falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
7621 Commits
32 Branches
303MB
Tree: 035314b683
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '035314b683'
${ noResults }
FFmpeg/libavcodec/h264.c

8571 lines
321KB
Raw Blame History 

  1. /*

  2.  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder

  3.  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>

  4.  *

  5.  * This file is part of FFmpeg.

  6.  *

  7.  * FFmpeg is free software; you can redistribute it and/or

  8.  * modify it under the terms of the GNU Lesser General Public

  9.  * License as published by the Free Software Foundation; either

  10.  * version 2.1 of the License, or (at your option) any later version.

  11.  *

  12.  * FFmpeg is distributed in the hope that it will be useful,

  13.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  14.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  15.  * Lesser General Public License for more details.

  16.  *

  17.  * You should have received a copy of the GNU Lesser General Public

  18.  * License along with FFmpeg; if not, write to the Free Software

  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  20.  *

  21.  */

  22. 

  23. /**

  24.  * @file h264.c

  25.  * H.264 / AVC / MPEG4 part10 codec.

  26.  * @author Michael Niedermayer <michaelni@gmx.at>

  27.  */

  28. 

  29. #include "common.h"

  30. #include "dsputil.h"

  31. #include "avcodec.h"

  32. #include "mpegvideo.h"

  33. #include "h264data.h"

  34. #include "golomb.h"

  35. 

  36. #include "cabac.h"

  37. 

  38. //#undef NDEBUG

  39. #include <assert.h>

  40. 

  41. #define interlaced_dct interlaced_dct_is_a_bad_name

  42. #define mb_intra mb_intra_isnt_initalized_see_mb_type

  43. 

  44. #define LUMA_DC_BLOCK_INDEX   25

  45. #define CHROMA_DC_BLOCK_INDEX 26

  46. 

  47. #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8

  48. #define COEFF_TOKEN_VLC_BITS           8

  49. #define TOTAL_ZEROS_VLC_BITS           9

  50. #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3

  51. #define RUN_VLC_BITS                   3

  52. #define RUN7_VLC_BITS                  6

  53. 

  54. #define MAX_SPS_COUNT 32

  55. #define MAX_PPS_COUNT 256

  56. 

  57. #define MAX_MMCO_COUNT 66

  58. 

  59. /* Compiling in interlaced support reduces the speed

  60.  * of progressive decoding by about 2%. */

  61. #define ALLOW_INTERLACE

  62. 

  63. #ifdef ALLOW_INTERLACE

  64. #define MB_MBAFF h->mb_mbaff

  65. #define MB_FIELD h->mb_field_decoding_flag

  66. #define FRAME_MBAFF h->mb_aff_frame

  67. #else

  68. #define MB_MBAFF 0

  69. #define MB_FIELD 0

  70. #define FRAME_MBAFF 0

  71. #undef  IS_INTERLACED

  72. #define IS_INTERLACED(mb_type) 0

  73. #endif

  74. 

  75. /**

  76.  * Sequence parameter set

  77.  */

  78. typedef struct SPS{

  79. 

  80.     int profile_idc;

  81.     int level_idc;

  82.     int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag

  83.     int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4

  84.     int poc_type;                      ///< pic_order_cnt_type

  85.     int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4

  86.     int delta_pic_order_always_zero_flag;

  87.     int offset_for_non_ref_pic;

  88.     int offset_for_top_to_bottom_field;

  89.     int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle

  90.     int ref_frame_count;               ///< num_ref_frames

  91.     int gaps_in_frame_num_allowed_flag;

  92.     int mb_width;                      ///< frame_width_in_mbs_minus1 + 1

  93.     int mb_height;                     ///< frame_height_in_mbs_minus1 + 1

  94.     int frame_mbs_only_flag;

  95.     int mb_aff;                        ///<mb_adaptive_frame_field_flag

  96.     int direct_8x8_inference_flag;

  97.     int crop;                   ///< frame_cropping_flag

  98.     int crop_left;              ///< frame_cropping_rect_left_offset

  99.     int crop_right;             ///< frame_cropping_rect_right_offset

  100.     int crop_top;               ///< frame_cropping_rect_top_offset

  101.     int crop_bottom;            ///< frame_cropping_rect_bottom_offset

  102.     int vui_parameters_present_flag;

  103.     AVRational sar;

  104.     int timing_info_present_flag;

  105.     uint32_t num_units_in_tick;

  106.     uint32_t time_scale;

  107.     int fixed_frame_rate_flag;

  108.     short offset_for_ref_frame[256]; //FIXME dyn aloc?

  109.     int bitstream_restriction_flag;

  110.     int num_reorder_frames;

  111.     int scaling_matrix_present;

  112.     uint8_t scaling_matrix4[6][16];

  113.     uint8_t scaling_matrix8[2][64];

  114. }SPS;

  115. 

  116. /**

  117.  * Picture parameter set

  118.  */

  119. typedef struct PPS{

  120.     int sps_id;

  121.     int cabac;                  ///< entropy_coding_mode_flag

  122.     int pic_order_present;      ///< pic_order_present_flag

  123.     int slice_group_count;      ///< num_slice_groups_minus1 + 1

  124.     int mb_slice_group_map_type;

  125.     int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1

  126.     int weighted_pred;          ///< weighted_pred_flag

  127.     int weighted_bipred_idc;

  128.     int init_qp;                ///< pic_init_qp_minus26 + 26

  129.     int init_qs;                ///< pic_init_qs_minus26 + 26

  130.     int chroma_qp_index_offset;

  131.     int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag

  132.     int constrained_intra_pred; ///< constrained_intra_pred_flag

  133.     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag

  134.     int transform_8x8_mode;     ///< transform_8x8_mode_flag

  135.     uint8_t scaling_matrix4[6][16];

  136.     uint8_t scaling_matrix8[2][64];

  137. }PPS;

  138. 

  139. /**

  140.  * Memory management control operation opcode.

  141.  */

  142. typedef enum MMCOOpcode{

  143.     MMCO_END=0,

  144.     MMCO_SHORT2UNUSED,

  145.     MMCO_LONG2UNUSED,

  146.     MMCO_SHORT2LONG,

  147.     MMCO_SET_MAX_LONG,

  148.     MMCO_RESET,

  149.     MMCO_LONG,

  150. } MMCOOpcode;

  151. 

  152. /**

  153.  * Memory management control operation.

  154.  */

  155. typedef struct MMCO{

  156.     MMCOOpcode opcode;

  157.     int short_frame_num;

  158.     int long_index;

  159. } MMCO;

  160. 

  161. /**

  162.  * H264Context

  163.  */

  164. typedef struct H264Context{

  165.     MpegEncContext s;

  166.     int nal_ref_idc;

  167.     int nal_unit_type;

  168.     uint8_t *rbsp_buffer;

  169.     unsigned int rbsp_buffer_size;

  170. 

  171.     /**

  172.       * Used to parse AVC variant of h264

  173.       */

  174.     int is_avc; ///< this flag is != 0 if codec is avc1

  175.     int got_avcC; ///< flag used to parse avcC data only once

  176.     int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)

  177. 

  178.     int chroma_qp; //QPc

  179. 

  180.     int prev_mb_skipped;

  181.     int next_mb_skipped;

  182. 

  183.     //prediction stuff

  184.     int chroma_pred_mode;

  185.     int intra16x16_pred_mode;

  186. 

  187.     int top_mb_xy;

  188.     int left_mb_xy[2];

  189. 

  190.     int8_t intra4x4_pred_mode_cache[5*8];

  191.     int8_t (*intra4x4_pred_mode)[8];

  192.     void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?

  193.     void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);

  194.     void (*pred8x8  [4+3])(uint8_t *src, int stride);

  195.     void (*pred16x16[4+3])(uint8_t *src, int stride);

  196.     unsigned int topleft_samples_available;

  197.     unsigned int top_samples_available;

  198.     unsigned int topright_samples_available;

  199.     unsigned int left_samples_available;

  200.     uint8_t (*top_borders[2])[16+2*8];

  201.     uint8_t left_border[2*(17+2*9)];

  202. 

  203.     /**

  204.      * non zero coeff count cache.

  205.      * is 64 if not available.

  206.      */

  207.     DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);

  208.     uint8_t (*non_zero_count)[16];

  209. 

  210.     /**

  211.      * Motion vector cache.

  212.      */

  213.     DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);

  214.     DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);

  215. #define LIST_NOT_USED -1 //FIXME rename?

  216. #define PART_NOT_AVAILABLE -2

  217. 

  218.     /**

  219.      * is 1 if the specific list MV&references are set to 0,0,-2.

  220.      */

  221.     int mv_cache_clean[2];

  222. 

  223.     /**

  224.      * number of neighbors (top and/or left) that used 8x8 dct

  225.      */

  226.     int neighbor_transform_size;

  227. 

  228.     /**

  229.      * block_offset[ 0..23] for frame macroblocks

  230.      * block_offset[24..47] for field macroblocks

  231.      */

  232.     int block_offset[2*(16+8)];

  233. 

  234.     uint32_t *mb2b_xy; //FIXME are these 4 a good idea?

  235.     uint32_t *mb2b8_xy;

  236.     int b_stride; //FIXME use s->b4_stride

  237.     int b8_stride;

  238. 

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

  240.     int mb_uvlinesize;

  241. 

  242.     int emu_edge_width;

  243.     int emu_edge_height;

  244. 

  245.     int halfpel_flag;

  246.     int thirdpel_flag;

  247. 

  248.     int unknown_svq3_flag;

  249.     int next_slice_index;

  250. 

  251.     SPS sps_buffer[MAX_SPS_COUNT];

  252.     SPS sps; ///< current sps

  253. 

  254.     PPS pps_buffer[MAX_PPS_COUNT];

  255.     /**

  256.      * current pps

  257.      */

  258.     PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?

  259. 

  260.     uint32_t dequant4_buffer[6][52][16];

  261.     uint32_t dequant8_buffer[2][52][64];

  262.     uint32_t (*dequant4_coeff[6])[16];

  263.     uint32_t (*dequant8_coeff[2])[64];

  264.     int dequant_coeff_pps;     ///< reinit tables when pps changes

  265. 

  266.     int slice_num;

  267.     uint8_t *slice_table_base;

  268.     uint8_t *slice_table;      ///< slice_table_base + 2*mb_stride + 1

  269.     int slice_type;

  270.     int slice_type_fixed;

  271. 

  272.     //interlacing specific flags

  273.     int mb_aff_frame;

  274.     int mb_field_decoding_flag;

  275.     int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag

  276. 

  277.     int sub_mb_type[4];

  278. 

  279.     //POC stuff

  280.     int poc_lsb;

  281.     int poc_msb;

  282.     int delta_poc_bottom;

  283.     int delta_poc[2];

  284.     int frame_num;

  285.     int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0

  286.     int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0

  287.     int frame_num_offset;         ///< for POC type 2

  288.     int prev_frame_num_offset;    ///< for POC type 2

  289.     int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2

  290. 

  291.     /**

  292.      * frame_num for frames or 2*frame_num for field pics.

  293.      */

  294.     int curr_pic_num;

  295. 

  296.     /**

  297.      * max_frame_num or 2*max_frame_num for field pics.

  298.      */

  299.     int max_pic_num;

  300. 

  301.     //Weighted pred stuff

  302.     int use_weight;

  303.     int use_weight_chroma;

  304.     int luma_log2_weight_denom;

  305.     int chroma_log2_weight_denom;

  306.     int luma_weight[2][48];

  307.     int luma_offset[2][48];

  308.     int chroma_weight[2][48][2];

  309.     int chroma_offset[2][48][2];

  310.     int implicit_weight[48][48];

  311. 

  312.     //deblock

  313.     int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0

  314.     int slice_alpha_c0_offset;

  315.     int slice_beta_offset;

  316. 

  317.     int redundant_pic_count;

  318. 

  319.     int direct_spatial_mv_pred;

  320.     int dist_scale_factor[16];

  321.     int dist_scale_factor_field[32];

  322.     int map_col_to_list0[2][16];

  323.     int map_col_to_list0_field[2][32];

  324. 

  325.     /**

  326.      * num_ref_idx_l0/1_active_minus1 + 1

  327.      */

  328.     int ref_count[2];            ///< counts frames or fields, depending on current mb mode

  329.     Picture *short_ref[32];

  330.     Picture *long_ref[32];

  331.     Picture default_ref_list[2][32];

  332.     Picture ref_list[2][48];     ///< 0..15: frame refs, 16..47: mbaff field refs

  333.     Picture *delayed_pic[16]; //FIXME size?

  334.     Picture *delayed_output_pic;

  335. 

  336.     /**

  337.      * memory management control operations buffer.

  338.      */

  339.     MMCO mmco[MAX_MMCO_COUNT];

  340.     int mmco_index;

  341. 

  342.     int long_ref_count;  ///< number of actual long term references

  343.     int short_ref_count; ///< number of actual short term references

  344. 

  345.     //data partitioning

  346.     GetBitContext intra_gb;

  347.     GetBitContext inter_gb;

  348.     GetBitContext *intra_gb_ptr;

  349.     GetBitContext *inter_gb_ptr;

  350. 

  351.     DECLARE_ALIGNED_8(DCTELEM, mb[16*24]);

  352. 

  353.     /**

  354.      * Cabac

  355.      */

  356.     CABACContext cabac;

  357.     uint8_t      cabac_state[460];

  358.     int          cabac_init_idc;

  359. 

  360.     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */

  361.     uint16_t     *cbp_table;

  362.     int cbp;

  363.     int top_cbp;

  364.     int left_cbp;

  365.     /* chroma_pred_mode for i4x4 or i16x16, else 0 */

  366.     uint8_t     *chroma_pred_mode_table;

  367.     int         last_qscale_diff;

  368.     int16_t     (*mvd_table[2])[2];

  369.     DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);

  370.     uint8_t     *direct_table;

  371.     uint8_t     direct_cache[5*8];

  372. 

  373.     uint8_t zigzag_scan[16];

  374.     uint8_t zigzag_scan8x8[64];

  375.     uint8_t zigzag_scan8x8_cavlc[64];

  376.     uint8_t field_scan[16];

  377.     uint8_t field_scan8x8[64];

  378.     uint8_t field_scan8x8_cavlc[64];

  379.     const uint8_t *zigzag_scan_q0;

  380.     const uint8_t *zigzag_scan8x8_q0;

  381.     const uint8_t *zigzag_scan8x8_cavlc_q0;

  382.     const uint8_t *field_scan_q0;

  383.     const uint8_t *field_scan8x8_q0;

  384.     const uint8_t *field_scan8x8_cavlc_q0;

  385. 

  386.     int x264_build;

  387. }H264Context;

  388. 

  389. static VLC coeff_token_vlc[4];

  390. static VLC chroma_dc_coeff_token_vlc;

  391. 

  392. static VLC total_zeros_vlc[15];

  393. static VLC chroma_dc_total_zeros_vlc[3];

  394. 

  395. static VLC run_vlc[6];

  396. static VLC run7_vlc;

  397. 

  398. static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);

  399. static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);

  400. static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);

  401. static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);

  402. 

  403. static always_inline uint32_t pack16to32(int a, int b){

  404. #ifdef WORDS_BIGENDIAN

  405.    return (b&0xFFFF) + (a<<16);

  406. #else

  407.    return (a&0xFFFF) + (b<<16);

  408. #endif

  409. }

  410. 

  411. /**

  412.  * fill a rectangle.

  413.  * @param h height of the rectangle, should be a constant

  414.  * @param w width of the rectangle, should be a constant

  415.  * @param size the size of val (1 or 4), should be a constant

  416.  */

  417. static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){

  418.     uint8_t *p= (uint8_t*)vp;

  419.     assert(size==1 || size==4);

  420.     assert(w<=4);

  421. 

  422.     w      *= size;

  423.     stride *= size;

  424. 

  425.     assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);

  426.     assert((stride&(w-1))==0);

  427.     if(w==2){

  428.         const uint16_t v= size==4 ? val : val*0x0101;

  429.         *(uint16_t*)(p + 0*stride)= v;

  430.         if(h==1) return;

  431.         *(uint16_t*)(p + 1*stride)= v;

  432.         if(h==2) return;

  433.         *(uint16_t*)(p + 2*stride)=

  434.         *(uint16_t*)(p + 3*stride)= v;

  435.     }else if(w==4){

  436.         const uint32_t v= size==4 ? val : val*0x01010101;

  437.         *(uint32_t*)(p + 0*stride)= v;

  438.         if(h==1) return;

  439.         *(uint32_t*)(p + 1*stride)= v;

  440.         if(h==2) return;

  441.         *(uint32_t*)(p + 2*stride)=

  442.         *(uint32_t*)(p + 3*stride)= v;

  443.     }else if(w==8){

  444.     //gcc can't optimize 64bit math on x86_32

  445. #if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)

  446.         const uint64_t v= val*0x0100000001ULL;

  447.         *(uint64_t*)(p + 0*stride)= v;

  448.         if(h==1) return;

  449.         *(uint64_t*)(p + 1*stride)= v;

  450.         if(h==2) return;

  451.         *(uint64_t*)(p + 2*stride)=

  452.         *(uint64_t*)(p + 3*stride)= v;

  453.     }else if(w==16){

  454.         const uint64_t v= val*0x0100000001ULL;

  455.         *(uint64_t*)(p + 0+0*stride)=

  456.         *(uint64_t*)(p + 8+0*stride)=

  457.         *(uint64_t*)(p + 0+1*stride)=

  458.         *(uint64_t*)(p + 8+1*stride)= v;

  459.         if(h==2) return;

  460.         *(uint64_t*)(p + 0+2*stride)=

  461.         *(uint64_t*)(p + 8+2*stride)=

  462.         *(uint64_t*)(p + 0+3*stride)=

  463.         *(uint64_t*)(p + 8+3*stride)= v;

  464. #else

  465.         *(uint32_t*)(p + 0+0*stride)=

  466.         *(uint32_t*)(p + 4+0*stride)= val;

  467.         if(h==1) return;

  468.         *(uint32_t*)(p + 0+1*stride)=

  469.         *(uint32_t*)(p + 4+1*stride)= val;

  470.         if(h==2) return;

  471.         *(uint32_t*)(p + 0+2*stride)=

  472.         *(uint32_t*)(p + 4+2*stride)=

  473.         *(uint32_t*)(p + 0+3*stride)=

  474.         *(uint32_t*)(p + 4+3*stride)= val;

  475.     }else if(w==16){

  476.         *(uint32_t*)(p + 0+0*stride)=

  477.         *(uint32_t*)(p + 4+0*stride)=

  478.         *(uint32_t*)(p + 8+0*stride)=

  479.         *(uint32_t*)(p +12+0*stride)=

  480.         *(uint32_t*)(p + 0+1*stride)=

  481.         *(uint32_t*)(p + 4+1*stride)=

  482.         *(uint32_t*)(p + 8+1*stride)=

  483.         *(uint32_t*)(p +12+1*stride)= val;

  484.         if(h==2) return;

  485.         *(uint32_t*)(p + 0+2*stride)=

  486.         *(uint32_t*)(p + 4+2*stride)=

  487.         *(uint32_t*)(p + 8+2*stride)=

  488.         *(uint32_t*)(p +12+2*stride)=

  489.         *(uint32_t*)(p + 0+3*stride)=

  490.         *(uint32_t*)(p + 4+3*stride)=

  491.         *(uint32_t*)(p + 8+3*stride)=

  492.         *(uint32_t*)(p +12+3*stride)= val;

  493. #endif

  494.     }else

  495.         assert(0);

  496.     assert(h==4);

  497. }

  498. 

  499. static void fill_caches(H264Context *h, int mb_type, int for_deblock){

  500.     MpegEncContext * const s = &h->s;

  501.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  502.     int topleft_xy, top_xy, topright_xy, left_xy[2];

  503.     int topleft_type, top_type, topright_type, left_type[2];

  504.     int left_block[8];

  505.     int i;

  506. 

  507.     //FIXME deblocking could skip the intra and nnz parts.

  508.     if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)

  509.         return;

  510. 

  511.     //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it

  512. 

  513.     top_xy     = mb_xy  - s->mb_stride;

  514.     topleft_xy = top_xy - 1;

  515.     topright_xy= top_xy + 1;

  516.     left_xy[1] = left_xy[0] = mb_xy-1;

  517.     left_block[0]= 0;

  518.     left_block[1]= 1;

  519.     left_block[2]= 2;

  520.     left_block[3]= 3;

  521.     left_block[4]= 7;

  522.     left_block[5]= 10;

  523.     left_block[6]= 8;

  524.     left_block[7]= 11;

  525.     if(FRAME_MBAFF){

  526.         const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;

  527.         const int top_pair_xy      = pair_xy     - s->mb_stride;

  528.         const int topleft_pair_xy  = top_pair_xy - 1;

  529.         const int topright_pair_xy = top_pair_xy + 1;

  530.         const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);

  531.         const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);

  532.         const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);

  533.         const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);

  534.         const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);

  535.         const int bottom = (s->mb_y & 1);

  536.         tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);

  537.         if (bottom

  538.                 ? !curr_mb_frame_flag // bottom macroblock

  539.                 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock

  540.                 ) {

  541.             top_xy -= s->mb_stride;

  542.         }

  543.         if (bottom

  544.                 ? !curr_mb_frame_flag // bottom macroblock

  545.                 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock

  546.                 ) {

  547.             topleft_xy -= s->mb_stride;

  548.         }

  549.         if (bottom

  550.                 ? !curr_mb_frame_flag // bottom macroblock

  551.                 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock

  552.                 ) {

  553.             topright_xy -= s->mb_stride;

  554.         }

  555.         if (left_mb_frame_flag != curr_mb_frame_flag) {

  556.             left_xy[1] = left_xy[0] = pair_xy - 1;

  557.             if (curr_mb_frame_flag) {

  558.                 if (bottom) {

  559.                     left_block[0]= 2;

  560.                     left_block[1]= 2;

  561.                     left_block[2]= 3;

  562.                     left_block[3]= 3;

  563.                     left_block[4]= 8;

  564.                     left_block[5]= 11;

  565.                     left_block[6]= 8;

  566.                     left_block[7]= 11;

  567.                 } else {

  568.                     left_block[0]= 0;

  569.                     left_block[1]= 0;

  570.                     left_block[2]= 1;

  571.                     left_block[3]= 1;

  572.                     left_block[4]= 7;

  573.                     left_block[5]= 10;

  574.                     left_block[6]= 7;

  575.                     left_block[7]= 10;

  576.                 }

  577.             } else {

  578.                 left_xy[1] += s->mb_stride;

  579.                 //left_block[0]= 0;

  580.                 left_block[1]= 2;

  581.                 left_block[2]= 0;

  582.                 left_block[3]= 2;

  583.                 //left_block[4]= 7;

  584.                 left_block[5]= 10;

  585.                 left_block[6]= 7;

  586.                 left_block[7]= 10;

  587.             }

  588.         }

  589.     }

  590. 

  591.     h->top_mb_xy = top_xy;

  592.     h->left_mb_xy[0] = left_xy[0];

  593.     h->left_mb_xy[1] = left_xy[1];

  594.     if(for_deblock){

  595.         topleft_type = 0;

  596.         topright_type = 0;

  597.         top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;

  598.         left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;

  599.         left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;

  600. 

  601.         if(FRAME_MBAFF && !IS_INTRA(mb_type)){

  602.             int list;

  603.             int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];

  604.             for(i=0; i<16; i++)

  605.                 h->non_zero_count_cache[scan8[i]] = (v>>i)&1;

  606.             for(list=0; list<1+(h->slice_type==B_TYPE); list++){

  607.                 if(USES_LIST(mb_type,list)){

  608.                     uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];

  609.                     uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];

  610.                     int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];

  611.                     for(i=0; i<4; i++, dst+=8, src+=h->b_stride){

  612.                         dst[0] = src[0];

  613.                         dst[1] = src[1];

  614.                         dst[2] = src[2];

  615.                         dst[3] = src[3];

  616.                     }

  617.                     *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =

  618.                     *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;

  619.                     ref += h->b8_stride;

  620.                     *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =

  621.                     *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;

  622.                 }else{

  623.                     fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);

  624.                     fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);

  625.                 }

  626.             }

  627.         }

  628.     }else{

  629.         topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;

  630.         top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;

  631.         topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;

  632.         left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;

  633.         left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;

  634.     }

  635. 

  636.     if(IS_INTRA(mb_type)){

  637.         h->topleft_samples_available=

  638.         h->top_samples_available=

  639.         h->left_samples_available= 0xFFFF;

  640.         h->topright_samples_available= 0xEEEA;

  641. 

  642.         if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){

  643.             h->topleft_samples_available= 0xB3FF;

  644.             h->top_samples_available= 0x33FF;

  645.             h->topright_samples_available= 0x26EA;

  646.         }

  647.         for(i=0; i<2; i++){

  648.             if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){

  649.                 h->topleft_samples_available&= 0xDF5F;

  650.                 h->left_samples_available&= 0x5F5F;

  651.             }

  652.         }

  653. 

  654.         if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))

  655.             h->topleft_samples_available&= 0x7FFF;

  656. 

  657.         if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))

  658.             h->topright_samples_available&= 0xFBFF;

  659. 

  660.         if(IS_INTRA4x4(mb_type)){

  661.             if(IS_INTRA4x4(top_type)){

  662.                 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];

  663.                 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];

  664.                 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];

  665.                 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];

  666.             }else{

  667.                 int pred;

  668.                 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))

  669.                     pred= -1;

  670.                 else{

  671.                     pred= 2;

  672.                 }

  673.                 h->intra4x4_pred_mode_cache[4+8*0]=

  674.                 h->intra4x4_pred_mode_cache[5+8*0]=

  675.                 h->intra4x4_pred_mode_cache[6+8*0]=

  676.                 h->intra4x4_pred_mode_cache[7+8*0]= pred;

  677.             }

  678.             for(i=0; i<2; i++){

  679.                 if(IS_INTRA4x4(left_type[i])){

  680.                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];

  681.                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];

  682.                 }else{

  683.                     int pred;

  684.                     if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))

  685.                         pred= -1;

  686.                     else{

  687.                         pred= 2;

  688.                     }

  689.                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=

  690.                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;

  691.                 }

  692.             }

  693.         }

  694.     }

  695. 

  696. 

  697. /*

  698. 0 . T T. T T T T

  699. 1 L . .L . . . .

  700. 2 L . .L . . . .

  701. 3 . T TL . . . .

  702. 4 L . .L . . . .

  703. 5 L . .. . . . .

  704. */

  705. //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)

  706.     if(top_type){

  707.         h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];

  708.         h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];

  709.         h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];

  710.         h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];

  711. 

  712.         h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];

  713.         h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];

  714. 

  715.         h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];

  716.         h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];

  717. 

  718.     }else{

  719.         h->non_zero_count_cache[4+8*0]=

  720.         h->non_zero_count_cache[5+8*0]=

  721.         h->non_zero_count_cache[6+8*0]=

  722.         h->non_zero_count_cache[7+8*0]=

  723. 

  724.         h->non_zero_count_cache[1+8*0]=

  725.         h->non_zero_count_cache[2+8*0]=

  726. 

  727.         h->non_zero_count_cache[1+8*3]=

  728.         h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;

  729. 

  730.     }

  731. 

  732.     for (i=0; i<2; i++) {

  733.         if(left_type[i]){

  734.             h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];

  735.             h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];

  736.             h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];

  737.             h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];

  738.         }else{

  739.             h->non_zero_count_cache[3+8*1 + 2*8*i]=

  740.             h->non_zero_count_cache[3+8*2 + 2*8*i]=

  741.             h->non_zero_count_cache[0+8*1 +   8*i]=

  742.             h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;

  743.         }

  744.     }

  745. 

  746.     if( h->pps.cabac ) {

  747.         // top_cbp

  748.         if(top_type) {

  749.             h->top_cbp = h->cbp_table[top_xy];

  750.         } else if(IS_INTRA(mb_type)) {

  751.             h->top_cbp = 0x1C0;

  752.         } else {

  753.             h->top_cbp = 0;

  754.         }

  755.         // left_cbp

  756.         if (left_type[0]) {

  757.             h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;

  758.         } else if(IS_INTRA(mb_type)) {

  759.             h->left_cbp = 0x1C0;

  760.         } else {

  761.             h->left_cbp = 0;

  762.         }

  763.         if (left_type[0]) {

  764.             h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;

  765.         }

  766.         if (left_type[1]) {

  767.             h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;

  768.         }

  769.     }

  770. 

  771. #if 1

  772.     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){

  773.         int list;

  774.         for(list=0; list<1+(h->slice_type==B_TYPE); list++){

  775.             if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){

  776.                 /*if(!h->mv_cache_clean[list]){

  777.                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?

  778.                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));

  779.                     h->mv_cache_clean[list]= 1;

  780.                 }*/

  781.                 continue;

  782.             }

  783.             h->mv_cache_clean[list]= 0;

  784. 

  785.             if(USES_LIST(top_type, list)){

  786.                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;

  787.                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;

  788.                 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];

  789.                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];

  790.                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];

  791.                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];

  792.                 h->ref_cache[list][scan8[0] + 0 - 1*8]=

  793.                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];

  794.                 h->ref_cache[list][scan8[0] + 2 - 1*8]=

  795.                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];

  796.             }else{

  797.                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=

  798.                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=

  799.                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=

  800.                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;

  801.                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;

  802.             }

  803. 

  804.             //FIXME unify cleanup or sth

  805.             if(USES_LIST(left_type[0], list)){

  806.                 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;

  807.                 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;

  808.                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]];

  809.                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]];

  810.                 h->ref_cache[list][scan8[0] - 1 + 0*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];

  811.                 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1]>>1)];

  812.             }else{

  813.                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=

  814.                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;

  815.                 h->ref_cache[list][scan8[0] - 1 + 0*8]=

  816.                 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;

  817.             }

  818. 

  819.             if(USES_LIST(left_type[1], list)){

  820.                 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;

  821.                 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;

  822.                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]];

  823.                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]];

  824.                 h->ref_cache[list][scan8[0] - 1 + 2*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];

  825.                 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[3]>>1)];

  826.             }else{

  827.                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=

  828.                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;

  829.                 h->ref_cache[list][scan8[0] - 1 + 2*8]=

  830.                 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;

  831.                 assert((!left_type[0]) == (!left_type[1]));

  832.             }

  833. 

  834.             if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)

  835.                 continue;

  836. 

  837.             if(USES_LIST(topleft_type, list)){

  838.                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;

  839.                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;

  840.                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];

  841.                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];

  842.             }else{

  843.                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;

  844.                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;

  845.             }

  846. 

  847.             if(USES_LIST(topright_type, list)){

  848.                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;

  849.                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;

  850.                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];

  851.                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];

  852.             }else{

  853.                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;

  854.                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;

  855.             }

  856. 

  857.             if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)

  858.                 continue;

  859. 

  860.             h->ref_cache[list][scan8[5 ]+1] =

  861.             h->ref_cache[list][scan8[7 ]+1] =

  862.             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)

  863.             h->ref_cache[list][scan8[4 ]] =

  864.             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;

  865.             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=

  866.             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=

  867.             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)

  868.             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=

  869.             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;

  870. 

  871.             if( h->pps.cabac ) {

  872.                 /* XXX beurk, Load mvd */

  873.                 if(USES_LIST(top_type, list)){

  874.                     const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;

  875.                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];

  876.                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];

  877.                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];

  878.                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];

  879.                 }else{

  880.                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=

  881.                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=

  882.                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=

  883.                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;

  884.                 }

  885.                 if(USES_LIST(left_type[0], list)){

  886.                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;

  887.                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];

  888.                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];

  889.                 }else{

  890.                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=

  891.                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;

  892.                 }

  893.                 if(USES_LIST(left_type[1], list)){

  894.                     const int b_xy= h->mb2b_xy[left_xy[1]] + 3;

  895.                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];

  896.                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];

  897.                 }else{

  898.                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=

  899.                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;

  900.                 }

  901.                 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=

  902.                 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=

  903.                 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)

  904.                 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=

  905.                 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;

  906. 

  907.                 if(h->slice_type == B_TYPE){

  908.                     fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);

  909. 

  910.                     if(IS_DIRECT(top_type)){

  911.                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;

  912.                     }else if(IS_8X8(top_type)){

  913.                         int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;

  914.                         h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];

  915.                         h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];

  916.                     }else{

  917.                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;

  918.                     }

  919. 

  920.                     if(IS_DIRECT(left_type[0]))

  921.                         h->direct_cache[scan8[0] - 1 + 0*8]= 1;

  922.                     else if(IS_8X8(left_type[0]))

  923.                         h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];

  924.                     else

  925.                         h->direct_cache[scan8[0] - 1 + 0*8]= 0;

  926. 

  927.                     if(IS_DIRECT(left_type[1]))

  928.                         h->direct_cache[scan8[0] - 1 + 2*8]= 1;

  929.                     else if(IS_8X8(left_type[1]))

  930.                         h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];

  931.                     else

  932.                         h->direct_cache[scan8[0] - 1 + 2*8]= 0;

  933.                 }

  934.             }

  935. 

  936.             if(FRAME_MBAFF){

  937. #define MAP_MVS\

  938.                     MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\

  939.                     MAP_F2F(scan8[0] + 0 - 1*8, top_type)\

  940.                     MAP_F2F(scan8[0] + 1 - 1*8, top_type)\

  941.                     MAP_F2F(scan8[0] + 2 - 1*8, top_type)\

  942.                     MAP_F2F(scan8[0] + 3 - 1*8, top_type)\

  943.                     MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\

  944.                     MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\

  945.                     MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\

  946.                     MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\

  947.                     MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])

  948.                 if(MB_FIELD){

  949. #define MAP_F2F(idx, mb_type)\

  950.                     if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\

  951.                         h->ref_cache[list][idx] <<= 1;\

  952.                         h->mv_cache[list][idx][1] /= 2;\

  953.                         h->mvd_cache[list][idx][1] /= 2;\

  954.                     }

  955.                     MAP_MVS

  956. #undef MAP_F2F

  957.                 }else{

  958. #define MAP_F2F(idx, mb_type)\

  959.                     if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\

  960.                         h->ref_cache[list][idx] >>= 1;\

  961.                         h->mv_cache[list][idx][1] <<= 1;\

  962.                         h->mvd_cache[list][idx][1] <<= 1;\

  963.                     }

  964.                     MAP_MVS

  965. #undef MAP_F2F

  966.                 }

  967.             }

  968.         }

  969.     }

  970. #endif

  971. 

  972.     h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);

  973. }

  974. 

  975. static inline void write_back_intra_pred_mode(H264Context *h){

  976.     MpegEncContext * const s = &h->s;

  977.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  978. 

  979.     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];

  980.     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];

  981.     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];

  982.     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];

  983.     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];

  984.     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];

  985.     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];

  986. }

  987. 

  988. /**

  989.  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.

  990.  */

  991. static inline int check_intra4x4_pred_mode(H264Context *h){

  992.     MpegEncContext * const s = &h->s;

  993.     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};

  994.     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};

  995.     int i;

  996. 

  997.     if(!(h->top_samples_available&0x8000)){

  998.         for(i=0; i<4; i++){

  999.             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];

  1000.             if(status<0){

  1001.                 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);

  1002.                 return -1;

  1003.             } else if(status){

  1004.                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;

  1005.             }

  1006.         }

  1007.     }

  1008. 

  1009.     if(!(h->left_samples_available&0x8000)){

  1010.         for(i=0; i<4; i++){

  1011.             int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];

  1012.             if(status<0){

  1013.                 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);

  1014.                 return -1;

  1015.             } else if(status){

  1016.                 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;

  1017.             }

  1018.         }

  1019.     }

  1020. 

  1021.     return 0;

  1022. } //FIXME cleanup like next

  1023. 

  1024. /**

  1025.  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.

  1026.  */

  1027. static inline int check_intra_pred_mode(H264Context *h, int mode){

  1028.     MpegEncContext * const s = &h->s;

  1029.     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};

  1030.     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};

  1031. 

  1032.     if(mode < 0 || mode > 6) {

  1033.         av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);

  1034.         return -1;

  1035.     }

  1036. 

  1037.     if(!(h->top_samples_available&0x8000)){

  1038.         mode= top[ mode ];

  1039.         if(mode<0){

  1040.             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);

  1041.             return -1;

  1042.         }

  1043.     }

  1044. 

  1045.     if(!(h->left_samples_available&0x8000)){

  1046.         mode= left[ mode ];

  1047.         if(mode<0){

  1048.             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);

  1049.             return -1;

  1050.         }

  1051.     }

  1052. 

  1053.     return mode;

  1054. }

  1055. 

  1056. /**

  1057.  * gets the predicted intra4x4 prediction mode.

  1058.  */

  1059. static inline int pred_intra_mode(H264Context *h, int n){

  1060.     const int index8= scan8[n];

  1061.     const int left= h->intra4x4_pred_mode_cache[index8 - 1];

  1062.     const int top = h->intra4x4_pred_mode_cache[index8 - 8];

  1063.     const int min= FFMIN(left, top);

  1064. 

  1065.     tprintf("mode:%d %d min:%d\n", left ,top, min);

  1066. 

  1067.     if(min<0) return DC_PRED;

  1068.     else      return min;

  1069. }

  1070. 

  1071. static inline void write_back_non_zero_count(H264Context *h){

  1072.     MpegEncContext * const s = &h->s;

  1073.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  1074. 

  1075.     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];

  1076.     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];

  1077.     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];

  1078.     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];

  1079.     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];

  1080.     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];

  1081.     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];

  1082. 

  1083.     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];

  1084.     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];

  1085.     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];

  1086. 

  1087.     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];

  1088.     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];

  1089.     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];

  1090. 

  1091.     if(FRAME_MBAFF){

  1092.         // store all luma nnzs, for deblocking

  1093.         int v = 0, i;

  1094.         for(i=0; i<16; i++)

  1095.             v += (!!h->non_zero_count_cache[scan8[i]]) << i;

  1096.         *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;

  1097.     }

  1098. }

  1099. 

  1100. /**

  1101.  * gets the predicted number of non zero coefficients.

  1102.  * @param n block index

  1103.  */

  1104. static inline int pred_non_zero_count(H264Context *h, int n){

  1105.     const int index8= scan8[n];

  1106.     const int left= h->non_zero_count_cache[index8 - 1];

  1107.     const int top = h->non_zero_count_cache[index8 - 8];

  1108.     int i= left + top;

  1109. 

  1110.     if(i<64) i= (i+1)>>1;

  1111. 

  1112.     tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);

  1113. 

  1114.     return i&31;

  1115. }

  1116. 

  1117. static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){

  1118.     const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];

  1119. 

  1120.     /* there is no consistent mapping of mvs to neighboring locations that will

  1121.      * make mbaff happy, so we can't move all this logic to fill_caches */

  1122.     if(FRAME_MBAFF){

  1123.         MpegEncContext *s = &h->s;

  1124.         const uint32_t *mb_types = s->current_picture_ptr->mb_type;

  1125.         const int16_t *mv;

  1126.         *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;

  1127.         *C = h->mv_cache[list][scan8[0]-2];

  1128. 

  1129.         if(!MB_FIELD

  1130.            && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){

  1131.             int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);

  1132.             if(IS_INTERLACED(mb_types[topright_xy])){

  1133. #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\

  1134.                 const int x4 = X4, y4 = Y4;\

  1135.                 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\

  1136.                 if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\

  1137.                     return LIST_NOT_USED;\

  1138.                 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\

  1139.                 h->mv_cache[list][scan8[0]-2][0] = mv[0];\

  1140.                 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\

  1141.                 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;

  1142. 

  1143.                 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);

  1144.             }

  1145.         }

  1146.         if(topright_ref == PART_NOT_AVAILABLE

  1147.            && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4

  1148.            && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){

  1149.             if(!MB_FIELD

  1150.                && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){

  1151.                 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);

  1152.             }

  1153.             if(MB_FIELD

  1154.                && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])

  1155.                && i >= scan8[0]+8){

  1156.                 // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.

