annotate h264.h @ 10852:86d7ab878805 libavcodec

Get rid of #include "svq3.c" functions called more than per mb are moved into the header, scan8 is also as it must be known at compiletime. The code after this patch duplicates h264data.h, this has been done to minimize the changes in this step and allow more fine grained benchmarking. Speedwise this is 1% faster on my pentium dual core with diegos cursed cathedral sample.

/*
 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
 * Copyright (c) 2003 Michael Niedermayer <[e-mail]>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file libavcodec/h264.h
 * H.264 / AVC / MPEG4 part10 codec.
 * @author Michael Niedermayer <[e-mail]>
 */

#ifndef AVCODEC_H264_H
#define AVCODEC_H264_H

#include "dsputil.h"
#include "cabac.h"
#include "mpegvideo.h"
#include "h264pred.h"

#define interlaced_dct interlaced_dct_is_a_bad_name
#define mb_intra mb_intra_is_not_initialized_see_mb_type

#define LUMA_DC_BLOCK_INDEX   25
#define CHROMA_DC_BLOCK_INDEX 26

#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
#define COEFF_TOKEN_VLC_BITS           8
#define TOTAL_ZEROS_VLC_BITS           9
#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
#define RUN_VLC_BITS                   3
#define RUN7_VLC_BITS                  6

#define MAX_SPS_COUNT 32
#define MAX_PPS_COUNT 256

#define MAX_MMCO_COUNT 66

#define MAX_DELAYED_PIC_COUNT 16

/* Compiling in interlaced support reduces the speed
 * of progressive decoding by about 2%. */
#define ALLOW_INTERLACE

#define ALLOW_NOCHROMA

/**
 * The maximum number of slices supported by the decoder.
 * must be a power of 2
 */
#define MAX_SLICES 16

#ifdef ALLOW_INTERLACE
#define MB_MBAFF h->mb_mbaff
#define MB_FIELD h->mb_field_decoding_flag
#define FRAME_MBAFF h->mb_aff_frame
#define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
#else
#define MB_MBAFF 0
#define MB_FIELD 0
#define FRAME_MBAFF 0
#define FIELD_PICTURE 0
#undef  IS_INTERLACED
#define IS_INTERLACED(mb_type) 0
#endif
#define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)

#ifdef ALLOW_NOCHROMA
#define CHROMA h->sps.chroma_format_idc
#else
#define CHROMA 1
#endif

#define EXTENDED_SAR          255

#define MB_TYPE_REF0       MB_TYPE_ACPRED //dirty but it fits in 16 bit
#define MB_TYPE_8x8DCT     0x01000000
#define IS_REF0(a)         ((a) & MB_TYPE_REF0)
#define IS_8x8DCT(a)       ((a) & MB_TYPE_8x8DCT)

/* NAL unit types */
enum {
    NAL_SLICE=1,
    NAL_DPA,
    NAL_DPB,
    NAL_DPC,
    NAL_IDR_SLICE,
    NAL_SEI,
    NAL_SPS,
    NAL_PPS,
    NAL_AUD,
    NAL_END_SEQUENCE,
    NAL_END_STREAM,
    NAL_FILLER_DATA,
    NAL_SPS_EXT,
    NAL_AUXILIARY_SLICE=19
};

/**
 * SEI message types
 */
typedef enum {
    SEI_BUFFERING_PERIOD             =  0, ///< buffering period (H.264, D.1.1)
    SEI_TYPE_PIC_TIMING              =  1, ///< picture timing
    SEI_TYPE_USER_DATA_UNREGISTERED  =  5, ///< unregistered user data
    SEI_TYPE_RECOVERY_POINT          =  6  ///< recovery point (frame # to decoder sync)
} SEI_Type;

/**
 * pic_struct in picture timing SEI message
 */
typedef enum {
    SEI_PIC_STRUCT_FRAME             = 0, ///<  0: %frame
    SEI_PIC_STRUCT_TOP_FIELD         = 1, ///<  1: top field
    SEI_PIC_STRUCT_BOTTOM_FIELD      = 2, ///<  2: bottom field
    SEI_PIC_STRUCT_TOP_BOTTOM        = 3, ///<  3: top field, bottom field, in that order
    SEI_PIC_STRUCT_BOTTOM_TOP        = 4, ///<  4: bottom field, top field, in that order
    SEI_PIC_STRUCT_TOP_BOTTOM_TOP    = 5, ///<  5: top field, bottom field, top field repeated, in that order
    SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///<  6: bottom field, top field, bottom field repeated, in that order
    SEI_PIC_STRUCT_FRAME_DOUBLING    = 7, ///<  7: %frame doubling
    SEI_PIC_STRUCT_FRAME_TRIPLING    = 8  ///<  8: %frame tripling
} SEI_PicStructType;

