270     int next_mb_skipped;
00271 
00272     // prediction stuff
00273     int chroma_pred_mode;
00274     int intra16x16_pred_mode;
00275 
00276     int topleft_mb_xy;
00277     int top_mb_xy;
00278     int topright_mb_xy;
00279     int left_mb_xy[LEFT_MBS];
00280 
00281     int topleft_type;
00282     int top_type;
00283     int topright_type;
00284     int left_type[LEFT_MBS];
00285 
00286     const uint8_t *left_block;
00287     int topleft_partition;
00288 
00289     int8_t intra4x4_pred_mode_cache[5 * 8];
00290     int8_t(*intra4x4_pred_mode);
00291     H264PredContext hpc;
00292     unsigned int topleft_samples_available;
00293     unsigned int top_samples_available;
00294     unsigned int topright_samples_available;
00295     unsigned int left_samples_available;
00296     uint8_t (*top_borders[2])[(16 * 3) * 2];
00297 
00302     DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
00303 
00304     uint8_t (*non_zero_count)[48];
00305 
00309     DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
00310     DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
00311 #define LIST_NOT_USED -1 // FIXME rename?
00312 #define PART_NOT_AVAILABLE -2
00313 
00317     int neighbor_transform_size;
00318 
00323     int block_offset[2 * (16 * 3)];
00324 
00325     uint32_t *mb2b_xy;  // FIXME are these 4 a good idea?
00326     uint32_t *mb2br_xy;
00327     int b_stride;       // FIXME use s->b4_stride
00328 
00329     int mb_linesize;    
00330     int mb_uvlinesize;
00331 
00332     int emu_edge_width;
00333     int emu_edge_height;
00334 
00335     SPS sps; 
00336 
00340     PPS pps; // FIXME move to Picture perhaps? (->no) do we need that?
00341 
00342     uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
00343     uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
00344     uint32_t(*dequant4_coeff[6])[16];
00345     uint32_t(*dequant8_coeff[6])[64];
00346 
00347     int slice_num;
00348     uint16_t *slice_table;      
00349     int slice_type;
00350     int slice_type_nos;         
00351     int slice_type_fixed;
00352 
00353     // interlacing specific flags
00354     int mb_aff_frame;
00355     int mb_field_decoding_flag;
00356     int mb_mbaff;               
00357 
00358     DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
00359 
00360     // Weighted pred stuff
00361     int use_weight;
00362     int use_weight_chroma;
00363     int luma_log2_weight_denom;
00364     int chroma_log2_weight_denom;
00365     // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
00366     int luma_weight[48][2][2];
00367     int chroma_weight[48][2][2][2];
00368     int implicit_weight[48][48][2];
00369 
00370     int direct_spatial_mv_pred;
00371     int col_parity;
00372     int col_fieldoff;
00373     int dist_scale_factor[16];
00374     int dist_scale_factor_field[2][32];
00375     int map_col_to_list0[2][16 + 32];
00376     int map_col_to_list0_field[2][2][16 + 32];
00377 
00381     unsigned int ref_count[2];          
00382     unsigned int list_count;
00383     uint8_t *list_counts;               
00384     Picture ref_list[2][48];            
00387     int ref2frm[MAX_SLICES][2][64];     
00388 
00389     // data partitioning
00390     GetBitContext intra_gb;
00391     GetBitContext inter_gb;
00392     GetBitContext *intra_gb_ptr;
00393     GetBitContext *inter_gb_ptr;
00394 
00395     DECLARE_ALIGNED(16, DCTELEM, mb)[16 * 48 * 2]; 
00396     DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[3][16 * 2];
00397     DCTELEM mb_padding[256 * 2];        
00398 
00402     CABACContext cabac;
00403     uint8_t cabac_state[1024];
00404 
00405     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
00406     uint16_t *cbp_table;
00407     int cbp;
00408     int top_cbp;
00409     int left_cbp;
00410     /* chroma_pred_mode for i4x4 or i16x16, else 0 */
00411     uint8_t *chroma_pred_mode_table;
00412     int last_qscale_diff;
00413     uint8_t (*mvd_table[2])[2];
00414     DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
00415     uint8_t *direct_table;
00416     uint8_t direct_cache[5 * 8];
00417 
00418     uint8_t zigzag_scan[16];
00419     uint8_t zigzag_scan8x8[64];
00420     uint8_t zigzag_scan8x8_cavlc[64];
00421     uint8_t field_scan[16];
00422     uint8_t field_scan8x8[64];
00423     uint8_t field_scan8x8_cavlc[64];
00424     uint8_t zigzag_scan_q0[16];
00425     uint8_t zigzag_scan8x8_q0[64];
00426     uint8_t zigzag_scan8x8_cavlc_q0[64];