  1157.                 SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);

  1158.             }

  1159.         }

  1160. #undef SET_DIAG_MV

  1161.     }

  1162. 

  1163.     if(topright_ref != PART_NOT_AVAILABLE){

  1164.         *C= h->mv_cache[list][ i - 8 + part_width ];

  1165.         return topright_ref;

  1166.     }else{

  1167.         tprintf("topright MV not available\n");

  1168. 

  1169.         *C= h->mv_cache[list][ i - 8 - 1 ];

  1170.         return h->ref_cache[list][ i - 8 - 1 ];

  1171.     }

  1172. }

  1173. 

  1174. /**

  1175.  * gets the predicted MV.

  1176.  * @param n the block index

  1177.  * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)

  1178.  * @param mx the x component of the predicted motion vector

  1179.  * @param my the y component of the predicted motion vector

  1180.  */

  1181. static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){

  1182.     const int index8= scan8[n];

  1183.     const int top_ref=      h->ref_cache[list][ index8 - 8 ];

  1184.     const int left_ref=     h->ref_cache[list][ index8 - 1 ];

  1185.     const int16_t * const A= h->mv_cache[list][ index8 - 1 ];

  1186.     const int16_t * const B= h->mv_cache[list][ index8 - 8 ];

  1187.     const int16_t * C;

  1188.     int diagonal_ref, match_count;

  1189. 

  1190.     assert(part_width==1 || part_width==2 || part_width==4);

  1191. 

  1192. /* mv_cache

  1193.   B . . A T T T T

  1194.   U . . L . . , .

  1195.   U . . L . . . .

  1196.   U . . L . . , .

  1197.   . . . L . . . .

  1198. */

  1199. 

  1200.     diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);

  1201.     match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);

  1202.     tprintf("pred_motion match_count=%d\n", match_count);

  1203.     if(match_count > 1){ //most common

  1204.         *mx= mid_pred(A[0], B[0], C[0]);

  1205.         *my= mid_pred(A[1], B[1], C[1]);

  1206.     }else if(match_count==1){

  1207.         if(left_ref==ref){

  1208.             *mx= A[0];

  1209.             *my= A[1];

  1210.         }else if(top_ref==ref){

  1211.             *mx= B[0];

  1212.             *my= B[1];

  1213.         }else{

  1214.             *mx= C[0];

  1215.             *my= C[1];

  1216.         }

  1217.     }else{

  1218.         if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){

  1219.             *mx= A[0];

  1220.             *my= A[1];

  1221.         }else{

  1222.             *mx= mid_pred(A[0], B[0], C[0]);

  1223.             *my= mid_pred(A[1], B[1], C[1]);

  1224.         }

  1225.     }

  1226. 

  1227.     tprintf("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);

  1228. }

  1229. 

  1230. /**

  1231.  * gets the directionally predicted 16x8 MV.

  1232.  * @param n the block index

  1233.  * @param mx the x component of the predicted motion vector

  1234.  * @param my the y component of the predicted motion vector

  1235.  */

  1236. static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){

  1237.     if(n==0){

  1238.         const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];

  1239.         const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];

  1240. 

  1241.         tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);

  1242. 

  1243.         if(top_ref == ref){

  1244.             *mx= B[0];

  1245.             *my= B[1];

  1246.             return;

  1247.         }

  1248.     }else{

  1249.         const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];

  1250.         const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];

  1251. 

  1252.         tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);

  1253. 

  1254.         if(left_ref == ref){

  1255.             *mx= A[0];

  1256.             *my= A[1];

  1257.             return;

  1258.         }

  1259.     }

  1260. 

  1261.     //RARE

  1262.     pred_motion(h, n, 4, list, ref, mx, my);

  1263. }

  1264. 

  1265. /**

  1266.  * gets the directionally predicted 8x16 MV.

  1267.  * @param n the block index

  1268.  * @param mx the x component of the predicted motion vector

  1269.  * @param my the y component of the predicted motion vector

  1270.  */

  1271. static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){

  1272.     if(n==0){

  1273.         const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];

  1274.         const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];

  1275. 

  1276.         tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);

  1277. 

  1278.         if(left_ref == ref){

  1279.             *mx= A[0];

  1280.             *my= A[1];

  1281.             return;

  1282.         }

  1283.     }else{

  1284.         const int16_t * C;

  1285.         int diagonal_ref;

  1286. 

  1287.         diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);

  1288. 

  1289.         tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);

  1290. 

  1291.         if(diagonal_ref == ref){

  1292.             *mx= C[0];

  1293.             *my= C[1];

  1294.             return;

  1295.         }

  1296.     }

  1297. 

  1298.     //RARE

  1299.     pred_motion(h, n, 2, list, ref, mx, my);

  1300. }

  1301. 

  1302. static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){

  1303.     const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];

  1304.     const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];

  1305. 

  1306.     tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);

  1307. 

  1308.     if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE

  1309.        || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)

  1310.        || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){

  1311. 

  1312.         *mx = *my = 0;

  1313.         return;

  1314.     }

  1315. 

  1316.     pred_motion(h, 0, 4, 0, 0, mx, my);

  1317. 

  1318.     return;

  1319. }

  1320. 

  1321. static inline void direct_dist_scale_factor(H264Context * const h){

  1322.     const int poc = h->s.current_picture_ptr->poc;

  1323.     const int poc1 = h->ref_list[1][0].poc;

  1324.     int i;

  1325.     for(i=0; i<h->ref_count[0]; i++){

  1326.         int poc0 = h->ref_list[0][i].poc;

  1327.         int td = clip(poc1 - poc0, -128, 127);

  1328.         if(td == 0 /* FIXME || pic0 is a long-term ref */){

  1329.             h->dist_scale_factor[i] = 256;

  1330.         }else{

  1331.             int tb = clip(poc - poc0, -128, 127);

  1332.             int tx = (16384 + (FFABS(td) >> 1)) / td;

  1333.             h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);

  1334.         }

  1335.     }

  1336.     if(FRAME_MBAFF){

  1337.         for(i=0; i<h->ref_count[0]; i++){

  1338.             h->dist_scale_factor_field[2*i] =

  1339.             h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];

  1340.         }

  1341.     }

  1342. }

  1343. static inline void direct_ref_list_init(H264Context * const h){

  1344.     MpegEncContext * const s = &h->s;

  1345.     Picture * const ref1 = &h->ref_list[1][0];

  1346.     Picture * const cur = s->current_picture_ptr;

  1347.     int list, i, j;

  1348.     if(cur->pict_type == I_TYPE)

  1349.         cur->ref_count[0] = 0;

  1350.     if(cur->pict_type != B_TYPE)

  1351.         cur->ref_count[1] = 0;

  1352.     for(list=0; list<2; list++){

  1353.         cur->ref_count[list] = h->ref_count[list];

  1354.         for(j=0; j<h->ref_count[list]; j++)

  1355.             cur->ref_poc[list][j] = h->ref_list[list][j].poc;

  1356.     }

  1357.     if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)

  1358.         return;

  1359.     for(list=0; list<2; list++){

  1360.         for(i=0; i<ref1->ref_count[list]; i++){

  1361.             const int poc = ref1->ref_poc[list][i];

  1362.             h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */

  1363.             for(j=0; j<h->ref_count[list]; j++)

  1364.                 if(h->ref_list[list][j].poc == poc){

  1365.                     h->map_col_to_list0[list][i] = j;

  1366.                     break;

  1367.                 }

  1368.         }

  1369.     }

  1370.     if(FRAME_MBAFF){

  1371.         for(list=0; list<2; list++){

  1372.             for(i=0; i<ref1->ref_count[list]; i++){

  1373.                 j = h->map_col_to_list0[list][i];

  1374.                 h->map_col_to_list0_field[list][2*i] = 2*j;

  1375.                 h->map_col_to_list0_field[list][2*i+1] = 2*j+1;

  1376.             }

  1377.         }

  1378.     }

  1379. }

  1380. 

  1381. static inline void pred_direct_motion(H264Context * const h, int *mb_type){

  1382.     MpegEncContext * const s = &h->s;

  1383.     const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;

  1384.     const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;

  1385.     const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;

  1386.     const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];

  1387.     const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];

  1388.     const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];

  1389.     const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];

  1390.     const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];

  1391.     const int is_b8x8 = IS_8X8(*mb_type);

  1392.     int sub_mb_type;

  1393.     int i8, i4;

  1394. 

  1395. #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)

  1396.     if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){

  1397.         /* FIXME save sub mb types from previous frames (or derive from MVs)

  1398.          * so we know exactly what block size to use */

  1399.         sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */

  1400.         *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;

  1401.     }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){

  1402.         sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  1403.         *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */

  1404.     }else{

  1405.         sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  1406.         *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;

  1407.     }

  1408.     if(!is_b8x8)

  1409.         *mb_type |= MB_TYPE_DIRECT2;

  1410.     if(MB_FIELD)

  1411.         *mb_type |= MB_TYPE_INTERLACED;

  1412. 

  1413.     tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);

  1414. 

  1415.     if(h->direct_spatial_mv_pred){

  1416.         int ref[2];

  1417.         int mv[2][2];

  1418.         int list;

  1419. 

  1420.         /* FIXME interlacing + spatial direct uses wrong colocated block positions */

  1421. 

  1422.         /* ref = min(neighbors) */

  1423.         for(list=0; list<2; list++){

  1424.             int refa = h->ref_cache[list][scan8[0] - 1];

  1425.             int refb = h->ref_cache[list][scan8[0] - 8];

  1426.             int refc = h->ref_cache[list][scan8[0] - 8 + 4];

  1427.             if(refc == -2)

  1428.                 refc = h->ref_cache[list][scan8[0] - 8 - 1];

  1429.             ref[list] = refa;

  1430.             if(ref[list] < 0 || (refb < ref[list] && refb >= 0))

  1431.                 ref[list] = refb;

  1432.             if(ref[list] < 0 || (refc < ref[list] && refc >= 0))

  1433.                 ref[list] = refc;

  1434.             if(ref[list] < 0)

  1435.                 ref[list] = -1;

  1436.         }

  1437. 

  1438.         if(ref[0] < 0 && ref[1] < 0){

  1439.             ref[0] = ref[1] = 0;

  1440.             mv[0][0] = mv[0][1] =

  1441.             mv[1][0] = mv[1][1] = 0;

  1442.         }else{

  1443.             for(list=0; list<2; list++){

  1444.                 if(ref[list] >= 0)

  1445.                     pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);

  1446.                 else

  1447.                     mv[list][0] = mv[list][1] = 0;

  1448.             }

  1449.         }

  1450. 

  1451.         if(ref[1] < 0){

  1452.             *mb_type &= ~MB_TYPE_P0L1;

  1453.             sub_mb_type &= ~MB_TYPE_P0L1;

  1454.         }else if(ref[0] < 0){

  1455.             *mb_type &= ~MB_TYPE_P0L0;

  1456.             sub_mb_type &= ~MB_TYPE_P0L0;

  1457.         }

  1458. 

  1459.         if(IS_16X16(*mb_type)){

  1460.             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);

  1461.             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);

  1462.             if(!IS_INTRA(mb_type_col)

  1463.                && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)

  1464.                    || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1

  1465.                        && (h->x264_build>33 || !h->x264_build)))){

  1466.                 if(ref[0] > 0)

  1467.                     fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);

  1468.                 else

  1469.                     fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);

  1470.                 if(ref[1] > 0)

  1471.                     fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);

  1472.                 else

  1473.                     fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);

  1474.             }else{

  1475.                 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);

  1476.                 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);

  1477.             }

  1478.         }else{

  1479.             for(i8=0; i8<4; i8++){

  1480.                 const int x8 = i8&1;

  1481.                 const int y8 = i8>>1;

  1482. 

  1483.                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  1484.                     continue;

  1485.                 h->sub_mb_type[i8] = sub_mb_type;

  1486. 

  1487.                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);

  1488.                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);

  1489.                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);

  1490.                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);

  1491. 

  1492.                 /* col_zero_flag */

  1493.                 if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0

  1494.                                               || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0

  1495.                                                   && (h->x264_build>33 || !h->x264_build)))){

  1496.                     const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;

  1497.                     if(IS_SUB_8X8(sub_mb_type)){

  1498.                         const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];

  1499.                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){

  1500.                             if(ref[0] == 0)

  1501.                                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

  1502.                             if(ref[1] == 0)

  1503.                                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);

  1504.                         }

  1505.                     }else

  1506.                     for(i4=0; i4<4; i4++){

  1507.                         const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];

  1508.                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){

  1509.                             if(ref[0] == 0)

  1510.                                 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;

  1511.                             if(ref[1] == 0)

  1512.                                 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;

  1513.                         }

  1514.                     }

  1515.                 }

  1516.             }

  1517.         }

  1518.     }else{ /* direct temporal mv pred */

  1519.         const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};

  1520.         const int *dist_scale_factor = h->dist_scale_factor;

  1521. 

  1522.         if(FRAME_MBAFF){

  1523.             if(IS_INTERLACED(*mb_type)){

  1524.                 map_col_to_list0[0] = h->map_col_to_list0_field[0];

  1525.                 map_col_to_list0[1] = h->map_col_to_list0_field[1];

  1526.                 dist_scale_factor = h->dist_scale_factor_field;

  1527.             }

  1528.             if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){

  1529.                 /* FIXME assumes direct_8x8_inference == 1 */

  1530.                 const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;

  1531.                 int mb_types_col[2];

  1532.                 int y_shift;

  1533. 

  1534.                 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1

  1535.                          | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)

  1536.                          | (*mb_type & MB_TYPE_INTERLACED);

  1537.                 sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;

  1538. 

  1539.                 if(IS_INTERLACED(*mb_type)){

  1540.                     /* frame to field scaling */

  1541.                     mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];

  1542.                     mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];

  1543.                     if(s->mb_y&1){

  1544.                         l1ref0 -= 2*h->b8_stride;

  1545.                         l1ref1 -= 2*h->b8_stride;

  1546.                         l1mv0 -= 4*h->b_stride;

  1547.                         l1mv1 -= 4*h->b_stride;

  1548.                     }

  1549.                     y_shift = 0;

  1550. 

  1551.                     if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)

  1552.                        && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)

  1553.                        && !is_b8x8)

  1554.                         *mb_type |= MB_TYPE_16x8;

  1555.                     else

  1556.                         *mb_type |= MB_TYPE_8x8;

  1557.                 }else{

  1558.                     /* field to frame scaling */

  1559.                     /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)

  1560.                      * but in MBAFF, top and bottom POC are equal */

  1561.                     int dy = (s->mb_y&1) ? 1 : 2;

  1562.                     mb_types_col[0] =

  1563.                     mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];

  1564.                     l1ref0 += dy*h->b8_stride;

  1565.                     l1ref1 += dy*h->b8_stride;

  1566.                     l1mv0 += 2*dy*h->b_stride;

  1567.                     l1mv1 += 2*dy*h->b_stride;

  1568.                     y_shift = 2;

  1569. 

  1570.                     if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))

  1571.                        && !is_b8x8)

  1572.                         *mb_type |= MB_TYPE_16x16;

  1573.                     else

  1574.                         *mb_type |= MB_TYPE_8x8;

  1575.                 }

  1576. 

  1577.                 for(i8=0; i8<4; i8++){

  1578.                     const int x8 = i8&1;

  1579.                     const int y8 = i8>>1;

  1580.                     int ref0, scale;

  1581.                     const int16_t (*l1mv)[2]= l1mv0;

  1582. 

  1583.                     if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  1584.                         continue;

  1585.                     h->sub_mb_type[i8] = sub_mb_type;

  1586. 

  1587.                     fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);

  1588.                     if(IS_INTRA(mb_types_col[y8])){

  1589.                         fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);

  1590.                         fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

  1591.                         fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);

  1592.                         continue;

  1593.                     }

  1594. 

  1595.                     ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];

  1596.                     if(ref0 >= 0)

  1597.                         ref0 = map_col_to_list0[0][ref0*2>>y_shift];

  1598.                     else{

  1599.                         ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];

  1600.                         l1mv= l1mv1;

  1601.                     }

  1602.                     scale = dist_scale_factor[ref0];

  1603.                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);

  1604. 

  1605.                     {

  1606.                         const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];

  1607.                         int my_col = (mv_col[1]<<y_shift)/2;

  1608.                         int mx = (scale * mv_col[0] + 128) >> 8;

  1609.                         int my = (scale * my_col + 128) >> 8;

  1610.                         fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);

  1611.                         fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);

  1612.                     }

  1613.                 }

  1614.                 return;

  1615.             }

  1616.         }

  1617. 

  1618.         /* one-to-one mv scaling */

  1619. 

  1620.         if(IS_16X16(*mb_type)){

  1621.             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);

  1622.             if(IS_INTRA(mb_type_col)){

  1623.                 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);

  1624.                 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);

  1625.                 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);

  1626.             }else{

  1627.                 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]

  1628.                                                 : map_col_to_list0[1][l1ref1[0]];

  1629.                 const int scale = dist_scale_factor[ref0];

  1630.                 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];

  1631.                 int mv_l0[2];

  1632.                 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;

  1633.                 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;

  1634.                 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);

  1635.                 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);

  1636.                 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);

  1637.             }

  1638.         }else{

  1639.             for(i8=0; i8<4; i8++){

  1640.                 const int x8 = i8&1;

  1641.                 const int y8 = i8>>1;

  1642.                 int ref0, scale;

  1643.                 const int16_t (*l1mv)[2]= l1mv0;

  1644. 

  1645.                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  1646.                     continue;

  1647.                 h->sub_mb_type[i8] = sub_mb_type;

  1648.                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);

  1649.                 if(IS_INTRA(mb_type_col)){

  1650.                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);

  1651.                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

  1652.                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);

  1653.                     continue;

  1654.                 }

  1655. 

  1656.                 ref0 = l1ref0[x8 + y8*h->b8_stride];

  1657.                 if(ref0 >= 0)

  1658.                     ref0 = map_col_to_list0[0][ref0];

  1659.                 else{

  1660.                     ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];

  1661.                     l1mv= l1mv1;

  1662.                 }

  1663.                 scale = dist_scale_factor[ref0];

  1664. 

  1665.                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);

  1666.                 if(IS_SUB_8X8(sub_mb_type)){

  1667.                     const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];

  1668.                     int mx = (scale * mv_col[0] + 128) >> 8;

  1669.                     int my = (scale * mv_col[1] + 128) >> 8;

  1670.                     fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);

  1671.                     fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);

  1672.                 }else

  1673.                 for(i4=0; i4<4; i4++){

  1674.                     const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];

  1675.                     int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];

  1676.                     mv_l0[0] = (scale * mv_col[0] + 128) >> 8;

  1677.                     mv_l0[1] = (scale * mv_col[1] + 128) >> 8;

  1678.                     *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =

  1679.                         pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);

  1680.                 }

  1681.             }

  1682.         }

  1683.     }

  1684. }

  1685. 

  1686. static inline void write_back_motion(H264Context *h, int mb_type){

  1687.     MpegEncContext * const s = &h->s;

  1688.     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;

  1689.     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;

  1690.     int list;

  1691. 

  1692.     if(!USES_LIST(mb_type, 0))

  1693.         fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);

  1694. 

  1695.     for(list=0; list<2; list++){

  1696.         int y;

  1697.         if(!USES_LIST(mb_type, list))

  1698.             continue;

  1699. 

  1700.         for(y=0; y<4; y++){

  1701.             *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];

  1702.             *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];

  1703.         }

  1704.         if( h->pps.cabac ) {

  1705.             if(IS_SKIP(mb_type))

  1706.                 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);

  1707.             else

  1708.             for(y=0; y<4; y++){

  1709.                 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];

  1710.                 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];

  1711.             }

  1712.         }

  1713. 

  1714.         {

  1715.             int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];

  1716.             ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];

  1717.             ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];

  1718.             ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];

  1719.             ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];

  1720.         }

  1721.     }

  1722. 

  1723.     if(h->slice_type == B_TYPE && h->pps.cabac){

  1724.         if(IS_8X8(mb_type)){

  1725.             uint8_t *direct_table = &h->direct_table[b8_xy];

  1726.             direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;

  1727.             direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;

  1728.             direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;

  1729.         }

  1730.     }

  1731. }

  1732. 

  1733. /**

  1734.  * Decodes a network abstraction layer unit.

  1735.  * @param consumed is the number of bytes used as input

  1736.  * @param length is the length of the array

  1737.  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?

  1738.  * @returns decoded bytes, might be src+1 if no escapes

  1739.  */

  1740. static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){

  1741.     int i, si, di;

  1742.     uint8_t *dst;

  1743. 

  1744. //    src[0]&0x80;                //forbidden bit

  1745.     h->nal_ref_idc= src[0]>>5;

  1746.     h->nal_unit_type= src[0]&0x1F;

  1747. 

  1748.     src++; length--;

  1749. #if 0

  1750.     for(i=0; i<length; i++)

  1751.         printf("%2X ", src[i]);

  1752. #endif

  1753.     for(i=0; i+1<length; i+=2){

  1754.         if(src[i]) continue;

  1755.         if(i>0 && src[i-1]==0) i--;

  1756.         if(i+2<length && src[i+1]==0 && src[i+2]<=3){

  1757.             if(src[i+2]!=3){

  1758.                 /* startcode, so we must be past the end */

  1759.                 length=i;

  1760.             }

  1761.             break;

  1762.         }

  1763.     }

  1764. 

  1765.     if(i>=length-1){ //no escaped 0

  1766.         *dst_length= length;

  1767.         *consumed= length+1; //+1 for the header

  1768.         return src;

  1769.     }

  1770. 

  1771.     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);

  1772.     dst= h->rbsp_buffer;

  1773. 

  1774. //printf("decoding esc\n");

  1775.     si=di=0;

  1776.     while(si<length){

  1777.         //remove escapes (very rare 1:2^22)

  1778.         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){

  1779.             if(src[si+2]==3){ //escape

  1780.                 dst[di++]= 0;

  1781.                 dst[di++]= 0;

  1782.                 si+=3;

  1783.                 continue;

  1784.             }else //next start code

  1785.                 break;

  1786.         }

  1787. 

  1788.         dst[di++]= src[si++];

  1789.     }

  1790. 

  1791.     *dst_length= di;

  1792.     *consumed= si + 1;//+1 for the header

  1793. //FIXME store exact number of bits in the getbitcontext (its needed for decoding)

  1794.     return dst;

  1795. }

  1796. 

  1797. /**

  1798.  * identifies the exact end of the bitstream

  1799.  * @return the length of the trailing, or 0 if damaged

  1800.  */

  1801. static int decode_rbsp_trailing(uint8_t *src){

  1802.     int v= *src;

  1803.     int r;

  1804. 

  1805.     tprintf("rbsp trailing %X\n", v);

  1806. 

  1807.     for(r=1; r<9; r++){

  1808.         if(v&1) return r;

  1809.         v>>=1;

  1810.     }

  1811.     return 0;

  1812. }

  1813. 

  1814. /**

  1815.  * idct tranforms the 16 dc values and dequantize them.

  1816.  * @param qp quantization parameter

  1817.  */

  1818. static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){

  1819. #define stride 16

  1820.     int i;

  1821.     int temp[16]; //FIXME check if this is a good idea

  1822.     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};

  1823.     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

  1824. 

  1825. //memset(block, 64, 2*256);

  1826. //return;

  1827.     for(i=0; i<4; i++){

  1828.         const int offset= y_offset[i];

  1829.         const int z0= block[offset+stride*0] + block[offset+stride*4];

  1830.         const int z1= block[offset+stride*0] - block[offset+stride*4];

  1831.         const int z2= block[offset+stride*1] - block[offset+stride*5];

  1832.         const int z3= block[offset+stride*1] + block[offset+stride*5];

  1833. 

  1834.         temp[4*i+0]= z0+z3;

  1835.         temp[4*i+1]= z1+z2;

  1836.         temp[4*i+2]= z1-z2;

  1837.         temp[4*i+3]= z0-z3;

  1838.     }

  1839. 

  1840.     for(i=0; i<4; i++){

  1841.         const int offset= x_offset[i];

  1842.         const int z0= temp[4*0+i] + temp[4*2+i];

  1843.         const int z1= temp[4*0+i] - temp[4*2+i];

  1844.         const int z2= temp[4*1+i] - temp[4*3+i];

  1845.         const int z3= temp[4*1+i] + temp[4*3+i];

  1846. 

  1847.         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual

  1848.         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));

  1849.         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));

  1850.         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));

  1851.     }

  1852. }

  1853. 

  1854. #if 0

  1855. /**

  1856.  * dct tranforms the 16 dc values.

  1857.  * @param qp quantization parameter ??? FIXME

  1858.  */

  1859. static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){

  1860. //    const int qmul= dequant_coeff[qp][0];

  1861.     int i;

  1862.     int temp[16]; //FIXME check if this is a good idea

  1863.     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};

  1864.     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

  1865. 

  1866.     for(i=0; i<4; i++){

  1867.         const int offset= y_offset[i];

  1868.         const int z0= block[offset+stride*0] + block[offset+stride*4];

  1869.         const int z1= block[offset+stride*0] - block[offset+stride*4];

  1870.         const int z2= block[offset+stride*1] - block[offset+stride*5];

  1871.         const int z3= block[offset+stride*1] + block[offset+stride*5];

  1872. 

  1873.         temp[4*i+0]= z0+z3;

  1874.         temp[4*i+1]= z1+z2;

  1875.         temp[4*i+2]= z1-z2;

  1876.         temp[4*i+3]= z0-z3;

  1877.     }

  1878. 

  1879.     for(i=0; i<4; i++){

  1880.         const int offset= x_offset[i];

  1881.         const int z0= temp[4*0+i] + temp[4*2+i];

  1882.         const int z1= temp[4*0+i] - temp[4*2+i];

  1883.         const int z2= temp[4*1+i] - temp[4*3+i];

  1884.         const int z3= temp[4*1+i] + temp[4*3+i];

  1885. 

  1886.         block[stride*0 +offset]= (z0 + z3)>>1;

  1887.         block[stride*2 +offset]= (z1 + z2)>>1;

  1888.         block[stride*8 +offset]= (z1 - z2)>>1;

  1889.         block[stride*10+offset]= (z0 - z3)>>1;

  1890.     }

  1891. }

  1892. #endif

  1893. 

  1894. #undef xStride

  1895. #undef stride

  1896. 

  1897. static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){

  1898.     const int stride= 16*2;

  1899.     const int xStride= 16;

  1900.     int a,b,c,d,e;

  1901. 

  1902.     a= block[stride*0 + xStride*0];

  1903.     b= block[stride*0 + xStride*1];

  1904.     c= block[stride*1 + xStride*0];

  1905.     d= block[stride*1 + xStride*1];

  1906. 

  1907.     e= a-b;

  1908.     a= a+b;

  1909.     b= c-d;

  1910.     c= c+d;

  1911. 

  1912.     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;

  1913.     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;

  1914.     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;

  1915.     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;

  1916. }

  1917. 

  1918. #if 0

  1919. static void chroma_dc_dct_c(DCTELEM *block){

  1920.     const int stride= 16*2;

  1921.     const int xStride= 16;

  1922.     int a,b,c,d,e;

  1923. 

  1924.     a= block[stride*0 + xStride*0];

  1925.     b= block[stride*0 + xStride*1];

  1926.     c= block[stride*1 + xStride*0];

  1927.     d= block[stride*1 + xStride*1];

  1928. 

  1929.     e= a-b;

  1930.     a= a+b;

  1931.     b= c-d;

  1932.     c= c+d;

  1933. 

  1934.     block[stride*0 + xStride*0]= (a+c);

  1935.     block[stride*0 + xStride*1]= (e+b);

  1936.     block[stride*1 + xStride*0]= (a-c);

  1937.     block[stride*1 + xStride*1]= (e-b);

  1938. }

  1939. #endif

  1940. 

  1941. /**

  1942.  * gets the chroma qp.

  1943.  */

  1944. static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){

  1945. 

  1946.     return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];

  1947. }

  1948. 

  1949. 

  1950. #if 0

  1951. static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){

  1952.     int i;

  1953.     //FIXME try int temp instead of block

  1954. 

  1955.     for(i=0; i<4; i++){

  1956.         const int d0= src1[0 + i*stride] - src2[0 + i*stride];

  1957.         const int d1= src1[1 + i*stride] - src2[1 + i*stride];

  1958.         const int d2= src1[2 + i*stride] - src2[2 + i*stride];

  1959.         const int d3= src1[3 + i*stride] - src2[3 + i*stride];

  1960.         const int z0= d0 + d3;

  1961.         const int z3= d0 - d3;

  1962.         const int z1= d1 + d2;

  1963.         const int z2= d1 - d2;

  1964. 

  1965.         block[0 + 4*i]=   z0 +   z1;

  1966.         block[1 + 4*i]= 2*z3 +   z2;

  1967.         block[2 + 4*i]=   z0 -   z1;

  1968.         block[3 + 4*i]=   z3 - 2*z2;

  1969.     }

  1970. 

  1971.     for(i=0; i<4; i++){

  1972.         const int z0= block[0*4 + i] + block[3*4 + i];

  1973.         const int z3= block[0*4 + i] - block[3*4 + i];

  1974.         const int z1= block[1*4 + i] + block[2*4 + i];

  1975.         const int z2= block[1*4 + i] - block[2*4 + i];

  1976. 

  1977.         block[0*4 + i]=   z0 +   z1;

  1978.         block[1*4 + i]= 2*z3 +   z2;

  1979.         block[2*4 + i]=   z0 -   z1;

  1980.         block[3*4 + i]=   z3 - 2*z2;

  1981.     }

  1982. }

  1983. #endif

  1984. 

  1985. //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close

  1986. //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)

  1987. static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){

  1988.     int i;

  1989.     const int * const quant_table= quant_coeff[qscale];

  1990.     const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;

  1991.     const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;

  1992.     const unsigned int threshold2= (threshold1<<1);

  1993.     int last_non_zero;

  1994. 

  1995.     if(seperate_dc){

  1996.         if(qscale<=18){

  1997.             //avoid overflows

  1998.             const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;

  1999.             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;

  2000.             const unsigned int dc_threshold2= (dc_threshold1<<1);

  2001. 

  2002.             int level= block[0]*quant_coeff[qscale+18][0];

  2003.             if(((unsigned)(level+dc_threshold1))>dc_threshold2){

  2004.                 if(level>0){

  2005.                     level= (dc_bias + level)>>(QUANT_SHIFT-2);

  2006.                     block[0]= level;

  2007.                 }else{

  2008.                     level= (dc_bias - level)>>(QUANT_SHIFT-2);

  2009.                     block[0]= -level;

  2010.                 }

  2011. //                last_non_zero = i;

  2012.             }else{

  2013.                 block[0]=0;

  2014.             }

  2015.         }else{

  2016.             const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;

  2017.             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;

  2018.             const unsigned int dc_threshold2= (dc_threshold1<<1);

  2019. 

  2020.             int level= block[0]*quant_table[0];

  2021.             if(((unsigned)(level+dc_threshold1))>dc_threshold2){

  2022.                 if(level>0){

  2023.                     level= (dc_bias + level)>>(QUANT_SHIFT+1);

  2024.                     block[0]= level;

  2025.                 }else{

  2026.                     level= (dc_bias - level)>>(QUANT_SHIFT+1);

  2027.                     block[0]= -level;

  2028.                 }

  2029. //                last_non_zero = i;

  2030.             }else{

  2031.                 block[0]=0;

  2032.             }

  2033.         }

  2034.         last_non_zero= 0;

  2035.         i=1;

  2036.     }else{

  2037.         last_non_zero= -1;

  2038.         i=0;

  2039.     }

  2040. 

  2041.     for(; i<16; i++){

  2042.         const int j= scantable[i];

  2043.         int level= block[j]*quant_table[j];

  2044. 

  2045. //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))

  2046. //           || bias-level >= (1<<(QMAT_SHIFT - 3))){

  2047.         if(((unsigned)(level+threshold1))>threshold2){

  2048.             if(level>0){

  2049.                 level= (bias + level)>>QUANT_SHIFT;

  2050.                 block[j]= level;

  2051.             }else{

  2052.                 level= (bias - level)>>QUANT_SHIFT;

  2053.                 block[j]= -level;

  2054.             }

  2055.             last_non_zero = i;

  2056.         }else{

  2057.             block[j]=0;

  2058.         }

  2059.     }

  2060. 

  2061.     return last_non_zero;

  2062. }

  2063. 

  2064. static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){

  2065.     const uint32_t a= ((uint32_t*)(src-stride))[0];

  2066.     ((uint32_t*)(src+0*stride))[0]= a;

  2067.     ((uint32_t*)(src+1*stride))[0]= a;

  2068.     ((uint32_t*)(src+2*stride))[0]= a;

  2069.     ((uint32_t*)(src+3*stride))[0]= a;

  2070. }

  2071. 

  2072. static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){

  2073.     ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;

  2074.     ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;

  2075.     ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;

  2076.     ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;

  2077. }

  2078. 

  2079. static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){

  2080.     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]

  2081.                    + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;

  2082. 

  2083.     ((uint32_t*)(src+0*stride))[0]=

  2084.     ((uint32_t*)(src+1*stride))[0]=

  2085.     ((uint32_t*)(src+2*stride))[0]=

  2086.     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;

  2087. }

  2088. 

  2089. static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){

  2090.     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;

  2091. 

  2092.     ((uint32_t*)(src+0*stride))[0]=

  2093.     ((uint32_t*)(src+1*stride))[0]=

  2094.     ((uint32_t*)(src+2*stride))[0]=

  2095.     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;

  2096. }

  2097. 

  2098. static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){

  2099.     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;

  2100. 

  2101.     ((uint32_t*)(src+0*stride))[0]=

  2102.     ((uint32_t*)(src+1*stride))[0]=

  2103.     ((uint32_t*)(src+2*stride))[0]=

  2104.     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;

  2105. }

  2106. 

  2107. static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){

  2108.     ((uint32_t*)(src+0*stride))[0]=

  2109.     ((uint32_t*)(src+1*stride))[0]=

  2110.     ((uint32_t*)(src+2*stride))[0]=

  2111.     ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;

  2112. }

  2113. 

  2114. 

  2115. #define LOAD_TOP_RIGHT_EDGE\

  2116.     const int t4= topright[0];\

  2117.     const int t5= topright[1];\

  2118.     const int t6= topright[2];\

  2119.     const int t7= topright[3];\

  2120. 

  2121. #define LOAD_LEFT_EDGE\

  2122.     const int l0= src[-1+0*stride];\

  2123.     const int l1= src[-1+1*stride];\

  2124.     const int l2= src[-1+2*stride];\

  2125.     const int l3= src[-1+3*stride];\

  2126. 

  2127. #define LOAD_TOP_EDGE\

  2128.     const int t0= src[ 0-1*stride];\

  2129.     const int t1= src[ 1-1*stride];\

  2130.     const int t2= src[ 2-1*stride];\

  2131.     const int t3= src[ 3-1*stride];\

  2132. 

  2133. static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){

  2134.     const int lt= src[-1-1*stride];

  2135.     LOAD_TOP_EDGE

  2136.     LOAD_LEFT_EDGE

  2137. 

  2138.     src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;

  2139.     src[0+2*stride]=

  2140.     src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;

  2141.     src[0+1*stride]=

  2142.     src[1+2*stride]=

  2143.     src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;

  2144.     src[0+0*stride]=

  2145.     src[1+1*stride]=

  2146.     src[2+2*stride]=

  2147.     src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;

  2148.     src[1+0*stride]=

  2149.     src[2+1*stride]=

  2150.     src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;

  2151.     src[2+0*stride]=

  2152.     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;

  2153.     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;

  2154. }

  2155. 

  2156. static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){

  2157.     LOAD_TOP_EDGE

  2158.     LOAD_TOP_RIGHT_EDGE

  2159. //    LOAD_LEFT_EDGE

  2160. 

  2161.     src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;

  2162.     src[1+0*stride]=

  2163.     src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;

  2164.     src[2+0*stride]=

  2165.     src[1+1*stride]=

  2166.     src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;

  2167.     src[3+0*stride]=

  2168.     src[2+1*stride]=

  2169.     src[1+2*stride]=

  2170.     src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;

  2171.     src[3+1*stride]=

  2172.     src[2+2*stride]=

  2173.     src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;

  2174.     src[3+2*stride]=

  2175.     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;

  2176.     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;

  2177. }

  2178. 

  2179. static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){

  2180.     const int lt= src[-1-1*stride];

  2181.     LOAD_TOP_EDGE

  2182.     LOAD_LEFT_EDGE

  2183.     const __attribute__((unused)) int unu= l3;

  2184. 

  2185.     src[0+0*stride]=

  2186.     src[1+2*stride]=(lt + t0 + 1)>>1;

  2187.     src[1+0*stride]=

  2188.     src[2+2*stride]=(t0 + t1 + 1)>>1;

  2189.     src[2+0*stride]=

  2190.     src[3+2*stride]=(t1 + t2 + 1)>>1;

  2191.     src[3+0*stride]=(t2 + t3 + 1)>>1;

  2192.     src[0+1*stride]=

  2193.     src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;

  2194.     src[1+1*stride]=

  2195.     src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;

  2196.     src[2+1*stride]=

  2197.     src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;

  2198.     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;

  2199.     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;

  2200.     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;

  2201. }

  2202. 

  2203. static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){

  2204.     LOAD_TOP_EDGE

  2205.     LOAD_TOP_RIGHT_EDGE

  2206.     const __attribute__((unused)) int unu= t7;

  2207. 

  2208.     src[0+0*stride]=(t0 + t1 + 1)>>1;

  2209.     src[1+0*stride]=

  2210.     src[0+2*stride]=(t1 + t2 + 1)>>1;

  2211.     src[2+0*stride]=

  2212.     src[1+2*stride]=(t2 + t3 + 1)>>1;

  2213.     src[3+0*stride]=

  2214.     src[2+2*stride]=(t3 + t4+ 1)>>1;

  2215.     src[3+2*stride]=(t4 + t5+ 1)>>1;

  2216.     src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;

  2217.     src[1+1*stride]=

  2218.     src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;

  2219.     src[2+1*stride]=

  2220.     src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;

  2221.     src[3+1*stride]=

  2222.     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;

  2223.     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;

  2224. }

  2225. 

  2226. static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){

  2227.     LOAD_LEFT_EDGE

  2228. 

  2229.     src[0+0*stride]=(l0 + l1 + 1)>>1;

  2230.     src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;

  2231.     src[2+0*stride]=

  2232.     src[0+1*stride]=(l1 + l2 + 1)>>1;

  2233.     src[3+0*stride]=

  2234.     src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;

  2235.     src[2+1*stride]=

  2236.     src[0+2*stride]=(l2 + l3 + 1)>>1;

  2237.     src[3+1*stride]=

  2238.     src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;

  2239.     src[3+2*stride]=

  2240.     src[1+3*stride]=

  2241.     src[0+3*stride]=

  2242.     src[2+2*stride]=

  2243.     src[2+3*stride]=

  2244.     src[3+3*stride]=l3;

  2245. }

  2246. 

  2247. static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){

  2248.     const int lt= src[-1-1*stride];

  2249.     LOAD_TOP_EDGE

  2250.     LOAD_LEFT_EDGE

  2251.     const __attribute__((unused)) int unu= t3;

  2252. 

  2253.     src[0+0*stride]=

  2254.     src[2+1*stride]=(lt + l0 + 1)>>1;

  2255.     src[1+0*stride]=

  2256.     src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;

  2257.     src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;

  2258.     src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;

  2259.     src[0+1*stride]=

  2260.     src[2+2*stride]=(l0 + l1 + 1)>>1;

  2261.     src[1+1*stride]=

  2262.     src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;

  2263.     src[0+2*stride]=

  2264.     src[2+3*stride]=(l1 + l2+ 1)>>1;

  2265.     src[1+2*stride]=

  2266.     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;

  2267.     src[0+3*stride]=(l2 + l3 + 1)>>1;

  2268.     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;

  2269. }

  2270. 