/**
 * Sequence parameter set
 */
typedef struct SPS{

    int profile_idc;
    int level_idc;
    int chroma_format_idc;
    int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
    int poc_type;                      ///< pic_order_cnt_type
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
    int delta_pic_order_always_zero_flag;
    int offset_for_non_ref_pic;
    int offset_for_top_to_bottom_field;
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
    int ref_frame_count;               ///< num_ref_frames
    int gaps_in_frame_num_allowed_flag;
    int mb_width;                      ///< pic_width_in_mbs_minus1 + 1
    int mb_height;                     ///< pic_height_in_map_units_minus1 + 1
    int frame_mbs_only_flag;
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
    int direct_8x8_inference_flag;
    int crop;                   ///< frame_cropping_flag
    unsigned int crop_left;            ///< frame_cropping_rect_left_offset
    unsigned int crop_right;           ///< frame_cropping_rect_right_offset
    unsigned int crop_top;             ///< frame_cropping_rect_top_offset
    unsigned int crop_bottom;          ///< frame_cropping_rect_bottom_offset
    int vui_parameters_present_flag;
    AVRational sar;
    int video_signal_type_present_flag;
    int full_range;
    int colour_description_present_flag;
    enum AVColorPrimaries color_primaries;
    enum AVColorTransferCharacteristic color_trc;
    enum AVColorSpace colorspace;
    int timing_info_present_flag;
    uint32_t num_units_in_tick;
    uint32_t time_scale;
    int fixed_frame_rate_flag;
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
    int bitstream_restriction_flag;
    int num_reorder_frames;
    int scaling_matrix_present;
    uint8_t scaling_matrix4[6][16];
    uint8_t scaling_matrix8[2][64];
    int nal_hrd_parameters_present_flag;
    int vcl_hrd_parameters_present_flag;
    int pic_struct_present_flag;
    int time_offset_length;
    int cpb_cnt;                       ///< See H.264 E.1.2
    int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
    int cpb_removal_delay_length;      ///< cpb_removal_delay_length_minus1 + 1
    int dpb_output_delay_length;       ///< dpb_output_delay_length_minus1 + 1
    int bit_depth_luma;                ///< bit_depth_luma_minus8 + 8
    int bit_depth_chroma;              ///< bit_depth_chroma_minus8 + 8
    int residual_color_transform_flag; ///< residual_colour_transform_flag
}SPS;

/**
 * Picture parameter set
 */
typedef struct PPS{
    unsigned int sps_id;
    int cabac;                  ///< entropy_coding_mode_flag
    int pic_order_present;      ///< pic_order_present_flag
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
    int mb_slice_group_map_type;
    unsigned int ref_count[2];  ///< num_ref_idx_l0/1_active_minus1 + 1
    int weighted_pred;          ///< weighted_pred_flag
    int weighted_bipred_idc;
    int init_qp;                ///< pic_init_qp_minus26 + 26
    int init_qs;                ///< pic_init_qs_minus26 + 26
    int chroma_qp_index_offset[2];
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
    int constrained_intra_pred; ///< constrained_intra_pred_flag
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
    uint8_t scaling_matrix4[6][16];
    uint8_t scaling_matrix8[2][64];
    uint8_t chroma_qp_table[2][64];  ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
    int chroma_qp_diff;
}PPS;

/**
 * Memory management control operation opcode.
 */
typedef enum MMCOOpcode{
    MMCO_END=0,
    MMCO_SHORT2UNUSED,
    MMCO_LONG2UNUSED,
    MMCO_SHORT2LONG,
    MMCO_SET_MAX_LONG,
    MMCO_RESET,
    MMCO_LONG,
} MMCOOpcode;

/**
 * Memory management control operation.
 */
typedef struct MMCO{
    MMCOOpcode opcode;
    int short_pic_num;  ///< pic_num without wrapping (pic_num & max_pic_num)
    int long_arg;       ///< index, pic_num, or num long refs depending on opcode
} MMCO;