00427     uint8_t field_scan_q0[16];
00428     uint8_t field_scan8x8_q0[64];
00429     uint8_t field_scan8x8_cavlc_q0[64];
00430 
00431     int x264_build;
00432 
00433     int mb_xy;
00434 
00435     int is_complex;
00436 
00437     // deblock
00438     int deblocking_filter;          
00439     int slice_alpha_c0_offset;
00440     int slice_beta_offset;
00441 
00442     // =============================================================
00443     // Things below are not used in the MB or more inner code
00444 
00445     int nal_ref_idc;
00446     int nal_unit_type;
00447     uint8_t *rbsp_buffer[2];
00448     unsigned int rbsp_buffer_size[2];
00449 
00453     int is_avc;           
00454     int nal_length_size;  
00455     int got_first;        
00456 
00457     SPS *sps_buffers[MAX_SPS_COUNT];
00458     PPS *pps_buffers[MAX_PPS_COUNT];
00459 
00460     int dequant_coeff_pps;      
00461 
00462     uint16_t *slice_table_base;
00463 
00464     // POC stuff
00465     int poc_lsb;
00466     int poc_msb;
00467     int delta_poc_bottom;
00468     int delta_poc[2];
00469     int frame_num;
00470     int prev_poc_msb;           
00471     int prev_poc_lsb;           
00472     int frame_num_offset;       
00473     int prev_frame_num_offset;  
00474     int prev_frame_num;         
00475 
00479     int curr_pic_num;
00480 
00484     int max_pic_num;
00485 
00486     int redundant_pic_count;
00487 
00488     Picture *short_ref[32];
00489     Picture *long_ref[32];
00490     Picture default_ref_list[2][32]; 
00491     Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
00492     int last_pocs[MAX_DELAYED_PIC_COUNT];
00493     Picture *next_output_pic;
00494     int outputed_poc;
00495     int next_outputed_poc;
00496 
00500     MMCO mmco[MAX_MMCO_COUNT];
00501     int mmco_index;
00502     int mmco_reset;
00503 
00504     int long_ref_count;     
00505     int short_ref_count;    
00506 
00507     int cabac_init_idc;
00508 
00513     struct H264Context *thread_context[MAX_THREADS];
00514 
00518     int current_slice;
00519 
00526     int max_contexts;
00527 
00532     int single_decode_warning;
00533 
00534     int last_slice_type;
00540     SEI_PicStructType sei_pic_struct;
00541 
00548     int prev_interlaced_frame;
00549 
00555     int sei_ct_type;
00556 
00560     int sei_dpb_output_delay;
00561 
00565     int sei_cpb_removal_delay;
00566 
00574     int sei_recovery_frame_cnt;
00581     int recovery_frame;
00582 
00586     int valid_recovery_point;
00587 
00588     int luma_weight_flag[2];    
00589     int chroma_weight_flag[2];  
00590 
00591     // Timestamp stuff
00592     int sei_buffering_period_present;   
00593     int initial_cpb_removal_delay[32];  
00594 
00595     int cur_chroma_format_idc;
00596 
00597     int16_t slice_row[MAX_SLICES]; 
00598 
00599     int sync;                      
00600 
00601     uint8_t parse_history[4];
00602     int parse_history_count;
00603     int parse_last_mb;
00604 } H264Context;
00605 
00606 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; 
00607 extern const uint16_t ff_h264_mb_sizes[4];
00608 
00612 int ff_h264_decode_sei(H264Context *h);
00613 
00617 int ff_h264_decode_seq_parameter_set(H264Context *h);
00618 
00622 int ff_h264_get_profile(SPS *sps);
00623 
00627 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
00628 
00637 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
00638                                   int *dst_length, int *consumed, int length);
00639 
00644 av_cold void ff_h264_free_context(H264Context *h);
00645 
00649 int ff_h264_get_slice_type(const H264Context *h);
00650 
00655 int ff_h264_alloc_tables(H264Context *h);
00656 
00660 int ff_h264_fill_default_ref_list(H264Context *h);
00661 
00662 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
00663 void ff_h264_fill_mbaff_ref_list(H264Context *h);
00664 void ff_h264_remove_all_refs(H264Context *h);
00665 
00669 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
00670 
00671 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
00672 
00673 void ff_generate_sliding_window_mmcos(H264Context *h);
00674 
00679 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
00680 
00685 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
00686 
00687 void ff_h264_hl_decode_mb(H264Context *h);
00688 int ff_h264_frame_start(H264Context *h);
00689 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size);