  2271. void ff_pred16x16_vertical_c(uint8_t *src, int stride){

  2272.     int i;

  2273.     const uint32_t a= ((uint32_t*)(src-stride))[0];

  2274.     const uint32_t b= ((uint32_t*)(src-stride))[1];

  2275.     const uint32_t c= ((uint32_t*)(src-stride))[2];

  2276.     const uint32_t d= ((uint32_t*)(src-stride))[3];

  2277. 

  2278.     for(i=0; i<16; i++){

  2279.         ((uint32_t*)(src+i*stride))[0]= a;

  2280.         ((uint32_t*)(src+i*stride))[1]= b;

  2281.         ((uint32_t*)(src+i*stride))[2]= c;

  2282.         ((uint32_t*)(src+i*stride))[3]= d;

  2283.     }

  2284. }

  2285. 

  2286. void ff_pred16x16_horizontal_c(uint8_t *src, int stride){

  2287.     int i;

  2288. 

  2289.     for(i=0; i<16; i++){

  2290.         ((uint32_t*)(src+i*stride))[0]=

  2291.         ((uint32_t*)(src+i*stride))[1]=

  2292.         ((uint32_t*)(src+i*stride))[2]=

  2293.         ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;

  2294.     }

  2295. }

  2296. 

  2297. void ff_pred16x16_dc_c(uint8_t *src, int stride){

  2298.     int i, dc=0;

  2299. 

  2300.     for(i=0;i<16; i++){

  2301.         dc+= src[-1+i*stride];

  2302.     }

  2303. 

  2304.     for(i=0;i<16; i++){

  2305.         dc+= src[i-stride];

  2306.     }

  2307. 

  2308.     dc= 0x01010101*((dc + 16)>>5);

  2309. 

  2310.     for(i=0; i<16; i++){

  2311.         ((uint32_t*)(src+i*stride))[0]=

  2312.         ((uint32_t*)(src+i*stride))[1]=

  2313.         ((uint32_t*)(src+i*stride))[2]=

  2314.         ((uint32_t*)(src+i*stride))[3]= dc;

  2315.     }

  2316. }

  2317. 

  2318. static void pred16x16_left_dc_c(uint8_t *src, int stride){

  2319.     int i, dc=0;

  2320. 

  2321.     for(i=0;i<16; i++){

  2322.         dc+= src[-1+i*stride];

  2323.     }

  2324. 

  2325.     dc= 0x01010101*((dc + 8)>>4);

  2326. 

  2327.     for(i=0; i<16; i++){

  2328.         ((uint32_t*)(src+i*stride))[0]=

  2329.         ((uint32_t*)(src+i*stride))[1]=

  2330.         ((uint32_t*)(src+i*stride))[2]=

  2331.         ((uint32_t*)(src+i*stride))[3]= dc;

  2332.     }

  2333. }

  2334. 

  2335. static void pred16x16_top_dc_c(uint8_t *src, int stride){

  2336.     int i, dc=0;

  2337. 

  2338.     for(i=0;i<16; i++){

  2339.         dc+= src[i-stride];

  2340.     }

  2341.     dc= 0x01010101*((dc + 8)>>4);

  2342. 

  2343.     for(i=0; i<16; i++){

  2344.         ((uint32_t*)(src+i*stride))[0]=

  2345.         ((uint32_t*)(src+i*stride))[1]=

  2346.         ((uint32_t*)(src+i*stride))[2]=

  2347.         ((uint32_t*)(src+i*stride))[3]= dc;

  2348.     }

  2349. }

  2350. 

  2351. void ff_pred16x16_128_dc_c(uint8_t *src, int stride){

  2352.     int i;

  2353. 

  2354.     for(i=0; i<16; i++){

  2355.         ((uint32_t*)(src+i*stride))[0]=

  2356.         ((uint32_t*)(src+i*stride))[1]=

  2357.         ((uint32_t*)(src+i*stride))[2]=

  2358.         ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;

  2359.     }

  2360. }

  2361. 

  2362. static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){

  2363.   int i, j, k;

  2364.   int a;

  2365.   uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  2366.   const uint8_t * const src0 = src+7-stride;

  2367.   const uint8_t *src1 = src+8*stride-1;

  2368.   const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;

  2369.   int H = src0[1] - src0[-1];

  2370.   int V = src1[0] - src2[ 0];

  2371.   for(k=2; k<=8; ++k) {

  2372.     src1 += stride; src2 -= stride;

  2373.     H += k*(src0[k] - src0[-k]);

  2374.     V += k*(src1[0] - src2[ 0]);

  2375.   }

  2376.   if(svq3){

  2377.     H = ( 5*(H/4) ) / 16;

  2378.     V = ( 5*(V/4) ) / 16;

  2379. 

  2380.     /* required for 100% accuracy */

  2381.     i = H; H = V; V = i;

  2382.   }else{

  2383.     H = ( 5*H+32 ) >> 6;

  2384.     V = ( 5*V+32 ) >> 6;

  2385.   }

  2386. 

  2387.   a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);

  2388.   for(j=16; j>0; --j) {

  2389.     int b = a;

  2390.     a += V;

  2391.     for(i=-16; i<0; i+=4) {

  2392.       src[16+i] = cm[ (b    ) >> 5 ];

  2393.       src[17+i] = cm[ (b+  H) >> 5 ];

  2394.       src[18+i] = cm[ (b+2*H) >> 5 ];

  2395.       src[19+i] = cm[ (b+3*H) >> 5 ];

  2396.       b += 4*H;

  2397.     }

  2398.     src += stride;

  2399.   }

  2400. }

  2401. 

  2402. void ff_pred16x16_plane_c(uint8_t *src, int stride){

  2403.     pred16x16_plane_compat_c(src, stride, 0);

  2404. }

  2405. 

  2406. void ff_pred8x8_vertical_c(uint8_t *src, int stride){

  2407.     int i;

  2408.     const uint32_t a= ((uint32_t*)(src-stride))[0];

  2409.     const uint32_t b= ((uint32_t*)(src-stride))[1];

  2410. 

  2411.     for(i=0; i<8; i++){

  2412.         ((uint32_t*)(src+i*stride))[0]= a;

  2413.         ((uint32_t*)(src+i*stride))[1]= b;

  2414.     }

  2415. }

  2416. 

  2417. void ff_pred8x8_horizontal_c(uint8_t *src, int stride){

  2418.     int i;

  2419. 

  2420.     for(i=0; i<8; i++){

  2421.         ((uint32_t*)(src+i*stride))[0]=

  2422.         ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;

  2423.     }

  2424. }

  2425. 

  2426. void ff_pred8x8_128_dc_c(uint8_t *src, int stride){

  2427.     int i;

  2428. 

  2429.     for(i=0; i<8; i++){

  2430.         ((uint32_t*)(src+i*stride))[0]=

  2431.         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;

  2432.     }

  2433. }

  2434. 

  2435. static void pred8x8_left_dc_c(uint8_t *src, int stride){

  2436.     int i;

  2437.     int dc0, dc2;

  2438. 

  2439.     dc0=dc2=0;

  2440.     for(i=0;i<4; i++){

  2441.         dc0+= src[-1+i*stride];

  2442.         dc2+= src[-1+(i+4)*stride];

  2443.     }

  2444.     dc0= 0x01010101*((dc0 + 2)>>2);

  2445.     dc2= 0x01010101*((dc2 + 2)>>2);

  2446. 

  2447.     for(i=0; i<4; i++){

  2448.         ((uint32_t*)(src+i*stride))[0]=

  2449.         ((uint32_t*)(src+i*stride))[1]= dc0;

  2450.     }

  2451.     for(i=4; i<8; i++){

  2452.         ((uint32_t*)(src+i*stride))[0]=

  2453.         ((uint32_t*)(src+i*stride))[1]= dc2;

  2454.     }

  2455. }

  2456. 

  2457. static void pred8x8_top_dc_c(uint8_t *src, int stride){

  2458.     int i;

  2459.     int dc0, dc1;

  2460. 

  2461.     dc0=dc1=0;

  2462.     for(i=0;i<4; i++){

  2463.         dc0+= src[i-stride];

  2464.         dc1+= src[4+i-stride];

  2465.     }

  2466.     dc0= 0x01010101*((dc0 + 2)>>2);

  2467.     dc1= 0x01010101*((dc1 + 2)>>2);

  2468. 

  2469.     for(i=0; i<4; i++){

  2470.         ((uint32_t*)(src+i*stride))[0]= dc0;

  2471.         ((uint32_t*)(src+i*stride))[1]= dc1;

  2472.     }

  2473.     for(i=4; i<8; i++){

  2474.         ((uint32_t*)(src+i*stride))[0]= dc0;

  2475.         ((uint32_t*)(src+i*stride))[1]= dc1;

  2476.     }

  2477. }

  2478. 

  2479. 

  2480. void ff_pred8x8_dc_c(uint8_t *src, int stride){

  2481.     int i;

  2482.     int dc0, dc1, dc2, dc3;

  2483. 

  2484.     dc0=dc1=dc2=0;

  2485.     for(i=0;i<4; i++){

  2486.         dc0+= src[-1+i*stride] + src[i-stride];

  2487.         dc1+= src[4+i-stride];

  2488.         dc2+= src[-1+(i+4)*stride];

  2489.     }

  2490.     dc3= 0x01010101*((dc1 + dc2 + 4)>>3);

  2491.     dc0= 0x01010101*((dc0 + 4)>>3);

  2492.     dc1= 0x01010101*((dc1 + 2)>>2);

  2493.     dc2= 0x01010101*((dc2 + 2)>>2);

  2494. 

  2495.     for(i=0; i<4; i++){

  2496.         ((uint32_t*)(src+i*stride))[0]= dc0;

  2497.         ((uint32_t*)(src+i*stride))[1]= dc1;

  2498.     }

  2499.     for(i=4; i<8; i++){

  2500.         ((uint32_t*)(src+i*stride))[0]= dc2;

  2501.         ((uint32_t*)(src+i*stride))[1]= dc3;

  2502.     }

  2503. }

  2504. 

  2505. void ff_pred8x8_plane_c(uint8_t *src, int stride){

  2506.   int j, k;

  2507.   int a;

  2508.   uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  2509.   const uint8_t * const src0 = src+3-stride;

  2510.   const uint8_t *src1 = src+4*stride-1;

  2511.   const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;

  2512.   int H = src0[1] - src0[-1];

  2513.   int V = src1[0] - src2[ 0];

  2514.   for(k=2; k<=4; ++k) {

  2515.     src1 += stride; src2 -= stride;

  2516.     H += k*(src0[k] - src0[-k]);

  2517.     V += k*(src1[0] - src2[ 0]);

  2518.   }

  2519.   H = ( 17*H+16 ) >> 5;

  2520.   V = ( 17*V+16 ) >> 5;

  2521. 

  2522.   a = 16*(src1[0] + src2[8]+1) - 3*(V+H);

  2523.   for(j=8; j>0; --j) {

  2524.     int b = a;

  2525.     a += V;

  2526.     src[0] = cm[ (b    ) >> 5 ];

  2527.     src[1] = cm[ (b+  H) >> 5 ];

  2528.     src[2] = cm[ (b+2*H) >> 5 ];

  2529.     src[3] = cm[ (b+3*H) >> 5 ];

  2530.     src[4] = cm[ (b+4*H) >> 5 ];

  2531.     src[5] = cm[ (b+5*H) >> 5 ];

  2532.     src[6] = cm[ (b+6*H) >> 5 ];

  2533.     src[7] = cm[ (b+7*H) >> 5 ];

  2534.     src += stride;

  2535.   }

  2536. }

  2537. 

  2538. #define SRC(x,y) src[(x)+(y)*stride]

  2539. #define PL(y) \

  2540.     const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;

  2541. #define PREDICT_8x8_LOAD_LEFT \

  2542.     const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \

  2543.                      + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \

  2544.     PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \

  2545.     const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2

  2546. 

  2547. #define PT(x) \

  2548.     const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;

  2549. #define PREDICT_8x8_LOAD_TOP \

  2550.     const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \

  2551.                      + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \

  2552.     PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \

  2553.     const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \

  2554.                      + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2

  2555. 

  2556. #define PTR(x) \

  2557.     t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;

  2558. #define PREDICT_8x8_LOAD_TOPRIGHT \

  2559.     int t8, t9, t10, t11, t12, t13, t14, t15; \

  2560.     if(has_topright) { \

  2561.         PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \

  2562.         t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \

  2563.     } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);

  2564. 

  2565. #define PREDICT_8x8_LOAD_TOPLEFT \

  2566.     const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2

  2567. 

  2568. #define PREDICT_8x8_DC(v) \

  2569.     int y; \

  2570.     for( y = 0; y < 8; y++ ) { \

  2571.         ((uint32_t*)src)[0] = \

  2572.         ((uint32_t*)src)[1] = v; \

  2573.         src += stride; \

  2574.     }

  2575. 

  2576. static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2577. {

  2578.     PREDICT_8x8_DC(0x80808080);

  2579. }

  2580. static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2581. {

  2582.     PREDICT_8x8_LOAD_LEFT;

  2583.     const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;

  2584.     PREDICT_8x8_DC(dc);

  2585. }

  2586. static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2587. {

  2588.     PREDICT_8x8_LOAD_TOP;

  2589.     const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;

  2590.     PREDICT_8x8_DC(dc);

  2591. }

  2592. static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2593. {

  2594.     PREDICT_8x8_LOAD_LEFT;

  2595.     PREDICT_8x8_LOAD_TOP;

  2596.     const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7

  2597.                          +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;

  2598.     PREDICT_8x8_DC(dc);

  2599. }

  2600. static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2601. {

  2602.     PREDICT_8x8_LOAD_LEFT;

  2603. #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\

  2604.                ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y

  2605.     ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);

  2606. #undef ROW

  2607. }

  2608. static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2609. {

  2610.     int y;

  2611.     PREDICT_8x8_LOAD_TOP;

  2612.     src[0] = t0;

  2613.     src[1] = t1;

  2614.     src[2] = t2;

  2615.     src[3] = t3;

  2616.     src[4] = t4;

  2617.     src[5] = t5;

  2618.     src[6] = t6;

  2619.     src[7] = t7;

  2620.     for( y = 1; y < 8; y++ )

  2621.         *(uint64_t*)(src+y*stride) = *(uint64_t*)src;

  2622. }

  2623. static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2624. {

  2625.     PREDICT_8x8_LOAD_TOP;

  2626.     PREDICT_8x8_LOAD_TOPRIGHT;

  2627.     SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;

  2628.     SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;

  2629.     SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;

  2630.     SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;

  2631.     SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;

  2632.     SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;

  2633.     SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2;

  2634.     SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2;

  2635.     SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2;

  2636.     SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;

  2637.     SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;

  2638.     SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;

  2639.     SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;

  2640.     SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;

  2641.     SRC(7,7)= (t14 + 3*t15 + 2) >> 2;

  2642. }

  2643. static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2644. {

  2645.     PREDICT_8x8_LOAD_TOP;

  2646.     PREDICT_8x8_LOAD_LEFT;

  2647.     PREDICT_8x8_LOAD_TOPLEFT;

  2648.     SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;

  2649.     SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;

  2650.     SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;

  2651.     SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;

  2652.     SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;

  2653.     SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;

  2654.     SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2;

  2655.     SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2;

  2656.     SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2;

  2657.     SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;

  2658.     SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;

  2659.     SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;

  2660.     SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;

  2661.     SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;

  2662.     SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;

  2663. 

  2664. }

  2665. static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2666. {

  2667.     PREDICT_8x8_LOAD_TOP;

  2668.     PREDICT_8x8_LOAD_LEFT;

  2669.     PREDICT_8x8_LOAD_TOPLEFT;

  2670.     SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;

  2671.     SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;

  2672.     SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;

  2673.     SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;

  2674.     SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;

  2675.     SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;

  2676.     SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;

  2677.     SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;

  2678.     SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;

  2679.     SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;

  2680.     SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;

  2681.     SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;

  2682.     SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;

  2683.     SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;

  2684.     SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;

  2685.     SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;

  2686.     SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;

  2687.     SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;

  2688.     SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;

  2689.     SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;

  2690.     SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;

  2691.     SRC(7,0)= (t6 + t7 + 1) >> 1;

  2692. }

  2693. static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2694. {

  2695.     PREDICT_8x8_LOAD_TOP;

  2696.     PREDICT_8x8_LOAD_LEFT;

  2697.     PREDICT_8x8_LOAD_TOPLEFT;

  2698.     SRC(0,7)= (l6 + l7 + 1) >> 1;

  2699.     SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;

  2700.     SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;

  2701.     SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;

  2702.     SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;

  2703.     SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;

  2704.     SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;

  2705.     SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;

  2706.     SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;

  2707.     SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;

  2708.     SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;

  2709.     SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;

  2710.     SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;

  2711.     SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;

  2712.     SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;

  2713.     SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;

  2714.     SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;

  2715.     SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;

  2716.     SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;

  2717.     SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;

  2718.     SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;

  2719.     SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;

  2720. }

  2721. static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2722. {

  2723.     PREDICT_8x8_LOAD_TOP;

  2724.     PREDICT_8x8_LOAD_TOPRIGHT;

  2725.     SRC(0,0)= (t0 + t1 + 1) >> 1;

  2726.     SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;

  2727.     SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;

  2728.     SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;

  2729.     SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;

  2730.     SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;

  2731.     SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;

  2732.     SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;

  2733.     SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;

  2734.     SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;

  2735.     SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;

  2736.     SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;

  2737.     SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;

  2738.     SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;

  2739.     SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;

  2740.     SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;

  2741.     SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;

  2742.     SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;

  2743.     SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;

  2744.     SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;

  2745.     SRC(7,6)= (t10 + t11 + 1) >> 1;

  2746.     SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;

  2747. }

  2748. static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2749. {

  2750.     PREDICT_8x8_LOAD_LEFT;

  2751.     SRC(0,0)= (l0 + l1 + 1) >> 1;

  2752.     SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;

  2753.     SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;

  2754.     SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;

  2755.     SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;

  2756.     SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;

  2757.     SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;

  2758.     SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;

  2759.     SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;

  2760.     SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;

  2761.     SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;

  2762.     SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;

  2763.     SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;

  2764.     SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;

  2765.     SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=

  2766.     SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=

  2767.     SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=

  2768.     SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;

  2769. }

  2770. #undef PREDICT_8x8_LOAD_LEFT

  2771. #undef PREDICT_8x8_LOAD_TOP

  2772. #undef PREDICT_8x8_LOAD_TOPLEFT

  2773. #undef PREDICT_8x8_LOAD_TOPRIGHT

  2774. #undef PREDICT_8x8_DC

  2775. #undef PTR

  2776. #undef PT

  2777. #undef PL

  2778. #undef SRC

  2779. 

  2780. static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,

  2781.                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  2782.                            int src_x_offset, int src_y_offset,

  2783.                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){

  2784.     MpegEncContext * const s = &h->s;

  2785.     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;

  2786.     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;

  2787.     const int luma_xy= (mx&3) + ((my&3)<<2);

  2788.     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;

  2789.     uint8_t * src_cb, * src_cr;

  2790.     int extra_width= h->emu_edge_width;

  2791.     int extra_height= h->emu_edge_height;

  2792.     int emu=0;

  2793.     const int full_mx= mx>>2;

  2794.     const int full_my= my>>2;

  2795.     const int pic_width  = 16*s->mb_width;

  2796.     const int pic_height = 16*s->mb_height >> MB_MBAFF;

  2797. 

  2798.     if(!pic->data[0])

  2799.         return;

  2800. 

  2801.     if(mx&7) extra_width -= 3;

  2802.     if(my&7) extra_height -= 3;

  2803. 

  2804.     if(   full_mx < 0-extra_width

  2805.        || full_my < 0-extra_height

  2806.        || full_mx + 16/*FIXME*/ > pic_width + extra_width

  2807.        || full_my + 16/*FIXME*/ > pic_height + extra_height){

  2808.         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);

  2809.             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;

  2810.         emu=1;

  2811.     }

  2812. 

  2813.     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?

  2814.     if(!square){

  2815.         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);

  2816.     }

  2817. 

  2818.     if(s->flags&CODEC_FLAG_GRAY) return;

  2819. 

  2820.     if(MB_MBAFF){

  2821.         // chroma offset when predicting from a field of opposite parity

  2822.         my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));

  2823.         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);

  2824.     }

  2825.     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;

  2826.     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;

  2827. 

  2828.     if(emu){

  2829.         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  2830.             src_cb= s->edge_emu_buffer;

  2831.     }

  2832.     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);

  2833. 

  2834.     if(emu){

  2835.         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);

  2836.             src_cr= s->edge_emu_buffer;

  2837.     }

  2838.     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);

  2839. }

  2840. 

  2841. static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,

  2842.                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  2843.                            int x_offset, int y_offset,

  2844.                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,

  2845.                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,

  2846.                            int list0, int list1){

  2847.     MpegEncContext * const s = &h->s;

  2848.     qpel_mc_func *qpix_op=  qpix_put;

  2849.     h264_chroma_mc_func chroma_op= chroma_put;

  2850. 

  2851.     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;

  2852.     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;

  2853.     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;

  2854.     x_offset += 8*s->mb_x;

  2855.     y_offset += 8*(s->mb_y >> MB_MBAFF);

  2856. 

  2857.     if(list0){

  2858.         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];

  2859.         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,

  2860.                            dest_y, dest_cb, dest_cr, x_offset, y_offset,

  2861.                            qpix_op, chroma_op);

  2862. 

  2863.         qpix_op=  qpix_avg;

  2864.         chroma_op= chroma_avg;

  2865.     }

  2866. 

  2867.     if(list1){

  2868.         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];

  2869.         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,

  2870.                            dest_y, dest_cb, dest_cr, x_offset, y_offset,

  2871.                            qpix_op, chroma_op);

  2872.     }

  2873. }

  2874. 

  2875. static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,

  2876.                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  2877.                            int x_offset, int y_offset,

  2878.                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,

  2879.                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,

  2880.                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,

  2881.                            int list0, int list1){

  2882.     MpegEncContext * const s = &h->s;

  2883. 

  2884.     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;

  2885.     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;

  2886.     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;

  2887.     x_offset += 8*s->mb_x;

  2888.     y_offset += 8*(s->mb_y >> MB_MBAFF);

  2889. 

  2890.     if(list0 && list1){

  2891.         /* don't optimize for luma-only case, since B-frames usually

  2892.          * use implicit weights => chroma too. */

  2893.         uint8_t *tmp_cb = s->obmc_scratchpad;

  2894.         uint8_t *tmp_cr = s->obmc_scratchpad + 8;

  2895.         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;

  2896.         int refn0 = h->ref_cache[0][ scan8[n] ];

  2897.         int refn1 = h->ref_cache[1][ scan8[n] ];

  2898. 

  2899.         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,

  2900.                     dest_y, dest_cb, dest_cr,

  2901.                     x_offset, y_offset, qpix_put, chroma_put);

  2902.         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,

  2903.                     tmp_y, tmp_cb, tmp_cr,

  2904.                     x_offset, y_offset, qpix_put, chroma_put);

  2905. 

  2906.         if(h->use_weight == 2){

  2907.             int weight0 = h->implicit_weight[refn0][refn1];

  2908.             int weight1 = 64 - weight0;

  2909.             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);

  2910.             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);

  2911.             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);

  2912.         }else{

  2913.             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,

  2914.                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],

  2915.                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);

  2916.             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,

  2917.                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],

  2918.                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);

  2919.             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,

  2920.                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],

  2921.                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);

  2922.         }

  2923.     }else{

  2924.         int list = list1 ? 1 : 0;

  2925.         int refn = h->ref_cache[list][ scan8[n] ];

  2926.         Picture *ref= &h->ref_list[list][refn];

  2927.         mc_dir_part(h, ref, n, square, chroma_height, delta, list,

  2928.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  2929.                     qpix_put, chroma_put);

  2930. 

  2931.         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,

  2932.                        h->luma_weight[list][refn], h->luma_offset[list][refn]);

  2933.         if(h->use_weight_chroma){

  2934.             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,

  2935.                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);

  2936.             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,

  2937.                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);

  2938.         }

  2939.     }

  2940. }

  2941. 

  2942. static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,

  2943.                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  2944.                            int x_offset, int y_offset,

  2945.                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,

  2946.                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,

  2947.                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,

  2948.                            int list0, int list1){

  2949.     if((h->use_weight==2 && list0 && list1

  2950.         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))

  2951.        || h->use_weight==1)

  2952.         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,

  2953.                          x_offset, y_offset, qpix_put, chroma_put,

  2954.                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);

  2955.     else

  2956.         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,

  2957.                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);

  2958. }

  2959. 

  2960. static inline void prefetch_motion(H264Context *h, int list){

  2961.     /* fetch pixels for estimated mv 4 macroblocks ahead

  2962.      * optimized for 64byte cache lines */

  2963.     MpegEncContext * const s = &h->s;

  2964.     const int refn = h->ref_cache[list][scan8[0]];

  2965.     if(refn >= 0){

  2966.         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;

  2967.         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;

  2968.         uint8_t **src= h->ref_list[list][refn].data;

  2969.         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;

  2970.         s->dsp.prefetch(src[0]+off, s->linesize, 4);

  2971.         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;

  2972.         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);

  2973.     }

  2974. }

  2975. 

  2976. static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  2977.                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),

  2978.                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),

  2979.                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){

  2980.     MpegEncContext * const s = &h->s;

  2981.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  2982.     const int mb_type= s->current_picture.mb_type[mb_xy];

  2983. 

  2984.     assert(IS_INTER(mb_type));

  2985. 

  2986.     prefetch_motion(h, 0);

  2987. 

  2988.     if(IS_16X16(mb_type)){

  2989.         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,

  2990.                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],

  2991.                 &weight_op[0], &weight_avg[0],

  2992.                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));

  2993.     }else if(IS_16X8(mb_type)){

  2994.         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,

  2995.                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],

  2996.                 &weight_op[1], &weight_avg[1],

  2997.                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));

  2998.         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,

  2999.                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],

  3000.                 &weight_op[1], &weight_avg[1],

  3001.                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));

  3002.     }else if(IS_8X16(mb_type)){

  3003.         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,

  3004.                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],

  3005.                 &weight_op[2], &weight_avg[2],

  3006.                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));

  3007.         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,

  3008.                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],

  3009.                 &weight_op[2], &weight_avg[2],

  3010.                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));

  3011.     }else{

  3012.         int i;

  3013. 

  3014.         assert(IS_8X8(mb_type));

  3015. 

  3016.         for(i=0; i<4; i++){

  3017.             const int sub_mb_type= h->sub_mb_type[i];

  3018.             const int n= 4*i;

  3019.             int x_offset= (i&1)<<2;

  3020.             int y_offset= (i&2)<<1;

  3021. 

  3022.             if(IS_SUB_8X8(sub_mb_type)){

  3023.                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,

  3024.                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],

  3025.                     &weight_op[3], &weight_avg[3],

  3026.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3027.             }else if(IS_SUB_8X4(sub_mb_type)){

  3028.                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,

  3029.                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],

  3030.                     &weight_op[4], &weight_avg[4],

  3031.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3032.                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,

  3033.                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],

  3034.                     &weight_op[4], &weight_avg[4],

  3035.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3036.             }else if(IS_SUB_4X8(sub_mb_type)){

  3037.                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,

  3038.                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],

  3039.                     &weight_op[5], &weight_avg[5],

  3040.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3041.                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,

  3042.                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],

  3043.                     &weight_op[5], &weight_avg[5],

  3044.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3045.             }else{

  3046.                 int j;

  3047.                 assert(IS_SUB_4X4(sub_mb_type));

  3048.                 for(j=0; j<4; j++){

  3049.                     int sub_x_offset= x_offset + 2*(j&1);

  3050.                     int sub_y_offset= y_offset +   (j&2);

  3051.                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,

  3052.                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],

  3053.                         &weight_op[6], &weight_avg[6],

  3054.                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3055.                 }

  3056.             }

  3057.         }

  3058.     }

  3059. 

  3060.     prefetch_motion(h, 1);

  3061. }

  3062. 

  3063. static void decode_init_vlc(){

  3064.     static int done = 0;

  3065. 

  3066.     if (!done) {

  3067.         int i;

  3068.         done = 1;

  3069. 

  3070.         init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,

  3071.                  &chroma_dc_coeff_token_len [0], 1, 1,

  3072.                  &chroma_dc_coeff_token_bits[0], 1, 1, 1);

  3073. 

  3074.         for(i=0; i<4; i++){

  3075.             init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,

  3076.                      &coeff_token_len [i][0], 1, 1,

  3077.                      &coeff_token_bits[i][0], 1, 1, 1);

  3078.         }

  3079. 

  3080.         for(i=0; i<3; i++){

  3081.             init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,

  3082.                      &chroma_dc_total_zeros_len [i][0], 1, 1,

  3083.                      &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);

  3084.         }

  3085.         for(i=0; i<15; i++){

  3086.             init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,

  3087.                      &total_zeros_len [i][0], 1, 1,

  3088.                      &total_zeros_bits[i][0], 1, 1, 1);

  3089.         }

  3090. 

  3091.         for(i=0; i<6; i++){

  3092.             init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,

  3093.                      &run_len [i][0], 1, 1,

  3094.                      &run_bits[i][0], 1, 1, 1);

  3095.         }

  3096.         init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,

  3097.                  &run_len [6][0], 1, 1,

  3098.                  &run_bits[6][0], 1, 1, 1);

  3099.     }

  3100. }

  3101. 

  3102. /**

  3103.  * Sets the intra prediction function pointers.

  3104.  */

  3105. static void init_pred_ptrs(H264Context *h){

  3106. //    MpegEncContext * const s = &h->s;

  3107. 

  3108.     h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;

  3109.     h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;

  3110.     h->pred4x4[DC_PRED             ]= pred4x4_dc_c;

  3111.     h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;

  3112.     h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;

  3113.     h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;

  3114.     h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;

  3115.     h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;

  3116.     h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;

  3117.     h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;

  3118.     h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;

  3119.     h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;

  3120. 

  3121.     h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;

  3122.     h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;

  3123.     h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;

  3124.     h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;

  3125.     h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;

  3126.     h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;

  3127.     h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;

  3128.     h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;

  3129.     h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;

  3130.     h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;

  3131.     h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;

  3132.     h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;

  3133. 

  3134.     h->pred8x8[DC_PRED8x8     ]= ff_pred8x8_dc_c;

  3135.     h->pred8x8[VERT_PRED8x8   ]= ff_pred8x8_vertical_c;

  3136.     h->pred8x8[HOR_PRED8x8    ]= ff_pred8x8_horizontal_c;

  3137.     h->pred8x8[PLANE_PRED8x8  ]= ff_pred8x8_plane_c;

  3138.     h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;

  3139.     h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;

  3140.     h->pred8x8[DC_128_PRED8x8 ]= ff_pred8x8_128_dc_c;

  3141. 

  3142.     h->pred16x16[DC_PRED8x8     ]= ff_pred16x16_dc_c;

  3143.     h->pred16x16[VERT_PRED8x8   ]= ff_pred16x16_vertical_c;

  3144.     h->pred16x16[HOR_PRED8x8    ]= ff_pred16x16_horizontal_c;

  3145.     h->pred16x16[PLANE_PRED8x8  ]= ff_pred16x16_plane_c;

  3146.     h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;

  3147.     h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;

  3148.     h->pred16x16[DC_128_PRED8x8 ]= ff_pred16x16_128_dc_c;

  3149. }

  3150. 

  3151. static void free_tables(H264Context *h){

  3152.     av_freep(&h->intra4x4_pred_mode);

  3153.     av_freep(&h->chroma_pred_mode_table);

  3154.     av_freep(&h->cbp_table);

  3155.     av_freep(&h->mvd_table[0]);

  3156.     av_freep(&h->mvd_table[1]);

  3157.     av_freep(&h->direct_table);

  3158.     av_freep(&h->non_zero_count);

  3159.     av_freep(&h->slice_table_base);

  3160.     av_freep(&h->top_borders[1]);

  3161.     av_freep(&h->top_borders[0]);

  3162.     h->slice_table= NULL;

  3163. 

  3164.     av_freep(&h->mb2b_xy);

  3165.     av_freep(&h->mb2b8_xy);

  3166. 

  3167.     av_freep(&h->s.obmc_scratchpad);

  3168. }

  3169. 

  3170. static void init_dequant8_coeff_table(H264Context *h){

  3171.     int i,q,x;

  3172.     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly

  3173.     h->dequant8_coeff[0] = h->dequant8_buffer[0];

  3174.     h->dequant8_coeff[1] = h->dequant8_buffer[1];

  3175. 

  3176.     for(i=0; i<2; i++ ){

  3177.         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){

  3178.             h->dequant8_coeff[1] = h->dequant8_buffer[0];

  3179.             break;

  3180.         }

  3181. 

  3182.         for(q=0; q<52; q++){

  3183.             int shift = div6[q];

  3184.             int idx = rem6[q];

  3185.             for(x=0; x<64; x++)

  3186.                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =

  3187.                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *

  3188.                     h->pps.scaling_matrix8[i][x]) << shift;

  3189.         }

  3190.     }

  3191. }

  3192. 

  3193. static void init_dequant4_coeff_table(H264Context *h){

  3194.     int i,j,q,x;

  3195.     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly

  3196.     for(i=0; i<6; i++ ){

  3197.         h->dequant4_coeff[i] = h->dequant4_buffer[i];

  3198.         for(j=0; j<i; j++){

  3199.             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){

  3200.                 h->dequant4_coeff[i] = h->dequant4_buffer[j];

  3201.                 break;

  3202.             }

  3203.         }

  3204.         if(j<i)

  3205.             continue;

  3206. 

  3207.         for(q=0; q<52; q++){

  3208.             int shift = div6[q] + 2;

  3209.             int idx = rem6[q];

  3210.             for(x=0; x<16; x++)

  3211.                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =

  3212.                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *

  3213.                     h->pps.scaling_matrix4[i][x]) << shift;

  3214.         }

  3215.     }

  3216. }

  3217. 

  3218. static void init_dequant_tables(H264Context *h){

  3219.     int i,x;

  3220.     init_dequant4_coeff_table(h);

  3221.     if(h->pps.transform_8x8_mode)

  3222.         init_dequant8_coeff_table(h);

  3223.     if(h->sps.transform_bypass){

  3224.         for(i=0; i<6; i++)

  3225.             for(x=0; x<16; x++)

  3226.                 h->dequant4_coeff[i][0][x] = 1<<6;

  3227.         if(h->pps.transform_8x8_mode)

  3228.             for(i=0; i<2; i++)

  3229.                 for(x=0; x<64; x++)

  3230.                     h->dequant8_coeff[i][0][x] = 1<<6;

  3231.     }

  3232. }

  3233. 

  3234. 

  3235. /**

  3236.  * allocates tables.

  3237.  * needs width/height

  3238.  */

  3239. static int alloc_tables(H264Context *h){

  3240.     MpegEncContext * const s = &h->s;

  3241.     const int big_mb_num= s->mb_stride * (s->mb_height+1);

  3242.     int x,y;

  3243. 

  3244.     CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))

  3245. 

  3246.     CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))

  3247.     CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))

  3248.     CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))

  3249.     CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))

  3250.     CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))

  3251. 

  3252.     if( h->pps.cabac ) {

  3253.         CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))

  3254.         CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));

  3255.         CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));

  3256.         CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));

  3257.     }

  3258. 

  3259.     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));

  3260.     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;

  3261. 

  3262.     CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));

  3263.     CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));

  3264.     for(y=0; y<s->mb_height; y++){

  3265.         for(x=0; x<s->mb_width; x++){

  3266.             const int mb_xy= x + y*s->mb_stride;

  3267.             const int b_xy = 4*x + 4*y*h->b_stride;

  3268.             const int b8_xy= 2*x + 2*y*h->b8_stride;

  3269. 

  3270.             h->mb2b_xy [mb_xy]= b_xy;

  3271.             h->mb2b8_xy[mb_xy]= b8_xy;

  3272.         }

  3273.     }

  3274. 

  3275.     s->obmc_scratchpad = NULL;

  3276. 

  3277.     if(!h->dequant4_coeff[0])

  3278.         init_dequant_tables(h);

  3279. 

  3280.     return 0;

  3281. fail:

  3282.     free_tables(h);

  3283.     return -1;

  3284. }

  3285. 

  3286. static void common_init(H264Context *h){

  3287.     MpegEncContext * const s = &h->s;

  3288. 

  3289.     s->width = s->avctx->width;

  3290.     s->height = s->avctx->height;

  3291.     s->codec_id= s->avctx->codec->id;

  3292. 

  3293.     init_pred_ptrs(h);

  3294. 

  3295.     h->dequant_coeff_pps= -1;

  3296.     s->unrestricted_mv=1;

  3297.     s->decode=1; //FIXME

  3298. 

  3299.     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));

  3300.     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));

  3301. }

  3302. 

  3303. static int decode_init(AVCodecContext *avctx){

  3304.     H264Context *h= avctx->priv_data;

  3305.     MpegEncContext * const s = &h->s;

  3306. 

  3307.     MPV_decode_defaults(s);

  3308. 

  3309.     s->avctx = avctx;

  3310.     common_init(h);

  3311. 

  3312.     s->out_format = FMT_H264;

  3313.     s->workaround_bugs= avctx->workaround_bugs;

  3314. 

  3315.     // set defaults

  3316. //    s->decode_mb= ff_h263_decode_mb;

  3317.     s->low_delay= 1;

  3318.     avctx->pix_fmt= PIX_FMT_YUV420P;

  3319. 

  3320.     decode_init_vlc();

  3321. 

  3322.     if(avctx->extradata_size > 0 && avctx->extradata &&

  3323.        *(char *)avctx->extradata == 1){

  3324.         h->is_avc = 1;

  3325.         h->got_avcC = 0;

  3326.     } else {

  3327.         h->is_avc = 0;

  3328.     }

  3329. 

  3330.     return 0;

  3331. }

  3332. 

  3333. static int frame_start(H264Context *h){

  3334.     MpegEncContext * const s = &h->s;

  3335.     int i;

  3336. 

  3337.     if(MPV_frame_start(s, s->avctx) < 0)

  3338.         return -1;

  3339.     ff_er_frame_start(s);

  3340. 

  3341.     assert(s->linesize && s->uvlinesize);

  3342. 

  3343.     for(i=0; i<16; i++){

  3344.         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);

  3345.         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);

  3346.     }

  3347.     for(i=0; i<4; i++){

  3348.         h->block_offset[16+i]=

  3349.         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);

  3350.         h->block_offset[24+16+i]=

  3351.         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);

  3352.     }

  3353. 

  3354.     /* can't be in alloc_tables because linesize isn't known there.

  3355.      * FIXME: redo bipred weight to not require extra buffer? */

  3356.     if(!s->obmc_scratchpad)

  3357.         s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);

  3358. 

  3359.     /* some macroblocks will be accessed before they're available */

  3360.     if(FRAME_MBAFF)

  3361.         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));

  3362. 

  3363. //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;

  3364.     return 0;

  3365. }

  3366. 

  3367. static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){

  3368.     MpegEncContext * const s = &h->s;

  3369.     int i;

  3370. 

  3371.     src_y  -=   linesize;

  3372.     src_cb -= uvlinesize;

  3373.     src_cr -= uvlinesize;

  3374. 

  3375.     // There are two lines saved, the line above the the top macroblock of a pair,

  3376.     // and the line above the bottom macroblock

  3377.     h->left_border[0]= h->top_borders[0][s->mb_x][15];

  3378.     for(i=1; i<17; i++){

  3379.         h->left_border[i]= src_y[15+i*  linesize];

  3380.     }

  3381. 

  3382.     *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);

  3383.     *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);

  3384. 