/**
 * H264Context
 */
typedef struct H264Context{
    MpegEncContext s;
    int nal_ref_idc;
    int nal_unit_type;
    uint8_t *rbsp_buffer[2];
    unsigned int rbsp_buffer_size[2];

    /**
      * Used to parse AVC variant of h264
      */
    int is_avc; ///< this flag is != 0 if codec is avc1
    int got_avcC; ///< flag used to parse avcC data only once
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)

    int chroma_qp[2]; //QPc

    int prev_mb_skipped;
    int next_mb_skipped;

    //prediction stuff
    int chroma_pred_mode;
    int intra16x16_pred_mode;

    int top_mb_xy;
    int left_mb_xy[2];

    int8_t intra4x4_pred_mode_cache[5*8];
    int8_t (*intra4x4_pred_mode)[8];
    H264PredContext hpc;
    unsigned int topleft_samples_available;
    unsigned int top_samples_available;
    unsigned int topright_samples_available;
    unsigned int left_samples_available;
    uint8_t (*top_borders[2])[16+2*8];
    uint8_t left_border[2*(17+2*9)];

    /**
     * non zero coeff count cache.
     * is 64 if not available.
     */
    DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
    uint8_t (*non_zero_count)[16];

    /**
     * Motion vector cache.
     */
    DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
    DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
#define LIST_NOT_USED -1 //FIXME rename?
#define PART_NOT_AVAILABLE -2

    /**
     * is 1 if the specific list MV&references are set to 0,0,-2.
     */
    int mv_cache_clean[2];

    /**
     * number of neighbors (top and/or left) that used 8x8 dct
     */
    int neighbor_transform_size;

    /**
     * block_offset[ 0..23] for frame macroblocks
     * block_offset[24..47] for field macroblocks
     */
    int block_offset[2*(16+8)];

    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
    uint32_t *mb2b8_xy;
    int b_stride; //FIXME use s->b4_stride
    int b8_stride;

    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
    int mb_uvlinesize;

    int emu_edge_width;
    int emu_edge_height;

    int halfpel_flag;
    int thirdpel_flag;

    int unknown_svq3_flag;
    int next_slice_index;

    SPS *sps_buffers[MAX_SPS_COUNT];
    SPS sps; ///< current sps

    PPS *pps_buffers[MAX_PPS_COUNT];
    /**
     * current pps
     */
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?

    uint32_t dequant4_buffer[6][52][16];
    uint32_t dequant8_buffer[2][52][64];
    uint32_t (*dequant4_coeff[6])[16];
    uint32_t (*dequant8_coeff[2])[64];
    int dequant_coeff_pps;     ///< reinit tables when pps changes

    int slice_num;
    uint16_t *slice_table_base;
    uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
    int slice_type;
    int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
    int slice_type_fixed;

    //interlacing specific flags
    int mb_aff_frame;
    int mb_field_decoding_flag;
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag

    DECLARE_ALIGNED_8(uint16_t, sub_mb_type[4]);

    //POC stuff
    int poc_lsb;
    int poc_msb;
    int delta_poc_bottom;
    int delta_poc[2];
    int frame_num;
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
    int frame_num_offset;         ///< for POC type 2
    int prev_frame_num_offset;    ///< for POC type 2
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2

    /**
     * frame_num for frames or 2*frame_num+1 for field pics.
     */
    int curr_pic_num;

    /**
     * max_frame_num or 2*max_frame_num for field pics.
     */
    int max_pic_num;

    //Weighted pred stuff
    int use_weight;
    int use_weight_chroma;
    int luma_log2_weight_denom;
    int chroma_log2_weight_denom;
    int luma_weight[2][48];
    int luma_offset[2][48];
    int chroma_weight[2][48][2];
    int chroma_offset[2][48][2];
    int implicit_weight[48][48];

    //deblock
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
    int slice_alpha_c0_offset;
    int slice_beta_offset;

    int redundant_pic_count;

    int direct_spatial_mv_pred;
    int dist_scale_factor[16];
    int dist_scale_factor_field[2][32];
    int map_col_to_list0[2][16+32];
    int map_col_to_list0_field[2][2][16+32];

    /**
     * num_ref_idx_l0/1_active_minus1 + 1
     */
    unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
    unsigned int list_count;
    Picture *short_ref[32];
    Picture *long_ref[32];
    Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
    Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
                                          Reordered version of default_ref_list
                                          according to picture reordering in slice header */
    int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
    Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
    int outputed_poc;