00690 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
00691 av_cold void ff_h264_decode_init_vlc(void);
00692 
00697 int ff_h264_decode_mb_cavlc(H264Context *h);
00698 
00703 int ff_h264_decode_mb_cabac(H264Context *h);
00704 
00705 void ff_h264_init_cabac_states(H264Context *h);
00706 
00707 void ff_h264_direct_dist_scale_factor(H264Context *const h);
00708 void ff_h264_direct_ref_list_init(H264Context *const h);
00709 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
00710 
00711 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
00712                             uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
00713                             unsigned int linesize, unsigned int uvlinesize);
00714 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
00715                        uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
00716                        unsigned int linesize, unsigned int uvlinesize);
00717 
00723 void ff_h264_reset_sei(H264Context *h);
00724 
00725 /*
00726  * o-o o-o
00727  *  / / /
00728  * o-o o-o
00729  *  ,---'
00730  * o-o o-o
00731  *  / / /
00732  * o-o o-o
00733  */
00734 
00735 /* Scan8 organization:
00736  *    0 1 2 3 4 5 6 7
00737  * 0  DY    y y y y y
00738  * 1        y Y Y Y Y
00739  * 2        y Y Y Y Y
00740  * 3        y Y Y Y Y
00741  * 4        y Y Y Y Y
00742  * 5  DU    u u u u u
00743  * 6        u U U U U
00744  * 7        u U U U U
00745  * 8        u U U U U
00746  * 9        u U U U U
00747  * 10 DV    v v v v v
00748  * 11       v V V V V
00749  * 12       v V V V V
00750  * 13       v V V V V
00751  * 14       v V V V V
00752  * DY/DU/DV are for luma/chroma DC.
00753  */
00754 
00755 #define LUMA_DC_BLOCK_INDEX   48
00756 #define CHROMA_DC_BLOCK_INDEX 49
00757 
00758 // This table must be here because scan8[constant] must be known at compiletime
00759 static const uint8_t scan8[16 * 3 + 3] = {
00760     4 +  1 * 8, 5 +  1 * 8, 4 +  2 * 8, 5 +  2 * 8,
00761     6 +  1 * 8, 7 +  1 * 8, 6 +  2 * 8, 7 +  2 * 8,
00762     4 +  3 * 8, 5 +  3 * 8, 4 +  4 * 8, 5 +  4 * 8,
00763     6 +  3 * 8, 7 +  3 * 8, 6 +  4 * 8, 7 +  4 * 8,
00764     4 +  6 * 8, 5 +  6 * 8, 4 +  7 * 8, 5 +  7 * 8,
00765     6 +  6 * 8, 7 +  6 * 8, 6 +  7 * 8, 7 +  7 * 8,
00766     4 +  8 * 8, 5 +  8 * 8, 4 +  9 * 8, 5 +  9 * 8,
00767     6 +  8 * 8, 7 +  8 * 8, 6 +  9 * 8, 7 +  9 * 8,
00768     4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
00769     6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
00770     4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
00771     6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
00772     0 +  0 * 8, 0 +  5 * 8, 0 + 10 * 8
00773 };
00774 
00775 static av_always_inline uint32_t pack16to32(int a, int b)
00776 {
00777 #if HAVE_BIGENDIAN
00778     return (b & 0xFFFF) + (a << 16);
00779 #else
00780     return (a & 0xFFFF) + (b << 16);
00781 #endif
00782 }
00783 
00784 static av_always_inline uint16_t pack8to16(int a, int b)
00785 {
00786 #if HAVE_BIGENDIAN
00787     return (b & 0xFF) + (a << 8);
00788 #else
00789     return (a & 0xFF) + (b << 8);
00790 #endif
00791 }
00792 
00796 static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale)
00797 {
00798     return h->pps.chroma_qp_table[t][qscale];
00799 }
00800 
00804 static av_always_inline int pred_intra_mode(H264Context *h, int n)
00805 {
00806     const int index8 = scan8[n];
00807     const int left   = h->intra4x4_pred_mode_cache[index8 - 1];
00808     const int top    = h->intra4x4_pred_mode_cache[index8 - 8];
00809     const int min    = FFMIN(left, top);
00810 
00811     tprintf(h->s.avctx, "mode:%d %d min:%d\n", left, top, min);
00812 
00813     if (min < 0)
00814         return DC_PRED;
00815     else
00816         return min;
00817 }
00818 
00819 static av_always_inline void write_back_intra_pred_mode(H264Context *h)
00820 {
00821     int8_t *i4x4       = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
00822     int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
00823 
00824     AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
00825     i4x4[4] = i4x4_cache[7 + 8 * 3];
00826     i4x4[5] = i4x4_cache[7 + 8 * 2];
00827     i4x4[6] = i4x4_cache[7 + 8 * 1];
00828 }
00829 
00830 static av_always_inline void write_back_non_zero_count(H264Context *h)
00831 {
00832     const int mb_xy    = h->mb_xy;
00833     uint8_t *nnz       = h->non_zero_count[mb_xy];
00834     uint8_t *nnz_cache = h->non_zero_count_cache;