  3385.     if(!(s->flags&CODEC_FLAG_GRAY)){

  3386.         h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];

  3387.         h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];

  3388.         for(i=1; i<9; i++){

  3389.             h->left_border[i+17  ]= src_cb[7+i*uvlinesize];

  3390.             h->left_border[i+17+9]= src_cr[7+i*uvlinesize];

  3391.         }

  3392.         *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);

  3393.         *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);

  3394.     }

  3395. }

  3396. 

  3397. static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){

  3398.     MpegEncContext * const s = &h->s;

  3399.     int temp8, i;

  3400.     uint64_t temp64;

  3401.     int deblock_left = (s->mb_x > 0);

  3402.     int deblock_top  = (s->mb_y > 0);

  3403. 

  3404.     src_y  -=   linesize + 1;

  3405.     src_cb -= uvlinesize + 1;

  3406.     src_cr -= uvlinesize + 1;

  3407. 

  3408. #define XCHG(a,b,t,xchg)\

  3409. t= a;\

  3410. if(xchg)\

  3411.     a= b;\

  3412. b= t;

  3413. 

  3414.     if(deblock_left){

  3415.         for(i = !deblock_top; i<17; i++){

  3416.             XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);

  3417.         }

  3418.     }

  3419. 

  3420.     if(deblock_top){

  3421.         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);

  3422.         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);

  3423.         if(s->mb_x+1 < s->mb_width){

  3424.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);

  3425.         }

  3426.     }

  3427. 

  3428.     if(!(s->flags&CODEC_FLAG_GRAY)){

  3429.         if(deblock_left){

  3430.             for(i = !deblock_top; i<9; i++){

  3431.                 XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);

  3432.                 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);

  3433.             }

  3434.         }

  3435.         if(deblock_top){

  3436.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);

  3437.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);

  3438.         }

  3439.     }

  3440. }

  3441. 

  3442. static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){

  3443.     MpegEncContext * const s = &h->s;

  3444.     int i;

  3445. 

  3446.     src_y  -= 2 *   linesize;

  3447.     src_cb -= 2 * uvlinesize;

  3448.     src_cr -= 2 * uvlinesize;

  3449. 

  3450.     // There are two lines saved, the line above the the top macroblock of a pair,

  3451.     // and the line above the bottom macroblock

  3452.     h->left_border[0]= h->top_borders[0][s->mb_x][15];

  3453.     h->left_border[1]= h->top_borders[1][s->mb_x][15];

  3454.     for(i=2; i<34; i++){

  3455.         h->left_border[i]= src_y[15+i*  linesize];

  3456.     }

  3457. 

  3458.     *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);

  3459.     *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);

  3460.     *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);

  3461.     *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);

  3462. 

  3463.     if(!(s->flags&CODEC_FLAG_GRAY)){

  3464.         h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];

  3465.         h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];

  3466.         h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];

  3467.         h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];

  3468.         for(i=2; i<18; i++){

  3469.             h->left_border[i+34   ]= src_cb[7+i*uvlinesize];

  3470.             h->left_border[i+34+18]= src_cr[7+i*uvlinesize];

  3471.         }

  3472.         *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);

  3473.         *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);

  3474.         *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);

  3475.         *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);

  3476.     }

  3477. }

  3478. 

  3479. static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){

  3480.     MpegEncContext * const s = &h->s;

  3481.     int temp8, i;

  3482.     uint64_t temp64;

  3483.     int deblock_left = (s->mb_x > 0);

  3484.     int deblock_top  = (s->mb_y > 1);

  3485. 

  3486.     tprintf("xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);

  3487. 

  3488.     src_y  -= 2 *   linesize + 1;

  3489.     src_cb -= 2 * uvlinesize + 1;

  3490.     src_cr -= 2 * uvlinesize + 1;

  3491. 

  3492. #define XCHG(a,b,t,xchg)\

  3493. t= a;\

  3494. if(xchg)\

  3495.     a= b;\

  3496. b= t;

  3497. 

  3498.     if(deblock_left){

  3499.         for(i = (!deblock_top)<<1; i<34; i++){

  3500.             XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);

  3501.         }

  3502.     }

  3503. 

  3504.     if(deblock_top){

  3505.         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);

  3506.         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);

  3507.         XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);

  3508.         XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);

  3509.         if(s->mb_x+1 < s->mb_width){

  3510.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);

  3511.             XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);

  3512.         }

  3513.     }

  3514. 

  3515.     if(!(s->flags&CODEC_FLAG_GRAY)){

  3516.         if(deblock_left){

  3517.             for(i = (!deblock_top) << 1; i<18; i++){

  3518.                 XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);

  3519.                 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);

  3520.             }

  3521.         }

  3522.         if(deblock_top){

  3523.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);

  3524.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);

  3525.             XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);

  3526.             XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);

  3527.         }

  3528.     }

  3529. }

  3530. 

  3531. static void hl_decode_mb(H264Context *h){

  3532.     MpegEncContext * const s = &h->s;

  3533.     const int mb_x= s->mb_x;

  3534.     const int mb_y= s->mb_y;

  3535.     const int mb_xy= mb_x + mb_y*s->mb_stride;

  3536.     const int mb_type= s->current_picture.mb_type[mb_xy];

  3537.     uint8_t  *dest_y, *dest_cb, *dest_cr;

  3538.     int linesize, uvlinesize /*dct_offset*/;

  3539.     int i;

  3540.     int *block_offset = &h->block_offset[0];

  3541.     const unsigned int bottom = mb_y & 1;

  3542.     const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);

  3543.     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);

  3544.     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);

  3545. 

  3546.     if(!s->decode)

  3547.         return;

  3548. 

  3549.     dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;

  3550.     dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;

  3551.     dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;

  3552. 

  3553.     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);

  3554.     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);

  3555. 

  3556.     if (MB_FIELD) {

  3557.         linesize   = h->mb_linesize   = s->linesize * 2;

  3558.         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;

  3559.         block_offset = &h->block_offset[24];

  3560.         if(mb_y&1){ //FIXME move out of this func?

  3561.             dest_y -= s->linesize*15;

  3562.             dest_cb-= s->uvlinesize*7;

  3563.             dest_cr-= s->uvlinesize*7;

  3564.         }

  3565.         if(FRAME_MBAFF) {

  3566.             int list;

  3567.             for(list=0; list<2; list++){

  3568.                 if(!USES_LIST(mb_type, list))

  3569.                     continue;

  3570.                 if(IS_16X16(mb_type)){

  3571.                     int8_t *ref = &h->ref_cache[list][scan8[0]];

  3572.                     fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);

  3573.                 }else{

  3574.                     for(i=0; i<16; i+=4){

  3575.                         //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?

  3576.                         int ref = h->ref_cache[list][scan8[i]];

  3577.                         if(ref >= 0)

  3578.                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);

  3579.                     }

  3580.                 }

  3581.             }

  3582.         }

  3583.     } else {

  3584.         linesize   = h->mb_linesize   = s->linesize;

  3585.         uvlinesize = h->mb_uvlinesize = s->uvlinesize;

  3586. //        dct_offset = s->linesize * 16;

  3587.     }

  3588. 

  3589.     if(transform_bypass){

  3590.         idct_dc_add =

  3591.         idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;

  3592.     }else if(IS_8x8DCT(mb_type)){

  3593.         idct_dc_add = s->dsp.h264_idct8_dc_add;

  3594.         idct_add = s->dsp.h264_idct8_add;

  3595.     }else{

  3596.         idct_dc_add = s->dsp.h264_idct_dc_add;

  3597.         idct_add = s->dsp.h264_idct_add;

  3598.     }

  3599. 

  3600.     if(FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)

  3601.        && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){

  3602.         int mbt_y = mb_y&~1;

  3603.         uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;

  3604.         uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;

  3605.         uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;

  3606.         xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);

  3607.     }

  3608. 

  3609.     if (IS_INTRA_PCM(mb_type)) {

  3610.         unsigned int x, y;

  3611. 

  3612.         // The pixels are stored in h->mb array in the same order as levels,

  3613.         // copy them in output in the correct order.

  3614.         for(i=0; i<16; i++) {

  3615.             for (y=0; y<4; y++) {

  3616.                 for (x=0; x<4; x++) {

  3617.                     *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];

  3618.                 }

  3619.             }

  3620.         }

  3621.         for(i=16; i<16+4; i++) {

  3622.             for (y=0; y<4; y++) {

  3623.                 for (x=0; x<4; x++) {

  3624.                     *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];

  3625.                 }

  3626.             }

  3627.         }

  3628.         for(i=20; i<20+4; i++) {

  3629.             for (y=0; y<4; y++) {

  3630.                 for (x=0; x<4; x++) {

  3631.                     *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];

  3632.                 }

  3633.             }

  3634.         }

  3635.     } else {

  3636.         if(IS_INTRA(mb_type)){

  3637.             if(h->deblocking_filter && !FRAME_MBAFF)

  3638.                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);

  3639. 

  3640.             if(!(s->flags&CODEC_FLAG_GRAY)){

  3641.                 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);

  3642.                 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);

  3643.             }

  3644. 

  3645.             if(IS_INTRA4x4(mb_type)){

  3646.                 if(!s->encoding){

  3647.                     if(IS_8x8DCT(mb_type)){

  3648.                         for(i=0; i<16; i+=4){

  3649.                             uint8_t * const ptr= dest_y + block_offset[i];

  3650.                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];

  3651.                             const int nnz = h->non_zero_count_cache[ scan8[i] ];

  3652.                             h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,

  3653.                                                    (h->topright_samples_available<<(i+1))&0x8000, linesize);

  3654.                             if(nnz){

  3655.                                 if(nnz == 1 && h->mb[i*16])

  3656.                                     idct_dc_add(ptr, h->mb + i*16, linesize);

  3657.                                 else

  3658.                                     idct_add(ptr, h->mb + i*16, linesize);

  3659.                             }

  3660.                         }

  3661.                     }else

  3662.                     for(i=0; i<16; i++){

  3663.                         uint8_t * const ptr= dest_y + block_offset[i];

  3664.                         uint8_t *topright;

  3665.                         const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];

  3666.                         int nnz, tr;

  3667. 

  3668.                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){

  3669.                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;

  3670.                             assert(mb_y || linesize <= block_offset[i]);

  3671.                             if(!topright_avail){

  3672.                                 tr= ptr[3 - linesize]*0x01010101;

  3673.                                 topright= (uint8_t*) &tr;

  3674.                             }else

  3675.                                 topright= ptr + 4 - linesize;

  3676.                         }else

  3677.                             topright= NULL;

  3678. 

  3679.                         h->pred4x4[ dir ](ptr, topright, linesize);

  3680.                         nnz = h->non_zero_count_cache[ scan8[i] ];

  3681.                         if(nnz){

  3682.                             if(s->codec_id == CODEC_ID_H264){

  3683.                                 if(nnz == 1 && h->mb[i*16])

  3684.                                     idct_dc_add(ptr, h->mb + i*16, linesize);

  3685.                                 else

  3686.                                     idct_add(ptr, h->mb + i*16, linesize);

  3687.                             }else

  3688.                                 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);

  3689.                         }

  3690.                     }

  3691.                 }

  3692.             }else{

  3693.                 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);

  3694.                 if(s->codec_id == CODEC_ID_H264){

  3695.                     if(!transform_bypass)

  3696.                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);

  3697.                 }else

  3698.                     svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);

  3699.             }

  3700.             if(h->deblocking_filter && !FRAME_MBAFF)

  3701.                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);

  3702.         }else if(s->codec_id == CODEC_ID_H264){

  3703.             hl_motion(h, dest_y, dest_cb, dest_cr,

  3704.                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,

  3705.                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,

  3706.                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);

  3707.         }

  3708. 

  3709. 

  3710.         if(!IS_INTRA4x4(mb_type)){

  3711.             if(s->codec_id == CODEC_ID_H264){

  3712.                 if(IS_INTRA16x16(mb_type)){

  3713.                     for(i=0; i<16; i++){

  3714.                         if(h->non_zero_count_cache[ scan8[i] ])

  3715.                             idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);

  3716.                         else if(h->mb[i*16])

  3717.                             idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);

  3718.                     }

  3719.                 }else{

  3720.                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;

  3721.                     for(i=0; i<16; i+=di){

  3722.                         int nnz = h->non_zero_count_cache[ scan8[i] ];

  3723.                         if(nnz){

  3724.                             if(nnz==1 && h->mb[i*16])

  3725.                                 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);

  3726.                             else

  3727.                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);

  3728.                         }

  3729.                     }

  3730.                 }

  3731.             }else{

  3732.                 for(i=0; i<16; i++){

  3733.                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below

  3734.                         uint8_t * const ptr= dest_y + block_offset[i];

  3735.                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);

  3736.                     }

  3737.                 }

  3738.             }

  3739.         }

  3740. 

  3741.         if(!(s->flags&CODEC_FLAG_GRAY)){

  3742.             uint8_t *dest[2] = {dest_cb, dest_cr};

  3743.             if(transform_bypass){

  3744.                 idct_add = idct_dc_add = s->dsp.add_pixels4;

  3745.             }else{

  3746.                 idct_add = s->dsp.h264_idct_add;

  3747.                 idct_dc_add = s->dsp.h264_idct_dc_add;

  3748.                 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp][0]);

  3749.                 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp][0]);

  3750.             }

  3751.             if(s->codec_id == CODEC_ID_H264){

  3752.                 for(i=16; i<16+8; i++){

  3753.                     if(h->non_zero_count_cache[ scan8[i] ])

  3754.                         idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);

  3755.                     else if(h->mb[i*16])

  3756.                         idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);

  3757.                 }

  3758.             }else{

  3759.                 for(i=16; i<16+8; i++){

  3760.                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){

  3761.                         uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];

  3762.                         svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);

  3763.                     }

  3764.                 }

  3765.             }

  3766.         }

  3767.     }

  3768.     if(h->deblocking_filter) {

  3769.         if (FRAME_MBAFF) {

  3770.             //FIXME try deblocking one mb at a time?

  3771.             // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border

  3772.             const int mb_y = s->mb_y - 1;

  3773.             uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;

  3774.             const int mb_xy= mb_x + mb_y*s->mb_stride;

  3775.             const int mb_type_top   = s->current_picture.mb_type[mb_xy];

  3776.             const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];

  3777.             if (!bottom) return;

  3778.             pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;

  3779.             pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;

  3780.             pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;

  3781. 

  3782.             if(IS_INTRA(mb_type_top | mb_type_bottom))

  3783.                 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);

  3784. 

  3785.             backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);

  3786.             // deblock a pair

  3787.             // top

  3788.             s->mb_y--;

  3789.             tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);

  3790.             fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb

  3791.             h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);

  3792.             filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);

  3793.             // bottom

  3794.             s->mb_y++;

  3795.             tprintf("call mbaff filter_mb\n");

  3796.             fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb

  3797.             h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);

  3798.             filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);

  3799.         } else {

  3800.             tprintf("call filter_mb\n");

  3801.             backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);

  3802.             fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb

  3803.             filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);

  3804.         }

  3805.     }

  3806. }

  3807. 

  3808. /**

  3809.  * fills the default_ref_list.

  3810.  */

  3811. static int fill_default_ref_list(H264Context *h){

  3812.     MpegEncContext * const s = &h->s;

  3813.     int i;

  3814.     int smallest_poc_greater_than_current = -1;

  3815.     Picture sorted_short_ref[32];

  3816. 

  3817.     if(h->slice_type==B_TYPE){

  3818.         int out_i;

  3819.         int limit= INT_MIN;

  3820. 

  3821.         /* sort frame according to poc in B slice */

  3822.         for(out_i=0; out_i<h->short_ref_count; out_i++){

  3823.             int best_i=INT_MIN;

  3824.             int best_poc=INT_MAX;

  3825. 

  3826.             for(i=0; i<h->short_ref_count; i++){

  3827.                 const int poc= h->short_ref[i]->poc;

  3828.                 if(poc > limit && poc < best_poc){

  3829.                     best_poc= poc;

  3830.                     best_i= i;

  3831.                 }

  3832.             }

  3833. 

  3834.             assert(best_i != INT_MIN);

  3835. 

  3836.             limit= best_poc;

  3837.             sorted_short_ref[out_i]= *h->short_ref[best_i];

  3838.             tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);

  3839.             if (-1 == smallest_poc_greater_than_current) {

  3840.                 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {

  3841.                     smallest_poc_greater_than_current = out_i;

  3842.                 }

  3843.             }

  3844.         }

  3845.     }

  3846. 

  3847.     if(s->picture_structure == PICT_FRAME){

  3848.         if(h->slice_type==B_TYPE){

  3849.             int list;

  3850.             tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);

  3851. 

  3852.             // find the largest poc

  3853.             for(list=0; list<2; list++){

  3854.                 int index = 0;

  3855.                 int j= -99;

  3856.                 int step= list ? -1 : 1;

  3857. 

  3858.                 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {

  3859.                     while(j<0 || j>= h->short_ref_count){

  3860.                         if(j != -99 && step == (list ? -1 : 1))

  3861.                             return -1;

  3862.                         step = -step;

  3863.                         j= smallest_poc_greater_than_current + (step>>1);

  3864.                     }

  3865.                     if(sorted_short_ref[j].reference != 3) continue;

  3866.                     h->default_ref_list[list][index  ]= sorted_short_ref[j];

  3867.                     h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;

  3868.                 }

  3869. 

  3870.                 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){

  3871.                     if(h->long_ref[i] == NULL) continue;

  3872.                     if(h->long_ref[i]->reference != 3) continue;

  3873. 

  3874.                     h->default_ref_list[ list ][index  ]= *h->long_ref[i];

  3875.                     h->default_ref_list[ list ][index++].pic_id= i;;

  3876.                 }

  3877. 

  3878.                 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){

  3879.                     // swap the two first elements of L1 when

  3880.                     // L0 and L1 are identical

  3881.                     Picture temp= h->default_ref_list[1][0];

  3882.                     h->default_ref_list[1][0] = h->default_ref_list[1][1];

  3883.                     h->default_ref_list[1][1] = temp;

  3884.                 }

  3885. 

  3886.                 if(index < h->ref_count[ list ])

  3887.                     memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));

  3888.             }

  3889.         }else{

  3890.             int index=0;

  3891.             for(i=0; i<h->short_ref_count; i++){

  3892.                 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit

  3893.                 h->default_ref_list[0][index  ]= *h->short_ref[i];

  3894.                 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;

  3895.             }

  3896.             for(i = 0; i < 16; i++){

  3897.                 if(h->long_ref[i] == NULL) continue;

  3898.                 if(h->long_ref[i]->reference != 3) continue;

  3899.                 h->default_ref_list[0][index  ]= *h->long_ref[i];

  3900.                 h->default_ref_list[0][index++].pic_id= i;;

  3901.             }

  3902.             if(index < h->ref_count[0])

  3903.                 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));

  3904.         }

  3905.     }else{ //FIELD

  3906.         if(h->slice_type==B_TYPE){

  3907.         }else{

  3908.             //FIXME second field balh

  3909.         }

  3910.     }

  3911. #ifdef TRACE

  3912.     for (i=0; i<h->ref_count[0]; i++) {

  3913.         tprintf("List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);

  3914.     }

  3915.     if(h->slice_type==B_TYPE){

  3916.         for (i=0; i<h->ref_count[1]; i++) {

  3917.             tprintf("List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);

  3918.         }

  3919.     }

  3920. #endif

  3921.     return 0;

  3922. }

  3923. 

  3924. static void print_short_term(H264Context *h);

  3925. static void print_long_term(H264Context *h);

  3926. 

  3927. static int decode_ref_pic_list_reordering(H264Context *h){

  3928.     MpegEncContext * const s = &h->s;

  3929.     int list, index;

  3930. 

  3931.     print_short_term(h);

  3932.     print_long_term(h);

  3933.     if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func

  3934. 

  3935.     for(list=0; list<2; list++){

  3936.         memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);

  3937. 

  3938.         if(get_bits1(&s->gb)){

  3939.             int pred= h->curr_pic_num;

  3940. 

  3941.             for(index=0; ; index++){

  3942.                 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);

  3943.                 int pic_id;

  3944.                 int i;

  3945.                 Picture *ref = NULL;

  3946. 

  3947.                 if(reordering_of_pic_nums_idc==3)

  3948.                     break;

  3949. 

  3950.                 if(index >= h->ref_count[list]){

  3951.                     av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");

  3952.                     return -1;

  3953.                 }

  3954. 

  3955.                 if(reordering_of_pic_nums_idc<3){

  3956.                     if(reordering_of_pic_nums_idc<2){

  3957.                         const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;

  3958. 

  3959.                         if(abs_diff_pic_num >= h->max_pic_num){

  3960.                             av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");

  3961.                             return -1;

  3962.                         }

  3963. 

  3964.                         if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;

  3965.                         else                                pred+= abs_diff_pic_num;

  3966.                         pred &= h->max_pic_num - 1;

  3967. 

  3968.                         for(i= h->short_ref_count-1; i>=0; i--){

  3969.                             ref = h->short_ref[i];

  3970.                             assert(ref->reference == 3);

  3971.                             assert(!ref->long_ref);

  3972.                             if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer

  3973.                                 break;

  3974.                         }

  3975.                         if(i>=0)

  3976.                             ref->pic_id= ref->frame_num;

  3977.                     }else{

  3978.                         pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx

  3979.                         ref = h->long_ref[pic_id];

  3980.                         ref->pic_id= pic_id;

  3981.                         assert(ref->reference == 3);

  3982.                         assert(ref->long_ref);

  3983.                         i=0;

  3984.                     }

  3985. 

  3986.                     if (i < 0) {

  3987.                         av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");

  3988.                         memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME

  3989.                     } else {

  3990.                         for(i=index; i+1<h->ref_count[list]; i++){

  3991.                             if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)

  3992.                                 break;

  3993.                         }

  3994.                         for(; i > index; i--){

  3995.                             h->ref_list[list][i]= h->ref_list[list][i-1];

  3996.                         }

  3997.                         h->ref_list[list][index]= *ref;

  3998.                     }

  3999.                 }else{

  4000.                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");

  4001.                     return -1;

  4002.                 }

  4003.             }

  4004.         }

  4005. 

  4006.         if(h->slice_type!=B_TYPE) break;

  4007.     }

  4008.     for(list=0; list<2; list++){

  4009.         for(index= 0; index < h->ref_count[list]; index++){

  4010.             if(!h->ref_list[list][index].data[0])

  4011.                 h->ref_list[list][index]= s->current_picture;

  4012.         }

  4013.         if(h->slice_type!=B_TYPE) break;

  4014.     }

  4015. 

  4016.     if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)

  4017.         direct_dist_scale_factor(h);

  4018.     direct_ref_list_init(h);

  4019.     return 0;

  4020. }

  4021. 

  4022. static void fill_mbaff_ref_list(H264Context *h){

  4023.     int list, i, j;

  4024.     for(list=0; list<2; list++){

  4025.         for(i=0; i<h->ref_count[list]; i++){

  4026.             Picture *frame = &h->ref_list[list][i];

  4027.             Picture *field = &h->ref_list[list][16+2*i];

  4028.             field[0] = *frame;

  4029.             for(j=0; j<3; j++)

  4030.                 field[0].linesize[j] <<= 1;

  4031.             field[1] = field[0];

  4032.             for(j=0; j<3; j++)

  4033.                 field[1].data[j] += frame->linesize[j];

  4034. 

  4035.             h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];

  4036.             h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];

  4037.             for(j=0; j<2; j++){

  4038.                 h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];

  4039.                 h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];

  4040.             }

  4041.         }

  4042.     }

  4043.     for(j=0; j<h->ref_count[1]; j++){

  4044.         for(i=0; i<h->ref_count[0]; i++)

  4045.             h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];

  4046.         memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));

  4047.         memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));

  4048.     }

  4049. }

  4050. 

  4051. static int pred_weight_table(H264Context *h){

  4052.     MpegEncContext * const s = &h->s;

  4053.     int list, i;

  4054.     int luma_def, chroma_def;

  4055. 

  4056.     h->use_weight= 0;

  4057.     h->use_weight_chroma= 0;

  4058.     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);

  4059.     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);

  4060.     luma_def = 1<<h->luma_log2_weight_denom;

  4061.     chroma_def = 1<<h->chroma_log2_weight_denom;

  4062. 

  4063.     for(list=0; list<2; list++){

  4064.         for(i=0; i<h->ref_count[list]; i++){

  4065.             int luma_weight_flag, chroma_weight_flag;

  4066. 

  4067.             luma_weight_flag= get_bits1(&s->gb);

  4068.             if(luma_weight_flag){

  4069.                 h->luma_weight[list][i]= get_se_golomb(&s->gb);

  4070.                 h->luma_offset[list][i]= get_se_golomb(&s->gb);

  4071.                 if(   h->luma_weight[list][i] != luma_def

  4072.                    || h->luma_offset[list][i] != 0)

  4073.                     h->use_weight= 1;

  4074.             }else{

  4075.                 h->luma_weight[list][i]= luma_def;

  4076.                 h->luma_offset[list][i]= 0;

  4077.             }

  4078. 

  4079.             chroma_weight_flag= get_bits1(&s->gb);

  4080.             if(chroma_weight_flag){

  4081.                 int j;

  4082.                 for(j=0; j<2; j++){

  4083.                     h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);

  4084.                     h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);

  4085.                     if(   h->chroma_weight[list][i][j] != chroma_def

  4086.                        || h->chroma_offset[list][i][j] != 0)

  4087.                         h->use_weight_chroma= 1;

  4088.                 }

  4089.             }else{

  4090.                 int j;

  4091.                 for(j=0; j<2; j++){

  4092.                     h->chroma_weight[list][i][j]= chroma_def;

  4093.                     h->chroma_offset[list][i][j]= 0;

  4094.                 }

  4095.             }

  4096.         }

  4097.         if(h->slice_type != B_TYPE) break;

  4098.     }

  4099.     h->use_weight= h->use_weight || h->use_weight_chroma;

  4100.     return 0;

  4101. }

  4102. 

  4103. static void implicit_weight_table(H264Context *h){

  4104.     MpegEncContext * const s = &h->s;

  4105.     int ref0, ref1;

  4106.     int cur_poc = s->current_picture_ptr->poc;

  4107. 

  4108.     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1

  4109.        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){

  4110.         h->use_weight= 0;

  4111.         h->use_weight_chroma= 0;

  4112.         return;

  4113.     }

  4114. 

  4115.     h->use_weight= 2;

  4116.     h->use_weight_chroma= 2;

  4117.     h->luma_log2_weight_denom= 5;

  4118.     h->chroma_log2_weight_denom= 5;

  4119. 

  4120.     for(ref0=0; ref0 < h->ref_count[0]; ref0++){

  4121.         int poc0 = h->ref_list[0][ref0].poc;

  4122.         for(ref1=0; ref1 < h->ref_count[1]; ref1++){

  4123.             int poc1 = h->ref_list[1][ref1].poc;

  4124.             int td = clip(poc1 - poc0, -128, 127);

  4125.             if(td){

  4126.                 int tb = clip(cur_poc - poc0, -128, 127);

  4127.                 int tx = (16384 + (FFABS(td) >> 1)) / td;

  4128.                 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;

  4129.                 if(dist_scale_factor < -64 || dist_scale_factor > 128)

  4130.                     h->implicit_weight[ref0][ref1] = 32;

  4131.                 else

  4132.                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;

  4133.             }else

  4134.                 h->implicit_weight[ref0][ref1] = 32;

  4135.         }

  4136.     }

  4137. }

  4138. 

  4139. static inline void unreference_pic(H264Context *h, Picture *pic){

  4140.     int i;

  4141.     pic->reference=0;

  4142.     if(pic == h->delayed_output_pic)

  4143.         pic->reference=1;

  4144.     else{

  4145.         for(i = 0; h->delayed_pic[i]; i++)

  4146.             if(pic == h->delayed_pic[i]){

  4147.                 pic->reference=1;

  4148.                 break;

  4149.             }

  4150.     }

  4151. }

  4152. 

  4153. /**

  4154.  * instantaneous decoder refresh.

  4155.  */

  4156. static void idr(H264Context *h){

  4157.     int i;

  4158. 

  4159.     for(i=0; i<16; i++){

  4160.         if (h->long_ref[i] != NULL) {

  4161.             unreference_pic(h, h->long_ref[i]);

  4162.             h->long_ref[i]= NULL;

  4163.         }

  4164.     }

  4165.     h->long_ref_count=0;

  4166. 

  4167.     for(i=0; i<h->short_ref_count; i++){

  4168.         unreference_pic(h, h->short_ref[i]);

  4169.         h->short_ref[i]= NULL;

  4170.     }

  4171.     h->short_ref_count=0;

  4172. }

  4173. 

  4174. /* forget old pics after a seek */

  4175. static void flush_dpb(AVCodecContext *avctx){

  4176.     H264Context *h= avctx->priv_data;

  4177.     int i;

  4178.     for(i=0; i<16; i++) {

  4179.         if(h->delayed_pic[i])

  4180.             h->delayed_pic[i]->reference= 0;

  4181.         h->delayed_pic[i]= NULL;

  4182.     }

  4183.     if(h->delayed_output_pic)

  4184.         h->delayed_output_pic->reference= 0;

  4185.     h->delayed_output_pic= NULL;

  4186.     idr(h);

  4187.     if(h->s.current_picture_ptr)

  4188.         h->s.current_picture_ptr->reference= 0;

  4189. }

  4190. 

  4191. /**

  4192.  *

  4193.  * @return the removed picture or NULL if an error occurs

  4194.  */

  4195. static Picture * remove_short(H264Context *h, int frame_num){

  4196.     MpegEncContext * const s = &h->s;

  4197.     int i;

  4198. 

  4199.     if(s->avctx->debug&FF_DEBUG_MMCO)

  4200.         av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);

  4201. 

  4202.     for(i=0; i<h->short_ref_count; i++){

  4203.         Picture *pic= h->short_ref[i];

  4204.         if(s->avctx->debug&FF_DEBUG_MMCO)

  4205.             av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);

  4206.         if(pic->frame_num == frame_num){

  4207.             h->short_ref[i]= NULL;

  4208.             memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));

  4209.             h->short_ref_count--;

  4210.             return pic;

  4211.         }

  4212.     }

  4213.     return NULL;

  4214. }

  4215. 

  4216. /**

  4217.  *

  4218.  * @return the removed picture or NULL if an error occurs

  4219.  */

  4220. static Picture * remove_long(H264Context *h, int i){

  4221.     Picture *pic;

  4222. 

  4223.     pic= h->long_ref[i];

  4224.     h->long_ref[i]= NULL;

  4225.     if(pic) h->long_ref_count--;

  4226. 

  4227.     return pic;

  4228. }

  4229. 

  4230. /**

  4231.  * print short term list

  4232.  */

  4233. static void print_short_term(H264Context *h) {

  4234.     uint32_t i;

  4235.     if(h->s.avctx->debug&FF_DEBUG_MMCO) {

  4236.         av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");

  4237.         for(i=0; i<h->short_ref_count; i++){

  4238.             Picture *pic= h->short_ref[i];

  4239.             av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);

  4240.         }

  4241.     }

  4242. }

  4243. 

  4244. /**

  4245.  * print long term list

  4246.  */

  4247. static void print_long_term(H264Context *h) {

  4248.     uint32_t i;

  4249.     if(h->s.avctx->debug&FF_DEBUG_MMCO) {

  4250.         av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");

  4251.         for(i = 0; i < 16; i++){

  4252.             Picture *pic= h->long_ref[i];

  4253.             if (pic) {

  4254.                 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);

  4255.             }

  4256.         }

  4257.     }

  4258. }

  4259. 

  4260. /**

  4261.  * Executes the reference picture marking (memory management control operations).

  4262.  */

  4263. static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){

  4264.     MpegEncContext * const s = &h->s;

  4265.     int i, j;

  4266.     int current_is_long=0;

  4267.     Picture *pic;

  4268. 

  4269.     if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)

  4270.         av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");

  4271. 

  4272.     for(i=0; i<mmco_count; i++){

  4273.         if(s->avctx->debug&FF_DEBUG_MMCO)

  4274.             av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);

  4275. 

  4276.         switch(mmco[i].opcode){

  4277.         case MMCO_SHORT2UNUSED:

  4278.             pic= remove_short(h, mmco[i].short_frame_num);

  4279.             if(pic)

  4280.                 unreference_pic(h, pic);

  4281.             else if(s->avctx->debug&FF_DEBUG_MMCO)

  4282.                 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");

  4283.             break;

  4284.         case MMCO_SHORT2LONG:

  4285.             pic= remove_long(h, mmco[i].long_index);

  4286.             if(pic) unreference_pic(h, pic);

  4287. 

  4288.             h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);

  4289.             h->long_ref[ mmco[i].long_index ]->long_ref=1;

  4290.             h->long_ref_count++;

  4291.             break;

  4292.         case MMCO_LONG2UNUSED:

  4293.             pic= remove_long(h, mmco[i].long_index);

  4294.             if(pic)

  4295.                 unreference_pic(h, pic);

  4296.             else if(s->avctx->debug&FF_DEBUG_MMCO)

  4297.                 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");

  4298.             break;

  4299.         case MMCO_LONG:

  4300.             pic= remove_long(h, mmco[i].long_index);

  4301.             if(pic) unreference_pic(h, pic);

  4302. 

  4303.             h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;

  4304.             h->long_ref[ mmco[i].long_index ]->long_ref=1;

  4305.             h->long_ref_count++;

  4306. 

  4307.             current_is_long=1;

  4308.             break;

  4309.         case MMCO_SET_MAX_LONG:

  4310.             assert(mmco[i].long_index <= 16);

  4311.             // just remove the long term which index is greater than new max

  4312.             for(j = mmco[i].long_index; j<16; j++){

  4313.                 pic = remove_long(h, j);

  4314.                 if (pic) unreference_pic(h, pic);

  4315.             }

  4316.             break;

  4317.         case MMCO_RESET:

  4318.             while(h->short_ref_count){

  4319.                 pic= remove_short(h, h->short_ref[0]->frame_num);

  4320.                 unreference_pic(h, pic);

  4321.             }

  4322.             for(j = 0; j < 16; j++) {

  4323.                 pic= remove_long(h, j);

  4324.                 if(pic) unreference_pic(h, pic);

  4325.             }

  4326.             break;

  4327.         default: assert(0);

  4328.         }

  4329.     }

  4330. 

  4331.     if(!current_is_long){

  4332.         pic= remove_short(h, s->current_picture_ptr->frame_num);

  4333.         if(pic){

  4334.             unreference_pic(h, pic);

  4335.             av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");

  4336.         }

  4337. 

  4338.         if(h->short_ref_count)

  4339.             memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));

  4340. 

  4341.         h->short_ref[0]= s->current_picture_ptr;

  4342.         h->short_ref[0]->long_ref=0;

  4343.         h->short_ref_count++;

  4344.     }

  4345. 

  4346.     print_short_term(h);

  4347.     print_long_term(h);

  4348.     return 0;

  4349. }

  4350. 

  4351. static int decode_ref_pic_marking(H264Context *h){

  4352.     MpegEncContext * const s = &h->s;

  4353.     int i;

  4354. 

  4355.     if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields

  4356.         s->broken_link= get_bits1(&s->gb) -1;

  4357.         h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx

  4358.         if(h->mmco[0].long_index == -1)

  4359.             h->mmco_index= 0;

  4360.         else{

  4361.             h->mmco[0].opcode= MMCO_LONG;

  4362.             h->mmco_index= 1;

  4363.         }

  4364.     }else{

  4365.         if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag

  4366.             for(i= 0; i<MAX_MMCO_COUNT; i++) {

  4367.                 MMCOOpcode opcode= get_ue_golomb(&s->gb);;

  4368. 

  4369.                 h->mmco[i].opcode= opcode;

  4370.                 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){

  4371.                     h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields

  4372. /*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){

  4373.                         av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);

  4374.                         return -1;

  4375.                     }*/

  4376.                 }

  4377.                 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){

  4378.                     h->mmco[i].long_index= get_ue_golomb(&s->gb);

  4379.                     if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ h->mmco[i].long_index >= 16){

  4380.                         av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);

  4381.                         return -1;

  4382.                     }

  4383.                 }

  4384. 

  4385.                 if(opcode > MMCO_LONG){

  4386.                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);

  4387.                     return -1;

  4388.                 }

  4389.                 if(opcode == MMCO_END)

  4390.                     break;

  4391.             }

  4392.             h->mmco_index= i;

  4393.         }else{

  4394.             assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);

  4395. 

  4396.             if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields

  4397.                 h->mmco[0].opcode= MMCO_SHORT2UNUSED;

  4398.                 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;

  4399.                 h->mmco_index= 1;

  4400.             }else

  4401.                 h->mmco_index= 0;

  4402.         }

  4403.     }

  4404. 

  4405.     return 0;

  4406. }

  4407. 

  4408. static int init_poc(H264Context *h){

  4409.     MpegEncContext * const s = &h->s;

  4410.     const int max_frame_num= 1<<h->sps.log2_max_frame_num;

  4411.     int field_poc[2];

  4412. 

  4413.     if(h->nal_unit_type == NAL_IDR_SLICE){

  4414.         h->frame_num_offset= 0;

  4415.     }else{

  4416.         if(h->frame_num < h->prev_frame_num)

  4417.             h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;

  4418.         else

  4419.             h->frame_num_offset= h->prev_frame_num_offset;

  4420.     }

  4421. 

  4422.     if(h->sps.poc_type==0){

  4423.         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;

  4424. 

  4425.         if(h->nal_unit_type == NAL_IDR_SLICE){

  4426.              h->prev_poc_msb=

  4427.              h->prev_poc_lsb= 0;

  4428.         }

  4429. 

  4430.         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)

  4431.             h->poc_msb = h->prev_poc_msb + max_poc_lsb;

  4432.         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)

  4433.             h->poc_msb = h->prev_poc_msb - max_poc_lsb;

  4434.         else

  4435.             h->poc_msb = h->prev_poc_msb;

  4436. //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);

  4437.         field_poc[0] =

  4438.         field_poc[1] = h->poc_msb + h->poc_lsb;

  4439.         if(s->picture_structure == PICT_FRAME)

  4440.             field_poc[1] += h->delta_poc_bottom;

  4441.     }else if(h->sps.poc_type==1){

  4442.         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;

  4443.         int i;

  4444. 

  4445.         if(h->sps.poc_cycle_length != 0)

  4446.             abs_frame_num = h->frame_num_offset + h->frame_num;

  4447.         else

  4448.             abs_frame_num = 0;

  4449. 

  4450.         if(h->nal_ref_idc==0 && abs_frame_num > 0)

  4451.             abs_frame_num--;

  4452. 

  4453.         expected_delta_per_poc_cycle = 0;

  4454.         for(i=0; i < h->sps.poc_cycle_length; i++)

  4455.             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse

  4456. 

  4457.         if(abs_frame_num > 0){

  4458.             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;

  4459.             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;

  4460. 

  4461.             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;

  4462.             for(i = 0; i <= frame_num_in_poc_cycle; i++)

  4463.                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];

  4464.         } else

  4465.             expectedpoc = 0;

  4466. 

  4467.         if(h->nal_ref_idc == 0)

  4468.             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;

  4469. 

  4470.         field_poc[0] = expectedpoc + h->delta_poc[0];

  4471.         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;

  4472. 

  4473.         if(s->picture_structure == PICT_FRAME)

  4474.             field_poc[1] += h->delta_poc[1];

  4475.     }else{

  4476.         int poc;

  4477.         if(h->nal_unit_type == NAL_IDR_SLICE){

  4478.             poc= 0;

  4479.         }else{

  4480.             if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);

  4481.             else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;

  4482.         }

  4483.         field_poc[0]= poc;

  4484.         field_poc[1]= poc;

  4485.     }

  4486. 