    /**
     * memory management control operations buffer.
     */
    MMCO mmco[MAX_MMCO_COUNT];
    int mmco_index;

    int long_ref_count;  ///< number of actual long term references
    int short_ref_count; ///< number of actual short term references

    //data partitioning
    GetBitContext intra_gb;
    GetBitContext inter_gb;
    GetBitContext *intra_gb_ptr;
    GetBitContext *inter_gb_ptr;

    DECLARE_ALIGNED_16(DCTELEM, mb[16*24]);
    DCTELEM mb_padding[256];        ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb

    /**
     * Cabac
     */
    CABACContext cabac;
    uint8_t      cabac_state[460];
    int          cabac_init_idc;

    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
    uint16_t     *cbp_table;
    int cbp;
    int top_cbp;
    int left_cbp;
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
    uint8_t     *chroma_pred_mode_table;
    int         last_qscale_diff;
    int16_t     (*mvd_table[2])[2];
    DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
    uint8_t     *direct_table;
    uint8_t     direct_cache[5*8];

    uint8_t zigzag_scan[16];
    uint8_t zigzag_scan8x8[64];
    uint8_t zigzag_scan8x8_cavlc[64];
    uint8_t field_scan[16];
    uint8_t field_scan8x8[64];
    uint8_t field_scan8x8_cavlc[64];
    const uint8_t *zigzag_scan_q0;
    const uint8_t *zigzag_scan8x8_q0;
    const uint8_t *zigzag_scan8x8_cavlc_q0;
    const uint8_t *field_scan_q0;
    const uint8_t *field_scan8x8_q0;
    const uint8_t *field_scan8x8_cavlc_q0;

    int x264_build;

    /**
     * @defgroup multithreading Members for slice based multithreading
     * @{
     */
    struct H264Context *thread_context[MAX_THREADS];

    /**
     * current slice number, used to initalize slice_num of each thread/context
     */
    int current_slice;

    /**
     * Max number of threads / contexts.
     * This is equal to AVCodecContext.thread_count unless
     * multithreaded decoding is impossible, in which case it is
     * reduced to 1.
     */
    int max_contexts;

    /**
     *  1 if the single thread fallback warning has already been
     *  displayed, 0 otherwise.
     */
    int single_decode_warning;

    int last_slice_type;
    /** @} */

    int mb_xy;

    uint32_t svq3_watermark_key;

    /**
     * pic_struct in picture timing SEI message
     */
    SEI_PicStructType sei_pic_struct;

    /**
     * Complement sei_pic_struct
     * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
     * However, soft telecined frames may have these values.
     * This is used in an attempt to flag soft telecine progressive.
     */
    int prev_interlaced_frame;

    /**
     * Bit set of clock types for fields/frames in picture timing SEI message.
     * For each found ct_type, appropriate bit is set (e.g., bit 1 for
     * interlaced).
     */
    int sei_ct_type;

    /**
     * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
     */
    int sei_dpb_output_delay;

    /**
     * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
     */
    int sei_cpb_removal_delay;

    /**
     * recovery_frame_cnt from SEI message
     *
     * Set to -1 if no recovery point SEI message found or to number of frames
     * before playback synchronizes. Frames having recovery point are key
     * frames.
     */
    int sei_recovery_frame_cnt;

    int is_complex;

    int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
    int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag

    // Timestamp stuff
    int sei_buffering_period_present;  ///< Buffering period SEI flag
    int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
}H264Context;

/**
 * Decode SEI
 */
int ff_h264_decode_sei(H264Context *h);

/**
 * Decode SPS
 */
int ff_h264_decode_seq_parameter_set(H264Context *h);

/**
 * Decode PPS
 */
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);

/**
 * Decodes a network abstraction layer unit.
 * @param consumed is the number of bytes used as input
 * @param length is the length of the array
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
 * @returns decoded bytes, might be src+1 if no escapes
 */
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);

/**
 * identifies the exact end of the bitstream
 * @return the length of the trailing, or 0 if damaged
 */
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);

/**
 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
 */
av_cold void ff_h264_free_context(H264Context *h);

/**
 * reconstructs bitstream slice_type.
 */
int ff_h264_get_slice_type(H264Context *h);

/**
 * allocates tables.
 * needs width/height
 */
int ff_h264_alloc_tables(H264Context *h);