00835 
00836     AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
00837     AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
00838     AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
00839     AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
00840     AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
00841     AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
00842     AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
00843     AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
00844 
00845     if (!h->s.chroma_y_shift) {
00846         AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
00847         AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
00848         AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
00849         AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
00850     }
00851 }
00852 
00853 static av_always_inline void write_back_motion_list(H264Context *h,
00854                                                     MpegEncContext *const s,
00855                                                     int b_stride,
00856                                                     int b_xy, int b8_xy,
00857                                                     int mb_type, int list)
00858 {
00859     int16_t(*mv_dst)[2] = &s->current_picture.f.motion_val[list][b_xy];
00860     int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
00861     AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
00862     AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
00863     AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
00864     AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
00865     if (CABAC) {
00866         uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
00867                                                         : h->mb2br_xy[h->mb_xy]];
00868         uint8_t(*mvd_src)[2]  = &h->mvd_cache[list][scan8[0]];
00869         if (IS_SKIP(mb_type)) {
00870             AV_ZERO128(mvd_dst);
00871         } else {
00872             AV_COPY64(mvd_dst, mvd_src + 8 * 3);
00873             AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
00874             AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
00875             AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
00876         }
00877     }
00878 
00879     {
00880         int8_t *ref_index = &s->current_picture.f.ref_index[list][b8_xy];
00881         int8_t *ref_cache = h->ref_cache[list];
00882         ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
00883         ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
00884         ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
00885         ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
00886     }
00887 }
00888 
00889 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
00890 {
00891     MpegEncContext *const s = &h->s;
00892     const int b_stride      = h->b_stride;
00893     const int b_xy  = 4 * s->mb_x + 4 * s->mb_y * h->b_stride; // try mb2b(8)_xy
00894     const int b8_xy = 4 * h->mb_xy;
00895 
00896     if (USES_LIST(mb_type, 0)) {
00897         write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
00898     } else {
00899         fill_rectangle(&s->current_picture.f.ref_index[0][b8_xy],
00900                        2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
00901     }
00902     if (USES_LIST(mb_type, 1))
00903         write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
00904 
00905     if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC) {
00906         if (IS_8X8(mb_type)) {
00907             uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
00908             direct_table[1] = h->sub_mb_type[1] >> 1;
00909             direct_table[2] = h->sub_mb_type[2] >> 1;
00910             direct_table[3] = h->sub_mb_type[3] >> 1;
00911         }
00912     }
00913 }
00914 
00915 static av_always_inline int get_dct8x8_allowed(H264Context *h)
00916 {
00917     if (h->sps.direct_8x8_inference_flag)
00918         return !(AV_RN64A(h->sub_mb_type) &
00919                  ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
00920                   0x0001000100010001ULL));
00921     else
00922         return !(AV_RN64A(h->sub_mb_type) &
00923                  ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
00924                   0x0001000100010001ULL));
00925 }
00926 
00927 #endif /* AVCODEC_H264_H */