  4487.     if(s->picture_structure != PICT_BOTTOM_FIELD)

  4488.         s->current_picture_ptr->field_poc[0]= field_poc[0];

  4489.     if(s->picture_structure != PICT_TOP_FIELD)

  4490.         s->current_picture_ptr->field_poc[1]= field_poc[1];

  4491.     if(s->picture_structure == PICT_FRAME) // FIXME field pix?

  4492.         s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);

  4493. 

  4494.     return 0;

  4495. }

  4496. 

  4497. /**

  4498.  * decodes a slice header.

  4499.  * this will allso call MPV_common_init() and frame_start() as needed

  4500.  */

  4501. static int decode_slice_header(H264Context *h){

  4502.     MpegEncContext * const s = &h->s;

  4503.     int first_mb_in_slice, pps_id;

  4504.     int num_ref_idx_active_override_flag;

  4505.     static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};

  4506.     int slice_type;

  4507.     int default_ref_list_done = 0;

  4508. 

  4509.     s->current_picture.reference= h->nal_ref_idc != 0;

  4510.     s->dropable= h->nal_ref_idc == 0;

  4511. 

  4512.     first_mb_in_slice= get_ue_golomb(&s->gb);

  4513. 

  4514.     slice_type= get_ue_golomb(&s->gb);

  4515.     if(slice_type > 9){

  4516.         av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);

  4517.         return -1;

  4518.     }

  4519.     if(slice_type > 4){

  4520.         slice_type -= 5;

  4521.         h->slice_type_fixed=1;

  4522.     }else

  4523.         h->slice_type_fixed=0;

  4524. 

  4525.     slice_type= slice_type_map[ slice_type ];

  4526.     if (slice_type == I_TYPE

  4527.         || (h->slice_num != 0 && slice_type == h->slice_type) ) {

  4528.         default_ref_list_done = 1;

  4529.     }

  4530.     h->slice_type= slice_type;

  4531. 

  4532.     s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though

  4533. 

  4534.     pps_id= get_ue_golomb(&s->gb);

  4535.     if(pps_id>255){

  4536.         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");

  4537.         return -1;

  4538.     }

  4539.     h->pps= h->pps_buffer[pps_id];

  4540.     if(h->pps.slice_group_count == 0){

  4541.         av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");

  4542.         return -1;

  4543.     }

  4544. 

  4545.     h->sps= h->sps_buffer[ h->pps.sps_id ];

  4546.     if(h->sps.log2_max_frame_num == 0){

  4547.         av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");

  4548.         return -1;

  4549.     }

  4550. 

  4551.     if(h->dequant_coeff_pps != pps_id){

  4552.         h->dequant_coeff_pps = pps_id;

  4553.         init_dequant_tables(h);

  4554.     }

  4555. 

  4556.     s->mb_width= h->sps.mb_width;

  4557.     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);

  4558. 

  4559.     h->b_stride=  s->mb_width*4;

  4560.     h->b8_stride= s->mb_width*2;

  4561. 

  4562.     s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );

  4563.     if(h->sps.frame_mbs_only_flag)

  4564.         s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);

  4565.     else

  4566.         s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck

  4567. 

  4568.     if (s->context_initialized

  4569.         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {

  4570.         free_tables(h);

  4571.         MPV_common_end(s);

  4572.     }

  4573.     if (!s->context_initialized) {

  4574.         if (MPV_common_init(s) < 0)

  4575.             return -1;

  4576. 

  4577.         if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly

  4578.             memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));

  4579.             memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));

  4580.         }else{

  4581.             int i;

  4582.             for(i=0; i<16; i++){

  4583. #define T(x) (x>>2) | ((x<<2) & 0xF)

  4584.                 h->zigzag_scan[i] = T(zigzag_scan[i]);

  4585.                 h-> field_scan[i] = T( field_scan[i]);

  4586. #undef T

  4587.             }

  4588.         }

  4589.         if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){

  4590.             memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));

  4591.             memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));

  4592.             memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));

  4593.             memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));

  4594.         }else{

  4595.             int i;

  4596.             for(i=0; i<64; i++){

  4597. #define T(x) (x>>3) | ((x&7)<<3)

  4598.                 h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);

  4599.                 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);

  4600.                 h->field_scan8x8[i]        = T(field_scan8x8[i]);

  4601.                 h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);

  4602. #undef T

  4603.             }

  4604.         }

  4605.         if(h->sps.transform_bypass){ //FIXME same ugly

  4606.             h->zigzag_scan_q0          = zigzag_scan;

  4607.             h->zigzag_scan8x8_q0       = zigzag_scan8x8;

  4608.             h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;

  4609.             h->field_scan_q0           = field_scan;

  4610.             h->field_scan8x8_q0        = field_scan8x8;

  4611.             h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;

  4612.         }else{

  4613.             h->zigzag_scan_q0          = h->zigzag_scan;

  4614.             h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;

  4615.             h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;

  4616.             h->field_scan_q0           = h->field_scan;

  4617.             h->field_scan8x8_q0        = h->field_scan8x8;

  4618.             h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;

  4619.         }

  4620. 

  4621.         alloc_tables(h);

  4622. 

  4623.         s->avctx->width = s->width;

  4624.         s->avctx->height = s->height;

  4625.         s->avctx->sample_aspect_ratio= h->sps.sar;

  4626.         if(!s->avctx->sample_aspect_ratio.den)

  4627.             s->avctx->sample_aspect_ratio.den = 1;

  4628. 

  4629.         if(h->sps.timing_info_present_flag){

  4630.             s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};

  4631.             if(h->x264_build > 0 && h->x264_build < 44)

  4632.                 s->avctx->time_base.den *= 2;

  4633.             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,

  4634.                       s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);

  4635.         }

  4636.     }

  4637. 

  4638.     if(h->slice_num == 0){

  4639.         if(frame_start(h) < 0)

  4640.             return -1;

  4641.     }

  4642. 

  4643.     s->current_picture_ptr->frame_num= //FIXME frame_num cleanup

  4644.     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);

  4645. 

  4646.     h->mb_mbaff = 0;

  4647.     h->mb_aff_frame = 0;

  4648.     if(h->sps.frame_mbs_only_flag){

  4649.         s->picture_structure= PICT_FRAME;

  4650.     }else{

  4651.         if(get_bits1(&s->gb)) { //field_pic_flag

  4652.             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag

  4653.             av_log(h->s.avctx, AV_LOG_ERROR, "PAFF interlacing is not implemented\n");

  4654.         } else {

  4655.             s->picture_structure= PICT_FRAME;

  4656.             h->mb_aff_frame = h->sps.mb_aff;

  4657.         }

  4658.     }

  4659. 

  4660.     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;

  4661.     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << h->mb_aff_frame;

  4662.     if(s->mb_y >= s->mb_height){

  4663.         return -1;

  4664.     }

  4665. 

  4666.     if(s->picture_structure==PICT_FRAME){

  4667.         h->curr_pic_num=   h->frame_num;

  4668.         h->max_pic_num= 1<< h->sps.log2_max_frame_num;

  4669.     }else{

  4670.         h->curr_pic_num= 2*h->frame_num;

  4671.         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);

  4672.     }

  4673. 

  4674.     if(h->nal_unit_type == NAL_IDR_SLICE){

  4675.         get_ue_golomb(&s->gb); /* idr_pic_id */

  4676.     }

  4677. 

  4678.     if(h->sps.poc_type==0){

  4679.         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);

  4680. 

  4681.         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){

  4682.             h->delta_poc_bottom= get_se_golomb(&s->gb);

  4683.         }

  4684.     }

  4685. 

  4686.     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){

  4687.         h->delta_poc[0]= get_se_golomb(&s->gb);

  4688. 

  4689.         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)

  4690.             h->delta_poc[1]= get_se_golomb(&s->gb);

  4691.     }

  4692. 

  4693.     init_poc(h);

  4694. 

  4695.     if(h->pps.redundant_pic_cnt_present){

  4696.         h->redundant_pic_count= get_ue_golomb(&s->gb);

  4697.     }

  4698. 

  4699.     //set defaults, might be overriden a few line later

  4700.     h->ref_count[0]= h->pps.ref_count[0];

  4701.     h->ref_count[1]= h->pps.ref_count[1];

  4702. 

  4703.     if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){

  4704.         if(h->slice_type == B_TYPE){

  4705.             h->direct_spatial_mv_pred= get_bits1(&s->gb);

  4706.             if(h->sps.mb_aff && h->direct_spatial_mv_pred)

  4707.                 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + spatial direct mode is not implemented\n");

  4708.         }

  4709.         num_ref_idx_active_override_flag= get_bits1(&s->gb);

  4710. 

  4711.         if(num_ref_idx_active_override_flag){

  4712.             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;

  4713.             if(h->slice_type==B_TYPE)

  4714.                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;

  4715. 

  4716.             if(h->ref_count[0] > 32 || h->ref_count[1] > 32){

  4717.                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");

  4718.                 return -1;

  4719.             }

  4720.         }

  4721.     }

  4722. 

  4723.     if(!default_ref_list_done){

  4724.         fill_default_ref_list(h);

  4725.     }

  4726. 

  4727.     if(decode_ref_pic_list_reordering(h) < 0)

  4728.         return -1;

  4729. 

  4730.     if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))

  4731.        || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )

  4732.         pred_weight_table(h);

  4733.     else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)

  4734.         implicit_weight_table(h);

  4735.     else

  4736.         h->use_weight = 0;

  4737. 

  4738.     if(s->current_picture.reference)

  4739.         decode_ref_pic_marking(h);

  4740. 

  4741.     if(FRAME_MBAFF)

  4742.         fill_mbaff_ref_list(h);

  4743. 

  4744.     if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )

  4745.         h->cabac_init_idc = get_ue_golomb(&s->gb);

  4746. 

  4747.     h->last_qscale_diff = 0;

  4748.     s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);

  4749.     if(s->qscale<0 || s->qscale>51){

  4750.         av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);

  4751.         return -1;

  4752.     }

  4753.     h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);

  4754.     //FIXME qscale / qp ... stuff

  4755.     if(h->slice_type == SP_TYPE){

  4756.         get_bits1(&s->gb); /* sp_for_switch_flag */

  4757.     }

  4758.     if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){

  4759.         get_se_golomb(&s->gb); /* slice_qs_delta */

  4760.     }

  4761. 

  4762.     h->deblocking_filter = 1;

  4763.     h->slice_alpha_c0_offset = 0;

  4764.     h->slice_beta_offset = 0;

  4765.     if( h->pps.deblocking_filter_parameters_present ) {

  4766.         h->deblocking_filter= get_ue_golomb(&s->gb);

  4767.         if(h->deblocking_filter < 2)

  4768.             h->deblocking_filter^= 1; // 1<->0

  4769. 

  4770.         if( h->deblocking_filter ) {

  4771.             h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;

  4772.             h->slice_beta_offset = get_se_golomb(&s->gb) << 1;

  4773.         }

  4774.     }

  4775.     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL

  4776.        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type != I_TYPE)

  4777.        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type == B_TYPE)

  4778.        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))

  4779.         h->deblocking_filter= 0;

  4780. 

  4781. #if 0 //FMO

  4782.     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)

  4783.         slice_group_change_cycle= get_bits(&s->gb, ?);

  4784. #endif

  4785. 

  4786.     h->slice_num++;

  4787. 

  4788.     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;

  4789.     h->emu_edge_height= FRAME_MBAFF ? 0 : h->emu_edge_width;

  4790. 

  4791.     if(s->avctx->debug&FF_DEBUG_PICT_INFO){

  4792.         av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s\n",

  4793.                h->slice_num,

  4794.                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),

  4795.                first_mb_in_slice,

  4796.                av_get_pict_type_char(h->slice_type),

  4797.                pps_id, h->frame_num,

  4798.                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],

  4799.                h->ref_count[0], h->ref_count[1],

  4800.                s->qscale,

  4801.                h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,

  4802.                h->use_weight,

  4803.                h->use_weight==1 && h->use_weight_chroma ? "c" : ""

  4804.                );

  4805.     }

  4806. 

  4807.     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !s->current_picture.reference){

  4808.         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;

  4809.         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;

  4810.     }else{

  4811.         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;

  4812.         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;

  4813.     }

  4814. 

  4815.     return 0;

  4816. }

  4817. 

  4818. /**

  4819.  *

  4820.  */

  4821. static inline int get_level_prefix(GetBitContext *gb){

  4822.     unsigned int buf;

  4823.     int log;

  4824. 

  4825.     OPEN_READER(re, gb);

  4826.     UPDATE_CACHE(re, gb);

  4827.     buf=GET_CACHE(re, gb);

  4828. 

  4829.     log= 32 - av_log2(buf);

  4830. #ifdef TRACE

  4831.     print_bin(buf>>(32-log), log);

  4832.     av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);

  4833. #endif

  4834. 

  4835.     LAST_SKIP_BITS(re, gb, log);

  4836.     CLOSE_READER(re, gb);

  4837. 

  4838.     return log-1;

  4839. }

  4840. 

  4841. static inline int get_dct8x8_allowed(H264Context *h){

  4842.     int i;

  4843.     for(i=0; i<4; i++){

  4844.         if(!IS_SUB_8X8(h->sub_mb_type[i])

  4845.            || (!h->sps.direct_8x8_inference_flag && IS_DIRECT(h->sub_mb_type[i])))

  4846.             return 0;

  4847.     }

  4848.     return 1;

  4849. }

  4850. 

  4851. /**

  4852.  * decodes a residual block.

  4853.  * @param n block index

  4854.  * @param scantable scantable

  4855.  * @param max_coeff number of coefficients in the block

  4856.  * @return <0 if an error occured

  4857.  */

  4858. static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){

  4859.     MpegEncContext * const s = &h->s;

  4860.     static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};

  4861.     int level[16];

  4862.     int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;

  4863. 

  4864.     //FIXME put trailing_onex into the context

  4865. 

  4866.     if(n == CHROMA_DC_BLOCK_INDEX){

  4867.         coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);

  4868.         total_coeff= coeff_token>>2;

  4869.     }else{

  4870.         if(n == LUMA_DC_BLOCK_INDEX){

  4871.             total_coeff= pred_non_zero_count(h, 0);

  4872.             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);

  4873.             total_coeff= coeff_token>>2;

  4874.         }else{

  4875.             total_coeff= pred_non_zero_count(h, n);

  4876.             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);

  4877.             total_coeff= coeff_token>>2;

  4878.             h->non_zero_count_cache[ scan8[n] ]= total_coeff;

  4879.         }

  4880.     }

  4881. 

  4882.     //FIXME set last_non_zero?

  4883. 

  4884.     if(total_coeff==0)

  4885.         return 0;

  4886.     if(total_coeff<0) {

  4887.         av_log(h->s.avctx, AV_LOG_ERROR, "corrupted macroblock %d %d (total_coeff<0)\n", s->mb_x, s->mb_y);

  4888.         return -1;

  4889.     }

  4890. 

  4891.     trailing_ones= coeff_token&3;

  4892.     tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);

  4893.     assert(total_coeff<=16);

  4894. 

  4895.     for(i=0; i<trailing_ones; i++){

  4896.         level[i]= 1 - 2*get_bits1(gb);

  4897.     }

  4898. 

  4899.     if(i<total_coeff) {

  4900.         int level_code, mask;

  4901.         int suffix_length = total_coeff > 10 && trailing_ones < 3;

  4902.         int prefix= get_level_prefix(gb);

  4903. 

  4904.         //first coefficient has suffix_length equal to 0 or 1

  4905.         if(prefix<14){ //FIXME try to build a large unified VLC table for all this

  4906.             if(suffix_length)

  4907.                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part

  4908.             else

  4909.                 level_code= (prefix<<suffix_length); //part

  4910.         }else if(prefix==14){

  4911.             if(suffix_length)

  4912.                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part

  4913.             else

  4914.                 level_code= prefix + get_bits(gb, 4); //part

  4915.         }else if(prefix==15){

  4916.             level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part

  4917.             if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense

  4918.         }else{

  4919.             av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);

  4920.             return -1;

  4921.         }

  4922. 

  4923.         if(trailing_ones < 3) level_code += 2;

  4924. 

  4925.         suffix_length = 1;

  4926.         if(level_code > 5)

  4927.             suffix_length++;

  4928.         mask= -(level_code&1);

  4929.         level[i]= (((2+level_code)>>1) ^ mask) - mask;

  4930.         i++;

  4931. 

  4932.         //remaining coefficients have suffix_length > 0

  4933.         for(;i<total_coeff;i++) {

  4934.             static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX };

  4935.             prefix = get_level_prefix(gb);

  4936.             if(prefix<15){

  4937.                 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);

  4938.             }else if(prefix==15){

  4939.                 level_code =  (prefix<<suffix_length) + get_bits(gb, 12);

  4940.             }else{

  4941.                 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);

  4942.                 return -1;

  4943.             }

  4944.             mask= -(level_code&1);

  4945.             level[i]= (((2+level_code)>>1) ^ mask) - mask;

  4946.             if(level_code > suffix_limit[suffix_length])

  4947.                 suffix_length++;

  4948.         }

  4949.     }

  4950. 

  4951.     if(total_coeff == max_coeff)

  4952.         zeros_left=0;

  4953.     else{

  4954.         if(n == CHROMA_DC_BLOCK_INDEX)

  4955.             zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);

  4956.         else

  4957.             zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);

  4958.     }

  4959. 

  4960.     coeff_num = zeros_left + total_coeff - 1;

  4961.     j = scantable[coeff_num];

  4962.     if(n > 24){

  4963.         block[j] = level[0];

  4964.         for(i=1;i<total_coeff;i++) {

  4965.             if(zeros_left <= 0)

  4966.                 run_before = 0;

  4967.             else if(zeros_left < 7){

  4968.                 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);

  4969.             }else{

  4970.                 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);

  4971.             }

  4972.             zeros_left -= run_before;

  4973.             coeff_num -= 1 + run_before;

  4974.             j= scantable[ coeff_num ];

  4975. 

  4976.             block[j]= level[i];

  4977.         }

  4978.     }else{

  4979.         block[j] = (level[0] * qmul[j] + 32)>>6;

  4980.         for(i=1;i<total_coeff;i++) {

  4981.             if(zeros_left <= 0)

  4982.                 run_before = 0;

  4983.             else if(zeros_left < 7){

  4984.                 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);

  4985.             }else{

  4986.                 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);

  4987.             }

  4988.             zeros_left -= run_before;

  4989.             coeff_num -= 1 + run_before;

  4990.             j= scantable[ coeff_num ];

  4991. 

  4992.             block[j]= (level[i] * qmul[j] + 32)>>6;

  4993.         }

  4994.     }

  4995. 

  4996.     if(zeros_left<0){

  4997.         av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);

  4998.         return -1;

  4999.     }

  5000. 

  5001.     return 0;

  5002. }

  5003. 

  5004. static void predict_field_decoding_flag(H264Context *h){

  5005.     MpegEncContext * const s = &h->s;

  5006.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  5007.     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)

  5008.                 ? s->current_picture.mb_type[mb_xy-1]

  5009.                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)

  5010.                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]

  5011.                 : 0;

  5012.     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;

  5013. }

  5014. 

  5015. /**

  5016.  * decodes a P_SKIP or B_SKIP macroblock

  5017.  */

  5018. static void decode_mb_skip(H264Context *h){

  5019.     MpegEncContext * const s = &h->s;

  5020.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  5021.     int mb_type=0;

  5022. 

  5023.     memset(h->non_zero_count[mb_xy], 0, 16);

  5024.     memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui

  5025. 

  5026.     if(MB_FIELD)

  5027.         mb_type|= MB_TYPE_INTERLACED;

  5028. 

  5029.     if( h->slice_type == B_TYPE )

  5030.     {

  5031.         // just for fill_caches. pred_direct_motion will set the real mb_type

  5032.         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;

  5033. 

  5034.         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...

  5035.         pred_direct_motion(h, &mb_type);

  5036.         mb_type|= MB_TYPE_SKIP;

  5037.     }

  5038.     else

  5039.     {

  5040.         int mx, my;

  5041.         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;

  5042. 

  5043.         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...

  5044.         pred_pskip_motion(h, &mx, &my);

  5045.         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);

  5046.         fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);

  5047.     }

  5048. 

  5049.     write_back_motion(h, mb_type);

  5050.     s->current_picture.mb_type[mb_xy]= mb_type;

  5051.     s->current_picture.qscale_table[mb_xy]= s->qscale;

  5052.     h->slice_table[ mb_xy ]= h->slice_num;

  5053.     h->prev_mb_skipped= 1;

  5054. }

  5055. 

  5056. /**

  5057.  * decodes a macroblock

  5058.  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed

  5059.  */

  5060. static int decode_mb_cavlc(H264Context *h){

  5061.     MpegEncContext * const s = &h->s;

  5062.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  5063.     int mb_type, partition_count, cbp;

  5064.     int dct8x8_allowed= h->pps.transform_8x8_mode;

  5065. 

  5066.     s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?

  5067. 

  5068.     tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);

  5069.     cbp = 0; /* avoid warning. FIXME: find a solution without slowing

  5070.                 down the code */

  5071.     if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){

  5072.         if(s->mb_skip_run==-1)

  5073.             s->mb_skip_run= get_ue_golomb(&s->gb);

  5074. 

  5075.         if (s->mb_skip_run--) {

  5076.             if(FRAME_MBAFF && (s->mb_y&1) == 0){

  5077.                 if(s->mb_skip_run==0)

  5078.                     h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);

  5079.                 else

  5080.                     predict_field_decoding_flag(h);

  5081.             }

  5082.             decode_mb_skip(h);

  5083.             return 0;

  5084.         }

  5085.     }

  5086.     if(FRAME_MBAFF){

  5087.         if( (s->mb_y&1) == 0 )

  5088.             h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);

  5089.     }else

  5090.         h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);

  5091. 

  5092.     h->prev_mb_skipped= 0;

  5093. 

  5094.     mb_type= get_ue_golomb(&s->gb);

  5095.     if(h->slice_type == B_TYPE){

  5096.         if(mb_type < 23){

  5097.             partition_count= b_mb_type_info[mb_type].partition_count;

  5098.             mb_type=         b_mb_type_info[mb_type].type;

  5099.         }else{

  5100.             mb_type -= 23;

  5101.             goto decode_intra_mb;

  5102.         }

  5103.     }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){

  5104.         if(mb_type < 5){

  5105.             partition_count= p_mb_type_info[mb_type].partition_count;

  5106.             mb_type=         p_mb_type_info[mb_type].type;

  5107.         }else{

  5108.             mb_type -= 5;

  5109.             goto decode_intra_mb;

  5110.         }

  5111.     }else{

  5112.        assert(h->slice_type == I_TYPE);

  5113. decode_intra_mb:

  5114.         if(mb_type > 25){

  5115.             av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);

  5116.             return -1;

  5117.         }

  5118.         partition_count=0;

  5119.         cbp= i_mb_type_info[mb_type].cbp;

  5120.         h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;

  5121.         mb_type= i_mb_type_info[mb_type].type;

  5122.     }

  5123. 

  5124.     if(MB_FIELD)

  5125.         mb_type |= MB_TYPE_INTERLACED;

  5126. 

  5127.     h->slice_table[ mb_xy ]= h->slice_num;

  5128. 

  5129.     if(IS_INTRA_PCM(mb_type)){

  5130.         unsigned int x, y;

  5131. 

  5132.         // we assume these blocks are very rare so we dont optimize it

  5133.         align_get_bits(&s->gb);

  5134. 

  5135.         // The pixels are stored in the same order as levels in h->mb array.

  5136.         for(y=0; y<16; y++){

  5137.             const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);

  5138.             for(x=0; x<16; x++){

  5139.                 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));

  5140.                 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);

  5141.             }

  5142.         }

  5143.         for(y=0; y<8; y++){

  5144.             const int index= 256 + 4*(y&3) + 32*(y>>2);

  5145.             for(x=0; x<8; x++){

  5146.                 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));

  5147.                 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);

  5148.             }

  5149.         }

  5150.         for(y=0; y<8; y++){

  5151.             const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);

  5152.             for(x=0; x<8; x++){

  5153.                 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));

  5154.                 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);

  5155.             }

  5156.         }

  5157. 

  5158.         // In deblocking, the quantizer is 0

  5159.         s->current_picture.qscale_table[mb_xy]= 0;

  5160.         h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);

  5161.         // All coeffs are present

  5162.         memset(h->non_zero_count[mb_xy], 16, 16);

  5163. 

  5164.         s->current_picture.mb_type[mb_xy]= mb_type;

  5165.         return 0;

  5166.     }

  5167. 

  5168.     if(MB_MBAFF){

  5169.         h->ref_count[0] <<= 1;

  5170.         h->ref_count[1] <<= 1;

  5171.     }

  5172. 

  5173.     fill_caches(h, mb_type, 0);

  5174. 

  5175.     //mb_pred

  5176.     if(IS_INTRA(mb_type)){

  5177. //            init_top_left_availability(h);

  5178.             if(IS_INTRA4x4(mb_type)){

  5179.                 int i;

  5180.                 int di = 1;

  5181.                 if(dct8x8_allowed && get_bits1(&s->gb)){

  5182.                     mb_type |= MB_TYPE_8x8DCT;

  5183.                     di = 4;

  5184.                 }

  5185. 

  5186. //                fill_intra4x4_pred_table(h);

  5187.                 for(i=0; i<16; i+=di){

  5188.                     int mode= pred_intra_mode(h, i);

  5189. 

  5190.                     if(!get_bits1(&s->gb)){

  5191.                         const int rem_mode= get_bits(&s->gb, 3);

  5192.                         mode = rem_mode + (rem_mode >= mode);

  5193.                     }

  5194. 

  5195.                     if(di==4)

  5196.                         fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );

  5197.                     else

  5198.                         h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;

  5199.                 }

  5200.                 write_back_intra_pred_mode(h);

  5201.                 if( check_intra4x4_pred_mode(h) < 0)

  5202.                     return -1;

  5203.             }else{

  5204.                 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);

  5205.                 if(h->intra16x16_pred_mode < 0)

  5206.                     return -1;

  5207.             }

  5208.             h->chroma_pred_mode= get_ue_golomb(&s->gb);

  5209. 

  5210.             h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);

  5211.             if(h->chroma_pred_mode < 0)

  5212.                 return -1;

  5213.     }else if(partition_count==4){

  5214.         int i, j, sub_partition_count[4], list, ref[2][4];

  5215. 

  5216.         if(h->slice_type == B_TYPE){

  5217.             for(i=0; i<4; i++){

  5218.                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);

  5219.                 if(h->sub_mb_type[i] >=13){

  5220.                     av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);

  5221.                     return -1;

  5222.                 }

  5223.                 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;

  5224.                 h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;

  5225.             }

  5226.             if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])

  5227.                || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {

  5228.                 pred_direct_motion(h, &mb_type);

  5229.                 h->ref_cache[0][scan8[4]] =

  5230.                 h->ref_cache[1][scan8[4]] =

  5231.                 h->ref_cache[0][scan8[12]] =

  5232.                 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;

  5233.             }

  5234.         }else{

  5235.             assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?

  5236.             for(i=0; i<4; i++){

  5237.                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);

  5238.                 if(h->sub_mb_type[i] >=4){

  5239.                     av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);

  5240.                     return -1;

  5241.                 }

  5242.                 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;

  5243.                 h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;

  5244.             }

  5245.         }

  5246. 

  5247.         for(list=0; list<2; list++){

  5248.             int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];

  5249.             if(ref_count == 0) continue;

  5250.             for(i=0; i<4; i++){

  5251.                 if(IS_DIRECT(h->sub_mb_type[i])) continue;

  5252.                 if(IS_DIR(h->sub_mb_type[i], 0, list)){

  5253.                     ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?

  5254.                 }else{

  5255.                  //FIXME

  5256.                     ref[list][i] = -1;

  5257.                 }

  5258.             }

  5259.         }

  5260. 

  5261.         if(dct8x8_allowed)

  5262.             dct8x8_allowed = get_dct8x8_allowed(h);

  5263. 

  5264.         for(list=0; list<2; list++){

  5265.             const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];

  5266.             if(ref_count == 0) continue;

  5267. 

  5268.             for(i=0; i<4; i++){

  5269.                 if(IS_DIRECT(h->sub_mb_type[i])) {

  5270.                     h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];

  5271.                     continue;

  5272.                 }

  5273.                 h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=

  5274.                 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];

  5275. 

  5276.                 if(IS_DIR(h->sub_mb_type[i], 0, list)){

  5277.                     const int sub_mb_type= h->sub_mb_type[i];

  5278.                     const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;

  5279.                     for(j=0; j<sub_partition_count[i]; j++){

  5280.                         int mx, my;

  5281.                         const int index= 4*i + block_width*j;

  5282.                         int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];

  5283.                         pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);

  5284.                         mx += get_se_golomb(&s->gb);

  5285.                         my += get_se_golomb(&s->gb);

  5286.                         tprintf("final mv:%d %d\n", mx, my);

  5287. 

  5288.                         if(IS_SUB_8X8(sub_mb_type)){

  5289.                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=

  5290.                             mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;

  5291.                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=

  5292.                             mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;

  5293.                         }else if(IS_SUB_8X4(sub_mb_type)){

  5294.                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;

  5295.                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;

  5296.                         }else if(IS_SUB_4X8(sub_mb_type)){

  5297.                             mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;

  5298.                             mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;

  5299.                         }else{

  5300.                             assert(IS_SUB_4X4(sub_mb_type));

  5301.                             mv_cache[ 0 ][0]= mx;

  5302.                             mv_cache[ 0 ][1]= my;

  5303.                         }

  5304.                     }

  5305.                 }else{

  5306.                     uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];

  5307.                     p[0] = p[1]=

  5308.                     p[8] = p[9]= 0;

  5309.                 }

  5310.             }

  5311.         }

  5312.     }else if(IS_DIRECT(mb_type)){

  5313.         pred_direct_motion(h, &mb_type);

  5314.         dct8x8_allowed &= h->sps.direct_8x8_inference_flag;

  5315.     }else{

  5316.         int list, mx, my, i;

  5317.          //FIXME we should set ref_idx_l? to 0 if we use that later ...

  5318.         if(IS_16X16(mb_type)){

  5319.             for(list=0; list<2; list++){

  5320.                 if(h->ref_count[list]>0){

  5321.                     if(IS_DIR(mb_type, 0, list)){

  5322.                         const int val= get_te0_golomb(&s->gb, h->ref_count[list]);

  5323.                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);

  5324.                     }else

  5325.                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1);

  5326.                 }

  5327.             }

  5328.             for(list=0; list<2; list++){

  5329.                 if(IS_DIR(mb_type, 0, list)){

  5330.                     pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);

  5331.                     mx += get_se_golomb(&s->gb);

  5332.                     my += get_se_golomb(&s->gb);

  5333.                     tprintf("final mv:%d %d\n", mx, my);

  5334. 

  5335.                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);

  5336.                 }else

  5337.                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);

  5338.             }

  5339.         }

  5340.         else if(IS_16X8(mb_type)){

  5341.             for(list=0; list<2; list++){

  5342.                 if(h->ref_count[list]>0){

  5343.                     for(i=0; i<2; i++){

  5344.                         if(IS_DIR(mb_type, i, list)){

  5345.                             const int val= get_te0_golomb(&s->gb, h->ref_count[list]);

  5346.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);

  5347.                         }else

  5348.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);

  5349.                     }

  5350.                 }

  5351.             }

  5352.             for(list=0; list<2; list++){

  5353.                 for(i=0; i<2; i++){

  5354.                     if(IS_DIR(mb_type, i, list)){

  5355.                         pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);

  5356.                         mx += get_se_golomb(&s->gb);

  5357.                         my += get_se_golomb(&s->gb);

  5358.                         tprintf("final mv:%d %d\n", mx, my);

  5359. 

  5360.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);

  5361.                     }else

  5362.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);

  5363.                 }

  5364.             }

  5365.         }else{

  5366.             assert(IS_8X16(mb_type));

  5367.             for(list=0; list<2; list++){

  5368.                 if(h->ref_count[list]>0){

  5369.                     for(i=0; i<2; i++){

  5370.                         if(IS_DIR(mb_type, i, list)){ //FIXME optimize

  5371.                             const int val= get_te0_golomb(&s->gb, h->ref_count[list]);

  5372.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);

  5373.                         }else

  5374.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);

  5375.                     }

  5376.                 }

  5377.             }

  5378.             for(list=0; list<2; list++){

  5379.                 for(i=0; i<2; i++){

  5380.                     if(IS_DIR(mb_type, i, list)){

  5381.                         pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);

  5382.                         mx += get_se_golomb(&s->gb);

  5383.                         my += get_se_golomb(&s->gb);

  5384.                         tprintf("final mv:%d %d\n", mx, my);

  5385. 

  5386.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);

  5387.                     }else

  5388.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);

  5389.                 }

  5390.             }

  5391.         }

  5392.     }

  5393. 

  5394.     if(IS_INTER(mb_type))

  5395.         write_back_motion(h, mb_type);

  5396. 

  5397.     if(!IS_INTRA16x16(mb_type)){

  5398.         cbp= get_ue_golomb(&s->gb);

  5399.         if(cbp > 47){

  5400.             av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);

  5401.             return -1;

  5402.         }

  5403. 

  5404.         if(IS_INTRA4x4(mb_type))

  5405.             cbp= golomb_to_intra4x4_cbp[cbp];

  5406.         else

  5407.             cbp= golomb_to_inter_cbp[cbp];

  5408.     }

  5409.     h->cbp = cbp;

  5410. 

  5411.     if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){

  5412.         if(get_bits1(&s->gb))

  5413.             mb_type |= MB_TYPE_8x8DCT;

  5414.     }

  5415.     s->current_picture.mb_type[mb_xy]= mb_type;

  5416. 

  5417.     if(cbp || IS_INTRA16x16(mb_type)){

  5418.         int i8x8, i4x4, chroma_idx;

  5419.         int chroma_qp, dquant;

  5420.         GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;

  5421.         const uint8_t *scan, *scan8x8, *dc_scan;

  5422. 

  5423. //        fill_non_zero_count_cache(h);

  5424. 

  5425.         if(IS_INTERLACED(mb_type)){

  5426.             scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;

  5427.             scan= s->qscale ? h->field_scan : h->field_scan_q0;

  5428.             dc_scan= luma_dc_field_scan;

  5429.         }else{

  5430.             scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;

  5431.             scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;

  5432.             dc_scan= luma_dc_zigzag_scan;

  5433.         }

  5434. 

  5435.         dquant= get_se_golomb(&s->gb);

  5436. 

  5437.         if( dquant > 25 || dquant < -26 ){

  5438.             av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);

  5439.             return -1;

  5440.         }

  5441. 

  5442.         s->qscale += dquant;

  5443.         if(((unsigned)s->qscale) > 51){

  5444.             if(s->qscale<0) s->qscale+= 52;

  5445.             else            s->qscale-= 52;

  5446.         }

  5447. 

  5448.         h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);

  5449.         if(IS_INTRA16x16(mb_type)){

  5450.             if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, h->dequant4_coeff[0][s->qscale], 16) < 0){

  5451.                 return -1; //FIXME continue if partitioned and other return -1 too

  5452.             }

  5453. 

  5454.             assert((cbp&15) == 0 || (cbp&15) == 15);

  5455. 

  5456.             if(cbp&15){

  5457.                 for(i8x8=0; i8x8<4; i8x8++){

  5458.                     for(i4x4=0; i4x4<4; i4x4++){

  5459.                         const int index= i4x4 + 4*i8x8;

  5460.                         if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){

  5461.                             return -1;

  5462.                         }

  5463.                     }

  5464.                 }

  5465.             }else{

  5466.                 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);

  5467.             }

  5468.         }else{

  5469.             for(i8x8=0; i8x8<4; i8x8++){

  5470.                 if(cbp & (1<<i8x8)){

  5471.                     if(IS_8x8DCT(mb_type)){

  5472.                         DCTELEM *buf = &h->mb[64*i8x8];

  5473.                         uint8_t *nnz;

  5474.                         for(i4x4=0; i4x4<4; i4x4++){

  5475.                             if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4,

  5476.                                                 h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 16) <0 )

  5477.                                 return -1;

  5478.                         }

  5479.                         nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];

  5480.                         nnz[0] += nnz[1] + nnz[8] + nnz[9];

  5481.                     }else{

  5482.                         for(i4x4=0; i4x4<4; i4x4++){

  5483.                             const int index= i4x4 + 4*i8x8;

  5484. 

  5485.                             if( decode_residual(h, gb, h->mb + 16*index, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){

  5486.                                 return -1;

  5487.                             }

  5488.                         }

  5489.                     }

  5490.                 }else{

  5491.                     uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];

  5492.                     nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;

  5493.                 }

  5494.             }

  5495.         }

  5496. 

  5497.         if(cbp&0x30){

  5498.             for(chroma_idx=0; chroma_idx<2; chroma_idx++)

  5499.                 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, NULL, 4) < 0){

  5500.                     return -1;

  5501.                 }

  5502.         }

  5503. 

  5504.         if(cbp&0x20){

  5505.             for(chroma_idx=0; chroma_idx<2; chroma_idx++){

  5506.                 for(i4x4=0; i4x4<4; i4x4++){

  5507.                     const int index= 16 + 4*chroma_idx + i4x4;

  5508.                     if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][chroma_qp], 15) < 0){

  5509.                         return -1;

  5510.                     }

  5511.                 }

  5512.             }

  5513.         }else{

  5514.             uint8_t * const nnz= &h->non_zero_count_cache[0];

  5515.             nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =

  5516.             nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;

  5517.         }

  5518.     }else{

  5519.         uint8_t * const nnz= &h->non_zero_count_cache[0];

  5520.         fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);

  5521.         nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =

  5522.         nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;

  5523.     }

  5524.     s->current_picture.qscale_table[mb_xy]= s->qscale;

  5525.     write_back_non_zero_count(h);

  5526. 

  5527.     if(MB_MBAFF){

  5528.         h->ref_count[0] >>= 1;

  5529.         h->ref_count[1] >>= 1;

  5530.     }

  5531. 

  5532.     return 0;

  5533. }

  5534. 

  5535. static int decode_cabac_field_decoding_flag(H264Context *h) {

  5536.     MpegEncContext * const s = &h->s;

  5537.     const int mb_x = s->mb_x;

  5538.     const int mb_y = s->mb_y & ~1;

  5539.     const int mba_xy = mb_x - 1 +  mb_y   *s->mb_stride;

  5540.     const int mbb_xy = mb_x     + (mb_y-2)*s->mb_stride;

  5541. 

  5542.     unsigned int ctx = 0;

  5543. 

  5544.     if( h->slice_table[mba_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) {

  5545.         ctx += 1;

  5546.     }

  5547.     if( h->slice_table[mbb_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) {

  5548.         ctx += 1;

  5549.     }

  5550. 

  5551.     return get_cabac_noinline( &h->cabac, &h->cabac_state[70 + ctx] );

  5552. }

  5553. 

  5554. static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {

  5555.     uint8_t *state= &h->cabac_state[ctx_base];

  5556.     int mb_type;

  5557. 

  5558.     if(intra_slice){

  5559.         MpegEncContext * const s = &h->s;

  5560.         const int mba_xy = h->left_mb_xy[0];

  5561.         const int mbb_xy = h->top_mb_xy;

  5562.         int ctx=0;

  5563.         if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )

  5564.             ctx++;

  5565.         if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )

  5566.             ctx++;

  5567.         if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 )

  5568.             return 0;   /* I4x4 */

  5569.         state += 2;

  5570.     }else{

  5571.         if( get_cabac_noinline( &h->cabac, &state[0] ) == 0 )

  5572.             return 0;   /* I4x4 */

  5573.     }

  5574. 

  5575.     if( get_cabac_terminate( &h->cabac ) )

  5576.         return 25;  /* PCM */

  5577. 