/**
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 */
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);

void ff_h264_write_back_intra_pred_mode(H264Context *h);
void ff_h264_hl_decode_mb(H264Context *h);
int ff_h264_frame_start(H264Context *h);
av_cold int ff_h264_decode_init(AVCodecContext *avctx);
av_cold int ff_h264_decode_end(AVCodecContext *avctx);

/*
o-o o-o
 / / /
o-o o-o
 ,---'
o-o o-o
 / / /
o-o o-o
*/
//This table must be here because scan8[constant] must be known at compiletime
static const uint8_t scan8[16 + 2*4]={
 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,
 1+1*8, 2+1*8,
 1+2*8, 2+2*8,
 1+4*8, 2+4*8,
 1+5*8, 2+5*8,
};

static av_always_inline uint32_t pack16to32(int a, int b){
#if HAVE_BIGENDIAN
   return (b&0xFFFF) + (a<<16);
#else
   return (a&0xFFFF) + (b<<16);
#endif
}

/**
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 */
static inline int check_intra4x4_pred_mode(H264Context *h){
    MpegEncContext * const s = &h->s;
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
    int i;

    if(!(h->top_samples_available&0x8000)){
        for(i=0; i<4; i++){
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
            if(status<0){
                av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
                return -1;
            } else if(status){
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
            }
        }
    }

    if((h->left_samples_available&0x8888)!=0x8888){
        static const int mask[4]={0x8000,0x2000,0x80,0x20};
        for(i=0; i<4; i++){
            if(!(h->left_samples_available&mask[i])){
                int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
                if(status<0){
                    av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
                    return -1;
                } else if(status){
                    h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
                }
            }
        }
    }

    return 0;
} //FIXME cleanup like ff_h264_check_intra_pred_mode

static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
    MpegEncContext *s = &h->s;

    /* there is no consistent mapping of mvs to neighboring locations that will
     * make mbaff happy, so we can't move all this logic to fill_caches */
    if(FRAME_MBAFF){
        const uint32_t *mb_types = s->current_picture_ptr->mb_type;
        const int16_t *mv;
        *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
        *C = h->mv_cache[list][scan8[0]-2];

        if(!MB_FIELD
           && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
            int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
            if(IS_INTERLACED(mb_types[topright_xy])){
#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
                const int x4 = X4, y4 = Y4;\
                const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
                if(!USES_LIST(mb_type,list))\
                    return LIST_NOT_USED;\
                mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
                h->mv_cache[list][scan8[0]-2][0] = mv[0];\
                h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
                return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;

                SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
            }
        }
        if(topright_ref == PART_NOT_AVAILABLE
           && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
           && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
            if(!MB_FIELD
               && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
                SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
            }
            if(MB_FIELD
               && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
               && i >= scan8[0]+8){
                // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
                SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
            }
        }
#undef SET_DIAG_MV
    }

    if(topright_ref != PART_NOT_AVAILABLE){
        *C= h->mv_cache[list][ i - 8 + part_width ];
        return topright_ref;
    }else{
        tprintf(s->avctx, "topright MV not available\n");

        *C= h->mv_cache[list][ i - 8 - 1 ];
        return h->ref_cache[list][ i - 8 - 1 ];
    }
}

/**
 * gets the predicted MV.
 * @param n the block index
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
 * @param mx the x component of the predicted motion vector
 * @param my the y component of the predicted motion vector
 */
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
    const int index8= scan8[n];
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
    const int16_t * C;
    int diagonal_ref, match_count;

    assert(part_width==1 || part_width==2 || part_width==4);

/* mv_cache
  B . . A T T T T
  U . . L . . , .
  U . . L . . . .
  U . . L . . , .
  . . . L . . . .
*/

    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
    tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
    if(match_count > 1){ //most common
        *mx= mid_pred(A[0], B[0], C[0]);
        *my= mid_pred(A[1], B[1], C[1]);
    }else if(match_count==1){
        if(left_ref==ref){
            *mx= A[0];
            *my= A[1];
        }else if(top_ref==ref){
            *mx= B[0];
            *my= B[1];
        }else{
            *mx= C[0];
            *my= C[1];
        }
    }else{
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
            *mx= A[0];
            *my= A[1];
        }else{
            *mx= mid_pred(A[0], B[0], C[0]);
            *my= mid_pred(A[1], B[1], C[1]);
        }
    }

    tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1],                    diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
}


#endif /* AVCODEC_H264_H */