  5578.     mb_type = 1; /* I16x16 */

  5579.     mb_type += 12 * get_cabac_noinline( &h->cabac, &state[1] ); /* cbp_luma != 0 */

  5580.     if( get_cabac_noinline( &h->cabac, &state[2] ) ) /* cbp_chroma */

  5581.         mb_type += 4 + 4 * get_cabac_noinline( &h->cabac, &state[2+intra_slice] );

  5582.     mb_type += 2 * get_cabac_noinline( &h->cabac, &state[3+intra_slice] );

  5583.     mb_type += 1 * get_cabac_noinline( &h->cabac, &state[3+2*intra_slice] );

  5584.     return mb_type;

  5585. }

  5586. 

  5587. static int decode_cabac_mb_type( H264Context *h ) {

  5588.     MpegEncContext * const s = &h->s;

  5589. 

  5590.     if( h->slice_type == I_TYPE ) {

  5591.         return decode_cabac_intra_mb_type(h, 3, 1);

  5592.     } else if( h->slice_type == P_TYPE ) {

  5593.         if( get_cabac_noinline( &h->cabac, &h->cabac_state[14] ) == 0 ) {

  5594.             /* P-type */

  5595.             if( get_cabac_noinline( &h->cabac, &h->cabac_state[15] ) == 0 ) {

  5596.                 /* P_L0_D16x16, P_8x8 */

  5597.                 return 3 * get_cabac_noinline( &h->cabac, &h->cabac_state[16] );

  5598.             } else {

  5599.                 /* P_L0_D8x16, P_L0_D16x8 */

  5600.                 return 2 - get_cabac_noinline( &h->cabac, &h->cabac_state[17] );

  5601.             }

  5602.         } else {

  5603.             return decode_cabac_intra_mb_type(h, 17, 0) + 5;

  5604.         }

  5605.     } else if( h->slice_type == B_TYPE ) {

  5606.         const int mba_xy = h->left_mb_xy[0];

  5607.         const int mbb_xy = h->top_mb_xy;

  5608.         int ctx = 0;

  5609.         int bits;

  5610. 

  5611.         if( h->slice_table[mba_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )

  5612.             ctx++;

  5613.         if( h->slice_table[mbb_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )

  5614.             ctx++;

  5615. 

  5616.         if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) )

  5617.             return 0; /* B_Direct_16x16 */

  5618. 

  5619.         if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+3] ) ) {

  5620.             return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */

  5621.         }

  5622. 

  5623.         bits = get_cabac_noinline( &h->cabac, &h->cabac_state[27+4] ) << 3;

  5624.         bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 2;

  5625.         bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 1;

  5626.         bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] );

  5627.         if( bits < 8 )

  5628.             return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */

  5629.         else if( bits == 13 ) {

  5630.             return decode_cabac_intra_mb_type(h, 32, 0) + 23;

  5631.         } else if( bits == 14 )

  5632.             return 11; /* B_L1_L0_8x16 */

  5633.         else if( bits == 15 )

  5634.             return 22; /* B_8x8 */

  5635. 

  5636.         bits= ( bits<<1 ) | get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] );

  5637.         return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */

  5638.     } else {

  5639.         /* TODO SI/SP frames? */

  5640.         return -1;

  5641.     }

  5642. }

  5643. 

  5644. static int decode_cabac_mb_skip( H264Context *h, int mb_x, int mb_y ) {

  5645.     MpegEncContext * const s = &h->s;

  5646.     int mba_xy, mbb_xy;

  5647.     int ctx = 0;

  5648. 

  5649.     if(FRAME_MBAFF){ //FIXME merge with the stuff in fill_caches?

  5650.         int mb_xy = mb_x + (mb_y&~1)*s->mb_stride;

  5651.         mba_xy = mb_xy - 1;

  5652.         if( (mb_y&1)

  5653.             && h->slice_table[mba_xy] == h->slice_num

  5654.             && MB_FIELD == !!IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) )

  5655.             mba_xy += s->mb_stride;

  5656.         if( MB_FIELD ){

  5657.             mbb_xy = mb_xy - s->mb_stride;

  5658.             if( !(mb_y&1)

  5659.                 && h->slice_table[mbb_xy] == h->slice_num

  5660.                 && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) )

  5661.                 mbb_xy -= s->mb_stride;

  5662.         }else

  5663.             mbb_xy = mb_x + (mb_y-1)*s->mb_stride;

  5664.     }else{

  5665.         int mb_xy = mb_x + mb_y*s->mb_stride;

  5666.         mba_xy = mb_xy - 1;

  5667.         mbb_xy = mb_xy - s->mb_stride;

  5668.     }

  5669. 

  5670.     if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))

  5671.         ctx++;

  5672.     if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))

  5673.         ctx++;

  5674. 

  5675.     if( h->slice_type == B_TYPE )

  5676.         ctx += 13;

  5677.     return get_cabac_noinline( &h->cabac, &h->cabac_state[11+ctx] );

  5678. }

  5679. 

  5680. static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {

  5681.     int mode = 0;

  5682. 

  5683.     if( get_cabac( &h->cabac, &h->cabac_state[68] ) )

  5684.         return pred_mode;

  5685. 

  5686.     mode += 1 * get_cabac( &h->cabac, &h->cabac_state[69] );

  5687.     mode += 2 * get_cabac( &h->cabac, &h->cabac_state[69] );

  5688.     mode += 4 * get_cabac( &h->cabac, &h->cabac_state[69] );

  5689. 

  5690.     if( mode >= pred_mode )

  5691.         return mode + 1;

  5692.     else

  5693.         return mode;

  5694. }

  5695. 

  5696. static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {

  5697.     const int mba_xy = h->left_mb_xy[0];

  5698.     const int mbb_xy = h->top_mb_xy;

  5699. 

  5700.     int ctx = 0;

  5701. 

  5702.     /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */

  5703.     if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )

  5704.         ctx++;

  5705. 

  5706.     if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )

  5707.         ctx++;

  5708. 

  5709.     if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )

  5710.         return 0;

  5711. 

  5712.     if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 )

  5713.         return 1;

  5714.     if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 )

  5715.         return 2;

  5716.     else

  5717.         return 3;

  5718. }

  5719. 

  5720. static const uint8_t block_idx_x[16] = {

  5721.     0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3

  5722. };

  5723. static const uint8_t block_idx_y[16] = {

  5724.     0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3

  5725. };

  5726. static const uint8_t block_idx_xy[4][4] = {

  5727.     { 0, 2, 8,  10},

  5728.     { 1, 3, 9,  11},

  5729.     { 4, 6, 12, 14},

  5730.     { 5, 7, 13, 15}

  5731. };

  5732. 

  5733. static int decode_cabac_mb_cbp_luma( H264Context *h) {

  5734.     int cbp = 0;

  5735.     int cbp_b = -1;

  5736.     int i8x8;

  5737. 

  5738.     if( h->slice_table[h->top_mb_xy] == h->slice_num ) {

  5739.         cbp_b = h->top_cbp;

  5740.         tprintf("cbp_b = top_cbp = %x\n", cbp_b);

  5741.     }

  5742. 

  5743.     for( i8x8 = 0; i8x8 < 4; i8x8++ ) {

  5744.         int cbp_a = -1;

  5745.         int x, y;

  5746.         int ctx = 0;

  5747. 

  5748.         x = block_idx_x[4*i8x8];

  5749.         y = block_idx_y[4*i8x8];

  5750. 

  5751.         if( x > 0 )

  5752.             cbp_a = cbp;

  5753.         else if( h->slice_table[h->left_mb_xy[0]] == h->slice_num ) {

  5754.             cbp_a = h->left_cbp;

  5755.             tprintf("cbp_a = left_cbp = %x\n", cbp_a);

  5756.         }

  5757. 

  5758.         if( y > 0 )

  5759.             cbp_b = cbp;

  5760. 

  5761.         /* No need to test for skip as we put 0 for skip block */

  5762.         /* No need to test for IPCM as we put 1 for IPCM block */

  5763.         if( cbp_a >= 0 ) {

  5764.             int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;

  5765.             if( ((cbp_a >> i8x8a)&0x01) == 0 )

  5766.                 ctx++;

  5767.         }

  5768. 

  5769.         if( cbp_b >= 0 ) {

  5770.             int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;

  5771.             if( ((cbp_b >> i8x8b)&0x01) == 0 )

  5772.                 ctx += 2;

  5773.         }

  5774. 

  5775.         if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {

  5776.             cbp |= 1 << i8x8;

  5777.         }

  5778.     }

  5779.     return cbp;

  5780. }

  5781. static int decode_cabac_mb_cbp_chroma( H264Context *h) {

  5782.     int ctx;

  5783.     int cbp_a, cbp_b;

  5784. 

  5785.     cbp_a = (h->left_cbp>>4)&0x03;

  5786.     cbp_b = (h-> top_cbp>>4)&0x03;

  5787. 

  5788.     ctx = 0;

  5789.     if( cbp_a > 0 ) ctx++;

  5790.     if( cbp_b > 0 ) ctx += 2;

  5791.     if( get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )

  5792.         return 0;

  5793. 

  5794.     ctx = 4;

  5795.     if( cbp_a == 2 ) ctx++;

  5796.     if( cbp_b == 2 ) ctx += 2;

  5797.     return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] );

  5798. }

  5799. static int decode_cabac_mb_dqp( H264Context *h) {

  5800.     MpegEncContext * const s = &h->s;

  5801.     int mbn_xy;

  5802.     int   ctx = 0;

  5803.     int   val = 0;

  5804. 

  5805.     if( s->mb_x > 0 )

  5806.         mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;

  5807.     else

  5808.         mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;

  5809. 

  5810.     if( h->last_qscale_diff != 0 )

  5811.         ctx++;

  5812. 

  5813.     while( get_cabac_noinline( &h->cabac, &h->cabac_state[60 + ctx] ) ) {

  5814.         if( ctx < 2 )

  5815.             ctx = 2;

  5816.         else

  5817.             ctx = 3;

  5818.         val++;

  5819.         if(val > 102) //prevent infinite loop

  5820.             return INT_MIN;

  5821.     }

  5822. 

  5823.     if( val&0x01 )

  5824.         return (val + 1)/2;

  5825.     else

  5826.         return -(val + 1)/2;

  5827. }

  5828. static int decode_cabac_p_mb_sub_type( H264Context *h ) {

  5829.     if( get_cabac( &h->cabac, &h->cabac_state[21] ) )

  5830.         return 0;   /* 8x8 */

  5831.     if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )

  5832.         return 1;   /* 8x4 */

  5833.     if( get_cabac( &h->cabac, &h->cabac_state[23] ) )

  5834.         return 2;   /* 4x8 */

  5835.     return 3;       /* 4x4 */

  5836. }

  5837. static int decode_cabac_b_mb_sub_type( H264Context *h ) {

  5838.     int type;

  5839.     if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )

  5840.         return 0;   /* B_Direct_8x8 */

  5841.     if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )

  5842.         return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */

  5843.     type = 3;

  5844.     if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {

  5845.         if( get_cabac( &h->cabac, &h->cabac_state[39] ) )

  5846.             return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */

  5847.         type += 4;

  5848.     }

  5849.     type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );

  5850.     type +=   get_cabac( &h->cabac, &h->cabac_state[39] );

  5851.     return type;

  5852. }

  5853. 

  5854. static inline int decode_cabac_mb_transform_size( H264Context *h ) {

  5855.     return get_cabac_noinline( &h->cabac, &h->cabac_state[399 + h->neighbor_transform_size] );

  5856. }

  5857. 

  5858. static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {

  5859.     int refa = h->ref_cache[list][scan8[n] - 1];

  5860.     int refb = h->ref_cache[list][scan8[n] - 8];

  5861.     int ref  = 0;

  5862.     int ctx  = 0;

  5863. 

  5864.     if( h->slice_type == B_TYPE) {

  5865.         if( refa > 0 && !h->direct_cache[scan8[n] - 1] )

  5866.             ctx++;

  5867.         if( refb > 0 && !h->direct_cache[scan8[n] - 8] )

  5868.             ctx += 2;

  5869.     } else {

  5870.         if( refa > 0 )

  5871.             ctx++;

  5872.         if( refb > 0 )

  5873.             ctx += 2;

  5874.     }

  5875. 

  5876.     while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {

  5877.         ref++;

  5878.         if( ctx < 4 )

  5879.             ctx = 4;

  5880.         else

  5881.             ctx = 5;

  5882.     }

  5883.     return ref;

  5884. }

  5885. 

  5886. static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {

  5887.     int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +

  5888.                abs( h->mvd_cache[list][scan8[n] - 8][l] );

  5889.     int ctxbase = (l == 0) ? 40 : 47;

  5890.     int ctx, mvd;

  5891. 

  5892.     if( amvd < 3 )

  5893.         ctx = 0;

  5894.     else if( amvd > 32 )

  5895.         ctx = 2;

  5896.     else

  5897.         ctx = 1;

  5898. 

  5899.     if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))

  5900.         return 0;

  5901. 

  5902.     mvd= 1;

  5903.     ctx= 3;

  5904.     while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {

  5905.         mvd++;

  5906.         if( ctx < 6 )

  5907.             ctx++;

  5908.     }

  5909. 

  5910.     if( mvd >= 9 ) {

  5911.         int k = 3;

  5912.         while( get_cabac_bypass( &h->cabac ) ) {

  5913.             mvd += 1 << k;

  5914.             k++;

  5915.         }

  5916.         while( k-- ) {

  5917.             if( get_cabac_bypass( &h->cabac ) )

  5918.                 mvd += 1 << k;

  5919.         }

  5920.     }

  5921.     return get_cabac_bypass_sign( &h->cabac, -mvd );

  5922. }

  5923. 

  5924. static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {

  5925.     int nza, nzb;

  5926.     int ctx = 0;

  5927. 

  5928.     if( cat == 0 ) {

  5929.         nza = h->left_cbp&0x100;

  5930.         nzb = h-> top_cbp&0x100;

  5931.     } else if( cat == 1 || cat == 2 ) {

  5932.         nza = h->non_zero_count_cache[scan8[idx] - 1];

  5933.         nzb = h->non_zero_count_cache[scan8[idx] - 8];

  5934.     } else if( cat == 3 ) {

  5935.         nza = (h->left_cbp>>(6+idx))&0x01;

  5936.         nzb = (h-> top_cbp>>(6+idx))&0x01;

  5937.     } else {

  5938.         assert(cat == 4);

  5939.         nza = h->non_zero_count_cache[scan8[16+idx] - 1];

  5940.         nzb = h->non_zero_count_cache[scan8[16+idx] - 8];

  5941.     }

  5942. 

  5943.     if( nza > 0 )

  5944.         ctx++;

  5945. 

  5946.     if( nzb > 0 )

  5947.         ctx += 2;

  5948. 

  5949.     return ctx + 4 * cat;

  5950. }

  5951. 

  5952. static const __attribute((used)) uint8_t last_coeff_flag_offset_8x8[63] = {

  5953.     0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,

  5954.     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  5955.     3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,

  5956.     5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8

  5957. };

  5958. 

  5959. static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff) {

  5960.     const int mb_xy  = h->s.mb_x + h->s.mb_y*h->s.mb_stride;

  5961.     static const int significant_coeff_flag_offset[2][6] = {

  5962.       { 105+0, 105+15, 105+29, 105+44, 105+47, 402 },

  5963.       { 277+0, 277+15, 277+29, 277+44, 277+47, 436 }

  5964.     };

  5965.     static const int last_coeff_flag_offset[2][6] = {

  5966.       { 166+0, 166+15, 166+29, 166+44, 166+47, 417 },

  5967.       { 338+0, 338+15, 338+29, 338+44, 338+47, 451 }

  5968.     };

  5969.     static const int coeff_abs_level_m1_offset[6] = {

  5970.         227+0, 227+10, 227+20, 227+30, 227+39, 426

  5971.     };

  5972.     static const uint8_t significant_coeff_flag_offset_8x8[2][63] = {

  5973.       { 0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5,

  5974.         4, 4, 4, 4, 3, 3, 6, 7, 7, 7, 8, 9,10, 9, 8, 7,

  5975.         7, 6,11,12,13,11, 6, 7, 8, 9,14,10, 9, 8, 6,11,

  5976.        12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12 },

  5977.       { 0, 1, 1, 2, 2, 3, 3, 4, 5, 6, 7, 7, 7, 8, 4, 5,

  5978.         6, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,11,12,11,

  5979.         9, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,13,13, 9,

  5980.         9,10,10, 8,13,13, 9, 9,10,10,14,14,14,14,14 }

  5981.     };

  5982. 

  5983.     int index[64];

  5984. 

  5985.     int last;

  5986.     int coeff_count = 0;

  5987. 

  5988.     int abslevel1 = 1;

  5989.     int abslevelgt1 = 0;

  5990. 

  5991.     uint8_t *significant_coeff_ctx_base;

  5992.     uint8_t *last_coeff_ctx_base;

  5993.     uint8_t *abs_level_m1_ctx_base;

  5994. 

  5995. #ifndef ARCH_X86

  5996. #define CABAC_ON_STACK

  5997. #endif

  5998. #ifdef CABAC_ON_STACK

  5999. #define CC &cc

  6000.     CABACContext cc;

  6001.     cc.range     = h->cabac.range;

  6002.     cc.low       = h->cabac.low;

  6003.     cc.bytestream= h->cabac.bytestream;

  6004. #else

  6005. #define CC &h->cabac

  6006. #endif

  6007. 

  6008. 

  6009.     /* cat: 0-> DC 16x16  n = 0

  6010.      *      1-> AC 16x16  n = luma4x4idx

  6011.      *      2-> Luma4x4   n = luma4x4idx

  6012.      *      3-> DC Chroma n = iCbCr

  6013.      *      4-> AC Chroma n = 4 * iCbCr + chroma4x4idx

  6014.      *      5-> Luma8x8   n = 4 * luma8x8idx

  6015.      */

  6016. 

  6017.     /* read coded block flag */

  6018.     if( cat != 5 ) {

  6019.         if( get_cabac( CC, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {

  6020.             if( cat == 1 || cat == 2 )

  6021.                 h->non_zero_count_cache[scan8[n]] = 0;

  6022.             else if( cat == 4 )

  6023.                 h->non_zero_count_cache[scan8[16+n]] = 0;

  6024. #ifdef CABAC_ON_STACK

  6025.             h->cabac.range     = cc.range     ;

  6026.             h->cabac.low       = cc.low       ;

  6027.             h->cabac.bytestream= cc.bytestream;

  6028. #endif

  6029.             return 0;

  6030.         }

  6031.     }

  6032. 

  6033.     significant_coeff_ctx_base = h->cabac_state

  6034.         + significant_coeff_flag_offset[MB_FIELD][cat];

  6035.     last_coeff_ctx_base = h->cabac_state

  6036.         + last_coeff_flag_offset[MB_FIELD][cat];

  6037.     abs_level_m1_ctx_base = h->cabac_state

  6038.         + coeff_abs_level_m1_offset[cat];

  6039. 

  6040.     if( cat == 5 ) {

  6041. #define DECODE_SIGNIFICANCE( coefs, sig_off, last_off ) \

  6042.         for(last= 0; last < coefs; last++) { \

  6043.             uint8_t *sig_ctx = significant_coeff_ctx_base + sig_off; \

  6044.             if( get_cabac( CC, sig_ctx )) { \

  6045.                 uint8_t *last_ctx = last_coeff_ctx_base + last_off; \

  6046.                 index[coeff_count++] = last; \

  6047.                 if( get_cabac( CC, last_ctx ) ) { \

  6048.                     last= max_coeff; \

  6049.                     break; \

  6050.                 } \

  6051.             } \

  6052.         }\

  6053.         if( last == max_coeff -1 ) {\

  6054.             index[coeff_count++] = last;\

  6055.         }

  6056.         const uint8_t *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD];

  6057. #if defined(ARCH_X86) && !(defined(PIC) && defined(__GNUC__))

  6058.         coeff_count= decode_significance_8x8_x86(CC, significant_coeff_ctx_base, index, sig_off);

  6059.     } else {

  6060.         coeff_count= decode_significance_x86(CC, max_coeff, significant_coeff_ctx_base, index);

  6061. #else

  6062.         DECODE_SIGNIFICANCE( 63, sig_off[last], last_coeff_flag_offset_8x8[last] );

  6063.     } else {

  6064.         DECODE_SIGNIFICANCE( max_coeff - 1, last, last );

  6065. #endif

  6066.     }

  6067.     assert(coeff_count > 0);

  6068. 

  6069.     if( cat == 0 )

  6070.         h->cbp_table[mb_xy] |= 0x100;

  6071.     else if( cat == 1 || cat == 2 )

  6072.         h->non_zero_count_cache[scan8[n]] = coeff_count;

  6073.     else if( cat == 3 )

  6074.         h->cbp_table[mb_xy] |= 0x40 << n;

  6075.     else if( cat == 4 )

  6076.         h->non_zero_count_cache[scan8[16+n]] = coeff_count;

  6077.     else {

  6078.         assert( cat == 5 );

  6079.         fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, coeff_count, 1);

  6080.     }

  6081. 

  6082.     for( coeff_count--; coeff_count >= 0; coeff_count-- ) {

  6083.         uint8_t *ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + abs_level_m1_ctx_base;

  6084.         int j= scantable[index[coeff_count]];

  6085. 

  6086.         if( get_cabac( CC, ctx ) == 0 ) {

  6087.             if( !qmul ) {

  6088.                 block[j] = get_cabac_bypass_sign( CC, -1);

  6089.             }else{

  6090.                 block[j] = (get_cabac_bypass_sign( CC, -qmul[j]) + 32) >> 6;;

  6091.             }

  6092. 

  6093.             abslevel1++;

  6094.         } else {

  6095.             int coeff_abs = 2;

  6096.             ctx = 5 + FFMIN( 4, abslevelgt1 ) + abs_level_m1_ctx_base;

  6097.             while( coeff_abs < 15 && get_cabac( CC, ctx ) ) {

  6098.                 coeff_abs++;

  6099.             }

  6100. 

  6101.             if( coeff_abs >= 15 ) {

  6102.                 int j = 0;

  6103.                 while( get_cabac_bypass( CC ) ) {

  6104.                     j++;

  6105.                 }

  6106. 

  6107.                 coeff_abs=1;

  6108.                 while( j-- ) {

  6109.                     coeff_abs += coeff_abs + get_cabac_bypass( CC );

  6110.                 }

  6111.                 coeff_abs+= 14;

  6112.             }

  6113. 

  6114.             if( !qmul ) {

  6115.                 if( get_cabac_bypass( CC ) ) block[j] = -coeff_abs;

  6116.                 else                                block[j] =  coeff_abs;

  6117.             }else{

  6118.                 if( get_cabac_bypass( CC ) ) block[j] = (-coeff_abs * qmul[j] + 32) >> 6;

  6119.                 else                                block[j] = ( coeff_abs * qmul[j] + 32) >> 6;

  6120.             }

  6121. 

  6122.             abslevelgt1++;

  6123.         }

  6124.     }

  6125. #ifdef CABAC_ON_STACK

  6126.             h->cabac.range     = cc.range     ;

  6127.             h->cabac.low       = cc.low       ;

  6128.             h->cabac.bytestream= cc.bytestream;

  6129. #endif

  6130.     return 0;

  6131. }

  6132. 

  6133. static void inline compute_mb_neighbors(H264Context *h)

  6134. {

  6135.     MpegEncContext * const s = &h->s;

  6136.     const int mb_xy  = s->mb_x + s->mb_y*s->mb_stride;

  6137.     h->top_mb_xy     = mb_xy - s->mb_stride;

  6138.     h->left_mb_xy[0] = mb_xy - 1;

  6139.     if(FRAME_MBAFF){

  6140.         const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;

  6141.         const int top_pair_xy      = pair_xy     - s->mb_stride;

  6142.         const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);

  6143.         const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);

  6144.         const int curr_mb_frame_flag = !MB_FIELD;

  6145.         const int bottom = (s->mb_y & 1);

  6146.         if (bottom

  6147.                 ? !curr_mb_frame_flag // bottom macroblock

  6148.                 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock

  6149.                 ) {

  6150.             h->top_mb_xy -= s->mb_stride;

  6151.         }

  6152.         if (left_mb_frame_flag != curr_mb_frame_flag) {

  6153.             h->left_mb_xy[0] = pair_xy - 1;

  6154.         }

  6155.     }

  6156.     return;

  6157. }

  6158. 

  6159. /**

  6160.  * decodes a macroblock

  6161.  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed

  6162.  */

  6163. static int decode_mb_cabac(H264Context *h) {

  6164.     MpegEncContext * const s = &h->s;

  6165.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  6166.     int mb_type, partition_count, cbp = 0;

  6167.     int dct8x8_allowed= h->pps.transform_8x8_mode;

  6168. 

  6169.     s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?)

  6170. 

  6171.     tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);

  6172.     if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {

  6173.         int skip;

  6174.         /* a skipped mb needs the aff flag from the following mb */

  6175.         if( FRAME_MBAFF && s->mb_x==0 && (s->mb_y&1)==0 )

  6176.             predict_field_decoding_flag(h);

  6177.         if( FRAME_MBAFF && (s->mb_y&1)==1 && h->prev_mb_skipped )

  6178.             skip = h->next_mb_skipped;

  6179.         else

  6180.             skip = decode_cabac_mb_skip( h, s->mb_x, s->mb_y );

  6181.         /* read skip flags */

  6182.         if( skip ) {

  6183.             if( FRAME_MBAFF && (s->mb_y&1)==0 ){

  6184.                 s->current_picture.mb_type[mb_xy] = MB_TYPE_SKIP;

  6185.                 h->next_mb_skipped = decode_cabac_mb_skip( h, s->mb_x, s->mb_y+1 );

  6186.                 if(h->next_mb_skipped)

  6187.                     predict_field_decoding_flag(h);

  6188.                 else

  6189.                     h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);

  6190.             }

  6191. 

  6192.             decode_mb_skip(h);

  6193. 

  6194.             h->cbp_table[mb_xy] = 0;

  6195.             h->chroma_pred_mode_table[mb_xy] = 0;

  6196.             h->last_qscale_diff = 0;

  6197. 

  6198.             return 0;

  6199. 

  6200.         }

  6201.     }

  6202.     if(FRAME_MBAFF){

  6203.         if( (s->mb_y&1) == 0 )

  6204.             h->mb_mbaff =

  6205.             h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);

  6206.     }else

  6207.         h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);

  6208. 

  6209.     h->prev_mb_skipped = 0;

  6210. 

  6211.     compute_mb_neighbors(h);

  6212.     if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {

  6213.         av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );

  6214.         return -1;

  6215.     }

  6216. 

  6217.     if( h->slice_type == B_TYPE ) {

  6218.         if( mb_type < 23 ){

  6219.             partition_count= b_mb_type_info[mb_type].partition_count;

  6220.             mb_type=         b_mb_type_info[mb_type].type;

  6221.         }else{

  6222.             mb_type -= 23;

  6223.             goto decode_intra_mb;

  6224.         }

  6225.     } else if( h->slice_type == P_TYPE ) {

  6226.         if( mb_type < 5) {

  6227.             partition_count= p_mb_type_info[mb_type].partition_count;

  6228.             mb_type=         p_mb_type_info[mb_type].type;

  6229.         } else {

  6230.             mb_type -= 5;

  6231.             goto decode_intra_mb;

  6232.         }

  6233.     } else {

  6234.        assert(h->slice_type == I_TYPE);

  6235. decode_intra_mb:

  6236.         partition_count = 0;

  6237.         cbp= i_mb_type_info[mb_type].cbp;

  6238.         h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;

  6239.         mb_type= i_mb_type_info[mb_type].type;

  6240.     }

  6241.     if(MB_FIELD)

  6242.         mb_type |= MB_TYPE_INTERLACED;

  6243. 

  6244.     h->slice_table[ mb_xy ]= h->slice_num;

  6245. 

  6246.     if(IS_INTRA_PCM(mb_type)) {

  6247.         const uint8_t *ptr;

  6248.         unsigned int x, y;

  6249. 

  6250.         // We assume these blocks are very rare so we dont optimize it.

  6251.         // FIXME The two following lines get the bitstream position in the cabac

  6252.         // decode, I think it should be done by a function in cabac.h (or cabac.c).

  6253.         ptr= h->cabac.bytestream;

  6254.         if (h->cabac.low&0x1) ptr-=CABAC_BITS/8;

  6255. 

  6256.         // The pixels are stored in the same order as levels in h->mb array.

  6257.         for(y=0; y<16; y++){

  6258.             const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);

  6259.             for(x=0; x<16; x++){

  6260.                 tprintf("LUMA ICPM LEVEL (%3d)\n", *ptr);

  6261.                 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++;

  6262.             }

  6263.         }

  6264.         for(y=0; y<8; y++){

  6265.             const int index= 256 + 4*(y&3) + 32*(y>>2);

  6266.             for(x=0; x<8; x++){

  6267.                 tprintf("CHROMA U ICPM LEVEL (%3d)\n", *ptr);

  6268.                 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;

  6269.             }

  6270.         }

  6271.         for(y=0; y<8; y++){

  6272.             const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);

  6273.             for(x=0; x<8; x++){

  6274.                 tprintf("CHROMA V ICPM LEVEL (%3d)\n", *ptr);

  6275.                 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;

  6276.             }

  6277.         }

  6278. 

  6279.         ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);

  6280. 

  6281.         // All blocks are present

  6282.         h->cbp_table[mb_xy] = 0x1ef;

  6283.         h->chroma_pred_mode_table[mb_xy] = 0;

  6284.         // In deblocking, the quantizer is 0

  6285.         s->current_picture.qscale_table[mb_xy]= 0;

  6286.         h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);

  6287.         // All coeffs are present

  6288.         memset(h->non_zero_count[mb_xy], 16, 16);

  6289.         s->current_picture.mb_type[mb_xy]= mb_type;

  6290.         return 0;

  6291.     }

  6292. 

  6293.     if(MB_MBAFF){

  6294.         h->ref_count[0] <<= 1;

  6295.         h->ref_count[1] <<= 1;

  6296.     }

  6297. 

  6298.     fill_caches(h, mb_type, 0);

  6299. 

  6300.     if( IS_INTRA( mb_type ) ) {

  6301.         int i;

  6302.         if( IS_INTRA4x4( mb_type ) ) {

  6303.             if( dct8x8_allowed && decode_cabac_mb_transform_size( h ) ) {

  6304.                 mb_type |= MB_TYPE_8x8DCT;

  6305.                 for( i = 0; i < 16; i+=4 ) {

  6306.                     int pred = pred_intra_mode( h, i );

  6307.                     int mode = decode_cabac_mb_intra4x4_pred_mode( h, pred );

  6308.                     fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );

  6309.                 }

  6310.             } else {

  6311.                 for( i = 0; i < 16; i++ ) {

  6312.                     int pred = pred_intra_mode( h, i );

  6313.                     h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );

  6314. 

  6315.                 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );

  6316.                 }

  6317.             }

  6318.             write_back_intra_pred_mode(h);

  6319.             if( check_intra4x4_pred_mode(h) < 0 ) return -1;

  6320.         } else {

  6321.             h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );

  6322.             if( h->intra16x16_pred_mode < 0 ) return -1;

  6323.         }

  6324.         h->chroma_pred_mode_table[mb_xy] =

  6325.             h->chroma_pred_mode          = decode_cabac_mb_chroma_pre_mode( h );

  6326. 

  6327.         h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );

  6328.         if( h->chroma_pred_mode < 0 ) return -1;

  6329.     } else if( partition_count == 4 ) {

  6330.         int i, j, sub_partition_count[4], list, ref[2][4];

  6331. 

  6332.         if( h->slice_type == B_TYPE ) {

  6333.             for( i = 0; i < 4; i++ ) {

  6334.                 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );

  6335.                 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;

  6336.                 h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;

  6337.             }

  6338.             if( IS_DIRECT(h->sub_mb_type[0] | h->sub_mb_type[1] |

  6339.                           h->sub_mb_type[2] | h->sub_mb_type[3]) ) {

  6340.                 pred_direct_motion(h, &mb_type);

  6341.                 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {

  6342.                     for( i = 0; i < 4; i++ )

  6343.                         if( IS_DIRECT(h->sub_mb_type[i]) )

  6344.                             fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );

  6345.                 }

  6346.             }

  6347.         } else {

  6348.             for( i = 0; i < 4; i++ ) {

  6349.                 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );

  6350.                 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;

  6351.                 h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;

  6352.             }

  6353.         }

  6354. 

  6355.         for( list = 0; list < 2; list++ ) {

  6356.             if( h->ref_count[list] > 0 ) {

  6357.                 for( i = 0; i < 4; i++ ) {

  6358.                     if(IS_DIRECT(h->sub_mb_type[i])) continue;

  6359.                     if(IS_DIR(h->sub_mb_type[i], 0, list)){

  6360.                         if( h->ref_count[list] > 1 )

  6361.                             ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );

  6362.                         else

  6363.                             ref[list][i] = 0;

  6364.                     } else {

  6365.                         ref[list][i] = -1;

  6366.                     }

  6367.                                                        h->ref_cache[list][ scan8[4*i]+1 ]=

  6368.                     h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];

  6369.                 }

  6370.             }

  6371.         }

  6372. 

  6373.         if(dct8x8_allowed)

  6374.             dct8x8_allowed = get_dct8x8_allowed(h);

  6375. 

  6376.         for(list=0; list<2; list++){

  6377.             for(i=0; i<4; i++){

  6378.                 if(IS_DIRECT(h->sub_mb_type[i])){

  6379.                     fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);

  6380.                     continue;

  6381.                 }

  6382.                 h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ];

  6383. 

  6384.                 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){

  6385.                     const int sub_mb_type= h->sub_mb_type[i];

  6386.                     const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;

  6387.                     for(j=0; j<sub_partition_count[i]; j++){

  6388.                         int mpx, mpy;

  6389.                         int mx, my;

  6390.                         const int index= 4*i + block_width*j;

  6391.                         int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];

  6392.                         int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];

  6393.                         pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);

  6394. 

  6395.                         mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );

  6396.                         my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );

  6397.                         tprintf("final mv:%d %d\n", mx, my);

  6398. 

  6399.                         if(IS_SUB_8X8(sub_mb_type)){

  6400.                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=

  6401.                             mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;

  6402.                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=

  6403.                             mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;

  6404. 

  6405.                             mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=

  6406.                             mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;

  6407.                             mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=

  6408.                             mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;

  6409.                         }else if(IS_SUB_8X4(sub_mb_type)){

  6410.                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;

  6411.                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;

  6412. 

  6413.                             mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;

  6414.                             mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;

  6415.                         }else if(IS_SUB_4X8(sub_mb_type)){

  6416.                             mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;

  6417.                             mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;

  6418. 

  6419.                             mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;

  6420.                             mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;

  6421.                         }else{

  6422.                             assert(IS_SUB_4X4(sub_mb_type));

  6423.                             mv_cache[ 0 ][0]= mx;

  6424.                             mv_cache[ 0 ][1]= my;

  6425. 

  6426.                             mvd_cache[ 0 ][0]= mx - mpx;

  6427.                             mvd_cache[ 0 ][1]= my - mpy;

  6428.                         }

  6429.                     }

  6430.                 }else{

  6431.                     uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];

  6432.                     uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];

  6433.                     p[0] = p[1] = p[8] = p[9] = 0;

  6434.                     pd[0]= pd[1]= pd[8]= pd[9]= 0;

  6435.                 }

  6436.             }

  6437.         }

  6438.     } else if( IS_DIRECT(mb_type) ) {

  6439.         pred_direct_motion(h, &mb_type);

  6440.         fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);

  6441.         fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);

  6442.         dct8x8_allowed &= h->sps.direct_8x8_inference_flag;

  6443.     } else {

  6444.         int list, mx, my, i, mpx, mpy;

  6445.         if(IS_16X16(mb_type)){

  6446.             for(list=0; list<2; list++){

  6447.                 if(IS_DIR(mb_type, 0, list)){

  6448.                     if(h->ref_count[list] > 0 ){

  6449.                         const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;

  6450.                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);

  6451.                     }

  6452.                 }else

  6453.                     fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);

  6454.             }

  6455.             for(list=0; list<2; list++){

  6456.                 if(IS_DIR(mb_type, 0, list)){

  6457.                     pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);

  6458. 

  6459.                     mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );

  6460.                     my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );

  6461.                     tprintf("final mv:%d %d\n", mx, my);

  6462. 

  6463.                     fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);

  6464.                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);

  6465.                 }else

  6466.                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);

  6467.             }

  6468.         }

  6469.         else if(IS_16X8(mb_type)){

  6470.             for(list=0; list<2; list++){

  6471.                 if(h->ref_count[list]>0){

  6472.                     for(i=0; i<2; i++){

  6473.                         if(IS_DIR(mb_type, i, list)){

  6474.                             const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;

  6475.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);

  6476.                         }else

  6477.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);

  6478.                     }

  6479.                 }

  6480.             }

  6481.             for(list=0; list<2; list++){

  6482.                 for(i=0; i<2; i++){

  6483.                     if(IS_DIR(mb_type, i, list)){

  6484.                         pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);

  6485.                         mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );

  6486.                         my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );

  6487.                         tprintf("final mv:%d %d\n", mx, my);

  6488. 

  6489.                         fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);

  6490.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);

  6491.                     }else{

  6492.                         fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);

  6493.                         fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);

  6494.                     }

  6495.                 }

  6496.             }

  6497.         }else{

  6498.             assert(IS_8X16(mb_type));

  6499.             for(list=0; list<2; list++){

  6500.                 if(h->ref_count[list]>0){

  6501.                     for(i=0; i<2; i++){

  6502.                         if(IS_DIR(mb_type, i, list)){ //FIXME optimize

  6503.                             const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;

  6504.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);

  6505.                         }else

  6506.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);

  6507.                     }

  6508.                 }

  6509.             }

  6510.             for(list=0; list<2; list++){

  6511.                 for(i=0; i<2; i++){

  6512.                     if(IS_DIR(mb_type, i, list)){

  6513.                         pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);

  6514.                         mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );

  6515.                         my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );

  6516. 

  6517.                         tprintf("final mv:%d %d\n", mx, my);

  6518.                         fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);

  6519.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);

  6520.                     }else{

  6521.                         fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);

  6522.                         fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);

  6523.                     }

  6524.                 }

  6525.             }

  6526.         }

  6527.     }

  6528. 

  6529.    if( IS_INTER( mb_type ) ) {

  6530.         h->chroma_pred_mode_table[mb_xy] = 0;

  6531.         write_back_motion( h, mb_type );

  6532.    }

  6533. 

  6534.     if( !IS_INTRA16x16( mb_type ) ) {

  6535.         cbp  = decode_cabac_mb_cbp_luma( h );

  6536.         cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;

  6537.     }

  6538. 

  6539.     h->cbp_table[mb_xy] = h->cbp = cbp;

  6540. 

  6541.     if( dct8x8_allowed && (cbp&15) && !IS_INTRA( mb_type ) ) {

  6542.         if( decode_cabac_mb_transform_size( h ) )

  6543.             mb_type |= MB_TYPE_8x8DCT;

  6544.     }

  6545.     s->current_picture.mb_type[mb_xy]= mb_type;

  6546. 

  6547.     if( cbp || IS_INTRA16x16( mb_type ) ) {

  6548.         const uint8_t *scan, *scan8x8, *dc_scan;

  6549.         int dqp;

  6550. 

  6551.         if(IS_INTERLACED(mb_type)){

  6552.             scan8x8= s->qscale ? h->field_scan8x8 : h->field_scan8x8_q0;

  6553.             scan= s->qscale ? h->field_scan : h->field_scan_q0;

  6554.             dc_scan= luma_dc_field_scan;

  6555.         }else{

  6556.             scan8x8= s->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0;

  6557.             scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;

  6558.             dc_scan= luma_dc_zigzag_scan;

  6559.         }

  6560. 

  6561.         h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );

  6562.         if( dqp == INT_MIN ){

  6563.             av_log(h->s.avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", s->mb_x, s->mb_y);

  6564.             return -1;

  6565.         }

  6566.         s->qscale += dqp;

  6567.         if(((unsigned)s->qscale) > 51){

  6568.             if(s->qscale<0) s->qscale+= 52;

  6569.             else            s->qscale-= 52;

  6570.         }

  6571.         h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);

  6572. 

  6573.         if( IS_INTRA16x16( mb_type ) ) {

  6574.             int i;

  6575.             //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );

  6576.             if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, NULL, 16) < 0)

  6577.                 return -1;

  6578.             if( cbp&15 ) {

  6579.                 for( i = 0; i < 16; i++ ) {

  6580.                     //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );

  6581.                     if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 )

  6582.                         return -1;

  6583.                 }

  6584.             } else {

  6585.                 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);

  6586.             }

  6587.         } else {

  6588.             int i8x8, i4x4;

  6589.             for( i8x8 = 0; i8x8 < 4; i8x8++ ) {

  6590.                 if( cbp & (1<<i8x8) ) {

  6591.                     if( IS_8x8DCT(mb_type) ) {

  6592.                         if( decode_cabac_residual(h, h->mb + 64*i8x8, 5, 4*i8x8,

  6593.                             scan8x8, h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 64) < 0 )

  6594.                             return -1;

  6595.                     } else

  6596.                     for( i4x4 = 0; i4x4 < 4; i4x4++ ) {

  6597.                         const int index = 4*i8x8 + i4x4;

  6598.                         //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );

  6599. //START_TIMER

  6600.                         if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) < 0 )

  6601.                             return -1;

  6602. //STOP_TIMER("decode_residual")

  6603.                     }

  6604.                 } else {

  6605.                     uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];

  6606.                     nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;

  6607.                 }

  6608.             }

  6609.         }

  6610. 

  6611.         if( cbp&0x30 ){

  6612.             int c;

  6613.             for( c = 0; c < 2; c++ ) {

  6614.                 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );

  6615.                 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, NULL, 4) < 0)

  6616.                     return -1;

  6617.             }

  6618.         }

  6619. 

  6620.         if( cbp&0x20 ) {

  6621.             int c, i;

  6622.             for( c = 0; c < 2; c++ ) {

  6623.                 for( i = 0; i < 4; i++ ) {

  6624.                     const int index = 16 + 4 * c + i;

  6625.                     //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );

  6626.                     if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->dequant4_coeff[c+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp], 15) < 0)

  6627.                         return -1;

  6628.                 }

  6629.             }

  6630.         } else {

  6631.             uint8_t * const nnz= &h->non_zero_count_cache[0];

  6632.             nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =

  6633.             nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;

  6634.         }

  6635.     } else {

  6636.         uint8_t * const nnz= &h->non_zero_count_cache[0];

  6637.         fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);

  6638.         nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =

  6639.         nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;

  6640.         h->last_qscale_diff = 0;

  6641.     }

  6642. 

  6643.     s->current_picture.qscale_table[mb_xy]= s->qscale;

  6644.     write_back_non_zero_count(h);

  6645. 

  6646.     if(MB_MBAFF){

  6647.         h->ref_count[0] >>= 1;

  6648.         h->ref_count[1] >>= 1;

  6649.     }

  6650. 

  6651.     return 0;

  6652. }

  6653. 

  6654. 

  6655. static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {

  6656.     int i, d;

  6657.     const int index_a = qp + h->slice_alpha_c0_offset;

  6658.     const int alpha = (alpha_table+52)[index_a];

  6659.     const int beta  = (beta_table+52)[qp + h->slice_beta_offset];

  6660. 

  6661.     if( bS[0] < 4 ) {

  6662.         int8_t tc[4];

  6663.         for(i=0; i<4; i++)

  6664.             tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] : -1;

  6665.         h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);

  6666.     } else {

  6667.         /* 16px edge length, because bS=4 is triggered by being at

  6668.          * the edge of an intra MB, so all 4 bS are the same */

  6669.             for( d = 0; d < 16; d++ ) {

  6670.                 const int p0 = pix[-1];

  6671.                 const int p1 = pix[-2];

  6672.                 const int p2 = pix[-3];

  6673. 

  6674.                 const int q0 = pix[0];

  6675.                 const int q1 = pix[1];

  6676.                 const int q2 = pix[2];

  6677. 

  6678.                 if( FFABS( p0 - q0 ) < alpha &&

  6679.                     FFABS( p1 - p0 ) < beta &&

  6680.                     FFABS( q1 - q0 ) < beta ) {

  6681. 

  6682.                     if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){

  6683.                         if( FFABS( p2 - p0 ) < beta)

  6684.                         {

  6685.                             const int p3 = pix[-4];

  6686.                             /* p0', p1', p2' */

  6687.                             pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;

  6688.                             pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;

  6689.                             pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;

  6690.                         } else {

  6691.                             /* p0' */

  6692.                             pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6693.                         }

  6694.                         if( FFABS( q2 - q0 ) < beta)

  6695.                         {

  6696.                             const int q3 = pix[3];

  6697.                             /* q0', q1', q2' */

  6698.                             pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;

  6699.                             pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;

  6700.                             pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;

  6701.                         } else {

  6702.                             /* q0' */

  6703.                             pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6704.                         }

  6705.                     }else{

  6706.                         /* p0', q0' */

  6707.                         pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6708.                         pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6709.                     }

  6710.                     tprintf("filter_mb_edgev i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, p2, p1, p0, q0, q1, q2, pix[-2], pix[-1], pix[0], pix[1]);

  6711.                 }

  6712.                 pix += stride;

  6713.             }

  6714.     }

  6715. }

  6716. static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {

  6717.     int i;

  6718.     const int index_a = qp + h->slice_alpha_c0_offset;

  6719.     const int alpha = (alpha_table+52)[index_a];

  6720.     const int beta  = (beta_table+52)[qp + h->slice_beta_offset];

  6721. 

  6722.     if( bS[0] < 4 ) {

  6723.         int8_t tc[4];

  6724.         for(i=0; i<4; i++)

  6725.             tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] + 1 : 0;

  6726.         h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);

  6727.     } else {

  6728.         h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);

  6729.     }

  6730. }

  6731. 

  6732. static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) {

  6733.     int i;

  6734.     for( i = 0; i < 16; i++, pix += stride) {

  6735.         int index_a;

  6736.         int alpha;

  6737.         int beta;

  6738. 

  6739.         int qp_index;

  6740.         int bS_index = (i >> 1);

  6741.         if (!MB_FIELD) {

  6742.             bS_index &= ~1;

  6743.             bS_index |= (i & 1);

  6744.         }

  6745. 

  6746.         if( bS[bS_index] == 0 ) {

  6747.             continue;

  6748.         }

  6749. 

  6750.         qp_index = MB_FIELD ? (i >> 3) : (i & 1);

  6751.         index_a = qp[qp_index] + h->slice_alpha_c0_offset;

  6752.         alpha = (alpha_table+52)[index_a];

  6753.         beta  = (beta_table+52)[qp[qp_index] + h->slice_beta_offset];

  6754. 

  6755.         if( bS[bS_index] < 4 ) {

  6756.             const int tc0 = (tc0_table+52)[index_a][bS[bS_index] - 1];

  6757.             const int p0 = pix[-1];

  6758.             const int p1 = pix[-2];

  6759.             const int p2 = pix[-3];

  6760.             const int q0 = pix[0];

  6761.             const int q1 = pix[1];

  6762.             const int q2 = pix[2];

  6763. 

  6764.             if( FFABS( p0 - q0 ) < alpha &&

  6765.                 FFABS( p1 - p0 ) < beta &&

  6766.                 FFABS( q1 - q0 ) < beta ) {

  6767.                 int tc = tc0;

  6768.                 int i_delta;

  6769. 

  6770.                 if( FFABS( p2 - p0 ) < beta ) {

  6771.                     pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );

  6772.                     tc++;

  6773.                 }

  6774.                 if( FFABS( q2 - q0 ) < beta ) {

  6775.                     pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );

  6776.                     tc++;

  6777.                 }

  6778. 

  6779.                 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );

  6780.                 pix[-1] = clip_uint8( p0 + i_delta );    /* p0' */

  6781.                 pix[0]  = clip_uint8( q0 - i_delta );    /* q0' */

  6782.                 tprintf("filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);

  6783.             }

  6784.         }else{

  6785.             const int p0 = pix[-1];

  6786.             const int p1 = pix[-2];

  6787.             const int p2 = pix[-3];

  6788. 

  6789.             const int q0 = pix[0];

  6790.             const int q1 = pix[1];

  6791.             const int q2 = pix[2];

  6792. 

  6793.             if( FFABS( p0 - q0 ) < alpha &&

  6794.                 FFABS( p1 - p0 ) < beta &&

  6795.                 FFABS( q1 - q0 ) < beta ) {

  6796. 

  6797.                 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){

  6798.                     if( FFABS( p2 - p0 ) < beta)

  6799.                     {

  6800.                         const int p3 = pix[-4];

  6801.                         /* p0', p1', p2' */

  6802.                         pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;

  6803.                         pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;

  6804.                         pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;

  6805.                     } else {

  6806.                         /* p0' */

  6807.                         pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6808.                     }

  6809.                     if( FFABS( q2 - q0 ) < beta)

  6810.                     {

  6811.                         const int q3 = pix[3];

  6812.                         /* q0', q1', q2' */

  6813.                         pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;

  6814.                         pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;

  6815.                         pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;

  6816.                     } else {

  6817.                         /* q0' */

  6818.                         pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6819.                     }

  6820.                 }else{

  6821.                     /* p0', q0' */

  6822.                     pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6823.                     pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6824.                 }

  6825.                 tprintf("filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);

  6826.             }

  6827.         }

  6828.     }

  6829. }

  6830. static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) {

  6831.     int i;

  6832.     for( i = 0; i < 8; i++, pix += stride) {

  6833.         int index_a;

  6834.         int alpha;

  6835.         int beta;

  6836. 

  6837.         int qp_index;

  6838.         int bS_index = i;

  6839. 

  6840.         if( bS[bS_index] == 0 ) {

  6841.             continue;

  6842.         }

  6843. 

  6844.         qp_index = MB_FIELD ? (i >> 2) : (i & 1);

  6845.         index_a = qp[qp_index] + h->slice_alpha_c0_offset;

  6846.         alpha = (alpha_table+52)[index_a];

  6847.         beta  = (beta_table+52)[qp[qp_index] + h->slice_beta_offset];

  6848. 

  6849.         if( bS[bS_index] < 4 ) {

  6850.             const int tc = (tc0_table+52)[index_a][bS[bS_index] - 1] + 1;

  6851.             const int p0 = pix[-1];

  6852.             const int p1 = pix[-2];

  6853.             const int q0 = pix[0];

  6854.             const int q1 = pix[1];

  6855. 

  6856.             if( FFABS( p0 - q0 ) < alpha &&

  6857.                 FFABS( p1 - p0 ) < beta &&

  6858.                 FFABS( q1 - q0 ) < beta ) {

  6859.                 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );

  6860. 

  6861.                 pix[-1] = clip_uint8( p0 + i_delta );    /* p0' */

  6862.                 pix[0]  = clip_uint8( q0 - i_delta );    /* q0' */

  6863.                 tprintf("filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);

  6864.             }

  6865.         }else{

  6866.             const int p0 = pix[-1];

  6867.             const int p1 = pix[-2];

  6868.             const int q0 = pix[0];

  6869.             const int q1 = pix[1];

  6870. 

  6871.             if( FFABS( p0 - q0 ) < alpha &&

  6872.                 FFABS( p1 - p0 ) < beta &&

  6873.                 FFABS( q1 - q0 ) < beta ) {

  6874. 

  6875.                 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */

  6876.                 pix[0]  = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */

  6877.                 tprintf("filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);

  6878.             }

  6879.         }

  6880.     }

  6881. }

  6882. 

  6883. static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {

  6884.     int i, d;

  6885.     const int index_a = qp + h->slice_alpha_c0_offset;

  6886.     const int alpha = (alpha_table+52)[index_a];

  6887.     const int beta  = (beta_table+52)[qp + h->slice_beta_offset];

  6888.     const int pix_next  = stride;

  6889. 

  6890.     if( bS[0] < 4 ) {

  6891.         int8_t tc[4];

  6892.         for(i=0; i<4; i++)

  6893.             tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] : -1;

  6894.         h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);

  6895.     } else {

  6896.         /* 16px edge length, see filter_mb_edgev */

  6897.             for( d = 0; d < 16; d++ ) {

  6898.                 const int p0 = pix[-1*pix_next];

  6899.                 const int p1 = pix[-2*pix_next];

  6900.                 const int p2 = pix[-3*pix_next];

  6901.                 const int q0 = pix[0];

  6902.                 const int q1 = pix[1*pix_next];

  6903.                 const int q2 = pix[2*pix_next];

  6904. 

  6905.                 if( FFABS( p0 - q0 ) < alpha &&

  6906.                     FFABS( p1 - p0 ) < beta &&

  6907.                     FFABS( q1 - q0 ) < beta ) {

  6908. 

  6909.                     const int p3 = pix[-4*pix_next];

  6910.                     const int q3 = pix[ 3*pix_next];

  6911. 

  6912.                     if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){

  6913.                         if( FFABS( p2 - p0 ) < beta) {

  6914.                             /* p0', p1', p2' */

  6915.                             pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;

  6916.                             pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;

  6917.                             pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;

  6918.                         } else {

  6919.                             /* p0' */

  6920.                             pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6921.                         }

  6922.                         if( FFABS( q2 - q0 ) < beta) {

  6923.                             /* q0', q1', q2' */

  6924.                             pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;

  6925.                             pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;

  6926.                             pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;

  6927.                         } else {

  6928.                             /* q0' */

  6929.                             pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6930.                         }

  6931.                     }else{

  6932.                         /* p0', q0' */

  6933.                         pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6934.                         pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6935.                     }

  6936.                     tprintf("filter_mb_edgeh i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, bS[i], p2, p1, p0, q0, q1, q2, pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]);

  6937.                 }

  6938.                 pix++;

  6939.             }

  6940.     }

  6941. }

  6942. 

  6943. static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {

  6944.     int i;

  6945.     const int index_a = qp + h->slice_alpha_c0_offset;

  6946.     const int alpha = (alpha_table+52)[index_a];

  6947.     const int beta  = (beta_table+52)[qp + h->slice_beta_offset];

  6948. 

  6949.     if( bS[0] < 4 ) {

  6950.         int8_t tc[4];

  6951.         for(i=0; i<4; i++)

  6952.             tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] + 1 : 0;

  6953.         h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);

  6954.     } else {

  6955.         h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);

  6956.     }

  6957. }

  6958. 

  6959. static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {

  6960.     MpegEncContext * const s = &h->s;

  6961.     int mb_xy, mb_type;

  6962.     int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;

  6963. 

  6964.     if(mb_x==0 || mb_y==0 || !s->dsp.h264_loop_filter_strength) {

  6965.         filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);

  6966.         return;

  6967.     }

  6968.     assert(!FRAME_MBAFF);

  6969. 

  6970.     mb_xy = mb_x + mb_y*s->mb_stride;

  6971.     mb_type = s->current_picture.mb_type[mb_xy];

  6972.     qp = s->current_picture.qscale_table[mb_xy];

  6973.     qp0 = s->current_picture.qscale_table[mb_xy-1];

  6974.     qp1 = s->current_picture.qscale_table[h->top_mb_xy];

  6975.     qpc = get_chroma_qp( h->pps.chroma_qp_index_offset, qp );

  6976.     qpc0 = get_chroma_qp( h->pps.chroma_qp_index_offset, qp0 );

  6977.     qpc1 = get_chroma_qp( h->pps.chroma_qp_index_offset, qp1 );

  6978.     qp0 = (qp + qp0 + 1) >> 1;

  6979.     qp1 = (qp + qp1 + 1) >> 1;

  6980.     qpc0 = (qpc + qpc0 + 1) >> 1;

  6981.     qpc1 = (qpc + qpc1 + 1) >> 1;

  6982.     qp_thresh = 15 - h->slice_alpha_c0_offset;

  6983.     if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&

  6984.        qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)

  6985.         return;

  6986. 

  6987.     if( IS_INTRA(mb_type) ) {

  6988.         int16_t bS4[4] = {4,4,4,4};

  6989.         int16_t bS3[4] = {3,3,3,3};

  6990.         if( IS_8x8DCT(mb_type) ) {

  6991.             filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );

  6992.             filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );

  6993.             filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bS4, qp1 );

  6994.             filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );

  6995.         } else {

  6996.             filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );

  6997.             filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp );

  6998.             filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );

  6999.             filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp );

  7000.             filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bS4, qp1 );

  7001.             filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp );

  7002.             filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );

  7003.             filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp );

  7004.         }

  7005.         filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 );

  7006.         filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc );

  7007.         filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 );

  7008.         filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc );

  7009.         filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bS4, qpc1 );

  7010.         filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc );

  7011.         filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bS4, qpc1 );

  7012.         filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc );

  7013.         return;

  7014.     } else {

  7015.         DECLARE_ALIGNED_8(int16_t, bS[2][4][4]);

  7016.         uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;

  7017.         int edges;

  7018.         if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {

  7019.             edges = 4;

  7020.             bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;

  7021.         } else {

  7022.             int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 :

  7023.                              (mb_type & MB_TYPE_16x8) ? 1 : 0;

  7024.             int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16))

  7025.                              && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16))

  7026.                              ? 3 : 0;

  7027.             int step = IS_8x8DCT(mb_type) ? 2 : 1;

  7028.             edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;

  7029.             s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,

  7030.                                               (h->slice_type == B_TYPE), edges, step, mask_edge0, mask_edge1 );

  7031.         }

  7032.         if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) )

  7033.             bSv[0][0] = 0x0004000400040004ULL;

  7034.         if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) )

  7035.             bSv[1][0] = 0x0004000400040004ULL;

  7036. 

  7037. #define FILTER(hv,dir,edge)\

  7038.         if(bSv[dir][edge]) {\

  7039.             filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\

  7040.             if(!(edge&1)) {\

  7041.                 filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\

  7042.                 filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\

  7043.             }\

  7044.         }

  7045.         if( edges == 1 ) {

  7046.             FILTER(v,0,0);

  7047.             FILTER(h,1,0);

  7048.         } else if( IS_8x8DCT(mb_type) ) {

  7049.             FILTER(v,0,0);

  7050.             FILTER(v,0,2);

  7051.             FILTER(h,1,0);

  7052.             FILTER(h,1,2);

  7053.         } else {

  7054.             FILTER(v,0,0);

  7055.             FILTER(v,0,1);

  7056.             FILTER(v,0,2);

  7057.             FILTER(v,0,3);

  7058.             FILTER(h,1,0);

  7059.             FILTER(h,1,1);

  7060.             FILTER(h,1,2);

  7061.             FILTER(h,1,3);

  7062.         }

  7063. #undef FILTER

  7064.     }

  7065. }

  7066. 

  7067. static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {

  7068.     MpegEncContext * const s = &h->s;

  7069.     const int mb_xy= mb_x + mb_y*s->mb_stride;

  7070.     const int mb_type = s->current_picture.mb_type[mb_xy];

  7071.     const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;

  7072.     int first_vertical_edge_done = 0;

  7073.     int dir;

  7074.     /* FIXME: A given frame may occupy more than one position in

  7075.      * the reference list. So ref2frm should be populated with

  7076.      * frame numbers, not indices. */

  7077.     static const int ref2frm[34] = {-1,-1,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,

  7078.                                     16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31};

  7079. 

  7080.     //for sufficiently low qp, filtering wouldn't do anything

  7081.     //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp

  7082.     if(!FRAME_MBAFF){

  7083.         int qp_thresh = 15 - h->slice_alpha_c0_offset - FFMAX(0, h->pps.chroma_qp_index_offset);

  7084.         int qp = s->current_picture.qscale_table[mb_xy];

  7085.         if(qp <= qp_thresh

  7086.            && (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh)

  7087.            && (mb_y == 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){

  7088.             return;

  7089.         }

  7090.     }

  7091. 

  7092.     if (FRAME_MBAFF

  7093.             // left mb is in picture

  7094.             && h->slice_table[mb_xy-1] != 255

  7095.             // and current and left pair do not have the same interlaced type

  7096.             && (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))

  7097.             // and left mb is in the same slice if deblocking_filter == 2

  7098.             && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {

  7099.         /* First vertical edge is different in MBAFF frames

  7100.          * There are 8 different bS to compute and 2 different Qp

  7101.          */

  7102.         const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;

  7103.         const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };

  7104.         int16_t bS[8];

  7105.         int qp[2];

  7106.         int chroma_qp[2];

  7107.         int mb_qp, mbn0_qp, mbn1_qp;

  7108.         int i;

  7109.         first_vertical_edge_done = 1;

  7110. 

  7111.         if( IS_INTRA(mb_type) )

  7112.             bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4;

  7113.         else {

  7114.             for( i = 0; i < 8; i++ ) {

  7115.                 int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];

  7116. 

  7117.                 if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )

  7118.                     bS[i] = 4;

  7119.                 else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||

  7120.                          /* FIXME: with 8x8dct + cavlc, should check cbp instead of nnz */

  7121.                          h->non_zero_count[mbn_xy][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2] )

  7122.                     bS[i] = 2;

  7123.                 else

  7124.                     bS[i] = 1;

  7125.             }

  7126.         }

  7127. 

  7128.         mb_qp = s->current_picture.qscale_table[mb_xy];

  7129.         mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];

  7130.         mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];

  7131.         qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;

  7132.         chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) +

  7133.                          get_chroma_qp( h->pps.chroma_qp_index_offset, mbn0_qp ) + 1 ) >> 1;

  7134.         qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;

  7135.         chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) +

  7136.                          get_chroma_qp( h->pps.chroma_qp_index_offset, mbn1_qp ) + 1 ) >> 1;

  7137. 

  7138.         /* Filter edge */

  7139.         tprintf("filter mb:%d/%d MBAFF, QPy:%d/%d, QPc:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], chroma_qp[0], chroma_qp[1], linesize, uvlinesize);

  7140.         { int i; for (i = 0; i < 8; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }

  7141.         filter_mb_mbaff_edgev ( h, &img_y [0], linesize,   bS, qp );

  7142.         filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp );

  7143.         filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp );

  7144.     }

  7145.     /* dir : 0 -> vertical edge, 1 -> horizontal edge */

  7146.     for( dir = 0; dir < 2; dir++ )

  7147.     {

  7148.         int edge;

  7149.         const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;

  7150.         const int mbm_type = s->current_picture.mb_type[mbm_xy];

  7151.         int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;

  7152. 

  7153.         const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP))

  7154.                                   == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4;

  7155.         // how often to recheck mv-based bS when iterating between edges

  7156.         const int mask_edge = (mb_type & (MB_TYPE_16x16 | (MB_TYPE_16x8 << dir))) ? 3 :

  7157.                               (mb_type & (MB_TYPE_8x16 >> dir)) ? 1 : 0;

  7158.         // how often to recheck mv-based bS when iterating along each edge

  7159.         const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));

  7160. 

  7161.         if (first_vertical_edge_done) {

  7162.             start = 1;

  7163.             first_vertical_edge_done = 0;

  7164.         }

  7165. 

  7166.         if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])

  7167.             start = 1;

  7168. 

  7169.         if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0

  7170.             && !IS_INTERLACED(mb_type)

  7171.             && IS_INTERLACED(mbm_type)

  7172.             ) {

  7173.             // This is a special case in the norm where the filtering must

  7174.             // be done twice (one each of the field) even if we are in a

  7175.             // frame macroblock.

  7176.             //

  7177.             static const int nnz_idx[4] = {4,5,6,3};

  7178.             unsigned int tmp_linesize   = 2 *   linesize;

  7179.             unsigned int tmp_uvlinesize = 2 * uvlinesize;

  7180.             int mbn_xy = mb_xy - 2 * s->mb_stride;

  7181.             int qp, chroma_qp;

  7182.             int i, j;

  7183.             int16_t bS[4];

  7184. 

  7185.             for(j=0; j<2; j++, mbn_xy += s->mb_stride){

  7186.                 if( IS_INTRA(mb_type) ||

  7187.                     IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) {

  7188.                     bS[0] = bS[1] = bS[2] = bS[3] = 3;

  7189.                 } else {

  7190.                     const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy];

  7191.                     for( i = 0; i < 4; i++ ) {

  7192.                         if( h->non_zero_count_cache[scan8[0]+i] != 0 ||

  7193.                             mbn_nnz[nnz_idx[i]] != 0 )

  7194.                             bS[i] = 2;

  7195.                         else

  7196.                             bS[i] = 1;

  7197.                     }

  7198.                 }

  7199.                 // Do not use s->qscale as luma quantizer because it has not the same

  7200.                 // value in IPCM macroblocks.

  7201.                 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;

  7202.                 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);

  7203.                 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }

  7204.                 filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp );

  7205.                 chroma_qp = ( h->chroma_qp +

  7206.                               get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;

  7207.                 filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp );

  7208.                 filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp );

  7209.             }

  7210. 

  7211.             start = 1;

  7212.         }

  7213. 

  7214.         /* Calculate bS */

  7215.         for( edge = start; edge < edges; edge++ ) {

  7216.             /* mbn_xy: neighbor macroblock */

  7217.             const int mbn_xy = edge > 0 ? mb_xy : mbm_xy;

  7218.             const int mbn_type = s->current_picture.mb_type[mbn_xy];

  7219.             int16_t bS[4];

  7220.             int qp;

  7221. 

  7222.             if( (edge&1) && IS_8x8DCT(mb_type) )

  7223.                 continue;

  7224. 

  7225.             if( IS_INTRA(mb_type) ||

  7226.                 IS_INTRA(mbn_type) ) {

  7227.                 int value;

  7228.                 if (edge == 0) {

  7229.                     if (   (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type))

  7230.                         || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))

  7231.                     ) {

  7232.                         value = 4;

  7233.                     } else {

  7234.                         value = 3;

  7235.                     }

  7236.                 } else {

  7237.                     value = 3;

  7238.                 }

  7239.                 bS[0] = bS[1] = bS[2] = bS[3] = value;

  7240.             } else {

  7241.                 int i, l;

  7242.                 int mv_done;

  7243. 

  7244.                 if( edge & mask_edge ) {

  7245.                     bS[0] = bS[1] = bS[2] = bS[3] = 0;

  7246.                     mv_done = 1;

  7247.                 }

  7248.                 else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) {

  7249.                     bS[0] = bS[1] = bS[2] = bS[3] = 1;

  7250.                     mv_done = 1;

  7251.                 }

  7252.                 else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {

  7253.                     int b_idx= 8 + 4 + edge * (dir ? 8:1);

  7254.                     int bn_idx= b_idx - (dir ? 8:1);

  7255.                     int v = 0;

  7256.                     for( l = 0; !v && l < 1 + (h->slice_type == B_TYPE); l++ ) {

  7257.                         v |= ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||

  7258.                              FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||

  7259.                              FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;

  7260.                     }

  7261.                     bS[0] = bS[1] = bS[2] = bS[3] = v;

  7262.                     mv_done = 1;

  7263.                 }

  7264.                 else

  7265.                     mv_done = 0;

  7266. 

  7267.                 for( i = 0; i < 4; i++ ) {

  7268.                     int x = dir == 0 ? edge : i;

  7269.                     int y = dir == 0 ? i    : edge;

  7270.                     int b_idx= 8 + 4 + x + 8*y;

  7271.                     int bn_idx= b_idx - (dir ? 8:1);

  7272. 

  7273.                     if( h->non_zero_count_cache[b_idx] != 0 ||

  7274.                         h->non_zero_count_cache[bn_idx] != 0 ) {

  7275.                         bS[i] = 2;

  7276.                     }

  7277.                     else if(!mv_done)

  7278.                     {

  7279.                         bS[i] = 0;

  7280.                         for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {

  7281.                             if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||

  7282.                                 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||

  7283.                                 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {

  7284.                                 bS[i] = 1;

  7285.                                 break;

  7286.                             }

  7287.                         }

  7288.                     }

  7289.                 }

  7290. 

  7291.                 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)

  7292.                     continue;

  7293.             }

  7294. 

  7295.             /* Filter edge */

  7296.             // Do not use s->qscale as luma quantizer because it has not the same

  7297.             // value in IPCM macroblocks.

  7298.             qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;

  7299.             //tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp, s->current_picture.qscale_table[mbn_xy]);

  7300.             tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);

  7301.             { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }

  7302.             if( dir == 0 ) {

  7303.                 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );

  7304.                 if( (edge&1) == 0 ) {

  7305.                     int chroma_qp = ( h->chroma_qp +

  7306.                                       get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;

  7307.                     filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );

  7308.                     filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );

  7309.                 }

  7310.             } else {

  7311.                 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );

  7312.                 if( (edge&1) == 0 ) {

  7313.                     int chroma_qp = ( h->chroma_qp +

  7314.                                       get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;

  7315.                     filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );

  7316.                     filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );

  7317.                 }

  7318.             }

  7319.         }

  7320.     }

  7321. }

  7322. 

  7323. static int decode_slice(H264Context *h){

  7324.     MpegEncContext * const s = &h->s;

  7325.     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;

  7326. 

  7327.     s->mb_skip_run= -1;

  7328. 

  7329.     if( h->pps.cabac ) {

  7330.         int i;

  7331. 

  7332.         /* realign */

  7333.         align_get_bits( &s->gb );

  7334. 

  7335.         /* init cabac */

  7336.         ff_init_cabac_states( &h->cabac);

  7337.         ff_init_cabac_decoder( &h->cabac,

  7338.                                s->gb.buffer + get_bits_count(&s->gb)/8,

  7339.                                ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);

  7340.         /* calculate pre-state */

  7341.         for( i= 0; i < 460; i++ ) {

  7342.             int pre;

  7343.             if( h->slice_type == I_TYPE )

  7344.                 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );

  7345.             else

  7346.                 pre = clip( ((cabac_context_init_PB[h->cabac_init_idc][i][0] * s->qscale) >>4 ) + cabac_context_init_PB[h->cabac_init_idc][i][1], 1, 126 );

  7347. 

  7348.             if( pre <= 63 )

  7349.                 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;

  7350.             else

  7351.                 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;

  7352.         }

  7353. 

  7354.         for(;;){

  7355. //START_TIMER

  7356.             int ret = decode_mb_cabac(h);

  7357.             int eos;

  7358. //STOP_TIMER("decode_mb_cabac")

  7359. 

  7360.             if(ret>=0) hl_decode_mb(h);

  7361. 

  7362.             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?

  7363.                 s->mb_y++;

  7364. 

  7365.                 if(ret>=0) ret = decode_mb_cabac(h);

  7366. 

  7367.                 if(ret>=0) hl_decode_mb(h);

  7368.                 s->mb_y--;

  7369.             }

  7370.             eos = get_cabac_terminate( &h->cabac );

  7371. 

  7372.             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {

  7373.                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%d)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);

  7374.                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);

  7375.                 return -1;

  7376.             }

  7377. 

  7378.             if( ++s->mb_x >= s->mb_width ) {

  7379.                 s->mb_x = 0;

  7380.                 ff_draw_horiz_band(s, 16*s->mb_y, 16);

  7381.                 ++s->mb_y;

  7382.                 if(FRAME_MBAFF) {

  7383.                     ++s->mb_y;

  7384.                 }

  7385.             }

  7386. 

  7387.             if( eos || s->mb_y >= s->mb_height ) {

  7388.                 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);

  7389.                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

  7390.                 return 0;

  7391.             }

  7392.         }

  7393. 

  7394.     } else {

  7395.         for(;;){

  7396.             int ret = decode_mb_cavlc(h);

  7397. 

  7398.             if(ret>=0) hl_decode_mb(h);

  7399. 

  7400.             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?

  7401.                 s->mb_y++;

  7402.                 ret = decode_mb_cavlc(h);

  7403. 

  7404.                 if(ret>=0) hl_decode_mb(h);

  7405.                 s->mb_y--;

  7406.             }

  7407. 

  7408.             if(ret<0){

  7409.                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  7410.                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);

  7411. 

  7412.                 return -1;

  7413.             }

  7414. 

  7415.             if(++s->mb_x >= s->mb_width){

  7416.                 s->mb_x=0;

  7417.                 ff_draw_horiz_band(s, 16*s->mb_y, 16);

  7418.                 ++s->mb_y;

  7419.                 if(FRAME_MBAFF) {

  7420.                     ++s->mb_y;

  7421.                 }

  7422.                 if(s->mb_y >= s->mb_height){

  7423.                     tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);

  7424. 

  7425.                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {

  7426.                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

  7427. 

  7428.                         return 0;

  7429.                     }else{

  7430.                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

  7431. 

  7432.                         return -1;

  7433.                     }

  7434.                 }

  7435.             }

  7436. 

  7437.             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){

  7438.                 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);

  7439.                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){

  7440.                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

  7441. 

  7442.                     return 0;

  7443.                 }else{

  7444.                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);

  7445. 

  7446.                     return -1;

  7447.                 }

  7448.             }

  7449.         }

  7450.     }

  7451. 

  7452. #if 0

  7453.     for(;s->mb_y < s->mb_height; s->mb_y++){

  7454.         for(;s->mb_x < s->mb_width; s->mb_x++){

  7455.             int ret= decode_mb(h);

  7456. 

  7457.             hl_decode_mb(h);

  7458. 

  7459.             if(ret<0){

  7460.                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  7461.                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);

  7462. 

  7463.                 return -1;

  7464.             }

  7465. 

  7466.             if(++s->mb_x >= s->mb_width){

  7467.                 s->mb_x=0;

  7468.                 if(++s->mb_y >= s->mb_height){

  7469.                     if(get_bits_count(s->gb) == s->gb.size_in_bits){

  7470.                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

  7471. 

  7472.                         return 0;

  7473.                     }else{

  7474.                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

  7475. 

  7476.                         return -1;

  7477.                     }

  7478.                 }

  7479.             }

  7480. 

  7481.             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){

  7482.                 if(get_bits_count(s->gb) == s->gb.size_in_bits){

  7483.                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);

  7484. 

  7485.                     return 0;

  7486.                 }else{

  7487.                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);

  7488. 

  7489.                     return -1;

  7490.                 }

  7491.             }

  7492.         }

  7493.         s->mb_x=0;

  7494.         ff_draw_horiz_band(s, 16*s->mb_y, 16);

  7495.     }

  7496. #endif

  7497.     return -1; //not reached

  7498. }

  7499. 

  7500. static int decode_unregistered_user_data(H264Context *h, int size){

  7501.     MpegEncContext * const s = &h->s;

  7502.     uint8_t user_data[16+256];

  7503.     int e, build, i;

  7504. 

  7505.     if(size<16)

  7506.         return -1;

  7507. 

  7508.     for(i=0; i<sizeof(user_data)-1 && i<size; i++){

  7509.         user_data[i]= get_bits(&s->gb, 8);

  7510.     }

  7511. 

  7512.     user_data[i]= 0;

  7513.     e= sscanf(user_data+16, "x264 - core %d"/*%s - H.264/MPEG-4 AVC codec - Copyleft 2005 - http://www.videolan.org/x264.html*/, &build);

  7514.     if(e==1 && build>=0)

  7515.         h->x264_build= build;

  7516. 

  7517.     if(s->avctx->debug & FF_DEBUG_BUGS)

  7518.         av_log(s->avctx, AV_LOG_DEBUG, "user data:\"%s\"\n", user_data+16);

  7519. 

  7520.     for(; i<size; i++)

  7521.         skip_bits(&s->gb, 8);

  7522. 

  7523.     return 0;

  7524. }

  7525. 

  7526. static int decode_sei(H264Context *h){

  7527.     MpegEncContext * const s = &h->s;

  7528. 

  7529.     while(get_bits_count(&s->gb) + 16 < s->gb.size_in_bits){

  7530.         int size, type;

  7531. 

  7532.         type=0;

  7533.         do{

  7534.             type+= show_bits(&s->gb, 8);

  7535.         }while(get_bits(&s->gb, 8) == 255);

  7536. 

  7537.         size=0;

  7538.         do{

  7539.             size+= show_bits(&s->gb, 8);

  7540.         }while(get_bits(&s->gb, 8) == 255);

  7541. 

  7542.         switch(type){

  7543.         case 5:

  7544.             if(decode_unregistered_user_data(h, size) < 0)

  7545.                 return -1;

  7546.             break;

  7547.         default:

  7548.             skip_bits(&s->gb, 8*size);

  7549.         }

  7550. 

  7551.         //FIXME check bits here

  7552.         align_get_bits(&s->gb);

  7553.     }

  7554. 

  7555.     return 0;

  7556. }

  7557. 

  7558. static inline void decode_hrd_parameters(H264Context *h, SPS *sps){

  7559.     MpegEncContext * const s = &h->s;

  7560.     int cpb_count, i;

  7561.     cpb_count = get_ue_golomb(&s->gb) + 1;

  7562.     get_bits(&s->gb, 4); /* bit_rate_scale */

  7563.     get_bits(&s->gb, 4); /* cpb_size_scale */

  7564.     for(i=0; i<cpb_count; i++){

  7565.         get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */

  7566.         get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */

  7567.         get_bits1(&s->gb);     /* cbr_flag */

  7568.     }

  7569.     get_bits(&s->gb, 5); /* initial_cpb_removal_delay_length_minus1 */

  7570.     get_bits(&s->gb, 5); /* cpb_removal_delay_length_minus1 */

  7571.     get_bits(&s->gb, 5); /* dpb_output_delay_length_minus1 */

  7572.     get_bits(&s->gb, 5); /* time_offset_length */

  7573. }

  7574. 

  7575. static inline int decode_vui_parameters(H264Context *h, SPS *sps){

  7576.     MpegEncContext * const s = &h->s;

  7577.     int aspect_ratio_info_present_flag, aspect_ratio_idc;

  7578.     int nal_hrd_parameters_present_flag, vcl_hrd_parameters_present_flag;

  7579. 

  7580.     aspect_ratio_info_present_flag= get_bits1(&s->gb);

  7581. 

  7582.     if( aspect_ratio_info_present_flag ) {

  7583.         aspect_ratio_idc= get_bits(&s->gb, 8);

  7584.         if( aspect_ratio_idc == EXTENDED_SAR ) {

  7585.             sps->sar.num= get_bits(&s->gb, 16);

  7586.             sps->sar.den= get_bits(&s->gb, 16);

  7587.         }else if(aspect_ratio_idc < 14){

  7588.             sps->sar=  pixel_aspect[aspect_ratio_idc];

  7589.         }else{

  7590.             av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");

  7591.             return -1;

  7592.         }

  7593.     }else{

  7594.         sps->sar.num=

  7595.         sps->sar.den= 0;

  7596.     }

  7597. //            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);

  7598. 

  7599.     if(get_bits1(&s->gb)){      /* overscan_info_present_flag */

  7600.         get_bits1(&s->gb);      /* overscan_appropriate_flag */

  7601.     }

  7602. 

  7603.     if(get_bits1(&s->gb)){      /* video_signal_type_present_flag */

  7604.         get_bits(&s->gb, 3);    /* video_format */

  7605.         get_bits1(&s->gb);      /* video_full_range_flag */

  7606.         if(get_bits1(&s->gb)){  /* colour_description_present_flag */

  7607.             get_bits(&s->gb, 8); /* colour_primaries */

  7608.             get_bits(&s->gb, 8); /* transfer_characteristics */

  7609.             get_bits(&s->gb, 8); /* matrix_coefficients */

  7610.         }

  7611.     }

  7612. 

  7613.     if(get_bits1(&s->gb)){      /* chroma_location_info_present_flag */

  7614.         get_ue_golomb(&s->gb);  /* chroma_sample_location_type_top_field */

  7615.         get_ue_golomb(&s->gb);  /* chroma_sample_location_type_bottom_field */

  7616.     }

  7617. 

  7618.     sps->timing_info_present_flag = get_bits1(&s->gb);

  7619.     if(sps->timing_info_present_flag){

  7620.         sps->num_units_in_tick = get_bits_long(&s->gb, 32);

  7621.         sps->time_scale = get_bits_long(&s->gb, 32);

  7622.         sps->fixed_frame_rate_flag = get_bits1(&s->gb);

  7623.     }

  7624. 

  7625.     nal_hrd_parameters_present_flag = get_bits1(&s->gb);

  7626.     if(nal_hrd_parameters_present_flag)

  7627.         decode_hrd_parameters(h, sps);

  7628.     vcl_hrd_parameters_present_flag = get_bits1(&s->gb);

  7629.     if(vcl_hrd_parameters_present_flag)

  7630.         decode_hrd_parameters(h, sps);

  7631.     if(nal_hrd_parameters_present_flag || vcl_hrd_parameters_present_flag)

  7632.         get_bits1(&s->gb);     /* low_delay_hrd_flag */

  7633.     get_bits1(&s->gb);         /* pic_struct_present_flag */

  7634. 

  7635.     sps->bitstream_restriction_flag = get_bits1(&s->gb);

  7636.     if(sps->bitstream_restriction_flag){

  7637.         get_bits1(&s->gb);     /* motion_vectors_over_pic_boundaries_flag */

  7638.         get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */

  7639.         get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */

  7640.         get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */

  7641.         get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */

  7642.         sps->num_reorder_frames = get_ue_golomb(&s->gb);

  7643.         get_ue_golomb(&s->gb); /* max_dec_frame_buffering */

  7644.     }

  7645. 

  7646.     return 0;

  7647. }

  7648. 

  7649. static void decode_scaling_list(H264Context *h, uint8_t *factors, int size,

  7650.                                 const uint8_t *jvt_list, const uint8_t *fallback_list){

  7651.     MpegEncContext * const s = &h->s;

  7652.     int i, last = 8, next = 8;

  7653.     const uint8_t *scan = size == 16 ? zigzag_scan : zigzag_scan8x8;

  7654.     if(!get_bits1(&s->gb)) /* matrix not written, we use the predicted one */

  7655.         memcpy(factors, fallback_list, size*sizeof(uint8_t));

  7656.     else

  7657.     for(i=0;i<size;i++){

  7658.         if(next)

  7659.             next = (last + get_se_golomb(&s->gb)) & 0xff;

  7660.         if(!i && !next){ /* matrix not written, we use the preset one */

  7661.             memcpy(factors, jvt_list, size*sizeof(uint8_t));

  7662.             break;

  7663.         }

  7664.         last = factors[scan[i]] = next ? next : last;

  7665.     }

  7666. }

  7667. 

  7668. static void decode_scaling_matrices(H264Context *h, SPS *sps, PPS *pps, int is_sps,

  7669.                                    uint8_t (*scaling_matrix4)[16], uint8_t (*scaling_matrix8)[64]){

  7670.     MpegEncContext * const s = &h->s;

  7671.     int fallback_sps = !is_sps && sps->scaling_matrix_present;

  7672.     const uint8_t *fallback[4] = {

  7673.         fallback_sps ? sps->scaling_matrix4[0] : default_scaling4[0],

  7674.         fallback_sps ? sps->scaling_matrix4[3] : default_scaling4[1],

  7675.         fallback_sps ? sps->scaling_matrix8[0] : default_scaling8[0],

  7676.         fallback_sps ? sps->scaling_matrix8[1] : default_scaling8[1]

  7677.     };

  7678.     if(get_bits1(&s->gb)){

  7679.         sps->scaling_matrix_present |= is_sps;

  7680.         decode_scaling_list(h,scaling_matrix4[0],16,default_scaling4[0],fallback[0]); // Intra, Y

  7681.         decode_scaling_list(h,scaling_matrix4[1],16,default_scaling4[0],scaling_matrix4[0]); // Intra, Cr

  7682.         decode_scaling_list(h,scaling_matrix4[2],16,default_scaling4[0],scaling_matrix4[1]); // Intra, Cb

  7683.         decode_scaling_list(h,scaling_matrix4[3],16,default_scaling4[1],fallback[1]); // Inter, Y

  7684.         decode_scaling_list(h,scaling_matrix4[4],16,default_scaling4[1],scaling_matrix4[3]); // Inter, Cr

  7685.         decode_scaling_list(h,scaling_matrix4[5],16,default_scaling4[1],scaling_matrix4[4]); // Inter, Cb

  7686.         if(is_sps || pps->transform_8x8_mode){

  7687.             decode_scaling_list(h,scaling_matrix8[0],64,default_scaling8[0],fallback[2]);  // Intra, Y

  7688.             decode_scaling_list(h,scaling_matrix8[1],64,default_scaling8[1],fallback[3]);  // Inter, Y

  7689.         }

  7690.     } else if(fallback_sps) {

  7691.         memcpy(scaling_matrix4, sps->scaling_matrix4, 6*16*sizeof(uint8_t));

  7692.         memcpy(scaling_matrix8, sps->scaling_matrix8, 2*64*sizeof(uint8_t));

  7693.     }

  7694. }

  7695. 

  7696. static inline int decode_seq_parameter_set(H264Context *h){

  7697.     MpegEncContext * const s = &h->s;

  7698.     int profile_idc, level_idc;

  7699.     int sps_id, i;

  7700.     SPS *sps;

  7701. 

  7702.     profile_idc= get_bits(&s->gb, 8);

  7703.     get_bits1(&s->gb);   //constraint_set0_flag

  7704.     get_bits1(&s->gb);   //constraint_set1_flag

  7705.     get_bits1(&s->gb);   //constraint_set2_flag

  7706.     get_bits1(&s->gb);   //constraint_set3_flag

  7707.     get_bits(&s->gb, 4); // reserved

  7708.     level_idc= get_bits(&s->gb, 8);

  7709.     sps_id= get_ue_golomb(&s->gb);

  7710. 

  7711.     sps= &h->sps_buffer[ sps_id ];

  7712.     sps->profile_idc= profile_idc;

  7713.     sps->level_idc= level_idc;

  7714. 

  7715.     if(sps->profile_idc >= 100){ //high profile

  7716.         if(get_ue_golomb(&s->gb) == 3) //chroma_format_idc

  7717.             get_bits1(&s->gb);  //residual_color_transform_flag

  7718.         get_ue_golomb(&s->gb);  //bit_depth_luma_minus8

  7719.         get_ue_golomb(&s->gb);  //bit_depth_chroma_minus8

  7720.         sps->transform_bypass = get_bits1(&s->gb);

  7721.         decode_scaling_matrices(h, sps, NULL, 1, sps->scaling_matrix4, sps->scaling_matrix8);

  7722.     }else

  7723.         sps->scaling_matrix_present = 0;

  7724. 

  7725.     sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;

  7726.     sps->poc_type= get_ue_golomb(&s->gb);

  7727. 

  7728.     if(sps->poc_type == 0){ //FIXME #define

  7729.         sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;

  7730.     } else if(sps->poc_type == 1){//FIXME #define

  7731.         sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);

  7732.         sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);

  7733.         sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);

  7734.         sps->poc_cycle_length= get_ue_golomb(&s->gb);

  7735. 

  7736.         for(i=0; i<sps->poc_cycle_length; i++)

  7737.             sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);

  7738.     }

  7739.     if(sps->poc_type > 2){

  7740.         av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);

  7741.         return -1;

  7742.     }

  7743. 

  7744.     sps->ref_frame_count= get_ue_golomb(&s->gb);

  7745.     if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){

  7746.         av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");

  7747.     }

  7748.     sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);

  7749.     sps->mb_width= get_ue_golomb(&s->gb) + 1;

  7750.     sps->mb_height= get_ue_golomb(&s->gb) + 1;

  7751.     if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 ||

  7752.        avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height))

  7753.         return -1;

  7754. 

  7755.     sps->frame_mbs_only_flag= get_bits1(&s->gb);

  7756.     if(!sps->frame_mbs_only_flag)

  7757.         sps->mb_aff= get_bits1(&s->gb);

  7758.     else

  7759.         sps->mb_aff= 0;

  7760. 

  7761.     sps->direct_8x8_inference_flag= get_bits1(&s->gb);

  7762. 

  7763. #ifndef ALLOW_INTERLACE

  7764.     if(sps->mb_aff)

  7765.         av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF support not included; enable it at compile-time.\n");

  7766. #endif

  7767.     if(!sps->direct_8x8_inference_flag && sps->mb_aff)

  7768.         av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + !direct_8x8_inference is not implemented\n");

  7769. 

  7770.     sps->crop= get_bits1(&s->gb);

  7771.     if(sps->crop){

  7772.         sps->crop_left  = get_ue_golomb(&s->gb);

  7773.         sps->crop_right = get_ue_golomb(&s->gb);

  7774.         sps->crop_top   = get_ue_golomb(&s->gb);

  7775.         sps->crop_bottom= get_ue_golomb(&s->gb);

  7776.         if(sps->crop_left || sps->crop_top){

  7777.             av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completely supported, this could look slightly wrong ...\n");

  7778.         }

  7779.     }else{

  7780.         sps->crop_left  =

  7781.         sps->crop_right =

  7782.         sps->crop_top   =

  7783.         sps->crop_bottom= 0;

  7784.     }

  7785. 

  7786.     sps->vui_parameters_present_flag= get_bits1(&s->gb);

  7787.     if( sps->vui_parameters_present_flag )

  7788.         decode_vui_parameters(h, sps);

  7789. 

  7790.     if(s->avctx->debug&FF_DEBUG_PICT_INFO){

  7791.         av_log(h->s.avctx, AV_LOG_DEBUG, "sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n",

  7792.                sps_id, sps->profile_idc, sps->level_idc,

  7793.                sps->poc_type,

  7794.                sps->ref_frame_count,

  7795.                sps->mb_width, sps->mb_height,

  7796.                sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),

  7797.                sps->direct_8x8_inference_flag ? "8B8" : "",

  7798.                sps->crop_left, sps->crop_right,

  7799.                sps->crop_top, sps->crop_bottom,

  7800.                sps->vui_parameters_present_flag ? "VUI" : ""

  7801.                );

  7802.     }

  7803.     return 0;

  7804. }

  7805. 

  7806. static inline int decode_picture_parameter_set(H264Context *h, int bit_length){

  7807.     MpegEncContext * const s = &h->s;

  7808.     int pps_id= get_ue_golomb(&s->gb);

  7809.     PPS *pps= &h->pps_buffer[pps_id];

  7810. 

  7811.     pps->sps_id= get_ue_golomb(&s->gb);

  7812.     pps->cabac= get_bits1(&s->gb);

  7813.     pps->pic_order_present= get_bits1(&s->gb);

  7814.     pps->slice_group_count= get_ue_golomb(&s->gb) + 1;

  7815.     if(pps->slice_group_count > 1 ){

  7816.         pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);

  7817.         av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");

  7818.         switch(pps->mb_slice_group_map_type){

  7819.         case 0:

  7820. #if 0

  7821. |   for( i = 0; i <= num_slice_groups_minus1; i++ ) |   |        |

  7822. |    run_length[ i ]                                |1  |ue(v)   |

  7823. #endif

  7824.             break;

  7825.         case 2:

  7826. #if 0

  7827. |   for( i = 0; i < num_slice_groups_minus1; i++ )  |   |        |

  7828. |{                                                  |   |        |

  7829. |    top_left_mb[ i ]                               |1  |ue(v)   |

  7830. |    bottom_right_mb[ i ]                           |1  |ue(v)   |

  7831. |   }                                               |   |        |

  7832. #endif

  7833.             break;

  7834.         case 3:

  7835.         case 4:

  7836.         case 5:

  7837. #if 0

  7838. |   slice_group_change_direction_flag               |1  |u(1)    |

  7839. |   slice_group_change_rate_minus1                  |1  |ue(v)   |

  7840. #endif

  7841.             break;

  7842.         case 6:

  7843. #if 0

  7844. |   slice_group_id_cnt_minus1                       |1  |ue(v)   |

  7845. |   for( i = 0; i <= slice_group_id_cnt_minus1; i++ |   |        |

  7846. |)                                                  |   |        |

  7847. |    slice_group_id[ i ]                            |1  |u(v)    |

  7848. #endif

  7849.             break;

  7850.         }

  7851.     }

  7852.     pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;

  7853.     pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;

  7854.     if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){

  7855.         av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");

  7856.         return -1;

  7857.     }

  7858. 

  7859.     pps->weighted_pred= get_bits1(&s->gb);

  7860.     pps->weighted_bipred_idc= get_bits(&s->gb, 2);

  7861.     pps->init_qp= get_se_golomb(&s->gb) + 26;

  7862.     pps->init_qs= get_se_golomb(&s->gb) + 26;

  7863.     pps->chroma_qp_index_offset= get_se_golomb(&s->gb);

  7864.     pps->deblocking_filter_parameters_present= get_bits1(&s->gb);

  7865.     pps->constrained_intra_pred= get_bits1(&s->gb);

  7866.     pps->redundant_pic_cnt_present = get_bits1(&s->gb);

  7867. 

  7868.     pps->transform_8x8_mode= 0;

  7869.     h->dequant_coeff_pps= -1; //contents of sps/pps can change even if id doesn't, so reinit

  7870.     memset(pps->scaling_matrix4, 16, 6*16*sizeof(uint8_t));

  7871.     memset(pps->scaling_matrix8, 16, 2*64*sizeof(uint8_t));

  7872. 

  7873.     if(get_bits_count(&s->gb) < bit_length){

  7874.         pps->transform_8x8_mode= get_bits1(&s->gb);

  7875.         decode_scaling_matrices(h, &h->sps_buffer[pps->sps_id], pps, 0, pps->scaling_matrix4, pps->scaling_matrix8);

  7876.         get_se_golomb(&s->gb);  //second_chroma_qp_index_offset

  7877.     }

  7878. 

  7879.     if(s->avctx->debug&FF_DEBUG_PICT_INFO){

  7880.         av_log(h->s.avctx, AV_LOG_DEBUG, "pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s %s\n",

  7881.                pps_id, pps->sps_id,

  7882.                pps->cabac ? "CABAC" : "CAVLC",

  7883.                pps->slice_group_count,

  7884.                pps->ref_count[0], pps->ref_count[1],

  7885.                pps->weighted_pred ? "weighted" : "",

  7886.                pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,

  7887.                pps->deblocking_filter_parameters_present ? "LPAR" : "",

  7888.                pps->constrained_intra_pred ? "CONSTR" : "",

  7889.                pps->redundant_pic_cnt_present ? "REDU" : "",

  7890.                pps->transform_8x8_mode ? "8x8DCT" : ""

  7891.                );

  7892.     }

  7893. 

  7894.     return 0;

  7895. }

  7896. 

  7897. /**

  7898.  * finds the end of the current frame in the bitstream.

  7899.  * @return the position of the first byte of the next frame, or -1

  7900.  */

  7901. static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){

  7902.     int i;

  7903.     uint32_t state;

  7904.     ParseContext *pc = &(h->s.parse_context);

  7905. //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);

  7906. //    mb_addr= pc->mb_addr - 1;

  7907.     state= pc->state;

  7908.     for(i=0; i<=buf_size; i++){

  7909.         if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){

  7910.             tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);

  7911.             if(pc->frame_start_found){

  7912.                 // If there isn't one more byte in the buffer

  7913.                 // the test on first_mb_in_slice cannot be done yet

  7914.                 // do it at next call.

  7915.                 if (i >= buf_size) break;

  7916.                 if (buf[i] & 0x80) {

  7917.                     // first_mb_in_slice is 0, probably the first nal of a new

  7918.                     // slice

  7919.                     tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);

  7920.                     pc->state=-1;

  7921.                     pc->frame_start_found= 0;

  7922.                     return i-4;

  7923.                 }

  7924.             }

  7925.             pc->frame_start_found = 1;

  7926.         }

  7927.         if((state&0xFFFFFF1F) == 0x107 || (state&0xFFFFFF1F) == 0x108 || (state&0xFFFFFF1F) == 0x109){

  7928.            if(pc->frame_start_found){

  7929.                 pc->state=-1;

  7930.                 pc->frame_start_found= 0;

  7931.                 return i-4;

  7932.            }

  7933.         }

  7934.         if (i<buf_size)

  7935.             state= (state<<8) | buf[i];

  7936.     }

  7937. 

  7938.     pc->state= state;

  7939.     return END_NOT_FOUND;

  7940. }

  7941. 

  7942. #ifdef CONFIG_H264_PARSER

  7943. static int h264_parse(AVCodecParserContext *s,

  7944.                       AVCodecContext *avctx,

  7945.                       uint8_t **poutbuf, int *poutbuf_size,

  7946.                       const uint8_t *buf, int buf_size)

  7947. {

  7948.     H264Context *h = s->priv_data;

  7949.     ParseContext *pc = &h->s.parse_context;

  7950.     int next;

  7951. 

  7952.     next= find_frame_end(h, buf, buf_size);

  7953. 

  7954.     if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {

  7955.         *poutbuf = NULL;

  7956.         *poutbuf_size = 0;

  7957.         return buf_size;

  7958.     }

  7959. 

  7960.     *poutbuf = (uint8_t *)buf;

  7961.     *poutbuf_size = buf_size;

  7962.     return next;

  7963. }

  7964. 

  7965. static int h264_split(AVCodecContext *avctx,

  7966.                       const uint8_t *buf, int buf_size)

  7967. {

  7968.     int i;

  7969.     uint32_t state = -1;

  7970.     int has_sps= 0;

  7971. 

  7972.     for(i=0; i<=buf_size; i++){

  7973.         if((state&0xFFFFFF1F) == 0x107)

  7974.             has_sps=1;

  7975. /*        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){

  7976.         }*/

  7977.         if((state&0xFFFFFF00) == 0x100 && (state&0xFFFFFF1F) != 0x107 && (state&0xFFFFFF1F) != 0x108 && (state&0xFFFFFF1F) != 0x109){

  7978.             if(has_sps){

  7979.                 while(i>4 && buf[i-5]==0) i--;

  7980.                 return i-4;

  7981.             }

  7982.         }

  7983.         if (i<buf_size)

  7984.             state= (state<<8) | buf[i];

  7985.     }

  7986.     return 0;

  7987. }

  7988. #endif /* CONFIG_H264_PARSER */

  7989. 

  7990. static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){

  7991.     MpegEncContext * const s = &h->s;

  7992.     AVCodecContext * const avctx= s->avctx;

  7993.     int buf_index=0;

  7994. #if 0

  7995.     int i;

  7996.     for(i=0; i<50; i++){

  7997.         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);

  7998.     }

  7999. #endif

  8000.     h->slice_num = 0;

  8001.     s->current_picture_ptr= NULL;

  8002.     for(;;){

  8003.         int consumed;

  8004.         int dst_length;

  8005.         int bit_length;

  8006.         uint8_t *ptr;

  8007.         int i, nalsize = 0;

  8008. 

  8009.       if(h->is_avc) {

  8010.         if(buf_index >= buf_size) break;

  8011.         nalsize = 0;

  8012.         for(i = 0; i < h->nal_length_size; i++)

  8013.             nalsize = (nalsize << 8) | buf[buf_index++];

  8014.         if(nalsize <= 1){

  8015.             if(nalsize == 1){

  8016.                 buf_index++;

  8017.                 continue;

  8018.             }else{

  8019.                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);

  8020.                 break;

  8021.             }

  8022.         }

  8023.       } else {

  8024.         // start code prefix search

  8025.         for(; buf_index + 3 < buf_size; buf_index++){

  8026.             // this should allways succeed in the first iteration

  8027.             if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)

  8028.                 break;

  8029.         }

  8030. 

  8031.         if(buf_index+3 >= buf_size) break;

  8032. 

  8033.         buf_index+=3;

  8034.       }

  8035. 

  8036.         ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);

  8037.         while(ptr[dst_length - 1] == 0 && dst_length > 1)

  8038.             dst_length--;

  8039.         bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);

  8040. 

  8041.         if(s->avctx->debug&FF_DEBUG_STARTCODE){

  8042.             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", h->nal_unit_type, buf_index, buf_size, dst_length);

  8043.         }

  8044. 

  8045.         if (h->is_avc && (nalsize != consumed))

  8046.             av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);

  8047. 

  8048.         buf_index += consumed;

  8049. 

  8050.         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME dont discard SEI id

  8051.            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))

  8052.             continue;

  8053. 

  8054.         switch(h->nal_unit_type){

  8055.         case NAL_IDR_SLICE:

  8056.             idr(h); //FIXME ensure we don't loose some frames if there is reordering

  8057.         case NAL_SLICE:

  8058.             init_get_bits(&s->gb, ptr, bit_length);

  8059.             h->intra_gb_ptr=

  8060.             h->inter_gb_ptr= &s->gb;

  8061.             s->data_partitioning = 0;

  8062. 

  8063.             if(decode_slice_header(h) < 0){

  8064.                 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");

  8065.                 break;

  8066.             }

  8067.             s->current_picture_ptr->key_frame= (h->nal_unit_type == NAL_IDR_SLICE);

  8068.             if(h->redundant_pic_count==0 && s->hurry_up < 5

  8069.                && (avctx->skip_frame < AVDISCARD_NONREF || h->nal_ref_idc)

  8070.                && (avctx->skip_frame < AVDISCARD_BIDIR  || h->slice_type!=B_TYPE)

  8071.                && (avctx->skip_frame < AVDISCARD_NONKEY || h->slice_type==I_TYPE)

  8072.                && avctx->skip_frame < AVDISCARD_ALL)

  8073.                 decode_slice(h);

  8074.             break;

  8075.         case NAL_DPA:

  8076.             init_get_bits(&s->gb, ptr, bit_length);

  8077.             h->intra_gb_ptr=

  8078.             h->inter_gb_ptr= NULL;

  8079.             s->data_partitioning = 1;

  8080. 

  8081.             if(decode_slice_header(h) < 0){

  8082.                 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");

  8083.             }

  8084.             break;

  8085.         case NAL_DPB:

  8086.             init_get_bits(&h->intra_gb, ptr, bit_length);

  8087.             h->intra_gb_ptr= &h->intra_gb;

  8088.             break;

  8089.         case NAL_DPC:

  8090.             init_get_bits(&h->inter_gb, ptr, bit_length);

  8091.             h->inter_gb_ptr= &h->inter_gb;

  8092. 

  8093.             if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning

  8094.                && s->hurry_up < 5

  8095.                && (avctx->skip_frame < AVDISCARD_NONREF || h->nal_ref_idc)

  8096.                && (avctx->skip_frame < AVDISCARD_BIDIR  || h->slice_type!=B_TYPE)

  8097.                && (avctx->skip_frame < AVDISCARD_NONKEY || h->slice_type==I_TYPE)

  8098.                && avctx->skip_frame < AVDISCARD_ALL)

  8099.                 decode_slice(h);

  8100.             break;

  8101.         case NAL_SEI:

  8102.             init_get_bits(&s->gb, ptr, bit_length);

  8103.             decode_sei(h);

  8104.             break;

  8105.         case NAL_SPS:

  8106.             init_get_bits(&s->gb, ptr, bit_length);

  8107.             decode_seq_parameter_set(h);

  8108. 

  8109.             if(s->flags& CODEC_FLAG_LOW_DELAY)

  8110.                 s->low_delay=1;

  8111. 

  8112.             if(avctx->has_b_frames < 2)

  8113.                 avctx->has_b_frames= !s->low_delay;

  8114.             break;

  8115.         case NAL_PPS:

  8116.             init_get_bits(&s->gb, ptr, bit_length);

  8117. 

  8118.             decode_picture_parameter_set(h, bit_length);

  8119. 

  8120.             break;

  8121.         case NAL_AUD:

  8122.         case NAL_END_SEQUENCE:

  8123.         case NAL_END_STREAM:

  8124.         case NAL_FILLER_DATA:

  8125.         case NAL_SPS_EXT:

  8126.         case NAL_AUXILIARY_SLICE:

  8127.             break;

  8128.         default:

  8129.             av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);

  8130.         }

  8131.     }

  8132. 

  8133.     if(!s->current_picture_ptr) return buf_index; //no frame

  8134. 

  8135.     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;

  8136.     s->current_picture_ptr->pict_type= s->pict_type;

  8137. 

  8138.     h->prev_frame_num_offset= h->frame_num_offset;

  8139.     h->prev_frame_num= h->frame_num;

  8140.     if(s->current_picture_ptr->reference){

  8141.         h->prev_poc_msb= h->poc_msb;

  8142.         h->prev_poc_lsb= h->poc_lsb;

  8143.     }

  8144.     if(s->current_picture_ptr->reference)

  8145.         execute_ref_pic_marking(h, h->mmco, h->mmco_index);

  8146. 

  8147.     ff_er_frame_end(s);

  8148. 

  8149.     MPV_frame_end(s);

  8150. 

  8151.     return buf_index;

  8152. }

  8153. 

  8154. /**

  8155.  * returns the number of bytes consumed for building the current frame

  8156.  */

  8157. static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){

  8158.     if(s->flags&CODEC_FLAG_TRUNCATED){

  8159.         pos -= s->parse_context.last_index;

  8160.         if(pos<0) pos=0; // FIXME remove (unneeded?)

  8161. 

  8162.         return pos;

  8163.     }else{

  8164.         if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)

  8165.         if(pos+10>buf_size) pos=buf_size; // oops ;)

  8166. 

  8167.         return pos;

  8168.     }

  8169. }

  8170. 

  8171. static int decode_frame(AVCodecContext *avctx,

  8172.                              void *data, int *data_size,

  8173.                              uint8_t *buf, int buf_size)

  8174. {

  8175.     H264Context *h = avctx->priv_data;

  8176.     MpegEncContext *s = &h->s;

  8177.     AVFrame *pict = data;

  8178.     int buf_index;

  8179. 

  8180.     s->flags= avctx->flags;

  8181.     s->flags2= avctx->flags2;

  8182. 

  8183.    /* no supplementary picture */

  8184.     if (buf_size == 0) {

  8185.         return 0;

  8186.     }

  8187. 

  8188.     if(s->flags&CODEC_FLAG_TRUNCATED){

  8189.         int next= find_frame_end(h, buf, buf_size);

  8190. 

  8191.         if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )

  8192.             return buf_size;

  8193. //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);

  8194.     }

  8195. 

  8196.     if(h->is_avc && !h->got_avcC) {

  8197.         int i, cnt, nalsize;

  8198.         unsigned char *p = avctx->extradata;

  8199.         if(avctx->extradata_size < 7) {

  8200.             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");

  8201.             return -1;

  8202.         }

  8203.         if(*p != 1) {

  8204.             av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);

  8205.             return -1;

  8206.         }

  8207.         /* sps and pps in the avcC always have length coded with 2 bytes,

  8208.            so put a fake nal_length_size = 2 while parsing them */

  8209.         h->nal_length_size = 2;

  8210.         // Decode sps from avcC

  8211.         cnt = *(p+5) & 0x1f; // Number of sps

  8212.         p += 6;

  8213.         for (i = 0; i < cnt; i++) {

  8214.             nalsize = BE_16(p) + 2;

  8215.             if(decode_nal_units(h, p, nalsize) < 0) {

  8216.                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);

  8217.                 return -1;

  8218.             }

  8219.             p += nalsize;

  8220.         }

  8221.         // Decode pps from avcC

  8222.         cnt = *(p++); // Number of pps

  8223.         for (i = 0; i < cnt; i++) {

  8224.             nalsize = BE_16(p) + 2;

  8225.             if(decode_nal_units(h, p, nalsize)  != nalsize) {

  8226.                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);

  8227.                 return -1;

  8228.             }

  8229.             p += nalsize;

  8230.         }

  8231.         // Now store right nal length size, that will be use to parse all other nals

  8232.         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;

  8233.         // Do not reparse avcC

  8234.         h->got_avcC = 1;

  8235.     }

  8236. 

  8237.     if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){

  8238.         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)

  8239.             return -1;

  8240.     }

  8241. 

  8242.     buf_index=decode_nal_units(h, buf, buf_size);

  8243.     if(buf_index < 0)

  8244.         return -1;

  8245. 

  8246.     //FIXME do something with unavailable reference frames

  8247. 

  8248. //    if(ret==FRAME_SKIPPED) return get_consumed_bytes(s, buf_index, buf_size);

  8249.     if(!s->current_picture_ptr){

  8250.         av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");

  8251.         return -1;

  8252.     }

  8253. 

  8254.     {

  8255.         Picture *out = s->current_picture_ptr;

  8256. #if 0 //decode order

  8257.         *data_size = sizeof(AVFrame);

  8258. #else

  8259.         /* Sort B-frames into display order */

  8260.         Picture *cur = s->current_picture_ptr;

  8261.         Picture *prev = h->delayed_output_pic;

  8262.         int i, pics, cross_idr, out_of_order, out_idx;

  8263. 

  8264.         if(h->sps.bitstream_restriction_flag

  8265.            && s->avctx->has_b_frames < h->sps.num_reorder_frames){

  8266.             s->avctx->has_b_frames = h->sps.num_reorder_frames;

  8267.             s->low_delay = 0;

  8268.         }

  8269. 

  8270.         pics = 0;

  8271.         while(h->delayed_pic[pics]) pics++;

  8272.         h->delayed_pic[pics++] = cur;

  8273.         if(cur->reference == 0)

  8274.             cur->reference = 1;

  8275. 

  8276.         cross_idr = 0;

  8277.         for(i=0; h->delayed_pic[i]; i++)

  8278.             if(h->delayed_pic[i]->key_frame || h->delayed_pic[i]->poc==0)

  8279.                 cross_idr = 1;

  8280. 

  8281.         out = h->delayed_pic[0];

  8282.         out_idx = 0;

  8283.         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++)

  8284.             if(h->delayed_pic[i]->poc < out->poc){

  8285.                 out = h->delayed_pic[i];

  8286.                 out_idx = i;

  8287.             }

  8288. 

  8289.         out_of_order = !cross_idr && prev && out->poc < prev->poc;

  8290.         if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)

  8291.             { }

  8292.         else if(prev && pics <= s->avctx->has_b_frames)

  8293.             out = prev;

  8294.         else if((out_of_order && pics-1 == s->avctx->has_b_frames && pics < 15)

  8295.            || (s->low_delay &&

  8296.             ((!cross_idr && prev && out->poc > prev->poc + 2)

  8297.              || cur->pict_type == B_TYPE)))

  8298.         {

  8299.             s->low_delay = 0;

  8300.             s->avctx->has_b_frames++;

  8301.             out = prev;

  8302.         }

  8303.         else if(out_of_order)

  8304.             out = prev;

  8305. 

  8306.         if(out_of_order || pics > s->avctx->has_b_frames){

  8307.             for(i=out_idx; h->delayed_pic[i]; i++)

  8308.                 h->delayed_pic[i] = h->delayed_pic[i+1];

  8309.         }

  8310. 

  8311.         if(prev == out)

  8312.             *data_size = 0;

  8313.         else

  8314.             *data_size = sizeof(AVFrame);

  8315.         if(prev && prev != out && prev->reference == 1)

  8316.             prev->reference = 0;

  8317.         h->delayed_output_pic = out;

  8318. #endif

  8319. 

  8320.         if(out)

  8321.             *pict= *(AVFrame*)out;

  8322.         else

  8323.             av_log(avctx, AV_LOG_DEBUG, "no picture\n");

  8324.     }

  8325. 

  8326.     assert(pict->data[0] || !*data_size);

  8327.     ff_print_debug_info(s, pict);

  8328. //printf("out %d\n", (int)pict->data[0]);

  8329. #if 0 //?

  8330. 

  8331.     /* Return the Picture timestamp as the frame number */

  8332.     /* we substract 1 because it is added on utils.c    */

  8333.     avctx->frame_number = s->picture_number - 1;

  8334. #endif

  8335.     return get_consumed_bytes(s, buf_index, buf_size);

  8336. }

  8337. #if 0

  8338. static inline void fill_mb_avail(H264Context *h){

  8339.     MpegEncContext * const s = &h->s;

  8340.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  8341. 

  8342.     if(s->mb_y){

  8343.         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;

  8344.         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;

  8345.         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;

  8346.     }else{

  8347.         h->mb_avail[0]=

  8348.         h->mb_avail[1]=

  8349.         h->mb_avail[2]= 0;

  8350.     }

  8351.     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;

  8352.     h->mb_avail[4]= 1; //FIXME move out

  8353.     h->mb_avail[5]= 0; //FIXME move out

  8354. }

  8355. #endif

  8356. 

  8357. #if 0 //selftest

  8358. #define COUNT 8000

  8359. #define SIZE (COUNT*40)

  8360. int main(){

  8361.     int i;

  8362.     uint8_t temp[SIZE];

  8363.     PutBitContext pb;

  8364.     GetBitContext gb;

  8365. //    int int_temp[10000];

  8366.     DSPContext dsp;

  8367.     AVCodecContext avctx;

  8368. 

  8369.     dsputil_init(&dsp, &avctx);

  8370. 

  8371.     init_put_bits(&pb, temp, SIZE);

  8372.     printf("testing unsigned exp golomb\n");

  8373.     for(i=0; i<COUNT; i++){

  8374.         START_TIMER

  8375.         set_ue_golomb(&pb, i);

  8376.         STOP_TIMER("set_ue_golomb");

  8377.     }

  8378.     flush_put_bits(&pb);

  8379. 

  8380.     init_get_bits(&gb, temp, 8*SIZE);

  8381.     for(i=0; i<COUNT; i++){

  8382.         int j, s;

  8383. 

  8384.         s= show_bits(&gb, 24);

  8385. 

  8386.         START_TIMER

  8387.         j= get_ue_golomb(&gb);

  8388.         if(j != i){

  8389.             printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);

  8390. //            return -1;

  8391.         }

  8392.         STOP_TIMER("get_ue_golomb");

  8393.     }

  8394. 

  8395. 

  8396.     init_put_bits(&pb, temp, SIZE);

  8397.     printf("testing signed exp golomb\n");

  8398.     for(i=0; i<COUNT; i++){

  8399.         START_TIMER

  8400.         set_se_golomb(&pb, i - COUNT/2);

  8401.         STOP_TIMER("set_se_golomb");

  8402.     }

  8403.     flush_put_bits(&pb);

  8404. 

  8405.     init_get_bits(&gb, temp, 8*SIZE);

  8406.     for(i=0; i<COUNT; i++){

  8407.         int j, s;

  8408. 

  8409.         s= show_bits(&gb, 24);

  8410. 

  8411.         START_TIMER

  8412.         j= get_se_golomb(&gb);

  8413.         if(j != i - COUNT/2){

  8414.             printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);

  8415. //            return -1;

  8416.         }

  8417.         STOP_TIMER("get_se_golomb");

  8418.     }

  8419. 

  8420.     printf("testing 4x4 (I)DCT\n");

  8421. 

  8422.     DCTELEM block[16];

  8423.     uint8_t src[16], ref[16];

  8424.     uint64_t error= 0, max_error=0;

  8425. 

  8426.     for(i=0; i<COUNT; i++){

  8427.         int j;

  8428. //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);

  8429.         for(j=0; j<16; j++){

  8430.             ref[j]= random()%255;

  8431.             src[j]= random()%255;

  8432.         }

  8433. 

  8434.         h264_diff_dct_c(block, src, ref, 4);

  8435. 

  8436.         //normalize

  8437.         for(j=0; j<16; j++){

  8438. //            printf("%d ", block[j]);

  8439.             block[j]= block[j]*4;

  8440.             if(j&1) block[j]= (block[j]*4 + 2)/5;

  8441.             if(j&4) block[j]= (block[j]*4 + 2)/5;

  8442.         }

  8443. //        printf("\n");

  8444. 

  8445.         s->dsp.h264_idct_add(ref, block, 4);

  8446. /*        for(j=0; j<16; j++){

  8447.             printf("%d ", ref[j]);

  8448.         }

  8449.         printf("\n");*/

  8450. 

  8451.         for(j=0; j<16; j++){

  8452.             int diff= FFABS(src[j] - ref[j]);

  8453. 

  8454.             error+= diff*diff;

  8455.             max_error= FFMAX(max_error, diff);

  8456.         }

  8457.     }

  8458.     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );

  8459. #if 0

  8460.     printf("testing quantizer\n");

  8461.     for(qp=0; qp<52; qp++){

  8462.         for(i=0; i<16; i++)

  8463.             src1_block[i]= src2_block[i]= random()%255;

  8464. 

  8465.     }

  8466. #endif

  8467.     printf("Testing NAL layer\n");

  8468. 

  8469.     uint8_t bitstream[COUNT];

  8470.     uint8_t nal[COUNT*2];

  8471.     H264Context h;

  8472.     memset(&h, 0, sizeof(H264Context));

  8473. 

  8474.     for(i=0; i<COUNT; i++){

  8475.         int zeros= i;

  8476.         int nal_length;

  8477.         int consumed;

  8478.         int out_length;

  8479.         uint8_t *out;

  8480.         int j;

  8481. 

  8482.         for(j=0; j<COUNT; j++){

  8483.             bitstream[j]= (random() % 255) + 1;

  8484.         }

  8485. 

  8486.         for(j=0; j<zeros; j++){

  8487.             int pos= random() % COUNT;

  8488.             while(bitstream[pos] == 0){

  8489.                 pos++;

  8490.                 pos %= COUNT;

  8491.             }

  8492.             bitstream[pos]=0;

  8493.         }

  8494. 

  8495.         START_TIMER

  8496. 

  8497.         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);

  8498.         if(nal_length<0){

  8499.             printf("encoding failed\n");

  8500.             return -1;

  8501.         }

  8502. 

  8503.         out= decode_nal(&h, nal, &out_length, &consumed, nal_length);

  8504. 

  8505.         STOP_TIMER("NAL")

  8506. 

  8507.         if(out_length != COUNT){

  8508.             printf("incorrect length %d %d\n", out_length, COUNT);

  8509.             return -1;

  8510.         }

  8511. 

  8512.         if(consumed != nal_length){

  8513.             printf("incorrect consumed length %d %d\n", nal_length, consumed);

  8514.             return -1;

  8515.         }

  8516. 

  8517.         if(memcmp(bitstream, out, COUNT)){

  8518.             printf("missmatch\n");

  8519.             return -1;

  8520.         }

  8521.     }

  8522. 

  8523.     printf("Testing RBSP\n");

  8524. 

  8525. 

  8526.     return 0;

  8527. }

  8528. #endif

  8529. 

  8530. 

  8531. static int decode_end(AVCodecContext *avctx)

  8532. {

  8533.     H264Context *h = avctx->priv_data;

  8534.     MpegEncContext *s = &h->s;

  8535. 

  8536.     av_freep(&h->rbsp_buffer);

  8537.     free_tables(h); //FIXME cleanup init stuff perhaps

  8538.     MPV_common_end(s);

  8539. 

  8540. //    memset(h, 0, sizeof(H264Context));

  8541. 

  8542.     return 0;

  8543. }

  8544. 

  8545. 

  8546. AVCodec h264_decoder = {

  8547.     "h264",

  8548.     CODEC_TYPE_VIDEO,

  8549.     CODEC_ID_H264,

  8550.     sizeof(H264Context),

  8551.     decode_init,

  8552.     NULL,

  8553.     decode_end,

  8554.     decode_frame,

  8555.     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY,

  8556.     .flush= flush_dpb,

  8557. };

  8558. 

  8559. #ifdef CONFIG_H264_PARSER

  8560. AVCodecParser h264_parser = {

  8561.     { CODEC_ID_H264 },

  8562.     sizeof(H264Context),

  8563.     NULL,

  8564.     h264_parse,

  8565.     ff_parse_close,

  8566.     h264_split,

  8567. };

  8568. #endif

  8569. 

  8570. #include "svq3.c"